[Users] Nested virt failing
I have Fedora 18 with ovirt running. I have vdsm-hook-nestedvt.noarch installed on all the nodes. The nodes are all identical with Intel i5-2400 CPUs (SandyBridge). When I try to boot a nested virtual machine (L2) instead of a ovirt virtual machine (L1), the kernel starts to load, and then that nested virtual machines (L2) pauses. On the host running Fedora 18 dmesg shows this message: nested_vmx_run: VMCS MSR_{LOAD,STORE} unsupported Reading this bug https://bugzilla.redhat.com/show_bug.cgi?id=892240 It looks like if the Fedora 18 ovirt nodes started the the L1 using the "core2duo" CPU flag instead of host pass through it might work? How do I do that, or are there other solutions? TIA, Dax Kelson
Dax Kelson wrote:
I have Fedora 18 with ovirt running. I have vdsm-hook-nestedvt.noarch installed on all the nodes.
The nodes are all identical with Intel i5-2400 CPUs (SandyBridge).
When I try to boot a nested virtual machine (L2) instead of a ovirt virtual machine (L1), the kernel starts to load, and then that nested virtual machines (L2) pauses.
On the host running Fedora 18 dmesg shows this message:
nested_vmx_run: VMCS MSR_{LOAD,STORE} unsupported
Reading this bug https://bugzilla.redhat.com/show_bug.cgi?id=892240
It looks like if the Fedora 18 ovirt nodes started the the L1 using the "core2duo" CPU flag instead of host pass through it might work?
How do I do that, or are there other solutions?
Are you sure nested is enabled? Check: cat /sys/module/kvm_intel/parameters/nested should say Y Check your cluster for what cpu is used. I have a test system where a core i3 isn't recognised as a sandy bridge but a core i7 is. Have tried fiddling with libvirts xml file but no go. Regards, Joop
Hi Dax, https://bugzilla.redhat.com/922075 -- Pavel On Thu, 06 Jun 2013, Dax Kelson wrote:
I have Fedora 18 with ovirt running. I have vdsm-hook-nestedvt.noarch installed on all the nodes.
The nodes are all identical with Intel i5-2400 CPUs (SandyBridge).
When I try to boot a nested virtual machine (L2) instead of a ovirt virtual machine (L1), the kernel starts to load, and then that nested virtual machines (L2) pauses.
On the host running Fedora 18 dmesg shows this message:
nested_vmx_run: VMCS MSR_{LOAD,STORE} unsupported
Reading this bug https://bugzilla.redhat.com/show_bug.cgi?id=892240
It looks like if the Fedora 18 ovirt nodes started the the L1 using the "core2duo" CPU flag instead of host pass through it might work?
How do I do that, or are there other solutions?
TIA, Dax Kelson _______________________________________________ Users mailing list Users@ovirt.org http://lists.ovirt.org/mailman/listinfo/users
On Thu, Jun 6, 2013 at 1:27 PM, Pavel Zhukov <pavel@zhukoff.net> wrote:
Hi Dax, https://bugzilla.redhat.com/922075
-- Pavel
Thanks for the pointer? Any workarounds? How can I fiddle on what -cpu flags are passed? Does it require creating a vdsm hook?
On 06/07/2013 12:56 AM, Dax Kelson wrote:
On Thu, Jun 6, 2013 at 1:27 PM, Pavel Zhukov <pavel@zhukoff.net> wrote:
Hi Dax, https://bugzilla.redhat.com/922075
-- Pavel
Thanks for the pointer?
Any workarounds? How can I fiddle on what -cpu flags are passed? Does it require creating a vdsm hook? _______________________________________________ Users mailing list Users@ovirt.org http://lists.ovirt.org/mailman/listinfo/users
you can change the flags at engine config (restart engine), but a hook is less hacky than changing this for all VMs in all clusters via the engine.
On Fri, Jun 7, 2013 at 6:55 AM, Itamar Heim <iheim@redhat.com> wrote:
you can change the flags at engine config (restart engine), but a hook is less hacky than changing this for all VMs in all clusters via the engine.
I'm OK with hacky. I poked around the documentation and didn't see anything mentioning how to do this. Do you have a pointer or more info? Thanks Dax
On 06/07/2013 08:57 PM, Dax Kelson wrote:
On Fri, Jun 7, 2013 at 6:55 AM, Itamar Heim <iheim@redhat.com <mailto:iheim@redhat.com>> wrote:
you can change the flags at engine config (restart engine), but a hook is less hacky than changing this for all VMs in all clusters via the engine.
I'm OK with hacky. I poked around the documentation and didn't see anything mentioning how to do this. Do you have a pointer or more info?
Thanks Dax
look at the ServerCPUList config value. you can change the cpu model and flags sent to the qemu (for all VMs)
On Sun, Jun 9, 2013 at 4:02 AM, Itamar Heim <iheim@redhat.com> wrote:
look at the ServerCPUList config value. you can change the cpu model and flags sent to the qemu (for all VMs)
Thanks! In the web interface (in both L1 and L2) I edited the Cluster properties and set the "CPU Name" to "Intel Conroe Family". Note my actual CPUs are SandyBridge family. This workaround seems be successful, nested virt is now working, no freezes/pauses. The next issue I ran into was that both layers were using SHIFT+F12 for the Spice release key causing difficulties. This fix for that was to change the release key for one of the layers, using the following commands: engine-config -s SpiceReleaseCursorkeys='f9' service ovirt-engine-restart
participants (4)
-
Dax Kelson -
Itamar Heim -
Joop -
Pavel Zhukov