[PATCH i-g-t] tests/xe_exec_threads: Fill in GT field for second balancer thread

[Lucas De Marchi]

On Wed, Mar 20, 2024 at 12:23:56PM -0700, Matt Roper wrote:
>The balancer subtests spawn two pthreads per engine class if there are
>multiple instances of the class.  The GT field of the data structure is
>filled in properly for the first thread, but not for the second,
>effectively leaving it set to "0."
>
>For platforms with standalone media, this will result in failures when
>the thread tries to find the instances of a media class on GT0 and trips
>the "igt_assert(num_placements > 1)" assertion in test_balancer().
>
>Signed-off-by: Matt Roper <matthew.d.roper at intel.com>


Reviewed-by: Lucas De Marchi <lucas.demarchi at intel.com>

do you know why we unrolled a loop in there and did a wrong copy and
paste? That together with vertical spacing on flow control seems to be a
good source of bugs. Good find.

The only difference between the first and second thread data I'm seeing
is flags VIRTUAL vs PARALLEL. Matt Brost, is this test from you? Could
you add an overview to its documentation? The boilerplate doc added
there after the fact looks less than helpful.

Lucas De Marchi

>---
> tests/intel/xe_exec_threads.c | 1 +
> 1 file changed, 1 insertion(+)
>
>diff --git a/tests/intel/xe_exec_threads.c b/tests/intel/xe_exec_threads.c
>index 55907e2b3..977e8c600 100644
>--- a/tests/intel/xe_exec_threads.c
>+++ b/tests/intel/xe_exec_threads.c
>@@ -1081,6 +1081,7 @@ static void threads(int fd, int flags)
> 						threads_data[i].fd = 0;
> 					else
> 						threads_data[i].fd = fd;
>+					threads_data[i].gt = gt;
> 					threads_data[i].vm_legacy_mode =
> 						vm_legacy_mode;
> 					threads_data[i].class = class;
>-- 
>2.43.0
>