On Fri, Jan 6, 2017 at 1:02 PM, Nicolai Hähnle <nhaeh...@gmail.com> wrote: > On 02.01.2017 21:20, Marek Olšák wrote: >> >> From: Marek Olšák <marek.ol...@amd.com> >> >> The CS thread is needed to ensure proper ordering of operations and can't >> be disabled (without complicating the code). >> >> Discovered by Nine CSMT, which ended up in a deadlock. > > > I'm curious why the thread makes a difference for the deadlock. Why isn't it > enough in the un-threaded case to extend the scope of the ws->bo_fence_lock > to cover the submit ioctl call?
You can't extend the bo_fence lock to the CS thread, so the main thread and the CS thread can be racing. Marek _______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev
