Hi,
hverkuil wrote:
> > + if (sev->ops && sev->ops->add) {
> > + int ret = sev->ops->add(sev);
> > + if (ret) {
> > + sev->ops = NULL;
> > + v4l2_event_unsubscribe(fh, sub);
> > + return ret;
> > + }
The problem here is that the event is basically available for use after the
spin_unlock_irqrestore(), but before the sev->ops->add() call. In the past I
just 'knew' that the event would never be generated by the control framework
until after v4l2_ctrl_add_event was called, but this should be formalized now
that these ops are added.
I see two options:
1) Have some method to mark the sev as being 'invalid' so functions sending
events can skip it (needed as well for your patch 4/6).
2) Document that drivers should never be able to send an event until after
the add() callback has succeeded.
I am leaning towards option 1 myself.
How about leaving sev->elems at 0 until after the add() op succeeds?
That's easy to test against and easy to implement.
I agree that option 1 is the best and that leaving sev->elems 0 is a good
way to do this. This will be in my next revision of this patch set.
Thanks & Regards,
Hans
--
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
