On 01/30, Konstantin Shkolnyy wrote: > mlx5_vdpa_dev_add() doesn’t initialize mvdev->actual_features. It’s > initialized later by mlx5_vdpa_set_driver_features(). However, > mlx5_vdpa_get_config() depends on the VIRTIO_F_VERSION_1 flag in > actual_features to return data with correct endianness. When it’s called > before mlx5_vdpa_set_driver_features(), the flag is clear, and the data are > returned as big-endian on big-endian machines, while QEMU interprets them > as little-endian. > Good catch! Thank you for the patch.
Could you add the clarification that mlx5_vdpa_dev_add() uses the endianess helpers (cpu_to_mlx5vdpa16()) which depend on actual_flags being set properly. > Signed-off-by: Konstantin Shkolnyy <k...@linux.ibm.com> > --- > drivers/vdpa/mlx5/net/mlx5_vnet.c | 1 + > 1 file changed, 1 insertion(+) > > diff --git a/drivers/vdpa/mlx5/net/mlx5_vnet.c > b/drivers/vdpa/mlx5/net/mlx5_vnet.c > index 36099047560d..f9ea240b8145 100644 > --- a/drivers/vdpa/mlx5/net/mlx5_vnet.c > +++ b/drivers/vdpa/mlx5/net/mlx5_vnet.c > @@ -3884,6 +3884,7 @@ static int mlx5_vdpa_dev_add(struct vdpa_mgmt_dev > *v_mdev, const char *name, > ndev->mvdev.max_vqs = max_vqs; > mvdev = &ndev->mvdev; > mvdev->mdev = mdev; > + mvdev->actual_features = device_features; > Maybe it is better to be conservative: only pick the VIRTIO_F_VERSION_1 flag out of device_features. Thanks, Dragos
