libvirt users - Aug 2014 - Passing literal -cpu model string to qemu

On aarch64 with -M virt, the default CPU model is cortex-a15 (a 32 bit
CPU).  This is IMHO a stupid default, but there we are.  Therefore
most users will need to pass the `-cpu cortex-a53' or `-cpu cortex-a57'
flag to qemu, depending on a complex formula of their host CPU and if
they are using TCG or not.

However I cannot work out how to pass this through libvirt.

The obvious one would be:

  <cpu><model>cortex-a57</model></cpu>

This passes `-cpu host' which is both wrong and likely to be a bug.

I tried adding various flags and studying the source, but it's
extremely non-obvious how to do this.  Also adding the following into
cpu_map.xml didn't make any difference:

--- a/src/cpu/cpu_map.xml
+++ b/src/cpu/cpu_map.xml
@@ -626,6 +626,11 @@
       <vendor name='IBM'/>
       <pvr value='0x004b0100'/>
     </model>
+  </arch>
 
+  <!-- ARM, 64 bit -->
+  <arch name='aarch64'>
+    <model name='cortex-a53'/>
+    <model name='cortex-a57'/>
   </arch>
 </cpus>

Ideas here?

Rich.

-- 
Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
Read my programming and virtualization blog: http://rwmj.wordpress.com
virt-df lists disk usage of guests without needing to install any
software inside the virtual machine.  Supports Linux and Windows.
http://people.redhat.com/~rjones/virt-df/

On Thu, Aug 07, 2014 at 02:28:05PM +0100, Richard W.M. Jones
wrote:
> On aarch64 with -M virt, the default CPU model is cortex-a15 (a 32 bit
> CPU).  This is IMHO a stupid default, but there we are.  Therefore
> most users will need to pass the `-cpu cortex-a53' or `-cpu
cortex-a57'
> flag to qemu, depending on a complex formula of their host CPU and if
> they are using TCG or not.
> 
> However I cannot work out how to pass this through libvirt.
> 
> The obvious one would be:
> 
>   <cpu><model>cortex-a57</model></cpu>
> 
> This passes `-cpu host' which is both wrong and likely to be a bug.
> 
> I tried adding various flags and studying the source, but it's
> extremely non-obvious how to do this.  Also adding the following into
> cpu_map.xml didn't make any difference:
IIRC we don't have the CPU model stuff written for ARM architecture
yet.


Regards,
Daniel
-- 
|: http://berrange.com      -o-    http://www.flickr.com/photos/dberrange/ :|
|: http://libvirt.org              -o-             http://virt-manager.org :|
|: http://autobuild.org       -o-         http://search.cpan.org/~danberr/ :|
|: http://entangle-photo.org       -o-       http://live.gnome.org/gtk-vnc :|