Tom Zoerner <tomzo@xxxxxxxxxx> writes:
> While debugging one of my apps I noticed that the saa7134 driver makes
> the VBI device readable in select() and returns EIO for ioctl DQBUF
> when there's no signal, i.e. when the card appearently doesn't generate
> an interrupt for a long time.
So does bttv 0.9.x btw, and it may happen for other reasons than "no
signal" too.
> Not sure what this is meant to be used for since users can implement
> a timeout if needed with select(), but anyways, the matter I'm
> worrying about is that DQBUF in this case does not simply abort, but
> actually does de-queue a buffer although it returns an error code.
> Is this a bug or a feature?
Its done intentionally.
> If it's a feature I don't understand how the user is supposed to deal
> with it, since video_usercopy() sees the error code returned by the
> ioctl and hence does not copy struct v4l2_buffer into user space, so
> the user doesn't even obtain the buffer index and cannot re-queue it.
Whoops. Good point. Maybe better use something like
V4L2_BUF_FLAG_IOERROR instead?
Gerd
PS: the event stuff is still sitting in my inbox, waiting for me
finding some time to look at it in detail ...
--
sigfault
