On Wed, Oct 03, 2018 at 11:29:32AM -0400, Dennis Dalessandro wrote: > On 10/2/2018 4:54 AM, Leon Romanovsky wrote: > > On Tue, Oct 02, 2018 at 11:39:43AM +0300, Kamal Heib wrote: > > > On Tue, Oct 02, 2018 at 11:07:38AM +0300, Leon Romanovsky wrote: > > > > On Tue, Oct 02, 2018 at 10:52:12AM +0300, Kamal Heib wrote: > > > > > To avoid checking if add_gid() and del_gid() are implemented adding a new > > > > > device cap flag that will indicate whether a specific device does > > > > > support managing his RoCE GIDs table or not. > > > > > > > > > > > > > Why? What will it give us? > > > > > > > > Thanks > > > > > > I plan to work on something similar to "verbs_context_ops" that > > > was introduced in rdma-core, so, this change will drop the > > > dependency on checking if the providers implemented the > > > call-backs or not. > > > > I would be happy to see this patch as part of whole feature, because > > this specific patch gives nothing to us in this current form. > > > Agree with Leon here. Patch has to stand on it's own, or at least come along > with something that makes it relevant. I also generally dislike the idea to test a bit to check if the function you are about to invoke is NULL or not. ops are to be NULL if they are not supported. Jason
