Re: [bug report] vdpa/mlx5: Support different address spaces for control and data

On Thu, Aug 11, 2022 at 01:11:11PM +0000, Eli Cohen wrote:

quoted

From: Michael S. Tsirkin <mst@redhat.com>
Sent: Thursday, August 11, 2022 4:09 PM
To: Eli Cohen <redacted>
Cc: Dan Carpenter <redacted>; Jason Wang <redacted>; Eugenio Perez Martin
[off-list ref]; virtualization@lists.linux-foundation.org
Subject: Re: [bug report] vdpa/mlx5: Support different address spaces for control and data

On Thu, Aug 11, 2022 at 12:40:09PM +0000, Eli Cohen wrote:

quoted

quoted

From: Dan Carpenter <redacted>
Sent: Thursday, August 11, 2022 1:40 PM
To: Eli Cohen <redacted>
Cc: virtualization@lists.linux-foundation.org
Subject: [bug report] vdpa/mlx5: Support different address spaces for control and data

Hello Eli Cohen,

The patch d5358cd0e369: "vdpa/mlx5: Support different address spaces
for control and data" from Jul 14, 2022, leads to the following
Smatch static checker warning:

	drivers/vdpa/mlx5/net/mlx5_vnet.c:2676 mlx5_vdpa_set_map()
	error: uninitialized symbol 'err'.

drivers/vdpa/mlx5/net/mlx5_vnet.c
    2657 static int mlx5_vdpa_set_map(struct vdpa_device *vdev, unsigned int asid,
    2658                              struct vhost_iotlb *iotlb)
    2659 {
    2660         struct mlx5_vdpa_dev *mvdev = to_mvdev(vdev);
    2661         struct mlx5_vdpa_net *ndev = to_mlx5_vdpa_ndev(mvdev);
    2662         int err;
    2663
    2664         down_write(&ndev->reslock);
    2665         if (mvdev->group2asid[MLX5_VDPA_DATAVQ_GROUP] == asid) {
    2666                 err = set_map_data(mvdev, iotlb);
    2667                 if (err)
    2668                         goto out;
    2669         }
    2670
    2671         if (mvdev->group2asid[MLX5_VDPA_CVQ_GROUP] == asid)
    2672                 err = set_map_control(mvdev, iotlb);

err not initialized on else path.  My guess is that one or both of these
conditions has to be true and this is a false positive but I don't know
the code well enough to be sure.

Thanks for reporting this.
I think it would be better to return an error if the provided asid is not recognized.

Therefore I am thinking about adding something like this:

        if (asid != mvdev->group2asid[MLX5_VDPA_DATAVQ_GROUP] &&
            asid != mvdev->group2asid[MLX5_VDPA_CVQ_GROUP]) {
                err = -EINVAL;
                goto out;
        }

I would probably chain the conditions:

if (mvdev->group2asid[MLX5_VDPA_DATAVQ_GROUP] == asid) {
} else if (mvdev->group2asid[MLX5_VDPA_CVQ_GROUP] == asid) {
} else {
                 err = -EINVAL;
                 goto out;
}


or alternatively initialize err with -EINVAL and be done with it.

This makes more sense.
Will send a patch in an hour.

Just making sure, the only result without this patch is that
and iotlb might be created without a mapping.
But nothing terrible bad will happen things just won't work
but this userspace is already buggy.
Right?

quoted

quoted

quoted

    2673
    2674 out:
    2675         up_write(&ndev->reslock);
--> 2676         return err;
    2677 }

regards,
dan carpenter

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