Re: [PATCH 4/4] iio: afe/rescale: Implement write_raw

[Paul Cercueil]

Hi Peter,

Le ven., juil. 22 2022 at 00:16:36 +0200, Peter Rosin <peda@xxxxxxxxxx> 
a écrit :
> Hi!
>
> 2022-07-21 at 21:15, Paul Cercueil wrote:
> >  Implement write_raw by converting the value if writing the scale, or
> >  just calling the managed channel driver's write_raw otherwise.
> >
> >  Signed-off-by: Paul Cercueil <paul@xxxxxxxxxxxxxxx>
> >  ---
> >   drivers/iio/afe/iio-rescale.c | 22 ++++++++++++++++++++++
> >   1 file changed, 22 insertions(+)
> >
> >  diff --git a/drivers/iio/afe/iio-rescale.c 
> > b/drivers/iio/afe/iio-rescale.c
> >  index 5c9970b93384..0edb62ee4508 100644
> >  --- a/drivers/iio/afe/iio-rescale.c
> >  +++ b/drivers/iio/afe/iio-rescale.c
> >  @@ -141,6 +141,27 @@ int rescale_process_offset(struct rescale 
> > *rescale, int scale_type,
> >   	}
> >   }
> >
> >  +static int rescale_write_raw(struct iio_dev *indio_dev,
> >  +			     struct iio_chan_spec const *chan,
> >  +			     int val, int val2, long mask)
> >  +{
> >  +	struct rescale *rescale = iio_priv(indio_dev);
> >  +	unsigned long long tmp;
> >  +
> >  +	switch (mask) {
> >  +	case IIO_CHAN_INFO_SCALE:
> >  +		tmp = val * 1000000000LL;
> >  +		do_div(tmp, rescale->numerator);
> >  +		tmp *= rescale->denominator;
> >  +		do_div(tmp, 1000000000LL);
>
> do_div is for unsigned operands. Can val never ever be negative?
> What about the numerator and denominator, can those be negative? I
> think this code should live in a new rescale_process_inverse_scale
> function, or something like that (and a few tests could be added to
> drivers/iio/test/iio-test-rescale.c)

I can do that.

> >  +		return iio_write_channel_attribute(rescale->source, tmp, 0,
> >  +						   IIO_CHAN_INFO_SCALE);
> >  +	default:
>
> What if the source driver has a .write_raw_get_fmt callback? That bit
> of info is silently dropped (with no comment that a shortcut has been
> taken). How does inverse rescaling mix with a .write_raw_get_fmt that
> returns e.g. IIO_VAL_INT_PLUS_MICRO_DB anyway? I think all cases might
> get a bit hairy to support, so I think you need to do some filtering
> and somehow fail the .write_raw call if the .write_raw_get_fmt of the
> source returns something that gets too difficult to support.

If the inverse rescale uses the same code as rescale_process_scale() 
then it becomes problematic, yes, as it likes to change the type of the 
value.

What I could try - compute the inverse of the value, then find the 
closest scale value+type that the source driver supports, and use this 
as the value+type. Then the only failure point would be if 
.write_raw_get_fmt returns something different than the formats 
returned by .read_avail, but that sounds unlikely to happen.

Cheers,
-Paul

> >  +		return iio_write_channel_attribute(rescale->source,
> >  +						   val, val2, mask);
> >  +	}
> >  +}
> >  +
> >   static int rescale_read_raw(struct iio_dev *indio_dev,
> >   			    struct iio_chan_spec const *chan,
> >   			    int *val, int *val2, long mask)
> >  @@ -250,6 +271,7 @@ static int rescale_read_avail(struct iio_dev 
> > *indio_dev,
> >   }
> >
> >   static const struct iio_info rescale_info = {
> >  +	.write_raw = rescale_write_raw,
> >   	.read_raw = rescale_read_raw,
> >   	.read_avail = rescale_read_avail,
> >   };