On Wed, Oct 29, 2014 at 05:04:22PM +0530, Prerna Saxena wrote:
On Tuesday 28 October 2014 05:32 PM, Daniel P. Berrange wrote:
> On Tue, Oct 28, 2014 at 04:43:59PM +0530, Prerna Saxena wrote:
>>
>> This patch allows libvirt to extend the "fallback" semantics of cpu
model to
>> describe this new mode for PowerKVM guests.
>> As an example:
>> When a user wants to request a power7 vm to run in compatibility mode on
>> a Power8 host, this can be described in XML as follows :
>> <cpu mode='host-model' match='exact'>
>> <model fallback='compat'>power7</model>
> I don't really see any point in adding the fallback='compat'
> attribute. That is basically always implied if you use
> mode=host-model in combination with a named model.
Hi Daniel,
Running in compatibility mode is not identical to running natively on an older
processor.
Reproducing the example from :
http://www.redhat.com/archives/libvir-list/2014-October/msg00359.html
On PowerKVM, we can run VMs in 2 possible vCPU configurations:
1) Host native mode: Where host is power7; guest VMs also see power7 vcpus.
Required qemu command line : "-cpu host"
2) Compat mode: In this case, the host CPU is Power8, but it can run
in binary compatibility mode with an older version of ISA(such as
power7). So it will exhibit only those features that were
available in an older generation of Power ISA. The guest will see a
"power7" vCPU even though it is running on a physical "power8" chip.
This is a hardware feature that is enabled via a qemu/kvm switch.
Required QEMU command line for PowerKVM : "-cpu host,compat=power7"
Sure, I don't debate that this is the QEMU command line syntax. I don't
see any reason why the fact that QEMU uses 'compat' here requires the
addition of 'compat' at the libvit XML level.
My patch attempts to define semantics for this second mode of execution.
A VM running in compat mode will not just differ in terms of qemu command line, but also
in behaviour when contrasted with the host native mode. In other words, applications
running on "power7"
native VM will see different CPU as compared to applications running on a compat-mode
"power7" VM over a Power8 host. From a VM management perspective, this
difference needs to be brought out in XML.
Also, "compat" mode is one way of emulating an older processor chip, but it may
not be the _only_ way.
Hence the virtualization stack needs to explicitly distinguish a compat-mode VM.
In my v2 post, you had suggested that model fallback 'compat' be based on
"host-model" cpu mode in place of "custom". I have sent out v3 and v4
based on this suggestion.
Hope this XML schema is acceptable:
<cpu mode='host-model'>
<model fallback='compat'>power7</model>
</cpu>
The current libvirt code only allows for
<cpu mode="host-model"/>
which maps to '-cpu host'
Given this starting point, it is sufficient to just have
<cpu mode='host-model'>
<model>power7</model>
</cpu>
mapping to '-cpu host,compat=power7'
ie the very existance of a <model> against host-model is
sufficient. We don't need to invent a pointless fallback=compat
attribute too.
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 :|