Linux Virtualization - Feb 2017 - [PATCH 2/2] x86/vdso: Add VCLOCK_HVCLOCK vDSO clock read method

> From: Thomas Gleixner [mailto:tglx at linutronix.de]
> Sent: Saturday, February 11, 2017 02:02
>  ...
> That's important if the stuff happens cross CPU. If the update happens
on
> the same CPU then this is a different story and as there are VMexits
> involved they might provide the required ordering already. But I can't
tell
> as I have no idea how that host side thing is done.
> 
> 	tglx
IMO Hyper-V TSC page clocksource here seems pretty similar to KVM's pvclock,
So I would guess "the structure is only updated just before reentering the
guest
after some VM event" (https://rwmj.wordpress.com/2010/10/15/kvm-pvclock/),
that is, the update should happen on the same CPU, I guess.

Thanks,
-- Dexuan

On Mon, 13 Feb 2017, Dexuan Cui wrote:
> > From: Thomas Gleixner [mailto:tglx at linutronix.de]
> > Sent: Saturday, February 11, 2017 02:02
> >  ...
> > That's important if the stuff happens cross CPU. If the update
happens on
> > the same CPU then this is a different story and as there are VMexits
> > involved they might provide the required ordering already. But I
can't tell
> > as I have no idea how that host side thing is done.
> > 
> 
> IMO Hyper-V TSC page clocksource here seems pretty similar to KVM's
pvclock,
> So I would guess "the structure is only updated just before reentering
the guest
> after some VM event"
(https://rwmj.wordpress.com/2010/10/15/kvm-pvclock/),
> that is, the update should happen on the same CPU, I guess.
Guessing does not help much. There are a gazillion ways to implement such a
thing. I'm sure that there exists proper documentation of the mechanism
inside your company.

Thanks,

	tglx

On Sun, Feb 12, 2017 at 11:49 PM, Dexuan Cui <decui at microsoft.com>
wrote:
>> From: Thomas Gleixner [mailto:tglx at linutronix.de]
>> Sent: Saturday, February 11, 2017 02:02
>>  ...
>> That's important if the stuff happens cross CPU. If the update
happens on
>> the same CPU then this is a different story and as there are VMexits
>> involved they might provide the required ordering already. But I
can't tell
>> as I have no idea how that host side thing is done.
>>
>>       tglx
>
> IMO Hyper-V TSC page clocksource here seems pretty similar to KVM's
pvclock,
> So I would guess "the structure is only updated just before reentering
the guest
> after some VM event"
(https://rwmj.wordpress.com/2010/10/15/kvm-pvclock/),
> that is, the update should happen on the same CPU, I guess.
If the patch is correct, there is one of these shared by all vCPUs, so
this is not a sufficient explanation.

On Mon, 13 Feb 2017, Andy Lutomirski wrote:
> On Sun, Feb 12, 2017 at 11:49 PM, Dexuan Cui <decui at microsoft.com>
wrote:
> >> From: Thomas Gleixner [mailto:tglx at linutronix.de]
> >> Sent: Saturday, February 11, 2017 02:02
> >>  ...
> >> That's important if the stuff happens cross CPU. If the update
happens on
> >> the same CPU then this is a different story and as there are
VMexits
> >> involved they might provide the required ordering already. But I
can't tell
> >> as I have no idea how that host side thing is done.
> >>
> >>       tglx
> >
> > IMO Hyper-V TSC page clocksource here seems pretty similar to
KVM's pvclock,
> > So I would guess "the structure is only updated just before
reentering the guest
> > after some VM event"
(https://rwmj.wordpress.com/2010/10/15/kvm-pvclock/),
> > that is, the update should happen on the same CPU, I guess.
> 
> If the patch is correct, there is one of these shared by all vCPUs, so
> this is not a sufficient explanation.
Right, because that's ony ONE TSC page for the whole guest and then the seq
stuff really lacks barriers.

Thanks,

	tglx