Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <1601583511-15138-1-git-send-email-si-wei.liu@oracle.com>
In-Reply-To: <1601583511-15138-1-git-send-email-si-wei.liu@oracle.com>
From:   Si-Wei Liu <siwliu.kernel@gmail.com>
Date:   Fri, 2 Oct 2020 13:17:00 -0700
Message-ID: <CAPWQSg1y8uvpiwxxp_ONGFs8GeuOY09q3AShfLCmhv77ePma-Q@mail.gmail.com>
Subject: Re: [PATCH] vdpa/mlx5: should keep avail_index despite device status
To:     elic@nvidia.com, mst@redhat.com, jasowang@redhat.com,
        netdev@vger.kernel.org
Cc:     joao.m.martins@oracle.com, boris.ostrovsky@oracle.com,
        linux-kernel@vger.kernel.org,
        virtualization@lists.linux-foundation.org,
        Si-Wei Liu <si-wei.liu@oracle.com>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

+ Eli.

On Thu, Oct 1, 2020 at 2:02 PM Si-Wei Liu <si-wei.liu@oracle.com> wrote:
>
> A VM with mlx5 vDPA has below warnings while being reset:
>
> vhost VQ 0 ring restore failed: -1: Resource temporarily unavailable (11)
> vhost VQ 1 ring restore failed: -1: Resource temporarily unavailable (11)
>
> We should allow userspace emulating the virtio device be
> able to get to vq's avail_index, regardless of vDPA device
> status. Save the index that was last seen when virtq was
> stopped, so that userspace doesn't complain.
>
> Signed-off-by: Si-Wei Liu <si-wei.liu@oracle.com>
> ---
>  drivers/vdpa/mlx5/net/mlx5_vnet.c | 20 ++++++++++++++------
>  1 file changed, 14 insertions(+), 6 deletions(-)
>
> diff --git a/drivers/vdpa/mlx5/net/mlx5_vnet.c b/drivers/vdpa/mlx5/net/mlx5_vnet.c
> index 70676a6..74264e59 100644
> --- a/drivers/vdpa/mlx5/net/mlx5_vnet.c
> +++ b/drivers/vdpa/mlx5/net/mlx5_vnet.c
> @@ -1133,15 +1133,17 @@ static void suspend_vq(struct mlx5_vdpa_net *ndev, struct mlx5_vdpa_virtqueue *m
>         if (!mvq->initialized)
>                 return;
>
> -       if (query_virtqueue(ndev, mvq, &attr)) {
> -               mlx5_vdpa_warn(&ndev->mvdev, "failed to query virtqueue\n");
> -               return;
> -       }
>         if (mvq->fw_state != MLX5_VIRTIO_NET_Q_OBJECT_STATE_RDY)
>                 return;
>
>         if (modify_virtqueue(ndev, mvq, MLX5_VIRTIO_NET_Q_OBJECT_STATE_SUSPEND))
>                 mlx5_vdpa_warn(&ndev->mvdev, "modify to suspend failed\n");
> +
> +       if (query_virtqueue(ndev, mvq, &attr)) {
> +               mlx5_vdpa_warn(&ndev->mvdev, "failed to query virtqueue\n");
> +               return;
> +       }
> +       mvq->avail_idx = attr.available_index;
>  }
>
>  static void suspend_vqs(struct mlx5_vdpa_net *ndev)
> @@ -1411,8 +1413,14 @@ static int mlx5_vdpa_get_vq_state(struct vdpa_device *vdev, u16 idx, struct vdpa
>         struct mlx5_virtq_attr attr;
>         int err;
>
> -       if (!mvq->initialized)
> -               return -EAGAIN;
> +       /* If the virtq object was destroyed, use the value saved at
> +        * the last minute of suspend_vq. This caters for userspace
> +        * that cares about emulating the index after vq is stopped.
> +        */
> +       if (!mvq->initialized) {
> +               state->avail_index = mvq->avail_idx;
> +               return 0;
> +       }
>
>         err = query_virtqueue(ndev, mvq, &attr);
>         if (err) {
> --
> 1.8.3.1
>