On 11/09/2018 15:55, Borislav Petkov wrote: > On Tue, Sep 11, 2018 at 01:07:06PM +0200, Paolo Bonzini wrote: >> If the host TSCs are unsynchronized then yes, that's what happens. And >> you can do live migration from synchronized to unsynchronized. > > Which brings us back to my original question: why would we *ever* want > to support unsynchronized TSCs in a guest? Such machines are a real > abomination for baremetal - it doesn't make *any* sense to me to have > that in guests too, if it can be helped... No, wait. The host TSC is unsynchronized, _so_ you need one kvmclock struct per vCPU. The resulting kvmclock is synchronized. Paolo
