On Wed, Nov 19, 2025 at 4:07 PM Michael S. Tsirkin <[email protected]> wrote:
>
> On Wed, Nov 19, 2025 at 10:49:11AM +0800, Jason Wang wrote:
> > > qemu already probes tap features.
> >
> > Only part of the features.
>
>
> the part we care about?

Yes and technically Qemu can probe all.

>
> > > To me, it seems natural
> > > for management to do the probing through qemu.
> > > in fact your patch is a way to do that, is it not?
> >
> > Yes and no.
> >
> > > what it lacks though is a structured way to tell management how
> > > to fix the problem.
> >
> > Probing through management seems to be better. For example it can
> > calculate the cluster in advance without the need to launch qemu
> > everywhere.
>
> it is basically replicating qemu code then.

Yes but libvirt has duplicated somehow, e.g create and destroy TAP and
it even uses TUNGETFEATURES in virNetDevMacVLanTapSetup(). But I'm not
even sure this is the work of the libvirt. I think it should be the
job of the one who is in charge of managing the cpu cluster.

>
> what's the big deal to launch qemu?

It looks more heavyweight than probing by libvirt but I'm not sure, we
can listen to libvirt guys.

>
>
>
> > Or consider the case when USO is not supported by the kernel in the
> > destination, even if qemu reports this, I'm not sure what is expected
> > to be done in the management layer?
> >
> > Thanks
>
> reports what?

USO is not supported by this kernel.

Btw, to not make things worse, I would revert this until we find a
solution. Do you agree?

commit 1c79ab6937ae938d3dfd4da1c01afc7eb599857e
Author: Paolo Abeni <[email protected]>
Date:   Fri Oct 10 16:12:57 2025 +0200

    virtio-net: Advertise UDP tunnel GSO support by default

    Allow bidirectional aggregated traffic for UDP encapsulated flows.

    Add the needed compatibility entries to avoid migration issues
    vs older QEMU instances.

    Signed-off-by: Paolo Abeni <[email protected]>
    Acked-by: Jason Wang <[email protected]>
    Tested-by: Lei Yang <[email protected]>
    Reviewed-by: Michael S. Tsirkin <[email protected]>
    Signed-off-by: Michael S. Tsirkin <[email protected]>
    Message-Id:
<9c500fbcd2cf29afd1826b1ac906f9d5beac3601.1760104079.git.pab...@redhat.com>

Otherwise we will suffer from a similar issue soon.

Thanks

>
> --
> MST
>


Reply via email to