Re: [Qemu-devel] [PATCH v4 0/3] virtio-net: Add support to MTU feature

[Maxime Coquelin]

Hi Daniel,

On 12/12/2016 11:02 AM, Daniel P. Berrange wrote:
> On Sat, Dec 10, 2016 at 04:30:35PM +0100, Maxime Coquelin wrote:
> > Thanks for the reviews,
> >
> > This series implements Virtio spec update from Aaron Conole which
> > defines a way for the host to expose its max MTU to the guest.
> >
> > "host_mtu" parameter is added to provide QEMU with the MTU value,
> > and the backend, if supported, gets notified of the MTU value when the
> > MTU feature neogotiation succeeds.
> >
> > Only user backend currently supports MTU notification. A new protocol
> > feature has been implemented for sending MTU value to the backend.
> >
> > For kernel backend, it is expected the management tool also configures
> > the tap/macvtap interface with same MTU value.
> > Daniel, I would be interrested about your feedback on this implementation
> > from management tool point of view.
>
> I can't give real feedback yet, as I'm not seeing clear information on
> what problem this series is designed to solve....

Right, I agree it is missing a bit of context here, I'll add more about
the background in next revision.

The goal of this series is to address two things:
1. Providing a way for the guests to use the same MTU as the host,
   in order to have a consistent MTU value across the infrastructure.

2. Improve Rx performance for small packets, as if all parties agrees
   on a MTU value for which packets fit into a single descriptor, then
   Rx mergeable buffers feature can be disabled (saving one
   cache-miss on guest side by not accessing the virtio header, when
   offloading has not been negotiated).

Thanks,
Maxime