update 'hw:cpu_model' meta data to support *-IBRS vcpu type #15
Conversation
after libvirt upgraded to 4.7.0, there are some new vcpu types for instance vcpu_model, if compute node is using *-IBRS cpu, Passthrough favor can not be used for creating instance. it will cause some like below error "No valid host was found. There are not enough hosts available. computer-0: (VCpuModelFilter) Host VCPU model Skylake-Server-IBRS required Passthrough Code 501" Closes-Bug: 1803280 Signed-off-by: Sun Austin <austin.sun@intel.com>
There was a problem hiding this comment.
There's a problem with this patch. The way the VCpuModelFilter is designed, the basic idea is that a host with a CPU model "later" in the list can support guests with all CPU models "earlier" in the list.
According to [1] the "spec-ctrl" and "ssbd" CPU features require host CPU microcode (i.e. that the host is detected as the "-IBRS" model) before it can be used for guest CPUs. This means that a "Skylake-Server" host cannot support a "Broadwell-IBRS" guest, which violates the design of the VCpuModelFilter.
I think as part of adding the -IBRS models we'll have to also modify the filter itself.
Chris
[1] https://www.berrange.com/posts/2018/06/29/cpu-model-configuration-for-qemu-kvm-on-x86-hosts/
|
Yes. Thanks your suggestion. I will make a new PR later. |
after libvirt upgraded to 4.7.0, there are some new vcpu types for
instance vcpu_model, if compute node is using *-IBRS cpu,
Passthrough favor can not be used for creating instance.
it will cause some like below error
"No valid host was found. There are not enough hosts available.
computer-0: (VCpuModelFilter) Host VCPU model Skylake-Server-IBRS
required Passthrough
Code 501"
Closes-Bug: 1803280
Signed-off-by: Sun Austin austin.sun@intel.com