pwm: imx: don't reprogram PWMSAR if PWM is disabled

Submitted by Dmitry Eremin-Solenikov on July 23, 2014, 8:09 a.m.

Details

Message ID 1406102987-14797-1-git-send-email-dbaryshkov@gmail.com
State Superseded
Headers show

Commit Message

Dmitry Eremin-Solenikov July 23, 2014, 8:09 a.m.
From: Dmitry Eremin-Solenikov <dmitry_eremin@mentor.com>

Writing several values to PWMSAR register with PWM being disabled can
lead to FIFO (connected to PWMSAR) being overflown. Then after enabling
PWM, hardware will use stale values. Instead cache the duty cycles and
write them to the hardware only before enabling PWM.

Signed-off-by: Dmitry Eremin-Solenikov <dmitry_eremin@mentor.com>
---
 drivers/pwm/pwm-imx.c | 11 ++++++++---
 1 file changed, 8 insertions(+), 3 deletions(-)

Comments

Dmitry Eremin-Solenikov Aug. 1, 2014, 6:02 p.m.
On Wed, Jul 23, 2014 at 12:09 PM, Dmitry Eremin-Solenikov
<dbaryshkov@gmail.com> wrote:
> From: Dmitry Eremin-Solenikov <dmitry_eremin@mentor.com>
>
> Writing several values to PWMSAR register with PWM being disabled can
> lead to FIFO (connected to PWMSAR) being overflown. Then after enabling
> PWM, hardware will use stale values. Instead cache the duty cycles and
> write them to the hardware only before enabling PWM.

What about this patch?

>
> Signed-off-by: Dmitry Eremin-Solenikov <dmitry_eremin@mentor.com>
> ---
>  drivers/pwm/pwm-imx.c | 11 ++++++++---
>  1 file changed, 8 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/pwm/pwm-imx.c b/drivers/pwm/pwm-imx.c
> index c735127..79c2b24 100644
> --- a/drivers/pwm/pwm-imx.c
> +++ b/drivers/pwm/pwm-imx.c
> @@ -46,6 +46,7 @@ struct imx_chip {
>         struct clk      *clk_ipg;
>
>         void __iomem    *mmio_base;
> +       unsigned long   duty_cycles;
>
>         struct pwm_chip chip;
>
> @@ -105,7 +106,7 @@ static int imx_pwm_config_v2(struct pwm_chip *chip,
>  {
>         struct imx_chip *imx = to_imx_chip(chip);
>         unsigned long long c;
> -       unsigned long period_cycles, duty_cycles, prescale;
> +       unsigned long period_cycles, prescale;
>         u32 cr;
>
>         c = clk_get_rate(imx->clk_per);
> @@ -118,7 +119,7 @@ static int imx_pwm_config_v2(struct pwm_chip *chip,
>         period_cycles /= prescale;
>         c = (unsigned long long)period_cycles * duty_ns;
>         do_div(c, period_ns);
> -       duty_cycles = c;
> +       imx->duty_cycles = c;
>
>         /*
>          * according to imx pwm RM, the real period value should be
> @@ -134,7 +135,8 @@ static int imx_pwm_config_v2(struct pwm_chip *chip,
>
>         period_cycles -= 2;
>
> -       writel(duty_cycles, imx->mmio_base + MX3_PWMSAR);
> +       if (test_bit(PWMF_ENABLED, &pwm->flags))
> +               writel(imx->duty_cycles, imx->mmio_base + MX3_PWMSAR);
>         writel(period_cycles, imx->mmio_base + MX3_PWMPR);
>
>         cr = readl(imx->mmio_base + MX3_PWMCR);
> @@ -157,6 +159,9 @@ static void imx_pwm_set_enable_v2(struct pwm_chip *chip, bool enable)
>         struct imx_chip *imx = to_imx_chip(chip);
>         u32 val;
>
> +       if (enable)
> +               writel(imx->duty_cycles, imx->mmio_base + MX3_PWMSAR);
> +
>         val = readl(imx->mmio_base + MX3_PWMCR);
>
>         if (enable)
> --
> 1.9.3
>
--
To unsubscribe from this list: send the line "unsubscribe linux-pwm" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Dmitry Eremin-Solenikov Aug. 5, 2014, 12:47 a.m.
On Fri, Aug 1, 2014 at 10:02 PM, Dmitry Eremin-Solenikov
<dmitry_eremin@mentor.com> wrote:
> On Wed, Jul 23, 2014 at 12:09 PM, Dmitry Eremin-Solenikov
> <dbaryshkov@gmail.com> wrote:
>> From: Dmitry Eremin-Solenikov <dmitry_eremin@mentor.com>
>>
>> Writing several values to PWMSAR register with PWM being disabled can
>> lead to FIFO (connected to PWMSAR) being overflown. Then after enabling
>> PWM, hardware will use stale values. Instead cache the duty cycles and
>> write them to the hardware only before enabling PWM.
>
> What about this patch?

For the reference: the issue with the FIFO can be easily demonstrated on i.MX6
with the help of pwm-backlight. Steps to reproduce:

1) Disable backlight: echo 1 > /sys/class/..../bl_power
2) Change config several times:
  echo 1 > /sys/class/.../brightness
  echo 2 > /sys/class/.../brightness
  echo 3 > /sys/class/.../brightness
  echo 4 > /sys/class/.../brightness
  echo 7 > /sys/class/.../brightness

3) Reenable backlight and pwm:
  echo 0 > /sys/class/.../bl_power

At this point you get PWMSAR = 0, because of FIFO overflow (can be
verified by reading PWM status register).
Shawn Guo Aug. 5, 2014, 1:36 a.m.
On Fri, Aug 01, 2014 at 10:02:17PM +0400, Dmitry Eremin-Solenikov wrote:
> On Wed, Jul 23, 2014 at 12:09 PM, Dmitry Eremin-Solenikov
> <dbaryshkov@gmail.com> wrote:
> > From: Dmitry Eremin-Solenikov <dmitry_eremin@mentor.com>
> >
> > Writing several values to PWMSAR register with PWM being disabled can
> > lead to FIFO (connected to PWMSAR) being overflown. Then after enabling
> > PWM, hardware will use stale values. Instead cache the duty cycles and
> > write them to the hardware only before enabling PWM.
> 
> What about this patch?

Copy Liu Ying who seems to have a patch [1] addressing the same problem?

Shawn

[1] http://thread.gmane.org/gmane.linux.pwm/837/focus=836

> 
> >
> > Signed-off-by: Dmitry Eremin-Solenikov <dmitry_eremin@mentor.com>
> > ---
> >  drivers/pwm/pwm-imx.c | 11 ++++++++---
> >  1 file changed, 8 insertions(+), 3 deletions(-)
> >
> > diff --git a/drivers/pwm/pwm-imx.c b/drivers/pwm/pwm-imx.c
> > index c735127..79c2b24 100644
> > --- a/drivers/pwm/pwm-imx.c
> > +++ b/drivers/pwm/pwm-imx.c
> > @@ -46,6 +46,7 @@ struct imx_chip {
> >         struct clk      *clk_ipg;
> >
> >         void __iomem    *mmio_base;
> > +       unsigned long   duty_cycles;
> >
> >         struct pwm_chip chip;
> >
> > @@ -105,7 +106,7 @@ static int imx_pwm_config_v2(struct pwm_chip *chip,
> >  {
> >         struct imx_chip *imx = to_imx_chip(chip);
> >         unsigned long long c;
> > -       unsigned long period_cycles, duty_cycles, prescale;
> > +       unsigned long period_cycles, prescale;
> >         u32 cr;
> >
> >         c = clk_get_rate(imx->clk_per);
> > @@ -118,7 +119,7 @@ static int imx_pwm_config_v2(struct pwm_chip *chip,
> >         period_cycles /= prescale;
> >         c = (unsigned long long)period_cycles * duty_ns;
> >         do_div(c, period_ns);
> > -       duty_cycles = c;
> > +       imx->duty_cycles = c;
> >
> >         /*
> >          * according to imx pwm RM, the real period value should be
> > @@ -134,7 +135,8 @@ static int imx_pwm_config_v2(struct pwm_chip *chip,
> >
> >         period_cycles -= 2;
> >
> > -       writel(duty_cycles, imx->mmio_base + MX3_PWMSAR);
> > +       if (test_bit(PWMF_ENABLED, &pwm->flags))
> > +               writel(imx->duty_cycles, imx->mmio_base + MX3_PWMSAR);
> >         writel(period_cycles, imx->mmio_base + MX3_PWMPR);
> >
> >         cr = readl(imx->mmio_base + MX3_PWMCR);
> > @@ -157,6 +159,9 @@ static void imx_pwm_set_enable_v2(struct pwm_chip *chip, bool enable)
> >         struct imx_chip *imx = to_imx_chip(chip);
> >         u32 val;
> >
> > +       if (enable)
> > +               writel(imx->duty_cycles, imx->mmio_base + MX3_PWMSAR);
> > +
> >         val = readl(imx->mmio_base + MX3_PWMCR);
> >
> >         if (enable)
> > --
> > 1.9.3
> >
--
To unsubscribe from this list: send the line "unsubscribe linux-pwm" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Liu Ying Aug. 5, 2014, 8:48 a.m.
On 08/05/2014 09:36 AM, Shawn Guo wrote:
> On Fri, Aug 01, 2014 at 10:02:17PM +0400, Dmitry Eremin-Solenikov wrote:
>> On Wed, Jul 23, 2014 at 12:09 PM, Dmitry Eremin-Solenikov
>> <dbaryshkov@gmail.com> wrote:
>>> From: Dmitry Eremin-Solenikov <dmitry_eremin@mentor.com>
>>>
>>> Writing several values to PWMSAR register with PWM being disabled can
>>> lead to FIFO (connected to PWMSAR) being overflown. Then after enabling
>>> PWM, hardware will use stale values. Instead cache the duty cycles and
>>> write them to the hardware only before enabling PWM.
>>
>> What about this patch?
>
> Copy Liu Ying who seems to have a patch [1] addressing the same problem?
>

Yes, my patch may address the same problem. And, my patch may cache the 
last duty cycle as well when the PWM is disabled. The difference is that 
my patch caches it in the register PWMSAR instead of a variable.

Regards,

Liu Ying

> Shawn
>
> [1] http://thread.gmane.org/gmane.linux.pwm/837/focus=836
>
>>
>>>
>>> Signed-off-by: Dmitry Eremin-Solenikov <dmitry_eremin@mentor.com>
>>> ---
>>>   drivers/pwm/pwm-imx.c | 11 ++++++++---
>>>   1 file changed, 8 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/drivers/pwm/pwm-imx.c b/drivers/pwm/pwm-imx.c
>>> index c735127..79c2b24 100644
>>> --- a/drivers/pwm/pwm-imx.c
>>> +++ b/drivers/pwm/pwm-imx.c
>>> @@ -46,6 +46,7 @@ struct imx_chip {
>>>          struct clk      *clk_ipg;
>>>
>>>          void __iomem    *mmio_base;
>>> +       unsigned long   duty_cycles;
>>>
>>>          struct pwm_chip chip;
>>>
>>> @@ -105,7 +106,7 @@ static int imx_pwm_config_v2(struct pwm_chip *chip,
>>>   {
>>>          struct imx_chip *imx = to_imx_chip(chip);
>>>          unsigned long long c;
>>> -       unsigned long period_cycles, duty_cycles, prescale;
>>> +       unsigned long period_cycles, prescale;
>>>          u32 cr;
>>>
>>>          c = clk_get_rate(imx->clk_per);
>>> @@ -118,7 +119,7 @@ static int imx_pwm_config_v2(struct pwm_chip *chip,
>>>          period_cycles /= prescale;
>>>          c = (unsigned long long)period_cycles * duty_ns;
>>>          do_div(c, period_ns);
>>> -       duty_cycles = c;
>>> +       imx->duty_cycles = c;
>>>
>>>          /*
>>>           * according to imx pwm RM, the real period value should be
>>> @@ -134,7 +135,8 @@ static int imx_pwm_config_v2(struct pwm_chip *chip,
>>>
>>>          period_cycles -= 2;
>>>
>>> -       writel(duty_cycles, imx->mmio_base + MX3_PWMSAR);
>>> +       if (test_bit(PWMF_ENABLED, &pwm->flags))
>>> +               writel(imx->duty_cycles, imx->mmio_base + MX3_PWMSAR);
>>>          writel(period_cycles, imx->mmio_base + MX3_PWMPR);
>>>
>>>          cr = readl(imx->mmio_base + MX3_PWMCR);
>>> @@ -157,6 +159,9 @@ static void imx_pwm_set_enable_v2(struct pwm_chip *chip, bool enable)
>>>          struct imx_chip *imx = to_imx_chip(chip);
>>>          u32 val;
>>>
>>> +       if (enable)
>>> +               writel(imx->duty_cycles, imx->mmio_base + MX3_PWMSAR);
>>> +
>>>          val = readl(imx->mmio_base + MX3_PWMCR);
>>>
>>>          if (enable)
>>> --
>>> 1.9.3
>>>
--
To unsubscribe from this list: send the line "unsubscribe linux-pwm" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Thierry Reding Aug. 25, 2014, 1:56 p.m.
On Tue, Aug 05, 2014 at 04:48:41PM +0800, Liu Ying wrote:
> On 08/05/2014 09:36 AM, Shawn Guo wrote:
> >On Fri, Aug 01, 2014 at 10:02:17PM +0400, Dmitry Eremin-Solenikov wrote:
> >>On Wed, Jul 23, 2014 at 12:09 PM, Dmitry Eremin-Solenikov
> >><dbaryshkov@gmail.com> wrote:
> >>>From: Dmitry Eremin-Solenikov <dmitry_eremin@mentor.com>
> >>>
> >>>Writing several values to PWMSAR register with PWM being disabled can
> >>>lead to FIFO (connected to PWMSAR) being overflown. Then after enabling
> >>>PWM, hardware will use stale values. Instead cache the duty cycles and
> >>>write them to the hardware only before enabling PWM.
> >>
> >>What about this patch?
> >
> >Copy Liu Ying who seems to have a patch [1] addressing the same problem?
> >
> 
> Yes, my patch may address the same problem. And, my patch may cache the last
> duty cycle as well when the PWM is disabled. The difference is that my patch
> caches it in the register PWMSAR instead of a variable.

Dmitry,

I've just pushed Liu's patch that might fix this problem. Can you give
it a quick spin to see if it fixes the issue that you're seeing? It's in
the for-next branch of the PWM tree[0].

Thanks,
Thierry

[0]: git://git.kernel.org/pub/scm/linux/kernel/git/thierry.reding/linux-pwm.git

Patch hide | download patch | download mbox

diff --git a/drivers/pwm/pwm-imx.c b/drivers/pwm/pwm-imx.c
index c735127..79c2b24 100644
--- a/drivers/pwm/pwm-imx.c
+++ b/drivers/pwm/pwm-imx.c
@@ -46,6 +46,7 @@  struct imx_chip {
 	struct clk	*clk_ipg;
 
 	void __iomem	*mmio_base;
+	unsigned long	duty_cycles;
 
 	struct pwm_chip	chip;
 
@@ -105,7 +106,7 @@  static int imx_pwm_config_v2(struct pwm_chip *chip,
 {
 	struct imx_chip *imx = to_imx_chip(chip);
 	unsigned long long c;
-	unsigned long period_cycles, duty_cycles, prescale;
+	unsigned long period_cycles, prescale;
 	u32 cr;
 
 	c = clk_get_rate(imx->clk_per);
@@ -118,7 +119,7 @@  static int imx_pwm_config_v2(struct pwm_chip *chip,
 	period_cycles /= prescale;
 	c = (unsigned long long)period_cycles * duty_ns;
 	do_div(c, period_ns);
-	duty_cycles = c;
+	imx->duty_cycles = c;
 
 	/*
 	 * according to imx pwm RM, the real period value should be
@@ -134,7 +135,8 @@  static int imx_pwm_config_v2(struct pwm_chip *chip,
 
 	period_cycles -= 2;
 
-	writel(duty_cycles, imx->mmio_base + MX3_PWMSAR);
+	if (test_bit(PWMF_ENABLED, &pwm->flags))
+		writel(imx->duty_cycles, imx->mmio_base + MX3_PWMSAR);
 	writel(period_cycles, imx->mmio_base + MX3_PWMPR);
 
 	cr = readl(imx->mmio_base + MX3_PWMCR);
@@ -157,6 +159,9 @@  static void imx_pwm_set_enable_v2(struct pwm_chip *chip, bool enable)
 	struct imx_chip *imx = to_imx_chip(chip);
 	u32 val;
 
+	if (enable)
+		writel(imx->duty_cycles, imx->mmio_base + MX3_PWMSAR);
+
 	val = readl(imx->mmio_base + MX3_PWMCR);
 
 	if (enable)