[libvirt] [PATCH 0/4] qemu: Use host-model CPU on s390 by default
Collin Walling
walling at linux.ibm.com
Mon Nov 25 22:55:11 UTC 2019
On 11/14/19 10:44 AM, Jiri Denemark wrote:
> On s390 machines host-passthrough and host-model CPUs result in the same
> guest ABI (with QEMU new enough to be able to tell us what "host" CPU is
> expanded to, which was implemented around 2.9.0). So instead of using
> host-passthrough CPU when there's no CPU specified in a domain XML we
> can safely use host-model and benefit from CPU compatibility checks
> during migration, snapshot restore and similar operations.
>
>
> This series applies on top of "qemu: Store default CPU in domain XML"
> which is already acked, but it's waiting for a QEMU patch to be applied
> to 4.2.0.
>
> You can fetch both series at once using
>
> git fetch https://gitlab.com/jirkade/libvirt cpu-default-type
>
> Jiri Denemark (4):
> cpu_conf: Fix default value for CPU match attribute
> cpu_conf: Don't format empty model for host-model CPUs
> cpu_s390: Don't check match attribute for host-model CPUs
> qemu: Use host-model CPU on s390 by default
>
> src/conf/cpu_conf.c | 25 ++++----------
> src/conf/cpu_conf.h | 2 +-
> src/cpu/cpu_s390.c | 18 +++++-----
> src/qemu/qemu_domain.c | 33 ++++++++++++-------
> .../ppc64-host+guest-compat-none.xml | 4 +--
> ...t-cpu-kvm-ccw-virtio-4.2.s390x-latest.args | 4 ++-
> .../cpu-check-default-partial.xml | 4 +--
> .../cpu-host-model-features.xml | 1 -
> ...lt-cpu-kvm-ccw-virtio-4.2.s390x-latest.xml | 2 +-
> 9 files changed, 45 insertions(+), 48 deletions(-)
>
Guest defined as expected. Once I start a guest, a dumpxml showed
the model name and associated features as expected. Was able to
successfully migrate the guest as well.
Much preferred over the old default host-passthrough way.
Built libvirt using the repo and branch you provided
Built QEMU using master 4.1.92 (v4.2.0-rc2-27-g65e05c82bd)
- guest used machine s390-ccw-virtio-4.2
Migrated between a weird mixed environment I'm chaotically working with :)
Perhaps the testing was a bit over-extensive, but it's good to make sure.
Tested-by: Collin Walling <walling at linux.ibm.com>
--
Respectfully,
- Collin Walling
More information about the libvir-list
mailing list