Hi Hans, On Mon, Jun 24, 2013 at 03:40:14PM +0200, Hans Verkuil wrote: ... > > Events that fit to regular 64 bytes will be delivered using that, and the > > user-provided pointer will only be used in the case a large event is > > delivered to the user. So in order to be able to receive large events, the > > user must always provide the pointer even if it's not always used. > > This is an option. The easiest approach would be to extend v4l2_kevent with > a pointer to a struct v4l2_event_payload, which would be refcounted (it's > basically a struct kref + size + payload). Since this would have to be allocated > you can't use this in interrupt context. Well, one can use GFP_NOWAIT. I would allow this. But surely better done outside interrupt context. > Since the payloads are larger I am less concerned about speed. There is one > problem, though: if you dequeue the event and the buffer that should receive > the payload is too small, then you have lost that payload. You can't allocate > a new, larger, buffer and retry. So this approach can only work if you really > know the maximum payload size. > > The advantage is also that you won't lose payloads. > > You are starting to convince me :-) Just don't change the current implementation > for small payloads, that's one that really works very well. I meant only large payload buffers above but failed to write it down. Smaller ones could stay as they are. -- Kind regards, Sakari Ailus e-mail: sakari.ailus@xxxxxx XMPP: sailus@xxxxxxxxxxxxxx -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
