Re: [PATCH v2 1/2] arm64: Provide an AMU-based version of arch_freq_get_on_cpu

[Ionela Voinescu]

Hi Beata,

On Monday 27 Nov 2023 at 16:08:37 (+0000), Beata Michalska wrote:
> With the Frequency Invariance Engine (FIE) being already wired up with
> sched tick and making use of relevant (core counter and constant
> counter) AMU counters, getting the current frequency for a given CPU
> on supported platforms, can be achieved by utilizing the frequency scale
> factor which reflects an average CPU frequency for the last tick period
> length.
> 
> Suggested-by: Ionela Voinescu <ionela.voinescu@xxxxxxx>
> Signed-off-by: Beata Michalska <beata.michalska@xxxxxxx>
> Reviewed-by: Sudeep Holla <sudeep.holla@xxxxxxx>
> ---
> 
> Notes:
>     Due to [1], if merged, there might be a need to modify the patch to
>     accommodate changes [1] introduces:
>     
>     	freq = cpufreq_get_hw_max_freq(cpu) >> SCHED_CAPACITY_SHIFT
>     	to
>     	freq = per_cpu(capacity_freq_ref, cpu); >> SCHED_CAPACITY_SHIFT
>     [1]
>     https://lore.kernel.org/linux-arm-kernel/20231121154349.GA1938@willie-the-truck/T/#mcb018d076dbce6f60ed2779634a9b6ffe622641e
> 
>  arch/arm64/kernel/topology.c | 39 ++++++++++++++++++++++++++++++++++++
>  1 file changed, 39 insertions(+)
> 
> diff --git a/arch/arm64/kernel/topology.c b/arch/arm64/kernel/topology.c
> index 615c1a20129f..ae2445f6e7da 100644
> --- a/arch/arm64/kernel/topology.c
> +++ b/arch/arm64/kernel/topology.c
> @@ -17,6 +17,7 @@
>  #include <linux/cpufreq.h>
>  #include <linux/init.h>
>  #include <linux/percpu.h>
> +#include <linux/sched/isolation.h>
>  
>  #include <asm/cpu.h>
>  #include <asm/cputype.h>
> @@ -186,6 +187,44 @@ static void amu_scale_freq_tick(void)
>  	this_cpu_write(arch_freq_scale, (unsigned long)scale);
>  }
>  
> +unsigned int arch_freq_get_on_cpu(int cpu)
> +{
> +	unsigned int freq;
> +	u64 scale;
> +
> +	if (!cpumask_test_cpu(cpu, amu_fie_cpus))
> +		return 0;
> +
> +	/*
> +	 * For those CPUs that are in full dynticks mode, try an alternative
> +	 * source for the counters (and thus freq scale),
> +	 * if available for given policy
> +	 */
> +	if (!housekeeping_cpu(cpu, HK_TYPE_TICK)) {
> +		struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
> +		int ref_cpu = nr_cpu_ids;
> +
> +		if (cpumask_intersects(housekeeping_cpumask(HK_TYPE_TICK),
> +				       policy->cpus))
> +			ref_cpu = cpumask_nth_and(cpu, policy->cpus,
> +						  housekeeping_cpumask(HK_TYPE_TICK));
> +		cpufreq_cpu_put(policy);
> +		if (ref_cpu >= nr_cpu_ids)
> +			return 0;
> +		cpu = ref_cpu;
> +	}
> +
> +	/*
> +	 * Reversed computation to the one used to determine
> +	 * the arch_freq_scale value
> +	 * (see amu_scale_freq_tick for details)
> +	 */
> +	scale = per_cpu(arch_freq_scale, cpu);

Any reason for not using arch_scale_freq_capacity() here?

To me it seems a bit nicer to use the "official" function to return the
frequency scale factor.

> +	freq = cpufreq_get_hw_max_freq(cpu) >> SCHED_CAPACITY_SHIFT;

Given Vincent's patch at [1] I think here might be best to call
arch_scale_freq_ref() instead. That's because the frequency scale factor
will use that frequency as the maximum frequency in its calculations and
we'd not want to use a different one here.

The annoyance is coping with capacity_freq_ref not having been set
yet, and that would be easy if capacity_freq_ref was initialized to 0.
Luckily with Vincent's changes it can now be 0. I'll comments on his
patches and ask him to make this change.

So I think you can safely use arch_scale_freq_ref() here. If
arch_scale_freq_ref() returns 0, arch_freq_get_on_cpu() will just return
0 as well.

[1] https://lore.kernel.org/lkml/20231109101438.1139696-8-vincent.guittot@xxxxxxxxxx/

> +	freq *= scale;

In some scenarios the frequencies visible to cpufreq might not look like
actual frequencies, but some scaled abstract performance values. One
example is cppc_cpufreq when one does not provide the optional frequency
information in the CPC objects but just the performance information.

Therefore the maximum frequency seen here can be quite a small value, so
it might be best to do the multiplication first and the shift after that.

Thanks,
Ionela.

> +	return freq;
> +}
> +
>  static struct scale_freq_data amu_sfd = {
>  	.source = SCALE_FREQ_SOURCE_ARCH,
>  	.set_freq_scale = amu_scale_freq_tick,
> -- 
> 2.25.1
>