Re: [PATCH v1 3/3] iio: adc: ad4130: add AD4130 driver

[Cosmin Tanislav <demonsingur@gmail.com>]

On 4/13/22 18:41, Andy Shevchenko wrote:
> On Wed, Apr 13, 2022 at 1:41 PM Cosmin Tanislav <demonsingur@gmail.com> wrote:
> >
>> +#include <linux/property.h>
>> +#include <linux/regmap.h>
>> +#include <linux/regulator/consumer.h>
>> +#include <linux/spi/spi.h>
>> +
>> +#include <linux/iio/iio.h>
>> +#include <linux/iio/sysfs.h>
> 
> ...
> 
>> +#define AD4130_8_NAME                  "ad4130-8"
> 
> What the meaning of -8 ? Is it number of channels? Or is it part of
> the official model (part number)? Can we see, btw, Datasheet: tag with
> a corresponding link in the commit message?
> 

That's just the name specified in the datasheet. I honestly don't have
much of an idea about why it is like that. Also, I already put the
datasheet in the .yaml documentation. Do you really also want it
in each commit message too?

> 
>> +#define AD4130_RESET_CLK_COUNT         64
>> +#define AD4130_RESET_BUF_SIZE          (AD4130_RESET_CLK_COUNT / 8)
> 
> To be more precise shouldn't the above need to have DIV_ROUND_UP() ?
> 

Does it look like 64 / 8 needs any rounding?

> ...
> 
>> +#define AD4130_FREQ_FACTOR             1000000000ull
>> +#define AD4130_DB3_FACTOR              1000
> 
> Ditto.

AD4130_DB3_FACTOR is unit-less. In the datasheet, the relation between
sampling frequency and 3db frequency is represented as a 0.xyz value,
hence why the db3_div values and 1000 factor.

> 
>> +enum ad4130_fifo_mode {
>> +       AD4130_FIFO_MODE_DISABLED = 0b00,
>> +       AD4130_FIFO_MODE_WATERMARK = 0b01,
>> +};
>> +
>> +enum ad4130_mode {
>> +       AD4130_MODE_CONTINUOUS = 0b0000,
>> +       AD4130_MODE_IDLE = 0b0100,
>> +};
> 
> 0b?! Hmm... Not that this is bad, just not so usual :-)
> 

I always use 0b to be consistent with the datasheet values which are
represented in binary. I think it makes it easier to not mess up
when initially translating the datasheet into code and later when
cross-checking with the datasheet.

> 
>> +struct ad4130_filter_config {
>> +       enum ad4130_filter_mode         filter_mode;
>> +       unsigned int                    odr_div;
>> +       unsigned int                    fs_max;
>> +       unsigned int                    db3_div;
>> +       enum iio_available_type         samp_freq_avail_type;
>> +       int                             samp_freq_avail_len;
>> +       int                             samp_freq_avail[3][2];
>> +       enum iio_available_type         db3_freq_avail_type;
>> +       int                             db3_freq_avail_len;
>> +       int                             db3_freq_avail[3][2];
> 
> These 3:s can be defined?
>
I could define IIO_AVAIL_RANGE_LEN and IIO_AVAIL_SINGLE_LEN and then
define another IIO_AVAIL_LEN that is the max between the two.
But that's just over-complicating it, really.

> ...
> 
> 
>> +static int ad4130_get_reg_size(struct ad4130_state *st, unsigned int reg,
>> +                              unsigned int *size)
>> +{
> 
>> +       if (reg >= ARRAY_SIZE(ad4130_reg_size))
>> +               return -EINVAL;
> 
> When this condition is true?

When the user tries reading a register from direct_reg_access
that hasn't had its size defined.

> 
>> +       regmap_update_bits(st->regmap, AD4130_REG_IO_CONTROL, mask,
>> +                          value ? mask : 0);
> 
> One line?
> 
> No error check?
> 

I actually can't think of a scenario where this would fail. It doesn't
if the chip is not even connected.

> 
>> +       if (setup_info->enabled_channels)
>> +               return -EINVAL;
> 
> -EBUSY?
> 

Eh, I don't think so. It would be pretty impossible for the code to hit
this if statement, taking into account the ad4130_find_slot() logic.
I could as well not have it at all.

> 
>> +static void ad4130_freq_to_fs(enum ad4130_filter_mode filter_mode,
>> +                             int val, int val2, unsigned int *fs, bool db3)
>> +{
>> +       const struct ad4130_filter_config *filter_config =
>> +               &ad4130_filter_configs[filter_mode];
>> +       unsigned long long dividend, divisor;
>> +       int temp;
>> +
>> +       dividend = filter_config->fs_max * filter_config->odr_div *
>> +                  (val * AD4130_FREQ_FACTOR + val2);
>> +       divisor = AD4130_MAX_ODR * AD4130_FREQ_FACTOR;
>> +
>> +       if (db3) {
>> +               dividend *= AD4130_DB3_FACTOR;
>> +               divisor *= filter_config->db3_div;
>> +       }
>> +
>> +       temp = AD4130_FS_MIN + filter_config->fs_max -
>> +              DIV64_U64_ROUND_CLOSEST(dividend, divisor);
>> +
>> +       if (temp < AD4130_FS_MIN)
>> +               temp = AD4130_FS_MIN;
>> +       else if (temp > filter_config->fs_max)
>> +               temp = filter_config->fs_max;
>> +
>> +       *fs = temp;
> 
> Would be nice to put a comment explaining the math behind this code.
> 
>> +}
>> +
>> +static void ad4130_fs_to_freq(enum ad4130_filter_mode filter_mode,
>> +                             unsigned int fs, int *val, int *val2, bool db3)
>> +{
>> +       const struct ad4130_filter_config *filter_config =
>> +               &ad4130_filter_configs[filter_mode];
>> +       unsigned int dividend, divisor;
>> +       u64 temp;
>> +
>> +       dividend = (filter_config->fs_max - fs + AD4130_FS_MIN) *
>> +                  AD4130_MAX_ODR;
>> +       divisor = filter_config->fs_max * filter_config->odr_div;
>> +
>> +       if (db3) {
>> +               dividend *= filter_config->db3_div;
>> +               divisor *= AD4130_DB3_FACTOR;
>> +       }
>> +
>> +       temp = div_u64(dividend * AD4130_FREQ_FACTOR, divisor);
>> +       *val = div_u64_rem(temp, AD4130_FREQ_FACTOR, val2);
> 
> 
> Ditto.
> 
I'll see what I can come up with.

> 
>> + out:
> 
> out_unlock: ?
> Ditto for similar cases.

There's a single label in the function, and there's a mutex being
taken, and, logically, the mutex must be released on the exit path.
It's clear what the label is for to me.

> 
>> +               *val = st->bipolar ? -(1 << (chan->scan_type.realbits - 1)) : 0;
> 
> Hmm... It seems like specific way to have a sign_extended, or actually
> reduced) mask.
> Can you rewrite it with the (potential)UB-free approach?
> 
> (Note, that if realbits == 32, this will have a lot of fun in
> accordance with C standard.)
> 

Can you elaborate on this? The purpose of this statement is to shift the
results so that, when bipolar configuration is enabled, the raw value is
offset with 1 << (realbits - 1) towards negative.

For the 24bit chips, 0x800000 becomes 0x000000.

Maybe you misread it as left shift on a negative number? The number
is turned negative only after the shift...

> ...
> 
>> +               *vals = (int *)st->scale_tbls[setup_info->ref_sel];
> 
> Can we get rid of casting here and in the similar cases?
> 

I feel like scale_tbls is best defined as an array of two-element
arrays. Because its type is IIO_VAL_FRACTIONAL.
But obviously the IIO framework can't take this case into account by
itself, so we cast it so it receives what it wants.

> 
>> +
>> +       if (val > AD4130_FIFO_SIZE)
>> +               return -EINVAL;
>> +
>> +       /*
>> +        * Always set watermark to a multiple of the number of enabled channels
>> +        * to avoid making the FIFO unaligned.
>> +        */
>> +       eff = rounddown(val, st->num_enabled_channels);
>> +
>> +       mutex_lock(&st->lock);
>> +
>> +       ret = regmap_update_bits(st->regmap, AD4130_REG_FIFO_CONTROL,
>> +                                AD4130_WATERMARK_MASK,
>> +                                FIELD_PREP(AD4130_WATERMARK_MASK,
>> +                                           ad4130_watermark_reg_val(eff)));
> 
> Temporary variable for mask?
> 

You mean for value?

> 
>> +static int ad4130_get_ref_voltage(struct ad4130_state *st,
>> +                                 enum ad4130_ref_sel ref_sel,
>> +                                 unsigned int *ref_uv)
>> +{
>> +       struct device *dev = &st->spi->dev;
>> +       int ret;
>> +
>> +       switch (ref_sel) {
>> +       case AD4130_REF_REFIN1:
>> +               ret = regulator_get_voltage(st->regulators[2].consumer);
>> +               break;
>> +       case AD4130_REF_REFIN2:
>> +               ret = regulator_get_voltage(st->regulators[3].consumer);
>> +               break;
>> +       case AD4130_REF_AVDD_AVSS:
>> +               ret = regulator_get_voltage(st->regulators[0].consumer);
>> +               break;
>> +       case AD4130_REF_REFOUT_AVSS:
> 
>> +               if (!st->int_ref_en) {
>> +                       ret = -EINVAL;
>> +                       break;
>> +               }
>> +
>> +               ret = st->int_ref_uv;
>> +               break;
> 
> Can be one if-else instead.
> 
>> +       default:
>> +               ret = -EINVAL;
>> +               break;
>> +       }
>> +
>> +       if (ret <= 0)
> 
> = 0 ?! Can you elaborate, please, this case taking into account below?
> 

I guess I just did it because voltage = 0 doesn't make sense and would
make scale be 0.0.


>> +               return dev_err_probe(dev, ret, "Cannot use reference %u\n",
>> +                                    ref_sel);
>> +
>> +       if (ref_uv)
>> +               *ref_uv = ret;
>> +
>> +       return 0;
>> +}
> 

> 
>> +       fwnode_property_read_u32(child, "adi,excitation-pin-0",
>> +                                &chan_info->iout0);
> 
> No default and/or error check?
> 
> ...
> 
>> +       fwnode_property_read_u32(child, "adi,excitation-pin-1",
>> +                                &chan_info->iout1);
> 
> Ditto.

Default is 0, just like the register defaults.

> 
> ...
> 
>> +static int ad4130_parse_fw_children(struct iio_dev *indio_dev)
>> +{
>> +       struct ad4130_state *st = iio_priv(indio_dev);
>> +       struct device *dev = &st->spi->dev;
>> +       struct fwnode_handle *child;
>> +       int ret;
>> +
>> +       indio_dev->channels = st->chans;
>> +
>> +       device_for_each_child_node(dev, child) {
>> +               ret = ad4130_parse_fw_channel(indio_dev, child);
>> +               if (ret)
>> +                       break;
>> +       }
> 
>> +       fwnode_handle_put(child);
> 
> There is no need to put fwnode if child is NULL. Moreover, the above
> pattern might be percepted wrongly, i.e. one may think that
> fwnode_handle_put() is a must after a loop.
> 

fwnode_handle_put already checks if the child is NULL. Why do the same
check twice?

> 
> Can you explain why regmap locking is needed?
> 

Am I supposed to set .disable_locking = true since SPI has its own
locking?