drm/i915/pmu: Wire up engine busy stats to PMU

We can use engine busy stats instead of the sampling timer for
better accuracy.

By doing this we replace the stohastic sampling with busyness
metric derived directly from engine activity. This is context
switch interrupt driven, so as accurate as we can get from
software tracking.

As a secondary benefit, we can also not run the sampling timer
in cases only busyness metric is enabled.

v2: Rebase.
v3:
 * Rebase, comments.
 * Leave engine busyness controls out of workers.
v4: Checkpatch cleanup.
v5: Added comment to pmu_needs_timer change.
v6:
 * Rebase.
 * Fix style of some comments. (Chris Wilson)
v7: Rebase and commit message update. (Chris Wilson)
v8: Add delayed stats disabling to improve accuracy in face of
    CPU hotplug events.
v9: Rebase.
v10: Rebase - i915_modparams.enable_execlists removal.

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20171121181852.16128-6-tvrtko.ursulin@linux.intel.com

1 files changed, 76 insertions, 2 deletions

diff --git a/drivers/gpu/drm/i915/i915_pmu.c b/drivers/gpu/drm/i915/i915_pmu.c
index fb95f0ac30ea..6a428e7218d2 100644
--- a/drivers/gpu/drm/i915/i915_pmu.c
+++ b/drivers/gpu/drm/i915/i915_pmu.c
@@ -90,6 +90,11 @@ static unsigned int event_enabled_bit(struct perf_event *event)
 	return config_enabled_bit(event->attr.config);
 }
 
+static bool supports_busy_stats(struct drm_i915_private *i915)
+{
+	return INTEL_GEN(i915) >= 8;
+}
+
 static bool pmu_needs_timer(struct drm_i915_private *i915, bool gpu_active)
 {
 	u64 enable;
@@ -115,6 +120,12 @@ static bool pmu_needs_timer(struct drm_i915_private *i915, bool gpu_active)
 	 */
 	if (!gpu_active)
 		enable &= ~ENGINE_SAMPLE_MASK;
+	/*
+	 * Also there is software busyness tracking available we do not
+	 * need the timer for I915_SAMPLE_BUSY counter.
+	 */
+	else if (supports_busy_stats(i915))
+		enable &= ~BIT(I915_SAMPLE_BUSY);
 
 	/*
 	 * If some bits remain it means we need the sampling timer running.
@@ -363,6 +374,9 @@ static u64 __i915_pmu_event_read(struct perf_event *event)
 
 		if (WARN_ON_ONCE(!engine)) {
 			/* Do nothing */
+		} else if (sample == I915_SAMPLE_BUSY &&
+			   engine->pmu.busy_stats) {
+			val = ktime_to_ns(intel_engine_get_busy_time(engine));
 		} else {
 			val = engine->pmu.sample[sample].cur;
 		}
@@ -399,6 +413,12 @@ again:
 	local64_add(new - prev, &event->count);
 }
 
+static bool engine_needs_busy_stats(struct intel_engine_cs *engine)
+{
+	return supports_busy_stats(engine->i915) &&
+	       (engine->pmu.enable & BIT(I915_SAMPLE_BUSY));
+}
+
 static void i915_pmu_enable(struct perf_event *event)
 {
 	struct drm_i915_private *i915 =
@@ -438,7 +458,21 @@ static void i915_pmu_enable(struct perf_event *event)
 
 		GEM_BUG_ON(sample >= I915_PMU_SAMPLE_BITS);
 		GEM_BUG_ON(engine->pmu.enable_count[sample] == ~0);
-		engine->pmu.enable_count[sample]++;
+		if (engine->pmu.enable_count[sample]++ == 0) {
+			/*
+			 * Enable engine busy stats tracking if needed or
+			 * alternatively cancel the scheduled disable.
+			 *
+			 * If the delayed disable was pending, cancel it and
+			 * in this case do not enable since it already is.
+			 */
+			if (engine_needs_busy_stats(engine) &&
+			    !engine->pmu.busy_stats) {
+				engine->pmu.busy_stats = true;
+				if (!cancel_delayed_work(&engine->pmu.disable_busy_stats))
+					intel_enable_engine_stats(engine);
+			}
+		}
 	}
 
 	/*
@@ -451,6 +485,14 @@ static void i915_pmu_enable(struct perf_event *event)
 	spin_unlock_irqrestore(&i915->pmu.lock, flags);
 }
 
+static void __disable_busy_stats(struct work_struct *work)
+{
+	struct intel_engine_cs *engine =
+	       container_of(work, typeof(*engine), pmu.disable_busy_stats.work);
+
+	intel_disable_engine_stats(engine);
+}
+
 static void i915_pmu_disable(struct perf_event *event)
 {
 	struct drm_i915_private *i915 =
@@ -474,8 +516,26 @@ static void i915_pmu_disable(struct perf_event *event)
 		 * Decrement the reference count and clear the enabled
 		 * bitmask when the last listener on an event goes away.
 		 */
-		if (--engine->pmu.enable_count[sample] == 0)
+		if (--engine->pmu.enable_count[sample] == 0) {
 			engine->pmu.enable &= ~BIT(sample);
+			if (!engine_needs_busy_stats(engine) &&
+			    engine->pmu.busy_stats) {
+				engine->pmu.busy_stats = false;
+				/*
+				 * We request a delayed disable to handle the
+				 * rapid on/off cycles on events, which can
+				 * happen when tools like perf stat start, in a
+				 * nicer way.
+				 *
+				 * In addition, this also helps with busy stats
+				 * accuracy with background CPU offline/online
+				 * migration events.
+				 */
+				queue_delayed_work(system_wq,
+						   &engine->pmu.disable_busy_stats,
+						   round_jiffies_up_relative(HZ));
+			}
+		}
 	}
 
 	GEM_BUG_ON(bit >= I915_PMU_MASK_BITS);
@@ -702,6 +762,8 @@ static void i915_pmu_unregister_cpuhp_state(struct drm_i915_private *i915)
 
 void i915_pmu_register(struct drm_i915_private *i915)
 {
+	struct intel_engine_cs *engine;
+	enum intel_engine_id id;
 	int ret;
 
 	if (INTEL_GEN(i915) <= 2) {
@@ -723,6 +785,10 @@ void i915_pmu_register(struct drm_i915_private *i915)
 	hrtimer_init(&i915->pmu.timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 	i915->pmu.timer.function = i915_sample;
 
+	for_each_engine(engine, i915, id)
+		INIT_DELAYED_WORK(&engine->pmu.disable_busy_stats,
+				  __disable_busy_stats);
+
 	ret = perf_pmu_register(&i915->pmu.base, "i915", -1);
 	if (ret)
 		goto err;
@@ -742,6 +808,9 @@ err:
 
 void i915_pmu_unregister(struct drm_i915_private *i915)
 {
+	struct intel_engine_cs *engine;
+	enum intel_engine_id id;
+
 	if (!i915->pmu.base.event_init)
 		return;
 
@@ -749,6 +818,11 @@ void i915_pmu_unregister(struct drm_i915_private *i915)
 
 	hrtimer_cancel(&i915->pmu.timer);
 
+	for_each_engine(engine, i915, id) {
+		GEM_BUG_ON(engine->pmu.busy_stats);
+		flush_delayed_work(&engine->pmu.disable_busy_stats);
+	}
+
 	i915_pmu_unregister_cpuhp_state(i915);
 
 	perf_pmu_unregister(&i915->pmu.base);