kernel: Set LTP_TAINT_EXPECTED for various memory tests

[pevik]

ltp_mm, ltp_mm_oom, ltp_numa, ltp_uevent on x86_64 need LTP_TAINT_EXPECTED=0x200 to avoid Kernel issued warning.

aarch64 and ppc64le don't issue these taints, s390x don't test them, thus not adding this setup.

ltp_mm is tested also on i586, where it's not needed atm, but overcommit_memory0[1-6] fail. Likely, once these tests are fixed, they will issue the same taint, thus set it now.

See:

• https://openqa.opensuse.org/tests/overview?distri=opensuse&version=Tumbleweed&build=20240825&groupid=32
• https://openqa.opensuse.org/tests/overview?distri=opensuse&version=Tumbleweed&build=20240823&groupid=32

@Vogtinator @mdoucha FYI

[Vogtinator]

IMO this should be done in the testsuite or test code and not the jobgroups.

Is there ever a case where this taint is not ok?

[pevik]

@Vogtinator

1. This is setup just for some archs (in the testsuite the setup is for all architectures). The reason, why this is done only for x86_64 and i586 is in the git commit message and in PR description.
2. We have lost some setup in the past DB data failure and following recovery from the backup, that's why we use this approach in both o3 and osd. We would prefer to have the same approach for kernel tests on both o3 and osd (I've been forced to do things differently for o3 and osd due different opinions of the reviewers and it's not really optimal for the team). Also we already merged some setup in LTP_TAINT_EXPECTED. Having in openQA code is_opensuse + comments that for osd we solve that in jobgroups yaml is not really optimal.

[Vogtinator]

This is setup just for some archs (in the testsuite the setup is for all architectures). The reason, why this is done only for x86_64 and i586 is in the git commit message and in PR description.

Yes, which I don't really agree with - just accept the taint for all architectures?

Having in openQA code is_opensuse + comments that for osd we solve that in jobgroups yaml is not really optimal.

osd and o3 use the same distri for testing, so to fix it in one place it can be done there.

[pevik]

Well, first I was incorrect, it's a real bug: https://bugzilla.suse.com/show_bug.cgi?id=1229671 (I thought it was caused by out of memory tests), thus closing it + I will tag the tests.

[mdoucha]

Is there ever a case where this taint is not ok?

I'd say there is no case where this taint would be OK. All kernel warnings need to be investigated and fixed. In this case, it's a QEMU configuration error. QEMU_NUMA=1 requires QEMUSOCKETS=%QEMUCPUS%. When I add the missing QEMUSOCKETS, the warning goes away.
https://openqa.opensuse.org/tests/4435106

[Vogtinator]

I thought this was about https://progress.opensuse.org/issues/165587#change-832358, but that's actually a different taint:

Kernel taint: 0x3000

Unexpected taint:
- Unsigned module was loaded