Re: [PATCH v4 3/3] media: uvcvideo: Add a unique suffix to camera names

[Mauro Carvalho Chehab]

Em Thu, 22 Dec 2022 00:24:44 +0200
Laurent Pinchart <laurent.pinchart@xxxxxxxxxxxxxxxx> escreveu:

> Hi Mauro,
> 
> On Wed, Dec 21, 2022 at 10:02:48PM +0000, Mauro Carvalho Chehab wrote:
> > Em Tue, 6 Dec 2022 00:02:22 +0100 Ricardo Ribalda escreveu:  
> > > On Mon, 5 Dec 2022 at 23:16, Laurent Pinchart wrote:  
> > > > On Fri, Dec 02, 2022 at 06:08:19PM +0100, Ricardo Ribalda wrote:    
> > > > > Some cameras have multiple data inputs (i.e. IR sensor and RGB sensor),
> > > > > append a unique number to the device name.
> > > > >
> > > > > Fixes v4l2-compliance:
> > > > >     Media Controller ioctls:
> > > > >          fail: v4l2-test-media.cpp(205): v2_entity_names_set.find(key) != v2_entity_names_set.end()
> > > > >        test MEDIA_IOC_G_TOPOLOGY: FAIL
> > > > >          fail: v4l2-test-media.cpp(394): num_data_links != num_links
> > > > >        test MEDIA_IOC_ENUM_ENTITIES/LINKS: FAIL
> > > > >
> > > > > Signed-off-by: Ricardo Ribalda <ribalda@xxxxxxxxxxxx>
> > > > > ---
> > > > >  drivers/media/usb/uvc/uvc_driver.c | 3 ++-
> > > > >  1 file changed, 2 insertions(+), 1 deletion(-)
> > > > >
> > > > > diff --git a/drivers/media/usb/uvc/uvc_driver.c b/drivers/media/usb/uvc/uvc_driver.c
> > > > > index 215fb483efb0..f4032ebb3689 100644
> > > > > --- a/drivers/media/usb/uvc/uvc_driver.c
> > > > > +++ b/drivers/media/usb/uvc/uvc_driver.c
> > > > > @@ -1963,7 +1963,8 @@ int uvc_register_video_device(struct uvc_device *dev,
> > > > >               break;
> > > > >       }
> > > > >
> > > > > -     strscpy(vdev->name, dev->name, sizeof(vdev->name));
> > > > > +     snprintf(vdev->name, sizeof(vdev->name), "%s %u", dev->name,
> > > > > +              stream->header.bTerminalLink);    
> > > >
> > > > This won't be perfect as the string is not guaranteed to fit in
> > > > vdev->name, but I suppose it will help as a quick fix for some devices.
> > > > How about the other devices ? Won't they still exhibit the above
> > > > v4l2-compliance failure ? Isn't that something that will still affect
> > > > Chrome OS devices ?    
> > > 
> > > We could place the id first... but that will look bad: Eg:
> > > 
> > > 1- My favorite camera
> > > 
> > > Another option is to remove the last chars to fit the id. Eg:
> > > 
> > > My favorite came-1
> > > 
> > > If you prefer any of those options or have a better idea I can implement that.
> > >   
> > > > The change should not cause any regression as big as in patch 1/3.
> > > > However, unless I'm mistaken users will notice a device name change,
> > > > especially when selecting a device in their web browser. Could that be a
> > > > problem ?    
> > > 
> > > I think the only side effect is that the first time that the kernel
> > > changes the naming convention, if there are more than one camera on
> > > the system, the video conference might pick a different camera.
> > > The good news is that the user will be presented with cameras with
> > > different names. Now some cameras show very confusing names:
> > > 
> > > ribalda@alco:~/work/linux$ for a in /dev/video* ; do yavta -l $a| grep
> > > "Dell Webcam"; done
> > > Device `Dell Webcam WB7022' on `usb-0000:2d:00.0-1.2.3.1' (driver
> > > 'uvcvideo') supports video, capture, without mplanes.
> > > Device `Dell Webcam WB7022' on `usb-0000:2d:00.0-1.2.3.1' (driver
> > > 'uvcvideo') supports meta-data, capture, without mplanes.
> > > Device `Dell Webcam WB7022' on `usb-0000:2d:00.0-1.2.3.1' (driver
> > > 'uvcvideo') supports video, capture, without mplanes.
> > > Device `Dell Webcam WB7022' on `usb-0000:2d:00.0-1.2.3.1' (driver
> > > 'uvcvideo') supports meta-data, capture, without mplanes.  
> > 
> > That is bad, as some apps like camorama actually store the user
> > preferences (last used resolution and fps), preserving them the next
> > time the camera with the same name is used.
> > 
> > In the specific case of camorama, it uses the by-id filename, which is
> > derived from the name, as stored at:
> > 
> > 	$ ls -la /dev/v4l/by-id/usb-Quanta_HD_Camera_0001-video-index*
> > 	lrwxrwxrwx 1 root root 12 dez 21 09:59 /dev/v4l/by-id/usb-Quanta_HD_Camera_0001-video-index0 -> ../../video0
> > 	lrwxrwxrwx 1 root root 12 dez 21 09:59 /dev/v4l/by-id/usb-Quanta_HD_Camera_0001-video-index1 -> ../../video1
> > 
> > With this change, not only the camera name will become bigger (and may
> > end losing the index0/index1 data), but it will also lost the stored 
> > preferences on apps like camorama, causing regressions.
> > 
> > It sounds a lot easier to teach udev to change the name on an unique
> > way, on machines where you need it.  
> 
> It's not a udev issue, it's an API compliance problem, entity names in a
> media controller graph must be unique, and they are not.

Then the right fix would be inside the media controller naming
logic, in a way that it will ensure unique names. It sounds to me
that mc core should then check, at device's register time, if the
name was already used. If so, change it to be unique.

> Losing the stored preference is possibly inconvenient in some cases, but
> I don't think it's a real blocker.

If it causes userspace regressions, then it is a blocker.

Regards,
Mauro