Re: [Qemu-devel] [PULL v4 02/29] vhost-user: add new vhost user messages to support virtio config space

[Michael S. Tsirkin]

On Wed, Mar 28, 2018 at 12:03:36PM +0200, Maxime Coquelin wrote:
> Hi,
> 
> On 01/18/2018 09:44 PM, Michael S. Tsirkin wrote:
> > From: Changpeng Liu <address@hidden>
> > 
> > Add VHOST_USER_GET_CONFIG/VHOST_USER_SET_CONFIG messages which can be
> > used for live migration of vhost user devices, also vhost user devices
> > can benefit from the messages to get/set virtio config space from/to the
> > I/O target. For the purpose to support virtio config space change,
> > VHOST_USER_SLAVE_CONFIG_CHANGE_MSG message is added as the event notifier
> > in case virtio config space change in the slave I/O target.
> > 
> > Signed-off-by: Changpeng Liu <address@hidden>
> > Reviewed-by: Marc-André Lureau <address@hidden>
> > Reviewed-by: Michael S. Tsirkin <address@hidden>
> > Signed-off-by: Michael S. Tsirkin <address@hidden>
> > ---
> >   docs/interop/vhost-user.txt       |  55 ++++++++++++++++++
> >   include/hw/virtio/vhost-backend.h |  12 ++++
> >   include/hw/virtio/vhost.h         |  15 +++++
> >   hw/virtio/vhost-user.c            | 118 
> > ++++++++++++++++++++++++++++++++++++++
> >   hw/virtio/vhost.c                 |  32 +++++++++++
> >   5 files changed, 232 insertions(+)
> > 
> 
> As discussed with Changpeng on DPDK ML, we need a vhost-user protocol
> feature bit to back these new requests, else QEMU cannot whether the
> user backend implement these.
> 
> With current and older DPDK version, if an unknown request is received,
> the socket connection is closed.
> 
> Do we still have time before QEMU v2.12 is out?
> 
> Thanks,
> Maxime

We certainly do, and I agree it's a concern as we get maintain
protocol extensions forever.

It's my bad that I forgot to check that there's a feature bit, but I
think we should act before the release.

There is probably time enough to add it - but if no one
can spare then time it would be better to disable the new messages in
QEMU for now than ship 2.12 without a feature bit.

I wonder what's Changpeng Liu's take on this.

-- 
MST