Re: [PATCH v4 2/2] thermal: thermal-generic-adc: add temperature sensor channel

[Svyatoslav Ryhel]

нд, 13 квіт. 2025 р. о 12:28 Jonathan Cameron <jic23@xxxxxxxxxx> пише:
>
> On Sat, 12 Apr 2025 14:06:06 +0300
> Svyatoslav Ryhel <clamor95@xxxxxxxxx> wrote:
>
> > сб, 12 квіт. 2025 р. о 13:54 Jonathan Cameron <jic23@xxxxxxxxxx> пише:
> > >
> > > On Sat, 5 Apr 2025 18:23:25 +0300
> > > Svyatoslav Ryhel <clamor95@xxxxxxxxx> wrote:
> > >
> > > > пн, 10 бер. 2025 р. о 09:57 Svyatoslav Ryhel <clamor95@xxxxxxxxx> пише:
> > > > >
> > > > > To avoid duplicating sensor functionality and conversion tables, this
> > > > > design allows converting an ADC IIO channel's output directly into a
> > > > > temperature IIO channel. This is particularly useful for devices where
> > > > > hwmon isn't suitable or where temperature data must be accessible through
> > > > > IIO.
> > > > >
> > > > > One such device is, for example, the MAX17040 fuel gauge.
> > > > >
> > > > > The temperature data, while technically a product of conversion and thus
> > > > > categorized as IIO_CHAN_INFO_PROCESSED, maintains its unscaled state
> > > > > (milli-degree). To account for this, IIO_CHAN_INFO_RAW is used along with
> > > > > IIO_CHAN_INFO_SCALE to provide different degrees of accuracy.
> > >
> > > You've lost me in this description.  The base units of an IIO temperature channel
> > > are milli-degrees so if the scaling is already right for that you would
> > > be fine using a IIO_CHAN_INFO_PROCESSED channel.
> > >
> > > A few other minor things inline.
> > >
> > > > >
> > > > > Signed-off-by: Svyatoslav Ryhel <clamor95@xxxxxxxxx>
> > > > > ---
> > > > >  drivers/thermal/thermal-generic-adc.c | 62 ++++++++++++++++++++++++++-
> > > > >  1 file changed, 61 insertions(+), 1 deletion(-)
> > > > >
> > > > > diff --git a/drivers/thermal/thermal-generic-adc.c b/drivers/thermal/thermal-generic-adc.c
> > > > > index ee3d0aa31406..7dcc2e1168a4 100644
> > > > > --- a/drivers/thermal/thermal-generic-adc.c
> > > > > +++ b/drivers/thermal/thermal-generic-adc.c
> > > > > @@ -7,6 +7,7 @@
> > > > >   * Author: Laxman Dewangan <ldewangan@xxxxxxxxxx>
> > > > >   */
> > > > >  #include <linux/iio/consumer.h>
> > > > > +#include <linux/iio/iio.h>
> > > > >  #include <linux/kernel.h>
> > > > >  #include <linux/module.h>
> > > > >  #include <linux/platform_device.h>
> > > > > @@ -73,6 +74,65 @@ static const struct thermal_zone_device_ops gadc_thermal_ops = {
> > > > >         .get_temp = gadc_thermal_get_temp,
> > > > >  };
> > > > >
> > > > > +static const struct iio_chan_spec gadc_thermal_iio_channel[] = {
> > > Even though there is only one. If it is an array use channels.
> > >
> > > or stop it being an array and just take a pointer to a single channel
> > > instance.
> > >
> >
> > There should not be more channels, but tbh, you never know, I will
> > rename to channels.
> >
> > > > > +       {
> > > > > +               .type = IIO_TEMP,
> > > > > +               .info_mask_separate = BIT(IIO_CHAN_INFO_RAW) |
> > > > > +                                     BIT(IIO_CHAN_INFO_SCALE),
> > > > > +       }
> > > > > +};
> > > > > +
> > > > > +static int gadc_thermal_read_raw(struct iio_dev *indio_dev,
> > > > > +                                struct iio_chan_spec const *chan,
> > > > > +                                int *val, int *val2, long mask)
> > > > > +{
> > > > > +       struct gadc_thermal_info *gtinfo = iio_priv(indio_dev);
> > > > > +       int ret;
> > > > > +
> > > > > +       switch (mask) {
> > > > > +       case IIO_CHAN_INFO_RAW:
> > > > > +               ret = gadc_thermal_get_temp(gtinfo->tz_dev, val);
> > > > > +               if (ret)
> > > > > +                       return ret;
> > > > > +
> > > > > +               return IIO_VAL_INT;
> > > > > +
> > > > > +       case IIO_CHAN_INFO_SCALE:
> > > > > +               /* scale to a degree centigrade */
> > >
> > > As above.  See Documentation/ABI/testing/sysfs-bus-iio
> > > entries from temperature. Scaling of a temperature channel is milli-degrees
> > >
> > > This is a bit of a historical artefact. Way back at the start of IIO
> > > when we had relatively few channel types, where possible I matched the
> > > scaling to hwmon.  With hindsight that made things a bit inconsistent
> > > but we are stuck with it as ABI :(
> > >
> >
> > RAW channel is in milli-degrees, or this is not enough? I don't get
> > your point here tbh. What is wrong? Battery driver requires
> > temperature in degree centigrade hence it is scaled to it.
>
> If the battery driver is assuming the temperature from a query to an IIO
> provide (to get the processed value) is in degrees centigrade then that
> is a bug that needs fixing and it probably implies a scaling problem
> in some other driver given I assume someone is using that driver!
>
> The scale here should be 1.  Simplest being to just make a processed
> channel and not provide scale at all.
>
> We never explicitly document the internal ABI but the use of
> scale and raw must match the userspace ABI which is documented
> in Documentation/ABI/testing/sysfs-bus-iio
>
> Jonathan
>

Acknowledged. I will return back to processed and send a fix regards
the battery.

>
> >
> > > Jonathan
> > >
> > > > > +               *val = 1;
> > > > > +               *val2 = 1000;
> > > > > +               return IIO_VAL_FRACTIONAL;
> > > > > +
> > > > > +       default:
> > > > > +               return -EINVAL;
> > > > > +       }
> > > > > +}
> > > > > +
> > > > > +static const struct iio_info gadc_thermal_iio_info = {
> > > > > +       .read_raw = gadc_thermal_read_raw,
> > > > > +};
> > > > > +
> > > > > +static int gadc_iio_register(struct device *dev, struct gadc_thermal_info *gti)
> > > > > +{
> > > > > +       struct gadc_thermal_info *gtinfo;
> > > > > +       struct iio_dev *indio_dev;
> > > > > +
> > > > > +       indio_dev = devm_iio_device_alloc(dev, sizeof(struct gadc_thermal_info));
> > > > > +       if (!indio_dev)
> > > > > +               return -ENOMEM;
> > > > > +
> > > > > +       gtinfo = iio_priv(indio_dev);
> > > > > +       memcpy(gtinfo, gti, sizeof(struct gadc_thermal_info));
> > >
> > > sizeof(*gtinfo) probably slightly better.
> > >
> > > > > +
> > > > > +       indio_dev->name = dev_name(dev);
> > >
> > > What does this end up as?  The convention in IIO is to name after
> > > a part number.  If you have duplicates this isn't how you tell them
> > > apart.  So I'd kind of expect thermal-generic-temp or
> > > something like that.
> > >
> >
> > it is "generic-adc-thermal" with this name, it is not present anywhere in IIO.
> >

Then this is acceptable?

> > > > > +       indio_dev->info = &gadc_thermal_iio_info;
> > > > > +       indio_dev->modes = INDIO_DIRECT_MODE;
> > > > > +       indio_dev->channels = gadc_thermal_iio_channel;
> > > As above, I'd stop that being an array and use
> > >         indio_dev->channels = &gadc_thermal_iio_chanel;
> > >         indio_dev->channels = 1;
> > >
> > > Unless you think maybe we will get more channels in future, in which case
> > > just rename it channels (which happens to have one element this time)
> > >
> > > > > +       indio_dev->num_channels = ARRAY_SIZE(gadc_thermal_iio_channel);
> > > > > +
> > > > > +       return devm_iio_device_register(dev, indio_dev);
> > > > > +}
> > > > > +
> > > > >  static int gadc_thermal_read_linear_lookup_table(struct device *dev,
> > > > >                                                  struct gadc_thermal_info *gti)
> > > > >  {
> > > > > @@ -153,7 +213,7 @@ static int gadc_thermal_probe(struct platform_device *pdev)
> > > > >
> > > > >         devm_thermal_add_hwmon_sysfs(dev, gti->tz_dev);
> > > > >
> > > > > -       return 0;
> > > > > +       return gadc_iio_register(&pdev->dev, gti);
> > > > >  }
> > > > >
> > > > >  static const struct of_device_id of_adc_thermal_match[] = {
> > > > > --
> > > > > 2.43.0
> > > > >
> > > >
> > > > Added Jonathan Cameron and linux-iio@xxxxxxxxxxxxxxx to list.
> > > >
> > > > Jonathan, this is newer version of the thermal-generic-adc you have
> > > > reviewed recently with channels adjusted like proposed in v3.
> > >
>