Re: [PATCH v3 0/5] Defer throttle when task exits to user

[Aaron Lu <ziqianlu@bytedance.com>]

On Mon, Aug 04, 2025 at 02:21:30PM +0530, K Prateek Nayak wrote:
> Hello Aaron,
> 
> On 7/15/2025 12:46 PM, Aaron Lu wrote:
> > v3:
> > - Keep throttled cfs_rq's PELT clock running as long as it still has
> >   entity queued, suggested by Benjamin Segall. I've folded this change
> >   into patch3;
> > - Rebased on top of tip/sched/core, commit 2885daf47081
> >   ("lib/smp_processor_id: Make migration check unconditional of SMP").
> > 
> > Hi Prateek,
> > I've kept your tested-by tag(Thanks!) for v2 since I believe this pelt
> > clock change should not affect things much, but let me know if you don't
> > think that is appropriate.
> 
> I've officially tested this series so it should be fine :)

Good to hear this :)

> 
> In addition to Jan's test, I also did some sanity test looking at PELT
> and everything looks good for the simplest case - once busy loop inside
> a cgroup that gets throttled. The per-task throttling behavior is
> identical to the current behavior for this simplest case.
> 
> If I find time, I'll look into nested hierarchies with wakeups to see
> if I can spot anything odd there. I don't really have a good control
> setup to compare against here but so far I haven't found anything odd
> and it works as intended.
> 

Thanks for all these tests.

Best regards,
Aaron

> > 
> > Tests I've done:
> > - Jan's rt deadlock reproducer[1]. Without this series, I saw rcu-stalls
> >   within 2 minutes and with this series, I do not see rcu-stalls after
> >   10 minutes.
> > - A stress test that creates a lot of pressure on fork/exit path and
> >   cgroup_threadgroup_rwsem. Without this series, the test will cause
> >   task hung in about 5 minutes and with this series, no problem found
> >   after several hours. Songtang wrote this test script and I've used it
> >   to verify the patches, thanks Songtang.
> 
> I just noticed this script. I'll give this a spin too when I test
> nested hierarchies.
> 
> -- 
> Thanks and Regards,
> Prateek