Re: [PATCH v2 4/4] hwmon: iio: add label for channels read by iio_hwmon

[Jonathan Cameron <jic23@kernel.org>]

On 19/07/16 07:55, Quentin Schulz wrote:
> On 18/07/2016 14:24, Jonathan Cameron wrote:
>> On 15/07/16 10:59, Quentin Schulz wrote:
>>> Currently, iio_hwmon only exposes values of the IIO channels it can read
>>> but no label by channel is exposed.
>>>
>>> This adds exposition of sysfs files containing label for IIO channels it
>>> can read based on extended_name field of the iio_chan_spec of the channel.
>>> If the extended_name field is empty, the sysfs file is not created by
>>> iio_hwmon.
>> Hmm. This is not the intent of extended name at all.  That exists to add
>> a small amount of information to an constructed IIO channel name.
>> Typically it's used to indicate physically wired stuff like:
>>
>> in_voltage0_vdd_raw for cases where that channel of the ADC is hard wired
>> to the vdd.  In this particular case the use might actually work as the
>> vdd makes it clear it's a voltage - in general that's not the case.
>>
>> Use of extended_name at all in IIO is only done after extensive review.
>> It adds nasty custom ABI to a device, so the gain has to be considerable
>> to use it.
>>
>> When I read your original suggestion of adding labels, I was expecting the
>> use of datasheet_name.  That has the advantage of being well defined by
>> the datasheet (if not it should not be provided) + not being used in
>> the construction of the IIO channel related attributes.  However, that
>> may still not correspond well to the expected labelling in hwmon.
> 
> Good to know for extend_name use cases. While doing further testing, I
> noticed the extend_name is also appended to the sysfs filename.. which
> is definitely not what we want.
> 
> I've checked for datasheet_name and it is only used to be compared to
> adc_channel_label from iio_map structure. Same for adc_channel_label
> (which has to be the same as datasheet_name of the iio_chan_spec it is
> linked to). So I could use this instead of extend_name to put a label on
> a channel.
> However, I thought of it to be more a way to identify the hardware in
> the datasheet more than giving users a hint on what it is. That's what
> "git grep adc_channel_label" told me. It's definitely better to use
> datasheet_name over extend_name for channel labeling but I don't know if
> it's really the good variable to use for labeling? In my understanding
> of datasheet_name, in my case it would be more "temp_gpadc" than "SoC
> temperature", that's what I mean.
If we are going to do this I think we need a new field in iio_chan_spec
for it.  The problem as ever is going to be that we'll end up with
'fuzzy' ABI which we can't change - even if we end up with spelling
mistakes sneaking through review - or entirely incorrect labels.
> 
>> Thinking more on this, the label is going to often be a function of how
>> the board is wired up...  Perhaps it should be a characteristic of the
>> channel_map (hence from DT or similar) rather than part of the IIO driver
>> itself?
> 
> Hmm.. I would not put a property in the DT only for labeling.
It's an odd one because in many cases the map (which is effectively
representing a wire) is the only relevant place to have such a label.

Can see what you mean about not putting things in DT which are basically
'help'!
> 
> [...]
>