Re: [PATCH v4] sched/psi: fix memory barrier without comment warnings

[Christophe JAILLET]

Le 06/10/2024 à 08:00, Pintu Kumar a écrit :
> These warnings were reported by checkpatch.
> Fix them with minor changes.
> No functional changes.
>
> WARNING: memory barrier without comment
> +       t = smp_load_acquire(trigger_ptr);
>
> WARNING: memory barrier without comment
> +       smp_store_release(&seq->private, new);
>
> Signed-off-by: Pintu Kumar <quic_pintu@xxxxxxxxxxx>
>
> ---
> Changes in V4:
> Added () in comment as well suggested by Christophe JAILLET.
> V3: https://lore.kernel.org/all/00aeb243-3d47-42be-b52c-08b39c5fef07@xxxxxxxxxx/
> Changes in V3:
> Removed signature of Joe as requested. No other change.
> V2: https://lore.kernel.org/all/CAOuPNLi1mUKW_vv0E6Ynzvdw_rHvCye+nAf2bWv6Qj9A8ofX1g@xxxxxxxxxxxxxx/
> Changes in V2:
> Retain printk_deferred warnings as suggested by Joe Perches.
> V1: https://lore.kernel.org/all/a848671f803ba2b4ab14b0f7b09f0f53a8dd1c4b.camel@xxxxxxxxxxx/
> ---
>   kernel/sched/psi.c | 2 ++
>   1 file changed, 2 insertions(+)
>
> diff --git a/kernel/sched/psi.c b/kernel/sched/psi.c
> index 020d58967d4e..175423716e4c 100644
> --- a/kernel/sched/psi.c
> +++ b/kernel/sched/psi.c
> @@ -1474,6 +1474,7 @@ __poll_t psi_trigger_poll(void **trigger_ptr,
>   	if (static_branch_likely(&psi_disabled))
>   		return DEFAULT_POLLMASK | EPOLLERR | EPOLLPRI;
>   
> +	/* Pairs with the smp_store_release() in psi_write */
>   	t = smp_load_acquire(trigger_ptr);
>   	if (!t)
>   		return DEFAULT_POLLMASK | EPOLLERR | EPOLLPRI;
> @@ -1557,6 +1558,7 @@ static ssize_t psi_write(struct file *file, const char __user *user_buf,
>   		return PTR_ERR(new);
>   	}
>   
> +	/* Pairs with the smp_store_acquire() in psi_trigger_poll */

Sorry if I was unclear in my previous comment, but my main point is that 
I think that it should be smp_load_acquire() and not smp_store_acquire().

(Also, if you add some (), you could also add them for psi_trigger_poll 
and psi_write)

CJ

>   	smp_store_release(&seq->private, new);
>   	mutex_unlock(&seq->lock);
>