Skip to contentSkip to conversation

vdpa/mlx5: Support different address spaces for control and data

bug report Dan Carpenter 2 participants 4 messages Aug 11, 2022 2 messages not in the archive virtio core

lore
Share & tools
lore
#0 lore raw

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.

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

regards,
dan carpenter
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

#2 lore raw
In reply to a message not in the archive
Quoted 40 lines · Eli Cohen
Quoted 33 lines

From: Dan Carpenter <[email protected]>
Sent: Thursday, August 11, 2022 1:40 PM
To: Eli Cohen <[email protected]>
Cc: [email protected]
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);
Quoted 3 lines

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.

Quoted 7 lines

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;
        }

Patch on top pls? Time's running short.

Quoted 9 lines
Quoted 9 lines
    2673
    2674 out:
    2675         up_write(&ndev->reslock);
--> 2676         return err;
    2677 }

regards,
dan carpenter

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

#3 lore raw
In reply to a message not in the archive
Quoted 40 lines · Eli Cohen
Quoted 33 lines

From: Dan Carpenter <[email protected]>
Sent: Thursday, August 11, 2022 1:40 PM
To: Eli Cohen <[email protected]>
Cc: [email protected]
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);
Quoted 3 lines

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.

Quoted 7 lines

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.

Quoted 9 lines
Quoted 9 lines
    2673
    2674 out:
    2675         up_write(&ndev->reslock);
--> 2676         return err;
    2677 }

regards,
dan carpenter

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

#5 lore raw
In reply to a message not in the archive
Quoted 66 lines · Eli Cohen
Quoted 66 lines

From: Michael S. Tsirkin <[email protected]>
Sent: Thursday, August 11, 2022 4:09 PM
To: Eli Cohen <[email protected]>
Cc: Dan Carpenter <[email protected]>; Jason Wang <[email protected]>; Eugenio Perez Martin <[email protected]>; [email protected]
Subject: Re: [bug report] vdpa/mlx5: Support different address spaces for control and data

Quoted 40 lines · Eli Cohen
Quoted 33 lines

From: Dan Carpenter <[email protected]>
Sent: Thursday, August 11, 2022 1:40 PM
To: Eli Cohen <[email protected]>
Cc: [email protected]
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);
Quoted 3 lines

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.

Quoted 7 lines

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;
}
Quoted 4 lines
Quoted 2 lines

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 10 lines
Quoted 10 lines
Quoted 10 lines
Quoted 10 lines

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

regards,
dan carpenter

_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization