From: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxx> We are not allowed to call intel_runtime_pm_get from the PMU counter read callback since the former can sleep, and the latter is running under IRQ context. To workaround this, we start our timer when we detect that we have failed to obtain a runtime PM reference during read, and approximate the growing RC6 counter from the timer. Once the timer manages to obtain the runtime PM reference, we stop the timer and go back to the above described behaviour. We have to be careful not to overshoot the RC6 estimate, so once resumed after a period of approximation, we only update the counter once it catches up. With the observation that RC6 is increasing while the device is suspended, this should not pose a problem and can only cause slight inaccuracies due clock base differences. Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxx> Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=104943 Fixes: 6060b6aec03c ("drm/i915/pmu: Add RC6 residency metrics") Testcase: igt/perf_pmu/rc6-runtime-pm Cc: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxx> Cc: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx> Cc: Imre Deak <imre.deak@xxxxxxxxx> Cc: Jani Nikula <jani.nikula@xxxxxxxxxxxxxxx> Cc: Joonas Lahtinen <joonas.lahtinen@xxxxxxxxxxxxxxx> Cc: Rodrigo Vivi <rodrigo.vivi@xxxxxxxxx> Cc: David Airlie <airlied@xxxxxxxx> Cc: intel-gfx@xxxxxxxxxxxxxxxxxxxxx Cc: dri-devel@xxxxxxxxxxxxxxxxxxxxx --- drivers/gpu/drm/i915/i915_pmu.c | 149 ++++++++++++++++++++++++++++++---------- drivers/gpu/drm/i915/i915_pmu.h | 1 + 2 files changed, 114 insertions(+), 36 deletions(-) diff --git a/drivers/gpu/drm/i915/i915_pmu.c b/drivers/gpu/drm/i915/i915_pmu.c index 1c440460255d..dca41c072a7c 100644 --- a/drivers/gpu/drm/i915/i915_pmu.c +++ b/drivers/gpu/drm/i915/i915_pmu.c @@ -90,23 +90,16 @@ static unsigned int event_enabled_bit(struct perf_event *event) return config_enabled_bit(event->attr.config); } -static bool pmu_needs_timer(struct drm_i915_private *i915, bool gpu_active) +static bool +__pmu_needs_timer(struct drm_i915_private *i915, u64 enable, bool gpu_active) { - u64 enable; - - /* - * Only some counters need the sampling timer. - * - * We start with a bitmask of all currently enabled events. - */ - enable = i915->pmu.enable; - /* - * Mask out all the ones which do not need the timer, or in + * Mask out all events which do not need the timer, or in * other words keep all the ones that could need the timer. */ enable &= config_enabled_mask(I915_PMU_ACTUAL_FREQUENCY) | config_enabled_mask(I915_PMU_REQUESTED_FREQUENCY) | + config_enabled_mask(I915_PMU_RC6_RESIDENCY) | ENGINE_SAMPLE_MASK; /* @@ -130,6 +123,11 @@ static bool pmu_needs_timer(struct drm_i915_private *i915, bool gpu_active) return enable; } +static bool pmu_needs_timer(struct drm_i915_private *i915, bool gpu_active) +{ + return __pmu_needs_timer(i915, i915->pmu.enable, gpu_active); +} + void i915_pmu_gt_parked(struct drm_i915_private *i915) { if (!i915->pmu.base.event_init) @@ -181,20 +179,20 @@ update_sample(struct i915_pmu_sample *sample, u32 unit, u32 val) sample->cur += mul_u32_u32(val, unit); } -static void engines_sample(struct drm_i915_private *dev_priv) +static bool engines_sample(struct drm_i915_private *dev_priv) { struct intel_engine_cs *engine; enum intel_engine_id id; bool fw = false; if ((dev_priv->pmu.enable & ENGINE_SAMPLE_MASK) == 0) - return; + return false; if (!dev_priv->gt.awake) - return; + return false; if (!intel_runtime_pm_get_if_in_use(dev_priv)) - return; + return false; for_each_engine(engine, dev_priv, id) { u32 current_seqno = intel_engine_get_seqno(engine); @@ -225,10 +223,51 @@ static void engines_sample(struct drm_i915_private *dev_priv) if (fw) intel_uncore_forcewake_put(dev_priv, FORCEWAKE_ALL); - intel_runtime_pm_put(dev_priv); + return true; +} + +static u64 read_rc6_residency(struct drm_i915_private *i915) +{ + u64 val; + + val = intel_rc6_residency_ns(i915, IS_VALLEYVIEW(i915) ? + VLV_GT_RENDER_RC6 : GEN6_GT_GFX_RC6); + if (HAS_RC6p(i915)) + val += intel_rc6_residency_ns(i915, GEN6_GT_GFX_RC6p); + if (HAS_RC6pp(i915)) + val += intel_rc6_residency_ns(i915, GEN6_GT_GFX_RC6pp); + + return val; +} + +static void +update_rc6_sample(struct drm_i915_private *i915, u64 val, bool locked) +{ + unsigned long flags; + + if (!locked) + spin_lock_irqsave(&i915->pmu.lock, flags); + + /* + * Update stored RC6 counter only if it is greater than the current + * value. This deals with periods of runtime suspend during which we are + * estimating the RC6 residency, so do not want to overshoot the real + * value read once the device is woken up. + */ + if (val > i915->pmu.sample[__I915_SAMPLE_RC6].cur) + i915->pmu.sample[__I915_SAMPLE_RC6].cur = val; + + /* We don't need to sample RC6 from the timer any more. */ + i915->pmu.timer_enabled = + __pmu_needs_timer(i915, + i915->pmu.enable & ~config_enabled_mask(I915_PMU_RC6_RESIDENCY), + READ_ONCE(i915->gt.awake)); + + if (!locked) + spin_unlock_irqrestore(&i915->pmu.lock, flags); } -static void frequency_sample(struct drm_i915_private *dev_priv) +static bool others_sample(struct drm_i915_private *dev_priv, bool pm) { if (dev_priv->pmu.enable & config_enabled_mask(I915_PMU_ACTUAL_FREQUENCY)) { @@ -236,10 +275,10 @@ static void frequency_sample(struct drm_i915_private *dev_priv) val = dev_priv->gt_pm.rps.cur_freq; if (dev_priv->gt.awake && - intel_runtime_pm_get_if_in_use(dev_priv)) { + (pm || intel_runtime_pm_get_if_in_use(dev_priv))) { + pm = true; val = intel_get_cagf(dev_priv, I915_READ_NOTRACE(GEN6_RPSTAT1)); - intel_runtime_pm_put(dev_priv); } update_sample(&dev_priv->pmu.sample[__I915_SAMPLE_FREQ_ACT], @@ -252,18 +291,48 @@ static void frequency_sample(struct drm_i915_private *dev_priv) intel_gpu_freq(dev_priv, dev_priv->gt_pm.rps.cur_freq)); } + + if (dev_priv->pmu.enable & + config_enabled_mask(I915_PMU_RC6_RESIDENCY)) { + if (pm || intel_runtime_pm_get_if_in_use(dev_priv)) { + update_rc6_sample(dev_priv, + read_rc6_residency(dev_priv), + false); + pm = true; + } else { + unsigned long flags; + + /* + * When device is runtime suspended we assume RC6 + * residency is increasing by the sampling timer period. + */ + spin_lock_irqsave(&dev_priv->pmu.lock, flags); + dev_priv->pmu.sample[__I915_SAMPLE_RC6].cur += PERIOD; + spin_unlock_irqrestore(&dev_priv->pmu.lock, flags); + } + } + + return pm; + } static enum hrtimer_restart i915_sample(struct hrtimer *hrtimer) { struct drm_i915_private *i915 = container_of(hrtimer, struct drm_i915_private, pmu.timer); + bool pm; if (!READ_ONCE(i915->pmu.timer_enabled)) return HRTIMER_NORESTART; - engines_sample(i915); - frequency_sample(i915); + pm = engines_sample(i915); + pm = others_sample(i915, pm); + + if (pm) + intel_runtime_pm_put(i915); + + if (!READ_ONCE(i915->pmu.timer_enabled)) + return HRTIMER_NORESTART; hrtimer_forward_now(hrtimer, ns_to_ktime(PERIOD)); return HRTIMER_RESTART; @@ -415,7 +484,7 @@ static int i915_pmu_event_init(struct perf_event *event) return 0; } -static u64 __i915_pmu_event_read(struct perf_event *event) +static u64 __i915_pmu_event_read(struct perf_event *event, bool locked) { struct drm_i915_private *i915 = container_of(event->pmu, typeof(*i915), pmu.base); @@ -453,18 +522,26 @@ static u64 __i915_pmu_event_read(struct perf_event *event) val = count_interrupts(i915); break; case I915_PMU_RC6_RESIDENCY: - intel_runtime_pm_get(i915); - val = intel_rc6_residency_ns(i915, - IS_VALLEYVIEW(i915) ? - VLV_GT_RENDER_RC6 : - GEN6_GT_GFX_RC6); - if (HAS_RC6p(i915)) - val += intel_rc6_residency_ns(i915, - GEN6_GT_GFX_RC6p); - if (HAS_RC6pp(i915)) - val += intel_rc6_residency_ns(i915, - GEN6_GT_GFX_RC6pp); - intel_runtime_pm_put(i915); + if (intel_runtime_pm_get_if_in_use(i915)) { + update_rc6_sample(i915, + read_rc6_residency(i915), + locked); + intel_runtime_pm_put(i915); + } else { + unsigned long flags; + + /* + * If we failed to read the actual value, start + * the timer which will be estimating it while + * device is suspended. + */ + if (!locked) + spin_lock_irqsave(&i915->pmu.lock, flags); + __i915_pmu_maybe_start_timer(i915); + if (!locked) + spin_unlock_irqrestore(&i915->pmu.lock, flags); + } + val = i915->pmu.sample[__I915_SAMPLE_RC6].cur; break; } } @@ -479,7 +556,7 @@ static void i915_pmu_event_read(struct perf_event *event) again: prev = local64_read(&hwc->prev_count); - new = __i915_pmu_event_read(event); + new = __i915_pmu_event_read(event, false); if (local64_cmpxchg(&hwc->prev_count, prev, new) != prev) goto again; @@ -534,7 +611,7 @@ static void i915_pmu_enable(struct perf_event *event) * for all listeners. Even when the event was already enabled and has * an existing non-zero value. */ - local64_set(&event->hw.prev_count, __i915_pmu_event_read(event)); + local64_set(&event->hw.prev_count, __i915_pmu_event_read(event, true)); spin_unlock_irqrestore(&i915->pmu.lock, flags); } diff --git a/drivers/gpu/drm/i915/i915_pmu.h b/drivers/gpu/drm/i915/i915_pmu.h index 5a2e013a56bb..249983ed3f08 100644 --- a/drivers/gpu/drm/i915/i915_pmu.h +++ b/drivers/gpu/drm/i915/i915_pmu.h @@ -27,6 +27,7 @@ enum { __I915_SAMPLE_FREQ_ACT = 0, __I915_SAMPLE_FREQ_REQ, + __I915_SAMPLE_RC6, __I915_NUM_PMU_SAMPLERS }; -- 2.14.1 _______________________________________________ dri-devel mailing list dri-devel@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/dri-devel
