mojo/gpu: use SharedBuffer instead of base::SharedMemory with hacks

mojo/services/gles2/command_buffer.mojom

Index: mojo/services/gles2/command_buffer.mojom
diff --git a/mojo/services/gles2/command_buffer.mojom b/mojo/services/gles2/command_buffer.mojom
index e54a6e14e6b24f0ac1b40a39edc0554af240ed83..c10a9a378a970b6b8df0984b6c5d26d9501be6ab 100644
--- a/mojo/services/gles2/command_buffer.mojom
+++ b/mojo/services/gles2/command_buffer.mojom
@@ -14,12 +14,6 @@ struct CommandBufferState {
   uint32 generation;
 };
 
-// TODO(piman): we need to support proper SHM handles, or refactor command
-// buffers to sit on top of mojo primitives (e.g. DataPipe, etc.).
-struct ShmHandle {
-  uint64 handle_hack;
-};
-
 interface CommandBufferSyncClient {
   DidInitialize(bool success);
   DidMakeProgress(CommandBufferState state);
@@ -28,11 +22,12 @@ interface CommandBufferSyncClient {
 [Peer=CommandBufferClient]
 interface CommandBuffer {
   Initialize(CommandBufferSyncClient sync_client,
-                  ShmHandle shared_state);
+             handle<shared_buffer> shared_state);
   SetGetBuffer(int32 buffer);
   Flush(int32 put_offset);
   MakeProgress(int32 last_get_offset);
-  RegisterTransferBuffer(int32 id, ShmHandle transfer_buffer, uint32 size);
+  RegisterTransferBuffer(
+      int32 id, handle<shared_buffer> transfer_buffer, uint32 size);
   DestroyTransferBuffer(int32 id);
   Echo() => ();