On Tue, Jun 26, 2012 at 10:06 AM, Marc Zyngier <marc.zyngier at arm.com> wrote: > On 26/06/12 14:48, Christoffer Dall wrote: >> On Tue, Jun 26, 2012 at 9:21 AM, Marc Zyngier <marc.zyngier at arm.com> wrote: >>> On 25/06/12 22:15, Christoffer Dall wrote: > >>>>> + ? ? ? cval = ((cycle_t)timer->cntv_cval_high << 32) | timer->cntv_cval_low; >>>>> + ? ? ? now = kvm_phys_timer_read() - vcpu->kvm->arch.timer.cntvoff; >>>> >>>> when is this adjusted and according to what? >>> >>> cntvoff is never adjusted. It is sampled exactly once when the VM is >>> created, and that's about it. And actually, the feature is disabled for >>> the time being. >>> >> >> ok, so the point is simply to say reasonably to the guest it's been >> running for X time, not counting time when the guest was not >> scheduled? > > No the point is actually to present the guest with an initial time > starting at 0. That's what I meant in my convoluted form. > I thought of using it to present some form of contiguous > time to the guest, but putting SMP into the equation hurts so much it's > not even funny. You'd end up with vcpus having a different view of time, > which may be correct from a relativist point of view, but not something > I wish to maintain... ;-) > right there with you;)
