On 8/21/24 8:16 AM, Jonathan Cameron wrote: > On Tue, 20 Aug 2024 10:58:36 -0500 > David Lechner <dlechner@xxxxxxxxxxxx> wrote: > >> The AD4695 has a calibration feature that allows the user to compensate >> for variations in the analog front end. This implements this feature in >> the driver using the standard `calibgain` and `calibbias` attributes. >> >> Signed-off-by: David Lechner <dlechner@xxxxxxxxxxxx> > Hi David, > > Whilst some of the messy value manipulation is unavoidable > (oh for signed integer zero!), I wonder if we can at least > move one case into the core. > > See below. > >> + >> +static int ad4695_write_raw(struct iio_dev *indio_dev, >> + struct iio_chan_spec const *chan, >> + int val, int val2, long mask) >> +{ >> + struct ad4695_state *st = iio_priv(indio_dev); >> + unsigned int reg_val; >> + int ret; >> + >> + iio_device_claim_direct_scoped(return -EBUSY, indio_dev) { >> + switch (mask) { >> + case IIO_CHAN_INFO_CALIBSCALE: >> + switch (chan->type) { >> + case IIO_VOLTAGE: >> + if (val < 0 || val2 < 0) >> + reg_val = 0; >> + else if (val > 1) >> + reg_val = U16_MAX; >> + else >> + reg_val = (val * (1 << 16) + >> + mul_u64_u32_div(val2, 1 << 16, >> + MICRO)) / 2; > Maybe worth extending iio_write_channel_info() to handle > IIO_VAL_FRACTIONAL_LOG2()? > It'll look much like this and you'll need to provide write_raw_get_fmt() > so the core know what to do with the value formatting. > > I don't really like the mixture here between the read path being able > to rely on the core code to deal with the /2^X and the write path not. Sounds like a good idea to me. It seems like we would need to add an extra out parameter to write_raw_get_fmt to say what we want the X in 2^X to be. For example, we would want to make sure the val2 we get in write_raw for this driver is 15. >> + >> + return regmap_write(st->regmap16, >> + AD4695_REG_GAIN_IN(chan->scan_index), >> + reg_val); >> + default: >> + return -EINVAL; > > >