On Tue, Jun 6, 2023 at 4:54 PM Herve Codina herve.codina@bootlin.com wrote:
On Sat, 3 Jun 2023 21:26:19 +0300 andy.shevchenko@gmail.com wrote:
Tue, May 23, 2023 at 05:12:21PM +0200, Herve Codina kirjoitti:
...
- int max;
- int min;
Wondering if there is already a data type for the ranges (like linear_range.h, but not sure it's applicable here).
Seems not applicable here.
- IIO does not use linear_range or something similar. It just uses simple int.
- ASoC does not use linear_range or something similar. It just uses simple long.
So, I keep the simple int min and max.
Sure.
...
- return 1; /* The value changed */
Perhaps this 1 needs a definition?
Yes but to be coherent, in ASoC code, many places need to be changed too in order to use the newly defined value. I don't think these modifications should be part of this series.
Yes, we are all for consistency.
...
- for (i = 0; i < iio_aux->num_chans; i++) {
iio_aux_chan = iio_aux->chans + i;
ret = of_property_read_string_index(np, "io-channel-names", i,
&iio_aux_chan->name);
if (ret < 0) {
dev_err(iio_aux->dev, "%pOF: failed to read io-channel-names[%d]\n", np, i);
return ret;
Ditto.
Will be changed in next iteration.
}
tmp = 0;
of_property_read_u32_index(np, "snd-control-invert-range", i, &tmp);
iio_aux_chan->is_invert_range = tmp;
You can use this variable directly.
Not sure, is_invert_range is a bool and tmp is a u32.
Ah, I see.
In previous iteration, I wrote iio_aux_chan->is_invert_range = !!tmp;
- }
Btw, can you avoid using OF APIs? It's better to have device property/fwnode API to be used from day 1.
Hum, this comment was raised in the previous iteration https://lore.kernel.org/linux-kernel/20230501162456.3448c494@jic23-huawei/
I didn't find any equivalent to of_property_read_u32_index() in the device_property_read_*() function family. I mean I did find anything available to get a value from an array using an index.
This is done by reading the entire array at once and then parsing as you wish in the code, device_property_read_u32_array() is for that.
In the previous iteration it was concluded that keeping OF APIs in this series seemed "reasonable".
Maybe, but consider the above.
-- With Best Regards, Andy Shevchenko