[igt-dev] [PATCH i-g-t v2] ts/core_setmaster: new test for drop/set master semantics

[Petri Latvala]

On Fri, Feb 21, 2020 at 01:55:18PM +0000, Emil Velikov wrote:
> Hi Petri,
> 
> Thank you for having a look.
> 
> On Fri, 21 Feb 2020 at 11:58, Petri Latvala <petri.latvala at intel.com> wrote:
> 
> > > +static void check_drop_set(void)
> > > +{
> > > +     int master;
> > > +
> > > +     master = __drm_open_driver(DRIVER_ANY);
> > > +
> > > +     /* Double-check if open has failed */
> > > +     igt_assert_neq(master, -1);
> >
> > Just use drm_open_driver(). For sure you don't want to produce a
> > 'FAIL' when running on a system without GPU drivers. 'SKIP' is correct
> > for that case.
> >
> This sounds very strange. Why would anyone run IGT if they lack _any_
> GPU drivers.

Accidentally, by breaking the driver module loading.

> If I'm running GPU tests and my GPU doesn't show up for any reason,
> I'd expect a hard failure.

Sure, some kind of a notification for that case is desired. However
the correct notification for that is a SKIP result. Granted, for
DRIVER_ANY open call the story is a bit more far-fetched, but consider
this:

If you get a FAIL from this test, you want to be able to say "I got
that because handling master in kernel is broken". If you can't open a
driver, you can't tell that kernel is broken. The only thing you can
tell is: You can't test the feature. SKIP is exactly that: Can't test
this.


> Even if we ignore that, a quick look around shows that there are
> multiple tests that will happily pass -1 to forward.
> If anything, the only way to trigger this is my dropping the chown()
> calls.

Which tests would those be? The only ones I can spot are ones that
want to test "can we still load the driver after doing $badthing".

Notice that drm_open_driver() stops testing if the open fails so it
cannot pass -1 along.


-- 
Petri Latvala