drm/i915/guc: Fix the fix for reset lock confusion

The previous fix for the circlular lock splat about the busyness
worker wasn't quite complete. Even though the reset-in-progress flag
is cleared at the start of intel_uc_reset_finish, the entire function
is still inside the reset mutex lock. Not sure why the patch appeared
to fix the issue both locally and in CI. However, it is now back
again.

There is a further complication that the wedge code path within
intel_gt_reset() jumps around so much that it results in nested
reset_prepare/_finish calls. That is, the call sequence is:
  intel_gt_reset
  | reset_prepare
  | __intel_gt_set_wedged
  | | reset_prepare
  | | reset_finish
  | reset_finish

The nested finish means that even if the clear of the in-progress flag
was moved to the end of _finish, it would still be clear for the
entire second call. Surprisingly, this does not seem to be causing any
other problems at present.

As an aside, a wedge on fini does not call the finish functions at
all. The reset_in_progress flag is left set (twice).

So instead of trying to cancel the worker anywhere at all in the reset
path, just add a cancel to intel_guc_submission_fini instead. Note
that it is not a problem if the worker is still active during a reset.
Either it will run before the reset path starts locking things and
will simply block the reset code for a tiny amount of time. Or it will
run after the locks have been acquired and will early exit due to the
try-lock.

Also, do not use the reset-in-progress flag to decide whether a
synchronous cancel is safe (from a lockdep perspective) or not.
Instead, use the actual reset mutex state (both the genuine one and
the custom rolled BACKOFF one).

Fixes: 0e00a88 ("drm/i915/guc: Avoid circular locking issue on busyness flush")
Signed-off-by: John Harrison <[email protected]>
Cc: Zhanjun Dong <[email protected]>
Cc: John Harrison <[email protected]>
Cc: Andi Shyti <[email protected]>
Cc: Daniel Vetter <[email protected]>
Cc: Daniel Vetter <[email protected]>
Cc: Rodrigo Vivi <[email protected]>
Cc: Nirmoy Das <[email protected]>
Cc: Tvrtko Ursulin <[email protected]>
Cc: Umesh Nerlige Ramappa <[email protected]>
Cc: Andrzej Hajda <[email protected]>
Cc: Matt Roper <[email protected]>
Cc: Jonathan Cavitt <[email protected]>
Cc: Prathap Kumar Valsan <[email protected]>
Cc: Alan Previn <[email protected]>
Cc: Madhumitha Tolakanahalli Pradeep <[email protected]>
Cc: Daniele Ceraolo Spurio <[email protected]>
Cc: Ashutosh Dixit <[email protected]>
Cc: Dnyaneshwar Bhadane <[email protected]>
Reviewed-by: Nirmoy Das <[email protected]>
Reviewed-by: Andi Shyti <[email protected]>
Link: https://patchwork.freedesktop.org/patch/msgid/[email protected]

Author: johnharr-intel

drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c

@@ -1403,14 +1403,17 @@ static void guc_cancel_busyness_worker(struct intel_guc *guc)
 	 * Trying to pass a 'need_sync' or 'in_reset' flag all the way down through
 	 * every possible call stack is unfeasible. It would be too intrusive to many
 	 * areas that really don't care about the GuC backend. However, there is the
-	 * 'reset_in_progress' flag available, so just use that.
+	 * I915_RESET_BACKOFF flag and the gt->reset.mutex can be tested for is_locked.
+	 * So just use those. Note that testing both is required due to the hideously
+	 * complex nature of the i915 driver's reset code paths.
 	 *
 	 * And note that in the case of a reset occurring during driver unload
-	 * (wedge_on_fini), skipping the cancel in _prepare (when the reset flag is set
-	 * is fine because there is another cancel in _finish (when the reset flag is
-	 * not).
+	 * (wedged_on_fini), skipping the cancel in reset_prepare/reset_fini (when the
+	 * reset flag/mutex are set) is fine because there is another explicit cancel in
+	 * intel_guc_submission_fini (when the reset flag/mutex are not).
 	 */
-	if (guc_to_gt(guc)->uc.reset_in_progress)
+	if (mutex_is_locked(&guc_to_gt(guc)->reset.mutex) ||
+	    test_bit(I915_RESET_BACKOFF, &guc_to_gt(guc)->reset.flags))
 		cancel_delayed_work(&guc->timestamp.work);
 	else
 		cancel_delayed_work_sync(&guc->timestamp.work);
@@ -1424,8 +1427,6 @@ static void __reset_guc_busyness_stats(struct intel_guc *guc)
 	unsigned long flags;
 	ktime_t unused;
 
-	guc_cancel_busyness_worker(guc);
-
 	spin_lock_irqsave(&guc->timestamp.lock, flags);
 
 	guc_update_pm_timestamp(guc, &unused);
@@ -2004,13 +2005,6 @@ void intel_guc_submission_cancel_requests(struct intel_guc *guc)
 
 void intel_guc_submission_reset_finish(struct intel_guc *guc)
 {
-	/*
-	 * Ensure the busyness worker gets cancelled even on a fatal wedge.
-	 * Note that reset_prepare is not allowed to because it confuses lockdep.
-	 */
-	if (guc_submission_initialized(guc))
-		guc_cancel_busyness_worker(guc);
-
 	/* Reset called during driver load or during wedge? */
 	if (unlikely(!guc_submission_initialized(guc) ||
 		     !intel_guc_is_fw_running(guc) ||
@@ -2136,6 +2130,7 @@ void intel_guc_submission_fini(struct intel_guc *guc)
 	if (!guc->submission_initialized)
 		return;
 
+	guc_fini_engine_stats(guc);
 	guc_flush_destroyed_contexts(guc);
 	guc_lrc_desc_pool_destroy_v69(guc);
 	i915_sched_engine_put(guc->sched_engine);

drivers/gpu/drm/i915/gt/uc/intel_uc.c

@@ -637,6 +637,10 @@ void intel_uc_reset_finish(struct intel_uc *uc)
 {
 	struct intel_guc *guc = &uc->guc;
 
+	/*
+	 * NB: The wedge code path results in prepare -> prepare -> finish -> finish.
+	 * So this function is sometimes called with the in-progress flag not set.
+	 */
 	uc->reset_in_progress = false;
 
 	/* Firmware expected to be running when this function is called */