On Wed, Aug 06, 2025 at 02:07:15PM +0200, Mathieu Dubois-Briand wrote:

On Fri Aug 1, 2025 at 12:11 PM CEST, Uwe Kleine-König wrote:

quoted

On Tue, Jul 22, 2025 at 06:23:48PM +0200, Mathieu Dubois-Briand wrote:

quoted

+static int max7360_pwm_round_waveform_tohw(struct pwm_chip *chip,
+					   struct pwm_device *pwm,
+					   const struct pwm_waveform *wf,
+					   void *_wfhw)
+{
+	struct max7360_pwm_waveform *wfhw = _wfhw;
+	u64 duty_steps;
+
+	/*
+	 * Ignore user provided values for period_length_ns and duty_offset_ns:
+	 * we only support fixed period of MAX7360_PWM_PERIOD_NS and offset of 0.
+	 * Values from 0 to 254 as duty_steps will provide duty cycles of 0/256
+	 * to 254/256, while value 255 will provide a duty cycle of 100%.
+	 */
+	if (wf->duty_length_ns >= MAX7360_PWM_PERIOD_NS) {
+		duty_steps = MAX7360_PWM_MAX;
+	} else {
+		duty_steps = (u32)wf->duty_length_ns * MAX7360_PWM_STEPS / MAX7360_PWM_PERIOD_NS;
+		if (duty_steps == MAX7360_PWM_MAX)
+			duty_steps = MAX7360_PWM_MAX - 1;
+	}
+
+	wfhw->duty_steps = min(MAX7360_PWM_MAX, duty_steps);
+	wfhw->enabled = !!wf->period_length_ns;
+
+	return 0;

The unconditional return 0 is wrong and testing with PWM_DEBUG enabled
should tell you that.

When you say should, does that mean the current version of PWM core will
tell me that with PWM_DEBUG enabled, or does that mean we should modify
the code so it does show a warning? As I did not see any warning when
specifying a wf->period_length_ns > MAX7360_PWM_PERIOD_NS, even with
PWM_DEBUG enabled.

On the other hand, if I specify a wf->period_length_ns value below
MAX7360_PWM_PERIOD_NS, I indeed get an error:
pwm pwmchip0: Wrong rounding: requested 1000000/1000000 [+0], result 1000000/2000000 [+0]

Yes, that's how I expect it.

quoted

I think the right thing to do here is:

	if (wf->period_length_ns > MAX7360_PWM_PERIOD_NS)
		return 1;
	else
		return 0;

I can definitely do that, but now I'm a bit confused by the meaning of
this return value: is it 0 on success, 1 if some rounding was made,
-errno on error? So I believe I should only return 0 if
wf->period_length_ns == MAX7360_PWM_PERIOD_NS, no?

Or reading this comment on pwm_round_waveform_might_sleep(), maybe we
only have to return 1 if some value is rounded UP. So I believe the test
should be (wf->period_length_ns < MAX7360_PWM_PERIOD_NS).

Right,

	if (wf->period_length_ns < MAX7360_PWM_PERIOD_NS)
		return 1;
	else
		return 0;

So 0 = request could be matched by only rounding down, 1 = request could
be matched but rounding up was needed, negative value = error.

quoted

 * Returns: 0 on success, 1 if at least one value had to be rounded up or a
 * negative errno.

This is kinda confirmed by this other comment, in the code checking the
above returned value in __pwm_apply(), even its just typical examples:

pwm_apply() has different rules. (.apply() fails when .period is too
small. This has the downside that finding a valid period is hard. For
that reason the waveform callbacks round up and signal that by returning
1.)

Best regards
Uwe