Bug#617421: xen-hypervisor-4.0-i386: can't migrate (or save/resume) lenny guests: BUG: recent printk recursion
Ferenc Wagner
Version: 4.0.1-2
Severity: normal
Hi,
Take the following config file (standard lenny kernel and initrd, with
some customization for iSCSI root):
name = "lenny3"
kernel = "vmlinuz-2.6.26-2-686-bigmem"
ramdisk = "initrd.img-2.6.26-2-686-bigmem"
memory = 100
vcpus = 1
vif = [ 'mac=00:16:3e:00:00:17, bridge=br695, vifname=lenny3.alma-g' ]
extra = "break=top"
on_poweroff = 'destroy'
on_reboot = 'restart'
on_crash = 'restart'
Start the domain:
xm create -c lenny3.conf
[...]
(initramfs)
From another terminal, save the guest:
xm save lenny3 /tmp/lenny3.save
The guest console says: Suspending xenbus... and detaches.
Restore the guest (paused to attach its console before continuing):
xm restore -p /tmp/lenny3.save
xm console lenny3
xm unpause lenny3
The console starts spewing out "BUG: recent printk recursion" and the
domain spins until destroyed. The same thing happens on (live or not)
migration. The very same domains could be migrated without issues under
Xen 3.2 (more precisely, before I upgraded my dom0 to squeeze --
previously it was running lenny with the etch dom0 kernel for stability).
Migrating squeeze guests isn't a problem in the same setup (just replace
26-2 with 32-5 in the above config).
This looks like a regression. On the other hand, it doesn't seem to
affect a lot of people, because I couldn't find anything relevant on the
web. Please tell me if I can help debugging the issue further.
--
Thanks,
Feri.
-- System Information:
Debian Release: 6.0
APT prefers stable
APT policy: (500, 'stable')
Architecture: i386 (i686)
Kernel: Linux 2.6.32-5-xen-686 (SMP w/4 CPU cores)
Locale: LANG=en_US.UTF-8, LC_CTYPE=en_US.UTF-8 (charmap=UTF-8)
Shell: /bin/sh linked to /bin/dash
xen-hypervisor-4.0-i386 depends on no packages.
Versions of packages xen-hypervisor-4.0-i386 recommends:
ii xen-utils-4.0 4.0.1-2 XEN administrative tools
Versions of packages xen-hypervisor-4.0-i386 suggests:
pn xen-docs-4.0 <none> (no description available)
-- no debconf information
--
To UNSUBSCRIBE, email to debian-bugs-...@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listm...@lists.debian.org
Thomas Goirand
Kernel 2.6.26 are designed for Xen 3, and 2.6.32 are for Xen 4. I'm not
really surprised with the result you have here, and I don't think you
can expect to have it fixed. Why don't you simply use kernel 2.6.32 for
your Lenny VMs? It does work well, even with a Lenny userland.
Thomas
Ian Campbell
> Kernel 2.6.26 are designed for Xen 3, and 2.6.32 are for Xen 4.
This is not at all true. There is no relationship of this sort between
Xen releases and guest kernel versions.
The guest visible hypervisor ABI has been stable since (IIRC) Xen 3.0.0
and Xen upstream maintains backwards compatibility for guest visible
interfaces -- if an old guest was broken by a hypervisor update then
this is most likely a bug.
On the other hand, the Xen enabled kernel in Lenny turned out to be a
bit of a disaster and I could well imagine that the new hypervisor/tools
are simply exposing a prexisting bug in that kernel. Due to that and the
fact that Lenny is now oldstable you may find there is not much interest
in tracking this issue down.
Ferenc should definitely make sure you are running the most recent Lenny
kernel though, the bug report doesn't include the version of this
package AFAICT. Failing that running a Squeeze kernel from the
lenny-backports suite is a good option IMHO.
Ian
--
Ian Campbell
You're being followed. Cut out the hanky-panky for a few days.
Ferenc Wagner
> The guest visible hypervisor ABI has been stable since (IIRC) Xen 3.0.0
> and Xen upstream maintains backwards compatibility for guest visible
> interfaces -- if an old guest was broken by a hypervisor update then
> this is most likely a bug.
Agreed.
> On the other hand, the Xen enabled kernel in Lenny turned out to be a
> bit of a disaster and I could well imagine that the new hypervisor/tools
> are simply exposing a prexisting bug in that kernel.
It wasn't entirely explicit, but I don't use (and haven't ever used) the
Xen-enabled lenny kernel: it was way too unstable for production use
from the very beginning. The guests I'm talking about run the latest
linux-image-2.6.26-2-686-bigmem kernel (version 2.6.26-26lenny2),
sporting upstream Xen domU support, and no dom0 support. The dom0 was
running the "latest" etch Xen kernel, linux-image-2.6.18-6-xen-686
(version 2.6.18.dfsg.1-26etch2).
> Due to that and the fact that Lenny is now oldstable you may find
> there is not much interest in tracking this issue down.
The latter part of course still applies, and I can accept that. This
report can then serve documentation purposes at least. :)
> Ferenc should definitely make sure you are running the most recent Lenny
> kernel though, the bug report doesn't include the version of this
> package AFAICT.
Sorry, I left that out indeed. It's the latest lenny kernel (Jan 27).
> Failing that running a Squeeze kernel from the lenny-backports suite
> is a good option IMHO.
On some machines it worked fine, on others it broke the iSCSI root setup:
[ 1.267399] iscsi: registered transport (tcp)
iscsistart: transport class version 2.0-870. iscsid version 2.0-870
iscsistart: Logging into **************************************** *************************
iscsistart: can not connect to iSCSI daemon (111)!
iscsistart: initiator reported error (18 - could not communicate to iscsid)
Which is rather strange, because the point of iscsistart is that it does
not require iscsid... Anyway, it's not Xen-related, and I'll hopefully
find a way out.
--
Thanks,
Feri.
Ian Campbell
> Ian Campbell <i...@hellion.org.uk> writes:
>
> > The guest visible hypervisor ABI has been stable since (IIRC) Xen 3.0.0
> > and Xen upstream maintains backwards compatibility for guest visible
> > interfaces -- if an old guest was broken by a hypervisor update then
> > this is most likely a bug.
>
> Agreed.
>
> > On the other hand, the Xen enabled kernel in Lenny turned out to be a
> > bit of a disaster and I could well imagine that the new hypervisor/tools
> > are simply exposing a prexisting bug in that kernel.
>
> It wasn't entirely explicit, but I don't use (and haven't ever used) the
> Xen-enabled lenny kernel: it was way too unstable for production use
> from the very beginning. The guests I'm talking about run the latest
> linux-image-2.6.26-2-686-bigmem kernel (version 2.6.26-26lenny2),
> sporting upstream Xen domU support, and no dom0 support.
Did pvops in 2.6.26 support migration? I must admit I thought it didn't
yet.
Ian.
--
Ian Campbell
Eat drink and be merry, for tomorrow we diet.
Ferenc Wagner
> On Wed, 2011-03-09 at 11:51 +0100, Ferenc Wagner wrote:
>
>> Ian Campbell <i...@hellion.org.uk> writes:
>>
>>> On the other hand, the Xen enabled kernel in Lenny turned out to be a
>>> bit of a disaster and I could well imagine that the new hypervisor/tools
>>> are simply exposing a prexisting bug in that kernel.
>>
>> It wasn't entirely explicit, but I don't use (and haven't ever used) the
>> Xen-enabled lenny kernel: it was way too unstable for production use
>> from the very beginning. The guests I'm talking about run the latest
>> linux-image-2.6.26-2-686-bigmem kernel (version 2.6.26-26lenny2),
>> sporting upstream Xen domU support, and no dom0 support.
>
> Did pvops in 2.6.26 support migration? I must admit I thought it didn't
> yet.
It apparently did... but it does not anymore. :(
--
Cheers,
Feri.
Ian Campbell
> Ian Campbell <i...@hellion.org.uk> writes:
>
> > On Wed, 2011-03-09 at 11:51 +0100, Ferenc Wagner wrote:
> >
> >> Ian Campbell <i...@hellion.org.uk> writes:
> >>
> >>> On the other hand, the Xen enabled kernel in Lenny turned out to be a
> >>> bit of a disaster and I could well imagine that the new hypervisor/tools
> >>> are simply exposing a prexisting bug in that kernel.
> >>
> >> It wasn't entirely explicit, but I don't use (and haven't ever used) the
> >> Xen-enabled lenny kernel: it was way too unstable for production use
> >> from the very beginning. The guests I'm talking about run the latest
> >> linux-image-2.6.26-2-686-bigmem kernel (version 2.6.26-26lenny2),
> >> sporting upstream Xen domU support, and no dom0 support.
> >
> > Did pvops in 2.6.26 support migration? I must admit I thought it didn't
> > yet.
>
> It apparently did... but it does not anymore. :(
I can confirm that suspend/restore/migrate support was not added to the
mainline kernel until commit 0e91398f2a5d which first appeared in
2.6.27-rc1.
Not sure what you were testing before but it wasn't 2.6.26 ;-)
Ian.
--
Ian Campbell
Current Noise: Heaven & Hell - Rock And Roll Angel
Good day to deal with people in high places; particularly lonely stewardesses.
Ferenc Wagner
> On Wed, 2011-03-09 at 13:01 +0100, Ferenc Wagner wrote:
>
>> Ian Campbell <i...@hellion.org.uk> writes:
>>
>>> On Wed, 2011-03-09 at 11:51 +0100, Ferenc Wagner wrote:
>>>
>>>> Ian Campbell <i...@hellion.org.uk> writes:
>>>>
>>>>> On the other hand, the Xen enabled kernel in Lenny turned out to be a
>>>>> bit of a disaster and I could well imagine that the new hypervisor/tools
>>>>> are simply exposing a prexisting bug in that kernel.
>>>>
>>>> It wasn't entirely explicit, but I don't use (and haven't ever used) the
>>>> Xen-enabled lenny kernel: it was way too unstable for production use
>>>> from the very beginning. The guests I'm talking about run the latest
>>>> linux-image-2.6.26-2-686-bigmem kernel (version 2.6.26-26lenny2),
>>>> sporting upstream Xen domU support, and no dom0 support.
>>>
>>> Did pvops in 2.6.26 support migration? I must admit I thought it didn't
>>> yet.
>>
>> It apparently did... but it does not anymore. :(
>
> I can confirm that suspend/restore/migrate support was not added to the
> mainline kernel until commit 0e91398f2a5d which first appeared in
> 2.6.27-rc1.
>
> Not sure what you were testing before but it wasn't 2.6.26 ;-)
Didn't you port it to the lenny kernel? I'm starting to remember...
--
Cheers,
Feri.
Ian Campbell
> Ian Campbell <i...@hellion.org.uk> writes:
>
> > On Wed, 2011-03-09 at 13:01 +0100, Ferenc Wagner wrote:
> >
> >> Ian Campbell <i...@hellion.org.uk> writes:
> >>
> >>> On Wed, 2011-03-09 at 11:51 +0100, Ferenc Wagner wrote:
> >>>
> >>>> Ian Campbell <i...@hellion.org.uk> writes:
> >>>>
> >>>>> On the other hand, the Xen enabled kernel in Lenny turned out to be a
> >>>>> bit of a disaster and I could well imagine that the new hypervisor/tools
> >>>>> are simply exposing a prexisting bug in that kernel.
> >>>>
> >>>> It wasn't entirely explicit, but I don't use (and haven't ever used) the
> >>>> Xen-enabled lenny kernel: it was way too unstable for production use
> >>>> from the very beginning. The guests I'm talking about run the latest
> >>>> linux-image-2.6.26-2-686-bigmem kernel (version 2.6.26-26lenny2),
> >>>> sporting upstream Xen domU support, and no dom0 support.
> >>>
> >>> Did pvops in 2.6.26 support migration? I must admit I thought it didn't
> >>> yet.
> >>
> >> It apparently did... but it does not anymore. :(
> >
> > I can confirm that suspend/restore/migrate support was not added to the
> > mainline kernel until commit 0e91398f2a5d which first appeared in
> > 2.6.27-rc1.
> >
> > Not sure what you were testing before but it wasn't 2.6.26 ;-)
>
> Didn't you port it to the lenny kernel? I'm starting to remember...
Hrm, debian/patches/features/all/xen/tip-x86.patch seems to suggest
someone did (and I moved the patch so I guess I knew about it at one
point).
Ian.
--
Ian Campbell
Current Noise: Heaven & Hell - Breaking Into Heaven
I want to kill everyone here with a cute colorful Hydrogen Bomb!!