Plumb GpuMemoryBuffer allocation to GPU process.

content/common/gpu/client/gpu_channel_host.h

Index: content/common/gpu/client/gpu_channel_host.h
diff --git a/content/common/gpu/client/gpu_channel_host.h b/content/common/gpu/client/gpu_channel_host.h
index 4f5b74f010d3ad22d4ee963a1751d5cf4ace05b1..2de8065dae17b4fe161312a3610969c16ae5a512 100644
--- a/content/common/gpu/client/gpu_channel_host.h
+++ b/content/common/gpu/client/gpu_channel_host.h
@@ -84,6 +84,24 @@ class CONTENT_EXPORT GpuChannelHostFactory {
       unsigned usage) = 0;
 };
 
+class CONTENT_EXPORT GpuChannelHostGpuBufferFactory {

[reveman, 2014/06/02 22:47:33]

We need a better name for this and I think it's probably best that we move it to
a separate file. We want to make it as easy as possible to move everything but
the implementation of this interface out of content/ in the future.

How about GpuMemoryBufferFactoryHost?

+ public:
+  typedef base::Callback<void(const gfx::GpuMemoryBufferHandle& handle)>
+      CreateGpuMemoryBufferCallback;
+
+  virtual ~GpuChannelHostGpuBufferFactory() {}

[reveman, 2014/06/02 22:47:33]

The dtor doesn't have to be part of the interface. Please make this protected as
that prevents the client from being able to delete it and removes the need for
CONTENT_EXPORT above.

+
+  virtual void CreateGpuMemoryBuffer(
+      size_t width,
+      size_t height,
+      unsigned internalformat,
+      unsigned usage,
+      const gfx::GpuMemoryBufferHandle& handle,

[reveman, 2014/06/02 22:47:33]

nit: maybe have the handle be first parameter. it's in some sense the most
important.

+      const CreateGpuMemoryBufferCallback& callback) = 0;
+  virtual void DeleteGpuMemoryBuffer(const gfx::GpuMemoryBufferHandle& handle,

[reveman, 2014/06/02 22:47:33]

Hm, should this be "Delete" or "Destroy"? Not sure why I chose Delete in case of
Create/DeleteImage but I feel like that was a mistake. Create/Destroy seem more
appropriate.

+                                     int32 sync_point) = 0;
+};
+
 // Encapsulates an IPC channel between the client and one GPU process.
 // On the GPU process side there's a corresponding GpuChannel.
 // Every method can be called on any thread with a message loop, except for the