Merging upstream version 6.10.6.

Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>

1 files changed, 11 insertions, 3 deletions

diff --git a/drivers/gpu/drm/xe/xe_preempt_fence.c b/drivers/gpu/drm/xe/xe_preempt_fence.c
index 7d50c6e89d..5b243b7feb 100644
--- a/drivers/gpu/drm/xe/xe_preempt_fence.c
+++ b/drivers/gpu/drm/xe/xe_preempt_fence.c
@@ -23,11 +23,19 @@ static void preempt_fence_work_func(struct work_struct *w)
 		q->ops->suspend_wait(q);
 
 	dma_fence_signal(&pfence->base);
-	dma_fence_end_signalling(cookie);
-
+	/*
+	 * Opt for keep everything in the fence critical section. This looks really strange since we
+	 * have just signalled the fence, however the preempt fences are all signalled via single
+	 * global ordered-wq, therefore anything that happens in this callback can easily block
+	 * progress on the entire wq, which itself may prevent other published preempt fences from
+	 * ever signalling.  Therefore try to keep everything here in the callback in the fence
+	 * critical section. For example if something below grabs a scary lock like vm->lock,
+	 * lockdep should complain since we also hold that lock whilst waiting on preempt fences to
+	 * complete.
+	 */
 	xe_vm_queue_rebind_worker(q->vm);
-
 	xe_exec_queue_put(q);
+	dma_fence_end_signalling(cookie);
 }
 
 static const char *