[v3.8-rc1] Multimedia regression, ioctl(17,..)-API changed ?

[v3.8-rc1] Multimedia regression, ioctl(17,..)-API changed ?

[Jörg Otte]

With kernel v3.8 all multimedia programs under KDE4 don't work (Kubuntu 12.04).
They alltogether ( at least Dragonplayer (Mediaplayer), Knotify4
(system-sound),
System-Settings-Multimedia,..) are looping forever producing 100% CPU-usage
and must be killed.

With kernel 3.7 there are no problems.

I compared an strace of Dragonplayer under 3.7 and 3.8 kernels. The
main difference
of both traces are the following corresponding outputs just before
looping in v3.8
begins:

v3.7:
ioctl(17, VIDIOC_ENUMSTD, 0x7fff6cce66a0) = -1 EINVAL (Invalid argument)
ioctl(17, VIDIOC_QUERYCTRL, 0x7fff6cce66f0) = -1 EINVAL (Invalid argument)

v3.8:
ioctl(17, VIDIOC_ENUMSTD, 0x7fffc3be6990) = -1 ENOTTY (Inappropriate
ioctl for device)
ioctl(17, VIDIOC_QUERYCTRL, 0x7fffc3be69e0) = -1 ENOENT (No such file
or directory)

So error number EINVAL was changed to ENOTTY/ENOENT

When Dragonplayer under v3.8 comes to ioctl(17, VIDIOC_QUERYCTRL,...)
and sees error
number ENOENT instead of EINVAL it loops forever producing 100% CPU
usage like so:

  .
  .
ioctl(17, VIDIOC_QUERYCTRL, 0x7fffc3be69e0) = -1 ENOENT (No such file
or directory)
ioctl(17, VIDIOC_QUERYCTRL, 0x7fffc3be69e0) = -1 ENOENT (No such file
or directory)
ioctl(17, VIDIOC_QUERYCTRL, 0x7fffc3be69e0) = -1 ENOENT (No such file
or directory)
ioctl(17, VIDIOC_QUERYCTRL, 0x7fffc3be69e0) = -1 ENOENT (No such file
or directory)
ioctl(17, VIDIOC_QUERYCTRL, 0x7fffc3be69e0) = -1 ENOENT (No such file
or directory)
ioctl(17, VIDIOC_QUERYCTRL, 0x7fffc3be69e0) = -1 ENOENT (No such file
or directory)
ioctl(17, VIDIOC_QUERYCTRL, 0x7fffc3be69e0) = -1 ENOENT (No such file
or directory)
ioctl(17, VIDIOC_QUERYCTRL, 0x7fffc3be69e0) = -1 ENOENT (No such file
or directory)
ioctl(17, VIDIOC_QUERYCTRL, 0x7fffc3be69e0) = -1 ENOENT (No such file
or directory)
ioctl(17, VIDIOC_QUERYCTRL, 0x7fffc3be69e0) = -1 ENOENT (No such file
or directory)
ioctl(17, VIDIOC_QUERYCTRL, 0x7fffc3be69e0) = -1 ENOENT (No such file
or directory)
  .
  and so on
  .

For me it looks like that KDE4 multimedia is not aware of the new error numbers.

Looking through the commits I found driver uvcvideo producing the changed
error numbers.

commit f0ed2ce840b3a59b587e8aa398538141a86e9588
Author: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
[media] uvcvideo: Set error_idx properly for extended controls API failures

To verify this I built a v3.8-kernel without uvcvideo (USB_VIDEO_CLASS=n)
and the problem disappeared!

Simply reverting the commit is not an option for me because then I am left
with merge conflicts and I don't know how to resolve.

Unfortunately without uvcvideo I lost my usb-camera support.

-- Jörg

Please CC me I'm not subscribed

Re: [v3.8-rc1] Multimedia regression, ioctl(17,..)-API changed ?

[Linus Torvalds]

Jörg - does current git work for you? It has a patch from Rafael that
just reverts the insane error code, and fixed something very similar
for him.

(I just pushed out, so it might take a few minutes to mirror out to
the public sites).

              Linus

On Sun, Dec 23, 2012 at 8:46 AM, Jörg Otte <jrg.otte@gmail.com> wrote:
> With kernel v3.8 all multimedia programs under KDE4 don't work (Kubuntu 12.04).
> They alltogether ( at least Dragonplayer (Mediaplayer), Knotify4
> (system-sound),
> System-Settings-Multimedia,..) are looping forever producing 100% CPU-usage
> and must be killed.

Re: [v3.8-rc1] Multimedia regression, ioctl(17,..)-API changed ?

[Mauro Carvalho Chehab]

Hi Jörg,

Em Sun, 23 Dec 2012 17:46:07 +0100
Jörg Otte <jrg.otte@gmail.com> escreveu:

> With kernel v3.8 all multimedia programs under KDE4 don't work (Kubuntu 12.04).
> They alltogether ( at least Dragonplayer (Mediaplayer), Knotify4
> (system-sound),
> System-Settings-Multimedia,..) are looping forever producing 100% CPU-usage
> and must be killed.
> 
> With kernel 3.7 there are no problems.

Do you have any other non-uvc device to test?

> I compared an strace of Dragonplayer under 3.7 and 3.8 kernels. The
> main difference
> of both traces are the following corresponding outputs just before
> looping in v3.8
> begins:
> 
> v3.7:
> ioctl(17, VIDIOC_ENUMSTD, 0x7fff6cce66a0) = -1 EINVAL (Invalid argument)

This ioctl returns -ENOTTY already with other media drivers on v3.7.

> ioctl(17, VIDIOC_QUERYCTRL, 0x7fff6cce66f0) = -1 EINVAL (Invalid argument)

This change is new, and, AFAIKT, only UVC returns -ENOENT on 3.8-rc1.

This is likely the source of the troubles.

> 
> v3.8:
> ioctl(17, VIDIOC_ENUMSTD, 0x7fffc3be6990) = -1 ENOTTY (Inappropriate
> ioctl for device)
> ioctl(17, VIDIOC_QUERYCTRL, 0x7fffc3be69e0) = -1 ENOENT (No such file
> or directory)
> 
> So error number EINVAL was changed to ENOTTY/ENOENT
> 
> When Dragonplayer under v3.8 comes to ioctl(17, VIDIOC_QUERYCTRL,...)
> and sees error
> number ENOENT instead of EINVAL it loops forever producing 100% CPU
> usage like so:
> 
>   .
>   .
> ioctl(17, VIDIOC_QUERYCTRL, 0x7fffc3be69e0) = -1 ENOENT (No such file
> or directory)
> ioctl(17, VIDIOC_QUERYCTRL, 0x7fffc3be69e0) = -1 ENOENT (No such file
> or directory)
> ioctl(17, VIDIOC_QUERYCTRL, 0x7fffc3be69e0) = -1 ENOENT (No such file
> or directory)
> ioctl(17, VIDIOC_QUERYCTRL, 0x7fffc3be69e0) = -1 ENOENT (No such file
> or directory)
> ioctl(17, VIDIOC_QUERYCTRL, 0x7fffc3be69e0) = -1 ENOENT (No such file
> or directory)
> ioctl(17, VIDIOC_QUERYCTRL, 0x7fffc3be69e0) = -1 ENOENT (No such file
> or directory)
> ioctl(17, VIDIOC_QUERYCTRL, 0x7fffc3be69e0) = -1 ENOENT (No such file
> or directory)
> ioctl(17, VIDIOC_QUERYCTRL, 0x7fffc3be69e0) = -1 ENOENT (No such file
> or directory)
> ioctl(17, VIDIOC_QUERYCTRL, 0x7fffc3be69e0) = -1 ENOENT (No such file
> or directory)
> ioctl(17, VIDIOC_QUERYCTRL, 0x7fffc3be69e0) = -1 ENOENT (No such file
> or directory)
> ioctl(17, VIDIOC_QUERYCTRL, 0x7fffc3be69e0) = -1 ENOENT (No such file
> or directory)

Yeah, it started an endless loop here, likely because kde4 seems to be
expecting either 0 or -EINVAL error code for VIDIOC_QUERYCTRL.

It should be noticed that there are other valid error codes here. For
example, if this ioctl is not implemented, -ENOTTY may also be returned.
Fortunately, almost all drivers do implement this ioctl.

The expected return error codes for this ioctl are described at:
	http://linuxtv.org/downloads/v4l-dvb-apis/vidioc-queryctrl.html

In practice, except for the uvc driver, the current return codes are
EINVAL/EACCES/ENOTTY.

>   .
>   and so on
>   .
> 
> For me it looks like that KDE4 multimedia is not aware of the new error numbers.
> 
> Looking through the commits I found driver uvcvideo producing the changed
> error numbers.
> 
> commit f0ed2ce840b3a59b587e8aa398538141a86e9588
> Author: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
> [media] uvcvideo: Set error_idx properly for extended controls API failures
> 
> To verify this I built a v3.8-kernel without uvcvideo (USB_VIDEO_CLASS=n)
> and the problem disappeared!
> 
> Simply reverting the commit is not an option for me because then I am left
> with merge conflicts and I don't know how to resolve.
> 
> Unfortunately without uvcvideo I lost my usb-camera support.

Rafael made a patch fixing it, and Linus should be applying it.

Regards,
Mauro

Re: [v3.8-rc1] Multimedia regression, ioctl(17,..)-API changed ?

[Jörg Otte]

Yes it works, incl. usb-camera. I am now at:

637704cbc95c02d18741b4a6e7a5d2397f8b28ce Merge branch
 'i2c-embedded/for-next' of git://git.pengutronix.de/git/wsa/linux

Thanks, Jörg


2012/12/23 Linus Torvalds <torvalds@linux-foundation.org>:
> Jörg - does current git work for you? It has a patch from Rafael that
> just reverts the insane error code, and fixed something very similar
> for him.
>
> (I just pushed out, so it might take a few minutes to mirror out to
> the public sites).
>
>               Linus
>
> On Sun, Dec 23, 2012 at 8:46 AM, Jörg Otte <jrg.otte@gmail.com> wrote:
>> With kernel v3.8 all multimedia programs under KDE4 don't work (Kubuntu 12.04).
>> They alltogether ( at least Dragonplayer (Mediaplayer), Knotify4
>> (system-sound),
>> System-Settings-Multimedia,..) are looping forever producing 100% CPU-usage
>> and must be killed.

Re: [v3.8-rc1] Multimedia regression, ioctl(17,..)-API changed ?

[Jörg Otte]

2012/12/23 Mauro Carvalho Chehab <mchehab@redhat.com>:
> Hi Jörg,
>
> Em Sun, 23 Dec 2012 17:46:07 +0100
> Jörg Otte <jrg.otte@gmail.com> escreveu:
>
>> With kernel v3.8 all multimedia programs under KDE4 don't work (Kubuntu 12.04).
>> They alltogether ( at least Dragonplayer (Mediaplayer), Knotify4
>> (system-sound),
>> System-Settings-Multimedia,..) are looping forever producing 100% CPU-usage
>> and must be killed.
>>
>> With kernel 3.7 there are no problems.
>
> Do you have any other non-uvc device to test?

No I haven't.

Thanks, Jörg