Re: [kvm-unit-tests PATCH v5 0/8] Multi-migration support

[Thomas Huth <thuth@redhat.com>]

On 26/02/2024 09.10, Nicholas Piggin wrote:
> On Fri Feb 23, 2024 at 5:06 PM AEST, Thomas Huth wrote:
>> On 21/02/2024 04.27, Nicholas Piggin wrote:
>>> Now that strange arm64 hang is found to be QEMU bug, I'll repost.
>>> Since arm64 requires Thomas's uart patch and it is worse affected
>>> by the QEMU bug, I will just not build it on arm. The QEMU bug
>>> still affects powerpc (and presumably s390x) but it's not causing
>>> so much trouble for this test case.
>>>
>>> I have another test case that can hit it reliably and doesn't
>>> cause crashes but that takes some harness and common lib work so
>>> I'll send that another time.
>>>
>>> Since v4:
>>> - Don't build selftest-migration on arm.
>>> - Reduce selftest-migration iterations from 100 to 30 to make the
>>>     test run faster (it's ~0.5s per migration).
>>
>> Thanks, I think the series is ready to go now ... we just have to wait for
>> your QEMU TCG migration fix to get merged first. Or should we maybe mark the
>> selftest-migration with "accel = kvm" for now and remove that line later
>> once QEMU has been fixed?
> 
> Could we merge it? I'm juggling a bunch of different things and prone to
> lose track of something :\ I'll need to drum up a bit of interest to
> review the QEMU fixes from those who know the code too, so that may take
> some time.

Ok, I merged it, but with "accel = kvm" for the time being (otherwise this 
would be quite a pitfall for people trying to run the k-u-t with TCG when 
they don't know that they have to fetch a patch from the mailing list to get 
it working).

> I left it out of arm unittests.cfg entirely, and s390 and powerpc seems
> to work by luck enough to be useful for gitlab CI so I don't think there
> is a chnage needed really unless you're paranoid.

At least the s390x test does not work reliably at all when running with TCG 
without your QEMU patch, so I think we really need the "accel = kvm" for the 
time being here.

> I do have a later patch that adds a memory tester that does trigger it
> right away on powerpc. I'll send that out after this series is merged...
> but we do still have the issue that the gitlab CI image has the old QEMU
> don't we? Until we update distro.

We only run selected tests in the gitlab-CI, so unless you add it to 
.gitlab-ci.yml, the selftest-migration test won't be run there.

  Thomas