On Thu, Feb 04, 2016 at 11:13:58AM +0100, Paolo Bonzini wrote: > On 04/02/2016 10:33, Roman Kagan wrote: > >> The bit that I was missing last week was how the guest could perceive > >> updates to the TSC page parameters as atomic. We actually _can_ > >> emulate seqlock-like behavior in Hyper-V by writing 0 to seq during > >> an update. Instead of looping like seqlocks do, the guest will simply > >> use the time reference counter and take a hypervisor exit. The result > >> however is still valid, because we want the time reference counter to > >> be perfectly synchronized with the Hyper-V clock and lets you handle > >> any kvmclock update scenario safely. > > > > That's smart, thanks for the idea! We'll only need to check that > > Windows goes through all these steps on every clock read, rather than > > remembering the failure and sticking with the MSR for good. > > Andrey pointed me to https://lkml.org/lkml/2015/11/2/655, which has the > source for Windows's clock read code. It should work. Indeed. (Shame on me for not remembering that as I participated in that analysis :) > I'll take a look at it if it's not super-high priority for you. Sure. Thanks, Roman. -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
