Re: #599161: Xen debug patch for the "clock shifts by 50 minutes" bug.

[Ian Campbell <ijc@hellion.org.uk>]

On Thu, 2012-11-08 at 12:54 +0000, Keir Fraser wrote:
> On 08/11/2012 11:43, "Ian Campbell" <ijc@hellion.org.uk> wrote:
> 
> >>> I'll leave this to Keir (who wrote the debugging patch) to answer but it
> >>> looks to me like it should be useful!
> >> 
> >> I'm scratching my head. plt_wrap is earlier than plt_now, which should be
> >> impossible.
> > 
> > impossible due to guarantees made by the h/w or by construction in Xen.
> 
> That's a question, right? 

Yes, sorry.

> By construction in Xen.
> 
> > There appears to be a certain amount of hardware-specificness to the
> > issue -- so I'm wondering if maybe there are some platforms whose tsc is
> > not as monotonically increasing as it needs to be...
> 
> plt_* timestamps are not derived from TSC at all.

I see, rather it is derived from the platform_timesource which could be
HPET, pmtimer, pit etc but in this case (according to the provided xm
dmesg) appears to be a 14MHz HPET.

So I guess s/tsc/HPET/ in my original thought...

> >>  plt_stamp64 oddly has low 32 bits identical to new_stamp. That
> >> seems very very improbable!
> > 
> > Does this code run on all cpus or just one? Is it always the same one?
> 
> Always cpu0.

So it's not cross cpu drift then.

Ian.