Re: [PATCH] kvmclock: clock should count only if vm is running

[Marcelo Tosatti <mtosatti@redhat.com>]

On Tue, Jun 18, 2013 at 11:02:27AM +0200, Paolo Bonzini wrote:
> Hi Marcelo, sorry for the late review.
> 
> Il 08/06/2013 04:00, Marcelo Tosatti ha scritto:
> > kvmclock should not count while vm is paused, because:
> > 
> > 1) if the vm is paused for long periods, timekeeping 
> > math can overflow while converting the (large) clocksource 
> > delta to nanoseconds.
> > 
> > 2) Users rely on CLOCK_MONOTONIC to count run time, that is, 
> > time which OS has been in a runnable state (see CLOCK_BOOTTIME).
> 
> Do you have any ideas on how to implement CLOCK_BOOTTIME for kvmclock?
> I think we need to add more fields for the delta between CLOCK_MONOTONIC
> and CLOCK_BOOTTIME.

Unsure. An alternative to new fields is to use MSR_KVM_WALL_CLOCK
interface. Looking at two possibilities for catching up with real time
on unpause or vmload:

1) Catch up realtime via guest agent.
2) Catch up realtime via kvmclock interface.

But it would be good to have proper CLOCK_BOOTTIME behaviour at the same
time.

> > Change kvmclock driver so as to save clock value when vm transitions
> > from runnable to stopped state, and to restore clock value from stopped
> > to runnable transition.
> > 
> > Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
> > 
> > diff --git a/hw/i386/kvm/clock.c b/hw/i386/kvm/clock.c
> > index 87d4d0f..7d2d005 100644
> > --- a/hw/i386/kvm/clock.c
> > +++ b/hw/i386/kvm/clock.c
> > @@ -28,38 +28,6 @@ typedef struct KVMClockState {
> >      bool clock_valid;
> >  } KVMClockState;
> >  
> > -static void kvmclock_pre_save(void *opaque)
> > -{
> > -    KVMClockState *s = opaque;
> > -    struct kvm_clock_data data;
> > -    int ret;
> > -
> > -    if (s->clock_valid) {
> > -        return;
> > -    }
> > -    ret = kvm_vm_ioctl(kvm_state, KVM_GET_CLOCK, &data);
> > -    if (ret < 0) {
> > -        fprintf(stderr, "KVM_GET_CLOCK failed: %s\n", strerror(ret));
> > -        data.clock = 0;
> > -    }
> > -    s->clock = data.clock;
> > -    /*
> > -     * If the VM is stopped, declare the clock state valid to avoid re-reading
> > -     * it on next vmsave (which would return a different value). Will be reset
> > -     * when the VM is continued.
> > -     */
> > -    s->clock_valid = !runstate_is_running();
> > -}
> > -
> > -static int kvmclock_post_load(void *opaque, int version_id)
> > -{
> > -    KVMClockState *s = opaque;
> > -    struct kvm_clock_data data;
> > -
> > -    data.clock = s->clock;
> > -    data.flags = 0;
> > -    return kvm_vm_ioctl(kvm_state, KVM_SET_CLOCK, &data);
> > -}
> >  
> >  static void kvmclock_vm_state_change(void *opaque, int running,
> >                                       RunState state)
> > @@ -70,8 +38,18 @@ static void kvmclock_vm_state_change(void *opaque, int running,
> >      int ret;
> >  
> >      if (running) {
> > +        struct kvm_clock_data data;
> > +
> 
> Do we need an "if (!s->clock_valid) return;" here, or an assertion?  (Or
> alternatively, what happens if s->clock_valid == false?)


s->clock_valid = true means the clock value at s->clock (saved in
QEMU) is valid. 

At the moment VM begins execution, it becomes invalid (which means
synchronizing that value is necessary).

For the stopped->running transition, it has no meaning (just as the load
functions assume the device state is valid when they are called).

That said, i see no reason for additional checks.

> > +         */
> > +        s->clock_valid = !runstate_is_running();
> 
> Here we know that !runstate_is_running() is true (we're in the else
> branch of "if (running)") so it can just be "s->clock_valid = true".
> This matches the false assignment in the other branch.  This is just a
> nit, but it makes the code clearer.
> 
> Paolo

Sure, resending.