Xen devel - Mar 2009 - [PATCH] linux-2.6.18-xen.hg: fix interaction of NTP and dom0->xen time updates

Don''t discard NTP sync when updating Xen wallclock time from dom0,
as that''s almost the first thing we do when we become synced.
Move the call to ntp_clear() into do_settimeofday(), which is the
only caller of __update_wallclock() that looks like it should break
NTP sync.  

This fixes the timer chain that sets Xen''s wallclock every minute when
dom0 is NTP synced, which in turn greatly improves wallclock accuracy in
PV domU.

Signed-off-by: Tim Deegan <Tim.Deegan@citrix.com>

-- 
Tim Deegan <Tim.Deegan@citrix.com>
Principal Software Engineer, Citrix Systems (R&D) Ltd.
[Company #02300071, SL9 0DZ, UK.]


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

Hi Tim --

Cool!

Just checking... this provides improvement when:

1) the patch is applied to dom0
2) dom0 is running NTP
3) pv domU has independent_wallclock=0
4) pv domU is NOT running NTP

Correct?  Does it also help when:

1) the patch is applied to pv domU
2) pv domU has independent_wallclock=1
3) pv domU IS running NTP

Are there any other configurations where it will
help?  Are there any configurations where it might
cause problems?

Thanks,
Dan
> -----Original Message-----
> From: Tim Deegan [mailto:Tim.Deegan@citrix.com]
> Sent: Wednesday, March 18, 2009 5:14 AM
> To: xen-devel@lists.xensource.com
> Subject: [Xen-devel] [PATCH] linux-2.6.18-xen.hg: fix 
> interaction of NTP
> and dom0->xen time updates
> 
> 
> Don''t discard NTP sync when updating Xen wallclock time from dom0,
> as that''s almost the first thing we do when we become synced.
> Move the call to ntp_clear() into do_settimeofday(), which is the
> only caller of __update_wallclock() that looks like it should break
> NTP sync.  
> 
> This fixes the timer chain that sets Xen''s wallclock every minute
when
> dom0 is NTP synced, which in turn greatly improves wallclock 
> accuracy in
> PV domU.
> 
> Signed-off-by: Tim Deegan <Tim.Deegan@citrix.com>
> 
> -- 
> Tim Deegan <Tim.Deegan@citrix.com>
> Principal Software Engineer, Citrix Systems (R&D) Ltd.
> [Company #02300071, SL9 0DZ, UK.]
>
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

At 14:31 +0000 on 18 Mar (1237386684), Dan Magenheimer
wrote:
> Just checking... this provides improvement when:
> 
> 1) the patch is applied to dom0
> 2) dom0 is running NTP
> 3) pv domU has independent_wallclock=0
> 4) pv domU is NOT running NTP
Yes, exactly that.  Xen''s wallclock tends to sawtooth, drifting with
the
underlying stime and being pulled back whenever dom0 updates it.
Without this patch, it pulls back only when ntpd in dom0 calls adjtime
(allowing for up to 100ms of drift on my test machine).  With the patch
it updates every minute, as intended (still sawtoothing but not as
wildly).
> Correct?  Does it also help when:
> 
> 1) the patch is applied to pv domU
> 2) pv domU has independent_wallclock=1
> 3) pv domU IS running NTP
Not sure, but I would guess not -- with independent_wallclock==1, domU
shuold be ignoring the wc_sec and wc_nsec values provided by Xen, which
are the only things affected by this patch.
> Are there any other configurations where it will
> help?  Are there any configurations where it might
> cause problems?
I don''t think so, to either question.  If there''s some bad
condition
that is triggered by the act of pulling time back in sync rather than
the size of the correction, it might be more frequent with the more
frequent updates.  Otherwise, it''s just making an existing sawtooth a
bit less blatant.

Tim. 

-- 
Tim Deegan <Tim.Deegan@citrix.com>
Principal Software Engineer, Citrix Systems (R&D) Ltd.
[Company #02300071, SL9 0DZ, UK.]

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