Re: [PATCH 0/12] virtio_net perf patches

[Avi Kivity <avi@qumranet.com>]

Mark McLoughlin wrote:
> Hi Avi,
>         Here's the set of patches that I think make sense to apply.
>
>         I've left out Anthony's zero-copy patch since I wasn't seeing
> a measurable difference with it, it's quite invasive and isn't safe
> with the "drop the global mutex during tapfd read()" patch.
>
>         I left that mutex patch 'til last since you may still be
> nervous about concurrency issues and also the posix-timers kernel race
> that it triggers.
>   

I'm nervous, but let's see what happens.  Worst case we make it conditional.

>         Another change since the first round of patches is that it
> took a bit of ugly hackery to handle the case where IFF_VNET_HDR
> is supported and we're using e.g. e1000. Since it's difficult to
> only enable IFF_VNET_HDR when we're using virtio_net_hdr, I went with
> this approach.
>
>   

Thanks, applied all...

>         Lastly, I need to add a "vnet_hdr=on" param to "-net tap" so
> that we can know if the supplied tap fd has IFF_VNET_HDR enabled.
> There's no interface to query that on the fd right now, and even if
> I did cook up a patch it probably wouldn't make 2.6.27.
>   

except for this, as I agree with Anthony.  I think it can make it into 
2.6.27, but if it doesn't, libvirt should simply not enable vnet_hdr 
unless it is sure qemu can query for the feature.

-- 
error compiling committee.c: too many arguments to function