Re: [Xen-devel] [Patch] x86/HVM: Fix RTC interrupt modelling

[Tim Deegan <tim@xxxxxxx>]

At 18:43 +0000 on 10 Feb (1392054204), Andrew Cooper wrote:
> REG_B and REG_C are to do with interrupts, and which events
> should(B)/have(C) generated interrupts.  The worst case is that a guest
> gets none/too-few/too-many interrupts when trying to drive the RTC. 
> None of this should lead to clock skew, as reading the time values
> directly will still provide the same information as before, although any
> guest which attempts to guess time based on counting periodic interrupts
> from the RTC is a) already broken and b) already having massive skew as
> a VM due to vcpu scheduling.

Sadly, not the case.  There absolutely are OSes that compute wallclock
time by counting timer ticks -- that's why we have the no-missed-ticks
mode.

Based on what these bugs have looked like in the past, the main risk
is that some kind of guest behaviour will cause clock skew or even
clock freeze in the guest.   In the past some of these have depended
on workload, e.g. something that changes the tick frequency. 

> XenRT does have tests for clock drift, but don't know for certain
> whether they have been run against the new code yet.
> 
> I will ensure they get run on v2 of the patch.

Excellent!

Tim.

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel