Re: [PATCH] sched/deadline: Fix race in push_dl_task

[Juri Lelli <juri.lelli@redhat.com>]

Hi Harshit,

Thanks for this!

On 07/03/25 20:42, Harshit Agarwal wrote:
> This fix is the deadline version of the change made to the rt scheduler
> here:
> https://lore.kernel.org/lkml/20250225180553.167995-1-harshit@nutanix.com/
> Please go through the original change for more details on the issue.

I don't think we want this kind of URLs in the changelog, as URL might
disappear while the history remains (at least usually a little longer
:). Maybe you could add a very condensed version of the description of
the problem you have on the other fix?
 
> In this fix we bail out or retry in the push_dl_task, if the task is no
> longer at the head of pushable tasks list because this list changed
> while trying to lock the runqueue of the other CPU.
> 
> Signed-off-by: Harshit Agarwal <harshit@nutanix.com>
> Cc: stable@vger.kernel.org
> ---
>  kernel/sched/deadline.c | 25 +++++++++++++++++++++----
>  1 file changed, 21 insertions(+), 4 deletions(-)
> 
> diff --git a/kernel/sched/deadline.c b/kernel/sched/deadline.c
> index 38e4537790af..c5048969c640 100644
> --- a/kernel/sched/deadline.c
> +++ b/kernel/sched/deadline.c
> @@ -2704,6 +2704,7 @@ static int push_dl_task(struct rq *rq)
>  {
>  	struct task_struct *next_task;
>  	struct rq *later_rq;
> +	struct task_struct *task;
>  	int ret = 0;
>  
>  	next_task = pick_next_pushable_dl_task(rq);
> @@ -2734,15 +2735,30 @@ static int push_dl_task(struct rq *rq)
>  
>  	/* Will lock the rq it'll find */
>  	later_rq = find_lock_later_rq(next_task, rq);
> -	if (!later_rq) {
> -		struct task_struct *task;
> +	task = pick_next_pushable_dl_task(rq);
> +	if (later_rq && (!task || task != next_task)) {
> +		/*
> +		 * We must check all this again, since
> +		 * find_lock_later_rq releases rq->lock and it is
> +		 * then possible that next_task has migrated and
> +		 * is no longer at the head of the pushable list.
> +		 */
> +		double_unlock_balance(rq, later_rq);
> +		if (!task) {
> +			/* No more tasks */
> +			goto out;
> +		}
>  
> +		put_task_struct(next_task);
> +		next_task = task;
> +		goto retry;

I fear we might hit a pathological condition that can lead us into a
never ending (or very long) loop. find_lock_later_rq() tries to find a
later_rq for at most DL_MAX_TRIES and it bails out if it can't.

Maybe to discern between find_lock_later_rq() callers we can use
dl_throttled flag in dl_se and still implement the fix in find_lock_
later_rq()? I.e., fix similar to the rt.c patch in case the task is not
throttled (so caller is push_dl_task()) and not rely on pick_next_
pushable_dl_task() if the task is throttled.

What do you think?

Thanks,
Juri