Re: [PATCH v2] vdpa/mlx5: Verify wq is a valid pointer in mlx5_vdpa_suspend

["Michael S. Tsirkin" <mst@redhat.com>]

On Tue, Mar 28, 2023 at 10:18:10AM +0300, Eli Cohen wrote:
> mlx5_vdpa_suspend() flushes the workqueue as part of its logic. However,
> if the device has been deleted while a VM was running, the workqueue
> will be destroyed first and wq will become null. After the VM is destroyed,
> suspend can be called and will access a null pointer.
> 
> Fix it by verifying wq is not NULL.
> 
> Fixes: cae15c2ed8e6 ("vdpa/mlx5: Implement susupend virtqueue callback")
> Signed-off-by: Eli Cohen <elic@nvidia.com>


This conflicts with the patch for not losing link state updates.
How do you want to handle this?

> ---
> v1 -> v2:
> Fix spelling errors
> 
>  drivers/vdpa/mlx5/net/mlx5_vnet.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/vdpa/mlx5/net/mlx5_vnet.c b/drivers/vdpa/mlx5/net/mlx5_vnet.c
> index 85866ace0061..b73c5943aefd 100644
> --- a/drivers/vdpa/mlx5/net/mlx5_vnet.c
> +++ b/drivers/vdpa/mlx5/net/mlx5_vnet.c
> @@ -2929,7 +2929,8 @@ static int mlx5_vdpa_suspend(struct vdpa_device *vdev)
>  	down_write(&ndev->reslock);
>  	ndev->nb_registered = false;
>  	mlx5_notifier_unregister(mvdev->mdev, &ndev->nb);
> -	flush_workqueue(ndev->mvdev.wq);
> +	if (ndev->mvdev.wq)
> +		flush_workqueue(ndev->mvdev.wq);
>  	for (i = 0; i < ndev->cur_num_vqs; i++) {
>  		mvq = &ndev->vqs[i];
>  		suspend_vq(ndev, mvq);
> -- 
> 2.38.1

_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization