Re: [PATCH 3/3] virtio-net: graceful fallback to vhost=off for tap netdev

[Jason Wang <jasowang@redhat.com>]

On 2021/2/9 上午3:46, Yuri Benditovich wrote:
> On Mon, Feb 8, 2021 at 5:15 AM Jason Wang <jasowang@redhat.com> wrote:
>>
>> On 2021/2/5 下午9:38, Michael S. Tsirkin wrote:
>>> On Thu, Feb 04, 2021 at 10:29:15PM +0200, Yuri Benditovich wrote:
>>>> Currently virtio-net silently clears features if they are
>>>> not supported by respective vhost. This may create migration
>>>> problems in future if vhost features on the source and destination
>>>> are different. Implement graceful fallback to no-vhost mode
>>>> when some acked features contradict with vhost. The decision is
>>>> taken on set_features call and the vhost will be disabled
>>>> till next reset (or migration).
>>>> Such fallback is currently enabled only for TAP netdev.
>>>>
>>>> Signed-off-by: Yuri Benditovich<yuri.benditovich@daynix.com>
>>> Sounds good, but I don't think we should do this if
>>> vhostforce=on is set.
>>
>> If we do this, does it mean we won't maintain migration compatibility
>> when vhostforce is on?
> AFAIU, the 'vhostforce=on' should mean the vhost can't be disabled (if
> I'm not mistaken this is typically used for vhost-user).
> So we can view this case as similar to vhost-vdpa and vhost-user.


Right, but since it was used by libivrt. Then it turns out to be a 
compatibility breaker.

Thanks


>
>> Thanks
>>
>>
>>> Also, let's document this behaviour with the vhost option so people
>>> are not suprized.
>>>