On Friday 01 Sep 2023 at 15:03:11 (+0200), Vincent Guittot wrote:
> cpuinfo_max_freq can change at runtime because of boost as example. This
> implies that the value could not be the same than the one that has been
> used when computing the capacity of a CPU.
>
> The new arch_scale_freq_ref() returns a fixed and coherent frequency
> reference that can be used when computing a frequency based on utilization.
>
> Use this arch_scale_freq_ref() when available and fallback to
> cpuinfo.max_freq otherwise.
>
> Signed-off-by: Vincent Guittot <[email protected]>
> ---
> kernel/sched/cpufreq_schedutil.c | 29 +++++++++++++++++++++++++++--
> 1 file changed, 27 insertions(+), 2 deletions(-)
>
> diff --git a/kernel/sched/cpufreq_schedutil.c b/kernel/sched/cpufreq_schedutil.c
> index 4492608b7d7f..9996ef429e2b 100644
> --- a/kernel/sched/cpufreq_schedutil.c
> +++ b/kernel/sched/cpufreq_schedutil.c
> @@ -114,6 +114,31 @@ static void sugov_deferred_update(struct sugov_policy *sg_policy)
> }
> }
>
> +#ifdef arch_scale_freq_ref
> +/**
> + * arch_scale_freq_ref_policy - get the reference frequency of a given CPU that
> + * has been used to correlate frequency and compute capacity.
> + * @cpu: the CPU in question.
> + *
> + * Return: the reference CPU frequency.
> + */
> +static __always_inline
> +unsigned long arch_scale_freq_ref_policy(struct cpufreq_policy *policy)
This should not be an arch_ function as it's only a wrapper over an
arch_ function and not a function that different architectures might
implement differently usually in architecture specific code.
> +{
> + return arch_scale_freq_ref(policy->cpu);
It might make it easier to read if arch_scale_freq_ref() had a default
implementation that returned 0.
Then this code would simply become:
freq = arch_scale_freq_ref(policy->cpu);
if (freq)
return freq;
else if (arch_scale_freq_invariant())
return ..
..
This approach is similar to the use of arch_freq_get_on_cpu() in
cpufreq.c, and, as there, having a chosen maximum frequency of 0 would
not be a valid value.
> +}
> +#else
> +static __always_inline
> +unsigned long arch_scale_freq_ref_policy(struct cpufreq_policy *policy)
> +{
> + if (arch_scale_freq_invariant())
> + return policy->cpuinfo.max_freq;
> +
> +
> + return policy->cur;
> +}
> +#endif
> +
> /**
> * get_next_freq - Compute a new frequency for a given cpufreq policy.
> * @sg_policy: schedutil policy object to compute the new frequency for.
> @@ -139,11 +164,11 @@ static void sugov_deferred_update(struct sugov_policy *sg_policy)
> static unsigned int get_next_freq(struct sugov_policy *sg_policy,
> unsigned long util, unsigned long max)
> {
> + unsigned int freq;
> struct cpufreq_policy *policy = sg_policy->policy;
> - unsigned int freq = arch_scale_freq_invariant() ?
> - policy->cpuinfo.max_freq : policy->cur;
>
> util = map_util_perf(util);
> + freq = arch_scale_freq_ref_policy(policy);
Given its single use here, it would likely be better to place the code
above directly here, rather than create a wrapper over a few lines of
code.
Hope it helps,
Ionela.
> freq = map_util_freq(util, freq, max);
>
> if (freq == sg_policy->cached_raw_freq && !sg_policy->need_freq_update)
> --
> 2.34.1
>
>
On Thu, 21 Sept 2023 at 11:19, Ionela Voinescu <[email protected]> wrote:
>
> On Friday 01 Sep 2023 at 15:03:11 (+0200), Vincent Guittot wrote:
> > cpuinfo_max_freq can change at runtime because of boost as example. This
> > implies that the value could not be the same than the one that has been
> > used when computing the capacity of a CPU.
> >
> > The new arch_scale_freq_ref() returns a fixed and coherent frequency
> > reference that can be used when computing a frequency based on utilization.
> >
> > Use this arch_scale_freq_ref() when available and fallback to
> > cpuinfo.max_freq otherwise.
> >
> > Signed-off-by: Vincent Guittot <[email protected]>
> > ---
> > kernel/sched/cpufreq_schedutil.c | 29 +++++++++++++++++++++++++++--
> > 1 file changed, 27 insertions(+), 2 deletions(-)
> >
> > diff --git a/kernel/sched/cpufreq_schedutil.c b/kernel/sched/cpufreq_schedutil.c
> > index 4492608b7d7f..9996ef429e2b 100644
> > --- a/kernel/sched/cpufreq_schedutil.c
> > +++ b/kernel/sched/cpufreq_schedutil.c
> > @@ -114,6 +114,31 @@ static void sugov_deferred_update(struct sugov_policy *sg_policy)
> > }
> > }
> >
> > +#ifdef arch_scale_freq_ref
> > +/**
> > + * arch_scale_freq_ref_policy - get the reference frequency of a given CPU that
> > + * has been used to correlate frequency and compute capacity.
> > + * @cpu: the CPU in question.
> > + *
> > + * Return: the reference CPU frequency.
> > + */
> > +static __always_inline
> > +unsigned long arch_scale_freq_ref_policy(struct cpufreq_policy *policy)
>
> This should not be an arch_ function as it's only a wrapper over an
> arch_ function and not a function that different architectures might
> implement differently usually in architecture specific code.
I expect this function to disappear at some point once all arch will
use it that why I named it arch_* but I can rename it
>
> > +{
> > + return arch_scale_freq_ref(policy->cpu);
>
> It might make it easier to read if arch_scale_freq_ref() had a default
> implementation that returned 0.
I will use the suggestion made by Peter to have only one function
>
> Then this code would simply become:
>
> freq = arch_scale_freq_ref(policy->cpu);
> if (freq)
> return freq;
> else if (arch_scale_freq_invariant())
> return ..
> ..
>
> This approach is similar to the use of arch_freq_get_on_cpu() in
> cpufreq.c, and, as there, having a chosen maximum frequency of 0 would
> not be a valid value.
>
> > +}
> > +#else
> > +static __always_inline
> > +unsigned long arch_scale_freq_ref_policy(struct cpufreq_policy *policy)
> > +{
> > + if (arch_scale_freq_invariant())
> > + return policy->cpuinfo.max_freq;
> > +
> > +
> > + return policy->cur;
> > +}
> > +#endif
> > +
> > /**
> > * get_next_freq - Compute a new frequency for a given cpufreq policy.
> > * @sg_policy: schedutil policy object to compute the new frequency for.
> > @@ -139,11 +164,11 @@ static void sugov_deferred_update(struct sugov_policy *sg_policy)
> > static unsigned int get_next_freq(struct sugov_policy *sg_policy,
> > unsigned long util, unsigned long max)
> > {
> > + unsigned int freq;
> > struct cpufreq_policy *policy = sg_policy->policy;
> > - unsigned int freq = arch_scale_freq_invariant() ?
> > - policy->cpuinfo.max_freq : policy->cur;
> >
> > util = map_util_perf(util);
> > + freq = arch_scale_freq_ref_policy(policy);
>
> Given its single use here, it would likely be better to place the code
> above directly here, rather than create a wrapper over a few lines of
> code.
>
> Hope it helps,
> Ionela.
>
> > freq = map_util_freq(util, freq, max);
> >
> > if (freq == sg_policy->cached_raw_freq && !sg_policy->need_freq_update)
> > --
> > 2.34.1
> >
> >