pwm: samsung: Fix output race on disabling

When disabling the Samsung PWM the output state remains at the level it
was at the end of a PWM cycle. In other words, calling pwm_disable()
when at 100% duty cycle will keep the output active, while at all other
settings the output will go/stay inactive. On top of that the Samsung
PWM settings are double-buffered, which means the new settings only get
applied at the start of a new PWM cycle.

This results in a race if the PWM is at 100% duty cycle and a driver
calls:

  pwm_config(pwm, 0, period);
  pwm_disable(pwm);

In this case the PWMs output will unexpectedly stay active, unless a new
PWM cycle happened to start between the register writes in pwm_config()
and pwm_disable(). As far as I can tell this is a regression introduced
by 3bdf878, before that a call to pwm_config() would call
pwm_samsung_enable() which, while heavy-handed, made sure the expected
settings were live.

To resolve this, while not re-introducing the issues 3bdf878 (flickering
as the PWM got reset while in a PWM cycle) fixed, only force an update
of the settings when at 100% duty cycle, which shouldn't have any
noticeable effect on the output but is enough to ensure the behaviour is
as expected on disable.

Signed-off-by: Sjoerd Simons <sjoerd.simons@collabora.co.uk>
Reviewed-by: Javier Martinez Canillas <javier.martinez@collabora.co.uk>
Acked-by: Lukasz Majewski <l.majewski@samsung.com>
Signed-off-by: Thierry Reding <thierry.reding@gmail.com>
diff --git a/drivers/pwm/pwm-samsung.c b/drivers/pwm/pwm-samsung.c
index 3e9b583..ff201e1 100644
--- a/drivers/pwm/pwm-samsung.c
+++ b/drivers/pwm/pwm-samsung.c
@@ -269,12 +269,31 @@
 	spin_unlock_irqrestore(&samsung_pwm_lock, flags);
 }
 
+static void pwm_samsung_manual_update(struct samsung_pwm_chip *chip,
+				      struct pwm_device *pwm)
+{
+	unsigned int tcon_chan = to_tcon_channel(pwm->hwpwm);
+	u32 tcon;
+	unsigned long flags;
+
+	spin_lock_irqsave(&samsung_pwm_lock, flags);
+
+	tcon = readl(chip->base + REG_TCON);
+	tcon |= TCON_MANUALUPDATE(tcon_chan);
+	writel(tcon, chip->base + REG_TCON);
+
+	tcon &= ~TCON_MANUALUPDATE(tcon_chan);
+	writel(tcon, chip->base + REG_TCON);
+
+	spin_unlock_irqrestore(&samsung_pwm_lock, flags);
+}
+
 static int pwm_samsung_config(struct pwm_chip *chip, struct pwm_device *pwm,
 			      int duty_ns, int period_ns)
 {
 	struct samsung_pwm_chip *our_chip = to_samsung_pwm_chip(chip);
 	struct samsung_pwm_channel *chan = pwm_get_chip_data(pwm);
-	u32 tin_ns = chan->tin_ns, tcnt, tcmp;
+	u32 tin_ns = chan->tin_ns, tcnt, tcmp, oldtcmp;
 
 	/*
 	 * We currently avoid using 64bit arithmetic by using the
@@ -288,6 +307,7 @@
 		return 0;
 
 	tcnt = readl(our_chip->base + REG_TCNTB(pwm->hwpwm));
+	oldtcmp = readl(our_chip->base + REG_TCMPB(pwm->hwpwm));
 
 	/* We need tick count for calculation, not last tick. */
 	++tcnt;
@@ -335,6 +355,16 @@
 	writel(tcnt, our_chip->base + REG_TCNTB(pwm->hwpwm));
 	writel(tcmp, our_chip->base + REG_TCMPB(pwm->hwpwm));
 
+	/*
+	 * In case the PWM is currently at 100% duty cycle, force a manual
+	 * update to prevent the signal staying high if the PWM is disabled
+	 * shortly afer this update (before it autoreloaded the new values).
+	 */
+	if (oldtcmp == (u32) -1) {
+		dev_dbg(our_chip->chip.dev, "Forcing manual update");
+		pwm_samsung_manual_update(our_chip, pwm);
+	}
+
 	chan->period_ns = period_ns;
 	chan->tin_ns = tin_ns;
 	chan->duty_ns = duty_ns;