[virt-tools-list] possible bug in passing host cpu configuration

[Cole Robinson]

On 11/08/2015 06:32 PM, Cole Robinson wrote:
> On 11/08/2015 10:37 AM, LK wrote:
>> Hi,
>>
>> I have banged my head for a few days on a problem with a virtual machine
>> configuration.
>>
>> I want to use Nested Virtualization, but virt-manager did not let me to pass
>> the host cpu configuration (I checked the "Copy host CPU configuration").
>>
>> After countless tests I was giving up (I also tried to modify manually the xml
>> file in /etc/libvirt/qemu/vm.xml), but as last attempt I discarded
>> virt-manager and used virsh instead.
>>
>> I modified the configuration of the VM (virsh edit vm) and added this cpu
>> configuration:
>>
>> ###########
>> <cpu mode='host-passthrough'>
>> </cpu>
>> ###########
>>
>> I saved the configuration and launched the VM and finally the guest operating
>> system (proxmox4) shows me that it can use the nested virtualization.
>>
>> I'm using as host Fedora 23 x86_64 and virt-manager  1.2.1.
>>
>> The funny thing is that after configuring the VM using virsh edit, now
>> virt-manager retain the correct configuration.
>>
> 
> Yes, for some reason mode=host-model isn't sufficient for virt passthrough on
> some machines. We don't expose host-passthrough in the UI because it's not
> generally recommended for libvirt usage since we can't insulate the VM from
> qemu logic changes. But you can specify it in the UI by manually entering
> host-passthrough in the CPU model field
> 

This virt-manager bug has links to other issues detailing the problems with
libvirt's host-model:

https://bugzilla.redhat.com/show_bug.cgi?id=1055002

- Cole