Re: [PATCH v10 0/3] Change PWM-controlled LED pin active mode and algorithm

[Uwe Kleine-König]

Hello Nylon,

I took another look in the driver and found another problem:

On Tue, Jan 21, 2025 at 07:12:10PM +0100, Uwe Kleine-König wrote:
> On Tue, Jan 21, 2025 at 04:47:46PM +0800, Nylon Chen wrote:
> > Uwe Kleine-König <u.kleine-koenig@xxxxxxxxxxxx> 於 2025年1月21日 週二 下午3:47寫道：
> > >
> > > Hello,
> > >
> > > On Sun, Jan 19, 2025 at 03:03:16PM +0800, Nylon Chen wrote:
> > > > I ran some basic tests by changing the period and duty cycle in both
> > > > decreasing and increasing sequences (see the script below).
> > >
> > > What is clk_get_rate(ddata->clk) for you?
> > 130 MHz
> 
> OK, so the possible period lengths are
> 
> 	(1 << (16 + scale)) / (130 MHz)
> 
> for scale in [0, .. 15], right? That's
> 
> 	2^scale * 504123.07692307694 ns
> 
> So testing period in the range between 5000 ns and 15000 ns isn't
> sensible? Did I get something wrong? If the above is right, switching
> between period=1008246 ns and 1008247 ns is likely to trigger a
> warning.

The possible periods are of the form

	2^scale * A

where A is constant and only depends on the input clock rate. scale
ranges over [0, ... 15]. (If I got it right in my last mail, we have A =
504123.07692307694 ns.)

If you request say:

	.period = 3.9 * A
	.duty_cycle = 1.9 * A

the period actually emitted by the PWM will be 2 * A. But to calculate
frac the originally requested period (i.e. 3.9 * A) is used. So frac
becomes 31927 resulting in .duty_cycle being ~0.974 A. A better value
would be frac = 62259 which results in .duty_cycle ≅ 1.9 * A.
(Depending on A the values for frac might be off by one due to rounding
differences.)

So I would expect that PWM_DEBUG is angry with you if you go from

	.period = 2 * A
	.duty_cycle = 1.9 * A

to

	.period = 3.9 * A
	.duty_cycle = 1.9 * A

.

Best regards
Uwe

Attachment: signature.asc
Description: PGP signature