[Qemu-devel] [PATCH v2] virtio-net: unbreak the minix guest

[Qemu-devel] [PATCH v2] virtio-net: unbreak the minix guest

[Jason Wang]

Multiqueue patchset conditionally add control vq only when guest negotiate the
feature. Though the spec is not clear on this but it breaks the minix guest
since it will identify the ctrl vq even if it does not support it. Though this
behavior seems a violation on the spec "If the VIRTIO_NET_F_CTRL_VQ feature bit
is negotiated, identify the control virtqueue.", to keep the backward
compatibility, always add the ctrl vq at end of the queues.

Reported-by: Aurelien Jarno <aurelien@aurel32.net>
Cc: Aurelien Jarno <aurelien@aurel32.net>
Signed-off-by: Jason Wang <jasowang@redhat.com>

---
Changes from V1:
- Drop the unrelated cleanups
- Drop the meaningless ctrl parameter of virtio_net_set_multiqueue()
---
 hw/net/virtio-net.c |   15 ++++++++-------
 1 files changed, 8 insertions(+), 7 deletions(-)

diff --git a/hw/net/virtio-net.c b/hw/net/virtio-net.c
index 4d2cdd2..1662f46 100644
--- a/hw/net/virtio-net.c
+++ b/hw/net/virtio-net.c
@@ -309,7 +309,7 @@ static void virtio_net_set_queues(VirtIONet *n)
     }
 }
 
-static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue, int ctrl);
+static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue);
 
 static uint32_t virtio_net_get_features(VirtIODevice *vdev, uint32_t features)
 {
@@ -364,8 +364,7 @@ static void virtio_net_set_features(VirtIODevice *vdev, uint32_t features)
     VirtIONet *n = VIRTIO_NET(vdev);
     int i;
 
-    virtio_net_set_multiqueue(n, !!(features & (1 << VIRTIO_NET_F_MQ)),
-                              !!(features & (1 << VIRTIO_NET_F_CTRL_VQ)));
+    virtio_net_set_multiqueue(n, !!(features & (1 << VIRTIO_NET_F_MQ)));
 
     virtio_net_set_mrg_rx_bufs(n, !!(features & (1 << VIRTIO_NET_F_MRG_RXBUF)));
 
@@ -1038,7 +1037,7 @@ static void virtio_net_tx_bh(void *opaque)
     }
 }
 
-static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue, int ctrl)
+static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue)
 {
     VirtIODevice *vdev = VIRTIO_DEVICE(n);
     int i, max = multiqueue ? n->max_queues : 1;
@@ -1067,9 +1066,11 @@ static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue, int ctrl)
         n->vqs[i].n = n;
     }
 
-    if (ctrl) {
-        n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
-    }
+    /* Note: Minux Guests (version 3.2.1) use ctrl vq but don't ack
+     * VIRTIO_NET_F_CTRL_VQ. Create ctrl vq unconditionally to avoid
+     * breaking them.
+     */
+    n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
 
     virtio_net_set_queues(n);
 }
-- 
1.7.1

Re: [Qemu-devel] [PATCH v2] virtio-net: unbreak the minix guest

[Michael S. Tsirkin]

On Thu, Apr 25, 2013 at 03:24:23PM +0800, Jason Wang wrote:
> Multiqueue patchset conditionally add control vq only when guest negotiate the
> feature. Though the spec is not clear on this but it breaks the minix guest
> since it will identify the ctrl vq even if it does not support it. Though this
> behavior seems a violation on the spec "If the VIRTIO_NET_F_CTRL_VQ feature bit
> is negotiated, identify the control virtqueue.", to keep the backward
> compatibility, always add the ctrl vq at end of the queues.
> 
> Reported-by: Aurelien Jarno <aurelien@aurel32.net>
> Cc: Aurelien Jarno <aurelien@aurel32.net>
> Signed-off-by: Jason Wang <jasowang@redhat.com>


Acked-by: Michael S. Tsirkin <mst@redhat.com>

> ---
> Changes from V1:
> - Drop the unrelated cleanups
> - Drop the meaningless ctrl parameter of virtio_net_set_multiqueue()
> ---
>  hw/net/virtio-net.c |   15 ++++++++-------
>  1 files changed, 8 insertions(+), 7 deletions(-)
> 
> diff --git a/hw/net/virtio-net.c b/hw/net/virtio-net.c
> index 4d2cdd2..1662f46 100644
> --- a/hw/net/virtio-net.c
> +++ b/hw/net/virtio-net.c
> @@ -309,7 +309,7 @@ static void virtio_net_set_queues(VirtIONet *n)
>      }
>  }
>  
> -static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue, int ctrl);
> +static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue);
>  
>  static uint32_t virtio_net_get_features(VirtIODevice *vdev, uint32_t features)
>  {
> @@ -364,8 +364,7 @@ static void virtio_net_set_features(VirtIODevice *vdev, uint32_t features)
>      VirtIONet *n = VIRTIO_NET(vdev);
>      int i;
>  
> -    virtio_net_set_multiqueue(n, !!(features & (1 << VIRTIO_NET_F_MQ)),
> -                              !!(features & (1 << VIRTIO_NET_F_CTRL_VQ)));
> +    virtio_net_set_multiqueue(n, !!(features & (1 << VIRTIO_NET_F_MQ)));
>  
>      virtio_net_set_mrg_rx_bufs(n, !!(features & (1 << VIRTIO_NET_F_MRG_RXBUF)));
>  
> @@ -1038,7 +1037,7 @@ static void virtio_net_tx_bh(void *opaque)
>      }
>  }
>  
> -static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue, int ctrl)
> +static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue)
>  {
>      VirtIODevice *vdev = VIRTIO_DEVICE(n);
>      int i, max = multiqueue ? n->max_queues : 1;
> @@ -1067,9 +1066,11 @@ static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue, int ctrl)
>          n->vqs[i].n = n;
>      }
>  
> -    if (ctrl) {
> -        n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
> -    }
> +    /* Note: Minux Guests (version 3.2.1) use ctrl vq but don't ack
> +     * VIRTIO_NET_F_CTRL_VQ. Create ctrl vq unconditionally to avoid
> +     * breaking them.
> +     */
> +    n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
>  
>      virtio_net_set_queues(n);
>  }
> -- 
> 1.7.1

Re: [Qemu-devel] [PATCH v2] virtio-net: unbreak the minix guest

[Aurelien Jarno]

On Thu, Apr 25, 2013 at 03:24:23PM +0800, Jason Wang wrote:
> Multiqueue patchset conditionally add control vq only when guest negotiate the
> feature. Though the spec is not clear on this but it breaks the minix guest
> since it will identify the ctrl vq even if it does not support it. Though this
> behavior seems a violation on the spec "If the VIRTIO_NET_F_CTRL_VQ feature bit
> is negotiated, identify the control virtqueue.", to keep the backward
> compatibility, always add the ctrl vq at end of the queues.
> 
> Reported-by: Aurelien Jarno <aurelien@aurel32.net>
> Cc: Aurelien Jarno <aurelien@aurel32.net>
> Signed-off-by: Jason Wang <jasowang@redhat.com>
> 
> ---
> Changes from V1:
> - Drop the unrelated cleanups
> - Drop the meaningless ctrl parameter of virtio_net_set_multiqueue()
> ---
>  hw/net/virtio-net.c |   15 ++++++++-------
>  1 files changed, 8 insertions(+), 7 deletions(-)


Thanks a lot for this patch. It fixes the issue observed on Minix. For
the record it also fixes virtio-net with NetBSD 6.0.1, before it was
triggering an assertion in QEMU.

Now that the problem is clear, I will try to work on fixing the problem
on the OS side when I find some time.

Tested-by: Aurelien Jarno <aurelien@aurel32.net>

> diff --git a/hw/net/virtio-net.c b/hw/net/virtio-net.c
> index 4d2cdd2..1662f46 100644
> --- a/hw/net/virtio-net.c
> +++ b/hw/net/virtio-net.c
> @@ -309,7 +309,7 @@ static void virtio_net_set_queues(VirtIONet *n)
>      }
>  }
>  
> -static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue, int ctrl);
> +static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue);
>  
>  static uint32_t virtio_net_get_features(VirtIODevice *vdev, uint32_t features)
>  {
> @@ -364,8 +364,7 @@ static void virtio_net_set_features(VirtIODevice *vdev, uint32_t features)
>      VirtIONet *n = VIRTIO_NET(vdev);
>      int i;
>  
> -    virtio_net_set_multiqueue(n, !!(features & (1 << VIRTIO_NET_F_MQ)),
> -                              !!(features & (1 << VIRTIO_NET_F_CTRL_VQ)));
> +    virtio_net_set_multiqueue(n, !!(features & (1 << VIRTIO_NET_F_MQ)));
>  
>      virtio_net_set_mrg_rx_bufs(n, !!(features & (1 << VIRTIO_NET_F_MRG_RXBUF)));
>  
> @@ -1038,7 +1037,7 @@ static void virtio_net_tx_bh(void *opaque)
>      }
>  }
>  
> -static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue, int ctrl)
> +static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue)
>  {
>      VirtIODevice *vdev = VIRTIO_DEVICE(n);
>      int i, max = multiqueue ? n->max_queues : 1;
> @@ -1067,9 +1066,11 @@ static void virtio_net_set_multiqueue(VirtIONet *n, int multiqueue, int ctrl)
>          n->vqs[i].n = n;
>      }
>  
> -    if (ctrl) {
> -        n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
> -    }
> +    /* Note: Minux Guests (version 3.2.1) use ctrl vq but don't ack
> +     * VIRTIO_NET_F_CTRL_VQ. Create ctrl vq unconditionally to avoid
> +     * breaking them.
> +     */
> +    n->ctrl_vq = virtio_add_queue(vdev, 64, virtio_net_handle_ctrl);
>  
>      virtio_net_set_queues(n);
>  }
> -- 
> 1.7.1
> 
> 

-- 
Aurelien Jarno                          GPG: 1024D/F1BCDB73
aurelien@aurel32.net                 http://www.aurel32.net

Re: [Qemu-devel] [PATCH v2] virtio-net: unbreak the minix guest

[Aurelien Jarno]

On Fri, Apr 26, 2013 at 12:49:53PM +0200, Aurelien Jarno wrote:
> On Thu, Apr 25, 2013 at 03:24:23PM +0800, Jason Wang wrote:
> > Multiqueue patchset conditionally add control vq only when guest negotiate the
> > feature. Though the spec is not clear on this but it breaks the minix guest
> > since it will identify the ctrl vq even if it does not support it. Though this
> > behavior seems a violation on the spec "If the VIRTIO_NET_F_CTRL_VQ feature bit
> > is negotiated, identify the control virtqueue.", to keep the backward
> > compatibility, always add the ctrl vq at end of the queues.
> > 
> > Reported-by: Aurelien Jarno <aurelien@aurel32.net>
> > Cc: Aurelien Jarno <aurelien@aurel32.net>
> > Signed-off-by: Jason Wang <jasowang@redhat.com>
> > 
> > ---
> > Changes from V1:
> > - Drop the unrelated cleanups
> > - Drop the meaningless ctrl parameter of virtio_net_set_multiqueue()
> > ---
> >  hw/net/virtio-net.c |   15 ++++++++-------
> >  1 files changed, 8 insertions(+), 7 deletions(-)
> 
> 
> Thanks a lot for this patch. It fixes the issue observed on Minix. For
> the record it also fixes virtio-net with NetBSD 6.0.1, before it was
> triggering an assertion in QEMU.
> 
> Now that the problem is clear, I will try to work on fixing the problem
> on the OS side when I find some time.
> 

For the record, now that I understand the problem, I have been able to
fix the issue on the guest side. Here are the links to the bug reports
and patches:

- Minix: https://gforge.cs.vu.nl/gf/project/minix/tracker/?action=TrackerItemEdit&tracker_item_id=653

- NetBSD: http://gnats.netbsd.org/47780

-- 
Aurelien Jarno                          GPG: 1024D/F1BCDB73
aurelien@aurel32.net                 http://www.aurel32.net

Re: [Qemu-devel] [PATCH v2] virtio-net: unbreak the minix guest

[Anthony Liguori]

Applied.  Thanks.

Regards,

Anthony Liguori