Re: [PATCH v4 6/6] iio: adc: ad7405: add ad7405 driver

[David Lechner <dlechner@baylibre.com>]

On 5/19/25 9:02 AM, Pop Ioan Daniel wrote:
> Add support for the AD7405/ADUM770x, a high performance isolated ADC,
> 1-channel, 16-bit with a second-order Σ-Δ modulator that converts an
> analog input signal into a high speed, single-bit data stream.
> 
> Signed-off-by: Pop Ioan Daniel <pop.ioan-daniel@analog.com>
> ---

...

> +static int ad7405_set_dec_rate(struct iio_dev *indio_dev,
> +			       const struct iio_chan_spec *chan,
> +			       unsigned int dec_rate)
> +{
> +	struct ad7405_state *st = iio_priv(indio_dev);
> +	int ret;

We probably want some range checking here to make sure dec_rate is exactly one of
ad7405_dec_rates before passing it to the backend.

> +
> +	ret = iio_backend_oversampling_ratio_set(st->back, chan->scan_index, dec_rate);
> +	if (ret)
> +		return ret;
> +
> +	st->dec_rate = dec_rate;

Otherwise, this will report back a wrong value to userspace in ad7405_read_raw().

> +
> +	return 0;
> +}
> +
> +static int ad7405_get_scale(struct ad7405_state *st, int *val, int *val2)
> +{
> +	*val = st->info->full_scale_mv;
> +	*val2 = st->info->channel.scan_type.realbits - 1;
> +
> +	return IIO_VAL_FRACTIONAL_LOG2;

Probably not worth having a separate function for this. Can just do it inline.

> +}
> +
> +static int ad7405_read_raw(struct iio_dev *indio_dev,
> +			   const struct iio_chan_spec *chan, int *val,
> +			   int *val2, long info)
> +{
> +	struct ad7405_state *st = iio_priv(indio_dev);
> +
> +	switch (info) {
> +	case IIO_CHAN_INFO_SCALE:
> +		return ad7405_get_scale(st, val, val2);
> +	case IIO_CHAN_INFO_OVERSAMPLING_RATIO:
> +		*val = st->dec_rate;
> +		return IIO_VAL_INT;
> +	case IIO_CHAN_INFO_SAMP_FREQ:
> +		*val = DIV_ROUND_CLOSEST_ULL(st->ref_frequency, st->dec_rate);
> +		return IIO_VAL_INT;
> +	case IIO_CHAN_INFO_OFFSET:
> +		*val = 2 << (st->info->channel.scan_type.realbits - 1);

Isn't this supposed to be:

		*val = -(1 << (st->info->channel.scan_type.realbits - 1));

2 was a typo by me, but hopefully, you are testing this and doing the
math to double-check it is correct! I.e. generate a constant voltage signal
and capture some data, then convert it to mV with (raw + offset) * scale to
make sure it matches the voltage being applied to the input.

(I'm only human, so believe the math if disagrees with what I say. :-p)

> +		return IIO_VAL_INT;
> +	default:
> +		return -EINVAL;
> +	}
> +}
> +
> +static int ad7405_write_raw(struct iio_dev *indio_dev,
> +			    struct iio_chan_spec const *chan, int val,
> +			    int val2, long info)
> +{
> +	switch (info) {
> +	case IIO_CHAN_INFO_OVERSAMPLING_RATIO:

val can be negative here, which can cause it to turn into a large positive
number when it gets changed to unsigned since the last parameter of
ad7405_set_dec_rate has unsigned type. So we should either check for negative
first here or change the parameter to signed and check in the function.

> +		return ad7405_set_dec_rate(indio_dev, chan, val);
> +	default:
> +		return -EINVAL;
> +	}
> +}
> +