On Mon, Mar 29, 2021 at 07:11:53PM +0200, Clemens Gruber wrote:
> On Mon, Mar 29, 2021 at 06:54:29PM +0200, Uwe Kleine-König wrote:
> > On Mon, Mar 29, 2021 at 02:57:02PM +0200, Clemens Gruber wrote:
> > > [...]
> > > +	/* Calculate (chip-wide) period from prescale value */
> > > +	regmap_read(pca->regmap, PCA9685_PRESCALE, &val);
> > > +	state->period = (PCA9685_COUNTER_RANGE * 1000 / PCA9685_OSC_CLOCK_MHZ) *
> > > +			(val + 1);
> > 
> > As we have PCA9685_OSC_CLOCK_MHZ = 25 this is an integer calculation
> > without loss of precision. It might be worth to point that out in a
> > comment. (Otherwise doing the division at the end might be more
> > sensible.)
> 
> What comment do you have in mind?
> /* 1 integer multiplication (without loss of precision) at runtime */ ?

Something like:

	/*
	 * PCA9685_OSC_CLOCK_MHZ is 25 and so an integer divider of
	 * 1000. So the calculation here is only a multiplication and
	 * we're not loosing precision.
	 */
 
> > > +	/* The (per-channel) polarity is fixed */
> > > +	state->polarity = PWM_POLARITY_NORMAL;
> > > +
> > > +	if (pwm->hwpwm >= PCA9685_MAXCHAN) {
> > > +		/*
> > > +		 * The "all LEDs" channel does not support HW readout
> > > +		 * Return 0 and disabled for backwards compatibility
> > > +		 */
> > > +		state->duty_cycle = 0;
> > > +		state->enabled = false;
> > > +		return;
> > > +	}
> > > +
> > > +	duty = pca9685_pwm_get_duty(pca, pwm->hwpwm);
> > > +
> > > +	state->enabled = !!duty;
> > > +	if (!state->enabled) {
> > > +		state->duty_cycle = 0;
> > > +		return;
> > > +	} else if (duty == PCA9685_COUNTER_RANGE) {
> > > +		state->duty_cycle = state->period;
> > > +		return;
> > > +	}
> > > +
> > > +	duty *= state->period;
> > > +	state->duty_cycle = duty / PCA9685_COUNTER_RANGE;
> > 
> > .apply uses ROUND_CLOSEST to calculate duty from state->duty_cycle,
> > still using / here (instead of ROUND_CLOSEST), but again as
> > PCA9685_OSC_CLOCK_MHZ is 25 this calculation doesn't suffer from
> > rounding errors. So if you feed the state returned here into .apply
> > again, there is (as I want it) no change.
> > 
> > The only annoyance is that if PCA9685_PRESCALE holds a value smaller
> > than 3, .apply() will fail. Not sure there is any saner way to handle
> > this.
> 
> According to the datasheet, "The hardware forces a minimum value that
> can be loaded into the PRE_SCALE register at '3'", so there should never
> be anything below 3 in that register.

Did you verify that the register reads back a 3 if you write a lower
value into the register?

Maybe the most defensive way would be:

+	regmap_read(pca->regmap, PCA9685_PRESCALE, &val);
+	/*
+	 * According to the datasheet, the hardware forces a minimum
+	 * value that can be loaded is 3, so if we read something lower
+	 * assume that the hardware actually implemented a 3.
+	 */
+	if (val < 3)
+		val = 3;
+	state->period = ...
	
Best regards
Uwe

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