zero-copy: Clarify to allocate/destroy GpuMemoryBuffer on any thread and use it on the main thread o

content/common/gpu/gpu_memory_buffer_factory_io_surface.cc

Index: content/common/gpu/gpu_memory_buffer_factory_io_surface.cc
diff --git a/content/common/gpu/gpu_memory_buffer_factory_io_surface.cc b/content/common/gpu/gpu_memory_buffer_factory_io_surface.cc
index 604c1b4927388f956c415939b023ce24b242932c..6d54be4c2ef55a7857e0f1788adcc36c7d40b759 100644
--- a/content/common/gpu/gpu_memory_buffer_factory_io_surface.cc
+++ b/content/common/gpu/gpu_memory_buffer_factory_io_surface.cc
@@ -159,7 +159,7 @@ GpuMemoryBufferFactoryIOSurface::CreateImageForGpuMemoryBuffer(
     unsigned internalformat,
     int client_id) {
   base::AutoLock lock(io_surfaces_lock_);
-
+  DCHECK(thread_checker_.CalledOnValidThread());

[reveman, 2015/04/07 13:07:05]

I don't think we want a thread checker here as that prevents this function from
being called on any thread.

[dshwang, 2015/04/07 14:04:26]

We allow CreateGpuMemoryBuffer() to be called on any thread, but we allow
CreateImageForGpuMemoryBuffer() to be called on only main thread because only
client API can create GLImage as following call stack

#2 0x7fe562138d61
content::GpuMemoryBufferFactoryIOSurface::CreateImageForGpuMemoryBuffer()
#3 0x7fe562bd608f content::GpuChannel::CreateImageForGpuMemoryBuffer()
#4 0x7fe562be06f2 content::GpuCommandBufferStub::OnCreateImage()
#5 0x7fe562be5ad5 content::GpuCommandBufferStub::OnMessageReceived()

[reveman, 2015/04/07 14:22:00]

Sorry I misread this change.

Yes, image creation is always happening on the same thread. However, there's
nothing about this code that restricts that. Wouldn't it make more sense to have
this thread checker in the GLImage implementations?  The GLImage implementations
with this restriction.

   DCHECK_EQ(handle.type, gfx::IO_SURFACE_BUFFER);
   IOSurfaceMapKey key(handle.id, client_id);
   IOSurfaceMap::iterator it = io_surfaces_.find(key);