[virt-tools-list] [virt-tool-list][PATCH]virt-manager: use host CPU configuration when creating new VM

[Chen Hanxiao]

> -----Original Message-----
> From: Cole Robinson [mailto:crobinso at redhat.com]
> Sent: Tuesday, November 19, 2013 9:23 PM
> To: Giuseppe Scrivano; Chen Hanxiao
> Cc: virt-tools-list at redhat.com
> Subject: Re: [virt-tools-list] [virt-tool-list][PATCH]virt-manager: use
host CPU
> configuration when creating new VM
> 
> On 11/19/2013 07:52 AM, Giuseppe Scrivano wrote:
> > Chen Hanxiao <chenhanxiao at cn.fujitsu.com> writes:
> >
> >> From: Chen Hanxiao <chenhanxiao at cn.fujitsu.com>
> >>
> >> This patch will use host CPU configuration automatically
> >> when creating new VM.
> >
> > on what version of virt-manager is your patch based?  I tried to apply
> > it to the current git master branch[1] but I get a conflict.
> >
> > In addition, I am not sure what is the impact of such a change, what are
> > for example the downsides for a migration if we default to a more
> > restrictive CPU configuration.
> >
> 
> Indeed, just doing this unconditionally is not an option.
> 
> I will be working on this within the next few weeks. My plan is to add an
> option 'Default CPU for New VMs' to the Preferences dialog. There will be
3
> options:

How about add this on vmm-create.ui: tab Memory?
> 
> 1) hypervisor default (no <cpu> block)
> 2) copy just the detected CPU model (so pentium3, Opteron_G4, etc)

use baselineCPU()?

> 3) full <cpu> copy (so either what this patch suggests, or use host-model

use copy_host_cpu()?

> 
> My plan is for the new virt-manager default to be option 2. Since
libvirt's
> host <cpu> detection has known issues with generating CPU configurations
> that
> don't match reality, #3 isn't safe yet. But this still gives the user a
chance
> to make it the default if they aren't affected by the issues. When libvirt
and
> qemu straighten it all out and host-model is a safe option, we can change
the
> default in virt-manager.
> 
> - Cole