Re: [PATCH v3 01/15] dt-bindings: iio: Add bindings for sx9310 sensor

[Rob Herring]

On Thu, Aug 6, 2020 at 12:14 PM Jonathan Cameron <jic23@xxxxxxxxxx> wrote:
>
> On Mon, 3 Aug 2020 20:01:06 -0600
> Rob Herring <robh+dt@xxxxxxxxxx> wrote:
>
> > On Mon, Aug 3, 2020 at 1:00 PM Stephen Boyd <swboyd@xxxxxxxxxxxx> wrote:
> > >
> > > Quoting Jonathan Cameron (2020-08-01 08:06:39)
> > > > On Fri, 31 Jul 2020 10:48:38 -0600
> > > > Daniel Campello <campello@xxxxxxxxxxxx> wrote:
> > > > > diff --git a/Documentation/devicetree/bindings/iio/proximity/semtech,sx9310.yaml b/Documentation/devicetree/bindings/iio/proximity/semtech,sx9310.yaml
> > > > > new file mode 100644
> > > > > index 00000000000000..5739074d3592fe
> > > > > --- /dev/null
> > > > > +++ b/Documentation/devicetree/bindings/iio/proximity/semtech,sx9310.yaml
> > > > > @@ -0,0 +1,65 @@
> > > [...]
> > > > > +
> > > > > +  "#io-channel-cells":
> > > > > +    const: 1
> > > > > +
> > > > > +required:
> > > > > +  - compatible
> > > > > +  - reg
> > > > > +  - "#io-channel-cells"
> > > >
> > > > Missed this in earlier review (only noticed when I saw whilst santity
> > > > checking earlier versions.
> > > >
> > > > Fairly sure we should only need #io-channel-cells if we have
> > > > a consumer of a channel somewhere else in DT.  So it's not
> > > > required as far as I can see.
> > > >
> > >
> > > This is mostly a decision for Rob to make, but I would make it required
> > > because the device is always an io channel provider. It may be that it
> > > isn't providing anything in the DT to something else in the DT but it is
> > > providing this information somewhere so always having to spell that out
> > > is simple and doesn't hurt.
> >
> > I agree. If the user is split in a board file or overlay, we don't
> > want to have to be adding it to the provider at that time.
>
> That is perhaps a reasonable view point for devices with channels that
> are likely to be used by consumer drivers, but in this particular case we
> are talking about a proximity sensor.  So far I don't think we
> have any consumer drivers for this type of sensor (I might have forgotten
> one of course!)

Indeed, I didn't consider whether it made sense in the first place. So
should it just not be specified at all in this case? I can't really
picture what the usecase for a consumer node would be.

Rob