Re: [PATCH 1/2] sched_ext: Add lazy preemption support

[Andrea Righi <arighi@nvidia.com>]

On Tue, Sep 15, 2026 at 07:57:52PM +0000, sashiko-bot@kernel.org wrote:
> Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
> - [Low] sched_ext: redundant idle kick processing for lazy preemptions
> --
> 
> commit 42fa214a424ea8999256e25137769c2bfce81537
> Author: Andrea Righi <arighi@nvidia.com>
> sched_ext: Add lazy preemption support
> 
> This commit adds lazy preemption support to sched_ext, allowing BPF
> schedulers to request deferred rescheduling. It introduces new enqueue and
> kick flags for lazy preemption, and updates the task slice expiry logic.
> 
> > diff --git a/kernel/sched/ext/ext.c b/kernel/sched/ext/ext.c
> > index f875c50f00bbe..819995667cc26 100644
> > --- a/kernel/sched/ext/ext.c
> > +++ b/kernel/sched/ext/ext.c
> [ ... ]
> > @@ -8574,6 +8626,14 @@ static void kick_cpus_irq_workfn(struct irq_work *irq_work)
> >  			cpumask_clear_cpu(cpu, pcpu->cpus_to_kick_if_idle);
> >  		}
> >  
> > +		/*
> > +		 * kick_one_cpu() clears the lazy bit of every cpu it visited
> > +		 * above; visit the remaining requests which contain lazy
> > +		 * preemption, see scx_kick_cpu().
> > +		 */
> > +		for_each_cpu(cpu, pcpu->cpus_to_preempt_lazy)
> > +			should_wait |= kick_one_cpu(cpu, pcpu, this_rq, ksyncs);
> > +
> 
> [Severity: Low]
> Does this loop in kick_cpus_irq_workfn() intentionally skip clearing
> cpus_to_kick_if_idle for the CPUs it processes?
> 
> If a BPF scheduler issues both an idle kick and a lazy preemption kick on
> the same target within a narrow window, but without a non-lazy kick, the
> target CPU is serviced by this loop. 
> 
> Because this loop fails to clear cpus_to_kick_if_idle for the visited CPU,
> the subsequent idle loop immediately below it will redundantly lock and
> evaluate the same CPU again. 
> 
> Could this cause a minor performance inefficiency due to the redundant
> evaluation, and should cpumask_clear_cpu(cpu, pcpu->cpus_to_kick_if_idle)
> be added here as well?

This looks like a false positive. SCX_KICK_IDLE and SCX_KICK_PREEMPT_LAZY may
accumulate from separate calls. Processing the lazy request does not always
satisfy the idle request: if the CPU is in SCX_RQ_IN_DISPATCH, the subsequent
idle pass intentionally upgrades it to an immediate reschedule. Unconditionally
clearing cpus_to_kick_if_idle in the lazy loop could lose that request.

-Andrea

> 
> >  		for_each_cpu(cpu, pcpu->cpus_to_kick_if_idle) {
> >  			kick_one_cpu_if_idle(cpu, pcpu, this_rq);
> >  			cpumask_clear_cpu(cpu, pcpu->cpus_to_kick_if_idle);
> >  		}
> 
> -- 
> Sashiko AI review · https://sashiko.dev/#/patchset/20260915194611.2674127-1-arighi@nvidia.com?part=1