[PATCH 6/7] drm/xe: Don't send a CLEANUP message on sched pause

[Stuart Summers <stuart.summers@intel.com>]

When the DRM scheduler has been "paused" (see xe_gpu_scheduler.c),
if we then send a message to cleanup an exec queue at that time,
we will never get a response for it if for some reason the
unpause never happens. This can occur if for some reason during
exec queue creation, the device becomes wedged. The wedge event
will cause that scheduler to become paused and the creation
then leaves a dangling LRC that will never get cleaned up since
the recovery is a driver rebind. So essentially this change
fixes a potential memory leak in the event the device is
wedged during a test unexpectedly (e.g. during an unexpected
hardware failure).

Signed-off-by: Stuart Summers <stuart.summers@intel.com>
---
 drivers/gpu/drm/xe/xe_guc_submit.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/xe/xe_guc_submit.c b/drivers/gpu/drm/xe/xe_guc_submit.c
index 55a50c46ea2b..45b72bebfc63 100644
--- a/drivers/gpu/drm/xe/xe_guc_submit.c
+++ b/drivers/gpu/drm/xe/xe_guc_submit.c
@@ -1732,7 +1732,8 @@ static void guc_exec_queue_destroy(struct xe_exec_queue *q)
 {
 	struct xe_sched_msg *msg = q->guc->static_msgs + STATIC_MSG_CLEANUP;
 
-	if (!(q->flags & EXEC_QUEUE_FLAG_PERMANENT) && !exec_queue_wedged(q))
+	if (!(q->flags & EXEC_QUEUE_FLAG_PERMANENT) && !exec_queue_wedged(q) &&
+	    !READ_ONCE(q->guc->sched.base.pause_submit))
 		guc_exec_queue_add_msg(q, msg, CLEANUP);
 	else
 		__guc_exec_queue_destroy(exec_queue_to_guc(q), q);
-- 
2.34.1