Re: Emulating device configuration / max_virtqueue_pairs in vhost-vdpa and vhost-user

[Jason Wang]

On Wed, Feb 1, 2023 at 2:56 PM Eugenio Perez Martin <eperezma@redhat.com> wrote:
>
> On Wed, Feb 1, 2023 at 4:27 AM Jason Wang <jasowang@redhat.com> wrote:
> >
> > On Wed, Feb 1, 2023 at 3:10 AM Eugenio Perez Martin <eperezma@redhat.com> 
> > wrote:
> > >
> > > Hi,
> > >
> > > The current approach of offering an emulated CVQ to the guest and map
> > > the commands to vhost-user is not scaling well:
> > > * Some devices already offer it, so the transformation is redundant.
> > > * There is no support for commands with variable length (RSS?)
> > >
> > > We can solve both of them by offering it through vhost-user the same
> > > way as vhost-vdpa do. With this approach qemu needs to track the
> > > commands, for similar reasons as vhost-vdpa: qemu needs to track the
> > > device status for live migration. vhost-user should use the same SVQ
> > > code for this, so we avoid duplications.
> >
> > Note that it really depends on the model we used. SVQ works well for
> > trap and emulation (without new API to be invented). But if save and
> > load API is invented, SVQ is not a must.
> >
>
> That's right, but the premise in the proposal is that control vq RSS
> messages are already complex enough to avoid adding more vhost-user
> messages. I cannot imagine expanding the vhost-user or virtio API to
> add the save and restore functions soon :).

Probably, but we know we need that some day.

>
> > >
> > > One of the challenges here is to know what virtqueue to shadow /
> > > isolate. The vhost-user device may not have the same queues as the
> > > device frontend:
> > > * The first depends on the actual vhost-user device, and qemu fetches
> > > it with VHOST_USER_GET_QUEUE_NUM at the moment.
> > > * The qemu device frontend's is set by netdev queues= cmdline parameter 
> > > in qemu
> > >
> > > For the device, the CVQ is the last one it offers, but for the guest
> > > it is the last one offered in config space.
> > >
> > > To create a new vhost-user command to decrease that maximum number of
> > > queues may be an option. But we can do it without adding more
> > > commands, remapping the CVQ index at virtqueue setup. I think it
> > > should be doable using (struct vhost_dev).vq_index and maybe a few
> > > adjustments here and there.
> >
> > It requires device specific knowledge, it might work for networking
> > devices but not others (or need new codes).
> >
>
> Yes, I didn't review all the other kinds of devices for the proposal,
> but I'm assuming:
> * There is no other device that has already implemented CVQ over
> vhost-user (or this problems would have been solved)

If I was not wrong, vhost-user protocol itself allows arbitrary types
of virtqueue to be implemented on top. So it might just be an
implementation issue.

> * All vhost-user devices config space are already offered by qemu like
> vhost-user net, and the cvq-alike index is well defined in the
> standard like -net.
>
> So this proposal should fit all those devices, isn't it?

If it could be done without extending vhost-user API, that would be fine.

Thanks

>
> Thanks!
>