Em Tue, 02 Sep 2014 03:16:00 +0000 (GMT)
Changbing Xiong <cb.xiong@xxxxxxxxxxx> escreveu:
>
> > Well, we may start returning -ENODEV when such event happens.
>
> > At the frontend, we could use fe->exit = DVB_FE_DEVICE_REMOVED to
> > signalize it. I don't think that the demod frontend has something
> > similar.
>
> > Yet, it should be up to the userspace application to properly handle
> > the error codes and close the devices on fatal non-recovery errors like
> > ENODEV.
>
> > So, what we can do, at Kernel level, is to always return -ENODEV when
> > the device is known to be removed, and double check libdvbv5 if it
> > handles such error properly.
>
> well, we do not use libdvbv5,
The upstream stuff I maintain, related to it, are the media subsystems
and libdvbv5. Of course, other apps will need to be patched as well.
> and -ENODEV can be returned by read syscall,
> but for poll syscall,
Actually, poll() may return an error as well (from poll() manpage):
"RETURN VALUE
On success, a positive number is returned; this is the number of struc‐
tures which have nonzero revents fields (in other words, those descrip‐
tors with events or errors reported). A value of 0 indicates that the
call timed out and no file descriptors were ready. On error, -1 is
returned, and errno is set appropriately."
So, if the Kernel returns -ENODEV, the glibc poll() wrapper would return -1
and errno will be ENODEV. Never actually tested if this works on poll()
though.
> -ENODEV can never be returned to user, as negative number
> is invalid type for poll returned value. please refer to my second patch.
>
> and in our usage, whether to read the device is up to the poll result. if tuner is plugged out,
> and there is no data in dvr ringbuffer. then user code will still go on polling the dvr device and never stop.
> if POLLERR is returned, then user will perform read dvr, and then -ENODEV can be got, and
> user will stop polling dvr device.
Your app should be also be handling poll() errors, as there are already
other errors that poll() can return.
> the first patch is enough to fix the deadlock issue.
> the second patch is used to correct the wrong type of returned value.
> the third patch is used to provide user a better controlling logic.
I'll take a deeper look and do some tests on your patches likely
tomorrow.
Regards,
Mauro
--
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
