On 11 Dec 2018, at 16:10, Gianluca Cecchi
<gianluca.cecchi(a)gmail.com> wrote:
On Tue, Dec 11, 2018 at 3:29 PM Markus Schaufler <markus.schaufler(a)digit-all.at
<mailto:markus.schaufler@digit-all.at>> wrote:
Centos 7.5, ovirt 4.2.7
it works when pinpointing the vm's to a host and enable cpu-passthrough.
but i presume it should work without this workaround too.
_______________________________________________
If I remember correctly there is no support/consistence for the L2 systems, that are
guests inside an L1 system that is migrated.
So perhaps this limitation drove to force host pinning and prevent having unpredictable
results for L2 guests
it’s not really done because of that - it’s rather that by default none of the named CPU
models are adding vmx and so the guest doesn’t look like capable of virtualization. Cpu
passthrough does, because it passes all the flags, but it also enforces host pinning.
Which is then a nice coincidence because indeed the migration wouldn’t work anyway:)
And it’s something to keep in mind now in later 4.2.z and 4.3 where we now allow migration
of VMs with cpu passthrough. But it’s not for this use case and there nested environment
still doesn’t support migration properly and you need to make sure not to migrate.