Re: [RFC PATCH v5 18/29] sched/core: Cgroup v2 support

[luca abeni]

On Thu, 7 May 2026 09:01:03 +0200
Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:

> On Wed, May 06, 2026 at 09:58:02PM +0200, luca abeni wrote:
> > Hi Peter,
> > 
> > On Tue, 5 May 2026 16:59:22 +0200
> > Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
> >   
> > > On Thu, Apr 30, 2026 at 11:38:22PM +0200, Yuri Andriaccio wrote:  
> > > > From: luca abeni <luca.abeni@xxxxxxxxxxxxxxx>
> > > > 
> > > > Make rt_runtime_us and rt_period_us virtual files accessible
> > > > also to the cgroup v2 controller, effectively enabling the
> > > > RT_GROUP_SCHED mechanism to cgroups v2.    
> > > 
> > > Can we have a blub about why only strict periodic servers; eg.
> > > why no sporadic? and such...  
> > 
> > Maybe I am misunderstanding your question, anyway: the file is
> > called "rt_runtime_us", but the scheduling algorithm used to
> > schedule the cgroup is SCHED_DEADLINE.
> > So, we do not use a strictly periodic server, but a CBS, that can
> > also support sporadic / non-periodic activations.  
> 
> The interface only exposes runtime and period, as such we can only
> configure strict periodic servers (with implicit deadline). And I'm
> thinking this makes sense, esp. to start off with, but I also think it
> makes sense to explicitly call that out.

Ah, I understand now: you are thinking about SCHED_DEADLINE with
deadline<period, right?
(sorry, I originally misunderstood and I was thinking about sporadic
activation patterns, which are already supported)

Yes, I think we can easily add the possibility to also set the
"deadline" parameter (with deafault "deadline=period")



			Thanks,
				Luca

> 
> State that this does not allow configuring sporadic servers, and
> hand-wave a reason for why not.
> 
> Or, if we struggle to justify it, perhaps add deadline, dunno.