Re: [PATCH v2 8/9] iio: dac: ad3552r: add axi platform driver

[Jonathan Cameron]

On Thu, 5 Sep 2024 15:40:11 -0500
David Lechner <dlechner@xxxxxxxxxxxx> wrote:

> On 9/5/24 10:17 AM, Angelo Dureghello wrote:
> 
> ...
One reply to a comment David made.

Jonathan

> 
> > +		 */
> > +		if (st->single_channel)
> > +			clk_rate = DIV_ROUND_CLOSEST(clk_rate, 4);
> > +		else
> > +			clk_rate = DIV_ROUND_CLOSEST(clk_rate, 8);
> > +  
> 
> Having the sample rate depend on how many channels are enabled in
> the buffer seems a bit odd. Sampling frequency is not strictly
> defined in IIO, so I think it would be fine to always return the
> same value no matter how many channels are enabled.
> 
> We will just need to document that the sampling frequency is the
> rate per sample, not per channel. So if two channels are enabled,
> the effective sampling rate per channel is 1/2 of the sampling
> rate reported by the sysfs attribute. 

There is an oddity around this that we've never cleared up fully.

In my head at least if there is a single sampling_frequency it
applies to 'scans', not individual channel reads (so would change
with the number of channels enabled).  If there
is a per channel attribute we do have documentation:

What:		/sys/bus/iio/devices/iio:deviceX/in_voltageX_sampling_frequency
What:		/sys/bus/iio/devices/iio:deviceX/in_powerY_sampling_frequency
What:		/sys/bus/iio/devices/iio:deviceX/in_currentZ_sampling_frequency
KernelVersion:	5.20
Contact:	linux-iio@xxxxxxxxxxxxxxx
Description:
		Some devices have separate controls of sampling frequency for
		individual channels. If multiple channels are enabled in a scan,
		then the sampling_frequency of the scan may be computed from the
		per channel sampling frequencies.

For many devices the sampling frequency isn't down to each sample taking
N microsecs and them running back to back, it is instead a function of
a periodic sampling start for samples that take the same time whatever
the sampling frequency.  Also for simultaneous sampling ADCs it is never
channel dependent.

So I think if you want to avoid the confusion, make your device fall into
the description above and provide a per channel attribute rather
than shared_by_all.

Or keep it as things stand and have it halve when you double the channels.


> 
> > +		*val = clk_rate;
> > +
> > +		return IIO_VAL_INT;
> > +	}

...