Re: [PATCH for-7.2] vhost: enable vrings in vhost_dev_start() for vhost-user devices

[Stefano Garzarella]

On Thu, Dec 01, 2022 at 04:49:37PM -0500, Michael S. Tsirkin wrote:
> On Thu, Dec 01, 2022 at 12:21:21PM +0000, Alex Bennée wrote:
> > "Michael S. Tsirkin" <mst@redhat.com> writes:
> >
> > > On Thu, Dec 01, 2022 at 10:14:39AM +0000, Alex Bennée wrote:
> > >> Do you think rust-vmm's vhost crates have enough of the state
> > >> management to manage vhost and vhost-user backends? Maybe it would be a
> > >> good experiment in replacing a (small well defined) piece of
> > >> functionality with rust?

I honestly don't know, but I think they are not 100% complete.
I do agree that an experiment would be nice though, maybe a GSoC?
Or maybe not, since it's not very clear whether we get to the end or not....

> > >>
> > >> That said there is a lot of deep magic in the vhost-net stuff which I
> > >> think is down to the interaction with things like vdpk and other network
> > >> optimisations that might be missing. For the rest of the devices most of
> > >> the code is basically boiler plate which has grown variations due to
> > >> code motion and change. This is the sort of thing that generics solves
> > >> well.
> > >
> > > Not sure what you want to replace with what though, libvhost-user or
> > > vhost-user bits in qemu?
> >
> > The vhost-user bits in the main QEMU binary. We already don't use
> > libvhost-user for most of our backends anyway ;-)
>
> Mixing C and Rust like this is far from trivial. I'd start with
> something much less ambitious that virtio.

We recently merged libblkio [1] support in QEMU [2], where we did
something similar.
libblkio provides a C API, but the library is totally written in Rust.
Writing the C interface was not too complicated.

The advantage here is that the Rust code is already part of the rust-vmm
crates.
I think it needs adjustments, but it's not a bad idea in my opinion.

Anyway, I don't think it's an easy job, though, so maybe GSoC could just
be about writing the C interface and adapting Rust crates. Touching QEMU
might require too much knowledge of the current code.

Thanks,
Stefano

[1] https://libblkio.gitlab.io/libblkio/
[2] https://lists.nongnu.org/archive/html/qemu-devel/2022-10/msg02215.html