Re: [PATCH v3 4/6] iio: frequency: adf41513: features on frequency change

[Andy Shevchenko <andriy.shevchenko@intel.com>]

On Mon, Jan 12, 2026 at 09:45:49AM +0000, Rodrigo Alencar wrote:
> On 26/01/09 09:07PM, Andy Shevchenko wrote:
> > On Thu, Jan 08, 2026 at 12:14:53PM +0000, Rodrigo Alencar via B4 Relay wrote:

First of all, remove the things you are agree with.

...

A side note: based on this discussion one may want to add a clarification
on how to use the unit-based multipliers to the documentation (top comment
on units.h also will work).

...

> > > +	bleed_value = div64_u64(st->settings.pfd_frequency_uhz * bleed_value,
> > > +				1600ULL * HZ_PER_MHZ * MICROHZ_PER_HZ);

You multiply Hz * Hz. One of them should be simply SI multiplier.
To me it sounds like one of

				1600ULL * MEGA * MICROHZ_PER_HZ);
				1600ULL * HZ_PER_MHZ * MICRO);

will be the correct one (and I lean towards the first one as you want units
to match).

The same is done in the definitions above somewhere.

> > > +	u16 ld_window_p1ns = div64_u64(10ULL * NANO * MICROHZ_PER_HZ,
> > > +				       st->settings.pfd_frequency_uhz << 1);
> > 
> > These multiplications (here and elsewhere) are (very) confusing.
> > 
> > I believe you want to have a frequency in Hz in µHz resolution. The second one
> > can be close to this if used GIGA instead of NANO. But I think the better way
> > to have something like the first one but with MICRO instead of MICROHZ_PER_HZ.
> > 
> > Please, put an order in these.
> 
> The first one: the numerator is in µHz, so the denominator is also in µHz so to
> cancel the units.
> 
> The second one: window size is nanoseconds with 0.1 precision in the datasheet.
> The numerator contains  MICROHZ_PER_HZ to convert µHz -> Hz = 1/s, and then
> 10ULL * NANO to convert 1/s into 0.1 ns.
> 
> How is that confusing? I am not sure GIGA is the right choice, as NANO shows
> that I am targeting nanoseconds, no? 

So, You wanted then one of

	u16 ld_window_p1ns = div64_u64(10ULL * NSEC_PER_SEC * MICROHZ_PER_HZ,
	u16 ld_window_p1ns = div64_u64(10ULL * NANO * MICRO,

(and I lean towards the first one as it may hint about the scale and resulting
 units).

Also make units in the name to be delimited with _.

	u16 ld_window_p1_ns = ...

...

> > > +	/* assuming both clock dividers hold similar values */
> > > +	total_div = mul_u64_u64_div_u64(st->settings.pfd_frequency_uhz,
> > > +					st->data.phase_resync_period_ns,
> > > +					1ULL * MICRO * NANO);
> > 
> > This sounds good as we multiply Hz by ns.
> 
> the numerator has a time in nanoseconds, so NANO 'cancels' that, as MICRO 'cancels'
> the micro under µHz.

Exactly, that's why I replied it sounds good.

-- 
With Best Regards,
Andy Shevchenko