Re: [PATCH] sched: Clarify wake_up_q()'s write to task->wake_q.next

[Peter Zijlstra]

On Wed, Jan 29, 2025 at 08:53:03PM +0100, Jann Horn wrote:
> Clarify that wake_up_q() does an atomic write to task->wake_q.next, after
> which a concurrent __wake_q_add() can immediately overwrite
> task->wake_q.next again.
> 
> Signed-off-by: Jann Horn <jannh@xxxxxxxxxx>
> ---
> a minor change; I figured we should at least be using WRITE_ONCE() here,
> and I might as well change the comments a little bit while I'm touching
> this
> ---
>  kernel/sched/core.c | 5 +++--
>  1 file changed, 3 insertions(+), 2 deletions(-)
> 
> diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> index 9142a0394d46605e96e10cef97cce02d741f6c93..ce64652858703826dca510479f563a28c2fb2405 100644
> --- a/kernel/sched/core.c
> +++ b/kernel/sched/core.c
> @@ -1063,9 +1063,10 @@ void wake_up_q(struct wake_q_head *head)
>  		struct task_struct *task;
>  
>  		task = container_of(node, struct task_struct, wake_q);
> -		/* Task can safely be re-inserted now: */
>  		node = node->next;
> -		task->wake_q.next = NULL;
> +		/* pairs with cmpxchg_relaxed() in __wake_q_add() */
> +		WRITE_ONCE(task->wake_q.next, NULL);
> +		/* Task can safely be re-inserted now. */

Right, so even if the store is shattered, the cmpxchg won't proceed
until all bits land -- eg. NULL becomes 'complete'.

That said, your patch makes sense, so let me go stick that in a queue
somewhere.

Thanks!