diff mbox series

[2/2] cpufreq: schedutil: consolidate capacity margin calculation

Message ID 1506121484-8260-2-git-send-email-leo.yan@linaro.org
State Superseded
Headers show
Series [1/2] sched/fair: make capacity_margin __read_mostly | expand

Commit Message

Leo Yan Sept. 22, 2017, 11:04 p.m. UTC 
Scheduler CFS class has variable 'capacity_margin' to calculate the
capacity margin, and schedutil governor also needs to compensate the
same margin for frequency tipping point. Below are formulas used in
CFS class and schedutil governor separately:

CFS:       U` = U * capacity_margin / 1024 = U * 1.25
Schedutil: U` = U + U >> 2 = U + U * 0.25  = U * 1.25

This patch consolidates the capacity margin calculation so let
schedutil to use same formula with CFS class. As result this can avoid
the mismatch issue between schedutil and CFS class after change
'capacity_margin' to other values.

Cc: Dietmar Eggemann <dietmar.eggemann@arm.com>
Cc: Morten Rasmussen <morten.rasmussen@arm.com>
Cc: Chris Redpath <Chris.Redpath@arm.com>
Cc: Joel Fernandes <joelaf@google.com>
Cc: Vincent Guittot <vincent.guittot@linaro.org>
Signed-off-by: Leo Yan <leo.yan@linaro.org>

---
 kernel/sched/cpufreq_schedutil.c | 5 +++--
 kernel/sched/sched.h             | 1 +
 2 files changed, 4 insertions(+), 2 deletions(-)

-- 
2.7.4

Comments

Peter Zijlstra Sept. 25, 2017, 8:15 a.m. UTC | #1 
On Sat, Sep 23, 2017 at 07:04:44AM +0800, Leo Yan wrote:
> +	freq = (freq * capacity_margin / 1024) * util / max;


We have SCHED_CAPACITY_SCALE for that..
Patrick Bellasi Sept. 25, 2017, 1:55 p.m. UTC | #2 
On 23-Sep 07:04, Leo Yan wrote:
> Scheduler CFS class has variable 'capacity_margin' to calculate the

> capacity margin, and schedutil governor also needs to compensate the

> same margin for frequency tipping point. Below are formulas used in

> CFS class and schedutil governor separately:

> 

> CFS:       U` = U * capacity_margin / 1024 = U * 1.25

> Schedutil: U` = U + U >> 2 = U + U * 0.25  = U * 1.25

> 

> This patch consolidates the capacity margin calculation so let

> schedutil to use same formula with CFS class. As result this can avoid

> the mismatch issue between schedutil and CFS class after change

> 'capacity_margin' to other values.

> 

> Cc: Dietmar Eggemann <dietmar.eggemann@arm.com>

> Cc: Morten Rasmussen <morten.rasmussen@arm.com>

> Cc: Chris Redpath <Chris.Redpath@arm.com>

> Cc: Joel Fernandes <joelaf@google.com>

> Cc: Vincent Guittot <vincent.guittot@linaro.org>

> Signed-off-by: Leo Yan <leo.yan@linaro.org>

> ---

>  kernel/sched/cpufreq_schedutil.c | 5 +++--

>  kernel/sched/sched.h             | 1 +

>  2 files changed, 4 insertions(+), 2 deletions(-)

> 

> diff --git a/kernel/sched/cpufreq_schedutil.c b/kernel/sched/cpufreq_schedutil.c

> index 9209d83..067abbe 100644

> --- a/kernel/sched/cpufreq_schedutil.c

> +++ b/kernel/sched/cpufreq_schedutil.c

> @@ -155,7 +155,8 @@ static void sugov_update_commit(struct sugov_policy *sg_policy, u64 time,

>   *

>   * next_freq = C * curr_freq * util_raw / max

>   *

> - * Take C = 1.25 for the frequency tipping point at (util / max) = 0.8.

> + * Take C = capacity_margin / 1024 = 1.25, so it's for the frequency tipping

> + * point at (util / max) = 0.8.

>   *

>   * The lowest driver-supported frequency which is equal or greater than the raw

>   * next_freq (as calculated above) is returned, subject to policy min/max and

> @@ -168,7 +169,7 @@ static unsigned int get_next_freq(struct sugov_policy *sg_policy,

>  	unsigned int freq = arch_scale_freq_invariant() ?

>  				policy->cpuinfo.max_freq : policy->cur;

>  

> -	freq = (freq + (freq >> 2)) * util / max;

> +	freq = (freq * capacity_margin / 1024) * util / max;


The compiler should be smart enough but perhaps you can better use:

        freq *= (capacity_margin >> SCHED_CAPACITY_SHIFT));
        freq *= util / max;

>  	if (freq == sg_policy->cached_raw_freq && sg_policy->next_freq != UINT_MAX)

>  		return sg_policy->next_freq;

> diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h

> index 14db76c..cf75bdc 100644

> --- a/kernel/sched/sched.h

> +++ b/kernel/sched/sched.h

> @@ -52,6 +52,7 @@ struct cpuidle_state;

>  #define TASK_ON_RQ_MIGRATING	2

>  

>  extern __read_mostly int scheduler_running;

> +extern unsigned int capacity_margin __read_mostly;

>  

>  extern unsigned long calc_load_update;

>  extern atomic_long_t calc_load_tasks;

> -- 

> 2.7.4

> 


-- 
#include <best/regards.h>

Patrick Bellasi
Leo Yan Sept. 28, 2017, 6:33 a.m. UTC | #3 
On Mon, Sep 25, 2017 at 02:55:07PM +0100, Patrick Bellasi wrote:
> On 23-Sep 07:04, Leo Yan wrote:

> > Scheduler CFS class has variable 'capacity_margin' to calculate the

> > capacity margin, and schedutil governor also needs to compensate the

> > same margin for frequency tipping point. Below are formulas used in

> > CFS class and schedutil governor separately:

> > 

> > CFS:       U` = U * capacity_margin / 1024 = U * 1.25

> > Schedutil: U` = U + U >> 2 = U + U * 0.25  = U * 1.25

> > 

> > This patch consolidates the capacity margin calculation so let

> > schedutil to use same formula with CFS class. As result this can avoid

> > the mismatch issue between schedutil and CFS class after change

> > 'capacity_margin' to other values.

> > 

> > Cc: Dietmar Eggemann <dietmar.eggemann@arm.com>

> > Cc: Morten Rasmussen <morten.rasmussen@arm.com>

> > Cc: Chris Redpath <Chris.Redpath@arm.com>

> > Cc: Joel Fernandes <joelaf@google.com>

> > Cc: Vincent Guittot <vincent.guittot@linaro.org>

> > Signed-off-by: Leo Yan <leo.yan@linaro.org>

> > ---

> >  kernel/sched/cpufreq_schedutil.c | 5 +++--

> >  kernel/sched/sched.h             | 1 +

> >  2 files changed, 4 insertions(+), 2 deletions(-)

> > 

> > diff --git a/kernel/sched/cpufreq_schedutil.c b/kernel/sched/cpufreq_schedutil.c

> > index 9209d83..067abbe 100644

> > --- a/kernel/sched/cpufreq_schedutil.c

> > +++ b/kernel/sched/cpufreq_schedutil.c

> > @@ -155,7 +155,8 @@ static void sugov_update_commit(struct sugov_policy *sg_policy, u64 time,

> >   *

> >   * next_freq = C * curr_freq * util_raw / max

> >   *

> > - * Take C = 1.25 for the frequency tipping point at (util / max) = 0.8.

> > + * Take C = capacity_margin / 1024 = 1.25, so it's for the frequency tipping

> > + * point at (util / max) = 0.8.

> >   *

> >   * The lowest driver-supported frequency which is equal or greater than the raw

> >   * next_freq (as calculated above) is returned, subject to policy min/max and

> > @@ -168,7 +169,7 @@ static unsigned int get_next_freq(struct sugov_policy *sg_policy,

> >  	unsigned int freq = arch_scale_freq_invariant() ?

> >  				policy->cpuinfo.max_freq : policy->cur;

> >  

> > -	freq = (freq + (freq >> 2)) * util / max;

> > +	freq = (freq * capacity_margin / 1024) * util / max;

> 

> The compiler should be smart enough but perhaps you can better use:

> 

>         freq *= (capacity_margin >> SCHED_CAPACITY_SHIFT));

>         freq *= util / max;


Thanks for the suggestion. The '*=' has low precedence than other
operators, will fix as below:

  freq = freq * capacity_margin >> SCHED_CAPACITY_SHIFT;
  freq = freq * util / max;

> >  	if (freq == sg_policy->cached_raw_freq && sg_policy->next_freq != UINT_MAX)

> >  		return sg_policy->next_freq;

> > diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h

> > index 14db76c..cf75bdc 100644

> > --- a/kernel/sched/sched.h

> > +++ b/kernel/sched/sched.h

> > @@ -52,6 +52,7 @@ struct cpuidle_state;

> >  #define TASK_ON_RQ_MIGRATING	2

> >  

> >  extern __read_mostly int scheduler_running;

> > +extern unsigned int capacity_margin __read_mostly;

> >  

> >  extern unsigned long calc_load_update;

> >  extern atomic_long_t calc_load_tasks;

> > -- 

> > 2.7.4

> > 

> 

> -- 

> #include <best/regards.h>

> 

> Patrick Bellasi
diff mbox series

Patch

diff --git a/kernel/sched/cpufreq_schedutil.c b/kernel/sched/cpufreq_schedutil.c
index 9209d83..067abbe 100644
--- a/kernel/sched/cpufreq_schedutil.c
+++ b/kernel/sched/cpufreq_schedutil.c
@@ -155,7 +155,8 @@  static void sugov_update_commit(struct sugov_policy *sg_policy, u64 time,
  *
  * next_freq = C * curr_freq * util_raw / max
  *
- * Take C = 1.25 for the frequency tipping point at (util / max) = 0.8.
+ * Take C = capacity_margin / 1024 = 1.25, so it's for the frequency tipping
+ * point at (util / max) = 0.8.
  *
  * The lowest driver-supported frequency which is equal or greater than the raw
  * next_freq (as calculated above) is returned, subject to policy min/max and
@@ -168,7 +169,7 @@  static unsigned int get_next_freq(struct sugov_policy *sg_policy,
 	unsigned int freq = arch_scale_freq_invariant() ?
 				policy->cpuinfo.max_freq : policy->cur;
 
-	freq = (freq + (freq >> 2)) * util / max;
+	freq = (freq * capacity_margin / 1024) * util / max;
 
 	if (freq == sg_policy->cached_raw_freq && sg_policy->next_freq != UINT_MAX)
 		return sg_policy->next_freq;
diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
index 14db76c..cf75bdc 100644
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
@@ -52,6 +52,7 @@  struct cpuidle_state;
 #define TASK_ON_RQ_MIGRATING	2
 
 extern __read_mostly int scheduler_running;
+extern unsigned int capacity_margin __read_mostly;
 
 extern unsigned long calc_load_update;
 extern atomic_long_t calc_load_tasks;