Re: [PATCH v3 6/7] media: i2c: imx290: Check for availability in probe()

[Laurent Pinchart]

On Mon, Sep 02, 2024 at 07:22:43PM +0100, Dave Stevenson wrote:
> Hi Benjamin
> 
> On Mon, 2 Sept 2024 at 16:58, <bbara93@xxxxxxxxx> wrote:
> >
> > From: Benjamin Bara <benjamin.bara@xxxxxxxxxxx>
> >
> > Currently, the V4L2 subdevice is also created when the device is not
> > available/connected. From userspace perspective, there is no visible
> > difference between a working and not-working subdevice (except when
> > trying it out).
> >
> > This commit adds a simple availability check before starting with the
> > subdev initialization to error out instead.
> >
> > Signed-off-by: Benjamin Bara <benjamin.bara@xxxxxxxxxxx>
> > ---
> > Changes since v2:
> > - the new 1/8 is split out
> > - use dev_err_probe() (thx Laurent)
> > ---
> >  drivers/media/i2c/imx290.c | 12 ++++++++++++
> >  1 file changed, 12 insertions(+)
> >
> > diff --git a/drivers/media/i2c/imx290.c b/drivers/media/i2c/imx290.c
> > index 9610e9fd2059..6b292bbb0856 100644
> > --- a/drivers/media/i2c/imx290.c
> > +++ b/drivers/media/i2c/imx290.c
> > @@ -1571,6 +1571,7 @@ static int imx290_probe(struct i2c_client *client)
> >  {
> >         struct device *dev = &client->dev;
> >         struct imx290 *imx290;
> > +       u64 val;
> >         int ret;
> >
> >         imx290 = devm_kzalloc(dev, sizeof(*imx290), GFP_KERNEL);
> > @@ -1631,6 +1632,17 @@ static int imx290_probe(struct i2c_client *client)
> >         pm_runtime_set_autosuspend_delay(dev, 1000);
> >         pm_runtime_use_autosuspend(dev);
> >
> > +       /* Make sure the sensor is available before V4L2 subdev init. */
> > +       ret = cci_read(imx290->regmap, IMX290_STANDBY, &val, NULL);
> > +       if (ret) {
> > +               ret = dev_err_probe(dev, -ENODEV, "Failed to detect sensor\n");
> > +               goto err_pm;
> > +       }
> > +       if (val != IMX290_STANDBY_STANDBY) {
> 
> As Laurent commented on v2, this is a slightly unsafe check. If the
> device isn't controlled via a regulator then there's no guarantee that
> the sensor will be in standby.
> The cci_read call will already have returned an error if the sensor
> isn't present which will be 99.999% of the error cases.
> 
> If you want to catch the case where it's not in standby, why not put
> it into standby as a recovery mechanism. It'd be a better user
> experience than just bombing out of the probe.

I would also just drop the value check. I don't think it would really
catch real world issues.

> > +               ret = dev_err_probe(dev, -ENODEV, "Sensor is not in standby\n");
> > +               goto err_pm;
> > +       }
> > +
> >         /* Initialize the V4L2 subdev. */
> >         ret = imx290_subdev_init(imx290);
> >         if (ret)

-- 
Regards,

Laurent Pinchart