Re: [Xen-devel] [PATCH v6 29/29] libxl: add support for migrating HVM guests without a device model

[Andrew Cooper <andrew.cooper3@xxxxxxxxxx>]

On 04/09/15 13:09, Roger Pau Monne wrote:
> Only some minor libxl changes are needed in order to be able to migrate HVM
> guests without a device model, no hypervisor changes are needed.
>
> This change adds support for LIBXL_DEVICE_MODEL_VERSION_NONE into the
> migration code.
>
> Signed-off-by: Roger Pau Monnà <roger.pau@xxxxxxxxxx>
IMO, this is conceptually wrong.

No device model should be signified by not sending an emulator record.  
Sending an emulator record saying "No emulator" is pointless in a single 
emulator setup, and wrong for a multi-emulator setup.
Furthermore, for situations like vGPU with no-dm guests, there will 
actually be an emulator record sent for demu (The component XenServer 
currently uses to make vGPU function).

On the save side, if the domain has no emulator, no EMULATOR_* records 
should be sent.
On the receive side the domain configuration has already been sent so 
the domain is known to not have an emulator.  The arrival of an 
EMULATOR_* record should be a fatal error for restore.
Support for migration of no-dm guests does not require any extension to 
the migration ABI.  Having said this, the libxl spec does not have a 
"LAYOUT" section like the libxc spec, and might benefit from having one.
~Andrew

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel