drivers/vdpa/mlx5/net/mlx5_vnet.c | 3 ++- drivers/vdpa/vdpa_sim/vdpa_sim.c | 3 ++- drivers/vhost/vdpa.c | 13 +++++++++++-- include/uapi/linux/vhost_types.h | 4 ++++ 4 files changed, 19 insertions(+), 4 deletions(-)
The commit in the fixes tag blocked VHOST_VDPA_SET_GROUP_ASID operations
once DRIVER_OK is set. That is too strict for devices which can safely
handle this during live migration flows.
Bring back this behavior under a new vhost backend feature flag. The
feature is supported by mlx5 and vdpa_sim devices.
Fixes: 3543b04a4ea3 ("vhost: forbid change vq groups ASID if DRIVER_OK is set")
Signed-off-by: Dragos Tatulea <dtatulea@nvidia.com>
Reviewed-by: Shahar Shitrit <shshitrit@nvidia.com>
---
drivers/vdpa/mlx5/net/mlx5_vnet.c | 3 ++-
drivers/vdpa/vdpa_sim/vdpa_sim.c | 3 ++-
drivers/vhost/vdpa.c | 13 +++++++++++--
include/uapi/linux/vhost_types.h | 4 ++++
4 files changed, 19 insertions(+), 4 deletions(-)
diff --git a/drivers/vdpa/mlx5/net/mlx5_vnet.c b/drivers/vdpa/mlx5/net/mlx5_vnet.c
index ad0d5fbbbca8..f89177957c76 100644
--- a/drivers/vdpa/mlx5/net/mlx5_vnet.c
+++ b/drivers/vdpa/mlx5/net/mlx5_vnet.c
@@ -2906,7 +2906,8 @@ static void unregister_link_notifier(struct mlx5_vdpa_net *ndev)
static u64 mlx5_vdpa_get_backend_features(const struct vdpa_device *vdpa)
{
- return BIT_ULL(VHOST_BACKEND_F_ENABLE_AFTER_DRIVER_OK);
+ return BIT_ULL(VHOST_BACKEND_F_ENABLE_AFTER_DRIVER_OK) |
+ BIT_ULL(VHOST_BACKEND_F_GROUP_ASID_AFTER_DRIVER_OK);
}
static int mlx5_vdpa_set_driver_features(struct vdpa_device *vdev, u64 features)
diff --git a/drivers/vdpa/vdpa_sim/vdpa_sim.c b/drivers/vdpa/vdpa_sim/vdpa_sim.c
index 8cb1cc2ea139..253c7fb35ea0 100644
--- a/drivers/vdpa/vdpa_sim/vdpa_sim.c
+++ b/drivers/vdpa/vdpa_sim/vdpa_sim.c
@@ -428,7 +428,8 @@ static u64 vdpasim_get_device_features(struct vdpa_device *vdpa)
static u64 vdpasim_get_backend_features(const struct vdpa_device *vdpa)
{
- return BIT_ULL(VHOST_BACKEND_F_ENABLE_AFTER_DRIVER_OK);
+ return BIT_ULL(VHOST_BACKEND_F_ENABLE_AFTER_DRIVER_OK) |
+ BIT_ULL(VHOST_BACKEND_F_GROUP_ASID_AFTER_DRIVER_OK);
}
static int vdpasim_set_driver_features(struct vdpa_device *vdpa, u64 features)
diff --git a/drivers/vhost/vdpa.c b/drivers/vhost/vdpa.c
index 692564b1bcbb..67b3f49fa709 100644
--- a/drivers/vhost/vdpa.c
+++ b/drivers/vhost/vdpa.c
@@ -682,7 +682,8 @@ static long vhost_vdpa_vring_ioctl(struct vhost_vdpa *v, unsigned int cmd,
return -EFAULT;
if (idx >= vdpa->ngroups || s.num >= vdpa->nas)
return -EINVAL;
- if (ops->get_status(vdpa) & VIRTIO_CONFIG_S_DRIVER_OK)
+ if ((ops->get_status(vdpa) & VIRTIO_CONFIG_S_DRIVER_OK) &&
+ !vhost_backend_has_feature(vq, VHOST_BACKEND_F_GROUP_ASID_AFTER_DRIVER_OK))
return -EBUSY;
if (!ops->set_group_asid)
return -EOPNOTSUPP;
@@ -791,7 +792,8 @@ static long vhost_vdpa_unlocked_ioctl(struct file *filep,
BIT_ULL(VHOST_BACKEND_F_IOTLB_PERSIST) |
BIT_ULL(VHOST_BACKEND_F_SUSPEND) |
BIT_ULL(VHOST_BACKEND_F_RESUME) |
- BIT_ULL(VHOST_BACKEND_F_ENABLE_AFTER_DRIVER_OK)))
+ BIT_ULL(VHOST_BACKEND_F_ENABLE_AFTER_DRIVER_OK) |
+ BIT_ULL(VHOST_BACKEND_F_GROUP_ASID_AFTER_DRIVER_OK)))
return -EOPNOTSUPP;
if ((features & BIT_ULL(VHOST_BACKEND_F_SUSPEND)) &&
!vhost_vdpa_can_suspend(v))
@@ -805,6 +807,13 @@ static long vhost_vdpa_unlocked_ioctl(struct file *filep,
if ((features & BIT_ULL(VHOST_BACKEND_F_DESC_ASID)) &&
!vhost_vdpa_has_desc_group(v))
return -EOPNOTSUPP;
+ if (features & BIT_ULL(VHOST_BACKEND_F_GROUP_ASID_AFTER_DRIVER_OK)) {
+ if (!(features & BIT_ULL(VHOST_BACKEND_F_IOTLB_ASID)))
+ return -EINVAL;
+ if (!(vhost_vdpa_get_backend_features(v) &
+ BIT_ULL(VHOST_BACKEND_F_GROUP_ASID_AFTER_DRIVER_OK)))
+ return -EOPNOTSUPP;
+ }
if ((features & BIT_ULL(VHOST_BACKEND_F_IOTLB_PERSIST)) &&
!vhost_vdpa_has_persistent_map(v))
return -EOPNOTSUPP;
diff --git a/include/uapi/linux/vhost_types.h b/include/uapi/linux/vhost_types.h
index 1c39cc5f5a31..ec1ff8a2e260 100644
--- a/include/uapi/linux/vhost_types.h
+++ b/include/uapi/linux/vhost_types.h
@@ -197,5 +197,9 @@ struct vhost_vdpa_iova_range {
#define VHOST_BACKEND_F_DESC_ASID 0x7
/* IOTLB don't flush memory mapping across device reset */
#define VHOST_BACKEND_F_IOTLB_PERSIST 0x8
+/* Device supports changing the group ASID after DRIVER_OK.
+ * Requires VHOST_BACKEND_F_IOTLB_ASID.
+ */
+#define VHOST_BACKEND_F_GROUP_ASID_AFTER_DRIVER_OK 0x9
#endif
--
2.54.0On Mon, May 11, 2026 at 11:46 AM Dragos Tatulea <dtatulea@nvidia.com> wrote:
>
> The commit in the fixes tag blocked VHOST_VDPA_SET_GROUP_ASID operations
> once DRIVER_OK is set. That is too strict for devices which can safely
> handle this during live migration flows.
>
> Bring back this behavior under a new vhost backend feature flag. The
> feature is supported by mlx5 and vdpa_sim devices.
>
> Fixes: 3543b04a4ea3 ("vhost: forbid change vq groups ASID if DRIVER_OK is set")
> Signed-off-by: Dragos Tatulea <dtatulea@nvidia.com>
> Reviewed-by: Shahar Shitrit <shshitrit@nvidia.com>
Acked-by: Eugenio Pérez <eperezma@redhat.com>
> ---
> drivers/vdpa/mlx5/net/mlx5_vnet.c | 3 ++-
> drivers/vdpa/vdpa_sim/vdpa_sim.c | 3 ++-
> drivers/vhost/vdpa.c | 13 +++++++++++--
> include/uapi/linux/vhost_types.h | 4 ++++
> 4 files changed, 19 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/vdpa/mlx5/net/mlx5_vnet.c b/drivers/vdpa/mlx5/net/mlx5_vnet.c
> index ad0d5fbbbca8..f89177957c76 100644
> --- a/drivers/vdpa/mlx5/net/mlx5_vnet.c
> +++ b/drivers/vdpa/mlx5/net/mlx5_vnet.c
> @@ -2906,7 +2906,8 @@ static void unregister_link_notifier(struct mlx5_vdpa_net *ndev)
>
> static u64 mlx5_vdpa_get_backend_features(const struct vdpa_device *vdpa)
> {
> - return BIT_ULL(VHOST_BACKEND_F_ENABLE_AFTER_DRIVER_OK);
> + return BIT_ULL(VHOST_BACKEND_F_ENABLE_AFTER_DRIVER_OK) |
> + BIT_ULL(VHOST_BACKEND_F_GROUP_ASID_AFTER_DRIVER_OK);
> }
>
> static int mlx5_vdpa_set_driver_features(struct vdpa_device *vdev, u64 features)
> diff --git a/drivers/vdpa/vdpa_sim/vdpa_sim.c b/drivers/vdpa/vdpa_sim/vdpa_sim.c
> index 8cb1cc2ea139..253c7fb35ea0 100644
> --- a/drivers/vdpa/vdpa_sim/vdpa_sim.c
> +++ b/drivers/vdpa/vdpa_sim/vdpa_sim.c
> @@ -428,7 +428,8 @@ static u64 vdpasim_get_device_features(struct vdpa_device *vdpa)
>
> static u64 vdpasim_get_backend_features(const struct vdpa_device *vdpa)
> {
> - return BIT_ULL(VHOST_BACKEND_F_ENABLE_AFTER_DRIVER_OK);
> + return BIT_ULL(VHOST_BACKEND_F_ENABLE_AFTER_DRIVER_OK) |
> + BIT_ULL(VHOST_BACKEND_F_GROUP_ASID_AFTER_DRIVER_OK);
> }
>
> static int vdpasim_set_driver_features(struct vdpa_device *vdpa, u64 features)
> diff --git a/drivers/vhost/vdpa.c b/drivers/vhost/vdpa.c
> index 692564b1bcbb..67b3f49fa709 100644
> --- a/drivers/vhost/vdpa.c
> +++ b/drivers/vhost/vdpa.c
> @@ -682,7 +682,8 @@ static long vhost_vdpa_vring_ioctl(struct vhost_vdpa *v, unsigned int cmd,
> return -EFAULT;
> if (idx >= vdpa->ngroups || s.num >= vdpa->nas)
> return -EINVAL;
> - if (ops->get_status(vdpa) & VIRTIO_CONFIG_S_DRIVER_OK)
> + if ((ops->get_status(vdpa) & VIRTIO_CONFIG_S_DRIVER_OK) &&
> + !vhost_backend_has_feature(vq, VHOST_BACKEND_F_GROUP_ASID_AFTER_DRIVER_OK))
> return -EBUSY;
> if (!ops->set_group_asid)
> return -EOPNOTSUPP;
> @@ -791,7 +792,8 @@ static long vhost_vdpa_unlocked_ioctl(struct file *filep,
> BIT_ULL(VHOST_BACKEND_F_IOTLB_PERSIST) |
> BIT_ULL(VHOST_BACKEND_F_SUSPEND) |
> BIT_ULL(VHOST_BACKEND_F_RESUME) |
> - BIT_ULL(VHOST_BACKEND_F_ENABLE_AFTER_DRIVER_OK)))
> + BIT_ULL(VHOST_BACKEND_F_ENABLE_AFTER_DRIVER_OK) |
> + BIT_ULL(VHOST_BACKEND_F_GROUP_ASID_AFTER_DRIVER_OK)))
> return -EOPNOTSUPP;
> if ((features & BIT_ULL(VHOST_BACKEND_F_SUSPEND)) &&
> !vhost_vdpa_can_suspend(v))
> @@ -805,6 +807,13 @@ static long vhost_vdpa_unlocked_ioctl(struct file *filep,
> if ((features & BIT_ULL(VHOST_BACKEND_F_DESC_ASID)) &&
> !vhost_vdpa_has_desc_group(v))
> return -EOPNOTSUPP;
Ouch to me here. By reading the errno manual:
Nit: By reading errno(3):
ENOTSUP - Operation not supported (POSIX.1-2001).
EOPNOTSUPP - Operation not supported on socket (POSIX.1-2001).
I picked the wrong constant even if they share the same errno value
(by the same page of the manual). MST, is it worth changing it?
> + if (features & BIT_ULL(VHOST_BACKEND_F_GROUP_ASID_AFTER_DRIVER_OK)) {
> + if (!(features & BIT_ULL(VHOST_BACKEND_F_IOTLB_ASID)))
> + return -EINVAL;
> + if (!(vhost_vdpa_get_backend_features(v) &
> + BIT_ULL(VHOST_BACKEND_F_GROUP_ASID_AFTER_DRIVER_OK)))
> + return -EOPNOTSUPP;
> + }
> if ((features & BIT_ULL(VHOST_BACKEND_F_IOTLB_PERSIST)) &&
> !vhost_vdpa_has_persistent_map(v))
> return -EOPNOTSUPP;
> diff --git a/include/uapi/linux/vhost_types.h b/include/uapi/linux/vhost_types.h
> index 1c39cc5f5a31..ec1ff8a2e260 100644
> --- a/include/uapi/linux/vhost_types.h
> +++ b/include/uapi/linux/vhost_types.h
> @@ -197,5 +197,9 @@ struct vhost_vdpa_iova_range {
> #define VHOST_BACKEND_F_DESC_ASID 0x7
> /* IOTLB don't flush memory mapping across device reset */
> #define VHOST_BACKEND_F_IOTLB_PERSIST 0x8
> +/* Device supports changing the group ASID after DRIVER_OK.
> + * Requires VHOST_BACKEND_F_IOTLB_ASID.
> + */
> +#define VHOST_BACKEND_F_GROUP_ASID_AFTER_DRIVER_OK 0x9
>
> #endif
> --
> 2.54.0
>
On Mon, May 11, 2026 at 12:45:03PM +0300, Dragos Tatulea wrote:
> The commit in the fixes tag blocked VHOST_VDPA_SET_GROUP_ASID operations
> once DRIVER_OK is set. That is too strict for devices which can safely
> handle this during live migration flows.
>
> Bring back this behavior under a new vhost backend feature flag. The
> feature is supported by mlx5 and vdpa_sim devices.
>
> Fixes: 3543b04a4ea3 ("vhost: forbid change vq groups ASID if DRIVER_OK is set")
> Signed-off-by: Dragos Tatulea <dtatulea@nvidia.com>
> Reviewed-by: Shahar Shitrit <shshitrit@nvidia.com>
> ---
Gentle ping.
Thanks,
Dragos
© 2016 - 2026 Red Hat, Inc.