Re: Running kunit tests on platform devices

[Tales Lelo da Aparecida]

On 17/08/2022 09:19, Ramon Fried wrote:
> On Wed, Aug 17, 2022 at 7:43 AM David Gow <davidgow@xxxxxxxxxx> wrote:
> > On Mon, Aug 15, 2022 at 5:16 PM Ramon Fried <rfried.dev@xxxxxxxxx> wrote:
> > > Hi.
> > > I implemented a test suite that checks a platform driver, that's the
> > > only way I can test interrupts behavior in the code.
> > > Because it's a platform, I can't use kunit_test_suite(), so I call
> > > __kunit_test_suites_init() as part of the platform driver probe
> > > function.
> > >
> > > This works fine but has the following problems.
> > > "TAP version 14" string is not printed and it's impossible to parse
> > > the results using the script.
> > > In addition, the suite is not displayed in /sys/kernel/debug/kunit.
> > >
> > > It would be my pleasure to provide a patch that fixes this, I just
> > > wanted to make sure that my testing strategy makes sense.
> > >
> > > Thanks,
> > > Ramon
> > Hi Ramon,
> >
> > Thanks for reaching out. Calling __kunit_test_suites_init() directly
> > is not something we'd recommend (and are trying desperately to remove
> > existing uses), so several of the issues re: the "TAP version 14"
> > string et al are side effects of calling what is supposed to be an
> > internal KUnit function manually.
> >
> > That being said, we definitely do want to make testing platform
> > drivers as convenient as possible. I'm not sure why kunit_test_suite()
> > doesn't work for you for platform drivers: are you just missing some
> > way of instantiating the device from within a test context?
> >
> > I know Brendan has experimented with some hardware faking code, for
> > which there's some example use here:
> > https://kunit-review.googlesource.com/c/linux/+/5275
> > (Note that you'll need to look at the other patches in the 'relation
> > chain' for dependencies.)
> >
> > Equally, I've experimented a bit with testing old soundcard drivers
> > (via a platform device) here, which may be an easier way to look
> > through:
> > https://github.com/sulix/linux/commit/4e1620c86553b9edde7f032318cf417dc13e4d26
> >
> > Note that neither of those examples are anything other than
> > experiments, so may not work as-is, or be ideal.
> >
> > Otherwise, we're always happy to accept patches, though again, if
> > there's any way of getting your tests working without a direct call to
> > __kunit_test_suites_init() --- even if that would require patches to
> > work --- that'd be preferable on our end.
> >
> > Cheers,
> > -- David
> Hi David,
> Thanks for replying.
> I looked at the examples you shared, and they all fake the actual device.
> My intention is to actually interact with the real device. - get a
> real interrupt, etc.
> Is it wrong, was the intention that the platform device be mocked ?
> Thanks,
> Ramon.

Hi, Ramon,

I particularly don't condemn writing tests that require hardware, but 
they're best avoided, mostly to facilitate running those tests.

Can you share your code?
I would be happy to take a look if its not a problem for you!

Kind regards,
Tales