On Fri, May 13, 2022 at 10:56 AM Daniel Latypov <dlatypov@xxxxxxxxxx> wrote:
>
> On Fri, May 13, 2022 at 1:51 AM David Gow <davidgow@xxxxxxxxxx> wrote:
> >
> > We're currently using the x86_64 qemu for i386 builds. While this is not
> > incorrect, it's probably more sensible to use the i386 one, which will
> > at least fail properly if we accidentally were to build a 64-bit kernel.
> >
> > Signed-off-by: David Gow <davidgow@xxxxxxxxxx>
>
> Tested-by: Daniel Latypov <dlatypov@xxxxxxxxxx>
>
> Works for me:
> [10:55:58] Testing complete. Ran 201 tests: passed: 165, skipped: 36
> [10:55:58] Elapsed time: 21.268s total, 0.003s configuring, 4.009s
> building, 16.967s running
>
> Side note:
> I ran into an unrelated bug where running on QEMU in general seems to
> hang when you try to filter to a test suite that doesn't exist.
> I've confirmed it happens on i386 before this patch, and it at least
> also happens on x86_64.
Ah, this "unrelated bug" is my fault, specifically commit a02353f49162
("kunit: bail out of test filtering logic quicker if OOM").
The use of _any_ filter glob is broken on kunit.py + QEMU.
Fix is https://lore.kernel.org/linux-kselftest/20220513183707.97290-1-dlatypov@xxxxxxxxxx/
