Re: [PATCH] cpu_map: Introduce A64FX
On Wed, Aug 31, 2022 at 15:11:41 +0100, Daniel P. Berrangé wrote:
On Wed, Aug 31, 2022 at 02:30:52PM +0200, Jiri Denemark wrote:
> On Wed, Aug 31, 2022 at 09:22:16 +0000, Liu Yiding wrote:
> > Add A64FX as a supported cpu model.
> >
> > Signed-off-by: Liu Yiding <liuyd.fnst(a)fujitsu.com>
> > ---
> > src/cpu_map/arm_A64FX.xml | 6 ++++++
> > src/cpu_map/arm_vendors.xml | 1 +
> > src/cpu_map/index.xml | 3 +++
> > src/cpu_map/meson.build | 1 +
> > 4 files changed, 11 insertions(+)
> > create mode 100644 src/cpu_map/arm_A64FX.xml
>
> CPU detection for ARM is a bit awkward and not very useful in practice.
> It will (in some cases) report the host CPU in host capabilities, but
> that's it. The CPU from host capabilities cannot really be used in
> general for specifying a guest CPU in domain XML and it would not even
> make a lot of sense as I believe the only really supported CPU for KVM
> is "host" (mode="host-passthrough" in domain XML).
This still lets it be used for a guest with TCG, which has no requirement
to use 'host'.
The host CPU is irrelevant for TCG and libvirt on ARM has always allowed
any CPU model to be used, even those that are not in cpu_map. And since
a lot of the ARM CPUs in cpu_map do not match anything in QEMU anyway,
which is the reason why host CPU detection on ARM is mostly a cosmetic
thing.
> I'm not saying the patch is wrong, I'm just trying to
make sure you
> don't expect something more from implementing this patch.
Jirka