Re: [PATCH v3 sched_ext/for-7.3 24/36] sched_ext: Add reject DSQ for cap-rejected dispatches

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] The CPU incorrectly goes idle and strands tasks when a consumed task is diverted to reject_dsq.
--

--- Patch [24]: [PATCH v3 sched_ext/for-7.3 24/36] sched_ext: Add reject DSQ for cap-rejected dispatches ---
commit 17b65480202fa0fa8a74c72c6813fa99184ffb6d
Author: Tejun Heo <tj@kernel.org>

sched_ext: Add reject DSQ for cap-rejected dispatches

This commit introduces a kernel-internal reject DSQ for the sched_ext
sub-scheduler. Tasks dispatched to a CPU where the sub-scheduler lacks
required capabilities are parked here instead of the local DSQ, and later
handed back to the BPF scheduler to re-decide.

> diff --git a/kernel/sched/ext/ext.c b/kernel/sched/ext/ext.c
> index 87a9f1ff2b0e1..d0966412ee421 100644
> --- a/kernel/sched/ext/ext.c
> +++ b/kernel/sched/ext/ext.c

[ ... ]

> @@ -2063,7 +2079,7 @@ static void move_local_task_to_local_dsq(struct scx_sched *sch,
>  					 struct scx_dispatch_q *src_dsq,
>  					 struct rq *dst_rq)
>  {
> -	struct scx_dispatch_q *dst_dsq = &dst_rq->scx.local_dsq;
> +	struct scx_dispatch_q *dst_dsq = scx_local_or_reject_dsq(sch, dst_rq, p, &enq_flags);

[Severity: High]
If the task is diverted to the reject_dsq here, how do the callers handle the
fact that the local_dsq might still be empty?

For example, when scx_consume_dispatch_q() processes a task:

if (rq == task_rq) {
    task_unlink_from_dsq(p, dsq);
    move_local_task_to_local_dsq(sch, p, enq_flags, dsq, rq);
    raw_spin_unlock(&dsq->lock);
    return true;
}

Does this cause scx_consume_dispatch_q() to unconditionally return true, 
leading balance_one() to believe it found a task and stop searching? 

If do_pick_task_scx() then finds the local_dsq empty, could the CPU 
incorrectly enter the idle state while runnable tasks remain in the global 
DSQ, eventually triggering the SCX watchdog timeout?

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260707001229.1410929-1-tj@kernel.org?part=24