On 11/23/15 11:41 AM, Gustavo Padovan wrote: > + - remove sw_sync, it is used only for testing/debugging and should not be > +upstreamed. > + - port sw_sync testcases to use debugfs somehow
A quick but important nitpick: sw_sync itself is just an in-kernel helper for creating fences, when you don't have something like sync timeline primitives baked into your hardware. CONFIG_SW_SYNC_USER adds the interface for creating and signaling sw_sync objects from userspace. This is the part that's dangerous and only intended for testing, etc. AFAIK CONFIG_SW_SYNC_USER is the only part people have been objecting to. I'm fine with removing it. Removing the kernel-facing side of sw_sync would be a problem for us, since many drivers use it to create their fences.
