Hi Kuniyuki.

On Fri, Jul 31, 2026 at 07:53:59AM +0200, Kuniyuki Iwashima <kuniyu@google.com> wrote:
> Why is this selftest backported to the kernel without the
> functional changes ?  That's the problem.

Let me join the topic and vouch for the changes from Sebastian.

I consider there's a scale of selftests from validation of bugfixes to
checking of features. Another axis is the discoverability of the
supported features by the kernel (none ... CONFIG_ check ... runtime
indicator).

I've found it quite useful to have one set of tests that can be run
across different kernel versions (think of stable kernels), the
maintenance is simpler, sometimes even same builds may be reused.

Occasionally a reported failure raises unnecessary alarms when it is
actually only a missing or disabled feature.

So when the behavior is on the right side of the above spaces
(feature*discoverability), it is relatively easy to add a guard in the
selftest proper. That is something what've been striving for in cgroup
selftests.

Obviously, the dividing line for a good guard is not always clear cut,
in such cases it remains up to the interpreter of the test output to
decide how serious the failures are.

Here I think the ENOPROTOOPT are suitable for the feature detection.

Regards,
Michal