Add NativePixmapHandle type & interface for exporting them

content/common/gpu/client/gpu_memory_buffer_impl_ozone_native_pixmap.cc

Index: content/common/gpu/client/gpu_memory_buffer_impl_ozone_native_pixmap.cc
diff --git a/content/common/gpu/client/gpu_memory_buffer_impl_ozone_native_pixmap.cc b/content/common/gpu/client/gpu_memory_buffer_impl_ozone_native_pixmap.cc
index 7acf59c8e51527215405b8ee7a18c12c2f90160c..40500a604f8a10ed40790cccb67a61348d455023 100644
--- a/content/common/gpu/client/gpu_memory_buffer_impl_ozone_native_pixmap.cc
+++ b/content/common/gpu/client/gpu_memory_buffer_impl_ozone_native_pixmap.cc
@@ -4,6 +4,7 @@
 
 #include "content/common/gpu/client/gpu_memory_buffer_impl_ozone_native_pixmap.h"
 
+#include "ui/ozone/public/client_native_pixmap_factory.h"
 #include "ui/ozone/public/surface_factory_ozone.h"
 
 namespace content {
@@ -12,8 +13,9 @@ GpuMemoryBufferImplOzoneNativePixmap::GpuMemoryBufferImplOzoneNativePixmap(
     gfx::GpuMemoryBufferId id,
     const gfx::Size& size,
     gfx::BufferFormat format,
-    const DestructionCallback& callback)
-    : GpuMemoryBufferImpl(id, size, format, callback) {}
+    const DestructionCallback& callback,
+    scoped_ptr<ui::ClientNativePixmap> pixmap)
+    : GpuMemoryBufferImpl(id, size, format, callback), pixmap_(pixmap.Pass()) {}
 
 GpuMemoryBufferImplOzoneNativePixmap::~GpuMemoryBufferImplOzoneNativePixmap() {}
 
@@ -25,22 +27,24 @@ GpuMemoryBufferImplOzoneNativePixmap::CreateFromHandle(
     gfx::BufferFormat format,
     gfx::BufferUsage usage,
     const DestructionCallback& callback) {
+  scoped_ptr<ui::ClientNativePixmap> native_pixmap =
+      ui::ClientNativePixmapFactory::GetInstance()->ImportFromHandle(

[no sievers, 2015/08/11 21:16:33]

Can this fail?

Below we are accessing |pixmap_| without null-checks, so maybe here is a good
place to check for null and return early?

[spang, 2015/08/11 22:03:31]

Done. However since cc does not check for allocation failure, it will still
crash later.

[reveman, 2015/08/12 08:20:29]

cc handles GMB allocation failure properly as it can happen during lost context
situations. However, I expect ImportFromHandle to not be affected by this as
after the browser has successfully allocating a buffer and returned a valid
handle there's no good reason for ImportFromHandle to fail, or is there? What
would be the reason for this to fail? I'd rather see a DCHECK or CHECK if it's
not something we expect to happen. 

[dshwang, 2015/08/12 08:28:09]

I think null check is not needed, as well as nullptr is possible in SCANOUT
case.
Until now, this class works well in SCANOUT case without |pixmap|. Because
Map/Unmap/Stride must not be called in SCANOUT case.
The next CL (https://codereview.chromium.org/1134993003/) will not create
|pixmap| in SCANOUT case as-is.
spang, reveman, Do I think same to you?

[dshwang, 2015/08/12 08:41:43]

I understand that ClientNativePixmapGbm is always created, so now I change my
mind. It's good to add DCHECK(native_pixmap) as sievers said.

+          handle.native_pixmap_handle, size, format, usage);
   return make_scoped_ptr<GpuMemoryBufferImpl>(
       new GpuMemoryBufferImplOzoneNativePixmap(handle.id, size, format,
-                                               callback));
+                                               callback, native_pixmap.Pass()));
 }
 
 bool GpuMemoryBufferImplOzoneNativePixmap::Map(void** data) {
-  NOTREACHED();
-  return false;
+  return pixmap_->Map(data);
 }
 
 void GpuMemoryBufferImplOzoneNativePixmap::Unmap() {
-  NOTREACHED();
+  pixmap_->Unmap();
 }
 
 void GpuMemoryBufferImplOzoneNativePixmap::GetStride(int* stride) const {
-  NOTREACHED();
+  pixmap_->GetStride(stride);
 }
 
 gfx::GpuMemoryBufferHandle GpuMemoryBufferImplOzoneNativePixmap::GetHandle()