Re: [PATCH 2/3] media: uvcvideo: Refactor uvc_query_ctrl

[Sakari Ailus]

Hi Ricardo,

On Tue, Oct 08, 2024 at 09:38:58PM +0800, Ricardo Ribalda wrote:
> On Tue, 8 Oct 2024 at 21:31, Sakari Ailus <sakari.ailus@xxxxxx> wrote:
> >
> > Hi Ricardo,
> >
> > On Tue, Oct 08, 2024 at 09:22:25PM +0800, Ricardo Ribalda wrote:
> > > Hi Sakari!
> > >
> > > On Tue, 8 Oct 2024 at 20:01, Sakari Ailus <sakari.ailus@xxxxxx> wrote:
> > > >
> > > > Hi Ricardo,
> > > >
> > > > On Tue, Oct 08, 2024 at 07:06:15AM +0000, Ricardo Ribalda wrote:
> > > > > Move the query control error logic to its own function.
> > > > > There is no functional change introduced by this patch.
> > > > >
> > > > > Signed-off-by: Ricardo Ribalda <ribalda@xxxxxxxxxxxx>
> > > > > ---
> > > > >  drivers/media/usb/uvc/uvc_video.c | 45 ++++++++++++++++++++++-----------------
> > > > >  1 file changed, 26 insertions(+), 19 deletions(-)
> > > > >
> > > > > diff --git a/drivers/media/usb/uvc/uvc_video.c b/drivers/media/usb/uvc/uvc_video.c
> > > > > index 853dfb7b5f7b..a57272a2c9e1 100644
> > > > > --- a/drivers/media/usb/uvc/uvc_video.c
> > > > > +++ b/drivers/media/usb/uvc/uvc_video.c
> > > > > @@ -67,30 +67,12 @@ static const char *uvc_query_name(u8 query)
> > > > >       }
> > > > >  }
> > > > >
> > > > > -int uvc_query_ctrl(struct uvc_device *dev, u8 query, u8 unit,
> > > > > -                     u8 intfnum, u8 cs, void *data, u16 size)
> > > > > +static int uvc_query_ctrl_error(struct uvc_device *dev, u8 intfnum, void *data)
> > > > >  {
> > > > >       int ret;
> > > > >       u8 error;
> > > > >       u8 tmp;
> > > > >
> > > > > -     ret = __uvc_query_ctrl(dev, query, unit, intfnum, cs, data, size,
> > > > > -                             UVC_CTRL_CONTROL_TIMEOUT);
> > > > > -     if (likely(ret == size))
> > > > > -             return 0;
> > > > > -
> > > > > -     if (ret > 0 && ret < size) {
> > > > > -             memset(data + ret, 0, size - ret);
> > > > > -             return 0;
> > > > > -     }
> > > > > -
> > > > > -     if (ret != -EPIPE) {
> > > > > -             dev_err(&dev->udev->dev,
> > > > > -                     "Failed to query (%s) UVC control %u on unit %u: %d (exp. %u).\n",
> > > > > -                     uvc_query_name(query), cs, unit, ret, size);
> > > > > -             return ret ? ret : -EPIPE;
> > > > > -     }
> > > > > -
> > > > >       /* Reuse data[0] to request the error code. */
> > > > >       tmp = *(u8 *)data;
> > > > >
> > > > > @@ -135,6 +117,31 @@ int uvc_query_ctrl(struct uvc_device *dev, u8 query, u8 unit,
> > > > >       return -EPIPE;
> > > > >  }
> > > > >
> > > > > +int uvc_query_ctrl(struct uvc_device *dev, u8 query, u8 unit,
> > > > > +                u8 intfnum, u8 cs, void *data, u16 size)
> > > > > +{
> > > > > +     int ret;
> > > > > +
> > > > > +     ret = __uvc_query_ctrl(dev, query, unit, intfnum, cs, data, size,
> > > > > +                            UVC_CTRL_CONTROL_TIMEOUT);
> > > > > +     if (likely(ret == size))
> > > > > +             return 0;
> > > > > +
> > > > > +     if (ret == -EPIPE)
> > > > > +             return uvc_query_ctrl_error(dev, intfnum, data);
> > > > > +
> > > > > +     dev_err(&dev->udev->dev,
> > > > > +             "Failed to query (%s) UVC control %u on unit %u: %d (exp. %u).\n",
> > > > > +             uvc_query_name(query), cs, unit, ret, size);
> > > >
> > > > This message should probably be printed after the check below.
> > >
> > > If the device is returning less bytes, the hardware is not behaving
> > > according to spec and it is good information, specially if you are
> > > bringing up a new device.
> > > I could make it  a dev_warn() (or even uvc_debug) if ret <size. WDYT?
> >
> > What I also came to think whether this is worth an explicit quirk flag.
> > There could well be devices that have other bugs that would still fall
> > under the same check.
> 
> Before https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/drivers/media/usb/uvc/uvc_video.c?id=a763b9fb58be869e252a7d33acb0a6390b01c801
> we were accepting short reads and showing the error.
> We are returning to the old behaviour, plus the memset.
> 
> A quirk will be difficult to maintain. If we can find a heuristics
> that works I would vote for that.

Seems reasonable.

> 
> 
> >
> > Either way, there should not be a message every single time this workaround
> > is applied. Isn't uvc_query_ctrl() also used in some IOCTLs outside probe?
> 
> I have only seen this behaviour for GET_DEF, and that value is cached.
> Meaning that the warning will only be shown once per control.

Could you add a comment with this information, also mentioning where this
has been noticed?

Perhaps we should wait for Laurent's comments still.

-- 
Kind regards,

Sakari Ailus