Re: [PATCH 01/15] sched: Make NOHZ CFS bandwidth checks follow proxy donor

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] NOHZ_FULL isolation is broken for RT tasks that preempt a constrained FAIR task because the tick dependency is erroneously kept active.
--

--- Patch [1]: [PATCH 01/15] sched: Make NOHZ CFS bandwidth checks follow proxy donor ---
commit 4ed25113b5b2751dc0d63ef337fd5d561e7a7554
Author: Andrea Righi <[email protected]>

sched: Make NOHZ CFS bandwidth checks follow proxy donor

This commit updates the NOHZ CFS bandwidth checks to evaluate the proxy
donor rather than the physical execution context. This ensures that a
retained proxy donor correctly enforces bandwidth accounting even when
the lock owner is queued in another scheduling class.

Link: https://lore.kernel.org/r/[email protected]

> diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> index 0697ed0f1c3d2..6e835d2780cfd 100644
> --- a/kernel/sched/core.c
> +++ b/kernel/sched/core.c

[ ... ]

> @@ -1431,6 +1428,14 @@ bool sched_can_stop_tick(struct rq *rq)
>  	if (rq->dl.dl_nr_running)
>  		return false;
>  
> +	/*
> +	 * The selected scheduling context can be a constrained FAIR donor even
> +	 * when rq->curr is an RT task. Check it before the RT fast paths below,
> +	 * which may report that the tick can stop for a throttled RT context.
> +	 */
> +	if (__need_bw_check(rq->donor) && cfs_task_bw_constrained(rq->donor))
> +		return false;
> +

[Severity: High]
Does evaluating rq->donor here during enqueue cause an issue when an RT task
wakes up on a NOHZ_FULL CPU that is currently running a bandwidth-constrained
FAIR task?

During the RT task wakeup, the enqueue path reaches this check before the
context switch:

enqueue_task_rt() -> add_nr_running() -> sched_update_tick_dependency() ->
sched_can_stop_tick()

At this point, rq->donor has not yet been updated and still points to the
currently running FAIR task. This causes the check to return false, which
keeps the TICK_DEP_BIT_SCHED dependency set.

After the context switch to the RT task, rq->donor is updated, but
sched_update_tick_dependency() is not called again on context switch, and
set_next_task_rt() does not actively clear the tick dependency.

Could this cause the incoming RT task to run with the timer tick erroneously
enabled, breaking NOHZ_FULL isolation?

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1