pwm-sun4i: Calculate the delay without rounding down to jiffies

This fixes a problem that was supposed to be addressed by commit
6eefb79d6f ("pwm: sun4i: Remove erroneous else branch") - backlight
could not be switched off on some Allwinner A20.  The commit was
correct, but was not a reliable fix for the problem, which was timing
related.

The real problem for the backlight switching problem was that sleeping
for a full period did not work, because delay_us is always zero.

It is zero because the period (plus 1 microsecond) is rounded down to
the next "jiffies", but the period is less than one jiffy.

On my Cubieboard 2, the period is 5ms, and 1 jiffy (at the default
HZ=100) is 10ms, so nsecs_to_jiffies(10ms+1us)=0.

The roundtrip from nanoseconds to jiffies and back to microseconds is
an unnecessary loss of precision; always rounding down (via
nsecs_to_jiffies()) then causes the breakage.

This patch eliminates this roundtrip, and directly converts from
nanoseconds to microseconds (for usleep_range()), using
DIV_ROUND_UP_ULL() to force rounding up.  This way, the sleep time is
never zero, and after the sleep, we are guaranteed to be in a
different period, and the device is ready for another control command
for sure.

Signed-off-by: Max Kellermann <max.kellermann@gmail.com>
Acked-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Signed-off-by: Thierry Reding <thierry.reding@gmail.com>

drivers/pwm/pwm-sun4i.c

@@ -234,7 +234,6 @@ static int sun4i_pwm_apply(struct pwm_chip *chip, struct pwm_device *pwm,
 	struct pwm_state cstate;
 	u32 ctrl, duty = 0, period = 0, val;
 	int ret;
-	unsigned int delay_jiffies;
 	unsigned int delay_us, prescaler = 0;
 	bool bypass;
 
@@ -302,8 +301,7 @@ static int sun4i_pwm_apply(struct pwm_chip *chip, struct pwm_device *pwm,
 		return 0;
 
 	/* We need a full period to elapse before disabling the channel. */
-	delay_jiffies = nsecs_to_jiffies(cstate.period + 1000);
-	delay_us = jiffies_to_usecs(delay_jiffies);
+	delay_us = DIV_ROUND_UP_ULL(cstate.period, NSEC_PER_USEC);
 	if ((delay_us / 500) > MAX_UDELAY_MS)
 		msleep(delay_us / 1000 + 1);
 	else