On Mon, Nov 23, 2020 at 10:15:11 +0000, Daniel P. Berrangé wrote:
On Fri, Nov 20, 2020 at 08:38:59PM +0100, Jiri Denemark wrote:
> Doing so would normally break migration to older libvirt, but most of
> the patches in this series prepare an infrastructure for removing
> features from existing CPU models while maintaining backward
> compatibility during migration.
>
> See patch 7 for more details.
QEMU introduced versioned CPU models to allow them to add/remove
arbitrary features within a named model. What's the timeframe for
supporting those in libvirt.
No idea, to be honest. QEMU currently does not provide all data we need
to start using it. I've been (occasionally, though) asking for QMP
extension for some time.
IIUC, that should stop us needing to introduce special workarounds
like this series ?
However, versioned CPU models would not save us in this case anyway as
QEMU did not introduce a new versions of models when removing "monitor".
This workaround is targeted at features which were included in CPU
models mostly accident and when they were not in fact enabled even
though the definitions of the CPU models contained them.
Jirka