RE: [Xen-devel] [PATCH] X86: Prefer TSC-deadline timer in Xen

["Wei, Gang" <gang.wei@xxxxxxxxx>]

Architecturally the TSC Deadline feature comes along with reliable TSC.

For hot-add physical CPU case, if original APIC timer & TSC calibration work 
well, then TSC Deadline should not have problem.

Jimmy

> -----Original Message-----
> From: Dan Magenheimer [mailto:dan.magenheimer@xxxxxxxxxx]
> Sent: Friday, October 29, 2010 6:29 AM
> To: Wei, Gang; Keir Fraser; Tim Deegan
> Cc: Brown, Len; xen-devel@xxxxxxxxxxxxxxxxxxx; Jan Beulich
> Subject: RE: [Xen-devel] [PATCH] X86: Prefer TSC-deadline timer in Xen
> 
> The change to cstate_restore_tsc reminds me...
> 
> TSC should never be written if it is being used to
> generate system time, and I think cstate_restore_tsc
> is only used on systems where another clocksource
> is used to generate system time.   (And it generates
> a rather poor approximation of TSC.)  Are there any
> machines where cstate_restore_tsc might get executed
> but TSC_Deadline is in use?  If so, might there be
> a race condition here?  Also, are there any
> issues with using TSC Deadline on a system with
> the potential for hot-add physical CPUs?
> 
> I suspect cstate_restore_tsc doesn't get executed
> if TSC is reliable, and the TSC Deadline feature
> is probably only available on systems where TSC
> is reliable, but am not sure.
> 
> Thanks,
> Dan

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel