Quoting Tvrtko Ursulin (2019-09-12 10:55:00) > > On 12/09/2019 10:39, Chris Wilson wrote: > > Quoting Tvrtko Ursulin (2019-09-12 10:20:39) > >> Don't we end up doing the irqsave spinlock needlessly when !CONFIG_PM? > > > > No, the intent is to serialise with i915_pmu_gt_parked and > > i915_pmu_gt_unparked (and the GT awake state), which are independent of > > CONFIG_PM. > > Yes but with !CONFIG_PM we can always read the real counters and don't > need to do any additional magic. In fact code in i915_pmu_gt_(un)parked > could be ifdef-ed out for that case as well.
Oh, you mean if we didn't have to worry about runtime-pm at all for mmio access. I was not thinking of that at all, just balancing parked vs sample. -Chris _______________________________________________ Intel-gfx mailing list Intelemail@example.com https://lists.freedesktop.org/mailman/listinfo/intel-gfx