Re: [PATCH] sched_ext: Fix inconsistent NUMA node lookup in scx_select_cpu_dfl()

[Andrea Righi <arighi@nvidia.com>]

Hi Cheng-Yang,

On Sat, Mar 21, 2026 at 06:54:58PM +0800, Cheng-Yang Chou wrote:
> In the WAKE_SYNC path of scx_select_cpu_dfl(), waker_node was computed
> with cpu_to_node(), while node (for prev_cpu) was computed with
> scx_cpu_node_if_enabled(). When scx_builtin_idle_per_node is disabled,
> node is NUMA_NO_NODE but waker_node would be the actual NUMA node,
> causing two issues:
> 
> 1. The (waker_node == node) check always fails when SCX_PICK_IDLE_IN_NODE
>    is set, preventing the waker CPU optimization from ever triggering.

When scx_builtin_idle_per_node is disabled, SCX_PICK_IDLE_IN_NODE won't be
set, which means !(flags & SCX_PICK_IDLE_IN_NODE) should be always true,
short-circuiting the ||, and the waker_node == node comparison is never
evaluated.

However, ...

> 2. idle_cpumask(waker_node) is called with a real node ID even though
>    per-node idle tracking is disabled, resulting in undefined behavior.

...this looks like a legit bug. I'm wondering how this fix impacts
performance, will do some testing. Nice catch!

> 
> Fix by using scx_cpu_node_if_enabled() for waker_node as well, ensuring
> both variables are computed consistently.
> 
> Fixes: 48849271e6611 ("sched_ext: idle: Per-node idle cpumasks")
> Signed-off-by: Cheng-Yang Chou <yphbchou0911@gmail.com>

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

We should also add:
Cc: stable@vger.kernel.org # v6.15+

Thanks!
-Andrea

> ---
>  kernel/sched/ext_idle.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/kernel/sched/ext_idle.c b/kernel/sched/ext_idle.c
> index c7e405262697..8436c7df0a56 100644
> --- a/kernel/sched/ext_idle.c
> +++ b/kernel/sched/ext_idle.c
> @@ -543,7 +543,7 @@ s32 scx_select_cpu_dfl(struct task_struct *p, s32 prev_cpu, u64 wake_flags,
>  		 * piled up on it even if there is an idle core elsewhere on
>  		 * the system.
>  		 */
> -		waker_node = cpu_to_node(cpu);
> +		waker_node = scx_cpu_node_if_enabled(cpu);
>  		if (!(current->flags & PF_EXITING) &&
>  		    cpu_rq(cpu)->scx.local_dsq.nr == 0 &&
>  		    (!(flags & SCX_PICK_IDLE_IN_NODE) || (waker_node == node)) &&
> -- 
> 2.48.1
>