Re: [PATCH i-g-t] tests/intel/xe_wedged: Manage verbosity of GT reset fault injection

[Rodrigo Vivi <rodrigo.vivi@intel.com>]

On Thu, Nov 28, 2024 at 06:05:44AM -0500, Rodrigo Vivi wrote:
> On Tue, Nov 26, 2024 at 03:02:40PM -0800, Matt Roper wrote:
> > igt@xe_wedged@basic-wedged triggers a GT reset and uses fault injection
> > to ensure that the reset artificially fails.  This works as expected,
> > but depending on the value in the fail_gt_reset/verbose debugfs
> > when the test starts, different levels of output may be printed to
> > dmesg.  The test already marks the general "declared wedged" message as
> > expected so that igt_runner won't try to flag those as test errors, but
> > if verbose=2, then the fault injection framework will also issue print
> > dump_stack() output which gets flagged as a problem by igt_runner.
> > 
> > Explicitly set verbose=1 when running the test so that we get the
> > general wedged message and fault injection notification (which we
> > already tell igt_runner to ignore), but skip the unnecessary stack dump
> > that just confuses the runner.
> > 
> 
> Cc: Kamil Konieczny <kamil.konieczny@linux.intel.com>
> Cc: Francois Dugast <francois.dugast@intel.com>
> 
> > Closes: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/2919
> > Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
> > ---
> >  tests/intel/xe_wedged.c | 1 +
> >  1 file changed, 1 insertion(+)
> > 
> > diff --git a/tests/intel/xe_wedged.c b/tests/intel/xe_wedged.c
> > index 88e5d47f2..0db8c8eba 100644
> > --- a/tests/intel/xe_wedged.c
> > +++ b/tests/intel/xe_wedged.c
> > @@ -213,6 +213,7 @@ igt_main
> >  	igt_subtest("basic-wedged") {
> >  		igt_require(igt_debugfs_exists(fd, "fail_gt_reset/probability",
> >  					       O_RDWR));
> > +		igt_debugfs_write(fd, "fail_gt_reset/verbose", "1");
> 
> Great finding! I believe this solves some of the cases that Kamil was working to mask.
> But also +Francois because it might have some use on the other fault-inject cases.

I forgot to add the rv-b tag here before hitting enter...

But that bought me some time :)

So, what about using verbose = 0 ? then we even skip the FAULT_INJECTION message
and don't need to filter for that as well, since by the test itself we know that
it is the injection.

But I'm okay on keeping 1 and filtering

Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

> 
> Thanks a lot,
> Rodrigo.
> 
> >  
> >  		igt_assert_eq(simple_ioctl(fd), 0);
> >  		ignore_wedged_in_dmesg();
> > -- 
> > 2.47.0
> >