On Tue, Apr 07, 2026 at 08:39:04AM -0400, Andrew Stellman wrote:
> vring_transport_features() whitelists known transport feature bits and
> clears the rest via __virtio_clear_bit(). VIRTIO_F_RING_RESET is
> missing from the whitelist, so it is unconditionally cleared during
> feature negotiation. Drivers that depend on ring reset capability
> silently lose the feature.
Hmm was this observed in practice or just from code analysis?
And on which transport?
Because
static void vp_transport_features(struct virtio_device *vdev, u64 features)
{
struct virtio_pci_device *vp_dev = to_vp_device(vdev);
struct pci_dev *pci_dev = vp_dev->pci_dev;
if ((features & BIT_ULL(VIRTIO_F_SR_IOV)) &&
pci_find_ext_capability(pci_dev, PCI_EXT_CAP_ID_SRIOV))
__virtio_set_bit(vdev, VIRTIO_F_SR_IOV);
if (features & BIT_ULL(VIRTIO_F_RING_RESET))
__virtio_set_bit(vdev, VIRTIO_F_RING_RESET);
...
}
> Add VIRTIO_F_RING_RESET to the switch statement, matching the other
> transport-level features.
>
> Signed-off-by: Andrew Stellman <[email protected]>
> ---
> drivers/virtio/virtio_ring.c | 2 ++
> 1 file changed, 2 insertions(+)
>
> diff --git a/drivers/virtio/virtio_ring.c b/drivers/virtio/virtio_ring.c
> index fbca7ce..2cb643f 100644
> --- a/drivers/virtio/virtio_ring.c
> +++ b/drivers/virtio/virtio_ring.c
> @@ -3524,6 +3524,8 @@ void vring_transport_features(struct virtio_device
> *vdev)
> break;
> case VIRTIO_F_IN_ORDER:
> break;
> + case VIRTIO_F_RING_RESET:
> + break;
> default:
> /* We don't understand this bit. */
> __virtio_clear_bit(vdev, i);
> --
> 2.34.1
