[PATCH] watchdog: s3c2410_wdt: Only register for cpufreq on CPU_FREQ_S3C24XX

[PATCH] watchdog: s3c2410_wdt: Only register for cpufreq on CPU_FREQ_S3C24XX

[Doug Anderson]

On modern SoCs the watchdog timer is parented on a clock that doesn't
change every time we have a cpufreq change.  That means we don't need
to constantly adjust the watchdog timer, so avoid registering for and
dealing with cpufreq transitions unless we've actually got
CPU_FREQ_S3C24XX defined.

Note that this is more than just an optimization.  The s3c2410
watchdog driver actually pats the watchdog on every CPU frequency
change.  On modern systems these happen many times per second (even in
a system where "nothing" is happening).  That effectively makes any
userspace watchdog program useless (the watchdog is constantly patted
by the kernel).  If we need CPU_FREQ_S3C24XX defined on a
multiplatform kernel we'll need to make sure that kernel supports
common clock and change this to user common clock framework.

Signed-off-by: Doug Anderson <dianders@chromium.org>
---
 drivers/watchdog/s3c2410_wdt.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/watchdog/s3c2410_wdt.c b/drivers/watchdog/s3c2410_wdt.c
index 7d8fd04..4980f84 100644
--- a/drivers/watchdog/s3c2410_wdt.c
+++ b/drivers/watchdog/s3c2410_wdt.c
@@ -264,7 +264,7 @@ static irqreturn_t s3c2410wdt_irq(int irqno, void *param)
 	return IRQ_HANDLED;
 }
 
-#ifdef CONFIG_CPU_FREQ
+#ifdef CONFIG_CPU_FREQ_S3C24XX
 
 static int s3c2410wdt_cpufreq_transition(struct notifier_block *nb,
 					  unsigned long val, void *data)
-- 
1.8.4.1

Re: [PATCH] watchdog: s3c2410_wdt: Only register for cpufreq on CPU_FREQ_S3C24XX

[Guenter Roeck]

On 11/25/2013 02:55 PM, Doug Anderson wrote:
> On modern SoCs the watchdog timer is parented on a clock that doesn't
> change every time we have a cpufreq change.  That means we don't need
> to constantly adjust the watchdog timer, so avoid registering for and
> dealing with cpufreq transitions unless we've actually got
> CPU_FREQ_S3C24XX defined.
>
> Note that this is more than just an optimization.  The s3c2410
> watchdog driver actually pats the watchdog on every CPU frequency
> change.  On modern systems these happen many times per second (even in
> a system where "nothing" is happening).  That effectively makes any
> userspace watchdog program useless (the watchdog is constantly patted
> by the kernel).  If we need CPU_FREQ_S3C24XX defined on a
> multiplatform kernel we'll need to make sure that kernel supports
> common clock and change this to user common clock framework.
>
> Signed-off-by: Doug Anderson <dianders@chromium.org>
> ---
>   drivers/watchdog/s3c2410_wdt.c | 2 +-
>   1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/watchdog/s3c2410_wdt.c b/drivers/watchdog/s3c2410_wdt.c
> index 7d8fd04..4980f84 100644
> --- a/drivers/watchdog/s3c2410_wdt.c
> +++ b/drivers/watchdog/s3c2410_wdt.c
> @@ -264,7 +264,7 @@ static irqreturn_t s3c2410wdt_irq(int irqno, void *param)
>   	return IRQ_HANDLED;
>   }
>
> -#ifdef CONFIG_CPU_FREQ
> +#ifdef CONFIG_CPU_FREQ_S3C24XX
>
Where is the CPU_FREQ_S3C24XX configuration option defined ? I don't see it
in the current upstream kernel, so it appears that this depends on some
out-of-tree changes.

Thanks,
Guenter

Re: [PATCH] watchdog: s3c2410_wdt: Only register for cpufreq on CPU_FREQ_S3C24XX

[Tomasz Figa]

On Monday 25 of November 2013 15:23:27 Guenter Roeck wrote:
> On 11/25/2013 02:55 PM, Doug Anderson wrote:
> > On modern SoCs the watchdog timer is parented on a clock that doesn't
> > change every time we have a cpufreq change.  That means we don't need
> > to constantly adjust the watchdog timer, so avoid registering for and
> > dealing with cpufreq transitions unless we've actually got
> > CPU_FREQ_S3C24XX defined.
> >
> > Note that this is more than just an optimization.  The s3c2410
> > watchdog driver actually pats the watchdog on every CPU frequency
> > change.  On modern systems these happen many times per second (even in
> > a system where "nothing" is happening).  That effectively makes any
> > userspace watchdog program useless (the watchdog is constantly patted
> > by the kernel).  If we need CPU_FREQ_S3C24XX defined on a
> > multiplatform kernel we'll need to make sure that kernel supports
> > common clock and change this to user common clock framework.
> >
> > Signed-off-by: Doug Anderson <dianders@chromium.org>
> > ---
> >   drivers/watchdog/s3c2410_wdt.c | 2 +-
> >   1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/drivers/watchdog/s3c2410_wdt.c b/drivers/watchdog/s3c2410_wdt.c
> > index 7d8fd04..4980f84 100644
> > --- a/drivers/watchdog/s3c2410_wdt.c
> > +++ b/drivers/watchdog/s3c2410_wdt.c
> > @@ -264,7 +264,7 @@ static irqreturn_t s3c2410wdt_irq(int irqno, void *param)
> >   	return IRQ_HANDLED;
> >   }
> >
> > -#ifdef CONFIG_CPU_FREQ
> > +#ifdef CONFIG_CPU_FREQ_S3C24XX
> >
> Where is the CPU_FREQ_S3C24XX configuration option defined ? I don't see it
> in the current upstream kernel, so it appears that this depends on some
> out-of-tree changes.

Seems like it got renamed to ARM_S3C24XX_CPUFREQ by f023f8dd59.

Best regards,
Tomasz

Re: [PATCH] watchdog: s3c2410_wdt: Only register for cpufreq on CPU_FREQ_S3C24XX

[Doug Anderson]

Guenter,

On Mon, Nov 25, 2013 at 3:23 PM, Guenter Roeck <linux@roeck-us.net> wrote:
> On 11/25/2013 02:55 PM, Doug Anderson wrote:
>>
>> On modern SoCs the watchdog timer is parented on a clock that doesn't
>> change every time we have a cpufreq change.  That means we don't need
>> to constantly adjust the watchdog timer, so avoid registering for and
>> dealing with cpufreq transitions unless we've actually got
>> CPU_FREQ_S3C24XX defined.
>>
>> Note that this is more than just an optimization.  The s3c2410
>> watchdog driver actually pats the watchdog on every CPU frequency
>> change.  On modern systems these happen many times per second (even in
>> a system where "nothing" is happening).  That effectively makes any
>> userspace watchdog program useless (the watchdog is constantly patted
>> by the kernel).  If we need CPU_FREQ_S3C24XX defined on a
>> multiplatform kernel we'll need to make sure that kernel supports
>> common clock and change this to user common clock framework.
>>
>> Signed-off-by: Doug Anderson <dianders@chromium.org>
>> ---
>>   drivers/watchdog/s3c2410_wdt.c | 2 +-
>>   1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/drivers/watchdog/s3c2410_wdt.c
>> b/drivers/watchdog/s3c2410_wdt.c
>> index 7d8fd04..4980f84 100644
>> --- a/drivers/watchdog/s3c2410_wdt.c
>> +++ b/drivers/watchdog/s3c2410_wdt.c
>> @@ -264,7 +264,7 @@ static irqreturn_t s3c2410wdt_irq(int irqno, void
>> *param)
>>         return IRQ_HANDLED;
>>   }
>>
>> -#ifdef CONFIG_CPU_FREQ
>> +#ifdef CONFIG_CPU_FREQ_S3C24XX
>>
> Where is the CPU_FREQ_S3C24XX configuration option defined ? I don't see it
> in the current upstream kernel, so it appears that this depends on some
> out-of-tree changes.

Whoops!  I guess I just trusted another patch and didn't dig.  I was
basing this on https://patchwork.kernel.org/patch/3023421/

Ah, I see that Tomasz just found the answer.  I'll update my patch.

-Doug

Re: [PATCH] watchdog: s3c2410_wdt: Only register for cpufreq on CPU_FREQ_S3C24XX

[Tomasz Figa]

On Monday 25 of November 2013 15:28:29 Doug Anderson wrote:
> Guenter,
> 
> On Mon, Nov 25, 2013 at 3:23 PM, Guenter Roeck <linux@roeck-us.net> wrote:
> > On 11/25/2013 02:55 PM, Doug Anderson wrote:
> >>
> >> On modern SoCs the watchdog timer is parented on a clock that doesn't
> >> change every time we have a cpufreq change.  That means we don't need
> >> to constantly adjust the watchdog timer, so avoid registering for and
> >> dealing with cpufreq transitions unless we've actually got
> >> CPU_FREQ_S3C24XX defined.
> >>
> >> Note that this is more than just an optimization.  The s3c2410
> >> watchdog driver actually pats the watchdog on every CPU frequency
> >> change.  On modern systems these happen many times per second (even in
> >> a system where "nothing" is happening).  That effectively makes any
> >> userspace watchdog program useless (the watchdog is constantly patted
> >> by the kernel).  If we need CPU_FREQ_S3C24XX defined on a
> >> multiplatform kernel we'll need to make sure that kernel supports
> >> common clock and change this to user common clock framework.
> >>
> >> Signed-off-by: Doug Anderson <dianders@chromium.org>
> >> ---
> >>   drivers/watchdog/s3c2410_wdt.c | 2 +-
> >>   1 file changed, 1 insertion(+), 1 deletion(-)
> >>
> >> diff --git a/drivers/watchdog/s3c2410_wdt.c
> >> b/drivers/watchdog/s3c2410_wdt.c
> >> index 7d8fd04..4980f84 100644
> >> --- a/drivers/watchdog/s3c2410_wdt.c
> >> +++ b/drivers/watchdog/s3c2410_wdt.c
> >> @@ -264,7 +264,7 @@ static irqreturn_t s3c2410wdt_irq(int irqno, void
> >> *param)
> >>         return IRQ_HANDLED;
> >>   }
> >>
> >> -#ifdef CONFIG_CPU_FREQ
> >> +#ifdef CONFIG_CPU_FREQ_S3C24XX
> >>
> > Where is the CPU_FREQ_S3C24XX configuration option defined ? I don't see it
> > in the current upstream kernel, so it appears that this depends on some
> > out-of-tree changes.
> 
> Whoops!  I guess I just trusted another patch and didn't dig.  I was
> basing this on https://patchwork.kernel.org/patch/3023421/
> 
> Ah, I see that Tomasz just found the answer.  I'll update my patch.

Seems like I originally gave you the wrong symbol name, looking at our
internal 3.10 tree, where the rename is not present yet, and nobody
noticed. Sorry for that.

Best regards,
Tomasz

[PATCH v2] watchdog: s3c2410_wdt: Only register for cpufreq on ARM_S3C24XX_CPUFREQ

[Doug Anderson]

On modern SoCs the watchdog timer is parented on a clock that doesn't
change every time we have a cpufreq change.  That means we don't need
to constantly adjust the watchdog timer, so avoid registering for and
dealing with cpufreq transitions unless we've actually got
CONFIG_ARM_S3C24XX_CPUFREQ defined.

Note that this is more than just an optimization.  The s3c2410
watchdog driver actually pats the watchdog on every CPU frequency
change.  On modern systems these happen many times per second (even in
a system where "nothing" is happening).  That effectively makes any
userspace watchdog program useless (the watchdog is constantly patted
by the kernel).  If we need ARM_S3C24XX_CPUFREQ defined on a
multiplatform kernel we'll need to make sure that kernel supports
common clock and change this to user common clock framework.

Signed-off-by: Doug Anderson <dianders@chromium.org>
---
Changes in v2:
- Use the updated config name.

 drivers/watchdog/s3c2410_wdt.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/watchdog/s3c2410_wdt.c b/drivers/watchdog/s3c2410_wdt.c
index 7d8fd04..c1ada33 100644
--- a/drivers/watchdog/s3c2410_wdt.c
+++ b/drivers/watchdog/s3c2410_wdt.c
@@ -264,7 +264,7 @@ static irqreturn_t s3c2410wdt_irq(int irqno, void *param)
 	return IRQ_HANDLED;
 }
 
-#ifdef CONFIG_CPU_FREQ
+#ifdef CONFIG_ARM_S3C24XX_CPUFREQ
 
 static int s3c2410wdt_cpufreq_transition(struct notifier_block *nb,
 					  unsigned long val, void *data)
-- 
1.8.4.1

Re: [PATCH v2] watchdog: s3c2410_wdt: Only register for cpufreq on ARM_S3C24XX_CPUFREQ

[Tomasz Figa]

On Monday 25 of November 2013 15:36:43 Doug Anderson wrote:
> On modern SoCs the watchdog timer is parented on a clock that doesn't
> change every time we have a cpufreq change.  That means we don't need
> to constantly adjust the watchdog timer, so avoid registering for and
> dealing with cpufreq transitions unless we've actually got
> CONFIG_ARM_S3C24XX_CPUFREQ defined.
> 
> Note that this is more than just an optimization.  The s3c2410
> watchdog driver actually pats the watchdog on every CPU frequency
> change.  On modern systems these happen many times per second (even in
> a system where "nothing" is happening).  That effectively makes any
> userspace watchdog program useless (the watchdog is constantly patted
> by the kernel).  If we need ARM_S3C24XX_CPUFREQ defined on a
> multiplatform kernel we'll need to make sure that kernel supports
> common clock and change this to user common clock framework.
> 
> Signed-off-by: Doug Anderson <dianders@chromium.org>
> ---
> Changes in v2:
> - Use the updated config name.
> 
>  drivers/watchdog/s3c2410_wdt.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)

Reviewed-by: Tomasz Figa <t.figa@samsung.com>

Best regards,
Tomasz

Re: [PATCH v2] watchdog: s3c2410_wdt: Only register for cpufreq on ARM_S3C24XX_CPUFREQ

[Guenter Roeck]

On 11/25/2013 03:36 PM, Doug Anderson wrote:
> On modern SoCs the watchdog timer is parented on a clock that doesn't
> change every time we have a cpufreq change.  That means we don't need
> to constantly adjust the watchdog timer, so avoid registering for and
> dealing with cpufreq transitions unless we've actually got
> CONFIG_ARM_S3C24XX_CPUFREQ defined.
>
> Note that this is more than just an optimization.  The s3c2410
> watchdog driver actually pats the watchdog on every CPU frequency
> change.  On modern systems these happen many times per second (even in
> a system where "nothing" is happening).  That effectively makes any
> userspace watchdog program useless (the watchdog is constantly patted
> by the kernel).  If we need ARM_S3C24XX_CPUFREQ defined on a
> multiplatform kernel we'll need to make sure that kernel supports
> common clock and change this to user common clock framework.
>
> Signed-off-by: Doug Anderson <dianders@chromium.org>

Reviewed-by: Guenter Roeck <linux@roeck-us.net>

Re: [PATCH v2] watchdog: s3c2410_wdt: Only register for cpufreq on ARM_S3C24XX_CPUFREQ

[Wim Van Sebroeck]

Hi Doug,

> On modern SoCs the watchdog timer is parented on a clock that doesn't
> change every time we have a cpufreq change.  That means we don't need
> to constantly adjust the watchdog timer, so avoid registering for and
> dealing with cpufreq transitions unless we've actually got
> CONFIG_ARM_S3C24XX_CPUFREQ defined.
> 
> Note that this is more than just an optimization.  The s3c2410
> watchdog driver actually pats the watchdog on every CPU frequency
> change.  On modern systems these happen many times per second (even in
> a system where "nothing" is happening).  That effectively makes any
> userspace watchdog program useless (the watchdog is constantly patted
> by the kernel).  If we need ARM_S3C24XX_CPUFREQ defined on a
> multiplatform kernel we'll need to make sure that kernel supports
> common clock and change this to user common clock framework.
> 
> Signed-off-by: Doug Anderson <dianders@chromium.org>
> ---
> Changes in v2:
> - Use the updated config name.

Patch added to linux-watchdog-next.

Kind regards,
Wim.

Re: [PATCH] watchdog: s3c2410_wdt: Only register for cpufreq on CPU_FREQ_S3C24XX

[Ben Dooks]

On Mon, Nov 25, 2013 at 02:55:37PM -0800, Doug Anderson wrote:
> On modern SoCs the watchdog timer is parented on a clock that doesn't
> change every time we have a cpufreq change.  That means we don't need
> to constantly adjust the watchdog timer, so avoid registering for and
> dealing with cpufreq transitions unless we've actually got
> CPU_FREQ_S3C24XX defined.

The S3C2410 to at-least S3C2440 there is only one PLL/clock-divider
set so changing any part of that can change frequency supplied to
the watchdog.

Re: [PATCH] watchdog: s3c2410_wdt: Only register for cpufreq on CPU_FREQ_S3C24XX

[Tomasz Figa]

On Tuesday 26 of November 2013 00:08:31 Ben Dooks wrote:
> On Mon, Nov 25, 2013 at 02:55:37PM -0800, Doug Anderson wrote:
> > On modern SoCs the watchdog timer is parented on a clock that doesn't
> > change every time we have a cpufreq change.  That means we don't need
> > to constantly adjust the watchdog timer, so avoid registering for and
> > dealing with cpufreq transitions unless we've actually got
> > CPU_FREQ_S3C24XX defined.
> 
> The S3C2410 to at-least S3C2440 there is only one PLL/clock-divider
> set so changing any part of that can change frequency supplied to
> the watchdog.

And so we're leaving the notifier code intact, just changing the ifdef
conditionm which decides whether it should be compiled in, from a (too)
generic one to an S3C24xx-specific one.

Best regards,
Tomasz