Quoting Chris Wilson (2018-04-25 13:45:45) > Quoting Chris Wilson (2018-04-25 12:31:29) > > Quoting Chris Wilson (2018-04-25 12:23:30) > > > Quoting Mika Kuoppala (2018-04-25 12:19:08) > > > > Did you try with WA_TAIL_DWORDS 16? > > > > > > Sure can try, but the error state doesn't indicate TAIL==HEAD as would > > > be the issue with WaIdleLiteRestore (restoring to an idle ring wouldn't > > > generate the arbitration event, so no more context switches). > > > > Watch https://patchwork.freedesktop.org/series/42284/ > > > > Actually surviving on my bdw > > And what appears to work just as well is > > @@ -1387,7 +1387,7 @@ static int execlists_request_alloc(struct i915_request *request) > */ > request->reserved_space += EXECLISTS_REQUEST_SIZE; > > - ret = intel_ring_wait_for_space(request->ring, request->reserved_space); > + ret = intel_ring_cacheline_align(request); > if (ret) > return ret; > > If that doesn't hint towards some weirdness in HW, I don't know what > would ;) Fwiw, I've now hit GPU hangs with WA_TAIL_DWORDS set to 16 and with the cacheline_align; scratch that theory. -Chris _______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/intel-gfx
