Re: [PATCH sched_ext/for-7.1-fixes] sched_ext: Fix deadlock between scx_root_disable() and concurrent forks

[Andrea Righi <arighi@nvidia.com>]

Hi Tejun,

On Sat, May 16, 2026 at 02:41:20PM -1000, Tejun Heo wrote:
> scx_root_disable() enters SCX_DISABLING before it grabs scx_enable_mutex to
> clear [__]scx_switching_all. task_should_scx() short-circuits on DISABLING,
> so forks in that window land on fair while next_active_class() still skips
> fair - the new tasks stall.
> 
> This can deadlock the disable path itself: scx_alloc_and_add_sched() runs
> under scx_enable_mutex and creates a helper kthread; if that new kthread is
> one of the stalled fair tasks, the mutex holder waits forever and
> scx_root_disable() can never make progress. Only sub-sched support exposes
> this, since sub-sched enables are the only path where
> scx_alloc_and_add_sched() can race the root's disable.
> 
> Move the DISABLING check after @scx_switching_all so that whenever
> @scx_switching_all is set, forks keep going to scx and stay in lockstep with
> __scx_switched_all. Once both are cleared (together under the mutex),
> DISABLING applies normally.
> 
> Fixes: 337ec00b1d9c ("sched_ext: Implement cgroup sub-sched enabling and disabling")
> Signed-off-by: Tejun Heo <tj@kernel.org>
> ---
>  kernel/sched/ext.c |   23 ++++++++++++++++++++++-
>  1 file changed, 22 insertions(+), 1 deletion(-)
> 
> --- a/kernel/sched/ext.c
> +++ b/kernel/sched/ext.c
> @@ -5092,10 +5092,31 @@ static const struct kset_uevent_ops scx_
>   */
>  bool task_should_scx(int policy)
>  {
> -	if (!scx_enabled() || unlikely(scx_enable_state() == SCX_DISABLING))
> +	/* if disabled, nothing should be on it */
> +	if (!scx_enabled())
>  		return false;
> +
> +	/* scx is taking over all SCHED_OTHER and SCHED_EXT tasks */
>  	if (READ_ONCE(scx_switching_all))
>  		return true;
> +
> +	/*
> +	 * scx is tearing down - keep new SCHED_EXT tasks out.
> +	 *
> +	 * Must come after scx_switching_all test. While both are set, we must
> +	 * return true via the branch above: [__]scx_switching_all are cleared
> +	 * together under scx_enable_mutex, and a fork routed to fair while
> +	 * __scx_switched_all is still on would stall because
> +	 * next_active_class() skips fair.

Just being extra picky: [__]scx_switching_all are cleared together sequentially,
but not atomically (in fact the order is what matters). To make it more clear,
how about rephrasing the comment block above like this:

  * Must come after the scx_switching_all test. scx_root_disable()
  * clears __scx_switched_all before scx_switching_all (both under
  * scx_enable_mutex), so while scx_switching_all is observed as true,
  * __scx_switched_all may still be on. A fork routed to fair in that
  * window would stall because next_active_class() skips fair.

> +	 *
> +	 * This can develop into a deadlock - scx holds scx_enable_mutex across
> +	 * kthread_create() in scx_alloc_and_add_sched(); if the new kthread is
> +	 * the stalled task, the disable path can never grab the mutex to clear
> +	 * scx_switching_all.
> +	 */
> +	if (unlikely(scx_enable_state() == SCX_DISABLING))
> +		return false;
> +
>  	return policy == SCHED_EXT;
>  }
> 

Other than that, looks good to me.

Reviewed-by: Andrea Righi <arighi@nvidia.com>

Thanks,
-Andrea