RE: sched/cpufreq: Rework schedutil governor performance estimation - Regression bisected

[Doug Smythies]

On 2024.02.24 06:31 Thorsten wrote:
> On 24.02.24 15:11, Rafael J. Wysocki wrote:
>> On Sat, Feb 24, 2024 at 2:44 PM Linux regression tracking (Thorsten
>>> Leemhuis) <regressions@xxxxxxxxxxxxx> wrote:
>>> On 16.02.24 14:17, Vincent Guittot wrote:
>>>> On Thu, 15 Feb 2024 at 23:53, Doug Smythies <dsmythies@xxxxxxxxx> wrote:
>>>>>
>>>>> This email thread appears as if it might be moving away from a regression
>>>>> caused by your commit towards a conclusion that your commit exposed
>>>>> a pre-existing bug in the intel_psate.c code.
>>>> Ok
>>>
>>> Well, even in that case it's a regression that must be fixed -- ideally
>>> before 6.8. Did anything happen towards that?
>>>
>>> I noticed that Doug send the fix "cpufreq: intel_pstate: fix pstate
>>> limits enforcement for adjust_perf call back":
>>> https://lore.kernel.org/all/20240217213010.2466-1-dsmythies@xxxxxxxxx/
>>>
>>> Is that supposed to fix the problem? 

Yes it fixes the preexisting issue exposed by 9c0b4bb7f630.

>>>Looks a bit like it, but I'm not
>>> totally sure. In that case I'd say it likely should be applied to 6.8,
>>> but Rafael apparently applied it to 6.9.
>> 
>> This hasn't reached linux-next yet, so I rebased it on top of -rc5 in
>> order to push it as a 6.8 fix.
>
> Ahh, great, many thx!

Yes, thanks.

>>> I'd also say that a Fixes: would be good as well (to ensure that fix is
>>> also backported in case anyone backports 9c0b4bb7f630), but I know that
>>> subsystems handle this differently.

I left the fixes tag off on purpose, because there was never anything wrong
with 9c0b4bb7f630. Apologies to Vincent for wasting his time, but thanks
for the help finding the actual issue.
 
>> So I added a Fixes: tag to it, but it points to the original change
>> that missed the check.
> Yeah, that totally works for me as well. Again: many thx!

Yes, thanks. A fix tag pointing to the original commit makes sense.

.. Doug