Re: scheduler performance regression since v6.11

[Dietmar Eggemann <dietmar.eggemann@arm.com>]

On 22/05/2025 10:48, Peter Zijlstra wrote:
> On Wed, May 21, 2025 at 04:54:47PM +0200, Peter Zijlstra wrote:
>> On Tue, May 20, 2025 at 09:38:31PM +0200, Peter Zijlstra wrote:
>>> On Tue, May 20, 2025 at 04:38:09PM +0200, Dietmar Eggemann wrote:
>>>
>>>> 3840cbe24cf0 - sched: psi: fix bogus pressure spikes from aggregation race
>>>>
>>>> With CONFIG_PSI enabled we call cpu_clock(cpu) now multiple times (up to
>>>> 4 times per task switch in my setup) in:
>>>>
>>>> __schedule() -> psi_sched_switch() -> psi_task_switch() ->
>>>> psi_group_change().
>>>>
>>>> There seems to be another/other v6.12 related patch(es) later which
>>>> cause(s) another 4% regression I yet have to discover.
>>>
>>> Urgh, let me add this to the pile to look at. Thanks!
>>
>> Does something like the compile tested only hackery below work?
> 
> possibly better hackery :-)

Ah OK, moving the seqlock protection outside of the single CPU clock read.

schbench number is up again ~9% with 'tip sched/core' on this VM thanks
to this single CPU clock read per task switch.

Doesn't show a big effect on AWS' 'hammerdb-mysqld' repro-collection
test though. (only 0.5% on NOPM (New Operations Per Minue)) even though
mysqld threads are running in '/system.slice/mysql.service'.

schbench is way more aggressive on task switch.

[...]