Re: [RFC PATCH] arm64: dts: qcom: Use labels with generic node names for ADC channels

[Jonathan Cameron]

On Sat, 10 Dec 2022 17:54:34 +0100
Marijn Suijten <marijn.suijten@xxxxxxxxxxxxxx> wrote:

> On 2022-12-10 12:02:03, Krzysztof Kozlowski wrote:
> > On 09/12/2022 22:53, Marijn Suijten wrote:  
> > > As discussed in [1] the DT should use labels to describe ADC channels,
> > > with generic node names, since the IIO drivers now moved to the fwnode
> > > API where node names include the `@xx` address suffix.
> > > 
> > > Especially for the ADC5 driver that uses extend_name - which cannot be
> > > removed for compatibility reasons - this results in sysfs files with the
> > > @xx name that wasn't previously present, and leads to an unpleasant
> > > file-browsing experience.
> > > 
> > > Also remove all the unused channel labels in pm660.dtsi.
> > > 
> > > [1]: https://lore.kernel.org/linux-arm-msm/20221106193018.270106-1-marijn.suijten@xxxxxxxxxxxxxx/T/#u
> > > 
> > > Signed-off-by: Marijn Suijten <marijn.suijten@xxxxxxxxxxxxxx>  
> > 
> > The talk was in context of bindings, not about changing all existing
> > users thus affecting DTS.  
> 
> And as a consequence, DTS.  The already-merged transition from OF to
> fwnode resulted in `@xx` to be included in the ADC channel name - and in
> the case of ADC5 even in sysfs filenames - so this seems like a
> necessary change to make.

Gah. We missed that at the time.  Arguably we should first fix that
particular issue as we will have lots of old DT out there.
(add a bit of code to strip the @xxx bit from that particular usecase).
It gets tricky because now we might have code relying on the new
broken behavior.

> 
> At the very least I would have changed the bindings submitted or
> co-authored /by myself/ since I initially decided to rely on this (now
> obviously) wrong behaviour, and should have used labels from the get go.
> 
> > What's more, to me "skin-temp-thermistor" is
> > quite generic name, maybe "thermistor" would be more and reflects the
> > purpose of the node, so it was more or less fine.  
> 
> Are you suggesting to not use "adc-chan", but "thermistor" as node name
> (and still use skin_temp as label)?  Or to keep the fully-written-out
> "thermistor" word in the label?
> 
> > Anyway I am against such changes without expressing it in the bindings.  
> 
> As expressed in [1] I suggested and am all for locking this change in
> via bindings, and you are right to expect that to have gone paired with
> this patch.
> 
> I'll submit that as the leading patch to this in v2, with the wildcard
> pattern changed to adc-chan (or something else pending the discussion
> above), and should I then also require the label property via `label:
> true`?
> 
> [1]: https://lore.kernel.org/linux-arm-msm/20221208101232.536i3cmjf4uk2z52@xxxxxxxxxxxxxx/

So the 'fun' here is what to do with old DTS as we need to support that
even if we update the binding docs and all in kernel users.

Probably right option in driver is:
a) Use label if present
b) Use node name if it's not adc-chan but strip the @xxx off it.
c) return an error.

p.s. Please add iio@xxxxxxxxxxxxxxx to future versions of this. If nothing
else I tend to lose direct emails about IIO stuff as they aren't in the
relevant patchwork instance.

> 
> - Marijn