On Wed, Nov 23, 2016 at 21:10 +0100, Stefan Fritsch wrote:
> On Wed, 23 Nov 2016, Rafael Zalamena wrote:
>
> > > Maybe something like this is enough already (untested):
> >
> > I tried your diff without Mike's if_vio diff and it doesn't panic anymore,
> > however it doesn't work.
> >
> > vioX can send packets to host, host receives them and reply, but vioX
> > doesn't see any packets back. I don't even need to touch the interface
> > up/down status to see this happening. Also when the interface comes
> > up after being shutdown it sends a bunch of packets to host.
>
> Sorry, device_status is a bitmask, not a plain value.
>
> Try the patch below. The first hunk is to fix the 'sends a bunch of
> packets'. If it causes any problems, leave it out.
>
> diff --git usr.sbin/vmd/virtio.c usr.sbin/vmd/virtio.c
> index 93def73..6436e6a 100644
> --- usr.sbin/vmd/virtio.c
> +++ usr.sbin/vmd/virtio.c
> @@ -703,6 +703,13 @@ virtio_net_io(int dir, uint16_t reg, uint32_t *data,
> uint8_t *intr,
> break;
> case VIRTIO_CONFIG_DEVICE_STATUS:
> dev->cfg.device_status = *data;
> + if (*data == 0) {
> + dev->vq[0].last_avail = 0;
> + dev->vq[0].notified_avail = 0;
> + dev->vq[1].last_avail = 0;
> + dev->vq[1].notified_avail = 0;
> + /* XXX do proper reset */
> + }
> break;
> default:
> break;
> @@ -796,6 +803,9 @@ vionet_enq_rx(struct vionet_dev *dev, char *pkt, ssize_t
> sz, int *spc)
>
> ret = 0;
>
> + if (!(dev->cfg.device_status & VIRTIO_CONFIG_DEVICE_STATUS_DRIVER_OK))
> + return ret;
> +
> vr_sz = vring_size(VIONET_QUEUE_SIZE);
> q_gpa = dev->vq[0].qa;
> q_gpa = q_gpa * VIRTIO_PAGE_SIZE;
>
This diff works fine, huge thanks for stepping in!
