Re: [PATCH v2 0/2] sched/cpufreq: fix schedutil's boost frequency handling

[Ananthu C V]

Hi Oleg,

On Thu, Sep 17, 2026 at 06:05:06PM +0200, Oleg Keri wrote:
> Hi Ananthu,
> 
> Tested both patches on a Lenovo Yoga Slim 7x Gen 11 (Glymur,
> scmi-cpufreq) on top of next-20260915.
> 
> 1/2: with boost off at boot the little cores get cpu_capacity=553,
> which matches the 4723200 boost maximum as the reference (4032000
> would give 647). A workload pinned to cpu6 runs at 4032000 with boost
> off and 4723200 with boost on, 3.14 s against 2.50 s.
> 
> 2/2:
> 
>   # cat policy6/scaling_max_freq
>   4032000
>   # echo 1 > boost
>   # cat policy6/scaling_max_freq
>   4723200
>   # echo 0 > boost
>   # cat policy6/scaling_max_freq
>   4032000
> 
> One observation: policy6/cpuinfo_max_freq still reads 4723200 after
> boost is switched off again.

cpuinfo->max_freq has a guard that only updates it upwards, so it is not
able to come down. This is by design, to preserve driver set values. That
is what the second patch is trying to work around. Runtime enforcement of
boost frequency handling will try not to rely on cpuinfo_max_freq whenever
possible with this series, allowing the value to come back to a non boost
freq.

> Tested-by: Oleg Keri <okerixx@xxxxxxxxx> # Lenovo Yoga Slim 7x Gen 11

Thanks for testing the patches.

Best,
Ananthu