Re: [RFC PATCH v1 0/2] hwmon: (iio_hwmon) optionally force iio channel type

[Guenter Roeck]

On 5/16/21 11:14 AM, Liam Beguin wrote:
> On Sun May 16, 2021 at 12:26 PM EDT, Jonathan Cameron wrote:
> > On Sun, 16 May 2021 08:54:06 -0700
> > Guenter Roeck <linux@xxxxxxxxxxxx> wrote:
> >
> > > On 5/16/21 8:02 AM, Liam Beguin wrote:
> > > > Hi Jonathan,
> > > >
> > > > On Sun May 16, 2021 at 5:06 AM EDT, Jonathan Cameron wrote:
> > > > > On Sun, 16 May 2021 00:43:13 -0400
> > > > > Liam Beguin <liambeguin@xxxxxxxxx> wrote:
> > > > >   
> > > > > > Add a devicetree binding to optionally force a different IIO channel
> > > > > > type.
> > > > > >
> > > > > > This is useful in cases where ADC channels are connected to a circuit
> > > > > > that represent another unit such as a temperature or a current.
> > > > > >
> > > > > > `channel-types` was chosen instead of `io-channel-types` as this is not
> > > > > > part of the iio consumer bindings.
> > > > > >
> > > > > > In the current form, this patch does what it's intended to do:
> > > > > > change the unit displayed by `sensors`, but feels like the wrong way to
> > > > > > address the problem.
> > > > > >
> > > > > > Would it be possible to force the type of different IIO channels for
> > > > > > this kind of use case with a devicetree binding from the IIO subsystem?
> > > > > >
> > > > > > It would be convenient to do it within the IIO subsystem to have the
> > > > > > right unit there too.
> > > > > >
> > > > > > Thanks for your time,
> > > > > > Liam
> > > > >
> > > > > Hi Liam,
> > > > >
> > > > > +CC Peter for AFE part.
> > > > >
> > > > > It's an interesting approach, but I would suggest we think about this
> > > > > a different way.
> > > > >
> > > > > Whenever a channel is being used to measure something 'different' from
> > > > > what it actually measures (e.g. a voltage ADC measuring a current) that
> > > > > reflects their being some analog component involved.
> > > > > If you look at drivers/iio/afe/iio-rescale.c you can see the approach
> > > > > we currently use to handle this.
> > > >
> > > > Many thanks for pointing out the AFE code. That look like what I was
> > > > hoping to accomplish, but in a much better way.
> > > >    
> > > > > Effectively what you add to devicetree is a consumer of the ADC channel
> > > > > which in turn provides services to other devices. For this current case
> > > > > it would be either a current-sense-amplifier or a current-sense-shunt
> > > > > depending on what the analog front end looks like. We have to describe
> > > > > the characteristics of that front end which isn't something that can
> > > > > be done via a simple channel type.
> > > > >   
> > > >
> > > > Understood. My original intention was to use sensors.conf to do the
> > > > conversions and take into accounts those parameters.
> > > >    
> > > > > That afe consumer device can then provide services to another consumer
> > > > > (e.g. iio-hwmon) which work for your usecase.
> > > > >
> > > > > The main limitation of this approach currently is you end up with
> > > > > one device per channel. That could be improved upon if you have a
> > > > > usecase
> > > > > where it matters.
> > > > >
> > > > > I don't think we currently have an equivalent for temperature sensing
> > > > > but it would be easy enough to do something similar.
> > > >
> > > > Wonderful, thanks again for pointing out the AFE!
> > > >    
> > >
> > > Please don't reinvent the ntc_thermistor driver.
>
> > Agreed, I'd forgotten it existed :( Had a feeling we'd solved that
> > problem before
> > but couldn't remember the name of the driver.
> >
> > The afe driver already deals with current / voltage scaling and
> > conversion
> > for common analog circuits. Potential dividers, current shunts etc, but
> > they
> > are all the linear cases IIRC.
> >
> > ntc_thermistor deals with the much more complex job of dealing with a
> > thermistor.
>
> I agree, no need to reinvent this.
>
> Like Jonathan said, the ntc_thermistor driver seems to handle much more
> complex cases. Where would be the best place to add support for PT100
> and PT1000? iio-rescale?

Those sensors don't seem to be even useful for hardware monitoring, so
if they are linear (and it looks like that that the case) iio would be
a better place.

Guenter