2016-10-19 Christian König <deathsimple@xxxxxxxxxxx>: > Am 19.10.2016 um 19:48 schrieb Gustavo Padovan: > > From: Gustavo Padovan <gustavo.padovan@xxxxxxxxxxxxxxx> > > > > When creating fence arrays we were not holding references to the fences > > in the array, however when destroy the array we were putting away a > > reference to these fences. > > > > This patch hold the ref for all fences in the array when creating the > > array. > > > > It then removes the code that was holding the fences on both amdgpu_vm and > > sync_file. For sync_file, specially, we worked on small referencing > > refactor for sync_file_merge(). > > > > Signed-off-by: Gustavo Padovan <gustavo.padovan@xxxxxxxxxxxxxxx> > > I would prefer it to keep it like it is, cause this is a bit inconsistent. > > With this change the fence array takes the ownership of the array, but not > of the fences inside it. I was thinking more in to keep consistency between all fence users. Every user should hold a ref to the fence assigned to it. That is what patch 1 is doing for sync_file and think it is a good idea do the same here. The array itself is not refcounted and the users calling fence_array_create() doesn't store the allocated array anywhere. The comment I errouneously removed already states that. Gustavo _______________________________________________ dri-devel mailing list dri-devel@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/dri-devel
