> I this specific case, one basically replaces /machine/unattached > orphanage with explicit /machine one and many 'cpuN' children, > which ain't any better than device[N]. > > and in future I can imagine that at least in x86 case vcpus > might have another parent depending on configuration. > (i.e. being parented to cores instead)
I remember that this was your idea all along, and I'm not sure if you're also referring to my previous patches about hybrid topology :-), which I'll continue to refresh afterward in future (after all, the hybrid architecture will continue in x86). And I think, since socket/core/thread are the three default QEMU topology hierarchies, I understand that it would be best for thread to always have core as parent.
