Hello,

just a bit of nitpicking left. If Lee is going to apply, I can care for
a followup patch if need be.

On Mon, Sep 07, 2020 at 11:37:55PM +0200, Michael Walle wrote:

[..]
+config PWM_SL28CPLD
+	tristate "Kontron sl28cpld PWM support"
+	depends on MFD_SL28CPLD ||  COMPILE_TEST

s/  / / (@Lee, maybe fixup during application?)

+	help
+	  Generic PWM framework driver for board management controller
+	  found on the Kontron sl28 CPLD.
[...]
+#define SL28CPLD_PWM_CLK			32000 /* 32 kHz */
+#define SL28CPLD_PWM_MAX_DUTY_CYCLE(prescaler)	(1 << (7 - (prescaler)))
+#define SL28CPLD_PWM_PERIOD(prescaler) \
+	(NSEC_PER_SEC / SL28CPLD_PWM_CLK * SL28CPLD_PWM_MAX_DUTY_CYCLE(prescaler))
+
+/*
+ * We calculate the duty cycle like this:
+ *   duty_cycle_ns = pwm_cycle_reg * max_period_ns / max_duty_cycle
+ *
+ * With
+ *   max_period_ns = 1 << (7 - prescaler) / pwm_clk * NSEC_PER_SEC
+ *   max_duty_cycle = 1 << (7 - prescaler)
+ * this then simplifies to:
+ *   duty_cycle_ns = pwm_cycle_reg / pwm_clk * NSEC_PER_SEC
+ *
+ * NSEC_PER_SEC is a multiple of SL28CPLD_PWM_CLK, therefore we're not losing
+ * precision by doing the divison first.

The division you're talking about is NSEC_PER_SEC / pwm_clk which isn't
obvious in the formula in the line above. Maybe:

	...
	this then simplifies to:

	  duty_cycle_ns = NSEC_PER_SEC / SL28CPLD_PWM_CLK * pwm_cycle_reg

	NSEC_PER_SEC is a multiple of SL28CPLD_PWM_CLK, therefor ...

to make it easier to understand the comment.

+ */
+#define SL28CPLD_PWM_TO_DUTY_CYCLE(reg) \
+	(NSEC_PER_SEC / SL28CPLD_PWM_CLK * (reg))
+#define SL28CPLD_PWM_FROM_DUTY_CYCLE(duty_cycle) \
+	(DIV_ROUND_DOWN_ULL((duty_cycle), NSEC_PER_SEC / SL28CPLD_PWM_CLK))
[...]
+	/*
+	 * To avoid glitches when we switch the prescaler, we have to make sure
+	 * we have a valid duty cycle for the new mode.
+	 *
+	 * Take the current prescaler (or the current period length) into
+	 * account to decide whether we have to write the duty cycle or the new
+	 * prescaler first. If the period length is decreasing we have to
+	 * write the duty cycle first.
+	 */
+	write_duty_cycle_first = pwm->state.period > state->period;
+
+	if (write_duty_cycle_first) {
+		ret = sl28cpld_pwm_write(priv, SL28CPLD_PWM_CYCLE, cycle);
+		if (ret)
+			return ret;
+	}
+
+	ret = sl28cpld_pwm_write(priv, SL28CPLD_PWM_CTRL, ctrl);
+	if (ret)
+		return ret;
+
+	if (!write_duty_cycle_first) {
+		ret = sl28cpld_pwm_write(priv, SL28CPLD_PWM_CYCLE, cycle);
+		if (ret)
+			return ret;
+	}

Nice! I didn't spend the necessary brain cycles to confirm this
algorithm, but it seems you did :-)

+
+	return 0;
+}
[...]

Best regards
Uwe

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