Re: [PATCH v2 1/1] virtio_net: notify MAC address change on device initialization

[Michael S. Tsirkin]

On Tue, Jan 24, 2023 at 12:04:24PM +0100, Laurent Vivier wrote:
> On 1/24/23 11:15, Michael S. Tsirkin wrote:
> > On Mon, Jan 23, 2023 at 01:00:22PM +0100, Laurent Vivier wrote:
> > > In virtnet_probe(), if the device doesn't provide a MAC address the
> > > driver assigns a random one.
> > > As we modify the MAC address we need to notify the device to allow it
> > > to update all the related information.
> > > 
> > > The problem can be seen with vDPA and mlx5_vdpa driver as it doesn't
> > > assign a MAC address by default. The virtio_net device uses a random
> > > MAC address (we can see it with "ip link"), but we can't ping a net
> > > namespace from another one using the virtio-vdpa device because the
> > > new MAC address has not been provided to the hardware.
> > 
> > And then what exactly happens? Does hardware drop the outgoing
> > or the incoming packets? Pls include in the commit log.
> 
> I don't know. There is nothing in the kernel logs.
> 
> The ping error is: "Destination Host Unreachable"
> 
> I found the problem with the mlx5 driver as in "it doesn't work when MAC
> address is not set"...
> 
> Perhaps Eli can explain what happens when the MAC address is not set?
> 
> > 
> > > Signed-off-by: Laurent Vivier <lvivier@xxxxxxxxxx>
> > > ---
> > >   drivers/net/virtio_net.c | 14 ++++++++++++++
> > >   1 file changed, 14 insertions(+)
> > > 
> > > diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
> > > index 7723b2a49d8e..4bdc8286678b 100644
> > > --- a/drivers/net/virtio_net.c
> > > +++ b/drivers/net/virtio_net.c
> > > @@ -3800,6 +3800,8 @@ static int virtnet_probe(struct virtio_device *vdev)
> > >   		eth_hw_addr_set(dev, addr);
> > >   	} else {
> > >   		eth_hw_addr_random(dev);
> > > +		dev_info(&vdev->dev, "Assigned random MAC address %pM\n",
> > > +			 dev->dev_addr);
> > >   	}
> > >   	/* Set up our device-specific information */
> > > @@ -3956,6 +3958,18 @@ static int virtnet_probe(struct virtio_device *vdev)
> > >   	pr_debug("virtnet: registered device %s with %d RX and TX vq's\n",
> > >   		 dev->name, max_queue_pairs);
> > > +	/* a random MAC address has been assigned, notify the device */
> > > +	if (!virtio_has_feature(vdev, VIRTIO_NET_F_MAC) &&
> > > +	    virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
> > 
> > Maybe add a comment explaining that we don't fail probe if
> > VIRTIO_NET_F_CTRL_MAC_ADDR is not there because
> > many devices work fine without getting MAC explicitly.
> 
> OK
> 
> > 
> > > +		struct scatterlist sg;
> > > +
> > > +		sg_init_one(&sg, dev->dev_addr, dev->addr_len);
> > > +		if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
> > > +					  VIRTIO_NET_CTRL_MAC_ADDR_SET, &sg)) {
> > > +			dev_warn(&vdev->dev, "Failed to update MAC address.\n");
> > 
> > Here, I'm not sure we want to proceed. Is it useful sometimes?
> 
> I think reporting an error is always useful, but I can remove that if you prefer.

No the question was whether we should fail probe not
whether we print the warning.


> > I note that we deny with virtnet_set_mac_address.
> > 
> > > +		}
> > > +	}
> > > +
> > >   	return 0;
> > 
> > 
> > 
> > Also, some code duplication with virtnet_set_mac_address here.
> > 
> > Also:
> > 	When using the legacy interface, \field{mac} is driver-writable
> > 	which provided a way for drivers to update the MAC without
> > 	negotiating VIRTIO_NET_F_CTRL_MAC_ADDR.
> > 
> > How about factoring out code in virtnet_set_mac_address
> > and reusing that?
> > 
> 
> In fact, we can write in the field only if we have VIRTIO_NET_F_MAC
> (according to virtnet_set_mac_address(), and this code is executed only if
> we do not have VIRTIO_NET_F_MAC. So I think it's better not factoring the
> code as we have only the control queue case to manage.
> 
> > This will also handle corner cases such as VIRTIO_NET_F_STANDBY
> > which are not currently addressed.
> 
> F_STANDBY is only enabled when virtio-net device MAC address is equal to the
> VFIO device MAC address, I don't think it can be enabled when the MAC
> address is randomly assigned (in this case it has already failed in
> net_failover_create(), as it has been called using the random mac address),
> it's why I didn't check for it.

But the spec did not say there's a dependency :(.
My point is what should we do if there's F_STANDBY but no MAC?
Maybe add a separate patch clearing F_STANDBY in this case?

> > 
> > 
> > >   free_unregister_netdev:
> > > -- 
> > > 2.39.0
> > 
> 
> Thanks,
> Laurent