Chromium Code Reviews| Index: gpu/command_buffer/common/discardable_handle.cc | 
| diff --git a/gpu/command_buffer/common/discardable_handle.cc b/gpu/command_buffer/common/discardable_handle.cc | 
| new file mode 100644 | 
| index 0000000000000000000000000000000000000000..716e8aa08d49f9c34465a9dfaa3184df5ad9299b | 
| --- /dev/null | 
| +++ b/gpu/command_buffer/common/discardable_handle.cc | 
| @@ -0,0 +1,123 @@ | 
| +// Copyright (c) 2017 The Chromium Authors. All rights reserved. | 
| +// Use of this source code is governed by a BSD-style license that can be | 
| +// found in the LICENSE file. | 
| + | 
| +#include "gpu/command_buffer/common/discardable_handle.h" | 
| + | 
| +#include "base/atomicops.h" | 
| +#include "gpu/command_buffer/common/buffer.h" | 
| + | 
| +namespace gpu { | 
| +namespace { | 
| +const int32_t kHandleDeleted = 0; | 
| +const int32_t kHandleUnlocked = 1; | 
| +const int32_t kHandleLockedStart = 2; | 
| + | 
| +} // namespace | 
| + | 
| +DiscardableHandleBase::DiscardableHandleBase(scoped_refptr<Buffer> buffer, | 
| + uint32_t byte_offset, | 
| + int32_t shm_id) | 
| + : buffer_(std::move(buffer)), byte_offset_(byte_offset), shm_id_(shm_id) {} | 
| + | 
| +DiscardableHandleBase::DiscardableHandleBase( | 
| + const DiscardableHandleBase& other) = default; | 
| +DiscardableHandleBase::DiscardableHandleBase(DiscardableHandleBase&& other) = | 
| + default; | 
| +DiscardableHandleBase::~DiscardableHandleBase() = default; | 
| +DiscardableHandleBase& DiscardableHandleBase::operator=( | 
| + const DiscardableHandleBase& other) = default; | 
| +DiscardableHandleBase& DiscardableHandleBase::operator=( | 
| + DiscardableHandleBase&& other) = default; | 
| + | 
| +bool DiscardableHandleBase::IsLockedForTesting() { | 
| + return kHandleLockedStart <= base::subtle::NoBarrier_Load(AsAtomic()); | 
| +} | 
| + | 
| +bool DiscardableHandleBase::IsDeletedForTesting() { | 
| + return kHandleDeleted == base::subtle::NoBarrier_Load(AsAtomic()); | 
| +} | 
| + | 
| +volatile base::subtle::Atomic32* DiscardableHandleBase::AsAtomic() const { | 
| + return reinterpret_cast<volatile base::subtle::Atomic32*>( | 
| + buffer_->GetDataAddress(byte_offset_, sizeof(base::subtle::Atomic32))); | 
| +} | 
| + | 
| +ClientDiscardableHandle::ClientDiscardableHandle(scoped_refptr<Buffer> buffer, | 
| + uint32_t byte_offset, | 
| + int32_t shm_id) | 
| + : DiscardableHandleBase(std::move(buffer), byte_offset, shm_id) { | 
| + // Handle always starts locked. | 
| + base::subtle::NoBarrier_Store(AsAtomic(), kHandleLockedStart); | 
| +} | 
| + | 
| +ClientDiscardableHandle::ClientDiscardableHandle( | 
| + const ClientDiscardableHandle& other) = default; | 
| +ClientDiscardableHandle::ClientDiscardableHandle( | 
| + ClientDiscardableHandle&& other) = default; | 
| +ClientDiscardableHandle& ClientDiscardableHandle::operator=( | 
| + const ClientDiscardableHandle& other) = default; | 
| +ClientDiscardableHandle& ClientDiscardableHandle::operator=( | 
| + ClientDiscardableHandle&& other) = default; | 
| + | 
| +bool ClientDiscardableHandle::Lock() { | 
| + while (true) { | 
| + base::subtle::Atomic32 current_value = | 
| + base::subtle::NoBarrier_Load(AsAtomic()); | 
| + if (current_value == kHandleDeleted) { | 
| + // Once a handle is deleted, it cannot be modified further. | 
| + return false; | 
| + } | 
| + base::subtle::Atomic32 new_value = current_value + 1; | 
| + // No barrier is needed, as any commands which depend on this operation | 
| + // will flow over the command buffer, which ensures a memory barrier | 
| + // between here and where these commands are executed on the GPU process. | 
| + base::subtle::Atomic32 previous_value = | 
| + base::subtle::NoBarrier_CompareAndSwap(AsAtomic(), current_value, | 
| + new_value); | 
| + if (current_value == previous_value) { | 
| + return true; | 
| + } | 
| + } | 
| +} | 
| + | 
| +bool ClientDiscardableHandle::CanBeReUsed() const { | 
| + return kHandleDeleted == base::subtle::Acquire_Load(AsAtomic()); | 
| +} | 
| + | 
| +ServiceDiscardableHandle::ServiceDiscardableHandle(scoped_refptr<Buffer> buffer, | 
| + uint32_t byte_offset, | 
| + int32_t shm_id) | 
| + : DiscardableHandleBase(std::move(buffer), byte_offset, shm_id) { | 
| + // Newly created service handles always start locked, as the client has no way | 
| + // to unlock them. | 
| + DCHECK(base::subtle::NoBarrier_Load(AsAtomic()) >= kHandleLockedStart); | 
| 
 
piman
2017/03/28 00:43:00
ditto on this DCHECK, we need to make sure we beha
 
ericrk
2017/03/28 01:12:10
same problems/mitigations as listed on the previou
 
 | 
| +} | 
| + | 
| +ServiceDiscardableHandle::ServiceDiscardableHandle( | 
| + const ServiceDiscardableHandle& other) = default; | 
| +ServiceDiscardableHandle::ServiceDiscardableHandle( | 
| + ServiceDiscardableHandle&& other) = default; | 
| +ServiceDiscardableHandle& ServiceDiscardableHandle::operator=( | 
| + const ServiceDiscardableHandle& other) = default; | 
| +ServiceDiscardableHandle& ServiceDiscardableHandle::operator=( | 
| + ServiceDiscardableHandle&& other) = default; | 
| + | 
| +void ServiceDiscardableHandle::Unlock() { | 
| + // No barrier is needed as all GPU process access happens on a single thread, | 
| + // and communication of dependent data between the GPU process and the | 
| + // renderer process happens across the command buffer and includes barriers. | 
| + base::subtle::Atomic32 value = | 
| + base::subtle::NoBarrier_AtomicIncrement(AsAtomic(), -1); | 
| + DCHECK(value >= kHandleLockedStart); | 
| 
 
piman
2017/03/28 00:43:00
ditto
 
ericrk
2017/03/28 01:12:10
same problems/mitigations as listed on the previou
 
 | 
| +} | 
| + | 
| +bool ServiceDiscardableHandle::Delete() { | 
| + // No barrier is needed as all GPU process access happens on a single thread, | 
| + // and communication of dependent data between the GPU process and the | 
| + // renderer process happens across the command buffer and includes barriers. | 
| + return kHandleUnlocked == base::subtle::NoBarrier_CompareAndSwap( | 
| + AsAtomic(), kHandleUnlocked, kHandleDeleted); | 
| 
 
piman
2017/03/28 00:43:00
nit: indent looks wrong
 
ericrk
2017/03/28 01:12:10
oddly enough, this is what git cl format produces
 
piman
2017/03/28 21:47:25
Seems like a clang-format bug, but let's not fight
 
 | 
| +} | 
| + | 
| +} // namespace gpu |