Re: [Qemu-devel] [PATCH 00/16] Add a -netdev option

["Michael S. Tsirkin" <mst@redhat.com>]

On Fri, Oct 09, 2009 at 10:26:10AM -0500, Anthony Liguori wrote:
> Hi Mark,
>
> Mark McLoughlin wrote:
>> Hi,
>>         Here's a series of patches which gets the ball rolling on adding
>> a -netdev option.
>>
>>         The idea is to de-emphasise the vlan support, and instead make
>> a nic directly connected to a host backend the default and recommended
>> configuration. We want this because it is only with this configuration
>> that we feasibly add optimizations like GSO support or vhost-net.
>>   
>
> I think this series is a good first step.  I've looked through the  
> patches and they all look pretty good.  I plan on pushing after some  
> testing assuming noone objects in the interim.
>
> With respect to moving forward, I wanted to bring up a couple of topics.
>
> To eliminate VLANs, I think the reasonable approach is to implemented an  
> emulated hub.  I'd suggest treating the hub like a device.  Instantiate  
> it through qdev, support it via -device, etc.  I think it would even be  
> reasonable to have it support a fixed number of ports.  I doubt there  
> would be any compatibility issues emulating a 32-bit port hub, for 
> instance.
>
> The current syntax would become just a convenient short cut for the  
> -netdev syntax along with creating a hub device for each vlan that gets  
> instantiated.
>
> I think merging vhost_net support will depend on finishing out the  
> netdev work.

More importantly, we need interrupt injection from kvm merged.
We can have userspace poll vhost net and inject the interrupt,
but I think this defeats the purpose of the in-kernel backend.

> Logically, I think vhost_net is a netdev backend and we  
> extend the netdev API to support async submission of packets (basically  
> mirroring the virtio-net ABI).  In the case of a virtio-net front-end to  
> a vhost_net backend, I think we special case that assuming we're running  
> kvm.
>
> Regards,
>
> Anthony Liguori