Re: [PATCH 4/7] drm/sched: Remove the single scheduler special case

[Sidong Yang <sidong.yang@furiosa.ai>]

On Fri, Nov 06, 2026 at 02:56:59PM +0000, Tvrtko Ursulin wrote:
> @@ -139,7 +141,10 @@ int drm_sched_entity_init(struct drm_sched_entity *entity,
>  	entity->guilty = guilty;
>  	entity->last_user = current->group_leader;
>  	entity->num_sched_list = num_sched_list;
> -	entity->sched_list = num_sched_list > 1 ? sched_list : NULL;
> +	if (num_sched_list == 1)
> +		entity->sched = sched_list[0];
> +	else
> +		entity->sched_list = sched_list;

This breaks rocket. rocket_job_open() passes a kmalloc'ed array of
rdev->num_cores schedulers, and rocket_job_close() does
kfree(entity->sched_list). With only one core bound when the file is
opened (e.g. the other cores probe later), the union now holds
&rdev->cores[0].sched, so close kfree()s a pointer into the devm
allocation. Before this patch it was kfree(NULL) (and the array leaked).

I can send a rocket fix that keeps its own pointer to the array and
frees that instead, so it can go in before this patch, unless you'd
rather carry it in this series.

Thanks,
Sidong