Patchwork cpufreq: scmi: Fix frequency invariance in slow path

login
register
mail settings
Submitter Quentin Perret
Date Jan. 9, 2019, 10:42 a.m.
Message ID <20190109104236.6532-1-quentin.perret@arm.com>
Download mbox | patch
Permalink /patch/695651/
State New
Headers show

Comments

Quentin Perret - Jan. 9, 2019, 10:42 a.m.
The scmi-cpufreq driver calls the arch_set_freq_scale() callback on
frequency changes to provide scale-invariant load-tracking signals to
the scheduler. However, in the slow path, it does so while specifying
the current and max frequencies in different units, hence resulting in a
broken freq_scale factor.

Fix this by passing all frequencies in KHz, as stored in the CPUFreq
frequency table.

Fixes: 99d6bdf33877 ("cpufreq: add support for CPU DVFS based on SCMI
message protocol")
Signed-off-by: Quentin Perret <quentin.perret@arm.com>
---
 drivers/cpufreq/scmi-cpufreq.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
Viresh Kumar - Jan. 9, 2019, 10:45 a.m.
On 09-01-19, 10:42, Quentin Perret wrote:
> The scmi-cpufreq driver calls the arch_set_freq_scale() callback on
> frequency changes to provide scale-invariant load-tracking signals to
> the scheduler. However, in the slow path, it does so while specifying
> the current and max frequencies in different units, hence resulting in a
> broken freq_scale factor.
> 
> Fix this by passing all frequencies in KHz, as stored in the CPUFreq
> frequency table.
> 
> Fixes: 99d6bdf33877 ("cpufreq: add support for CPU DVFS based on SCMI
> message protocol")
> Signed-off-by: Quentin Perret <quentin.perret@arm.com>
> ---
>  drivers/cpufreq/scmi-cpufreq.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/cpufreq/scmi-cpufreq.c b/drivers/cpufreq/scmi-cpufreq.c
> index 50b1551ba894..3f0693439486 100644
> --- a/drivers/cpufreq/scmi-cpufreq.c
> +++ b/drivers/cpufreq/scmi-cpufreq.c
> @@ -52,9 +52,9 @@ scmi_cpufreq_set_target(struct cpufreq_policy *policy, unsigned int index)
>  	int ret;
>  	struct scmi_data *priv = policy->driver_data;
>  	struct scmi_perf_ops *perf_ops = handle->perf_ops;
> -	u64 freq = policy->freq_table[index].frequency * 1000;
> +	u64 freq = policy->freq_table[index].frequency;
>  
> -	ret = perf_ops->freq_set(handle, priv->domain_id, freq, false);
> +	ret = perf_ops->freq_set(handle, priv->domain_id, freq * 1000, false);
>  	if (!ret)
>  		arch_set_freq_scale(policy->related_cpus, freq,
>  				    policy->cpuinfo.max_freq);

Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Rafael J. Wysocki - Jan. 9, 2019, 10:56 a.m.
On Wednesday, January 9, 2019 11:45:11 AM CET Viresh Kumar wrote:
> On 09-01-19, 10:42, Quentin Perret wrote:
> > The scmi-cpufreq driver calls the arch_set_freq_scale() callback on
> > frequency changes to provide scale-invariant load-tracking signals to
> > the scheduler. However, in the slow path, it does so while specifying
> > the current and max frequencies in different units, hence resulting in a
> > broken freq_scale factor.
> > 
> > Fix this by passing all frequencies in KHz, as stored in the CPUFreq
> > frequency table.
> > 
> > Fixes: 99d6bdf33877 ("cpufreq: add support for CPU DVFS based on SCMI
> > message protocol")
> > Signed-off-by: Quentin Perret <quentin.perret@arm.com>
> > ---
> >  drivers/cpufreq/scmi-cpufreq.c | 4 ++--
> >  1 file changed, 2 insertions(+), 2 deletions(-)
> > 
> > diff --git a/drivers/cpufreq/scmi-cpufreq.c b/drivers/cpufreq/scmi-cpufreq.c
> > index 50b1551ba894..3f0693439486 100644
> > --- a/drivers/cpufreq/scmi-cpufreq.c
> > +++ b/drivers/cpufreq/scmi-cpufreq.c
> > @@ -52,9 +52,9 @@ scmi_cpufreq_set_target(struct cpufreq_policy *policy, unsigned int index)
> >  	int ret;
> >  	struct scmi_data *priv = policy->driver_data;
> >  	struct scmi_perf_ops *perf_ops = handle->perf_ops;
> > -	u64 freq = policy->freq_table[index].frequency * 1000;
> > +	u64 freq = policy->freq_table[index].frequency;
> >  
> > -	ret = perf_ops->freq_set(handle, priv->domain_id, freq, false);
> > +	ret = perf_ops->freq_set(handle, priv->domain_id, freq * 1000, false);
> >  	if (!ret)
> >  		arch_set_freq_scale(policy->related_cpus, freq,
> >  				    policy->cpuinfo.max_freq);
> 
> Acked-by: Viresh Kumar <viresh.kumar@linaro.org>

This would be stable-candidate I guess?
Quentin Perret - Jan. 9, 2019, 10:59 a.m.
On Wednesday 09 Jan 2019 at 11:56:06 (+0100), Rafael J. Wysocki wrote:
> On Wednesday, January 9, 2019 11:45:11 AM CET Viresh Kumar wrote:
> > On 09-01-19, 10:42, Quentin Perret wrote:
> > > The scmi-cpufreq driver calls the arch_set_freq_scale() callback on
> > > frequency changes to provide scale-invariant load-tracking signals to
> > > the scheduler. However, in the slow path, it does so while specifying
> > > the current and max frequencies in different units, hence resulting in a
> > > broken freq_scale factor.
> > > 
> > > Fix this by passing all frequencies in KHz, as stored in the CPUFreq
> > > frequency table.
> > > 
> > > Fixes: 99d6bdf33877 ("cpufreq: add support for CPU DVFS based on SCMI
> > > message protocol")
> > > Signed-off-by: Quentin Perret <quentin.perret@arm.com>
> > > ---
> > >  drivers/cpufreq/scmi-cpufreq.c | 4 ++--
> > >  1 file changed, 2 insertions(+), 2 deletions(-)
> > > 
> > > diff --git a/drivers/cpufreq/scmi-cpufreq.c b/drivers/cpufreq/scmi-cpufreq.c
> > > index 50b1551ba894..3f0693439486 100644
> > > --- a/drivers/cpufreq/scmi-cpufreq.c
> > > +++ b/drivers/cpufreq/scmi-cpufreq.c
> > > @@ -52,9 +52,9 @@ scmi_cpufreq_set_target(struct cpufreq_policy *policy, unsigned int index)
> > >  	int ret;
> > >  	struct scmi_data *priv = policy->driver_data;
> > >  	struct scmi_perf_ops *perf_ops = handle->perf_ops;
> > > -	u64 freq = policy->freq_table[index].frequency * 1000;
> > > +	u64 freq = policy->freq_table[index].frequency;
> > >  
> > > -	ret = perf_ops->freq_set(handle, priv->domain_id, freq, false);
> > > +	ret = perf_ops->freq_set(handle, priv->domain_id, freq * 1000, false);
> > >  	if (!ret)
> > >  		arch_set_freq_scale(policy->related_cpus, freq,
> > >  				    policy->cpuinfo.max_freq);
> > 
> > Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
> 
> This would be stable-candidate I guess?

I think so yes. I was hoping the 'Fixes:' tag would be enough ? Or do I
still need to CC stable too ?

Thanks,
Quentin
Viresh Kumar - Jan. 9, 2019, 10:59 a.m.
On 09-01-19, 11:56, Rafael J. Wysocki wrote:
> On Wednesday, January 9, 2019 11:45:11 AM CET Viresh Kumar wrote:
> > On 09-01-19, 10:42, Quentin Perret wrote:
> > > The scmi-cpufreq driver calls the arch_set_freq_scale() callback on
> > > frequency changes to provide scale-invariant load-tracking signals to
> > > the scheduler. However, in the slow path, it does so while specifying
> > > the current and max frequencies in different units, hence resulting in a
> > > broken freq_scale factor.
> > > 
> > > Fix this by passing all frequencies in KHz, as stored in the CPUFreq
> > > frequency table.
> > > 
> > > Fixes: 99d6bdf33877 ("cpufreq: add support for CPU DVFS based on SCMI
> > > message protocol")
> > > Signed-off-by: Quentin Perret <quentin.perret@arm.com>
> > > ---
> > >  drivers/cpufreq/scmi-cpufreq.c | 4 ++--
> > >  1 file changed, 2 insertions(+), 2 deletions(-)
> > > 
> > > diff --git a/drivers/cpufreq/scmi-cpufreq.c b/drivers/cpufreq/scmi-cpufreq.c
> > > index 50b1551ba894..3f0693439486 100644
> > > --- a/drivers/cpufreq/scmi-cpufreq.c
> > > +++ b/drivers/cpufreq/scmi-cpufreq.c
> > > @@ -52,9 +52,9 @@ scmi_cpufreq_set_target(struct cpufreq_policy *policy, unsigned int index)
> > >  	int ret;
> > >  	struct scmi_data *priv = policy->driver_data;
> > >  	struct scmi_perf_ops *perf_ops = handle->perf_ops;
> > > -	u64 freq = policy->freq_table[index].frequency * 1000;
> > > +	u64 freq = policy->freq_table[index].frequency;
> > >  
> > > -	ret = perf_ops->freq_set(handle, priv->domain_id, freq, false);
> > > +	ret = perf_ops->freq_set(handle, priv->domain_id, freq * 1000, false);
> > >  	if (!ret)
> > >  		arch_set_freq_scale(policy->related_cpus, freq,
> > >  				    policy->cpuinfo.max_freq);
> > 
> > Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
> 
> This would be stable-candidate I guess?

Cc: 4.17+ <stable@vger.kernel.org> # v4.17+
Sudeep Holla - Jan. 9, 2019, 10:59 a.m.
On Wed, Jan 09, 2019 at 10:42:36AM +0000, Quentin Perret wrote:
> The scmi-cpufreq driver calls the arch_set_freq_scale() callback on
> frequency changes to provide scale-invariant load-tracking signals to
> the scheduler. However, in the slow path, it does so while specifying
> the current and max frequencies in different units, hence resulting in a
> broken freq_scale factor.
> 
> Fix this by passing all frequencies in KHz, as stored in the CPUFreq
> frequency table.
> 
> Fixes: 99d6bdf33877 ("cpufreq: add support for CPU DVFS based on SCMI
> message protocol")

Good find.

Acked-by: Sudeep Holla <sudeep.holla@arm.com>

--
Regards,
Sudeep
Rafael J. Wysocki - Jan. 9, 2019, 11:03 a.m.
On Wednesday, January 9, 2019 11:59:05 AM CET Quentin Perret wrote:
> On Wednesday 09 Jan 2019 at 11:56:06 (+0100), Rafael J. Wysocki wrote:
> > On Wednesday, January 9, 2019 11:45:11 AM CET Viresh Kumar wrote:
> > > On 09-01-19, 10:42, Quentin Perret wrote:
> > > > The scmi-cpufreq driver calls the arch_set_freq_scale() callback on
> > > > frequency changes to provide scale-invariant load-tracking signals to
> > > > the scheduler. However, in the slow path, it does so while specifying
> > > > the current and max frequencies in different units, hence resulting in a
> > > > broken freq_scale factor.
> > > > 
> > > > Fix this by passing all frequencies in KHz, as stored in the CPUFreq
> > > > frequency table.
> > > > 
> > > > Fixes: 99d6bdf33877 ("cpufreq: add support for CPU DVFS based on SCMI
> > > > message protocol")
> > > > Signed-off-by: Quentin Perret <quentin.perret@arm.com>
> > > > ---
> > > >  drivers/cpufreq/scmi-cpufreq.c | 4 ++--
> > > >  1 file changed, 2 insertions(+), 2 deletions(-)
> > > > 
> > > > diff --git a/drivers/cpufreq/scmi-cpufreq.c b/drivers/cpufreq/scmi-cpufreq.c
> > > > index 50b1551ba894..3f0693439486 100644
> > > > --- a/drivers/cpufreq/scmi-cpufreq.c
> > > > +++ b/drivers/cpufreq/scmi-cpufreq.c
> > > > @@ -52,9 +52,9 @@ scmi_cpufreq_set_target(struct cpufreq_policy *policy, unsigned int index)
> > > >  	int ret;
> > > >  	struct scmi_data *priv = policy->driver_data;
> > > >  	struct scmi_perf_ops *perf_ops = handle->perf_ops;
> > > > -	u64 freq = policy->freq_table[index].frequency * 1000;
> > > > +	u64 freq = policy->freq_table[index].frequency;
> > > >  
> > > > -	ret = perf_ops->freq_set(handle, priv->domain_id, freq, false);
> > > > +	ret = perf_ops->freq_set(handle, priv->domain_id, freq * 1000, false);
> > > >  	if (!ret)
> > > >  		arch_set_freq_scale(policy->related_cpus, freq,
> > > >  				    policy->cpuinfo.max_freq);
> > > 
> > > Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
> > 
> > This would be stable-candidate I guess?
> 
> I think so yes. I was hoping the 'Fixes:' tag would be enough ?

It doesn't cause -stable to pick up commits automatically if that's
what you're asking about.

> Or do I still need to CC stable too ?

No, you don't need to (and should not) CC -stable yourself.

You can, however, add a Cc: <stable..> tag to the patch which helps
maintainers a bit.
Quentin Perret - Jan. 9, 2019, 11:07 a.m.
On Wednesday 09 Jan 2019 at 12:03:49 (+0100), Rafael J. Wysocki wrote:
> On Wednesday, January 9, 2019 11:59:05 AM CET Quentin Perret wrote:
> > On Wednesday 09 Jan 2019 at 11:56:06 (+0100), Rafael J. Wysocki wrote:
> > > On Wednesday, January 9, 2019 11:45:11 AM CET Viresh Kumar wrote:
> > > > On 09-01-19, 10:42, Quentin Perret wrote:
> > > > > The scmi-cpufreq driver calls the arch_set_freq_scale() callback on
> > > > > frequency changes to provide scale-invariant load-tracking signals to
> > > > > the scheduler. However, in the slow path, it does so while specifying
> > > > > the current and max frequencies in different units, hence resulting in a
> > > > > broken freq_scale factor.
> > > > > 
> > > > > Fix this by passing all frequencies in KHz, as stored in the CPUFreq
> > > > > frequency table.
> > > > > 
> > > > > Fixes: 99d6bdf33877 ("cpufreq: add support for CPU DVFS based on SCMI
> > > > > message protocol")
> > > > > Signed-off-by: Quentin Perret <quentin.perret@arm.com>
> > > > > ---
> > > > >  drivers/cpufreq/scmi-cpufreq.c | 4 ++--
> > > > >  1 file changed, 2 insertions(+), 2 deletions(-)
> > > > > 
> > > > > diff --git a/drivers/cpufreq/scmi-cpufreq.c b/drivers/cpufreq/scmi-cpufreq.c
> > > > > index 50b1551ba894..3f0693439486 100644
> > > > > --- a/drivers/cpufreq/scmi-cpufreq.c
> > > > > +++ b/drivers/cpufreq/scmi-cpufreq.c
> > > > > @@ -52,9 +52,9 @@ scmi_cpufreq_set_target(struct cpufreq_policy *policy, unsigned int index)
> > > > >  	int ret;
> > > > >  	struct scmi_data *priv = policy->driver_data;
> > > > >  	struct scmi_perf_ops *perf_ops = handle->perf_ops;
> > > > > -	u64 freq = policy->freq_table[index].frequency * 1000;
> > > > > +	u64 freq = policy->freq_table[index].frequency;
> > > > >  
> > > > > -	ret = perf_ops->freq_set(handle, priv->domain_id, freq, false);
> > > > > +	ret = perf_ops->freq_set(handle, priv->domain_id, freq * 1000, false);
> > > > >  	if (!ret)
> > > > >  		arch_set_freq_scale(policy->related_cpus, freq,
> > > > >  				    policy->cpuinfo.max_freq);
> > > > 
> > > > Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
> > > 
> > > This would be stable-candidate I guess?
> > 
> > I think so yes. I was hoping the 'Fixes:' tag would be enough ?
> 
> It doesn't cause -stable to pick up commits automatically if that's
> what you're asking about.
>
> > Or do I still need to CC stable too ?
> 
> No, you don't need to (and should not) CC -stable yourself.
> 
> You can, however, add a Cc: <stable..> tag to the patch which helps
> maintainers a bit.

OK, will do from now on.

Thanks !
Quentin
Rafael J. Wysocki - Jan. 11, 2019, 10:41 a.m.
On Wednesday, January 9, 2019 11:59:45 AM CET Sudeep Holla wrote:
> On Wed, Jan 09, 2019 at 10:42:36AM +0000, Quentin Perret wrote:
> > The scmi-cpufreq driver calls the arch_set_freq_scale() callback on
> > frequency changes to provide scale-invariant load-tracking signals to
> > the scheduler. However, in the slow path, it does so while specifying
> > the current and max frequencies in different units, hence resulting in a
> > broken freq_scale factor.
> > 
> > Fix this by passing all frequencies in KHz, as stored in the CPUFreq
> > frequency table.
> > 
> > Fixes: 99d6bdf33877 ("cpufreq: add support for CPU DVFS based on SCMI
> > message protocol")
> 
> Good find.
> 
> Acked-by: Sudeep Holla <sudeep.holla@arm.com>

Applied, thanks!

Patch

diff --git a/drivers/cpufreq/scmi-cpufreq.c b/drivers/cpufreq/scmi-cpufreq.c
index 50b1551ba894..3f0693439486 100644
--- a/drivers/cpufreq/scmi-cpufreq.c
+++ b/drivers/cpufreq/scmi-cpufreq.c
@@ -52,9 +52,9 @@  scmi_cpufreq_set_target(struct cpufreq_policy *policy, unsigned int index)
 	int ret;
 	struct scmi_data *priv = policy->driver_data;
 	struct scmi_perf_ops *perf_ops = handle->perf_ops;
-	u64 freq = policy->freq_table[index].frequency * 1000;
+	u64 freq = policy->freq_table[index].frequency;
 
-	ret = perf_ops->freq_set(handle, priv->domain_id, freq, false);
+	ret = perf_ops->freq_set(handle, priv->domain_id, freq * 1000, false);
 	if (!ret)
 		arch_set_freq_scale(policy->related_cpus, freq,
 				    policy->cpuinfo.max_freq);