Convert Gamepad IPC messages into mojo interface.

device/gamepad/public/interfaces/gamepad.mojom

Index: device/gamepad/public/interfaces/gamepad.mojom
diff --git a/device/gamepad/public/interfaces/gamepad.mojom b/device/gamepad/public/interfaces/gamepad.mojom
index 5fbd79f824f7edeb38b6a30801bdaad273c8f7e7..8632387038a3166bc3895590c0604d15eb8488e1 100644
--- a/device/gamepad/public/interfaces/gamepad.mojom
+++ b/device/gamepad/public/interfaces/gamepad.mojom
@@ -49,3 +49,22 @@ struct Gamepad {
   GamepadHand hand;
   uint32 display_id;
 };
+
+interface GamepadObserver {
+  GamepadConnected(int32 index, Gamepad gamepad);
+  GamepadDisconnected(int32 index, Gamepad gamepad);
+};
+
+// Asks the browser process to start polling, and return a shared memory
+// handles that will hold the data from the hardware. See

[Tom Sepez, 2016/12/06 17:53:40]

nit: handle that

[ke.he, 2016/12/07 08:05:10]

Done.

+// gamepad_hardware_buffer.h for a description of how synchronization is
+// handled. The number of Starts should match the number of Stops.

[Tom Sepez, 2016/12/06 17:53:40]

nit: or what happens if they don't?  Does the browser poll forever?  Do we clean
up on child termination?

[ke.he, 2016/12/07 08:05:10]

Hi, Tom Sepez, Thanks very much for considering this.
The GamepadService owns a std::Set of consumers. The renderer does
register/unregister consumer in GamepadService by calling
StartPolling/StopPolling. If GamepadService finds the std::Set<consumers> are
not empty, it will ask GamepadProvider to do Polling in the polling thread, to
notify consumers if any updates.
So if a renderer calls Starts, and doesn't call Stops, the polling thread will
keep polling, it means that renderer wants to received updates before it calls
Stops.
We do cleanup on child termination. Each render process "StrongBinds" a
GamepadMonitor(on behalf a consumer) instance in browser side, GamepadMonitor
will un-register in its destructor to avoid cause unnecessary polling.
And, mismatch of Starts and Stops may cause DCHECK() complains too.
 
Actually mojoifiction didn't change the logic of the register/unregister and the
polling. Comments here are copied from the removed
"content/common/gamepad_messages.h".

+interface GamepadMonitor {
+  [Sync]
+  GamepadStartPolling() => (handle<shared_buffer> memory_handle);
+
+  [Sync]
+  GamepadStopPolling() => ();
+
+  SetObserver(GamepadObserver gamepad_observer);
+};