Re: [RFC PATCH 3/5] sched/fair: Update overloaded mask in presence of pushable task

[K Prateek Nayak]

Hello Aaron,

On 4/14/2025 7:58 AM, Aaron Lu wrote:
> Hi Prateek,
>
> On Wed, Apr 09, 2025 at 11:15:37AM +0000, K Prateek Nayak wrote:
> > In presence of pushable tasks on the CPU, set it on the newly introduced
> > "overloaded+mask" in sched_domain_shared struct. This will be used by
> > the newidle balance to limit the scanning to these overloaded CPUs since
> > they contain tasks that could be run on the newly idle target.
> >
> > Signed-off-by: K Prateek Nayak <kprateek.nayak@xxxxxxx>
> > ---
> >   kernel/sched/fair.c | 24 ++++++++++++++++++++++++
> >   1 file changed, 24 insertions(+)
> >
> > diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> > index 98d3ed2078cd..834fcdd15cac 100644
> > --- a/kernel/sched/fair.c
> > +++ b/kernel/sched/fair.c
> > @@ -8559,6 +8559,24 @@ static int find_energy_efficient_cpu(struct task_struct *p, int prev_cpu)
> >   	return target;
> >   }
> >   
> > +static inline void update_overloaded_mask(int cpu, bool contains_pushable)
> > +{
> > +	struct sched_domain_shared *sd_share = rcu_dereference(per_cpu(sd_llc_shared, cpu));
>
> I got a suspicious RCU usage warning for this line while testing your
> series. Since rq lock is held here, rcu_dereference() should not be
> necessary.

Thank you for reporting this. I'll make sure to run with LOCKDEP next
time around. Note: The performance aspect is still quite bad with this
series an the intent for the RFC was to vet the idea and to understand
if I got the basic implementation details right.

> > +	cpumask_var_t overloaded_mask;
> > +
> > +	if (!sd_share)
> > +		return;
> > +
> > +	overloaded_mask = sd_share->overloaded_mask;
> > +	if (!overloaded_mask)
> > +		return;
> > +
> > +	if (contains_pushable)
> > +		cpumask_set_cpu(cpu, overloaded_mask);
> > +	else
> > +		cpumask_clear_cpu(cpu, overloaded_mask);
> > +}

--
Thanks and Regards,
Prateek