cc: Use worker context for one-copy tile initialization.

This moves management of staging resources to
OneCopyTileTaskWorkerPool class. This makes it possible
to use a worker context to issue and detect when copy
operations complete.

BUG=490295
CQ_INCLUDE_TRYBOTS=tryserver.blink:linux_blink_rel

Review URL: https://codereview.chromium.org/1230203007

Cr-Commit-Position: refs/heads/master@{#342235}

55 files changed, 746 insertions, 803 deletions

diff --git a/cc/output/context_provider.h b/cc/output/context_provider.h
index aa4367d..c024e02 100644
--- a/cc/output/context_provider.h
+++ b/cc/output/context_provider.h
@@ -7,6 +7,7 @@
 
 #include "base/callback.h"
 #include "base/memory/ref_counted.h"
+#include "base/synchronization/lock.h"
 #include "cc/base/cc_export.h"
 #include "gpu/command_buffer/common/capabilities.h"
 
@@ -26,6 +27,27 @@ struct ManagedMemoryPolicy;
 
 class ContextProvider : public base::RefCountedThreadSafe<ContextProvider> {
  public:
+  class ScopedContextLock {
+   public:
+    explicit ScopedContextLock(ContextProvider* context_provider)
+        : context_provider_(context_provider),
+          context_lock_(*context_provider_->GetLock()) {
+      // Allow current thread to bind to |context_provider|.
+      context_provider_->DetachFromThread();
+    }
+    ~ScopedContextLock() {
+      // Allow a different thread to bind to |context_provider|.
+      context_provider_->DetachFromThread();
+    }
+
+    gpu::gles2::GLES2Interface* ContextGL() {
+      return context_provider_->ContextGL();
+    }
+
+   private:
+    ContextProvider* const context_provider_;
+    base::AutoLock context_lock_;
+  };
   // Bind the 3d context to the current thread. This should be called before
   // accessing the contexts. Calling it more than once should have no effect.
   // Once this function has been called, the class should only be accessed
diff --git a/cc/output/output_surface.cc b/cc/output/output_surface.cc
index a597a87..0786d66 100644
--- a/cc/output/output_surface.cc
+++ b/cc/output/output_surface.cc
@@ -107,10 +107,6 @@ OutputSurface::~OutputSurface() {
     context_provider_->SetMemoryPolicyChangedCallback(
         ContextProvider::MemoryPolicyChangedCallback());
   }
-  if (worker_context_provider_.get()) {
-    worker_context_provider_->SetLostContextCallback(
-        ContextProvider::LostContextCallback());
-  }
 }
 
 bool OutputSurface::HasExternalStencilTest() const {
@@ -134,14 +130,8 @@ bool OutputSurface::BindToClient(OutputSurfaceClient* client) {
 
   if (success && worker_context_provider_.get()) {
     success = worker_context_provider_->BindToCurrentThread();
-    if (success) {
+    if (success)
       worker_context_provider_->SetupLock();
-      // The destructor resets the context lost callback, so base::Unretained
-      // is safe, as long as the worker threads stop using the context before
-      // the output surface is destroyed.
-      worker_context_provider_->SetLostContextCallback(base::Bind(
-          &OutputSurface::DidLoseOutputSurface, base::Unretained(this)));
-    }
   }
 
   if (!success)
@@ -217,11 +207,7 @@ void OutputSurface::SetWorkerContextShouldAggressivelyFreeResources(
                "OutputSurface::SetWorkerContextShouldAggressivelyFreeResources",
                "aggressively_free_resources", aggressively_free_resources);
   if (auto* context_provider = worker_context_provider()) {
-    // The context lock must be held while accessing the worker context.
-    base::AutoLock context_lock(*context_provider->GetLock());
-
-    // Allow context to bind to current thread.
-    context_provider->DetachFromThread();
+    ContextProvider::ScopedContextLock scoped_context(context_provider);
 
     if (aggressively_free_resources) {
       context_provider->DeleteCachedResources();
@@ -231,9 +217,6 @@ void OutputSurface::SetWorkerContextShouldAggressivelyFreeResources(
       context_support->SetAggressivelyFreeResources(
           aggressively_free_resources);
     }
-
-    // Allow context to bind to other threads.
-    context_provider->DetachFromThread();
   }
 }
 
diff --git a/cc/raster/gpu_tile_task_worker_pool.cc b/cc/raster/gpu_tile_task_worker_pool.cc
index 3567308..2380455 100644
--- a/cc/raster/gpu_tile_task_worker_pool.cc
+++ b/cc/raster/gpu_tile_task_worker_pool.cc
@@ -46,13 +46,9 @@ class RasterBufferImpl : public RasterBuffer {
     ContextProvider* context_provider = rasterizer_->resource_provider()
                                             ->output_surface()
                                             ->worker_context_provider();
+    DCHECK(context_provider);
 
-    // The context lock must be held while accessing the context on a
-    // worker thread.
-    base::AutoLock context_lock(*context_provider->GetLock());
-
-    // Allow this worker thread to bind to context_provider.
-    context_provider->DetachFromThread();
+    ContextProvider::ScopedContextLock scoped_context(context_provider);
 
     gfx::Rect playback_rect = raster_full_rect;
     if (resource_has_previous_content_) {
@@ -67,10 +63,7 @@ class RasterBufferImpl : public RasterBuffer {
                                  playback_rect, scale);
 
     // Barrier to sync worker context output to cc context.
-    context_provider->ContextGL()->OrderingBarrierCHROMIUM();
-
-    // Allow compositor thread to bind to context_provider.
-    context_provider->DetachFromThread();
+    scoped_context.ContextGL()->OrderingBarrierCHROMIUM();
   }
 
  private:
diff --git a/cc/raster/one_copy_tile_task_worker_pool.cc b/cc/raster/one_copy_tile_task_worker_pool.cc
index 64da37e..3284fe9 100644
--- a/cc/raster/one_copy_tile_task_worker_pool.cc
+++ b/cc/raster/one_copy_tile_task_worker_pool.cc
@@ -8,16 +8,20 @@
 #include <limits>
 
 #include "base/strings/stringprintf.h"
+#include "base/thread_task_runner_handle.h"
+#include "base/trace_event/memory_dump_manager.h"
 #include "base/trace_event/trace_event.h"
 #include "base/trace_event/trace_event_argument.h"
 #include "cc/base/math_util.h"
 #include "cc/debug/traced_value.h"
 #include "cc/raster/raster_buffer.h"
 #include "cc/resources/platform_color.h"
-#include "cc/resources/resource_pool.h"
+#include "cc/resources/resource_format.h"
+#include "cc/resources/resource_util.h"
 #include "cc/resources/scoped_resource.h"
+#include "gpu/GLES2/gl2extchromium.h"
 #include "gpu/command_buffer/client/gles2_interface.h"
-#include "ui/gfx/gpu_memory_buffer.h"
+#include "gpu/command_buffer/client/gpu_memory_buffer_manager.h"
 
 namespace cc {
 namespace {
@@ -26,49 +30,15 @@ class RasterBufferImpl : public RasterBuffer {
  public:
   RasterBufferImpl(OneCopyTileTaskWorkerPool* worker_pool,
                    ResourceProvider* resource_provider,
-                   ResourcePool* resource_pool,
                    ResourceFormat resource_format,
-                   const Resource* output_resource,
+                   const Resource* resource,
                    uint64_t previous_content_id)
       : worker_pool_(worker_pool),
-        resource_provider_(resource_provider),
-        resource_pool_(resource_pool),
-        output_resource_(output_resource),
-        raster_content_id_(0),
-        sequence_(0) {
-    if (worker_pool->have_persistent_gpu_memory_buffers() &&
-        previous_content_id) {
-      raster_resource_ =
-          resource_pool->TryAcquireResourceWithContentId(previous_content_id);
-    }
-    if (raster_resource_) {
-      raster_content_id_ = previous_content_id;
-      DCHECK_EQ(resource_format, raster_resource_->format());
-      DCHECK_EQ(output_resource->size().ToString(),
-                raster_resource_->size().ToString());
-    } else {
-      raster_resource_ = resource_pool->AcquireResource(output_resource->size(),
-                                                        resource_format);
-    }
-
-    lock_.reset(new ResourceProvider::ScopedWriteLockGpuMemoryBuffer(
-        resource_provider_, raster_resource_->id()));
-  }
+        resource_(resource),
+        lock_(resource_provider, resource->id()),
+        previous_content_id_(previous_content_id) {}
 
-  ~RasterBufferImpl() override {
-    // Release write lock in case a copy was never scheduled.
-    lock_.reset();
-
-    // Make sure any scheduled copy operations are issued before we release the
-    // raster resource.
-    if (sequence_)
-      worker_pool_->AdvanceLastIssuedCopyTo(sequence_);
-
-    // Return resources to pool so they can be used by another RasterBuffer
-    // instance.
-    resource_pool_->ReleaseResource(raster_resource_.Pass(),
-                                    raster_content_id_);
-  }
+  ~RasterBufferImpl() override {}
 
   // Overridden from RasterBuffer:
   void Playback(const RasterSource* raster_source,
@@ -77,55 +47,117 @@ class RasterBufferImpl : public RasterBuffer {
                 uint64_t new_content_id,
                 float scale,
                 bool include_images) override {
-    // If there's a raster_content_id_, we are reusing a resource with that
-    // content id.
-    bool reusing_raster_resource = raster_content_id_ != 0;
-    sequence_ = worker_pool_->PlaybackAndScheduleCopyOnWorkerThread(
-        reusing_raster_resource, lock_.Pass(), raster_resource_.get(),
-        output_resource_, raster_source, raster_full_rect, raster_dirty_rect,
-        scale, include_images);
-    // Store the content id of the resource to return to the pool.
-    raster_content_id_ = new_content_id;
+    worker_pool_->PlaybackAndCopyOnWorkerThread(
+        resource_, &lock_, raster_source, raster_full_rect, raster_dirty_rect,
+        scale, include_images, previous_content_id_, new_content_id);
   }
 
  private:
   OneCopyTileTaskWorkerPool* worker_pool_;
-  ResourceProvider* resource_provider_;
-  ResourcePool* resource_pool_;
-  const Resource* output_resource_;
-  uint64_t raster_content_id_;
-  scoped_ptr<ScopedResource> raster_resource_;
-  scoped_ptr<ResourceProvider::ScopedWriteLockGpuMemoryBuffer> lock_;
-  CopySequenceNumber sequence_;
+  const Resource* resource_;
+  ResourceProvider::ScopedWriteLockGL lock_;
+  uint64_t previous_content_id_;
 
   DISALLOW_COPY_AND_ASSIGN(RasterBufferImpl);
 };
 
-// Number of in-flight copy operations to allow.
-const int kMaxCopyOperations = 32;
-
-// Delay been checking for copy operations to complete.
-const int kCheckForCompletedCopyOperationsTickRateMs = 1;
+// Delay between checking for query result to be available.
+const int kCheckForQueryResultAvailableTickRateMs = 1;
 
-// Number of failed attempts to allow before we perform a check that will
-// wait for copy operations to complete if needed.
-const int kFailedAttemptsBeforeWaitIfNeeded = 256;
+// Number of attempts to allow before we perform a check that will wait for
+// query to complete.
+const int kMaxCheckForQueryResultAvailableAttempts = 256;
 
 // 4MiB is the size of 4 512x512 tiles, which has proven to be a good
 // default batch size for copy operations.
 const int kMaxBytesPerCopyOperation = 1024 * 1024 * 4;
 
+// Delay before a staging buffer might be released.
+const int kStagingBufferExpirationDelayMs = 1000;
+
+bool CheckForQueryResult(gpu::gles2::GLES2Interface* gl, unsigned query_id) {
+  unsigned complete = 1;
+  gl->GetQueryObjectuivEXT(query_id, GL_QUERY_RESULT_AVAILABLE_EXT, &complete);
+  return !!complete;
+}
+
+void WaitForQueryResult(gpu::gles2::GLES2Interface* gl, unsigned query_id) {
+  TRACE_EVENT0("cc", "WaitForQueryResult");
+
+  int attempts_left = kMaxCheckForQueryResultAvailableAttempts;
+  while (attempts_left--) {
+    if (CheckForQueryResult(gl, query_id))
+      break;
+
+    base::PlatformThread::Sleep(base::TimeDelta::FromMilliseconds(
+        kCheckForQueryResultAvailableTickRateMs));
+  }
+
+  unsigned result = 0;
+  gl->GetQueryObjectuivEXT(query_id, GL_QUERY_RESULT_EXT, &result);
+}
+
 }  // namespace
 
-OneCopyTileTaskWorkerPool::CopyOperation::CopyOperation(
-    scoped_ptr<ResourceProvider::ScopedWriteLockGpuMemoryBuffer> src_write_lock,
-    const Resource* src,
-    const Resource* dst,
-    const gfx::Rect& rect)
-    : src_write_lock(src_write_lock.Pass()), src(src), dst(dst), rect(rect) {
+OneCopyTileTaskWorkerPool::StagingBuffer::StagingBuffer(const gfx::Size& size)
+    : size(size), texture_id(0), image_id(0), query_id(0), content_id(0) {}
+
+OneCopyTileTaskWorkerPool::StagingBuffer::~StagingBuffer() {
+  DCHECK_EQ(texture_id, 0u);
+  DCHECK_EQ(image_id, 0u);
+  DCHECK_EQ(query_id, 0u);
 }
 
-OneCopyTileTaskWorkerPool::CopyOperation::~CopyOperation() {
+void OneCopyTileTaskWorkerPool::StagingBuffer::DestroyGLResources(
+    gpu::gles2::GLES2Interface* gl) {
+  if (query_id) {
+    gl->DeleteQueriesEXT(1, &query_id);
+    query_id = 0;
+  }
+  if (image_id) {
+    gl->DestroyImageCHROMIUM(image_id);
+    image_id = 0;
+  }
+  if (texture_id) {
+    gl->DeleteTextures(1, &texture_id);
+    texture_id = 0;
+  }
+}
+
+void OneCopyTileTaskWorkerPool::StagingBuffer::OnMemoryDump(
+    base::trace_event::ProcessMemoryDump* pmd,
+    ResourceFormat format,
+    bool in_free_list) const {
+  if (!gpu_memory_buffer)
+    return;
+
+  gfx::GpuMemoryBufferId buffer_id = gpu_memory_buffer->GetId();
+  std::string buffer_dump_name =
+      base::StringPrintf("cc/one_copy/staging_memory/buffer_%d", buffer_id);
+  base::trace_event::MemoryAllocatorDump* buffer_dump =
+      pmd->CreateAllocatorDump(buffer_dump_name);
+
+  uint64_t buffer_size_in_bytes =
+      ResourceUtil::UncheckedSizeInBytes<uint64_t>(size, format);
+  buffer_dump->AddScalar(base::trace_event::MemoryAllocatorDump::kNameSize,
+                         base::trace_event::MemoryAllocatorDump::kUnitsBytes,
+                         buffer_size_in_bytes);
+  buffer_dump->AddScalar("free_size",
+                         base::trace_event::MemoryAllocatorDump::kUnitsBytes,
+                         in_free_list ? buffer_size_in_bytes : 0);
+
+  // Emit an ownership edge towards a global allocator dump node.
+  const uint64 tracing_process_id =
+      base::trace_event::MemoryDumpManager::GetInstance()
+          ->GetTracingProcessId();
+  base::trace_event::MemoryAllocatorDumpGuid shared_buffer_guid =
+      gfx::GetGpuMemoryBufferGUIDForTracing(tracing_process_id, buffer_id);
+  pmd->CreateSharedGlobalAllocatorDump(shared_buffer_guid);
+
+  // By creating an edge with a higher |importance| (w.r.t. browser-side dumps)
+  // the tracing UI will account the effective size of the buffer to the child.
+  const int kImportance = 2;
+  pmd->AddOwnershipEdge(buffer_dump->guid(), shared_buffer_guid, kImportance);
 }
 
 // static
@@ -134,51 +166,52 @@ scoped_ptr<TileTaskWorkerPool> OneCopyTileTaskWorkerPool::Create(
     TaskGraphRunner* task_graph_runner,
     ContextProvider* context_provider,
     ResourceProvider* resource_provider,
-    ResourcePool* resource_pool,
     int max_copy_texture_chromium_size,
-    bool have_persistent_gpu_memory_buffers) {
+    bool use_persistent_gpu_memory_buffers,
+    unsigned image_target,
+    int max_staging_buffers) {
   return make_scoped_ptr<TileTaskWorkerPool>(new OneCopyTileTaskWorkerPool(
-      task_runner, task_graph_runner, context_provider, resource_provider,
-      resource_pool, max_copy_texture_chromium_size,
-      have_persistent_gpu_memory_buffers));
+      task_runner, task_graph_runner, resource_provider,
+      max_copy_texture_chromium_size, use_persistent_gpu_memory_buffers,
+      image_target, max_staging_buffers));
 }
 
 OneCopyTileTaskWorkerPool::OneCopyTileTaskWorkerPool(
     base::SequencedTaskRunner* task_runner,
     TaskGraphRunner* task_graph_runner,
-    ContextProvider* context_provider,
     ResourceProvider* resource_provider,
-    ResourcePool* resource_pool,
     int max_copy_texture_chromium_size,
-    bool have_persistent_gpu_memory_buffers)
+    bool use_persistent_gpu_memory_buffers,
+    unsigned image_target,
+    int max_staging_buffers)
     : task_runner_(task_runner),
       task_graph_runner_(task_graph_runner),
       namespace_token_(task_graph_runner->GetNamespaceToken()),
-      context_provider_(context_provider),
       resource_provider_(resource_provider),
-      resource_pool_(resource_pool),
       max_bytes_per_copy_operation_(
           max_copy_texture_chromium_size
               ? std::min(kMaxBytesPerCopyOperation,
                          max_copy_texture_chromium_size)
               : kMaxBytesPerCopyOperation),
-      have_persistent_gpu_memory_buffers_(have_persistent_gpu_memory_buffers),
-      last_issued_copy_operation_(0),
-      last_flushed_copy_operation_(0),
-      lock_(),
-      copy_operation_count_cv_(&lock_),
+      use_persistent_gpu_memory_buffers_(use_persistent_gpu_memory_buffers),
+      image_target_(image_target),
       bytes_scheduled_since_last_flush_(0),
-      issued_copy_operation_count_(0),
-      next_copy_operation_sequence_(1),
-      check_for_completed_copy_operations_pending_(false),
-      shutdown_(false),
+      max_staging_buffers_(max_staging_buffers),
+      staging_buffer_expiration_delay_(
+          base::TimeDelta::FromMilliseconds(kStagingBufferExpirationDelayMs)),
+      reduce_memory_usage_pending_(false),
       weak_ptr_factory_(this),
       task_set_finished_weak_ptr_factory_(this) {
-  DCHECK(context_provider_);
+  base::trace_event::MemoryDumpManager::GetInstance()->RegisterDumpProvider(
+      this, base::ThreadTaskRunnerHandle::Get());
+  reduce_memory_usage_callback_ =
+      base::Bind(&OneCopyTileTaskWorkerPool::ReduceMemoryUsage,
+                 weak_ptr_factory_.GetWeakPtr());
 }
 
 OneCopyTileTaskWorkerPool::~OneCopyTileTaskWorkerPool() {
-  DCHECK_EQ(pending_copy_operations_.size(), 0u);
+  base::trace_event::MemoryDumpManager::GetInstance()->UnregisterDumpProvider(
+      this);
 }
 
 TileTaskRunner* OneCopyTileTaskWorkerPool::AsTileTaskRunner() {
@@ -192,28 +225,21 @@ void OneCopyTileTaskWorkerPool::SetClient(TileTaskRunnerClient* client) {
 void OneCopyTileTaskWorkerPool::Shutdown() {
   TRACE_EVENT0("cc", "OneCopyTileTaskWorkerPool::Shutdown");
 
-  {
-    base::AutoLock lock(lock_);
-
-    shutdown_ = true;
-    copy_operation_count_cv_.Signal();
-  }
-
   TaskGraph empty;
   task_graph_runner_->ScheduleTasks(namespace_token_, &empty);
   task_graph_runner_->WaitForTasksToFinishRunning(namespace_token_);
+
+  base::AutoLock lock(lock_);
+
+  if (buffers_.empty())
+    return;
+
+  ReleaseBuffersNotUsedSince(base::TimeTicks() + base::TimeDelta::Max());
 }
 
 void OneCopyTileTaskWorkerPool::ScheduleTasks(TileTaskQueue* queue) {
   TRACE_EVENT0("cc", "OneCopyTileTaskWorkerPool::ScheduleTasks");
 
-#if DCHECK_IS_ON()
-  {
-    base::AutoLock lock(lock_);
-    DCHECK(!shutdown_);
-  }
-#endif
-
   if (tasks_pending_.none())
     TRACE_EVENT_ASYNC_BEGIN0("cc", "ScheduledTasks", this);
 
@@ -238,8 +264,6 @@ void OneCopyTileTaskWorkerPool::ScheduleTasks(TileTaskQueue* queue) {
                    task_set_finished_weak_ptr_factory_.GetWeakPtr(), task_set));
   }
 
-  resource_pool_->CheckBusyResources(false);
-
   for (TileTaskQueue::Item::Vector::const_iterator it = queue->items.begin();
        it != queue->items.end(); ++it) {
     const TileTaskQueue::Item& item = *it;
@@ -266,14 +290,19 @@ void OneCopyTileTaskWorkerPool::ScheduleTasks(TileTaskQueue* queue) {
   }
 
   ScheduleTasksOnOriginThread(this, &graph_);
+
+  // Barrier to sync any new resources to the worker context.
+  resource_provider_->output_surface()
+      ->context_provider()
+      ->ContextGL()
+      ->OrderingBarrierCHROMIUM();
+
   task_graph_runner_->ScheduleTasks(namespace_token_, &graph_);
 
   std::copy(new_task_set_finished_tasks,
             new_task_set_finished_tasks + kNumberOfTaskSets,
             task_set_finished_tasks_);
 
-  resource_pool_->ReduceResourceUsage();
-
   TRACE_EVENT_ASYNC_STEP_INTO1("cc", "ScheduledTasks", this, "running", "state",
                                StateAsValue());
 }
@@ -312,10 +341,9 @@ scoped_ptr<RasterBuffer> OneCopyTileTaskWorkerPool::AcquireBufferForRaster(
   // TODO(danakj): If resource_content_id != 0, we only need to copy/upload
   // the dirty rect.
   DCHECK_EQ(resource->format(), resource_provider_->best_texture_format());
-  return make_scoped_ptr<RasterBuffer>(
-      new RasterBufferImpl(this, resource_provider_, resource_pool_,
-                           resource_provider_->best_texture_format(), resource,
-                           previous_content_id));
+  return make_scoped_ptr<RasterBuffer>(new RasterBufferImpl(
+      this, resource_provider_, resource_provider_->best_texture_format(),
+      resource, previous_content_id));
 }
 
 void OneCopyTileTaskWorkerPool::ReleaseBufferForRaster(
@@ -323,238 +351,363 @@ void OneCopyTileTaskWorkerPool::ReleaseBufferForRaster(
   // Nothing to do here. RasterBufferImpl destructor cleans up after itself.
 }
 
-CopySequenceNumber
-OneCopyTileTaskWorkerPool::PlaybackAndScheduleCopyOnWorkerThread(
-    bool reusing_raster_resource,
-    scoped_ptr<ResourceProvider::ScopedWriteLockGpuMemoryBuffer>
-        raster_resource_write_lock,
-    const Resource* raster_resource,
-    const Resource* output_resource,
+void OneCopyTileTaskWorkerPool::PlaybackAndCopyOnWorkerThread(
+    const Resource* resource,
+    const ResourceProvider::ScopedWriteLockGL* resource_lock,
     const RasterSource* raster_source,
     const gfx::Rect& raster_full_rect,
     const gfx::Rect& raster_dirty_rect,
     float scale,
-    bool include_images) {
-  gfx::GpuMemoryBuffer* gpu_memory_buffer =
-      raster_resource_write_lock->GetGpuMemoryBuffer();
-  if (gpu_memory_buffer) {
-    void* data = NULL;
-    bool rv = gpu_memory_buffer->Map(&data);
-    DCHECK(rv);
-    int stride;
-    gpu_memory_buffer->GetStride(&stride);
-    // TileTaskWorkerPool::PlaybackToMemory only supports unsigned strides.
-    DCHECK_GE(stride, 0);
+    bool include_images,
+    uint64_t previous_content_id,
+    uint64_t new_content_id) {
+  base::AutoLock lock(lock_);
 
-    gfx::Rect playback_rect = raster_full_rect;
-    if (reusing_raster_resource) {
-      playback_rect.Intersect(raster_dirty_rect);
-    }
-    DCHECK(!playback_rect.IsEmpty())
-        << "Why are we rastering a tile that's not dirty?";
-    TileTaskWorkerPool::PlaybackToMemory(
-        data, raster_resource->format(), raster_resource->size(),
-        static_cast<size_t>(stride), raster_source, raster_full_rect,
-        playback_rect, scale, include_images);
-    gpu_memory_buffer->Unmap();
-  }
+  scoped_ptr<StagingBuffer> staging_buffer =
+      AcquireStagingBuffer(resource, previous_content_id);
+  DCHECK(staging_buffer);
 
-  base::AutoLock lock(lock_);
+  {
+    base::AutoUnlock unlock(lock_);
+
+    // Allocate GpuMemoryBuffer if necessary.
+    if (!staging_buffer->gpu_memory_buffer) {
+      staging_buffer->gpu_memory_buffer =
+          resource_provider_->gpu_memory_buffer_manager()
+              ->AllocateGpuMemoryBuffer(
+                  staging_buffer->size,
+                  BufferFormat(resource_provider_->best_texture_format()),
+                  use_persistent_gpu_memory_buffers_
+                      ? gfx::BufferUsage::PERSISTENT_MAP
+                      : gfx::BufferUsage::MAP);
+    }
 
-  CopySequenceNumber sequence = 0;
-  int bytes_per_row = (BitsPerPixel(raster_resource->format()) *
-                       raster_resource->size().width()) /
-                      8;
-  int chunk_size_in_rows =
-      std::max(1, max_bytes_per_copy_operation_ / bytes_per_row);
-  // Align chunk size to 4. Required to support compressed texture formats.
-  chunk_size_in_rows = MathUtil::UncheckedRoundUp(chunk_size_in_rows, 4);
-  int y = 0;
-  int height = raster_resource->size().height();
-  while (y < height) {
-    int failed_attempts = 0;
-    while ((pending_copy_operations_.size() + issued_copy_operation_count_) >=
-           kMaxCopyOperations) {
-      // Ignore limit when shutdown is set.
-      if (shutdown_)
-        break;
+    gfx::Rect playback_rect = raster_full_rect;
+    if (use_persistent_gpu_memory_buffers_ && previous_content_id) {
+      // Reduce playback rect to dirty region if the content id of the staging
+      // buffer matches the prevous content id.
+      if (previous_content_id == staging_buffer->content_id)
+        playback_rect.Intersect(raster_dirty_rect);
+    }
 
-      ++failed_attempts;
+    if (staging_buffer->gpu_memory_buffer) {
+      void* data = nullptr;
+      bool rv = staging_buffer->gpu_memory_buffer->Map(&data);
+      DCHECK(rv);
+      int stride;
+      staging_buffer->gpu_memory_buffer->GetStride(&stride);
+      // TileTaskWorkerPool::PlaybackToMemory only supports unsigned strides.
+      DCHECK_GE(stride, 0);
+
+      DCHECK(!playback_rect.IsEmpty())
+          << "Why are we rastering a tile that's not dirty?";
+      TileTaskWorkerPool::PlaybackToMemory(
+          data, resource_provider_->best_texture_format(), staging_buffer->size,
+          static_cast<size_t>(stride), raster_source, raster_full_rect,
+          playback_rect, scale, include_images);
+      staging_buffer->gpu_memory_buffer->Unmap();
+      staging_buffer->content_id = new_content_id;
+    }
+  }
 
-      // Schedule a check that will also wait for operations to complete
-      // after too many failed attempts.
-      bool wait_if_needed = failed_attempts > kFailedAttemptsBeforeWaitIfNeeded;
+  ContextProvider* context_provider =
+      resource_provider_->output_surface()->worker_context_provider();
+  DCHECK(context_provider);
 
-      // Schedule a check for completed copy operations if too many operations
-      // are currently in-flight.
-      ScheduleCheckForCompletedCopyOperationsWithLockAcquired(wait_if_needed);
+  {
+    ContextProvider::ScopedContextLock scoped_context(context_provider);
+
+    gpu::gles2::GLES2Interface* gl = scoped_context.ContextGL();
+    DCHECK(gl);
+
+    if (!staging_buffer->texture_id) {
+      gl->GenTextures(1, &staging_buffer->texture_id);
+      gl->BindTexture(image_target_, staging_buffer->texture_id);
+      gl->TexParameteri(image_target_, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+      gl->TexParameteri(image_target_, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
+      gl->TexParameteri(image_target_, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
+      gl->TexParameteri(image_target_, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
+      if (staging_buffer->gpu_memory_buffer) {
+        DCHECK(!staging_buffer->image_id);
+        staging_buffer->image_id = gl->CreateImageCHROMIUM(
+            staging_buffer->gpu_memory_buffer->AsClientBuffer(),
+            staging_buffer->size.width(), staging_buffer->size.height(),
+            GLInternalFormat(resource_provider_->best_texture_format()));
+        gl->BindTexImage2DCHROMIUM(image_target_, staging_buffer->image_id);
+      } else {
+        gl->BindTexture(image_target_, staging_buffer->texture_id);
+        gl->ReleaseTexImage2DCHROMIUM(image_target_, staging_buffer->image_id);
+        gl->BindTexImage2DCHROMIUM(image_target_, staging_buffer->image_id);
+      }
+    }
 
-      {
-        TRACE_EVENT0("cc", "WaitingForCopyOperationsToComplete");
+    if (resource_provider_->use_sync_query()) {
+      if (!staging_buffer->query_id)
+        gl->GenQueriesEXT(1, &staging_buffer->query_id);
+
+#if defined(OS_CHROMEOS)
+      // TODO(reveman): This avoids a performance problem on some ChromeOS
+      // devices. This needs to be removed to support native GpuMemoryBuffer
+      // implementations. crbug.com/436314
+      gl->BeginQueryEXT(GL_COMMANDS_ISSUED_CHROMIUM, staging_buffer->query_id);
+#else
+      gl->BeginQueryEXT(GL_COMMANDS_COMPLETED_CHROMIUM,
+                        staging_buffer->query_id);
+#endif
+    }
 
-        // Wait for in-flight copy operations to drop below limit.
-        copy_operation_count_cv_.Wait();
+    int bytes_per_row =
+        (BitsPerPixel(resource_provider_->best_texture_format()) *
+         resource->size().width()) /
+        8;
+    int chunk_size_in_rows =
+        std::max(1, max_bytes_per_copy_operation_ / bytes_per_row);
+    // Align chunk size to 4. Required to support compressed texture formats.
+    chunk_size_in_rows = MathUtil::UncheckedRoundUp(chunk_size_in_rows, 4);
+    int y = 0;
+    int height = resource->size().height();
+    while (y < height) {
+      // Copy at most |chunk_size_in_rows|.
+      int rows_to_copy = std::min(chunk_size_in_rows, height - y);
+      DCHECK_GT(rows_to_copy, 0);
+
+      gl->CopySubTextureCHROMIUM(GL_TEXTURE_2D, staging_buffer->texture_id,
+                                 resource_lock->texture_id(), 0, y, 0, y,
+                                 resource->size().width(), rows_to_copy, false,
+                                 false, false);
+      y += rows_to_copy;
+
+      // Increment |bytes_scheduled_since_last_flush_| by the amount of memory
+      // used for this copy operation.
+      bytes_scheduled_since_last_flush_ += rows_to_copy * bytes_per_row;
+
+      if (bytes_scheduled_since_last_flush_ >= max_bytes_per_copy_operation_) {
+        gl->ShallowFlushCHROMIUM();
+        bytes_scheduled_since_last_flush_ = 0;
       }
     }
 
-    // There may be more work available, so wake up another worker thread.
-    copy_operation_count_cv_.Signal();
-
-    // Copy at most |chunk_size_in_rows|.
-    int rows_to_copy = std::min(chunk_size_in_rows, height - y);
-    DCHECK_GT(rows_to_copy, 0);
-
-    // |raster_resource_write_lock| is passed to the first copy operation as it
-    // needs to be released before we can issue a copy.
-    pending_copy_operations_.push_back(make_scoped_ptr(new CopyOperation(
-        raster_resource_write_lock.Pass(), raster_resource, output_resource,
-        gfx::Rect(0, y, raster_resource->size().width(), rows_to_copy))));
-    y += rows_to_copy;
-
-    // Acquire a sequence number for this copy operation.
-    sequence = next_copy_operation_sequence_++;
-
-    // Increment |bytes_scheduled_since_last_flush_| by the amount of memory
-    // used for this copy operation.
-    bytes_scheduled_since_last_flush_ += rows_to_copy * bytes_per_row;
-
-    // Post task that will advance last flushed copy operation to |sequence|
-    // when |bytes_scheduled_since_last_flush_| has reached
-    // |max_bytes_per_copy_operation_|.
-    if (bytes_scheduled_since_last_flush_ >= max_bytes_per_copy_operation_) {
-      task_runner_->PostTask(
-          FROM_HERE,
-          base::Bind(&OneCopyTileTaskWorkerPool::AdvanceLastFlushedCopyTo,
-                     weak_ptr_factory_.GetWeakPtr(), sequence));
-      bytes_scheduled_since_last_flush_ = 0;
+    if (resource_provider_->use_sync_query()) {
+#if defined(OS_CHROMEOS)
+      gl->EndQueryEXT(GL_COMMANDS_ISSUED_CHROMIUM);
+#else
+      gl->EndQueryEXT(GL_COMMANDS_COMPLETED_CHROMIUM);
+#endif
     }
+
+    // Barrier to sync worker context output to cc context.
+    gl->OrderingBarrierCHROMIUM();
   }
 
-  return sequence;
+  staging_buffer->last_usage = base::TimeTicks::Now();
+  busy_buffers_.push_back(staging_buffer.Pass());
+
+  ScheduleReduceMemoryUsage();
 }
 
-void OneCopyTileTaskWorkerPool::AdvanceLastIssuedCopyTo(
-    CopySequenceNumber sequence) {
-  if (last_issued_copy_operation_ >= sequence)
-    return;
+bool OneCopyTileTaskWorkerPool::OnMemoryDump(
+    const base::trace_event::MemoryDumpArgs& args,
+    base::trace_event::ProcessMemoryDump* pmd) {
+  base::AutoLock lock(lock_);
 
-  IssueCopyOperations(sequence - last_issued_copy_operation_);
-  last_issued_copy_operation_ = sequence;
+  for (const auto& buffer : buffers_) {
+    buffer->OnMemoryDump(pmd, resource_provider_->best_texture_format(),
+                         std::find(free_buffers_.begin(), free_buffers_.end(),
+                                   buffer) != free_buffers_.end());
+  }
+
+  return true;
 }
 
-void OneCopyTileTaskWorkerPool::AdvanceLastFlushedCopyTo(
-    CopySequenceNumber sequence) {
-  if (last_flushed_copy_operation_ >= sequence)
-    return;
+scoped_ptr<OneCopyTileTaskWorkerPool::StagingBuffer>
+OneCopyTileTaskWorkerPool::AcquireStagingBuffer(const Resource* resource,
+                                                uint64_t previous_content_id) {
+  lock_.AssertAcquired();
 
-  AdvanceLastIssuedCopyTo(sequence);
+  scoped_ptr<StagingBuffer> staging_buffer;
 
-  // Flush all issued copy operations.
-  context_provider_->ContextGL()->ShallowFlushCHROMIUM();
-  last_flushed_copy_operation_ = last_issued_copy_operation_;
-}
+  ContextProvider* context_provider =
+      resource_provider_->output_surface()->worker_context_provider();
+  DCHECK(context_provider);
 
-void OneCopyTileTaskWorkerPool::OnTaskSetFinished(TaskSet task_set) {
-  TRACE_EVENT1("cc", "OneCopyTileTaskWorkerPool::OnTaskSetFinished", "task_set",
-               task_set);
+  ContextProvider::ScopedContextLock scoped_context(context_provider);
 
-  DCHECK(tasks_pending_[task_set]);
-  tasks_pending_[task_set] = false;
-  if (tasks_pending_.any()) {
-    TRACE_EVENT_ASYNC_STEP_INTO1("cc", "ScheduledTasks", this, "running",
-                                 "state", StateAsValue());
-  } else {
-    TRACE_EVENT_ASYNC_END0("cc", "ScheduledTasks", this);
-  }
-  client_->DidFinishRunningTileTasks(task_set);
-}
+  gpu::gles2::GLES2Interface* gl = scoped_context.ContextGL();
+  DCHECK(gl);
 
-void OneCopyTileTaskWorkerPool::IssueCopyOperations(int64 count) {
-  TRACE_EVENT1("cc", "OneCopyTileTaskWorkerPool::IssueCopyOperations", "count",
-               count);
+  // Check if any busy buffers have become available.
+  if (resource_provider_->use_sync_query()) {
+    while (!busy_buffers_.empty()) {
+      if (!CheckForQueryResult(gl, busy_buffers_.front()->query_id))
+        break;
 
-  CopyOperation::Deque copy_operations;
+      free_buffers_.push_back(busy_buffers_.take_front());
+    }
+  }
 
-  {
-    base::AutoLock lock(lock_);
+  // Wait for number of non-free buffers to become less than the limit.
+  while ((buffers_.size() - free_buffers_.size()) >= max_staging_buffers_) {
+    // Stop when there are no more busy buffers to wait for.
+    if (busy_buffers_.empty())
+      break;
 
-    for (int64 i = 0; i < count; ++i) {
-      DCHECK(!pending_copy_operations_.empty());
-      copy_operations.push_back(pending_copy_operations_.take_front());
+    if (resource_provider_->use_sync_query()) {
+      WaitForQueryResult(gl, busy_buffers_.front()->query_id);
+      free_buffers_.push_back(busy_buffers_.take_front());
+    } else {
+      // Fall-back to glFinish if CHROMIUM_sync_query is not available.
+      gl->Finish();
+      while (!busy_buffers_.empty())
+        free_buffers_.push_back(busy_buffers_.take_front());
     }
+  }
+
+  // Find a staging buffer that allows us to perform partial raster when
+  // using persistent GpuMemoryBuffers.
+  if (use_persistent_gpu_memory_buffers_ && previous_content_id) {
+    StagingBufferDeque::iterator it =
+        std::find_if(free_buffers_.begin(), free_buffers_.end(),
+                     [previous_content_id](const StagingBuffer* buffer) {
+                       return buffer->content_id == previous_content_id;
+                     });
+    if (it != free_buffers_.end())
+      staging_buffer = free_buffers_.take(it);
+  }
 
-    // Increment |issued_copy_operation_count_| to reflect the transition of
-    // copy operations from "pending" to "issued" state.
-    issued_copy_operation_count_ += copy_operations.size();
+  // Find staging buffer of correct size.
+  if (!staging_buffer) {
+    StagingBufferDeque::iterator it =
+        std::find_if(free_buffers_.begin(), free_buffers_.end(),
+                     [resource](const StagingBuffer* buffer) {
+                       return buffer->size == resource->size();
+                     });
+    if (it != free_buffers_.end())
+      staging_buffer = free_buffers_.take(it);
   }
 
-  while (!copy_operations.empty()) {
-    scoped_ptr<CopyOperation> copy_operation = copy_operations.take_front();
+  // Create new staging buffer if necessary.
+  if (!staging_buffer) {
+    staging_buffer = make_scoped_ptr(new StagingBuffer(resource->size()));
+    buffers_.insert(staging_buffer.get());
+  }
 
-    // Remove the write lock.
-    copy_operation->src_write_lock.reset();
+  // Release enough free buffers to stay within the limit.
+  while (buffers_.size() > max_staging_buffers_) {
+    if (free_buffers_.empty())
+      break;
 
-    // Copy contents of source resource to destination resource.
-    resource_provider_->CopyResource(copy_operation->src->id(),
-                                     copy_operation->dst->id(),
-                                     copy_operation->rect);
+    free_buffers_.front()->DestroyGLResources(gl);
+    buffers_.erase(free_buffers_.front());
+    free_buffers_.take_front();
   }
+
+  return staging_buffer.Pass();
 }
 
-void OneCopyTileTaskWorkerPool::
-    ScheduleCheckForCompletedCopyOperationsWithLockAcquired(
-        bool wait_if_needed) {
+base::TimeTicks OneCopyTileTaskWorkerPool::GetUsageTimeForLRUBuffer() {
   lock_.AssertAcquired();
 
-  if (check_for_completed_copy_operations_pending_)
-    return;
+  if (!free_buffers_.empty())
+    return free_buffers_.front()->last_usage;
+
+  if (!busy_buffers_.empty())
+    return busy_buffers_.front()->last_usage;
+
+  return base::TimeTicks();
+}
+
+void OneCopyTileTaskWorkerPool::ScheduleReduceMemoryUsage() {
+  lock_.AssertAcquired();
 
-  base::TimeTicks now = base::TimeTicks::Now();
+  if (reduce_memory_usage_pending_)
+    return;
 
-  // Schedule a check for completed copy operations as soon as possible but
-  // don't allow two consecutive checks to be scheduled to run less than the
-  // tick rate apart.
-  base::TimeTicks next_check_for_completed_copy_operations_time =
-      std::max(last_check_for_completed_copy_operations_time_ +
-                   base::TimeDelta::FromMilliseconds(
-                       kCheckForCompletedCopyOperationsTickRateMs),
-               now);
+  reduce_memory_usage_pending_ = true;
 
+  // Schedule a call to ReduceMemoryUsage at the time when the LRU buffer
+  // should be released.
+  base::TimeTicks reduce_memory_usage_time =
+      GetUsageTimeForLRUBuffer() + staging_buffer_expiration_delay_;
   task_runner_->PostDelayedTask(
-      FROM_HERE,
-      base::Bind(&OneCopyTileTaskWorkerPool::CheckForCompletedCopyOperations,
-                 weak_ptr_factory_.GetWeakPtr(), wait_if_needed),
-      next_check_for_completed_copy_operations_time - now);
-
-  last_check_for_completed_copy_operations_time_ =
-      next_check_for_completed_copy_operations_time;
-  check_for_completed_copy_operations_pending_ = true;
+      FROM_HERE, reduce_memory_usage_callback_,
+      reduce_memory_usage_time - base::TimeTicks::Now());
 }
 
-void OneCopyTileTaskWorkerPool::CheckForCompletedCopyOperations(
-    bool wait_if_needed) {
-  TRACE_EVENT1("cc",
-               "OneCopyTileTaskWorkerPool::CheckForCompletedCopyOperations",
-               "wait_if_needed", wait_if_needed);
+void OneCopyTileTaskWorkerPool::ReduceMemoryUsage() {
+  base::AutoLock lock(lock_);
+
+  reduce_memory_usage_pending_ = false;
+
+  if (free_buffers_.empty() && busy_buffers_.empty())
+    return;
+
+  base::TimeTicks current_time = base::TimeTicks::Now();
+  ReleaseBuffersNotUsedSince(current_time - staging_buffer_expiration_delay_);
 
-  resource_pool_->CheckBusyResources(wait_if_needed);
+  if (free_buffers_.empty() && busy_buffers_.empty())
+    return;
+
+  reduce_memory_usage_pending_ = true;
+
+  // Schedule another call to ReduceMemoryUsage at the time when the next
+  // buffer should be released.
+  base::TimeTicks reduce_memory_usage_time =
+      GetUsageTimeForLRUBuffer() + staging_buffer_expiration_delay_;
+  task_runner_->PostDelayedTask(FROM_HERE, reduce_memory_usage_callback_,
+                                reduce_memory_usage_time - current_time);
+}
+
+void OneCopyTileTaskWorkerPool::ReleaseBuffersNotUsedSince(
+    base::TimeTicks time) {
+  lock_.AssertAcquired();
+
+  ContextProvider* context_provider =
+      resource_provider_->output_surface()->worker_context_provider();
+  DCHECK(context_provider);
 
   {
-    base::AutoLock lock(lock_);
+    ContextProvider::ScopedContextLock scoped_context(context_provider);
+
+    gpu::gles2::GLES2Interface* gl = scoped_context.ContextGL();
+    DCHECK(gl);
 
-    DCHECK(check_for_completed_copy_operations_pending_);
-    check_for_completed_copy_operations_pending_ = false;
+    // Note: Front buffer is guaranteed to be LRU so we can stop releasing
+    // buffers as soon as we find a buffer that has been used since |time|.
+    while (!free_buffers_.empty()) {
+      if (free_buffers_.front()->last_usage > time)
+        return;
 
-    // The number of busy resources in the pool reflects the number of issued
-    // copy operations that have not yet completed.
-    issued_copy_operation_count_ = resource_pool_->busy_resource_count();
+      free_buffers_.front()->DestroyGLResources(gl);
+      buffers_.erase(free_buffers_.front());
+      free_buffers_.take_front();
+    }
+
+    while (!busy_buffers_.empty()) {
+      if (busy_buffers_.front()->last_usage > time)
+        return;
 
-    // There may be work blocked on too many in-flight copy operations, so wake
-    // up a worker thread.
-    copy_operation_count_cv_.Signal();
+      busy_buffers_.front()->DestroyGLResources(gl);
+      buffers_.erase(busy_buffers_.front());
+      busy_buffers_.take_front();
+    }
   }
 }
 
+void OneCopyTileTaskWorkerPool::OnTaskSetFinished(TaskSet task_set) {
+  TRACE_EVENT1("cc", "OneCopyTileTaskWorkerPool::OnTaskSetFinished", "task_set",
+               task_set);
+
+  DCHECK(tasks_pending_[task_set]);
+  tasks_pending_[task_set] = false;
+  if (tasks_pending_.any()) {
+    TRACE_EVENT_ASYNC_STEP_INTO1("cc", "ScheduledTasks", this, "running",
+                                 "state", StateAsValue());
+  } else {
+    TRACE_EVENT_ASYNC_END0("cc", "ScheduledTasks", this);
+  }
+  client_->DidFinishRunningTileTasks(task_set);
+}
+
 scoped_refptr<base::trace_event::ConvertableToTraceFormat>
 OneCopyTileTaskWorkerPool::StateAsValue() const {
   scoped_refptr<base::trace_event::TracedValue> state =
@@ -573,20 +726,14 @@ OneCopyTileTaskWorkerPool::StateAsValue() const {
 
 void OneCopyTileTaskWorkerPool::StagingStateAsValueInto(
     base::trace_event::TracedValue* staging_state) const {
-  staging_state->SetInteger(
-      "staging_resource_count",
-      static_cast<int>(resource_pool_->total_resource_count()));
-  staging_state->SetInteger(
-      "bytes_used_for_staging_resources",
-      static_cast<int>(resource_pool_->total_memory_usage_bytes()));
-  staging_state->SetInteger(
-      "pending_copy_count",
-      static_cast<int>(resource_pool_->total_resource_count() -
-                       resource_pool_->acquired_resource_count()));
-  staging_state->SetInteger(
-      "bytes_pending_copy",
-      static_cast<int>(resource_pool_->total_memory_usage_bytes() -
-                       resource_pool_->acquired_memory_usage_bytes()));
+  base::AutoLock lock(lock_);
+
+  staging_state->SetInteger("staging_buffer_count",
+                            static_cast<int>(buffers_.size()));
+  staging_state->SetInteger("busy_count",
+                            static_cast<int>(busy_buffers_.size()));
+  staging_state->SetInteger("free_count",
+                            static_cast<int>(free_buffers_.size()));
 }
 
 }  // namespace cc
diff --git a/cc/raster/one_copy_tile_task_worker_pool.h b/cc/raster/one_copy_tile_task_worker_pool.h
index 4e050b4..6554fcc 100644
--- a/cc/raster/one_copy_tile_task_worker_pool.h
+++ b/cc/raster/one_copy_tile_task_worker_pool.h
@@ -5,8 +5,12 @@
 #ifndef CC_RASTER_ONE_COPY_TILE_TASK_WORKER_POOL_H_
 #define CC_RASTER_ONE_COPY_TILE_TASK_WORKER_POOL_H_
 
+#include <set>
+
 #include "base/memory/weak_ptr.h"
 #include "base/synchronization/lock.h"
+#include "base/time/time.h"
+#include "base/trace_event/memory_dump_provider.h"
 #include "base/values.h"
 #include "cc/base/scoped_ptr_deque.h"
 #include "cc/output/context_provider.h"
@@ -21,15 +25,20 @@ class TracedValue;
 }
 }
 
+namespace gpu {
+namespace gles2 {
+class GLES2Interface;
+}
+}
+
 namespace cc {
 class ResourcePool;
-class ScopedResource;
-
-typedef int64 CopySequenceNumber;
 
-class CC_EXPORT OneCopyTileTaskWorkerPool : public TileTaskWorkerPool,
-                                            public TileTaskRunner,
-                                            public TileTaskClient {
+class CC_EXPORT OneCopyTileTaskWorkerPool
+    : public TileTaskWorkerPool,
+      public TileTaskRunner,
+      public TileTaskClient,
+      public base::trace_event::MemoryDumpProvider {
  public:
   ~OneCopyTileTaskWorkerPool() override;
 
@@ -38,9 +47,10 @@ class CC_EXPORT OneCopyTileTaskWorkerPool : public TileTaskWorkerPool,
       TaskGraphRunner* task_graph_runner,
       ContextProvider* context_provider,
       ResourceProvider* resource_provider,
-      ResourcePool* resource_pool,
       int max_copy_texture_chromium_size,
-      bool have_persistent_gpu_memory_buffers);
+      bool use_persistent_gpu_memory_buffers,
+      unsigned image_target,
+      int max_staging_buffers);
 
   // Overridden from TileTaskWorkerPool:
   TileTaskRunner* AsTileTaskRunner() override;
@@ -60,61 +70,58 @@ class CC_EXPORT OneCopyTileTaskWorkerPool : public TileTaskWorkerPool,
       uint64_t previous_content_id) override;
   void ReleaseBufferForRaster(scoped_ptr<RasterBuffer> buffer) override;
 
-  // Playback raster source and schedule copy of |raster_resource| resource to
-  // |output_resource|. Returns a non-zero sequence number for this copy
-  // operation.
-  CopySequenceNumber PlaybackAndScheduleCopyOnWorkerThread(
-      bool reusing_raster_resource,
-      scoped_ptr<ResourceProvider::ScopedWriteLockGpuMemoryBuffer>
-          raster_resource_write_lock,
-      const Resource* raster_resource,
-      const Resource* output_resource,
+  // Overridden from base::trace_event::MemoryDumpProvider:
+  bool OnMemoryDump(const base::trace_event::MemoryDumpArgs& args,
+                    base::trace_event::ProcessMemoryDump* pmd) override;
+
+  // Playback raster source and copy result into |resource|.
+  void PlaybackAndCopyOnWorkerThread(
+      const Resource* resource,
+      const ResourceProvider::ScopedWriteLockGL* resource_lock,
       const RasterSource* raster_source,
       const gfx::Rect& raster_full_rect,
       const gfx::Rect& raster_dirty_rect,
       float scale,
-      bool include_images);
-
-  // Issues copy operations until |sequence| has been processed. This will
-  // return immediately if |sequence| has already been processed.
-  void AdvanceLastIssuedCopyTo(CopySequenceNumber sequence);
-
-  bool have_persistent_gpu_memory_buffers() const {
-    return have_persistent_gpu_memory_buffers_;
-  }
+      bool include_images,
+      uint64_t resource_content_id,
+      uint64_t previous_content_id);
 
  protected:
   OneCopyTileTaskWorkerPool(base::SequencedTaskRunner* task_runner,
                             TaskGraphRunner* task_graph_runner,
-                            ContextProvider* context_provider,
                             ResourceProvider* resource_provider,
-                            ResourcePool* resource_pool,
                             int max_copy_texture_chromium_size,
-                            bool have_persistent_gpu_memory_buffers);
+                            bool use_persistent_gpu_memory_buffers,
+                            unsigned image_target,
+                            int max_staging_buffers);
 
  private:
-  struct CopyOperation {
-    typedef ScopedPtrDeque<CopyOperation> Deque;
-
-    CopyOperation(scoped_ptr<ResourceProvider::ScopedWriteLockGpuMemoryBuffer>
-                      src_write_lock,
-                  const Resource* src,
-                  const Resource* dst,
-                  const gfx::Rect& rect);
-    ~CopyOperation();
-
-    scoped_ptr<ResourceProvider::ScopedWriteLockGpuMemoryBuffer> src_write_lock;
-    const Resource* src;
-    const Resource* dst;
-    const gfx::Rect rect;
+  struct StagingBuffer {
+    explicit StagingBuffer(const gfx::Size& size);
+    ~StagingBuffer();
+
+    void DestroyGLResources(gpu::gles2::GLES2Interface* gl);
+    void OnMemoryDump(base::trace_event::ProcessMemoryDump* pmd,
+                      ResourceFormat format,
+                      bool is_free) const;
+
+    const gfx::Size size;
+    scoped_ptr<gfx::GpuMemoryBuffer> gpu_memory_buffer;
+    base::TimeTicks last_usage;
+    unsigned texture_id;
+    unsigned image_id;
+    unsigned query_id;
+    uint64_t content_id;
   };
 
+  scoped_ptr<StagingBuffer> AcquireStagingBuffer(const Resource* resource,
+                                                 uint64_t previous_content_id);
+  base::TimeTicks GetUsageTimeForLRUBuffer();
+  void ScheduleReduceMemoryUsage();
+  void ReduceMemoryUsage();
+  void ReleaseBuffersNotUsedSince(base::TimeTicks time);
+
   void OnTaskSetFinished(TaskSet task_set);
-  void AdvanceLastFlushedCopyTo(CopySequenceNumber sequence);
-  void IssueCopyOperations(int64 count);
-  void ScheduleCheckForCompletedCopyOperationsWithLockAcquired(
-      bool wait_if_needed);
-  void CheckForCompletedCopyOperations(bool wait_if_needed);
   scoped_refptr<base::trace_event::ConvertableToTraceFormat> StateAsValue()
       const;
   void StagingStateAsValueInto(
@@ -124,31 +131,30 @@ class CC_EXPORT OneCopyTileTaskWorkerPool : public TileTaskWorkerPool,
   TaskGraphRunner* task_graph_runner_;
   const NamespaceToken namespace_token_;
   TileTaskRunnerClient* client_;
-  ContextProvider* context_provider_;
-  ResourceProvider* resource_provider_;
-  ResourcePool* resource_pool_;
+  ResourceProvider* const resource_provider_;
   const int max_bytes_per_copy_operation_;
-  const bool have_persistent_gpu_memory_buffers_;
+  const bool use_persistent_gpu_memory_buffers_;
+  const unsigned image_target_;
   TaskSetCollection tasks_pending_;
   scoped_refptr<TileTask> task_set_finished_tasks_[kNumberOfTaskSets];
-  CopySequenceNumber last_issued_copy_operation_;
-  CopySequenceNumber last_flushed_copy_operation_;
 
   // Task graph used when scheduling tasks and vector used to gather
   // completed tasks.
   TaskGraph graph_;
   Task::Vector completed_tasks_;
 
-  base::Lock lock_;
+  mutable base::Lock lock_;
   // |lock_| must be acquired when accessing the following members.
-  base::ConditionVariable copy_operation_count_cv_;
+  using StagingBufferSet = std::set<StagingBuffer*>;
+  StagingBufferSet buffers_;
+  using StagingBufferDeque = ScopedPtrDeque<StagingBuffer>;
+  StagingBufferDeque free_buffers_;
+  StagingBufferDeque busy_buffers_;
   int bytes_scheduled_since_last_flush_;
-  size_t issued_copy_operation_count_;
-  CopyOperation::Deque pending_copy_operations_;
-  CopySequenceNumber next_copy_operation_sequence_;
-  bool check_for_completed_copy_operations_pending_;
-  base::TimeTicks last_check_for_completed_copy_operations_time_;
-  bool shutdown_;
+  size_t max_staging_buffers_;
+  const base::TimeDelta staging_buffer_expiration_delay_;
+  bool reduce_memory_usage_pending_;
+  base::Closure reduce_memory_usage_callback_;
 
   base::WeakPtrFactory<OneCopyTileTaskWorkerPool> weak_ptr_factory_;
   // "raster finished" tasks need their own factory as they need to be
diff --git a/cc/raster/tile_task_worker_pool_perftest.cc b/cc/raster/tile_task_worker_pool_perftest.cc
index b49c498..88322c1 100644
--- a/cc/raster/tile_task_worker_pool_perftest.cc
+++ b/cc/raster/tile_task_worker_pool_perftest.cc
@@ -265,13 +265,11 @@ class TileTaskWorkerPoolPerfTest
         break;
       case TILE_TASK_WORKER_POOL_TYPE_ONE_COPY:
         Create3dOutputSurfaceAndResourceProvider();
-        staging_resource_pool_ = ResourcePool::Create(resource_provider_.get(),
-                                                      GL_TEXTURE_2D);
         tile_task_worker_pool_ = OneCopyTileTaskWorkerPool::Create(
             task_runner_.get(), task_graph_runner_.get(),
             context_provider_.get(), resource_provider_.get(),
-            staging_resource_pool_.get(), std::numeric_limits<int>::max(),
-            false);
+            std::numeric_limits<int>::max(), false, GL_TEXTURE_2D,
+            std::numeric_limits<int>::max());
         break;
       case TILE_TASK_WORKER_POOL_TYPE_GPU:
         Create3dOutputSurfaceAndResourceProvider();
@@ -431,7 +429,6 @@ class TileTaskWorkerPoolPerfTest
     return std::string();
   }
 
-  scoped_ptr<ResourcePool> staging_resource_pool_;
   scoped_ptr<TileTaskWorkerPool> tile_task_worker_pool_;
   TestGpuMemoryBufferManager gpu_memory_buffer_manager_;
   TestSharedBitmapManager shared_bitmap_manager_;
diff --git a/cc/raster/tile_task_worker_pool_unittest.cc b/cc/raster/tile_task_worker_pool_unittest.cc
index 6038b00..5e2d1af 100644
--- a/cc/raster/tile_task_worker_pool_unittest.cc
+++ b/cc/raster/tile_task_worker_pool_unittest.cc
@@ -41,6 +41,7 @@ namespace {
 
 const size_t kMaxTransferBufferUsageBytes = 10000U;
 const size_t kMaxBytesPerCopyOperation = 1000U;
+const size_t kMaxStagingBuffers = 32U;
 
 // A resource of this dimension^2 * 4 must be greater than the above transfer
 // buffer constant.
@@ -165,12 +166,11 @@ class TileTaskWorkerPoolTest
         break;
       case TILE_TASK_WORKER_POOL_TYPE_ONE_COPY:
         Create3dOutputSurfaceAndResourceProvider();
-        staging_resource_pool_ = ResourcePool::Create(resource_provider_.get(),
-                                                      GL_TEXTURE_2D);
         tile_task_worker_pool_ = OneCopyTileTaskWorkerPool::Create(
             base::ThreadTaskRunnerHandle::Get().get(), &task_graph_runner_,
             context_provider_.get(), resource_provider_.get(),
-            staging_resource_pool_.get(), kMaxBytesPerCopyOperation, false);
+            kMaxBytesPerCopyOperation, false, GL_TEXTURE_2D,
+            kMaxStagingBuffers);
         break;
       case TILE_TASK_WORKER_POOL_TYPE_GPU:
         Create3dOutputSurfaceAndResourceProvider();
@@ -332,7 +332,6 @@ class TileTaskWorkerPoolTest
   FakeOutputSurfaceClient output_surface_client_;
   scoped_ptr<FakeOutputSurface> output_surface_;
   scoped_ptr<ResourceProvider> resource_provider_;
-  scoped_ptr<ResourcePool> staging_resource_pool_;
   scoped_ptr<TileTaskWorkerPool> tile_task_worker_pool_;
   TestGpuMemoryBufferManager gpu_memory_buffer_manager_;
   TestSharedBitmapManager shared_bitmap_manager_;
diff --git a/cc/resources/resource_format.cc b/cc/resources/resource_format.cc
index ce9fe7d..c62ae93 100644
--- a/cc/resources/resource_format.cc
+++ b/cc/resources/resource_format.cc
@@ -87,4 +87,23 @@ GLenum GLInternalFormat(ResourceFormat format) {
   return GLDataFormat(format);
 }
 
+gfx::BufferFormat BufferFormat(ResourceFormat format) {
+  switch (format) {
+    case RGBA_8888:
+      return gfx::BufferFormat::RGBA_8888;
+    case BGRA_8888:
+      return gfx::BufferFormat::BGRA_8888;
+    case RGBA_4444:
+      return gfx::BufferFormat::RGBA_4444;
+    case ALPHA_8:
+    case LUMINANCE_8:
+    case RGB_565:
+    case ETC1:
+    case RED_8:
+      break;
+  }
+  NOTREACHED();
+  return gfx::BufferFormat::RGBA_8888;
+}
+
 }  // namespace cc
diff --git a/cc/resources/resource_format.h b/cc/resources/resource_format.h
index d910bb0..79b815f 100644
--- a/cc/resources/resource_format.h
+++ b/cc/resources/resource_format.h
@@ -8,6 +8,7 @@
 #include "base/logging.h"
 #include "cc/base/cc_export.h"
 #include "third_party/skia/include/core/SkBitmap.h"
+#include "ui/gfx/gpu_memory_buffer.h"
 
 // TODO(prashant.n): Including third_party/khronos/GLES2/gl2.h causes
 // redefinition errors as macros/functions defined in it conflict with
@@ -35,6 +36,7 @@ CC_EXPORT int BitsPerPixel(ResourceFormat format);
 CC_EXPORT GLenum GLDataType(ResourceFormat format);
 CC_EXPORT GLenum GLDataFormat(ResourceFormat format);
 CC_EXPORT GLenum GLInternalFormat(ResourceFormat format);
+CC_EXPORT gfx::BufferFormat BufferFormat(ResourceFormat format);
 
 }  // namespace cc
 
diff --git a/cc/resources/resource_pool.cc b/cc/resources/resource_pool.cc
index f3e2090..c09fa14 100644
--- a/cc/resources/resource_pool.cc
+++ b/cc/resources/resource_pool.cc
@@ -172,15 +172,12 @@ void ResourcePool::DeleteResource(ScopedResource* resource) {
   delete resource;
 }
 
-void ResourcePool::CheckBusyResources(bool wait_if_needed) {
+void ResourcePool::CheckBusyResources() {
   ResourceList::iterator it = busy_resources_.begin();
 
   while (it != busy_resources_.end()) {
     ScopedResource* resource = it->resource;
 
-    if (wait_if_needed)
-      resource_provider_->WaitReadLockIfNeeded(resource->id());
-
     if (resource_provider_->CanLockForWrite(resource->id())) {
       DidFinishUsingResource(resource, it->content_id);
       it = busy_resources_.erase(it);
diff --git a/cc/resources/resource_pool.h b/cc/resources/resource_pool.h
index efd1e01..6ea9329c 100644
--- a/cc/resources/resource_pool.h
+++ b/cc/resources/resource_pool.h
@@ -37,10 +37,7 @@ class CC_EXPORT ResourcePool : public base::trace_event::MemoryDumpProvider {
                               size_t max_resource_count);
 
   void ReduceResourceUsage();
-  // This might block if |wait_if_needed| is true and one of the currently
-  // busy resources has a read lock fence that needs to be waited upon before
-  // it can be locked for write again.
-  void CheckBusyResources(bool wait_if_needed);
+  void CheckBusyResources();
 
   size_t total_memory_usage_bytes() const { return memory_usage_bytes_; }
   size_t acquired_memory_usage_bytes() const {
diff --git a/cc/resources/resource_pool_unittest.cc b/cc/resources/resource_pool_unittest.cc
index 035b19e..f280f49 100644
--- a/cc/resources/resource_pool_unittest.cc
+++ b/cc/resources/resource_pool_unittest.cc
@@ -71,8 +71,7 @@ TEST_F(ResourcePoolTest, AccountingSingleResource) {
   EXPECT_EQ(1u, resource_pool_->total_resource_count());
   EXPECT_EQ(1u, resource_pool_->busy_resource_count());
 
-  bool wait_if_needed = false;
-  resource_pool_->CheckBusyResources(wait_if_needed);
+  resource_pool_->CheckBusyResources();
   EXPECT_EQ(resource_bytes, resource_pool_->total_memory_usage_bytes());
   EXPECT_EQ(0u, resource_pool_->acquired_memory_usage_bytes());
   EXPECT_EQ(1u, resource_pool_->total_resource_count());
@@ -96,26 +95,25 @@ TEST_F(ResourcePoolTest, SimpleResourceReuse) {
 
   gfx::Size size(100, 100);
   ResourceFormat format = RGBA_8888;
-  bool wait_if_needed = false;
 
   scoped_ptr<ScopedResource> resource =
       resource_pool_->AcquireResource(size, format);
   resource_pool_->ReleaseResource(resource.Pass(), 0u);
-  resource_pool_->CheckBusyResources(wait_if_needed);
+  resource_pool_->CheckBusyResources();
   EXPECT_EQ(1u, resource_provider_->num_resources());
 
   // Same size/format should re-use resource.
   resource = resource_pool_->AcquireResource(size, format);
   EXPECT_EQ(1u, resource_provider_->num_resources());
   resource_pool_->ReleaseResource(resource.Pass(), 0u);
-  resource_pool_->CheckBusyResources(wait_if_needed);
+  resource_pool_->CheckBusyResources();
   EXPECT_EQ(1u, resource_provider_->num_resources());
 
   // Different size/format should alloate new resource.
   resource = resource_pool_->AcquireResource(gfx::Size(50, 50), LUMINANCE_8);
   EXPECT_EQ(2u, resource_provider_->num_resources());
   resource_pool_->ReleaseResource(resource.Pass(), 0u);
-  resource_pool_->CheckBusyResources(wait_if_needed);
+  resource_pool_->CheckBusyResources();
   EXPECT_EQ(2u, resource_provider_->num_resources());
 }
 
@@ -127,7 +125,6 @@ TEST_F(ResourcePoolTest, LostResource) {
 
   gfx::Size size(100, 100);
   ResourceFormat format = RGBA_8888;
-  bool wait_if_needed = false;
 
   scoped_ptr<ScopedResource> resource =
       resource_pool_->AcquireResource(size, format);
@@ -135,7 +132,7 @@ TEST_F(ResourcePoolTest, LostResource) {
 
   resource_provider_->LoseResourceForTesting(resource->id());
   resource_pool_->ReleaseResource(resource.Pass(), 0u);
-  resource_pool_->CheckBusyResources(wait_if_needed);
+  resource_pool_->CheckBusyResources();
   EXPECT_EQ(0u, resource_provider_->num_resources());
 }
 
diff --git a/cc/resources/resource_provider.cc b/cc/resources/resource_provider.cc
index 204422c..d09901e 100644
--- a/cc/resources/resource_provider.cc
+++ b/cc/resources/resource_provider.cc
@@ -32,7 +32,6 @@
 #include "third_party/skia/include/gpu/GrTextureProvider.h"
 #include "ui/gfx/geometry/rect.h"
 #include "ui/gfx/geometry/vector2d.h"
-#include "ui/gfx/gpu_memory_buffer.h"
 #include "ui/gl/trace_util.h"
 
 using gpu::gles2::GLES2Interface;
@@ -117,25 +116,6 @@ GrPixelConfig ToGrPixelConfig(ResourceFormat format) {
   return kSkia8888_GrPixelConfig;
 }
 
-gfx::BufferFormat ToGpuMemoryBufferFormat(ResourceFormat format) {
-  switch (format) {
-    case RGBA_8888:
-      return gfx::BufferFormat::RGBA_8888;
-    case BGRA_8888:
-      return gfx::BufferFormat::BGRA_8888;
-    case RGBA_4444:
-      return gfx::BufferFormat::RGBA_4444;
-    case ALPHA_8:
-    case LUMINANCE_8:
-    case RGB_565:
-    case ETC1:
-    case RED_8:
-      break;
-  }
-  NOTREACHED();
-  return gfx::BufferFormat::RGBA_8888;
-}
-
 class ScopedSetActiveTexture {
  public:
   ScopedSetActiveTexture(GLES2Interface* gl, GLenum unit)
@@ -206,46 +186,6 @@ class BufferIdAllocator : public IdAllocator {
   DISALLOW_COPY_AND_ASSIGN(BufferIdAllocator);
 };
 
-// Query object based fence implementation used to detect completion of copy
-// texture operations. Fence has passed when query result is available.
-class CopyTextureFence : public ResourceProvider::Fence {
- public:
-  CopyTextureFence(gpu::gles2::GLES2Interface* gl, unsigned query_id)
-      : gl_(gl), query_id_(query_id) {}
-
-  // Overridden from ResourceProvider::Fence:
-  void Set() override {}
-  bool HasPassed() override {
-    unsigned available = 1;
-    gl_->GetQueryObjectuivEXT(
-        query_id_, GL_QUERY_RESULT_AVAILABLE_EXT, &available);
-    if (!available)
-      return false;
-
-    ProcessResult();
-    return true;
-  }
-  void Wait() override {
-    // ProcessResult() will wait for result to become available.
-    ProcessResult();
-  }
-
- private:
-  ~CopyTextureFence() override {}
-
-  void ProcessResult() {
-    unsigned time_elapsed_us = 0;
-    gl_->GetQueryObjectuivEXT(query_id_, GL_QUERY_RESULT_EXT, &time_elapsed_us);
-    UMA_HISTOGRAM_CUSTOM_COUNTS("Renderer4.CopyTextureLatency", time_elapsed_us,
-                                0, 256000, 50);
-  }
-
-  gpu::gles2::GLES2Interface* gl_;
-  unsigned query_id_;
-
-  DISALLOW_COPY_AND_ASSIGN(CopyTextureFence);
-};
-
 }  // namespace
 
 ResourceProvider::Resource::~Resource() {}
@@ -396,13 +336,11 @@ scoped_ptr<ResourceProvider> ResourceProvider::Create(
     BlockingTaskRunner* blocking_main_thread_task_runner,
     int highp_threshold_min,
     bool use_rgba_4444_texture_format,
-    size_t id_allocation_chunk_size,
-    bool use_persistent_map_for_gpu_memory_buffers) {
+    size_t id_allocation_chunk_size) {
   scoped_ptr<ResourceProvider> resource_provider(new ResourceProvider(
       output_surface, shared_bitmap_manager, gpu_memory_buffer_manager,
       blocking_main_thread_task_runner, highp_threshold_min,
-      use_rgba_4444_texture_format, id_allocation_chunk_size,
-      use_persistent_map_for_gpu_memory_buffers));
+      use_rgba_4444_texture_format, id_allocation_chunk_size));
   resource_provider->Initialize();
   return resource_provider;
 }
@@ -993,13 +931,9 @@ gfx::GpuMemoryBuffer*
 ResourceProvider::ScopedWriteLockGpuMemoryBuffer::GetGpuMemoryBuffer() {
   if (gpu_memory_buffer_)
     return gpu_memory_buffer_;
-  gfx::BufferUsage usage =
-      resource_provider_->use_persistent_map_for_gpu_memory_buffers()
-          ? gfx::BufferUsage::PERSISTENT_MAP
-          : gfx::BufferUsage::MAP;
   scoped_ptr<gfx::GpuMemoryBuffer> gpu_memory_buffer =
       gpu_memory_buffer_manager_->AllocateGpuMemoryBuffer(
-          size_, ToGpuMemoryBufferFormat(format_), usage);
+          size_, BufferFormat(format_), gfx::BufferUsage::MAP);
   gpu_memory_buffer_ = gpu_memory_buffer.release();
   return gpu_memory_buffer_;
 }
@@ -1090,8 +1024,7 @@ ResourceProvider::ResourceProvider(
     BlockingTaskRunner* blocking_main_thread_task_runner,
     int highp_threshold_min,
     bool use_rgba_4444_texture_format,
-    size_t id_allocation_chunk_size,
-    bool use_persistent_map_for_gpu_memory_buffers)
+    size_t id_allocation_chunk_size)
     : output_surface_(output_surface),
       shared_bitmap_manager_(shared_bitmap_manager),
       gpu_memory_buffer_manager_(gpu_memory_buffer_manager),
@@ -1110,10 +1043,7 @@ ResourceProvider::ResourceProvider(
       best_texture_format_(RGBA_8888),
       best_render_buffer_format_(RGBA_8888),
       use_rgba_4444_texture_format_(use_rgba_4444_texture_format),
-      id_allocation_chunk_size_(id_allocation_chunk_size),
-      use_sync_query_(false),
-      use_persistent_map_for_gpu_memory_buffers_(
-          use_persistent_map_for_gpu_memory_buffers) {
+      id_allocation_chunk_size_(id_allocation_chunk_size) {
   DCHECK(output_surface_->HasClient());
   DCHECK(id_allocation_chunk_size_);
 }
@@ -1847,79 +1777,6 @@ void ResourceProvider::BindImageForSampling(Resource* resource) {
   resource->dirty_image = false;
 }
 
-void ResourceProvider::CopyResource(ResourceId source_id,
-                                    ResourceId dest_id,
-                                    const gfx::Rect& rect) {
-  TRACE_EVENT0("cc", "ResourceProvider::CopyResource");
-
-  Resource* source_resource = GetResource(source_id);
-  DCHECK(!source_resource->lock_for_read_count);
-  DCHECK(source_resource->origin == Resource::INTERNAL);
-  DCHECK_EQ(source_resource->exported_count, 0);
-  DCHECK_EQ(RESOURCE_TYPE_GL_TEXTURE, source_resource->type);
-  LazyAllocate(source_resource);
-
-  Resource* dest_resource = GetResource(dest_id);
-  DCHECK(!dest_resource->locked_for_write);
-  DCHECK(!dest_resource->lock_for_read_count);
-  DCHECK(dest_resource->origin == Resource::INTERNAL);
-  DCHECK_EQ(dest_resource->exported_count, 0);
-  DCHECK_EQ(RESOURCE_TYPE_GL_TEXTURE, dest_resource->type);
-  LazyAllocate(dest_resource);
-
-  DCHECK_EQ(source_resource->type, dest_resource->type);
-  DCHECK_EQ(source_resource->format, dest_resource->format);
-  DCHECK(source_resource->size == dest_resource->size);
-  DCHECK(gfx::Rect(dest_resource->size).Contains(rect));
-
-  GLES2Interface* gl = ContextGL();
-  DCHECK(gl);
-  if (source_resource->image_id && source_resource->dirty_image) {
-    gl->BindTexture(source_resource->target, source_resource->gl_id);
-    BindImageForSampling(source_resource);
-  }
-  if (use_sync_query_) {
-    if (!source_resource->gl_read_lock_query_id)
-      gl->GenQueriesEXT(1, &source_resource->gl_read_lock_query_id);
-#if defined(OS_CHROMEOS)
-    // TODO(reveman): This avoids a performance problem on some ChromeOS
-    // devices. This needs to be removed to support native GpuMemoryBuffer
-    // implementations. crbug.com/436314
-    gl->BeginQueryEXT(GL_COMMANDS_ISSUED_CHROMIUM,
-                      source_resource->gl_read_lock_query_id);
-#else
-    gl->BeginQueryEXT(GL_COMMANDS_COMPLETED_CHROMIUM,
-                      source_resource->gl_read_lock_query_id);
-#endif
-  }
-  DCHECK(!dest_resource->image_id);
-  dest_resource->allocated = true;
-  gl->CopySubTextureCHROMIUM(dest_resource->target, source_resource->gl_id,
-                             dest_resource->gl_id, rect.x(), rect.y(), rect.x(),
-                             rect.y(), rect.width(), rect.height(),
-                             false, false, false);
-  if (source_resource->gl_read_lock_query_id) {
-    // End query and create a read lock fence that will prevent access to
-// source resource until CopySubTextureCHROMIUM command has completed.
-#if defined(OS_CHROMEOS)
-    gl->EndQueryEXT(GL_COMMANDS_ISSUED_CHROMIUM);
-#else
-    gl->EndQueryEXT(GL_COMMANDS_COMPLETED_CHROMIUM);
-#endif
-    source_resource->read_lock_fence = make_scoped_refptr(
-        new CopyTextureFence(gl, source_resource->gl_read_lock_query_id));
-  } else {
-    // Create a SynchronousFence when CHROMIUM_sync_query extension is missing.
-    // Try to use one synchronous fence for as many CopyResource operations as
-    // possible as that reduce the number of times we have to synchronize with
-    // the GL.
-    if (!synchronous_fence_.get() || synchronous_fence_->has_synchronized())
-      synchronous_fence_ = make_scoped_refptr(new SynchronousFence(gl));
-    source_resource->read_lock_fence = synchronous_fence_;
-    source_resource->read_lock_fence->Set();
-  }
-}
-
 void ResourceProvider::WaitSyncPointIfNeeded(ResourceId id) {
   Resource* resource = GetResource(id);
   DCHECK_EQ(resource->exported_count, 0);
@@ -1935,15 +1792,6 @@ void ResourceProvider::WaitSyncPointIfNeeded(ResourceId id) {
   resource->mailbox.set_sync_point(0);
 }
 
-void ResourceProvider::WaitReadLockIfNeeded(ResourceId id) {
-  Resource* resource = GetResource(id);
-  DCHECK_EQ(resource->exported_count, 0);
-  if (!resource->read_lock_fence.get())
-    return;
-
-  resource->read_lock_fence->Wait();
-}
-
 GLint ResourceProvider::GetActiveTextureUnit(GLES2Interface* gl) {
   GLint active_unit = 0;
   gl->GetIntegerv(GL_ACTIVE_TEXTURE, &active_unit);
diff --git a/cc/resources/resource_provider.h b/cc/resources/resource_provider.h
index bee3ce1..2a778d3 100644
--- a/cc/resources/resource_provider.h
+++ b/cc/resources/resource_provider.h
@@ -35,6 +35,7 @@
 #include "third_party/skia/include/core/SkBitmap.h"
 #include "third_party/skia/include/core/SkCanvas.h"
 #include "ui/gfx/geometry/size.h"
+#include "ui/gfx/gpu_memory_buffer.h"
 
 class GrContext;
 
@@ -46,7 +47,6 @@ class GLES2Interface;
 }
 
 namespace gfx {
-class GpuMemoryBuffer;
 class Rect;
 class Vector2d;
 }
@@ -87,8 +87,7 @@ class CC_EXPORT ResourceProvider
       BlockingTaskRunner* blocking_main_thread_task_runner,
       int highp_threshold_min,
       bool use_rgba_4444_texture_format,
-      size_t id_allocation_chunk_size,
-      bool use_persistent_map_for_gpu_memory_buffers);
+      size_t id_allocation_chunk_size);
   ~ResourceProvider() override;
 
   void DidLoseOutputSurface() { lost_output_surface_ = true; }
@@ -103,8 +102,8 @@ class CC_EXPORT ResourceProvider
   }
   ResourceFormat yuv_resource_format() const { return yuv_resource_format_; }
   bool use_sync_query() const { return use_sync_query_; }
-  bool use_persistent_map_for_gpu_memory_buffers() const {
-    return use_persistent_map_for_gpu_memory_buffers_;
+  gpu::GpuMemoryBufferManager* gpu_memory_buffer_manager() {
+    return gpu_memory_buffer_manager_;
   }
   size_t num_resources() const { return resources_.size(); }
 
@@ -426,15 +425,8 @@ class CC_EXPORT ResourceProvider
   // Indicates if we can currently lock this resource for write.
   bool CanLockForWrite(ResourceId id);
 
-  // Copy |rect| pixels from source to destination.
-  void CopyResource(ResourceId source_id,
-                    ResourceId dest_id,
-                    const gfx::Rect& rect);
-
   void WaitSyncPointIfNeeded(ResourceId id);
 
-  void WaitReadLockIfNeeded(ResourceId id);
-
   static GLint GetActiveTextureUnit(gpu::gles2::GLES2Interface* gl);
 
   OutputSurface* output_surface() { return output_surface_; }
@@ -452,8 +444,7 @@ class CC_EXPORT ResourceProvider
                    BlockingTaskRunner* blocking_main_thread_task_runner,
                    int highp_threshold_min,
                    bool use_rgba_4444_texture_format,
-                   size_t id_allocation_chunk_size,
-                   bool use_persistent_map_for_gpu_memory_buffers);
+                   size_t id_allocation_chunk_size);
   void Initialize();
 
  private:
@@ -608,9 +599,6 @@ class CC_EXPORT ResourceProvider
   scoped_ptr<IdAllocator> buffer_id_allocator_;
 
   bool use_sync_query_;
-  bool use_persistent_map_for_gpu_memory_buffers_;
-  // Fence used for CopyResource if CHROMIUM_sync_query is not supported.
-  scoped_refptr<SynchronousFence> synchronous_fence_;
 
   DISALLOW_COPY_AND_ASSIGN(ResourceProvider);
 };
diff --git a/cc/resources/resource_provider_unittest.cc b/cc/resources/resource_provider_unittest.cc
index 1368558..3d5adc0 100644
--- a/cc/resources/resource_provider_unittest.cc
+++ b/cc/resources/resource_provider_unittest.cc
@@ -421,11 +421,11 @@ class ResourceProviderTest
     resource_provider_ = ResourceProvider::Create(
         output_surface_.get(), shared_bitmap_manager_.get(),
         gpu_memory_buffer_manager_.get(), main_thread_task_runner_.get(), 0,
-        false, 1, false);
+        false, 1);
     child_resource_provider_ = ResourceProvider::Create(
         child_output_surface_.get(), shared_bitmap_manager_.get(),
         gpu_memory_buffer_manager_.get(), main_thread_task_runner_.get(), 0,
-        false, 1, false);
+        false, 1);
   }
 
   ResourceProviderTest() : ResourceProviderTest(true) {}
@@ -1367,7 +1367,7 @@ TEST_P(ResourceProviderTest, TransferGLToSoftware) {
 
   scoped_ptr<ResourceProvider> child_resource_provider(ResourceProvider::Create(
       child_output_surface.get(), shared_bitmap_manager_.get(),
-      gpu_memory_buffer_manager_.get(), NULL, 0, false, 1, false));
+      gpu_memory_buffer_manager_.get(), NULL, 0, false, 1));
 
   gfx::Size size(1, 1);
   ResourceFormat format = RGBA_8888;
@@ -1851,7 +1851,7 @@ class ResourceProviderTestTextureFilters : public ResourceProviderTest {
     scoped_ptr<ResourceProvider> child_resource_provider(
         ResourceProvider::Create(child_output_surface.get(),
                                  shared_bitmap_manager.get(), NULL, NULL, 0,
-                                 false, 1, false));
+                                 false, 1));
 
     scoped_ptr<TextureStateTrackingContext> parent_context_owned(
         new TextureStateTrackingContext);
@@ -1865,7 +1865,7 @@ class ResourceProviderTestTextureFilters : public ResourceProviderTest {
     scoped_ptr<ResourceProvider> parent_resource_provider(
         ResourceProvider::Create(parent_output_surface.get(),
                                  shared_bitmap_manager.get(), NULL, NULL, 0,
-                                 false, 1, false));
+                                 false, 1));
 
     gfx::Size size(1, 1);
     ResourceFormat format = RGBA_8888;
@@ -2497,7 +2497,7 @@ TEST_P(ResourceProviderTest, ScopedSampler) {
 
   scoped_ptr<ResourceProvider> resource_provider(ResourceProvider::Create(
       output_surface.get(), shared_bitmap_manager_.get(),
-      gpu_memory_buffer_manager_.get(), NULL, 0, false, 1, false));
+      gpu_memory_buffer_manager_.get(), NULL, 0, false, 1));
 
   gfx::Size size(1, 1);
   ResourceFormat format = RGBA_8888;
@@ -2579,7 +2579,7 @@ TEST_P(ResourceProviderTest, ManagedResource) {
 
   scoped_ptr<ResourceProvider> resource_provider(ResourceProvider::Create(
       output_surface.get(), shared_bitmap_manager_.get(),
-      gpu_memory_buffer_manager_.get(), NULL, 0, false, 1, false));
+      gpu_memory_buffer_manager_.get(), NULL, 0, false, 1));
 
   gfx::Size size(1, 1);
   ResourceFormat format = RGBA_8888;
@@ -2626,7 +2626,7 @@ TEST_P(ResourceProviderTest, TextureWrapMode) {
 
   scoped_ptr<ResourceProvider> resource_provider(ResourceProvider::Create(
       output_surface.get(), shared_bitmap_manager_.get(),
-      gpu_memory_buffer_manager_.get(), NULL, 0, false, 1, false));
+      gpu_memory_buffer_manager_.get(), NULL, 0, false, 1));
 
   gfx::Size size(1, 1);
   ResourceFormat format = RGBA_8888;
@@ -2676,7 +2676,7 @@ TEST_P(ResourceProviderTest, TextureHint) {
 
   scoped_ptr<ResourceProvider> resource_provider(ResourceProvider::Create(
       output_surface.get(), shared_bitmap_manager_.get(),
-      gpu_memory_buffer_manager_.get(), NULL, 0, false, 1, false));
+      gpu_memory_buffer_manager_.get(), NULL, 0, false, 1));
 
   gfx::Size size(1, 1);
   ResourceFormat format = RGBA_8888;
@@ -2741,7 +2741,7 @@ TEST_P(ResourceProviderTest, TextureMailbox_SharedMemory) {
   scoped_ptr<ResourceProvider> resource_provider(ResourceProvider::Create(
       output_surface.get(), shared_bitmap_manager_.get(),
       gpu_memory_buffer_manager_.get(), main_thread_task_runner_.get(), 0,
-      false, 1, false));
+      false, 1));
 
   uint32 release_sync_point = 0;
   bool lost_resource = false;
@@ -2790,7 +2790,7 @@ class ResourceProviderTestTextureMailboxGLFilters
 
     scoped_ptr<ResourceProvider> resource_provider(ResourceProvider::Create(
         output_surface.get(), shared_bitmap_manager, gpu_memory_buffer_manager,
-        main_thread_task_runner, 0, false, 1, false));
+        main_thread_task_runner, 0, false, 1));
 
     unsigned texture_id = 1;
     uint32 sync_point = 30;
@@ -2933,7 +2933,7 @@ TEST_P(ResourceProviderTest, TextureMailbox_GLTextureExternalOES) {
 
   scoped_ptr<ResourceProvider> resource_provider(ResourceProvider::Create(
       output_surface.get(), shared_bitmap_manager_.get(),
-      gpu_memory_buffer_manager_.get(), NULL, 0, false, 1, false));
+      gpu_memory_buffer_manager_.get(), NULL, 0, false, 1));
 
   uint32 sync_point = 30;
   unsigned target = GL_TEXTURE_EXTERNAL_OES;
@@ -3002,7 +3002,7 @@ TEST_P(ResourceProviderTest,
 
   scoped_ptr<ResourceProvider> resource_provider(ResourceProvider::Create(
       output_surface.get(), shared_bitmap_manager_.get(),
-      gpu_memory_buffer_manager_.get(), NULL, 0, false, 1, false));
+      gpu_memory_buffer_manager_.get(), NULL, 0, false, 1));
 
   uint32 sync_point = 30;
   unsigned target = GL_TEXTURE_2D;
@@ -3055,7 +3055,7 @@ TEST_P(ResourceProviderTest, TextureMailbox_WaitSyncPointIfNeeded_NoSyncPoint) {
 
   scoped_ptr<ResourceProvider> resource_provider(ResourceProvider::Create(
       output_surface.get(), shared_bitmap_manager_.get(),
-      gpu_memory_buffer_manager_.get(), NULL, 0, false, 1, false));
+      gpu_memory_buffer_manager_.get(), NULL, 0, false, 1));
 
   uint32 sync_point = 0;
   unsigned target = GL_TEXTURE_2D;
@@ -3175,7 +3175,7 @@ TEST_P(ResourceProviderTest, TextureAllocation) {
 
   scoped_ptr<ResourceProvider> resource_provider(ResourceProvider::Create(
       output_surface.get(), shared_bitmap_manager_.get(),
-      gpu_memory_buffer_manager_.get(), NULL, 0, false, 1, false));
+      gpu_memory_buffer_manager_.get(), NULL, 0, false, 1));
 
   gfx::Size size(2, 2);
   gfx::Vector2d offset(0, 0);
@@ -3249,7 +3249,7 @@ TEST_P(ResourceProviderTest, TextureAllocationHint) {
 
   scoped_ptr<ResourceProvider> resource_provider(ResourceProvider::Create(
       output_surface.get(), shared_bitmap_manager_.get(),
-      gpu_memory_buffer_manager_.get(), NULL, 0, false, 1, false));
+      gpu_memory_buffer_manager_.get(), NULL, 0, false, 1));
 
   gfx::Size size(2, 2);
 
@@ -3304,7 +3304,7 @@ TEST_P(ResourceProviderTest, TextureAllocationHint_BGRA) {
 
   scoped_ptr<ResourceProvider> resource_provider(ResourceProvider::Create(
       output_surface.get(), shared_bitmap_manager_.get(),
-      gpu_memory_buffer_manager_.get(), NULL, 0, false, 1, false));
+      gpu_memory_buffer_manager_.get(), NULL, 0, false, 1));
 
   gfx::Size size(2, 2);
   const ResourceFormat formats[2] = {RGBA_8888, BGRA_8888};
@@ -3358,7 +3358,7 @@ TEST_P(ResourceProviderTest, PixelBuffer_GLTexture) {
 
   scoped_ptr<ResourceProvider> resource_provider(ResourceProvider::Create(
       output_surface.get(), shared_bitmap_manager_.get(),
-      gpu_memory_buffer_manager_.get(), NULL, 0, false, 1, false));
+      gpu_memory_buffer_manager_.get(), NULL, 0, false, 1));
 
   id = resource_provider->CreateResource(
       size, GL_CLAMP_TO_EDGE, ResourceProvider::TEXTURE_HINT_IMMUTABLE, format);
@@ -3400,7 +3400,7 @@ TEST_P(ResourceProviderTest, ForcingAsyncUploadToComplete) {
 
   scoped_ptr<ResourceProvider> resource_provider(ResourceProvider::Create(
       output_surface.get(), shared_bitmap_manager_.get(),
-      gpu_memory_buffer_manager_.get(), NULL, 0, false, 1, false));
+      gpu_memory_buffer_manager_.get(), NULL, 0, false, 1));
 
   id = resource_provider->CreateResource(
       size, GL_CLAMP_TO_EDGE, ResourceProvider::TEXTURE_HINT_IMMUTABLE, format);
@@ -3442,7 +3442,7 @@ TEST_P(ResourceProviderTest, PixelBufferLostContext) {
 
   scoped_ptr<ResourceProvider> resource_provider(ResourceProvider::Create(
       output_surface.get(), shared_bitmap_manager_.get(),
-      gpu_memory_buffer_manager_.get(), NULL, 0, false, 1, false));
+      gpu_memory_buffer_manager_.get(), NULL, 0, false, 1));
 
   EXPECT_CALL(*context, NextTextureId()).WillRepeatedly(Return(texture_id));
 
@@ -3482,7 +3482,7 @@ TEST_P(ResourceProviderTest, Image_GLTexture) {
 
   scoped_ptr<ResourceProvider> resource_provider(ResourceProvider::Create(
       output_surface.get(), shared_bitmap_manager_.get(),
-      gpu_memory_buffer_manager_.get(), NULL, 0, false, 1, false));
+      gpu_memory_buffer_manager_.get(), NULL, 0, false, 1));
 
   id = resource_provider->CreateResource(
       size, GL_CLAMP_TO_EDGE, ResourceProvider::TEXTURE_HINT_IMMUTABLE, format);
@@ -3542,87 +3542,6 @@ TEST_P(ResourceProviderTest, Image_GLTexture) {
       .RetiresOnSaturation();
 }
 
-TEST_P(ResourceProviderTest, CopyResource_GLTexture) {
-  if (GetParam() != ResourceProvider::RESOURCE_TYPE_GL_TEXTURE)
-    return;
-  scoped_ptr<AllocationTrackingContext3D> context_owned(
-      new StrictMock<AllocationTrackingContext3D>);
-  AllocationTrackingContext3D* context = context_owned.get();
-  context_owned->set_support_sync_query(true);
-
-  FakeOutputSurfaceClient output_surface_client;
-  scoped_ptr<OutputSurface> output_surface(
-      FakeOutputSurface::Create3d(context_owned.Pass()));
-  ASSERT_TRUE(output_surface->BindToClient(&output_surface_client));
-
-  const int kWidth = 2;
-  const int kHeight = 2;
-  gfx::Size size(kWidth, kHeight);
-  ResourceFormat format = RGBA_8888;
-  ResourceId source_id = 0;
-  ResourceId dest_id = 0;
-  const unsigned kSourceTextureId = 123u;
-  const unsigned kDestTextureId = 321u;
-  const unsigned kImageId = 234u;
-
-  scoped_ptr<ResourceProvider> resource_provider(ResourceProvider::Create(
-      output_surface.get(), shared_bitmap_manager_.get(),
-      gpu_memory_buffer_manager_.get(), NULL, 0, false, 1, false));
-
-  source_id = resource_provider->CreateResource(
-      size, GL_CLAMP_TO_EDGE, ResourceProvider::TEXTURE_HINT_IMMUTABLE, format);
-
-  EXPECT_CALL(*context, NextTextureId())
-      .WillOnce(Return(kSourceTextureId))
-      .RetiresOnSaturation();
-  EXPECT_CALL(*context, bindTexture(GL_TEXTURE_2D, kSourceTextureId))
-      .Times(1)
-      .RetiresOnSaturation();
-  EXPECT_CALL(*context, createImageCHROMIUM(_, kWidth, kHeight, GL_RGBA))
-      .WillOnce(Return(kImageId))
-      .RetiresOnSaturation();
-  {
-    ResourceProvider::ScopedWriteLockGpuMemoryBuffer lock(
-        resource_provider.get(), source_id);
-    EXPECT_TRUE(lock.GetGpuMemoryBuffer());
-  }
-  Mock::VerifyAndClearExpectations(context);
-
-  dest_id = resource_provider->CreateResource(
-      size, GL_CLAMP_TO_EDGE, ResourceProvider::TEXTURE_HINT_IMMUTABLE, format);
-
-  EXPECT_CALL(*context, NextTextureId())
-      .WillOnce(Return(kDestTextureId))
-      .RetiresOnSaturation();
-  EXPECT_CALL(*context, bindTexture(GL_TEXTURE_2D, kDestTextureId))
-      .Times(2)
-      .RetiresOnSaturation();
-  EXPECT_CALL(*context, texImage2D(GL_TEXTURE_2D, 0, GL_RGBA, 2, 2, 0, GL_RGBA,
-                                   GL_UNSIGNED_BYTE, nullptr))
-      .Times(1)
-      .RetiresOnSaturation();
-  EXPECT_CALL(*context, bindTexture(GL_TEXTURE_2D, kSourceTextureId))
-      .Times(1)
-      .RetiresOnSaturation();
-  EXPECT_CALL(*context, bindTexImage2DCHROMIUM(GL_TEXTURE_2D, kImageId))
-      .Times(1)
-      .RetiresOnSaturation();
-  resource_provider->CopyResource(source_id, dest_id, gfx::Rect(size));
-  Mock::VerifyAndClearExpectations(context);
-
-  EXPECT_CALL(*context, destroyImageCHROMIUM(kImageId))
-      .Times(1)
-      .RetiresOnSaturation();
-  EXPECT_CALL(*context, RetireTextureId(kSourceTextureId))
-      .Times(1)
-      .RetiresOnSaturation();
-  EXPECT_CALL(*context, RetireTextureId(kDestTextureId))
-      .Times(1)
-      .RetiresOnSaturation();
-  resource_provider->DeleteResource(source_id);
-  resource_provider->DeleteResource(dest_id);
-}
-
 TEST_P(ResourceProviderTest, CompressedTextureETC1Allocate) {
   if (GetParam() != ResourceProvider::RESOURCE_TYPE_GL_TEXTURE)
     return;
@@ -3640,7 +3559,7 @@ TEST_P(ResourceProviderTest, CompressedTextureETC1Allocate) {
   gfx::Size size(4, 4);
   scoped_ptr<ResourceProvider> resource_provider(ResourceProvider::Create(
       output_surface.get(), shared_bitmap_manager_.get(),
-      gpu_memory_buffer_manager_.get(), NULL, 0, false, 1, false));
+      gpu_memory_buffer_manager_.get(), NULL, 0, false, 1));
   int texture_id = 123;
 
   ResourceId id = resource_provider->CreateResource(
@@ -3671,7 +3590,7 @@ TEST_P(ResourceProviderTest, CompressedTextureETC1Upload) {
   gfx::Size size(4, 4);
   scoped_ptr<ResourceProvider> resource_provider(ResourceProvider::Create(
       output_surface.get(), shared_bitmap_manager_.get(),
-      gpu_memory_buffer_manager_.get(), NULL, 0, false, 1, false));
+      gpu_memory_buffer_manager_.get(), NULL, 0, false, 1));
   int texture_id = 123;
   uint8_t pixels[8];
 
@@ -3727,7 +3646,7 @@ TEST(ResourceProviderTest, TextureAllocationChunkSize) {
     size_t kTextureAllocationChunkSize = 1;
     scoped_ptr<ResourceProvider> resource_provider(ResourceProvider::Create(
         output_surface.get(), shared_bitmap_manager.get(), NULL, NULL, 0, false,
-        kTextureAllocationChunkSize, false));
+        kTextureAllocationChunkSize));
 
     ResourceId id = resource_provider->CreateResource(
         size, GL_CLAMP_TO_EDGE, ResourceProvider::TEXTURE_HINT_IMMUTABLE,
@@ -3743,7 +3662,7 @@ TEST(ResourceProviderTest, TextureAllocationChunkSize) {
     size_t kTextureAllocationChunkSize = 8;
     scoped_ptr<ResourceProvider> resource_provider(ResourceProvider::Create(
         output_surface.get(), shared_bitmap_manager.get(), NULL, NULL, 0, false,
-        kTextureAllocationChunkSize, false));
+        kTextureAllocationChunkSize));
 
     ResourceId id = resource_provider->CreateResource(
         size, GL_CLAMP_TO_EDGE, ResourceProvider::TEXTURE_HINT_IMMUTABLE,
diff --git a/cc/surfaces/display.cc b/cc/surfaces/display.cc
index 7e80854..586bfad 100644
--- a/cc/surfaces/display.cc
+++ b/cc/surfaces/display.cc
@@ -101,13 +101,10 @@ void Display::InitializeRenderer() {
   if (resource_provider_)
     return;
 
-  // Display does not use GpuMemoryBuffers, so persistent map is not relevant.
-  bool use_persistent_map_for_gpu_memory_buffers = false;
   scoped_ptr<ResourceProvider> resource_provider = ResourceProvider::Create(
       output_surface_.get(), bitmap_manager_, gpu_memory_buffer_manager_,
       nullptr, settings_.highp_threshold_min, settings_.use_rgba_4444_textures,
-      settings_.texture_id_allocation_chunk_size,
-      use_persistent_map_for_gpu_memory_buffers);
+      settings_.texture_id_allocation_chunk_size);
   if (!resource_provider)
     return;
 
diff --git a/cc/surfaces/surface_display_output_surface.cc b/cc/surfaces/surface_display_output_surface.cc
index e21d15d..6a2779d 100644
--- a/cc/surfaces/surface_display_output_surface.cc
+++ b/cc/surfaces/surface_display_output_surface.cc
@@ -17,8 +17,9 @@ namespace cc {
 SurfaceDisplayOutputSurface::SurfaceDisplayOutputSurface(
     SurfaceManager* surface_manager,
     SurfaceIdAllocator* allocator,
-    const scoped_refptr<ContextProvider>& context_provider)
-    : OutputSurface(context_provider),
+    const scoped_refptr<ContextProvider>& context_provider,
+    const scoped_refptr<ContextProvider>& worker_context_provider)
+    : OutputSurface(context_provider, worker_context_provider),
       display_client_(NULL),
       factory_(surface_manager, this),
       allocator_(allocator) {
diff --git a/cc/surfaces/surface_display_output_surface.h b/cc/surfaces/surface_display_output_surface.h
index c11bb86..2cdaadc 100644
--- a/cc/surfaces/surface_display_output_surface.h
+++ b/cc/surfaces/surface_display_output_surface.h
@@ -27,7 +27,8 @@ class CC_SURFACES_EXPORT SurfaceDisplayOutputSurface
   SurfaceDisplayOutputSurface(
       SurfaceManager* surface_manager,
       SurfaceIdAllocator* allocator,
-      const scoped_refptr<ContextProvider>& context_provider);
+      const scoped_refptr<ContextProvider>& context_provider,
+      const scoped_refptr<ContextProvider>& worker_context_provider);
   ~SurfaceDisplayOutputSurface() override;
 
   void set_display_client(OnscreenDisplayClient* display_client) {
diff --git a/cc/surfaces/surface_display_output_surface_unittest.cc b/cc/surfaces/surface_display_output_surface_unittest.cc
index 9b67f1f..3145729 100644
--- a/cc/surfaces/surface_display_output_surface_unittest.cc
+++ b/cc/surfaces/surface_display_output_surface_unittest.cc
@@ -60,7 +60,8 @@ class SurfaceDisplayOutputSurfaceTest : public testing::Test {
         context_provider_(TestContextProvider::Create()),
         surface_display_output_surface_(&surface_manager_,
                                         &allocator_,
-                                        context_provider_) {
+                                        context_provider_,
+                                        nullptr) {
     output_surface_ = display_client_.output_surface();
     display_client_.set_surface_output_surface(
         &surface_display_output_surface_);
diff --git a/cc/test/fake_output_surface.h b/cc/test/fake_output_surface.h
index 7041a44..8edec08 100644
--- a/cc/test/fake_output_surface.h
+++ b/cc/test/fake_output_surface.h
@@ -60,13 +60,15 @@ class FakeOutputSurface : public OutputSurface {
 
   static scoped_ptr<FakeOutputSurface> CreateDelegating3d(
       scoped_refptr<TestContextProvider> context_provider) {
-    return make_scoped_ptr(new FakeOutputSurface(context_provider, true));
+    return make_scoped_ptr(new FakeOutputSurface(
+        context_provider, TestContextProvider::Create(), true));
   }
 
   static scoped_ptr<FakeOutputSurface> CreateDelegating3d(
       scoped_ptr<TestWebGraphicsContext3D> context) {
-    return make_scoped_ptr(new FakeOutputSurface(
-        TestContextProvider::Create(context.Pass()), true));
+    return make_scoped_ptr(
+        new FakeOutputSurface(TestContextProvider::Create(context.Pass()),
+                              TestContextProvider::Create(), true));
   }
 
   static scoped_ptr<FakeOutputSurface> CreateDelegatingSoftware(
diff --git a/cc/test/fake_resource_provider.h b/cc/test/fake_resource_provider.h
index c96beba..5570703 100644
--- a/cc/test/fake_resource_provider.h
+++ b/cc/test/fake_resource_provider.h
@@ -14,9 +14,8 @@ class FakeResourceProvider : public ResourceProvider {
   static scoped_ptr<FakeResourceProvider> Create(
       OutputSurface* output_surface,
       SharedBitmapManager* shared_bitmap_manager) {
-    scoped_ptr<FakeResourceProvider> provider(
-        new FakeResourceProvider(output_surface, shared_bitmap_manager, nullptr,
-                                 nullptr, 0, false, 1, false));
+    scoped_ptr<FakeResourceProvider> provider(new FakeResourceProvider(
+        output_surface, shared_bitmap_manager, nullptr, nullptr, 0, false, 1));
     provider->Initialize();
     return provider;
   }
@@ -27,7 +26,7 @@ class FakeResourceProvider : public ResourceProvider {
       gpu::GpuMemoryBufferManager* gpu_memory_buffer_manager) {
     scoped_ptr<FakeResourceProvider> provider(new FakeResourceProvider(
         output_surface, shared_bitmap_manager, gpu_memory_buffer_manager,
-        nullptr, 0, false, 1, false));
+        nullptr, 0, false, 1));
     provider->Initialize();
     return provider;
   }
@@ -39,16 +38,14 @@ class FakeResourceProvider : public ResourceProvider {
                        BlockingTaskRunner* blocking_main_thread_task_runner,
                        int highp_threshold_min,
                        bool use_rgba_4444_texture_format,
-                       size_t id_allocation_chunk_size,
-                       bool use_persistent_map_for_gpu_memory_buffers)
+                       size_t id_allocation_chunk_size)
       : ResourceProvider(output_surface,
                          shared_bitmap_manager,
                          gpu_memory_buffer_manager,
                          blocking_main_thread_task_runner,
                          highp_threshold_min,
                          use_rgba_4444_texture_format,
-                         id_allocation_chunk_size,
-                         use_persistent_map_for_gpu_memory_buffers) {}
+                         id_allocation_chunk_size) {}
 };
 
 }  // namespace cc
diff --git a/cc/test/layer_tree_pixel_resource_test.cc b/cc/test/layer_tree_pixel_resource_test.cc
index bc323f1..1bd39892 100644
--- a/cc/test/layer_tree_pixel_resource_test.cc
+++ b/cc/test/layer_tree_pixel_resource_test.cc
@@ -127,8 +127,7 @@ void LayerTreeHostPixelResourceTest::InitializeFromTestCase(
 void LayerTreeHostPixelResourceTest::CreateResourceAndTileTaskWorkerPool(
     LayerTreeHostImpl* host_impl,
     scoped_ptr<TileTaskWorkerPool>* tile_task_worker_pool,
-    scoped_ptr<ResourcePool>* resource_pool,
-    scoped_ptr<ResourcePool>* staging_resource_pool) {
+    scoped_ptr<ResourcePool>* resource_pool) {
   base::SingleThreadTaskRunner* task_runner =
       proxy()->HasImplThread() ? proxy()->ImplThreadTaskRunner()
                                : proxy()->MainThreadTaskRunner();
@@ -140,6 +139,7 @@ void LayerTreeHostPixelResourceTest::CreateResourceAndTileTaskWorkerPool(
   ResourceProvider* resource_provider = host_impl->resource_provider();
   size_t max_transfer_buffer_usage_bytes = 1024u * 1024u * 60u;
   int max_bytes_per_copy_operation = 1024 * 1024;
+  int max_staging_buffers = 32;
 
   switch (resource_pool_option_) {
     case BITMAP_TILE_TASK_WORKER_POOL:
@@ -177,15 +177,13 @@ void LayerTreeHostPixelResourceTest::CreateResourceAndTileTaskWorkerPool(
       EXPECT_TRUE(context_provider);
       EXPECT_EQ(PIXEL_TEST_GL, test_type_);
       EXPECT_TRUE(host_impl->GetRendererCapabilities().using_image);
-      // We need to create a staging resource pool when using copy rasterizer.
-      *staging_resource_pool =
-          ResourcePool::Create(resource_provider, staging_texture_target_);
       *resource_pool =
           ResourcePool::Create(resource_provider, draw_texture_target_);
 
       *tile_task_worker_pool = OneCopyTileTaskWorkerPool::Create(
           task_runner, task_graph_runner(), context_provider, resource_provider,
-          staging_resource_pool->get(), max_bytes_per_copy_operation, false);
+          max_bytes_per_copy_operation, false, staging_texture_target_,
+          max_staging_buffers);
       break;
     case PIXEL_BUFFER_TILE_TASK_WORKER_POOL:
       EXPECT_TRUE(context_provider);
diff --git a/cc/test/layer_tree_pixel_resource_test.h b/cc/test/layer_tree_pixel_resource_test.h
index 16b7ca9..405690e 100644
--- a/cc/test/layer_tree_pixel_resource_test.h
+++ b/cc/test/layer_tree_pixel_resource_test.h
@@ -38,8 +38,7 @@ class LayerTreeHostPixelResourceTest : public LayerTreePixelTest {
   void CreateResourceAndTileTaskWorkerPool(
       LayerTreeHostImpl* host_impl,
       scoped_ptr<TileTaskWorkerPool>* tile_task_worker_pool,
-      scoped_ptr<ResourcePool>* resource_pool,
-      scoped_ptr<ResourcePool>* staging_resource_pool) override;
+      scoped_ptr<ResourcePool>* resource_pool) override;
 
   void RunPixelResourceTest(scoped_refptr<Layer> content_root,
                             base::FilePath file_name);
diff --git a/cc/test/layer_tree_pixel_test.cc b/cc/test/layer_tree_pixel_test.cc
index c145735..77399b9 100644
--- a/cc/test/layer_tree_pixel_test.cc
+++ b/cc/test/layer_tree_pixel_test.cc
@@ -381,4 +381,10 @@ void LayerTreePixelTest::CopyBitmapToTextureMailboxAsTexture(
                  texture_id));
 }
 
+void LayerTreePixelTest::Finish() {
+  scoped_ptr<gpu::GLInProcessContext> context = CreateTestInProcessContext();
+  GLES2Interface* gl = context->GetImplementation();
+  gl->Finish();
+}
+
 }  // namespace cc
diff --git a/cc/test/layer_tree_pixel_test.h b/cc/test/layer_tree_pixel_test.h
index f2352b0..fcaf1b4 100644
--- a/cc/test/layer_tree_pixel_test.h
+++ b/cc/test/layer_tree_pixel_test.h
@@ -91,6 +91,8 @@ class LayerTreePixelTest : public LayerTreeTest {
                              uint32 sync_point,
                              bool lost_resource);
 
+  void Finish();
+
   void set_enlarge_texture_amount(const gfx::Vector2d& enlarge_texture_amount) {
     enlarge_texture_amount_ = enlarge_texture_amount;
   }
diff --git a/cc/test/layer_tree_test.cc b/cc/test/layer_tree_test.cc
index 7823ef1..e81f1e3 100644
--- a/cc/test/layer_tree_test.cc
+++ b/cc/test/layer_tree_test.cc
@@ -107,10 +107,9 @@ DrawResult TestHooks::PrepareToDrawOnThread(
 void TestHooks::CreateResourceAndTileTaskWorkerPool(
     LayerTreeHostImpl* host_impl,
     scoped_ptr<TileTaskWorkerPool>* tile_task_worker_pool,
-    scoped_ptr<ResourcePool>* resource_pool,
-    scoped_ptr<ResourcePool>* staging_resource_pool) {
+    scoped_ptr<ResourcePool>* resource_pool) {
   host_impl->LayerTreeHostImpl::CreateResourceAndTileTaskWorkerPool(
-      tile_task_worker_pool, resource_pool, staging_resource_pool);
+      tile_task_worker_pool, resource_pool);
 }
 
 // Adapts ThreadProxy for test. Injects test hooks for testing.
@@ -303,10 +302,9 @@ class LayerTreeHostImplForTesting : public LayerTreeHostImpl {
 
   void CreateResourceAndTileTaskWorkerPool(
       scoped_ptr<TileTaskWorkerPool>* tile_task_worker_pool,
-      scoped_ptr<ResourcePool>* resource_pool,
-      scoped_ptr<ResourcePool>* staging_resource_pool) override {
+      scoped_ptr<ResourcePool>* resource_pool) override {
     test_hooks_->CreateResourceAndTileTaskWorkerPool(
-        this, tile_task_worker_pool, resource_pool, staging_resource_pool);
+        this, tile_task_worker_pool, resource_pool);
   }
 
   void WillBeginImplFrame(const BeginFrameArgs& args) override {
diff --git a/cc/test/layer_tree_test.h b/cc/test/layer_tree_test.h
index 828850c..9f6d369 100644
--- a/cc/test/layer_tree_test.h
+++ b/cc/test/layer_tree_test.h
@@ -55,8 +55,7 @@ class TestHooks : public AnimationDelegate {
   virtual void CreateResourceAndTileTaskWorkerPool(
       LayerTreeHostImpl* host_impl,
       scoped_ptr<TileTaskWorkerPool>* tile_task_worker_pool,
-      scoped_ptr<ResourcePool>* resource_pool,
-      scoped_ptr<ResourcePool>* staging_resource_pool);
+      scoped_ptr<ResourcePool>* resource_pool);
   virtual void WillBeginImplFrameOnThread(LayerTreeHostImpl* host_impl,
                                           const BeginFrameArgs& args) {}
   virtual void DidFinishImplFrameOnThread(LayerTreeHostImpl* host_impl) {}
diff --git a/cc/test/pixel_test.cc b/cc/test/pixel_test.cc
index ae3cef1..e7e76ff 100644
--- a/cc/test/pixel_test.cc
+++ b/cc/test/pixel_test.cc
@@ -136,7 +136,7 @@ void PixelTest::SetUpGLRenderer(bool use_skia_gpu_backend,
   resource_provider_ = ResourceProvider::Create(
       output_surface_.get(), shared_bitmap_manager_.get(),
       gpu_memory_buffer_manager_.get(), main_thread_task_runner_.get(), 0,
-      false, 1, false);
+      false, 1);
 
   texture_mailbox_deleter_ = make_scoped_ptr(
       new TextureMailboxDeleter(base::ThreadTaskRunnerHandle::Get()));
@@ -177,7 +177,7 @@ void PixelTest::SetUpSoftwareRenderer() {
   resource_provider_ = ResourceProvider::Create(
       output_surface_.get(), shared_bitmap_manager_.get(),
       gpu_memory_buffer_manager_.get(), main_thread_task_runner_.get(), 0,
-      false, 1, false);
+      false, 1);
   renderer_ =
       SoftwareRenderer::Create(this, &settings_.renderer_settings,
                                output_surface_.get(), resource_provider_.get());
diff --git a/cc/test/test_in_process_context_provider.cc b/cc/test/test_in_process_context_provider.cc
index efbbeec..79d53d2 100644
--- a/cc/test/test_in_process_context_provider.cc
+++ b/cc/test/test_in_process_context_provider.cc
@@ -142,6 +142,7 @@ TestInProcessContextProvider::ContextCapabilities() {
   ContextProvider::Capabilities capabilities;
   capabilities.gpu.image = true;
   capabilities.gpu.texture_rectangle = true;
+  capabilities.gpu.sync_query = true;
 
   switch (PlatformColor::Format()) {
     case PlatformColor::SOURCE_FORMAT_RGBA8:
diff --git a/cc/tiles/tile_manager.cc b/cc/tiles/tile_manager.cc
index 3ff2885..6d192a1 100644
--- a/cc/tiles/tile_manager.cc
+++ b/cc/tiles/tile_manager.cc
@@ -494,7 +494,7 @@ void TileManager::AssignGpuMemoryToTiles(
   // or deleted. If this operation becomes expensive too, only do this after
   // some resource(s) was returned. Note that in that case, one also need to
   // invalidate when releasing some resource from the pool.
-  resource_pool_->CheckBusyResources(false);
+  resource_pool_->CheckBusyResources();
 
   // Now give memory out to the tiles until we're out, and build
   // the needs-to-be-rasterized queue.
diff --git a/cc/trees/layer_tree_host_impl.cc b/cc/trees/layer_tree_host_impl.cc
index ae0c943..f638bd4 100644
--- a/cc/trees/layer_tree_host_impl.cc
+++ b/cc/trees/layer_tree_host_impl.cc
@@ -153,11 +153,6 @@ size_t GetMaxTransferBufferUsageBytes(
                   max_transfer_buffer_usage_bytes);
 }
 
-size_t GetMaxStagingResourceCount() {
-  // Upper bound for number of staging resource to allow.
-  return 32;
-}
-
 size_t GetDefaultMemoryAllocationLimit() {
   // TODO(ccameron): (http://crbug.com/137094) This 64MB default is a straggler
   // from the old texture manager and is just to give us a default memory
@@ -1235,7 +1230,7 @@ void LayerTreeHostImpl::UpdateTileManagerMemoryPolicy(
       100);
 
   DCHECK(resource_pool_);
-  resource_pool_->CheckBusyResources(false);
+  resource_pool_->CheckBusyResources();
   // Soft limit is used for resource pool such that memory returns to soft
   // limit after going over.
   resource_pool_->SetResourceUsageLimits(
@@ -1243,15 +1238,6 @@ void LayerTreeHostImpl::UpdateTileManagerMemoryPolicy(
       unused_memory_limit_in_bytes,
       global_tile_state_.num_resources_limit);
 
-  // Release all staging resources when invisible.
-  if (staging_resource_pool_) {
-    staging_resource_pool_->CheckBusyResources(false);
-    staging_resource_pool_->SetResourceUsageLimits(
-        std::numeric_limits<size_t>::max(),
-        std::numeric_limits<size_t>::max(),
-        visible_ ? GetMaxStagingResourceCount() : 0);
-  }
-
   DidModifyTilePriorities();
 }
 
@@ -1465,7 +1451,7 @@ void LayerTreeHostImpl::ReclaimResources(const CompositorFrameAck* ack) {
   // In OOM, we now might be able to release more resources that were held
   // because they were exported.
   if (resource_pool_) {
-    resource_pool_->CheckBusyResources(false);
+    resource_pool_->CheckBusyResources();
     resource_pool_->ReduceResourceUsage();
   }
   // If we're not visible, we likely released resources, so we want to
@@ -2061,8 +2047,7 @@ void LayerTreeHostImpl::CreateAndSetRenderer() {
 }
 
 void LayerTreeHostImpl::CreateTileManagerResources() {
-  CreateResourceAndTileTaskWorkerPool(&tile_task_worker_pool_, &resource_pool_,
-                                      &staging_resource_pool_);
+  CreateResourceAndTileTaskWorkerPool(&tile_task_worker_pool_, &resource_pool_);
   // TODO(vmpstr): Initialize tile task limit at ctor time.
   tile_manager_->SetResources(
       resource_pool_.get(), tile_task_worker_pool_->AsTileTaskRunner(),
@@ -2073,8 +2058,7 @@ void LayerTreeHostImpl::CreateTileManagerResources() {
 
 void LayerTreeHostImpl::CreateResourceAndTileTaskWorkerPool(
     scoped_ptr<TileTaskWorkerPool>* tile_task_worker_pool,
-    scoped_ptr<ResourcePool>* resource_pool,
-    scoped_ptr<ResourcePool>* staging_resource_pool) {
+    scoped_ptr<ResourcePool>* resource_pool) {
   DCHECK(GetTaskRunner());
   // TODO(vmpstr): Make this a DCHECK (or remove) when crbug.com/419086 is
   // resolved.
@@ -2131,14 +2115,6 @@ void LayerTreeHostImpl::CreateResourceAndTileTaskWorkerPool(
   }
 
   if (settings_.use_one_copy) {
-    // Synchronous single-threaded mode depends on tiles being ready to
-    // draw when raster is complete.  Therefore, it must use one of zero
-    // copy, software raster, or GPU raster.
-    DCHECK(!is_synchronous_single_threaded_);
-
-    // We need to create a staging resource pool when using copy rasterizer.
-    *staging_resource_pool =
-        ResourcePool::Create(resource_provider_.get(), image_target);
     *resource_pool =
         ResourcePool::Create(resource_provider_.get(), GL_TEXTURE_2D);
 
@@ -2148,9 +2124,9 @@ void LayerTreeHostImpl::CreateResourceAndTileTaskWorkerPool(
 
     *tile_task_worker_pool = OneCopyTileTaskWorkerPool::Create(
         GetTaskRunner(), task_graph_runner, context_provider,
-        resource_provider_.get(), staging_resource_pool_.get(),
-        max_copy_texture_chromium_size,
-        settings_.use_persistent_map_for_gpu_memory_buffers);
+        resource_provider_.get(), max_copy_texture_chromium_size,
+        settings_.use_persistent_map_for_gpu_memory_buffers, image_target,
+        settings_.max_staging_buffers);
     return;
   }
 
@@ -2193,7 +2169,6 @@ void LayerTreeHostImpl::PostFrameTimingEvents(
 void LayerTreeHostImpl::CleanUpTileManager() {
   tile_manager_->FinishTasksAndCleanUp();
   resource_pool_ = nullptr;
-  staging_resource_pool_ = nullptr;
   tile_task_worker_pool_ = nullptr;
   single_thread_synchronous_task_graph_runner_ = nullptr;
 }
@@ -2226,8 +2201,7 @@ bool LayerTreeHostImpl::InitializeRenderer(
       proxy_->blocking_main_thread_task_runner(),
       settings_.renderer_settings.highp_threshold_min,
       settings_.renderer_settings.use_rgba_4444_textures,
-      settings_.renderer_settings.texture_id_allocation_chunk_size,
-      settings_.use_persistent_map_for_gpu_memory_buffers);
+      settings_.renderer_settings.texture_id_allocation_chunk_size);
 
   CreateAndSetRenderer();
 
diff --git a/cc/trees/layer_tree_host_impl.h b/cc/trees/layer_tree_host_impl.h
index b9a07f6..7936eb5 100644
--- a/cc/trees/layer_tree_host_impl.h
+++ b/cc/trees/layer_tree_host_impl.h
@@ -559,8 +559,7 @@ class CC_EXPORT LayerTreeHostImpl
 
   virtual void CreateResourceAndTileTaskWorkerPool(
       scoped_ptr<TileTaskWorkerPool>* tile_task_worker_pool,
-      scoped_ptr<ResourcePool>* resource_pool,
-      scoped_ptr<ResourcePool>* staging_resource_pool);
+      scoped_ptr<ResourcePool>* resource_pool);
 
   bool prepare_tiles_needed() const { return tile_priorities_dirty_; }
 
@@ -713,7 +712,6 @@ class CC_EXPORT LayerTreeHostImpl
   bool tree_resources_for_gpu_rasterization_dirty_;
   scoped_ptr<TileTaskWorkerPool> tile_task_worker_pool_;
   scoped_ptr<ResourcePool> resource_pool_;
-  scoped_ptr<ResourcePool> staging_resource_pool_;
   scoped_ptr<Renderer> renderer_;
 
   GlobalStateThatImpactsTilePriority global_tile_state_;
diff --git a/cc/trees/layer_tree_host_pixeltest_tiles.cc b/cc/trees/layer_tree_host_pixeltest_tiles.cc
index a07f100..ede77df 100644
--- a/cc/trees/layer_tree_host_pixeltest_tiles.cc
+++ b/cc/trees/layer_tree_host_pixeltest_tiles.cc
@@ -161,6 +161,7 @@ class LayerTreeHostTilesTestPartialInvalidation
         // only re-raster the stuff in the rect. If it doesn't do partial raster
         // it would re-raster the whole thing instead.
         client_.set_blue_top(false);
+        Finish();
         picture_layer_->SetNeedsDisplayRect(gfx::Rect(50, 50, 100, 100));
 
         // Add a copy request to see what happened!
diff --git a/cc/trees/layer_tree_settings.cc b/cc/trees/layer_tree_settings.cc
index 0869e92..d4150c7 100644
--- a/cc/trees/layer_tree_settings.cc
+++ b/cc/trees/layer_tree_settings.cc
@@ -71,7 +71,8 @@ LayerTreeSettings::LayerTreeSettings()
       gather_pixel_refs(false),
       use_compositor_animation_timelines(false),
       invert_viewport_scroll_order(false),
-      wait_for_beginframe_interval(true) {}
+      wait_for_beginframe_interval(true),
+      max_staging_buffers(32) {}
 
 LayerTreeSettings::~LayerTreeSettings() {}
 
diff --git a/cc/trees/layer_tree_settings.h b/cc/trees/layer_tree_settings.h
index 27e80af..bad1d3b 100644
--- a/cc/trees/layer_tree_settings.h
+++ b/cc/trees/layer_tree_settings.h
@@ -86,6 +86,7 @@ class CC_EXPORT LayerTreeSettings {
   bool use_compositor_animation_timelines;
   bool invert_viewport_scroll_order;
   bool wait_for_beginframe_interval;
+  int max_staging_buffers;
 
   LayerTreeDebugState initial_debug_state;
 
diff --git a/content/browser/compositor/browser_compositor_output_surface.cc b/content/browser/compositor/browser_compositor_output_surface.cc
index 0e014c9..bdf3a1b 100644
--- a/content/browser/compositor/browser_compositor_output_surface.cc
+++ b/content/browser/compositor/browser_compositor_output_surface.cc
@@ -17,14 +17,16 @@ namespace content {
 
 BrowserCompositorOutputSurface::BrowserCompositorOutputSurface(
     const scoped_refptr<cc::ContextProvider>& context_provider,
+    const scoped_refptr<cc::ContextProvider>& worker_context_provider,
     const scoped_refptr<ui::CompositorVSyncManager>& vsync_manager,
     scoped_ptr<BrowserCompositorOverlayCandidateValidator>
         overlay_candidate_validator)
-    : OutputSurface(context_provider),
+    : OutputSurface(context_provider, worker_context_provider),
       vsync_manager_(vsync_manager),
       reflector_(nullptr),
-      use_begin_frame_scheduling_(base::CommandLine::ForCurrentProcess()->
-          HasSwitch(cc::switches::kEnableBeginFrameScheduling)) {
+      use_begin_frame_scheduling_(
+          base::CommandLine::ForCurrentProcess()
+              ->HasSwitch(cc::switches::kEnableBeginFrameScheduling)) {
   overlay_candidate_validator_ = overlay_candidate_validator.Pass();
   Initialize();
 }
diff --git a/content/browser/compositor/browser_compositor_output_surface.h b/content/browser/compositor/browser_compositor_output_surface.h
index f582776..f5da02d 100644
--- a/content/browser/compositor/browser_compositor_output_surface.h
+++ b/content/browser/compositor/browser_compositor_output_surface.h
@@ -56,6 +56,7 @@ class CONTENT_EXPORT BrowserCompositorOutputSurface
   // Constructor used by the accelerated implementation.
   BrowserCompositorOutputSurface(
       const scoped_refptr<cc::ContextProvider>& context,
+      const scoped_refptr<cc::ContextProvider>& worker_context,
       const scoped_refptr<ui::CompositorVSyncManager>& vsync_manager,
       scoped_ptr<BrowserCompositorOverlayCandidateValidator>
           overlay_candidate_validator);
diff --git a/content/browser/compositor/gpu_browser_compositor_output_surface.cc b/content/browser/compositor/gpu_browser_compositor_output_surface.cc
index 81da36d..70725a2 100644
--- a/content/browser/compositor/gpu_browser_compositor_output_surface.cc
+++ b/content/browser/compositor/gpu_browser_compositor_output_surface.cc
@@ -17,10 +17,12 @@ namespace content {
 
 GpuBrowserCompositorOutputSurface::GpuBrowserCompositorOutputSurface(
     const scoped_refptr<ContextProviderCommandBuffer>& context,
+    const scoped_refptr<ContextProviderCommandBuffer>& worker_context,
     const scoped_refptr<ui::CompositorVSyncManager>& vsync_manager,
     scoped_ptr<BrowserCompositorOverlayCandidateValidator>
         overlay_candidate_validator)
     : BrowserCompositorOutputSurface(context,
+                                     worker_context,
                                      vsync_manager,
                                      overlay_candidate_validator.Pass()),
 #if defined(OS_MACOSX)
diff --git a/content/browser/compositor/gpu_browser_compositor_output_surface.h b/content/browser/compositor/gpu_browser_compositor_output_surface.h
index 3bc2a3a..7b4f242 100644
--- a/content/browser/compositor/gpu_browser_compositor_output_surface.h
+++ b/content/browser/compositor/gpu_browser_compositor_output_surface.h
@@ -26,6 +26,7 @@ class GpuBrowserCompositorOutputSurface
  public:
   GpuBrowserCompositorOutputSurface(
       const scoped_refptr<ContextProviderCommandBuffer>& context,
+      const scoped_refptr<ContextProviderCommandBuffer>& worker_context,
       const scoped_refptr<ui::CompositorVSyncManager>& vsync_manager,
       scoped_ptr<BrowserCompositorOverlayCandidateValidator>
           overlay_candidate_validator);
diff --git a/content/browser/compositor/gpu_process_transport_factory.cc b/content/browser/compositor/gpu_process_transport_factory.cc
index fcb0a3e..f03eb9e 100644
--- a/content/browser/compositor/gpu_process_transport_factory.cc
+++ b/content/browser/compositor/gpu_process_transport_factory.cc
@@ -237,6 +237,13 @@ void GpuProcessTransportFactory::EstablishedGpuChannel(
 
   scoped_refptr<ContextProviderCommandBuffer> context_provider;
   if (create_gpu_output_surface) {
+    // Try to reuse existing worker context provider.
+    if (shared_worker_context_provider_) {
+      base::AutoLock lock(*shared_worker_context_provider_->GetLock());
+      if (shared_worker_context_provider_->ContextGL()
+              ->GetGraphicsResetStatusKHR() != GL_NO_ERROR)
+        shared_worker_context_provider_ = nullptr;
+    }
     scoped_refptr<GpuChannelHost> gpu_channel_host =
         BrowserGpuChannelHostFactory::instance()->GetGpuChannel();
     if (gpu_channel_host.get()) {
@@ -246,12 +253,24 @@ void GpuProcessTransportFactory::EstablishedGpuChannel(
           BROWSER_COMPOSITOR_ONSCREEN_CONTEXT);
       if (context_provider && !context_provider->BindToCurrentThread())
         context_provider = nullptr;
+      if (!shared_worker_context_provider_) {
+        shared_worker_context_provider_ = ContextProviderCommandBuffer::Create(
+            GpuProcessTransportFactory::CreateContextCommon(gpu_channel_host,
+                                                            0),
+            BROWSER_WORKER_CONTEXT);
+        if (shared_worker_context_provider_ &&
+            !shared_worker_context_provider_->BindToCurrentThread())
+          shared_worker_context_provider_ = nullptr;
+      }
     }
 
+    bool created_gpu_browser_compositor =
+        !!context_provider && !!shared_worker_context_provider_;
+
     UMA_HISTOGRAM_BOOLEAN("Aura.CreatedGpuBrowserCompositor",
-                          !!context_provider.get());
+                          created_gpu_browser_compositor);
 
-    if (!context_provider) {
+    if (!created_gpu_browser_compositor) {
       // Try again.
       CauseForGpuLaunch cause =
           CAUSE_FOR_GPU_LAUNCH_WEBGRAPHICSCONTEXT3DCOMMANDBUFFERIMPL_INITIALIZE;
@@ -274,7 +293,8 @@ void GpuProcessTransportFactory::EstablishedGpuChannel(
         context_provider->ContextCapabilities();
     if (!data->surface_id) {
       surface = make_scoped_ptr(new OffscreenBrowserCompositorOutputSurface(
-          context_provider, compositor->vsync_manager(),
+          context_provider, shared_worker_context_provider_,
+          compositor->vsync_manager(),
           scoped_ptr<BrowserCompositorOverlayCandidateValidator>()));
     } else if (capabilities.gpu.surfaceless) {
       GLenum target = GL_TEXTURE_2D;
@@ -285,13 +305,15 @@ void GpuProcessTransportFactory::EstablishedGpuChannel(
 #endif
       surface =
           make_scoped_ptr(new GpuSurfacelessBrowserCompositorOutputSurface(
-              context_provider, data->surface_id, compositor->vsync_manager(),
+              context_provider, shared_worker_context_provider_,
+              data->surface_id, compositor->vsync_manager(),
               CreateOverlayCandidateValidator(compositor->widget()), target,
               format, BrowserGpuMemoryBufferManager::current()));
     } else {
       if (!surface) {
         surface = make_scoped_ptr(new GpuBrowserCompositorOutputSurface(
-            context_provider, compositor->vsync_manager(),
+            context_provider, shared_worker_context_provider_,
+            compositor->vsync_manager(),
             CreateOverlayCandidateValidator(compositor->widget())));
       }
     }
@@ -322,7 +344,8 @@ void GpuProcessTransportFactory::EstablishedGpuChannel(
 
   scoped_ptr<cc::SurfaceDisplayOutputSurface> output_surface(
       new cc::SurfaceDisplayOutputSurface(
-          manager, compositor->surface_id_allocator(), context_provider));
+          manager, compositor->surface_id_allocator(), context_provider,
+          shared_worker_context_provider_));
   display_client->set_surface_output_surface(output_surface.get());
   output_surface->set_display_client(display_client.get());
   display_client->display()->Resize(compositor->size());
diff --git a/content/browser/compositor/gpu_process_transport_factory.h b/content/browser/compositor/gpu_process_transport_factory.h
index c16b36e..564b3cd 100644
--- a/content/browser/compositor/gpu_process_transport_factory.h
+++ b/content/browser/compositor/gpu_process_transport_factory.h
@@ -102,6 +102,7 @@ class GpuProcessTransportFactory
   uint32_t next_surface_id_namespace_;
   scoped_ptr<cc::TaskGraphRunner> task_graph_runner_;
   scoped_ptr<base::SimpleThread> raster_thread_;
+  scoped_refptr<ContextProviderCommandBuffer> shared_worker_context_provider_;
 
 #if defined(OS_WIN)
   scoped_ptr<OutputDeviceBacking> software_backing_;
diff --git a/content/browser/compositor/gpu_surfaceless_browser_compositor_output_surface.cc b/content/browser/compositor/gpu_surfaceless_browser_compositor_output_surface.cc
index 9507cc3..d28e264 100644
--- a/content/browser/compositor/gpu_surfaceless_browser_compositor_output_surface.cc
+++ b/content/browser/compositor/gpu_surfaceless_browser_compositor_output_surface.cc
@@ -19,6 +19,7 @@ namespace content {
 GpuSurfacelessBrowserCompositorOutputSurface::
     GpuSurfacelessBrowserCompositorOutputSurface(
         const scoped_refptr<ContextProviderCommandBuffer>& context,
+        const scoped_refptr<ContextProviderCommandBuffer>& worker_context,
         int surface_id,
         const scoped_refptr<ui::CompositorVSyncManager>& vsync_manager,
         scoped_ptr<BrowserCompositorOverlayCandidateValidator>
@@ -27,6 +28,7 @@ GpuSurfacelessBrowserCompositorOutputSurface::
         unsigned int internalformat,
         BrowserGpuMemoryBufferManager* gpu_memory_buffer_manager)
     : GpuBrowserCompositorOutputSurface(context,
+                                        worker_context,
                                         vsync_manager,
                                         overlay_candidate_validator.Pass()),
       internalformat_(internalformat),
diff --git a/content/browser/compositor/gpu_surfaceless_browser_compositor_output_surface.h b/content/browser/compositor/gpu_surfaceless_browser_compositor_output_surface.h
index 27bb335..ed5618d 100644
--- a/content/browser/compositor/gpu_surfaceless_browser_compositor_output_surface.h
+++ b/content/browser/compositor/gpu_surfaceless_browser_compositor_output_surface.h
@@ -18,6 +18,7 @@ class GpuSurfacelessBrowserCompositorOutputSurface
  public:
   GpuSurfacelessBrowserCompositorOutputSurface(
       const scoped_refptr<ContextProviderCommandBuffer>& context,
+      const scoped_refptr<ContextProviderCommandBuffer>& worker_context,
       int surface_id,
       const scoped_refptr<ui::CompositorVSyncManager>& vsync_manager,
       scoped_ptr<BrowserCompositorOverlayCandidateValidator>
diff --git a/content/browser/compositor/offscreen_browser_compositor_output_surface.cc b/content/browser/compositor/offscreen_browser_compositor_output_surface.cc
index 3a98015..606b9fa 100644
--- a/content/browser/compositor/offscreen_browser_compositor_output_surface.cc
+++ b/content/browser/compositor/offscreen_browser_compositor_output_surface.cc
@@ -30,10 +30,12 @@ namespace content {
 OffscreenBrowserCompositorOutputSurface::
     OffscreenBrowserCompositorOutputSurface(
         const scoped_refptr<ContextProviderCommandBuffer>& context,
+        const scoped_refptr<ContextProviderCommandBuffer>& worker_context,
         const scoped_refptr<ui::CompositorVSyncManager>& vsync_manager,
         scoped_ptr<BrowserCompositorOverlayCandidateValidator>
             overlay_candidate_validator)
     : BrowserCompositorOutputSurface(context,
+                                     worker_context,
                                      vsync_manager,
                                      overlay_candidate_validator.Pass()),
       fbo_(0),
diff --git a/content/browser/compositor/offscreen_browser_compositor_output_surface.h b/content/browser/compositor/offscreen_browser_compositor_output_surface.h
index 67fc224..c9e32e7 100644
--- a/content/browser/compositor/offscreen_browser_compositor_output_surface.h
+++ b/content/browser/compositor/offscreen_browser_compositor_output_surface.h
@@ -23,6 +23,7 @@ class OffscreenBrowserCompositorOutputSurface
  public:
   OffscreenBrowserCompositorOutputSurface(
       const scoped_refptr<ContextProviderCommandBuffer>& context,
+      const scoped_refptr<ContextProviderCommandBuffer>& worker_context,
       const scoped_refptr<ui::CompositorVSyncManager>& vsync_manager,
       scoped_ptr<BrowserCompositorOverlayCandidateValidator>
           overlay_candidate_validator);
diff --git a/content/browser/compositor/reflector_impl_unittest.cc b/content/browser/compositor/reflector_impl_unittest.cc
index ce9fbb9..3377b8a 100644
--- a/content/browser/compositor/reflector_impl_unittest.cc
+++ b/content/browser/compositor/reflector_impl_unittest.cc
@@ -74,6 +74,7 @@ class TestOutputSurface : public BrowserCompositorOutputSurface {
       const scoped_refptr<cc::ContextProvider>& context_provider,
       const scoped_refptr<ui::CompositorVSyncManager>& vsync_manager)
       : BrowserCompositorOutputSurface(context_provider,
+                                       nullptr,
                                        vsync_manager,
                                        CreateTestValidatorOzone().Pass()) {}
 
diff --git a/content/browser/gpu/compositor_util.cc b/content/browser/gpu/compositor_util.cc
index 2811a0c..e80973b 100644
--- a/content/browser/gpu/compositor_util.cc
+++ b/content/browser/gpu/compositor_util.cc
@@ -242,14 +242,6 @@ bool IsOneCopyUploadEnabled() {
 bool IsZeroCopyUploadEnabled() {
   const base::CommandLine& command_line =
       *base::CommandLine::ForCurrentProcess();
-  // Single-threaded mode in the renderer process (for layout tests) is
-  // synchronous, which depends on tiles being ready to draw when raster is
-  // complete.  Therefore, it must use one of zero copy, software raster, or
-  // GPU raster. So we force zero-copy on for the case where software/GPU raster
-  // is not used.
-  // TODO(reveman): One-copy can work with sync compositing: crbug.com/490295.
-  if (command_line.HasSwitch(switches::kDisableThreadedCompositing))
-    return true;
   return command_line.HasSwitch(switches::kEnableZeroCopy);
 }
 
diff --git a/content/browser/renderer_host/compositor_impl_android.cc b/content/browser/renderer_host/compositor_impl_android.cc
index 358ac51..d502f23 100644
--- a/content/browser/renderer_host/compositor_impl_android.cc
+++ b/content/browser/renderer_host/compositor_impl_android.cc
@@ -674,7 +674,7 @@ void CompositorImpl::CreateOutputSurface() {
         base::ThreadTaskRunnerHandle::Get()));
     scoped_ptr<cc::SurfaceDisplayOutputSurface> surface_output_surface(
         new cc::SurfaceDisplayOutputSurface(
-            manager, surface_id_allocator_.get(), context_provider));
+            manager, surface_id_allocator_.get(), context_provider, nullptr));
 
     display_client_->set_surface_output_surface(surface_output_surface.get());
     surface_output_surface->set_display_client(display_client_.get());
diff --git a/content/common/gpu/client/command_buffer_metrics.cc b/content/common/gpu/client/command_buffer_metrics.cc
index d16f276..5065ca5 100644
--- a/content/common/gpu/client/command_buffer_metrics.cc
+++ b/content/common/gpu/client/command_buffer_metrics.cc
@@ -79,6 +79,10 @@ void RecordContextLost(CommandBufferContextType type,
       UMA_HISTOGRAM_ENUMERATION("GPU.ContextLost.BrowserMainThread", reason,
                                 CONTEXT_LOST_REASON_MAX_ENUM);
       break;
+    case BROWSER_WORKER_CONTEXT:
+      UMA_HISTOGRAM_ENUMERATION("GPU.ContextLost.BrowserWorker", reason,
+                                CONTEXT_LOST_REASON_MAX_ENUM);
+      break;
     case RENDER_COMPOSITOR_CONTEXT:
       UMA_HISTOGRAM_ENUMERATION("GPU.ContextLost.RenderCompositor", reason,
                                 CONTEXT_LOST_REASON_MAX_ENUM);
@@ -120,6 +124,8 @@ std::string CommandBufferContextTypeToString(CommandBufferContextType type) {
       return "Compositor";
     case BROWSER_OFFSCREEN_MAINTHREAD_CONTEXT:
       return "Offscreen-MainThread";
+    case BROWSER_WORKER_CONTEXT:
+      return "CompositorWorker";
     case RENDER_COMPOSITOR_CONTEXT:
       return "RenderCompositor";
     case RENDER_WORKER_CONTEXT:
diff --git a/content/common/gpu/client/command_buffer_metrics.h b/content/common/gpu/client/command_buffer_metrics.h
index e198d85..0b4790c 100644
--- a/content/common/gpu/client/command_buffer_metrics.h
+++ b/content/common/gpu/client/command_buffer_metrics.h
@@ -14,6 +14,7 @@ namespace content {
 enum CommandBufferContextType {
   BROWSER_COMPOSITOR_ONSCREEN_CONTEXT,
   BROWSER_OFFSCREEN_MAINTHREAD_CONTEXT,
+  BROWSER_WORKER_CONTEXT,
   RENDER_COMPOSITOR_CONTEXT,
   RENDER_WORKER_CONTEXT,
   RENDERER_MAINTHREAD_CONTEXT,
diff --git a/content/renderer/gpu/render_widget_compositor.cc b/content/renderer/gpu/render_widget_compositor.cc
index 942470b..aec0e5dc 100644
--- a/content/renderer/gpu/render_widget_compositor.cc
+++ b/content/renderer/gpu/render_widget_compositor.cc
@@ -467,6 +467,11 @@ void RenderWidgetCompositor::Initialize() {
     settings.use_external_begin_frame_source = false;
   }
 
+  settings.max_staging_buffers = 32;
+  // Use 1/4th of staging buffers on low-end devices.
+  if (base::SysInfo::IsLowEndDevice())
+    settings.max_staging_buffers /= 4;
+
   scoped_refptr<base::SingleThreadTaskRunner> compositor_thread_task_runner =
       compositor_deps_->GetCompositorImplThreadTaskRunner();
   scoped_refptr<base::SingleThreadTaskRunner>
diff --git a/gpu/command_buffer/service/query_manager.cc b/gpu/command_buffer/service/query_manager.cc
index 49f111f..a269f35 100644
--- a/gpu/command_buffer/service/query_manager.cc
+++ b/gpu/command_buffer/service/query_manager.cc
@@ -785,6 +785,7 @@ QueryManager::~QueryManager() {
 }
 
 void QueryManager::Destroy(bool have_context) {
+  active_queries_.clear();
   pending_queries_.clear();
   pending_transfer_queries_.clear();
   active_queries_.clear();
diff --git a/ui/compositor/test/in_process_context_factory.cc b/ui/compositor/test/in_process_context_factory.cc
index 13ddd90..942947d 100644
--- a/ui/compositor/test/in_process_context_factory.cc
+++ b/ui/compositor/test/in_process_context_factory.cc
@@ -41,9 +41,11 @@ class FakeReflector : public Reflector {
 // GL surface.
 class DirectOutputSurface : public cc::OutputSurface {
  public:
-  explicit DirectOutputSurface(
-      const scoped_refptr<cc::ContextProvider>& context_provider)
-      : cc::OutputSurface(context_provider), weak_ptr_factory_(this) {}
+  DirectOutputSurface(
+      const scoped_refptr<cc::ContextProvider>& context_provider,
+      const scoped_refptr<cc::ContextProvider>& worker_context_provider)
+      : cc::OutputSurface(context_provider, worker_context_provider),
+        weak_ptr_factory_(this) {}
 
   ~DirectOutputSurface() override {}
 
@@ -110,16 +112,19 @@ void InProcessContextFactory::CreateOutputSurface(
       InProcessContextProvider::Create(attribs, &gpu_memory_buffer_manager_,
                                        &image_factory_,
                                        compositor->widget(), "UICompositor");
+  scoped_refptr<InProcessContextProvider> worker_context_provider =
+      InProcessContextProvider::CreateOffscreen(&gpu_memory_buffer_manager_,
+                                                &image_factory_);
 
   scoped_ptr<cc::OutputSurface> real_output_surface;
 
   if (use_test_surface_) {
     bool flipped_output_surface = false;
     real_output_surface = make_scoped_ptr(new cc::PixelTestOutputSurface(
-        context_provider, flipped_output_surface));
+        context_provider, worker_context_provider, flipped_output_surface));
   } else {
-    real_output_surface =
-        make_scoped_ptr(new DirectOutputSurface(context_provider));
+    real_output_surface = make_scoped_ptr(
+        new DirectOutputSurface(context_provider, worker_context_provider));
   }
 
   if (surface_manager_) {
@@ -129,9 +134,9 @@ void InProcessContextFactory::CreateOutputSurface(
             GetSharedBitmapManager(), GetGpuMemoryBufferManager(),
             compositor->GetRendererSettings(), compositor->task_runner()));
     scoped_ptr<cc::SurfaceDisplayOutputSurface> surface_output_surface(
-        new cc::SurfaceDisplayOutputSurface(surface_manager_,
-                                            compositor->surface_id_allocator(),
-                                            context_provider));
+        new cc::SurfaceDisplayOutputSurface(
+            surface_manager_, compositor->surface_id_allocator(),
+            context_provider, worker_context_provider));
     display_client->set_surface_output_surface(surface_output_surface.get());
     surface_output_surface->set_display_client(display_client.get());
 
diff --git a/ui/compositor/test/in_process_context_provider.cc b/ui/compositor/test/in_process_context_provider.cc
index c5f950f..76cd19d 100644
--- a/ui/compositor/test/in_process_context_provider.cc
+++ b/ui/compositor/test/in_process_context_provider.cc
@@ -122,6 +122,10 @@ bool InProcessContextProvider::BindToCurrentThread() {
   return true;
 }
 
+void InProcessContextProvider::DetachFromThread() {
+  context_thread_checker_.DetachFromThread();
+}
+
 cc::ContextProvider::Capabilities
 InProcessContextProvider::ContextCapabilities() {
   DCHECK(context_thread_checker_.CalledOnValidThread());
diff --git a/ui/compositor/test/in_process_context_provider.h b/ui/compositor/test/in_process_context_provider.h
index bc3e494..0f5f35f 100644
--- a/ui/compositor/test/in_process_context_provider.h
+++ b/ui/compositor/test/in_process_context_provider.h
@@ -49,6 +49,7 @@ class InProcessContextProvider : public cc::ContextProvider {
 
   // cc::ContextProvider:
   bool BindToCurrentThread() override;
+  void DetachFromThread() override;
   Capabilities ContextCapabilities() override;
   gpu::gles2::GLES2Interface* ContextGL() override;
   gpu::ContextSupport* ContextSupport() override;