2015-02-19 16:34:52

by Peter Zijlstra

[permalink] [raw]
Subject: Re: [PATCH RESEND v9 04/10] sched: Make sched entity usage tracking scale-invariant

On Thu, Jan 15, 2015 at 11:09:24AM +0100, Vincent Guittot wrote:
> From: Morten Rasmussen <[email protected]>
>
> Apply frequency scale-invariance correction factor to usage tracking.
> Each segment of the running_load_avg geometric series is now scaled by the
> current frequency so the utilization_avg_contrib of each entity will be
> invariant with frequency scaling.

> As a result, utilization_load_avg which is
> the sum of utilization_avg_contrib, becomes invariant too. So the usage level
> that is returned by get_cpu_usage, stays relative to the max frequency as the
> cpu_capacity which is is compared against.

> Then, we want the keep the load tracking values in a 32bits type, which implies
> that the max value of {runnable|running}_avg_sum must be lower than
> 2^32/88761=48388 (88761 is the max weigth of a task).

> As LOAD_AVG_MAX = 47742,
> arch_scale_freq_capacity must return a value less than
> (48388/47742) << SCHED_CAPACITY_SHIFT = 1037 (SCHED_SCALE_CAPACITY = 1024).
> So we define the range to [0..SCHED_SCALE_CAPACITY] in order to avoid overflow.

> cc: Paul Turner <[email protected]>
> cc: Ben Segall <[email protected]>
>
> Signed-off-by: Morten Rasmussen <[email protected]>
> Signed-off-by: Vincent Guittot <[email protected]>
> ---
> kernel/sched/fair.c | 21 ++++++++++++++-------
> 1 file changed, 14 insertions(+), 7 deletions(-)
>
> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index a96affd..a5039da 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -2266,6 +2266,8 @@ static u32 __compute_runnable_contrib(u64 n)
> return contrib + runnable_avg_yN_sum[n];
> }
>
> +unsigned long __weak arch_scale_freq_capacity(struct sched_domain *sd, int cpu);

There is no actual function definition in this patch; this would hinder
linking, no?


> if (running)
> + sa->running_avg_sum += delta_w * scale_freq
> + >> SCHED_CAPACITY_SHIFT;


> if (running)
> + sa->running_avg_sum += runnable_contrib * scale_freq
> + >> SCHED_CAPACITY_SHIFT;

> if (running)
> + sa->running_avg_sum += delta * scale_freq
> + >> SCHED_CAPACITY_SHIFT;

If you're going to respin; please but {} around that.


2015-02-19 17:01:39

by Morten Rasmussen

[permalink] [raw]
Subject: Re: [PATCH RESEND v9 04/10] sched: Make sched entity usage tracking scale-invariant

On Thu, Feb 19, 2015 at 04:34:42PM +0000, Peter Zijlstra wrote:
> On Thu, Jan 15, 2015 at 11:09:24AM +0100, Vincent Guittot wrote:
> > From: Morten Rasmussen <[email protected]>
> >
> > Apply frequency scale-invariance correction factor to usage tracking.
> > Each segment of the running_load_avg geometric series is now scaled by the
> > current frequency so the utilization_avg_contrib of each entity will be
> > invariant with frequency scaling.
>
> > As a result, utilization_load_avg which is
> > the sum of utilization_avg_contrib, becomes invariant too. So the usage level
> > that is returned by get_cpu_usage, stays relative to the max frequency as the
> > cpu_capacity which is is compared against.
>
> > Then, we want the keep the load tracking values in a 32bits type, which implies
> > that the max value of {runnable|running}_avg_sum must be lower than
> > 2^32/88761=48388 (88761 is the max weigth of a task).
>
> > As LOAD_AVG_MAX = 47742,
> > arch_scale_freq_capacity must return a value less than
> > (48388/47742) << SCHED_CAPACITY_SHIFT = 1037 (SCHED_SCALE_CAPACITY = 1024).
> > So we define the range to [0..SCHED_SCALE_CAPACITY] in order to avoid overflow.
>
> > cc: Paul Turner <[email protected]>
> > cc: Ben Segall <[email protected]>
> >
> > Signed-off-by: Morten Rasmussen <[email protected]>
> > Signed-off-by: Vincent Guittot <[email protected]>
> > ---
> > kernel/sched/fair.c | 21 ++++++++++++++-------
> > 1 file changed, 14 insertions(+), 7 deletions(-)
> >
> > diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> > index a96affd..a5039da 100644
> > --- a/kernel/sched/fair.c
> > +++ b/kernel/sched/fair.c
> > @@ -2266,6 +2266,8 @@ static u32 __compute_runnable_contrib(u64 n)
> > return contrib + runnable_avg_yN_sum[n];
> > }
> >
> > +unsigned long __weak arch_scale_freq_capacity(struct sched_domain *sd, int cpu);
>
> There is no actual function definition in this patch; this would hinder
> linking, no?

The function definition already lives further down in fair.c. I can move
it up here if you prefer?

>
>
> > if (running)
> > + sa->running_avg_sum += delta_w * scale_freq
> > + >> SCHED_CAPACITY_SHIFT;
>
>
> > if (running)
> > + sa->running_avg_sum += runnable_contrib * scale_freq
> > + >> SCHED_CAPACITY_SHIFT;
>
> > if (running)
> > + sa->running_avg_sum += delta * scale_freq
> > + >> SCHED_CAPACITY_SHIFT;
>
> If you're going to respin; please but {} around that.

Sure.

2015-02-19 17:05:35

by Peter Zijlstra

[permalink] [raw]
Subject: Re: [PATCH RESEND v9 04/10] sched: Make sched entity usage tracking scale-invariant

On Thu, Feb 19, 2015 at 05:02:36PM +0000, Morten Rasmussen wrote:
> > There is no actual function definition in this patch; this would hinder
> > linking, no?
>
> The function definition already lives further down in fair.c. I can move
> it up here if you prefer?

argh.. ok.

With previous patches removing all that I had missed you ;eft this one.

I thought everything got ripped and we went with new stuff.

2015-02-20 09:20:55

by Morten Rasmussen

[permalink] [raw]
Subject: Re: [PATCH RESEND v9 04/10] sched: Make sched entity usage tracking scale-invariant

On Thu, Feb 19, 2015 at 05:05:28PM +0000, Peter Zijlstra wrote:
> On Thu, Feb 19, 2015 at 05:02:36PM +0000, Morten Rasmussen wrote:
> > > There is no actual function definition in this patch; this would hinder
> > > linking, no?
> >
> > The function definition already lives further down in fair.c. I can move
> > it up here if you prefer?
>
> argh.. ok.
>
> With previous patches removing all that I had missed you ;eft this one.
>
> I thought everything got ripped and we went with new stuff.

I see. Vincent's patch only removes the old calls to the function, not
the function itself. The history might get more clear if we wipe it
completely before reintroducing again with a slightly different meaning.