Re: [PATCH 2/3] PM: EM: Make em_cpu_energy() able to return bigger values

[Peter Zijlstra <peterz@infradead.org>]

On Fri, Jun 25, 2021 at 04:26:02PM +0100, Lukasz Luba wrote:
> The Energy Model (EM) em_cpu_energy() is responsible for providing good
> estimation regarding CPUs energy. It contains proper data structures which
> are then used during calculation. The values stored in there are in
> milli-Watts precision (or in abstract scale) smaller that 0xffff, which use
> sufficient unsigned long even on 32-bit machines. There are scenarios where
> we would like to provide calculated estimations in a better precision and
> the values might be 1000 times bigger. This patch makes possible to use
> quite big values for also 32-bit machines.
> 
> Signed-off-by: Lukasz Luba <lukasz.luba@arm.com>
> ---
>  include/linux/energy_model.h | 6 +++---
>  1 file changed, 3 insertions(+), 3 deletions(-)
> 
> diff --git a/include/linux/energy_model.h b/include/linux/energy_model.h
> index 3f221dbf5f95..2016f5a706e0 100644
> --- a/include/linux/energy_model.h
> +++ b/include/linux/energy_model.h
> @@ -101,7 +101,7 @@ void em_dev_unregister_perf_domain(struct device *dev);
>   * Return: the sum of the energy consumed by the CPUs of the domain assuming
>   * a capacity state satisfying the max utilization of the domain.
>   */
> -static inline unsigned long em_cpu_energy(struct em_perf_domain *pd,
> +static inline u64 em_cpu_energy(struct em_perf_domain *pd,
>  				unsigned long max_util, unsigned long sum_util,
>  				unsigned long allowed_cpu_cap)
>  {
> @@ -180,7 +180,7 @@ static inline unsigned long em_cpu_energy(struct em_perf_domain *pd,
>  	 *   pd_nrg = ------------------------                       (4)
>  	 *                  scale_cpu
>  	 */
> -	return ps->cost * sum_util / scale_cpu;
> +	return div_u64((u64)ps->cost * sum_util, scale_cpu);

So these patches are all rather straight forward, however.. the above is
pretty horrific on a 32bit box, and we do quite a few of them per
wakeup. Is this really worth the performance penalty on 32bit CPUs?

Do you really still care about 32bit CPUs, or is this mostly an artifact
of wanting to unconditionally increase the precision?