On Tue, Oct 03, 2023, David Woodhouse wrote: > On Mon, 2023-10-02 at 17:53 -0700, Sean Christopherson wrote: > > > > This is *very* lightly tested, as in it compiles and doesn't explode, but that's > > about all I've tested. > > I don't think it's working, if I understand what it's supposed to be > doing. ... > When I run xen_shinfo_test, the kvmclock still drifts from the > "monotonic_raw" I get from kvm_get_time_and_clockread(), even with your > patch. It "works", in that it does what I intended it to do. What I missed, and what's obvious in hindsight, is that the timekeeping code doesn't *just* update what KVM uses as the "base" when there "big events" like suspend/resume, the timekeeping code updates the base clock on every tick. My hack only smoothed away the delta, it didn't address the time that had gotten moved into the base. I though the base would be stable if the host was in a steady state, but see above...
