colbyt <colby@colbyt.com> writes:

> Update Spectrum's patched Cloud Hypervisor package to 52.0 and rebase
> the local virtio-gpu/vhost-user GPU support onto that release.
>
> Cloud Hypervisor 52.0 uses vhost 0.16.0 and vhost-user-backend 0.22.0,
> so update the paired rust-vmm/vhost checkout as well.  The local vhost
> shared-memory patches are no longer applied because vhost-user-backend
> 0.22.0 already has the protocol support used by the GPU frontend.
>
> The cargo vendor derivation now vendors dependencies after applying the
> local Cloud Hypervisor and vhost patches, so the patched vhost checkout
> is included in the fixed-output dependency tree.
>
> Signed-off-by: colbyt <colby@colbyt.com>
> ---

For Cloud Hypervisor 52.0, my plan had been to drop the custom GPU
device, and just extend the new generic vhost-user device Demi
upstreamed to support SHMEM, reducing the divergence with upstream and
easing the path towards upstreaming the rest.  I have this version of
the patch locally, but had been waiting for the Nixpkgs update that will
bring in the new Cloud Hypervisor.  I expect that Nixpkgs update will
come in the next week or two, after the next Nixpkgs staging cycle
(which will bring some package fixes we need).  Keeping Nixpkgs up to date got
away from me this year so far, because I was focused on some big
organizational/long-term things for the first few months (that I hope
will visibly bear fruit shortly…), but I've been back on it in the last
few weeks getting Nixpkgs back into shape, so it should be back to
regular updates shortly.

If you'd like though, we could decouple our Cloud Hypervisor version
from Nixpkgs' like you've done here, and I could update Cloud Hypervisor
to 52.0 with my patch to extend the generic vhost-user device, rather
than waiting another week or two for the Nixpkgs update.  What do you
think?