Re: [PATCH v3 3/5] sched/fair: Switch to task based throttle model

[K Prateek Nayak <kprateek.nayak@amd.com>]

Hello Florian,

On 9/4/2025 1:03 PM, Bezdeka, Florian wrote:
> Hi Aaron,
> 
> On Wed, 2025-09-03 at 19:35 +0800, Aaron Lu wrote:
>>
>>> [..snip..]
>>>>
>>
>> Just want to add that with the above diff applied, I also tested
>> songtang's stress test[0] and Jan's rt deadlock reproducer[1] and didn't
>> see any problem.
> 
> Thanks a lot for taking the reproducer into account. But: To trigger
> PREEMPT_RT needs to be enabled, otherwise the rwlock in the epoll
> infrastructure (that we highly stress) won't sleep. Just to be sure:
> PREEMPT_RT was enabled on your end?

I've too tested this series with Jan's reproducer on PREEMPT_RT and
can confirm no hang was triggered over a full weekend run lasting
more than 72Hrs with Aaron's changes applied on top of v6.16-rc6
based kernel.

> 
>>
>> [0]: https://lore.kernel.org/lkml/20250715072218.GA304@bytedance/
>> [1]: https://lore.kernel.org/all/7483d3ae-5846-4067-b9f7-390a614ba408@siemens.com/

I used the same reproducer from [1] with small modification to
run.sh to periodically move epoll-stall and epoll-stall-writer
to one particular CPU and keep changing this CPU ever minute. 

-- 
Thanks and Regards,
Prateek