RE: [PATCH i-g-t v3 2/3] tests/intel/kms_dp_linktrain_fallback: fix MST traversal state leak between subtests

Hi Kunal,

> -----Original Message-----
> From: Joshi, Kunal1 <[email protected]>
> Sent: 10 August 2026 10:24
> To: S, Sowmiya <[email protected]>; [email protected]
> Cc: B S, Karthik <[email protected]>
> Subject: Re: [PATCH i-g-t v3 2/3] tests/intel/kms_dp_linktrain_fallback: fix MST
> traversal state leak between subtests
> 
> Hello Sowmiya,
> 
> On 05-08-2026 14:14, Sowmiya S wrote:
> > traversed_mst_output_count is file-static and was never reset between
> > subtests. In a full-binary run the count from dp-fallback causes
> > setup_mst_outputs() to skip all MST outputs in subsequent subtests.
> >
> > Reset it at the top of run_lt_fallback_test(). The DSC path never
> > reaches setup_mst_outputs() so run_dsc_sst_fallaback_test() does not
> > need the reset.
> In the tree this patch applies to there are exactly two subtests, dp-fallback and
> dsc-fallback. setup_mst_outputs() is only reachable from setup_outputs()
> (tests/intel/kms_dp_linktrain_fallback.c:184), setup_outputs() is only called
> from test_fallback() (:365), and test_fallback() is only called from
> run_lt_fallback_test(). So run_lt_fallback_test() is the only path that ever
> touches traversed_mst_output_count, and igt_main() calls it exactly once. The
> counter starts at 0 (file-static), and as you say yourself in the next paragraph the
> DSC path never reaches setup_mst_outputs().
> 
> So "the count from dp-fallback causes setup_mst_outputs() to skip all MST
> outputs in subsequent subtests" describes something that cannot happen, and
> this patch is a no-op standalone. It only becomes live in 3/3, where
> run_lt_fallback_test() acquires a second caller.
> 
> That is fine for a prep patch - please just say so:
> 
>          run_lt_fallback_test() tracks which MST outputs it has already
>          traversed in a file-static counter that is never cleared. Today it
>          has a single caller so the counter is implicitly zero, but a second
>          caller would silently inherit the first run's state and skip every
>          MST sibling as "already visited".
> 
>          Reset the counter on entry so the function is safe to call more than
>          once per binary. The DSC path never reaches setup_mst_outputs(), so
>          run_dsc_sst_fallaback_test() does not need the same treatment.
> 
> 
> Thanks and Regards
> Kunal Joshi
> 

I will correct the commit message as this is a no-op without 3/3 patch.
Regards,
Sowmiya.

> >
> > Signed-off-by: Sowmiya S <[email protected]>
> > ---
> >   tests/intel/kms_dp_linktrain_fallback.c | 7 +++++++
> >   1 file changed, 7 insertions(+)
> >
> > diff --git a/tests/intel/kms_dp_linktrain_fallback.c
> b/tests/intel/kms_dp_linktrain_fallback.c
> > index 115183d2d..44b875ab9 100644
> > --- a/tests/intel/kms_dp_linktrain_fallback.c
> > +++ b/tests/intel/kms_dp_linktrain_fallback.c
> > @@ -431,6 +431,13 @@ static bool run_lt_fallback_test(data_t *data)
> >   	bool ran = false;
> >   	igt_output_t *output;
> >
> > +	/*
> > +	 * Reset per invocation so MST traversal state from a previous subtest
> > +	 * (e.g. dp-fallback followed by dsc-fallback) doesn't leak and cause
> > +	 * MST siblings to be silently skipped as "already visited".
> > +	 */
> > +	traversed_mst_output_count = 0;
> > +
> >   	for_each_connected_output(&data->display, output) {
> >   		data->output = output;
> >