Xen devel - Apr 2013 - Occasional clock/time problems on HVM domU after live migrate on Xen 4.2.{1, 2}

I am seeing occasional clock/time problems on an HVM domU with upstream 
qemu-dm on Xen 4.2 (more precisely 4.2.2-rc1 with my live migrate
patches).

I''m migrating between two machines with identical hardware.
I''m using an
Ubuntu Precise dom0 and domU. The machines run ntp in dom0 for what its
worth, but their RTCs may not be synced.

When I migrate a machine in one direction it very frequently works. However,
sometimes, executing a ping command on the migrated machine appears to fail.
Close inspection suggests the networking works (I can see traffic going in
and out), but the machine''s idea of time is all screwed up so ping gets
confused into sending nothing. It''s as if time is not moving forward on
the migrated machine.

Possibly for a similar reason, the paravirtualised domU vNIC which is
attached to a bridge in dom0 sometimes comes up ''NO CARRIER''
for no
discernable reason. It appears to be a member of the relevant bridge
according to brctl. I''m not sure how the pv nic detects the presence
of ''carrier''.

Any ideas?

-- 
Alex Bligh

On Sat, 2013-04-27 at 18:11 +0100, Alex Bligh wrote:
> I am seeing occasional clock/time problems on an HVM domU with upstream 
> qemu-dm on Xen 4.2 (more precisely 4.2.2-rc1 with my live migrate
> patches).
Does this work with Xen unstable? (IOW can we isolate this to an
upstream bug or a backporting bug/lack of a fix)
> I''m migrating between two machines with identical hardware.
I''m using an
> Ubuntu Precise dom0 and domU. The machines run ntp in dom0 for what its
> worth, but their RTCs may not be synced.
> 
> When I migrate a machine in one direction it very frequently works.
However,
> sometimes, executing a ping command on the migrated machine appears to
fail.
> Close inspection suggests the networking works (I can see traffic going in
> and out), but the machine''s idea of time is all screwed up so ping
gets
> confused into sending nothing. It''s as if time is not moving
forward on
> the migrated machine.
> 
> Possibly for a similar reason, the paravirtualised domU vNIC which is
> attached to a bridge in dom0 sometimes comes up ''NO
CARRIER'' for no
> discernable reason. It appears to be a member of the relevant bridge
> according to brctl. I''m not sure how the pv nic detects the
presence
> of ''carrier''.
> 
> Any ideas?
>

--On 29 April 2013 09:36:34 +0100 Ian Campbell <Ian.Campbell@citrix.com> 
wrote:
> On Sat, 2013-04-27 at 18:11 +0100, Alex Bligh wrote:
>> I am seeing occasional clock/time problems on an HVM domU with upstream
>> qemu-dm on Xen 4.2 (more precisely 4.2.2-rc1 with my live migrate
>> patches).
>
> Does this work with Xen unstable? (IOW can we isolate this to an
> upstream bug or a backporting bug/lack of a fix)
That''s a good question. And given this appears /more/ reproducible
(for once) with xl command line than with our own code, it should
be easy to find out.

I''ll also get this tested against release 4.2.2.

-- 
Alex Bligh

Ian,

--On 29 April 2013 10:54:38 +0000 Alex Bligh <alex@alex.org.uk> wrote:
>> Does this work with Xen unstable? (IOW can we isolate this to an
>> upstream bug or a backporting bug/lack of a fix)
>
> That''s a good question. And given this appears /more/ reproducible
> (for once) with xl command line than with our own code, it should
> be easy to find out.
>
> I''ll also get this tested against release 4.2.2.
Sorry about the extended delay. My colleague Diana Crisan has
replicated this problem on 4.3-rc1 and release 4.2.2, and has sent
three emails to this list describing three related problems
more fully than I did. We think it''s something to do with ACPI
handling.

The subject lines are:
* HVM Migration of domU on Qemu-upstream DM causes stuck system clock with 
ACPI
* HVM Migration of domU on Qemu-upstream DM loses ACPI data in xenstore
* Early ACPI events prevent subsequent ACPI functionality on xen 4.3 + HVM 
domU

-- 
Alex Bligh