drm/i915/pmu: Cheat when reading the actual frequency to avoid fw
authorChris Wilson <chris@chris-wilson.co.uk>
Fri, 8 Nov 2019 10:35:10 +0000 (10:35 +0000)
committerChris Wilson <chris@chris-wilson.co.uk>
Fri, 8 Nov 2019 11:36:10 +0000 (11:36 +0000)
We want to avoid taking forcewake when querying the performance stats,
as we wish to avoid perturbing the system under observation. (And with
the forcewake being kept alive for 1ms after use, sampling the frequency
from a 200Hz timer keeps forcewake 40% active.)

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191108103511.20951-1-chris@chris-wilson.co.uk
drivers/gpu/drm/i915/i915_pmu.c

index 0539501..8e74f40 100644 (file)
@@ -366,8 +366,21 @@ frequency_sample(struct intel_gt *gt, unsigned int period_ns)
 
                val = rps->cur_freq;
                if (intel_gt_pm_get_if_awake(gt)) {
-                       val = intel_uncore_read_notrace(uncore, GEN6_RPSTAT1);
-                       val = intel_get_cagf(rps, val);
+                       u32 stat;
+
+                       /*
+                        * We take a quick peek here without using forcewake
+                        * so that we don't perturb the system under observation
+                        * (forcewake => !rc6 => increased power use). We expect
+                        * that if the read fails because it is outside of the
+                        * mmio power well, then it will return 0 -- in which
+                        * case we assume the system is running at the intended
+                        * frequency. Fortunately, the read should rarely fail!
+                        */
+                       stat = intel_uncore_read_fw(uncore, GEN6_RPSTAT1);
+                       if (stat)
+                               val = intel_get_cagf(rps, stat);
+
                        intel_gt_pm_put(gt);
                }