Re: [Qemu-devel] [PATCH 4/4] virtio-net: add linkspeed and duplex settings to virtio-net

[Jason Baron]

On 03/02/2018 03:22 PM, Michael S. Tsirkin wrote:
> On Fri, Mar 02, 2018 at 11:59:00AM -0500, Jason Baron wrote:
>> On 03/02/2018 02:14 AM, Jason Wang wrote:
>>>
>>>
>>> On 2018年03月02日 11:46, Jason Baron wrote:
>>>> Although linkspeed and duplex can be set in a linux guest via 'ethtool
>>>> -s',
>>>> this requires custom ethtool commands for virtio-net by default.
>>>>
>>>> Introduce a new feature flag, VIRTIO_NET_F_SPEED_DUPLEX, which allows
>>>> the hypervisor to export a linkspeed and duplex setting. The user can
>>>> subsequently overwrite it later if desired via: 'ethtool -s'.
>>>>
>>>> Linkspeed and duplex settings can be set as:
>>>> '-device virtio-net,speed=10000,duplex=full'
>>>
>>> I was thinking whether or not it's better to decide the duplex by the
>>> type of backends.
>>>
>>> E.g userspace and vhost-kernel implement a in fact half duplex. But dpdk
>>> implement a full duplex.
>>
>> Interesting - could this be derived only from the backend 'type'. IE:
>> NET_CLIENT_DRIVER_TAP, NET_CLIENT_DRIVER_VHOST_USER...
>>
>>
>> I was also thinking this could be specified as 'duplex=backend', in
>> addition to the proposed 'duplex=full' or 'duplex=half'?
>>
>> Thanks,
>>
>> -Jason
> 
> I'd say it would make more sense to teach backends to obey what's
> specified by the user. E.g. if vhost gets a duplex config,
> create two threads.
> 
> But I think all that's for future, we can just fake it for
> now - the current uses don't seem to particularly care about whether
> virtio actually is or isn't a duplex.
> 
> 

Ok, I wouldn't add 'duplex=backend' when I re-post, and will leave any
automatic settings of duplex to 'future work'.

Thanks,

-Jason