Re: [PATCH] media: usb: uvc: fill in description for unknown pixelformats

[Laurent Pinchart <laurent.pinchart@ideasonboard.com>]

Hi Hans,

On Wed, Apr 05, 2023 at 09:40:20AM +0300, Laurent Pinchart wrote:
> On Wed, Mar 29, 2023 at 07:20:59PM +0200, Hans Verkuil wrote:
> > On 29/03/2023 18:05, Ricardo Ribalda wrote:
> > > Hi Hans
> > > 
> > > Thanks for the patch.
> > > 
> > > I believe the user can fetch this info from lsusb, so this is kind of
> > > duplicated info, and this is why it was removed.
> > 
> > You got to set some description, so using the GUID this seems best.
> > 
> > > Is there an app that uses this unknown format code ? Or the only
> > > complaint is that WARN() is too loud for the user?
> > 
> > Normally drivers do not pass on unknown formats, but if a driver does,
> > then I want a WARN. If a driver does this legitimately (and I understand
> > that's the case for UVC), then the driver should fill in the description
> > to avoid this WARN.
> 
> In hindsight we shouldn't have added a text description to formats :-)
> 
> > > On Wed, 29 Mar 2023 at 14:39, Hans Verkuil wrote:
> > >>
> > >> If the fcc is 0 (indicating an unknown GUID format), then fill in the
> > >> description field in ENUM_FMT. Otherwise the V4L2 core will WARN.
> > >>
> > >> Signed-off-by: Hans Verkuil <hverkuil-cisco@xs4all.nl>
> > >> Fixes: 50459f103edf ("media: uvcvideo: Remove format descriptions")
> > >> ---
> > >> diff --git a/drivers/media/usb/uvc/uvc_driver.c b/drivers/media/usb/uvc/uvc_driver.c
> > >> index 7aefa76a42b3..2f1ced1212cd 100644
> > >> --- a/drivers/media/usb/uvc/uvc_driver.c
> > >> +++ b/drivers/media/usb/uvc/uvc_driver.c
> > >> @@ -256,6 +256,9 @@ static int uvc_parse_format(struct uvc_device *dev,
> > >>                 } else {
> > >>                         dev_info(&streaming->intf->dev,
> > >>                                  "Unknown video format %pUl\n", &buffer[5]);
> > >> +                       snprintf(format->name, sizeof(format->name), "%pUl\n",
> > >> +                                &buffer[5]);
> > > Don't we need at least 38 chars for this?
> > 
> > Yes. But all we have is 31 chars, so we take what we can :-)
> > 
> > This is what uvc did before this was removed.
> > 
> > Regards,
> > 
> > 	Hans
> > 
> > > https://docs.kernel.org/core-api/printk-formats.html#uuid-guid-addresses
> > > 
> > >> +
> > >>                         format->fcc = 0;
> > >>                 }
> > >>
> > >> diff --git a/drivers/media/usb/uvc/uvc_v4l2.c b/drivers/media/usb/uvc/uvc_v4l2.c
> > >> index 35453f81c1d9..fc6f9e7d8506 100644
> > >> --- a/drivers/media/usb/uvc/uvc_v4l2.c
> > >> +++ b/drivers/media/usb/uvc/uvc_v4l2.c
> > >> @@ -713,6 +713,10 @@ static int uvc_ioctl_enum_fmt(struct uvc_streaming *stream,
> > >>         if (format->flags & UVC_FMT_FLAG_COMPRESSED)
> > >>                 fmt->flags |= V4L2_FMT_FLAG_COMPRESSED;
> > >>         fmt->pixelformat = format->fcc;
> > >> +       if (format->name[0])
> > >> +               strscpy(fmt->description, format->name,
> > >> +                       sizeof(fmt->description));
> > >> +
> > >>         return 0;
> > >>  }
> > >>
> > >> diff --git a/drivers/media/usb/uvc/uvcvideo.h b/drivers/media/usb/uvc/uvcvideo.h
> > >> index 9a596c8d894a..22656755a801 100644
> > >> --- a/drivers/media/usb/uvc/uvcvideo.h
> > >> +++ b/drivers/media/usb/uvc/uvcvideo.h
> > >> @@ -264,6 +264,8 @@ struct uvc_format {
> > >>         u32 fcc;
> > >>         u32 flags;
> > >>
> > >> +       char name[32];
> > >> +
> 
> I'd not really nice to have to store the name for every format, when we
> know it will very rarely be used.
> 
> One alternative option would be to store the GUID, which would halve the
> amount of memory. Another option would be to stop reporting those
> formats to userspace in uvc_ioctl_enum_fmt(). They can't be selected
> anyway, they have no unique 4CC.

Any opinion on this ? I'm increasingly tempted by not reporting
unsupported formats to userspace.

> > >>         unsigned int nframes;
> > >>         struct uvc_frame *frame;
> > >>  };

-- 
Regards,

Laurent Pinchart