On Thu, Sep 11, 2025 at 09:56:41PM +0530, Kaushlendra Kumar wrote:
> Use WRITE_ONCE() for rclp->len decrement in rcu_cblist_dequeue() to
> maintain consistency with rcu_cblist_enqueue(), which already uses
> WRITE_ONCE() for rclp->len increment. It maintains consistent access
> patterns to the rclp->len field across both enqueue and dequeue operations.
> 
> Signed-off-by: Kaushlendra Kumar <kaushlendra.ku...@intel.com>

Again, why is this patch needed?

Please keep in mind that unmarked normal C-language loads and stores have
value in that KCSAN can detect data races involving them.  If we decorate
all loads and stores with READ_ONCE() and WRITE_ONCE(), respectively,
KCSAN is unable to detect buggy concurrent accesses.

                                                        Thanx, Paul

> ---
> Changes in v2:
> - Revised justification to focus on code consistency rather than
>   concurrent access claims.
> - Revised commit message to use imperative mood throughout per kernel
>   documentation guidelines (suggested by Markus Elfring).
> 
>  kernel/rcu/rcu_segcblist.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/kernel/rcu/rcu_segcblist.c b/kernel/rcu/rcu_segcblist.c
> index 1693ea22ef1b..e10b36e9de54 100644
> --- a/kernel/rcu/rcu_segcblist.c
> +++ b/kernel/rcu/rcu_segcblist.c
> @@ -71,7 +71,7 @@ struct rcu_head *rcu_cblist_dequeue(struct rcu_cblist *rclp)
>       rhp = rclp->head;
>       if (!rhp)
>               return NULL;
> -     rclp->len--;
> +     WRITE_ONCE(rclp->len, rclp->len - 1);
>       rclp->head = rhp->next;
>       if (!rclp->head)
>               rclp->tail = &rclp->head;
> -- 
> 2.34.1
> 

Reply via email to