Messages in this thread | | | From | Paul Durrant <> | Date | Tue, 31 Oct 2023 12:22:08 +0000 | Subject | Re: [PATCH] KVM: x86/xen: improve accuracy of Xen timers |
| |
On 31/10/2023 12:11, David Woodhouse wrote: > > > On 31 October 2023 12:06:17 GMT, Paul Durrant <xadimgnik@gmail.com> wrote: >> On 31/10/2023 11:42, David Woodhouse wrote: >>> Secondly, it's also wrong thing to do in the general case. Let's say KVM does its thing and snaps the kvmclock backwards in time on a KVM_REQ_CLOCK_UPDATE... do we really want to reinterpret existing timers against the new kvmclock? They were best left alone, I think. >> >> Do we not want to do exactly that? If the master clock is changed, why would we not want to re-interpret the guest's idea of time? That update will be visible to the guest when it re-reads the PV clock source. > > Well no, because the guest set that timer *before* we yanked the clock from under it, and probably wants it interpreted in the time scale which was in force at the time it was set. > > But more to the point, KVM shouldn't be doing that! We need to *fix* the kvmclock brokenness, not design further band-aids around it.
Ok, fair enough.
> > As I said, this patch stands even *after* we fix kvmclock, because it handles the timer delta calculation from an single TSC read. > > But overengineering a timer reset on KVM_REQ_CLOCK_UPDATE would not.
I'm not sure what you intend to do to kvmlock, so not sure whether we'll still need the __pvclock_read_cycles(&vcpu->arch.hv_clock, guest_tsc) but this patch (with the extra check on validity of hv_clock) does fix the drift so...
Reviewed-by: Paul Durrant <paul@xen.org>
| |