libvirt & virtio_net - host.freeze@reset.domain

libvirt & virtio_net - host.freeze@reset.domain

[poma]

Hello people,

libvirtd (libvirt) 1.0.5.2
virsh 1.0.5.2
virt-manager 0.10.0

Host:
Linux localhost 3.9.8-300.fc19.x86_64 #1 SMP Thu Jun 27 19:24:23 UTC
2013 x86_64 x86_64 x86_64 GNU/Linux
Guest1:
Linux localhost 3.9.8-300.fc19.i686.PAE #1 SMP Thu Jun 27 19:29:30 UTC
2013 i686 (none)
Guest2:
Linux localhost 3.9.8-300.fc19.x86_64 #1 SMP Thu Jun 27 19:24:23 UTC
2013 x86_64 x86_64 x86_64 GNU/Linux


Virtual NIC - source & model:
macvtap/NAT/bridge & virtio(virtio_net)

Host freeze at "virsh reset <domain>" or "virt-manager - Force Reset"
Need kernel.sysrq or power reset.


poma

Re: [libvirt-users] libvirt & virtio_net - host.freeze@reset.domain

[Daniel P. Berrange]

On Tue, Jul 02, 2013 at 01:25:21PM +0200, poma wrote:
> Hello people,
> 
> libvirtd (libvirt) 1.0.5.2
> virsh 1.0.5.2
> virt-manager 0.10.0
> 
> Host:
> Linux localhost 3.9.8-300.fc19.x86_64 #1 SMP Thu Jun 27 19:24:23 UTC
> 2013 x86_64 x86_64 x86_64 GNU/Linux
> Guest1:
> Linux localhost 3.9.8-300.fc19.i686.PAE #1 SMP Thu Jun 27 19:29:30 UTC
> 2013 i686 (none)
> Guest2:
> Linux localhost 3.9.8-300.fc19.x86_64 #1 SMP Thu Jun 27 19:24:23 UTC
> 2013 x86_64 x86_64 x86_64 GNU/Linux
> 
> 
> Virtual NIC - source & model:
> macvtap/NAT/bridge & virtio(virtio_net)
> 
> Host freeze at "virsh reset <domain>" or "virt-manager - Force Reset"
> Need kernel.sysrq or power reset.

I don't believe this is a libvirt issue - the 'virsh reset' command
will issue the 'system_reset' QEMU monitor command. This in turn
does an immediate reset of the guest CPUs/machine.

Even if QEMU is doing the wrong thing, the kernel should obviously
never freeze/crash in this way - it should be robust against a
malicious QEMU process.

You should probably send this message to the main QEMU and/or KVM
mailing lists so that it comes to the attention of people who are
more familiar with QEMU + virtio-net


Regards,
Daniel
-- 
|: http://berrange.com      -o-    http://www.flickr.com/photos/dberrange/ :|
|: http://libvirt.org              -o-             http://virt-manager.org :|
|: http://autobuild.org       -o-         http://search.cpan.org/~danberr/ :|
|: http://entangle-photo.org       -o-       http://live.gnome.org/gtk-vnc :|

Re: [libvirt-users] libvirt & virtio_net - host.freeze@reset.domain

[poma]

On 03.07.2013 13:43, Daniel P. Berrange wrote:
> On Tue, Jul 02, 2013 at 01:25:21PM +0200, poma wrote:
>> Hello people,
>>
>> libvirtd (libvirt) 1.0.5.2
>> virsh 1.0.5.2
>> virt-manager 0.10.0
>>
>> Host:
>> Linux localhost 3.9.8-300.fc19.x86_64 #1 SMP Thu Jun 27 19:24:23 UTC
>> 2013 x86_64 x86_64 x86_64 GNU/Linux
>> Guest1:
>> Linux localhost 3.9.8-300.fc19.i686.PAE #1 SMP Thu Jun 27 19:29:30 UTC
>> 2013 i686 (none)
>> Guest2:
>> Linux localhost 3.9.8-300.fc19.x86_64 #1 SMP Thu Jun 27 19:24:23 UTC
>> 2013 x86_64 x86_64 x86_64 GNU/Linux
>>
>>
>> Virtual NIC - source & model:
>> macvtap/NAT/bridge & virtio(virtio_net)
>>
>> Host freeze at "virsh reset <domain>" or "virt-manager - Force Reset"
>> Need kernel.sysrq or power reset.
> 
> I don't believe this is a libvirt issue - the 'virsh reset' command
> will issue the 'system_reset' QEMU monitor command. This in turn
> does an immediate reset of the guest CPUs/machine.
> 
> Even if QEMU is doing the wrong thing, the kernel should obviously
> never freeze/crash in this way - it should be robust against a
> malicious QEMU process.
> 
> You should probably send this message to the main QEMU and/or KVM
> mailing lists so that it comes to the attention of people who are
> more familiar with QEMU + virtio-net
> 
> 
> Regards,
> Daniel
> 

Thanks for your response.
Mateusz hit the same issue[1] as well.
OK, here we go.


poma


[1] https://lists.fedoraproject.org/pipermail/users/2013-July/436984.html

Re: [libvirt-users] libvirt & virtio_net - host.freeze@reset.domain

[poma]

On 04.07.2013 11:14, poma wrote:
> On 03.07.2013 13:43, Daniel P. Berrange wrote:
>> On Tue, Jul 02, 2013 at 01:25:21PM +0200, poma wrote:
>>> Hello people,
>>>
>>> libvirtd (libvirt) 1.0.5.2
>>> virsh 1.0.5.2
>>> virt-manager 0.10.0
>>>
>>> Host:
>>> Linux localhost 3.9.8-300.fc19.x86_64 #1 SMP Thu Jun 27 19:24:23 UTC
>>> 2013 x86_64 x86_64 x86_64 GNU/Linux
>>> Guest1:
>>> Linux localhost 3.9.8-300.fc19.i686.PAE #1 SMP Thu Jun 27 19:29:30 UTC
>>> 2013 i686 (none)
>>> Guest2:
>>> Linux localhost 3.9.8-300.fc19.x86_64 #1 SMP Thu Jun 27 19:24:23 UTC
>>> 2013 x86_64 x86_64 x86_64 GNU/Linux
>>>
>>>
>>> Virtual NIC - source & model:
>>> macvtap/NAT/bridge & virtio(virtio_net)
>>>
>>> Host freeze at "virsh reset <domain>" or "virt-manager - Force Reset"
>>> Need kernel.sysrq or power reset.
>>
>> I don't believe this is a libvirt issue - the 'virsh reset' command
>> will issue the 'system_reset' QEMU monitor command. This in turn
>> does an immediate reset of the guest CPUs/machine.
>>
>> Even if QEMU is doing the wrong thing, the kernel should obviously
>> never freeze/crash in this way - it should be robust against a
>> malicious QEMU process.
>>
>> You should probably send this message to the main QEMU and/or KVM
>> mailing lists so that it comes to the attention of people who are
>> more familiar with QEMU + virtio-net
>>
>>
>> Regards,
>> Daniel
>>
> 
> Thanks for your response.
> Mateusz hit the same issue[1] as well.
> OK, here we go.
> 
> 
> poma
> 
> 
> [1] https://lists.fedoraproject.org/pipermail/users/2013-July/436984.html

OK, is this a side effect or not, but certainly kernel[1] with
'bridge-timer-fix.patch'[2] resolves issue aforementioned, so far.
Thanks Cong, Josh.


poma


[1] http://koji.fedoraproject.org/koji/taskinfo?taskID=5569632
    kernel-3.9.8-300.7.fc19.x86_64.rpm
[2] https://bugzilla.redhat.com/show_bug.cgi?id=880035#c53

Ref.
    "fix for unreliable guest->host multicast triggers oops"
    https://bugzilla.redhat.com/show_bug.cgi?id=980254