On Wed, Jul 2, 2014 at 7:37 AM, Greg KH gregkh@linuxfoundation.org wrote:
Android can expose fences to userspace. It's possible to make the new fence mechanism expose the same fences to userspace by changing sync_fence_create to take a struct fence instead of a struct sync_pt. No other change is needed, because only the fence parts of struct sync_pt are used. But because the userspace fences are a separate problem and I haven't really looked at it yet I feel it should stay in staging, for now.
Ok, that's reasonable.
At first glance, this all looks "sane" to me, any objection from anyone if I merge this through my driver-core tree for 3.17?
Ack from my side fwiw.
Just for the record (again) about android syncpts. I think using android syncpts stuff as the official userspace interfaces for fences for userspace that wants to do explicit syncing is the sane approach - after all the (only) big platform using explicit fencing is Android since opencl is kinda not yet there really on linux (which would be the other guy really interested in explicit fencing).
But before we de-stage android syncpts (and so bake in the userspace abi forever) I really want to see a full upstream gpu driver implementation using fences and running on both X+DRI userspace (implicit syncing) and android (expclicit syncing), including good test coverage for corner-cases to make sure we've addressed all warts and hidden dragons. i915 looks like the most likely candidate to get there first (we have the code for both use-cases out-of-tree after all) so I'll keep on pestering people. No promises though ;-)
Cheers, Daniel