| Message ID | 20220207131932.2164560-2-mperttunen@nvidia.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [1/2] gpu: host1x: Always return syncpoint value when waiting | expand |
diff --git a/drivers/gpu/host1x/bus.c b/drivers/gpu/host1x/bus.c index bdee16a0bb8e..66ba04501c94 100644 --- a/drivers/gpu/host1x/bus.c +++ b/drivers/gpu/host1x/bus.c @@ -938,9 +938,6 @@ struct host1x_bo_mapping *host1x_bo_pin(struct device *dev, struct host1x_bo *bo mapping->cache = cache; list_add_tail(&mapping->entry, &cache->mappings); - - /* bump reference count to track the copy in the cache */ - kref_get(&mapping->ref); } unlock:
The recently introduced buffer cache is causing cached BOs to leak, as cache entries are dropped when a BO's refcount goes to zero but the cache itself keeps a reference to each cache BO, causing a circular reference. Fix this by not taking a reference to the cached BOs within the cache itself. Fixes: 1f39b1dfa53c ("drm/tegra: Implement buffer object cache") Signed-off-by: Mikko Perttunen <mperttunen@nvidia.com> --- NOTE As is, applying this patch reveals a BO refcounting issue in the DC driver. A fix for that bug should be applied before applying this patch. I believe Thierry will be posting a fix for that shortly. --- drivers/gpu/host1x/bus.c | 3 --- 1 file changed, 3 deletions(-)