Quoting Tvrtko Ursulin (2019-04-10 12:06:49) > > On 10/04/2019 11:13, Chris Wilson wrote: > > Quoting Tvrtko Ursulin (2019-04-10 11:05:13) > >> And some lockdep_assert_held in all three? > > > > Read on :( > > > > The plan is for intel_context_enter/_exit to be under the > > timeline->mutex, but that isn't realised for about another 30 patches. > > > > mark_active has special protection because it gets used from the > > serialised portion of intel_engine_park. > > Ok, but bug on for zero active_count makes sense right? Maybe even > intel_context_pin_active to signify that it can only act on an already > active context?
It get's called on the kernel_context with ce->active_count==0 and not even holding the mutex from inside the wakeref handling code. (On no, I've invented a new BKL --- well a parking brake.) -Chris _______________________________________________ Intel-gfx mailing list Intel-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/intel-gfx
