Message-ID: <ZQwGHBGg9B8SjzeS@arm.com>
Date:   Thu, 21 Sep 2023 10:00:37 +0100
From:   Ionela Voinescu <ionela.voinescu@....com>
To:     Vincent Guittot <vincent.guittot@...aro.org>
Cc:     linux@...linux.org.uk, catalin.marinas@....com, will@...nel.org,
        paul.walmsley@...ive.com, palmer@...belt.com,
        aou@...s.berkeley.edu, sudeep.holla@....com,
        gregkh@...uxfoundation.org, rafael@...nel.org, mingo@...hat.com,
        peterz@...radead.org, juri.lelli@...hat.com,
        dietmar.eggemann@....com, rostedt@...dmis.org, bsegall@...gle.com,
        mgorman@...e.de, bristot@...hat.com, vschneid@...hat.com,
        viresh.kumar@...aro.org, linux-arm-kernel@...ts.infradead.org,
        linux-kernel@...r.kernel.org, linux-riscv@...ts.infradead.org,
        linux-pm@...r.kernel.org, conor.dooley@...rochip.com,
        suagrfillet@...il.com, ajones@...tanamicro.com, lftan@...nel.org
Subject: Re: [PATCH 2/4] topology: add a new arch_scale_freq_reference

Hi Vincent,

On Friday 01 Sep 2023 at 15:03:10 (+0200), Vincent Guittot wrote:
> Create a new method to get a unique and fixed max frequency. Currently
> cpuinfo.max_freq or last item of performance domain are used as the max
> frequency when computing the frequency for a level of utilization but:
> - cpuinfo_max_freq can change at runtime. boost is one example of
>   such change.
> - cpuinfo.max_freq and last item of the PD can be different leading to
>   different results betwen cpufreq and energy model.
> 
> We need to save the max frequency that has been used when computing the
> CPUs capacity and use this fixed and coherent value to convert between
> frequency and CPU's capacity.
> 
> In fact, we already save the frequency that has been used when computing
> the capacity of each CPU. We extend the precision to save khZ instead of
> Mhz currently and we modify the type to be aligned with other variables
> used when converting frequency to capacity and the other way.
> 
> Signed-off-by: Vincent Guittot <vincent.guittot@...aro.org>
> ---
>  arch/arm/include/asm/topology.h   | 1 +
>  arch/arm64/include/asm/topology.h | 1 +
>  arch/riscv/include/asm/topology.h | 1 +
>  drivers/base/arch_topology.c      | 9 +++------
>  include/linux/arch_topology.h     | 7 +++++++
>  5 files changed, 13 insertions(+), 6 deletions(-)
> 
> diff --git a/arch/arm/include/asm/topology.h b/arch/arm/include/asm/topology.h
> index c7d2510e5a78..853c4f81ba4a 100644
> --- a/arch/arm/include/asm/topology.h
> +++ b/arch/arm/include/asm/topology.h
> @@ -13,6 +13,7 @@
>  #define arch_set_freq_scale topology_set_freq_scale
>  #define arch_scale_freq_capacity topology_get_freq_scale
>  #define arch_scale_freq_invariant topology_scale_freq_invariant
> +#define arch_scale_freq_ref topology_get_freq_ref

The "reference" frequency or performance of a CPU has a specific meaning
in the ACPI specification as a fixed frequency of a constant clock and its
associated performance level, usually used for performance to frequency
conversions. This is not guaranteed to be the maximum
performance/frequency and it's definitely not the case for arm systems
where this is tied to the system timer - 1-50Mhz.

So I believe it might create some confusion to call this "reference"
frequency.

Any reason for not calling it arch_scale_freq_max? I know not all max
frequencies are equal :) but it would still drive the point that this is
intended to act as a chosen maximum for the scheduler's use, especially
if there's a well place comment.

>  #endif
>  
>  /* Replace task scheduler's default cpu-invariant accounting */
> diff --git a/arch/arm64/include/asm/topology.h b/arch/arm64/include/asm/topology.h
> index 9fab663dd2de..a323b109b9c4 100644
> --- a/arch/arm64/include/asm/topology.h
> +++ b/arch/arm64/include/asm/topology.h
> @@ -23,6 +23,7 @@ void update_freq_counters_refs(void);
>  #define arch_set_freq_scale topology_set_freq_scale
>  #define arch_scale_freq_capacity topology_get_freq_scale
>  #define arch_scale_freq_invariant topology_scale_freq_invariant
> +#define arch_scale_freq_ref topology_get_freq_ref
>  
>  #ifdef CONFIG_ACPI_CPPC_LIB
>  #define arch_init_invariance_cppc topology_init_cpu_capacity_cppc
> diff --git a/arch/riscv/include/asm/topology.h b/arch/riscv/include/asm/topology.h
> index e316ab3b77f3..61183688bdd5 100644
> --- a/arch/riscv/include/asm/topology.h
> +++ b/arch/riscv/include/asm/topology.h
> @@ -9,6 +9,7 @@
>  #define arch_set_freq_scale		topology_set_freq_scale
>  #define arch_scale_freq_capacity	topology_get_freq_scale
>  #define arch_scale_freq_invariant	topology_scale_freq_invariant
> +#define arch_scale_freq_ref		topology_get_freq_ref
>  
>  /* Replace task scheduler's default cpu-invariant accounting */
>  #define arch_scale_cpu_capacity	topology_get_cpu_scale
> diff --git a/drivers/base/arch_topology.c b/drivers/base/arch_topology.c
> index b741b5ba82bd..75fa67477a9d 100644
> --- a/drivers/base/arch_topology.c
> +++ b/drivers/base/arch_topology.c
> @@ -26,7 +26,7 @@
>  static DEFINE_PER_CPU(struct scale_freq_data __rcu *, sft_data);
>  static struct cpumask scale_freq_counters_mask;
>  static bool scale_freq_invariant;
> -static DEFINE_PER_CPU(u32, freq_factor) = 1;
> +DEFINE_PER_CPU(unsigned long, freq_factor) = 1;

Given its new wider use, it might be good for this to get a more
relevant name as well. Previously freq_factor made sense for its role
in using the dmips/mhz values to obtain CPU capacities. But if this is
now returned from arch_scale_freq_ref() it would be more difficult still
to easily understand what this value is supposed to reflect, when
reading the function name or the per-cpu variable name alone.

Thanks,
Ionela.

>  
>  static bool supports_scale_freq_counters(const struct cpumask *cpus)
>  {
> @@ -183,10 +183,7 @@ void topology_update_thermal_pressure(const struct cpumask *cpus,
>  
>  	cpu = cpumask_first(cpus);
>  	max_capacity = arch_scale_cpu_capacity(cpu);
> -	max_freq = per_cpu(freq_factor, cpu);
> -
> -	/* Convert to MHz scale which is used in 'freq_factor' */
> -	capped_freq /= 1000;
> +	max_freq = arch_scale_freq_ref(cpu);
>  
>  	/*
>  	 * Handle properly the boost frequencies, which should simply clean
> @@ -411,7 +408,7 @@ init_cpu_capacity_callback(struct notifier_block *nb,
>  	cpumask_andnot(cpus_to_visit, cpus_to_visit, policy->related_cpus);
>  
>  	for_each_cpu(cpu, policy->related_cpus)
> -		per_cpu(freq_factor, cpu) = policy->cpuinfo.max_freq / 1000;
> +		per_cpu(freq_factor, cpu) = policy->cpuinfo.max_freq;
>  
>  	if (cpumask_empty(cpus_to_visit)) {
>  		topology_normalize_cpu_scale();
> diff --git a/include/linux/arch_topology.h b/include/linux/arch_topology.h
> index a07b510e7dc5..7a2dba9c3dc0 100644
> --- a/include/linux/arch_topology.h
> +++ b/include/linux/arch_topology.h
> @@ -27,6 +27,13 @@ static inline unsigned long topology_get_cpu_scale(int cpu)
>  
>  void topology_set_cpu_scale(unsigned int cpu, unsigned long capacity);
>  
> +DECLARE_PER_CPU(unsigned long, freq_factor);
> +
> +static inline unsigned long topology_get_freq_ref(int cpu)
> +{
> +	return per_cpu(freq_factor, cpu);
> +}
> +
>  DECLARE_PER_CPU(unsigned long, arch_freq_scale);
>  
>  static inline unsigned long topology_get_freq_scale(int cpu)
> -- 
> 2.34.1
> 
> 

Powered by blists - more mailing lists