Alan Cox wrote:

- Stacked hypervisors stomping each others functions
  

Possibly an issue, but why would you ever want stacked paravirt-ops?  
You're only talking to the hypervisor directly above you, and there is 
only one of those.

- Locking required to do updates: and remember our lock functions use
methods in the array
  

Yes, locking is an issue, but it is possible to do.  You just need to 
stop interrupts, NMIs, and faults on all processors simultaneously.  
Actually, it's not that scary - since you'll be doing it in a hypervisor.

- If we boot patch inline code to get performance natively its almost
impossible to then revert that.
  

You can patch back over it.  I've already implemented the locking and 
repatching bits for VMI.

Zach