On 28/11/2017 12:48, Chris Wilson wrote:
Quoting Tvrtko Ursulin (2017-11-28 12:41:27)From: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxx> Move the execlists specific setup out of intel_engine_setup_common. This was supposed to be only for backend agnostic bits. At the same time rename it to intel_engine_setup_execlist to follow the setup vs init naming convetion we have. Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxx> --- +static void +intel_engine_setup_execlist(struct intel_engine_cs *engine) +{ + struct intel_engine_execlists * const execlists = &engine->execlists; + + execlists->csb_use_mmio = csb_force_mmio(engine->i915); + + execlists->port_mask = 1; + BUILD_BUG_ON_NOT_POWER_OF_2(execlists_num_ports(execlists)); + GEM_BUG_ON(execlists_num_ports(execlists) > EXECLIST_MAX_PORTS); + + execlists->queue = RB_ROOT; + execlists->first = NULL; +}The only problem here was that we wanted to be sure that some fields were initialised for the common paths, i.e. so we could iterate over the queue without worrying first if it was execlists (if it wasn't execlists the queue would be empty). Now, I think we could just rely on zero initialisation, but that was the rationale for it ending up early. Now we could split it between setup_execlists and init_execlists if we want the pedantry.
Common paths as in ringbuffer submission? I grepped around and don't see it used there.
Then about setup vs init, we said init is for hw access so I don't follow how you would split the above?
Regards, Tvrtko _______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/intel-gfx
