lkml.org 
[lkml]   [2022]   [Nov]   [16]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
SubjectRe: [PATCH v1] Revert "cpufreq: schedutil: Move max CPU capacity to sugov_policy"
From


On 11/15/22 22:35, Saravana Kannan wrote:
> On Thu, Nov 10, 2022 at 11:57 AM Sam Wu <wusamuel@google.com> wrote:
>>
>> This reverts commit 6d5afdc97ea71958287364a1f1d07e59ef151b11.
>>
>> On a Pixel 6 device, it is observed that this commit increases
>> latency by approximately 50ms, or 20%, in migrating a task
>> that requires full CPU utilization from a LITTLE CPU to Fmax
>> on a big CPU. Reverting this change restores the latency back
>> to its original baseline value.
>>
>> Fixes: 6d5afdc97ea7 ("cpufreq: schedutil: Move max CPU capacity to sugov_policy")
>> Cc: Lukasz Luba <lukasz.luba@arm.com>
>> Cc: Saravana Kannan <saravanak@google.com>
>> Cc: Isaac J. Manjarres <isaacmanjarres@google.com>
>> Signed-off-by: Sam Wu <wusamuel@google.com>
>
> Rafael, can we pick this up please?
>
> Lukasz, no objections to the idea itself, but it's causing regression
> and we'd like to revert and then fix it.

If you see this in mainline kernel, then I'm fine with reverting it.
Then I will have to trace why this CPU capacity value can change over
time in mainline kernel (while it shouldn't, because we register the
cpufreq policy and the governor later, after we calculate the capacity
in arch_topology.c). Maybe something has changed in mainline in the
meantime in this CPU capacity setup code, which caused this side effect.

I know that android-mainline has some different setup code for those
custom vendor governors. I just want to eliminate this bit and be on the
same page.

Regards,
Lukasz

>
> -Saravana
>

\
 
 \ /
  Last update: 2022-11-16 12:55    [W:0.097 / U:1.132 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site