Re: [PATCH] selftests/mm: Have the harness run each test category separately

[Andrew Morton]

On Tue, 20 Jan 2026 13:25:32 +0000 Mark Brown <broonie@xxxxxxxxxx> wrote:

> At present the mm selftests are integrated into the kselftest harness by
> having it run run_vmtest.sh and letting it pick it's default set of
> tests to invoke, rather than by telling the kselftest framework about
> each test program individually as is more standard.   This has some
> unfortunate interactions with the kselftest harness:
> 
>  - If any of the tests hangs the harness will kill the entire mm
>    selftests run rather than just the individual test, meaning no
>    further tests get run.
>  - The timeout applied by the harness is applied to the whole run rather
>    than an individual test which frequently leads to the suite not being
>    completed in production testing.
> 
> Deploy a crude but effective mitigation for these issues by telling the
> kselftest framework to run each of the test categories that run_vmtests.sh
> has separately.  Since kselftest really wants to run test programs this
> is done by providing a trivial wrapper script for each categorty that
> invokes run_vmtest.sh, this is not a thing of great elegence but it is
> clear and simple.  Since run_vmtests.sh is doing runtime support
> detection, scenario enumeration and setup for many of the tests we can't
> consistently tell the framework about the individual test programs.
> 
> This has the side effect of reordering the tests, hopefully the testing
> is not overly sensitive to this.

Thanks, let's see what people think.

What happens with tests which are newly added but which don't integrate
into this new framework?  eg,
https://lkml.kernel.org/r/20260120123239.909882-2-linmiaohe@xxxxxxxxxx