[PATCH 1/1] sched/deadline: Fix fair_server runtime calculation formula

[Juri Lelli]

On 17/06/25 10:55, Peter Zijlstra wrote:
> On Sat, Jun 14, 2025 at 10:04:55AM +0800, Kuyo Chang wrote:
> > From: kuyo chang <kuyo.chang at mediatek.com>
> > 
> > [Symptom]
> > The calculation formula for fair_server runtime is based on
> > Frequency/CPU scale-invariance.
> > This will cause excessive RT latency (expect absolute time).
> > 
> > [Analysis]
> > Consider the following case under a Big.LITTLE architecture:
> > 
> > Assume the runtime is : 50,000,000 ns, and FIE/CIE as below
> > FIE: 100
> > CIE:50
> > First by FIE, the runtime is scaled to 50,000,000 * 100 >> 10 = 4,882,812
> > Then by CIE, it is further scaled to 4,882,812 * 50 >> 10 = 238,418.
> 
> What's this FIE/CIE stuff? Is that some ARM lingo?
> 
> 
> > diff --git a/kernel/sched/deadline.c b/kernel/sched/deadline.c
> > index ad45a8fea245..8bfa846cf0dc 100644
> > --- a/kernel/sched/deadline.c
> > +++ b/kernel/sched/deadline.c
> > @@ -1504,7 +1504,10 @@ static void update_curr_dl_se(struct rq *rq, struct sched_dl_entity *dl_se, s64
> >  	if (dl_entity_is_special(dl_se))
> >  		return;
> >  
> > -	scaled_delta_exec = dl_scaled_delta_exec(rq, dl_se, delta_exec);
> > +	if (dl_se == &rq->fair_server)
> > +		scaled_delta_exec = delta_exec;
> > +	else
> > +		scaled_delta_exec = dl_scaled_delta_exec(rq, dl_se, delta_exec);
> 
> Juri, the point it a bit moot atm, but is this something specific to the
> fair_server in particular, or all servers?

I believe for other servers (i.e., rt-server work from Yuri and Luca) it
might be useful to have it configurable somehow. I actually had a recent
discussion about this concerning single task entities (traditional
deadline servers) for which as well there might be cases where one might
want not to scale considering frequency and capacity.

> Because if this is something all servers require then the above is
> ofcourse wrong.

To me it looks like we want this (no scaling) for fair_server (and
possibly scx_server?) as for them we are only looking into a 'fixed
time' type of isolation. Full fledged servers (hierarchical scheduling)
maybe have it configurable, or enabled by default as a start (as we have
it today).

Best,
Juri