[Qemu-devel] Re: [PATCH v2 14/17] kvm: x86: Introduce kvmclock device to save/restore its state

[Glauber Costa]

On Mon, 2011-01-03 at 17:30 +0100, Jan Kiszka wrote:
> Am 03.01.2011 17:04, Avi Kivity wrote:
> > On 01/03/2011 10:33 AM, Jan Kiszka wrote:
> >> From: Jan Kiszka<address@hidden>
> >>
> >> If kvmclock is used, which implies the kernel supports it, register a
> >> kvmclock device with the sysbus. Its main purpose is to save and restore
> >> the kernel state on migration, but this will also allow to visualize it
> >> one day.
> >>
> > 
> > kvmclock is a per-cpu affair.
> 
> Nope, it's state (the one save/restored here) is per VM.
> 
> > 
> >>
> >> @@ -534,6 +599,10 @@ int kvm_arch_init(int smp_cpus)
> >>       int ret;
> >>       struct utsname utsname;
> >>
> >> +#ifdef KVM_CAP_ADJUST_CLOCK
> >> +    sysbus_register_withprop(&kvmclock_info);
> >> +#endif
> >> +
> > 
> > So this doesn't look right.  I think we're fine with just migrating the
> > MSRs, like we migrate anything else that has to do with the cpu.
> > 
> 
> The kvmclock state is not contained in any MSR. It's an independent
> machine state that can be indirectly obtained via MSR access. Therefore,
> qemu-kvm currently registers only one vmstate entry per machine, and
> this patch just turns this into a clean device - because that's what
> kvmclock is in the end, something like an HPET.

Jan is right, the per-cpu information only cares about which piece of
memory will be written to.