[PATCH v4 0/5] gpio: mvebu: pwm fixes and improvements

[PATCH v4 0/5] gpio: mvebu: pwm fixes and improvements

[Baruch Siach]

This series adds a few related fixes to the pwm .apply and .get_state 
callbacks.

The first patch was originally part of the series adding Armada 8K/7K pwm 
support. I split it out to a separate series following review comments from 
Uwe Kleine-König who spotted a few more issues. There is no dependency between 
this and the Armada 8K/7K series.

v4:

  * Take advantage of zero value being treated as 2^32 by hardware. Rewrite 
    patch 5/5 (Uwe).

v3:

  * Improve patch 3/5 description (Uwe)

  * Add more Reviewed-by tags from Uwe

v2:

Address Uwe Kleine-König comments.

  * Improve patch 1/5 summary line

  * Add more information to patch 1/5 description

  * Add more information to patch 2/5 description

  * Don't round period/duty_cycle up in .apply (patch 3/5)

  * Expand the comment in path 5/5 based on RMK's analysis of hardware 
    behaviour

  * Add Uwe's Reviewed-by tags

Baruch Siach (5):
  gpio: mvebu: fix pwm .get_state period calculation
  gpio: mvebu: improve pwm period calculation accuracy
  gpio: mvebu: make pwm .get_state closer to idempotent
  gpio: mvebu: don't limit pwm period/duty_cycle to UINT_MAX
  gpio: mvebu: improve handling of pwm zero on/off values

 drivers/gpio/gpio-mvebu.c | 51 +++++++++++++++++++--------------------
 1 file changed, 25 insertions(+), 26 deletions(-)

-- 
2.29.2

[PATCH v4 1/5] gpio: mvebu: fix pwm .get_state period calculation

[Baruch Siach]

The period is the sum of on and off values. That is, calculate period as

  ($on + $off) / clkrate

instead of

  $off / clkrate - $on / clkrate

that makes no sense.

Reported-by: Russell King <linux@armlinux.org.uk>
Reviewed-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Fixes: 757642f9a584e ("gpio: mvebu: Add limited PWM support")
Signed-off-by: Baruch Siach <baruch@tkos.co.il>
---
 drivers/gpio/gpio-mvebu.c | 19 ++++++++-----------
 1 file changed, 8 insertions(+), 11 deletions(-)

diff --git a/drivers/gpio/gpio-mvebu.c b/drivers/gpio/gpio-mvebu.c
index 672681a976f5..a912a8fed197 100644
--- a/drivers/gpio/gpio-mvebu.c
+++ b/drivers/gpio/gpio-mvebu.c
@@ -676,20 +676,17 @@ static void mvebu_pwm_get_state(struct pwm_chip *chip,
 	else
 		state->duty_cycle = 1;
 
+	val = (unsigned long long) u; /* on duration */
 	regmap_read(mvpwm->regs, mvebu_pwmreg_blink_off_duration(mvpwm), &u);
-	val = (unsigned long long) u * NSEC_PER_SEC;
+	val += (unsigned long long) u; /* period = on + off duration */
+	val *= NSEC_PER_SEC;
 	do_div(val, mvpwm->clk_rate);
-	if (val < state->duty_cycle) {
+	if (val > UINT_MAX)
+		state->period = UINT_MAX;
+	else if (val)
+		state->period = val;
+	else
 		state->period = 1;
-	} else {
-		val -= state->duty_cycle;
-		if (val > UINT_MAX)
-			state->period = UINT_MAX;
-		else if (val)
-			state->period = val;
-		else
-			state->period = 1;
-	}
 
 	regmap_read(mvchip->regs, GPIO_BLINK_EN_OFF + mvchip->offset, &u);
 	if (u)
-- 
2.29.2

Re: [PATCH v4 1/5] gpio: mvebu: fix pwm .get_state period calculation

[Bartosz Golaszewski]

On Sun, Jan 17, 2021 at 2:17 PM Baruch Siach <baruch@tkos.co.il> wrote:
>
> The period is the sum of on and off values. That is, calculate period as
>
>   ($on + $off) / clkrate
>
> instead of
>
>   $off / clkrate - $on / clkrate
>
> that makes no sense.
>
> Reported-by: Russell King <linux@armlinux.org.uk>
> Reviewed-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
> Fixes: 757642f9a584e ("gpio: mvebu: Add limited PWM support")
> Signed-off-by: Baruch Siach <baruch@tkos.co.il>
> ---
>  drivers/gpio/gpio-mvebu.c | 19 ++++++++-----------
>  1 file changed, 8 insertions(+), 11 deletions(-)
>
> diff --git a/drivers/gpio/gpio-mvebu.c b/drivers/gpio/gpio-mvebu.c
> index 672681a976f5..a912a8fed197 100644
> --- a/drivers/gpio/gpio-mvebu.c
> +++ b/drivers/gpio/gpio-mvebu.c
> @@ -676,20 +676,17 @@ static void mvebu_pwm_get_state(struct pwm_chip *chip,
>         else
>                 state->duty_cycle = 1;
>
> +       val = (unsigned long long) u; /* on duration */
>         regmap_read(mvpwm->regs, mvebu_pwmreg_blink_off_duration(mvpwm), &u);
> -       val = (unsigned long long) u * NSEC_PER_SEC;
> +       val += (unsigned long long) u; /* period = on + off duration */
> +       val *= NSEC_PER_SEC;
>         do_div(val, mvpwm->clk_rate);
> -       if (val < state->duty_cycle) {
> +       if (val > UINT_MAX)
> +               state->period = UINT_MAX;
> +       else if (val)
> +               state->period = val;
> +       else
>                 state->period = 1;
> -       } else {
> -               val -= state->duty_cycle;
> -               if (val > UINT_MAX)
> -                       state->period = UINT_MAX;
> -               else if (val)
> -                       state->period = val;
> -               else
> -                       state->period = 1;
> -       }
>
>         regmap_read(mvchip->regs, GPIO_BLINK_EN_OFF + mvchip->offset, &u);
>         if (u)
> --
> 2.29.2
>

I applied this for fixes and will send out a PR to Linus T this week.
Once that's upstream I'll apply the rest for the next release.

Bartosz

[PATCH v4 2/5] gpio: mvebu: improve pwm period calculation accuracy

[Baruch Siach]

Change 'off' register value calculation from

  $off = (period - duty_cycle) * clkrate / NSEC_PER_SEC

to

  $off = (period * clkrate / NSEC_PER_SEC) - $on

That is, divide the full period value to reduce rounding error.

Reported-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Reviewed-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Signed-off-by: Baruch Siach <baruch@tkos.co.il>
---
 drivers/gpio/gpio-mvebu.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/gpio/gpio-mvebu.c b/drivers/gpio/gpio-mvebu.c
index a912a8fed197..c424d88e9e2b 100644
--- a/drivers/gpio/gpio-mvebu.c
+++ b/drivers/gpio/gpio-mvebu.c
@@ -715,9 +715,9 @@ static int mvebu_pwm_apply(struct pwm_chip *chip, struct pwm_device *pwm,
 	else
 		on = 1;
 
-	val = (unsigned long long) mvpwm->clk_rate *
-		(state->period - state->duty_cycle);
+	val = (unsigned long long) mvpwm->clk_rate * state->period;
 	do_div(val, NSEC_PER_SEC);
+	val -= on;
 	if (val > UINT_MAX)
 		return -EINVAL;
 	if (val)
-- 
2.29.2

[PATCH v4 3/5] gpio: mvebu: make pwm .get_state closer to idempotent

[Baruch Siach]

Round up the divisions in .get_state() to make applying the read out
configuration idempotent in most cases as .apply rounds down.

Reported-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Reviewed-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Signed-off-by: Baruch Siach <baruch@tkos.co.il>
---
 drivers/gpio/gpio-mvebu.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/gpio/gpio-mvebu.c b/drivers/gpio/gpio-mvebu.c
index c424d88e9e2b..8673ba77af5a 100644
--- a/drivers/gpio/gpio-mvebu.c
+++ b/drivers/gpio/gpio-mvebu.c
@@ -668,7 +668,7 @@ static void mvebu_pwm_get_state(struct pwm_chip *chip,
 
 	regmap_read(mvpwm->regs, mvebu_pwmreg_blink_on_duration(mvpwm), &u);
 	val = (unsigned long long) u * NSEC_PER_SEC;
-	do_div(val, mvpwm->clk_rate);
+	val = DIV_ROUND_UP_ULL(val, mvpwm->clk_rate);
 	if (val > UINT_MAX)
 		state->duty_cycle = UINT_MAX;
 	else if (val)
@@ -680,7 +680,7 @@ static void mvebu_pwm_get_state(struct pwm_chip *chip,
 	regmap_read(mvpwm->regs, mvebu_pwmreg_blink_off_duration(mvpwm), &u);
 	val += (unsigned long long) u; /* period = on + off duration */
 	val *= NSEC_PER_SEC;
-	do_div(val, mvpwm->clk_rate);
+	val = DIV_ROUND_UP_ULL(val, mvpwm->clk_rate);
 	if (val > UINT_MAX)
 		state->period = UINT_MAX;
 	else if (val)
-- 
2.29.2

[PATCH v4 4/5] gpio: mvebu: don't limit pwm period/duty_cycle to UINT_MAX

[Baruch Siach]

PWM on/off registers are limited to UINT_MAX. However the state period
and duty_cycle fields are ns values of type u64. There is no reason to
limit them to UINT_MAX.

Reported-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Reviewed-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Signed-off-by: Baruch Siach <baruch@tkos.co.il>
---
 drivers/gpio/gpio-mvebu.c | 8 ++------
 1 file changed, 2 insertions(+), 6 deletions(-)

diff --git a/drivers/gpio/gpio-mvebu.c b/drivers/gpio/gpio-mvebu.c
index 8673ba77af5a..6b017854ce61 100644
--- a/drivers/gpio/gpio-mvebu.c
+++ b/drivers/gpio/gpio-mvebu.c
@@ -669,9 +669,7 @@ static void mvebu_pwm_get_state(struct pwm_chip *chip,
 	regmap_read(mvpwm->regs, mvebu_pwmreg_blink_on_duration(mvpwm), &u);
 	val = (unsigned long long) u * NSEC_PER_SEC;
 	val = DIV_ROUND_UP_ULL(val, mvpwm->clk_rate);
-	if (val > UINT_MAX)
-		state->duty_cycle = UINT_MAX;
-	else if (val)
+	if (val)
 		state->duty_cycle = val;
 	else
 		state->duty_cycle = 1;
@@ -681,9 +679,7 @@ static void mvebu_pwm_get_state(struct pwm_chip *chip,
 	val += (unsigned long long) u; /* period = on + off duration */
 	val *= NSEC_PER_SEC;
 	val = DIV_ROUND_UP_ULL(val, mvpwm->clk_rate);
-	if (val > UINT_MAX)
-		state->period = UINT_MAX;
-	else if (val)
+	if (val)
 		state->period = val;
 	else
 		state->period = 1;
-- 
2.29.2

[PATCH v4 5/5] gpio: mvebu: improve handling of pwm zero on/off values

[Baruch Siach]

Hardware appears to treat zero value as 2^32. Take advantage of this
fact to support on/off values of up to UINT_MAX+1 == 2^32. Adjust both
.apply and .get_sate to handle zero as a special case.

Rounded up division result in .get_state can't be zero, since the
dividend is now larger than 0. Remove check for this case.

Reported-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Analyzed-by: Russell King <linux@armlinux.org.uk>
Signed-off-by: Baruch Siach <baruch@tkos.co.il>
---
 drivers/gpio/gpio-mvebu.c | 38 ++++++++++++++++++++++----------------
 1 file changed, 22 insertions(+), 16 deletions(-)

diff --git a/drivers/gpio/gpio-mvebu.c b/drivers/gpio/gpio-mvebu.c
index 6b017854ce61..37f5bd65062f 100644
--- a/drivers/gpio/gpio-mvebu.c
+++ b/drivers/gpio/gpio-mvebu.c
@@ -667,22 +667,20 @@ static void mvebu_pwm_get_state(struct pwm_chip *chip,
 	spin_lock_irqsave(&mvpwm->lock, flags);
 
 	regmap_read(mvpwm->regs, mvebu_pwmreg_blink_on_duration(mvpwm), &u);
-	val = (unsigned long long) u * NSEC_PER_SEC;
-	val = DIV_ROUND_UP_ULL(val, mvpwm->clk_rate);
-	if (val)
-		state->duty_cycle = val;
+	/* Hardware treats zero as 2^32. See mvebu_pwm_apply(). */
+	if (u > 0)
+		val = (unsigned long long) u * NSEC_PER_SEC;
 	else
-		state->duty_cycle = 1;
+		val = ((unsigned long long) UINT_MAX+1) * NSEC_PER_SEC;
+	state->duty_cycle = DIV_ROUND_UP_ULL(val, mvpwm->clk_rate);
 
-	val = (unsigned long long) u; /* on duration */
 	regmap_read(mvpwm->regs, mvebu_pwmreg_blink_off_duration(mvpwm), &u);
-	val += (unsigned long long) u; /* period = on + off duration */
-	val *= NSEC_PER_SEC;
-	val = DIV_ROUND_UP_ULL(val, mvpwm->clk_rate);
-	if (val)
-		state->period = val;
+	/* period = on + off duration */
+	if (u > 0)
+		val += (unsigned long long) u * NSEC_PER_SEC;
 	else
-		state->period = 1;
+		val += ((unsigned long long) UINT_MAX+1) * NSEC_PER_SEC;
+	state->period = DIV_ROUND_UP_ULL(val, mvpwm->clk_rate);
 
 	regmap_read(mvchip->regs, GPIO_BLINK_EN_OFF + mvchip->offset, &u);
 	if (u)
@@ -704,9 +702,15 @@ static int mvebu_pwm_apply(struct pwm_chip *chip, struct pwm_device *pwm,
 
 	val = (unsigned long long) mvpwm->clk_rate * state->duty_cycle;
 	do_div(val, NSEC_PER_SEC);
-	if (val > UINT_MAX)
+	if (val > (unsigned long long) UINT_MAX+1)
 		return -EINVAL;
-	if (val)
+	/*
+	 * Zero on/off values don't work as expected. Experimentation shows
+	 * that zero value is treated as 2^32. This behavior is not documented.
+	 */
+	if (val == (unsigned long long) UINT_MAX+1)
+		on = 0;
+	else if (val)
 		on = val;
 	else
 		on = 1;
@@ -714,9 +718,11 @@ static int mvebu_pwm_apply(struct pwm_chip *chip, struct pwm_device *pwm,
 	val = (unsigned long long) mvpwm->clk_rate * state->period;
 	do_div(val, NSEC_PER_SEC);
 	val -= on;
-	if (val > UINT_MAX)
+	if (val > (unsigned long long) UINT_MAX+1)
 		return -EINVAL;
-	if (val)
+	if (val == (unsigned long long) UINT_MAX+1)
+		off = 0;
+	else if (val)
 		off = val;
 	else
 		off = 1;
-- 
2.29.2

Re: [PATCH v4 5/5] gpio: mvebu: improve handling of pwm zero on/off values

[Uwe Kleine-König]

[-- Attachment #1: Type: text/plain, Size: 3159 bytes --]

Hello Baruch,

On Sun, Jan 17, 2021 at 03:17:06PM +0200, Baruch Siach wrote:
> Hardware appears to treat zero value as 2^32. Take advantage of this
> fact to support on/off values of up to UINT_MAX+1 == 2^32. Adjust both
> .apply and .get_sate to handle zero as a special case.

s/get_sate/get_state/

> Rounded up division result in .get_state can't be zero, since the
> dividend is now larger than 0. Remove check for this case.
> 
> Reported-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
> Analyzed-by: Russell King <linux@armlinux.org.uk>
> Signed-off-by: Baruch Siach <baruch@tkos.co.il>
> ---
>  drivers/gpio/gpio-mvebu.c | 38 ++++++++++++++++++++++----------------
>  1 file changed, 22 insertions(+), 16 deletions(-)
> 
> diff --git a/drivers/gpio/gpio-mvebu.c b/drivers/gpio/gpio-mvebu.c
> index 6b017854ce61..37f5bd65062f 100644
> --- a/drivers/gpio/gpio-mvebu.c
> +++ b/drivers/gpio/gpio-mvebu.c
> @@ -667,22 +667,20 @@ static void mvebu_pwm_get_state(struct pwm_chip *chip,
>  	spin_lock_irqsave(&mvpwm->lock, flags);
>  
>  	regmap_read(mvpwm->regs, mvebu_pwmreg_blink_on_duration(mvpwm), &u);
> -	val = (unsigned long long) u * NSEC_PER_SEC;
> -	val = DIV_ROUND_UP_ULL(val, mvpwm->clk_rate);
> -	if (val)
> -		state->duty_cycle = val;
> +	/* Hardware treats zero as 2^32. See mvebu_pwm_apply(). */
> +	if (u > 0)
> +		val = (unsigned long long) u * NSEC_PER_SEC;
>  	else
> -		state->duty_cycle = 1;
> +		val = ((unsigned long long) UINT_MAX+1) * NSEC_PER_SEC;
> +	state->duty_cycle = DIV_ROUND_UP_ULL(val, mvpwm->clk_rate);

Instead of

	if (u > 0)
		val = (unsigned long long) u * NSEC_PER_SEC;
	else
		val = ((unsigned long long) UINT_MAX+1) * NSEC_PER_SEC;

	state->duty_cycle = DIV_ROUND_UP_ULL(val, mvpwm->clk_rate);

you could also write

	if (u > 0)
		val = u;
	else
		val = UINT_MAX + 1;

	state->duty_cycle = DIV_ROUND_UP_ULL(val * NSEC_PER_SEC, mvpwm->clk_rate);

which is a bit lighter (IMHO).

>  
> -	val = (unsigned long long) u; /* on duration */
>  	regmap_read(mvpwm->regs, mvebu_pwmreg_blink_off_duration(mvpwm), &u);
> -	val += (unsigned long long) u; /* period = on + off duration */
> -	val *= NSEC_PER_SEC;
> -	val = DIV_ROUND_UP_ULL(val, mvpwm->clk_rate);
> -	if (val)
> -		state->period = val;
> +	/* period = on + off duration */
> +	if (u > 0)
> +		val += (unsigned long long) u * NSEC_PER_SEC;
>  	else
> -		state->period = 1;
> +		val += ((unsigned long long) UINT_MAX+1) * NSEC_PER_SEC;
> +	state->period = DIV_ROUND_UP_ULL(val, mvpwm->clk_rate);

>  
>  	regmap_read(mvchip->regs, GPIO_BLINK_EN_OFF + mvchip->offset, &u);
>  	if (u)
> @@ -704,9 +702,15 @@ static int mvebu_pwm_apply(struct pwm_chip *chip, struct pwm_device *pwm,
>  
>  	val = (unsigned long long) mvpwm->clk_rate * state->duty_cycle;
>  	do_div(val, NSEC_PER_SEC);
> -	if (val > UINT_MAX)
> +	if (val > (unsigned long long) UINT_MAX+1)

Please add whitespace around the +

Best regards
Uwe

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

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 488 bytes --]