On 13/06/19 19:36, Jim Mattson wrote: >> Also, while this may vary dynamically based on the L2 guest that is >> running, this is much less true for unrestricted-guest processors. >> Without data on _which_ scenarios are bad for a static set of shadowed >> fields, I'm not really happy to add even more complexity. > > Data supporting which scenarios would lead you to entertain more > complexity? For example it would be interesting if some L1 hypervisor had 2x slower vmexits on some L2 guests, but otherwise fits the current set of shadowed fields. Paolo > Is it even worth collecting data on L3 performance, for > example? :-)
