Hello Clemens,

On Mon, Mar 29, 2021 at 07:33:56PM +0200, Clemens Gruber wrote:
> On Mon, Mar 29, 2021 at 07:15:59PM +0200, Uwe Kleine-König wrote:
> > On Mon, Mar 29, 2021 at 02:57:06PM +0200, Clemens Gruber wrote:
> > > @@ -330,14 +345,22 @@ static int pca9685_pwm_apply(struct pwm_chip *chip, struct pwm_device *pwm,
> > >  
> > >  	if (!state->enabled || duty < 1) {
> > >  		pca9685_pwm_set_duty(pca, pwm->hwpwm, 0);
> > > +		clear_bit(pwm->hwpwm, pca->prescaler_users);
> > 
> > Hmm, so if "my" channel runs at say
> > 
> > 	.duty_cycle = 2539520 ns
> > 	.period = 5079040 ns
> > 
> > and I call pwm_apply_state(mypwm, { .duty_cycle = 0, .period = 5079040,
> > enabled = true }); it might happen that another channel modifies the
> > period and I won't be able to return to the initial setting.
> 
> Yes, that's correct.
> 
> But that also applies to PWMs set to 100%:
> 
> pwm_apply_state(mypwm, { .duty_cycle = 5079040, .period = 5079040,
> enabled = true });
> 
> As this sets the full ON bit and does not use the prescaler, with the
> current code, another channel could modify the period and you wouldn't
> be able to return to the initial setting of 50%.
> 
> > So I think it's sensible to only clear the user bit if the PWM is
> > disabled, but not if it is configured for duty_cycle = 0.
> > 
> > Does this make sense?
> 
> With both cases in mind, you are suggesting we block modifications of
> the prescaler if other PWMs are enabled and not if other PWMs are using
> the prescaler?

That sounds sensible, yes.

Best regards
Uwe

-- 
Pengutronix e.K.                           | Uwe Kleine-König            |
Industrial Linux Solutions                 | https://www.pengutronix.de/ |