From: =?UTF-8?q?Christian=20K=C3=B6nig?= <christian.koenig@amd.com>
Date: Fri, 15 Dec 2017 13:36:49 +0100
Subject: drm/ttm: use try_lock in ttm_bo_delayed_delete again
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Git-commit: 827ed2b06b05a9e9791e10f03096cc5ddd85bc44
Patch-mainline: v4.16-rc1
References: FATE#326289 FATE#326079 FATE#326049 FATE#322398 FATE#326166

We only need to wait for the contended lock when the reservation object is
shared or when we want to remove everything. A trylock should be sufficient
in all other cases.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Roger He <Hongbo.He@amd.com>
Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Acked-by: Petr Tesarik <ptesarik@suse.com>
---
 drivers/gpu/drm/ttm/ttm_bo.c |   13 +++++++++----
 1 file changed, 9 insertions(+), 4 deletions(-)

--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -586,12 +586,17 @@ static bool ttm_bo_delayed_delete(struct
 				      ddestroy);
 		kref_get(&bo->list_kref);
 		list_move_tail(&bo->ddestroy, &removed);
-		spin_unlock(&glob->lru_lock);
 
-		reservation_object_lock(bo->resv, NULL);
+		if (remove_all || bo->resv != &bo->ttm_resv) {
+			spin_unlock(&glob->lru_lock);
+			reservation_object_lock(bo->resv, NULL);
 
-		spin_lock(&glob->lru_lock);
-		ttm_bo_cleanup_refs(bo, false, !remove_all, true);
+			spin_lock(&glob->lru_lock);
+			ttm_bo_cleanup_refs(bo, false, !remove_all, true);
+
+		} else if (reservation_object_trylock(bo->resv)) {
+			ttm_bo_cleanup_refs(bo, false, !remove_all, true);
+		}
 
 		kref_put(&bo->list_kref, ttm_bo_release_list);
 		spin_lock(&glob->lru_lock);