[Users] fedora qemu modern CPU recognition issues cause ovirt manged host to go non-operational

[Ayal Baron]

Sounds like vdsm should just require a newer version of qemu-kvm as well?
Have you tested with a newer qemu-kvm?

----- Original Message -----
> 
> Seeing an issue wherein ovirt moves a managed host to non-operational
> state.
> This occurs with the currently released version of ovirt and the
> latest development builds.
> The ovirt host is loaded with Fedora core 16 and equipped with most
> current development version of the vdsm.
> *Editorial node*
> The latest vdsm to work on the FC16 host required building and adding
> newer versions of the sanlock, libvirt, lvm2 and device-mapper
> packages than what FC16 provides.
> Ultimately however none of the newer packages have any bearing on
> this failure mode.
> 
> The failure mode is as follows.
> Upon successfully adding the host and setting the cluster CPU
> compatibility level oVirt will offline the host with the following
> message:
> --> Host ovirtnode moved to Non-Operational state as host does not
> meet the cluster's minimum CPU level. Missing CPU features :
> model_Nehalem
> 
> Under the hood the actual cause of this failure is that qemu is not
> correctly able to identify the host CPU feature flags.
> This can be observed by doing: qemu-system-x86_64 -cpu Nehalem,check
> which fails with:
> warning: host cpuid 0000_0000 lacks requested flag 'fpu' [0x00000001]
> warning: host cpuid 0000_0000 lacks requested flag 'de' [0x00000004]
> and on and on...
> 
> A simple check of "cat /proc/cpuinfo | grep flags"
> flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov
> pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx
> rdtscp lm constant_tsc arch_perfmon pebs bts rep_good nopl xtopology
> nonstop_tsc aperfmperf pni dtes64 monitor ds_cpl vmx est tm2 ssse3
> cx16 xtpr pdcm dca sse4_1 sse4_2 popcnt lahf_lm ida dts tpr_shadow
> vnmi flexpriority ept vpid
> 
> Thus the CPU is more than capable of everything being asked of it via
> the cpudef for "Nehalem" in /etc/qemu/target-x86_64.conf.
> 
> This is only an issue on Fedora hosts. RHEL/CentOS/SL hosts work
> fine. This was recognized as an issue RHEL and fixed there but has
> not been fixed in Fedora.
> See: http://wiki.qemu.org/Features/CPUModels#Examples and this:
> https://bugzilla.redhat.com/show_bug.cgi?format=multiple&id=689665
> and this related discussion in qemu-devel:
> http://www.mail-archive.com/qemu-devel@nongnu.org/msg101360.html
> 
> Thus is appears that the changes were made to the RHEL qemu (eg: cpu
> type rhel6) AKA the change one needs to make to the ovirt engine
> database to have ovirt manage a EL based host.
> Fedora hosts--> psql -U postgres engine -c "update vdc_options set
> option_value='pc-0.14' where option_name='EmulatedMachine' and
> version='3.0';"
> EL hosts --> psql -U postgres engine -c "update vdc_options set
> option_value='rhel6.2.0' where option_name='EmulatedMachine' and
> version='3.0';"
> 
> Thus at the moment any host loaded with Fedora and manged by oVirt
> utilizing a Sandy Bridge, Nehalem or Westmere processor will be dead
> in the water.
> 
> -DHC
> 
> _______________________________________________
> Users mailing list
> Users at ovirt.org
> http://lists.ovirt.org/mailman/listinfo/users
>