On Fri, Dec 16, 2016 at 12:10:55 -0500, Collin L. Walling wrote:
On 12/15/2016 11:10 AM, Jiri Denemark wrote:
> We already have a lot of tests that use the ±feature syntax so there's
> no real need to add another one. However, we could use it as a negative
> test since s390 does not support the old syntax. So how about:
>
> DO_TEST("cpu-s390-features", QEMU_CAPS_KVM,
> QEMU_CAPS_QUERY_CPU_MODEL_EXPANSION);
> DO_TEST_FAILURE("cpu-s390-features", QEMU_CAPS_KVM);
>
> Of course, patch 5/11 would need to be modified to actually refuse to
> start an s390 domain with CPU features in the XML on QEMU without
> QEMU_CAPS_QUERY_CPU_MODEL_EXPANSION capability.
I agree with not testing for +/- on s390x. And since s390x doesn't
even support CPU models until 2.8 (which is conveniently when
query-cpu-model-expac was introduced!), we don't necessarily
need to do any kind of error checking in the CPU feature
argument generation code. We will encounter an error
relating to "no support for CPU models" first.
Oh nice. In that case we should add a domain capabilities test case for
QEMU 2.7.0 to make sure we do not advertise support for any CPU mode.
Jirka