在 2021/8/24 下午8:59, Eugenio Perez Martin 写道:
@@ -1378,11 +1508,19 @@ static int mlx5_vdpa_set_vq_address(struct vdpa_device *vdev, u16 idx, u64 desc_
{
struct mlx5_vdpa_dev *mvdev = to_mvdev(vdev);
struct mlx5_vdpa_net *ndev = to_mlx5_vdpa_ndev(mvdev);
- struct mlx5_vdpa_virtqueue *mvq = &ndev->vqs[idx];
+ struct mlx5_vdpa_virtqueue *mvq;
if (!is_index_valid(mvdev, idx))
return -EINVAL;
+ if (is_ctrl_vq_idx(mvdev, idx)) {
+ mvdev->cvq.desc_addr = desc_area;
+ mvdev->cvq.device_addr = device_area;
+ mvdev->cvq.driver_addr = driver_area;
Hi Eli,
Some mechanism for updating vringh addresses is also needed here.
setup_virtqueues is called a lot earlier than cvq address set, and
when cvq is kicked it tries to read from an outdated/invalid address.
Hi Eugenio:
Usually the driver should set the cvq address before set DRIVER_OK.
And during DRIVER_OK the mlx5 driver will call setup_virtqueues().
Did you see the cvq is set after DRIVER_OK?
Thanks
In my environment I just set mvdev->cvq.vring.vring.{desc,avail,used}
as a workaround and then I have cvq, but I think more protection is
needed because of the workqueue.
Sorry for the dripping of comments, it took a while to know what was
failing in the environment.
Thanks!
_______________________________________________
Virtualization mailing list
Virtualization@xxxxxxxxxxxxxxxxxxxxxxxxxx
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
