Re: [PATCH] afs: Add missing memory barriers in afs_manage_cell()

[Andrea Parri]

On Mon, Nov 26, 2018 at 05:44:12PM +0100, Andrea Parri wrote:
> As the comments for wake_up_bit() and waitqueue_active() point out,
> the barriers are needed to order the clearing of the _FL_NOT_READY
> bit and the waitqueue_active() load; match the implicit barrier in
> prepare_to_wait().
> 
> Signed-off-by: Andrea Parri <andrea.parri@xxxxxxxxxxxxxxxxxxxx>

Hi David, do you have any feedback on this?

  Andrea


> ---
>  fs/afs/cell.c | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/fs/afs/cell.c b/fs/afs/cell.c
> index cf445dbd5f2e0..a9488abe56029 100644
> --- a/fs/afs/cell.c
> +++ b/fs/afs/cell.c
> @@ -637,6 +637,7 @@ static void afs_manage_cell(struct work_struct *work)
>  		cell->state = AFS_CELL_ACTIVE;
>  		smp_wmb();
>  		clear_bit(AFS_CELL_FL_NOT_READY, &cell->flags);
> +		smp_mb__after_atomic(); /* see comment for wake_up_bit() */
>  		wake_up_bit(&cell->flags, AFS_CELL_FL_NOT_READY);
>  		goto again;
>  
> @@ -678,6 +679,7 @@ static void afs_manage_cell(struct work_struct *work)
>  	cell->state = AFS_CELL_ACTIVE;
>  	smp_wmb();
>  	clear_bit(AFS_CELL_FL_NOT_READY, &cell->flags);
> +	smp_mb__after_atomic(); /* see comment for wake_up_bit() */
>  	wake_up_bit(&cell->flags, AFS_CELL_FL_NOT_READY);
>  	_leave(" [deact->act]");
>  	return;
> -- 
> 2.17.1
>