Quoting Tvrtko Ursulin (2020-11-17 12:44:00)
>
> On 17/11/2020 11:30, Chris Wilson wrote:
> > Since we wake the GT up before executing a request, and go to sleep as
> > soon as it is retired, the GT wake time not only represents how long the
> > device is powered up, but also provides a summary, albeit an overestimate,
> > of the device runtime.
> >
> > Signed-off-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
> > ---
> > drivers/gpu/drm/i915/gt/intel_gt_pm.c | 45 ++++++++++++++++++++++++
> > drivers/gpu/drm/i915/gt/intel_gt_pm.h | 2 ++
> > drivers/gpu/drm/i915/gt/intel_gt_types.h | 24 +++++++++++++
> > drivers/gpu/drm/i915/i915_debugfs.c | 2 ++
> > drivers/gpu/drm/i915/i915_pmu.c | 6 ++++
> > include/uapi/drm/i915_drm.h | 3 +-
> > 6 files changed, 81 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/gpu/drm/i915/gt/intel_gt_pm.c b/drivers/gpu/drm/i915/gt/intel_gt_pm.c
> > index 274aa0dd7050..dd2f88bed65a 100644
> > --- a/drivers/gpu/drm/i915/gt/intel_gt_pm.c
> > +++ b/drivers/gpu/drm/i915/gt/intel_gt_pm.c
> > @@ -39,6 +39,24 @@ static void user_forcewake(struct intel_gt *gt, bool suspend)
> > intel_gt_pm_put(gt);
> > }
> >
> > +static void runtime_begin(struct intel_gt *gt)
> > +{
> > + write_seqcount_begin(>->stats.lock);
> > + gt->stats.start = ktime_get();
> > + gt->stats.active = true;
> > + write_seqcount_end(>->stats.lock);
> > +}
> > +
> > +static void runtime_end(struct intel_gt *gt)
> > +{
> > + write_seqcount_begin(>->stats.lock);
> > + gt->stats.active = false;
> > + gt->stats.total =
> > + ktime_add(gt->stats.total,
> > + ktime_sub(ktime_get(), gt->stats.start));
> > + write_seqcount_end(>->stats.lock);
> > +}
> > +
> > static int __gt_unpark(struct intel_wakeref *wf)
> > {
> > struct intel_gt *gt = container_of(wf, typeof(*gt), wakeref);
> > @@ -67,6 +85,7 @@ static int __gt_unpark(struct intel_wakeref *wf)
> > i915_pmu_gt_unparked(i915);
> >
> > intel_gt_unpark_requests(gt);
> > + runtime_begin(gt);
> >
> > return 0;
> > }
> > @@ -79,6 +98,7 @@ static int __gt_park(struct intel_wakeref *wf)
> >
> > GT_TRACE(gt, "\n");
> >
> > + runtime_end(gt);
> > intel_gt_park_requests(gt);
> >
> > i915_vma_parked(gt);
> > @@ -106,6 +126,7 @@ static const struct intel_wakeref_ops wf_ops = {
> > void intel_gt_pm_init_early(struct intel_gt *gt)
> > {
> > intel_wakeref_init(>->wakeref, gt->uncore->rpm, &wf_ops);
> > + seqcount_mutex_init(>->stats.lock, >->wakeref.mutex);
> > }
> >
> > void intel_gt_pm_init(struct intel_gt *gt)
> > @@ -339,6 +360,30 @@ int intel_gt_runtime_resume(struct intel_gt *gt)
> > return intel_uc_runtime_resume(>->uc);
> > }
> >
> > +static ktime_t __intel_gt_get_busy_time(const struct intel_gt *gt)
> > +{
> > + ktime_t total = gt->stats.total;
> > +
> > + if (gt->stats.active)
> > + total = ktime_add(total,
> > + ktime_sub(ktime_get(), gt->stats.start));
> > +
> > + return total;
> > +}
> > +
> > +ktime_t intel_gt_get_busy_time(const struct intel_gt *gt)
> > +{
> > + unsigned int seq;
> > + ktime_t total;
> > +
> > + do {
> > + seq = read_seqcount_begin(>->stats.lock);
>
> Any specific reasons for read_seqcount_being vs read_seqbegin etc, the
> latter being used in engine stats seems to have some kcsan integration
> and former a bit more low level.
The brand spanking new seqlock.h offers a lot more variety in lockdep
analysis, in particular we declare these as being guarded by the
wakeref.mutex (see seqlock_mutex_init()). read_seqcount_being() is not
as low-level as it once was, and now includes the lockdep and kcsan
checks (and is quite picky about being init'ed correctly.)
> > + total = __intel_gt_get_busy_time(gt);
> > + } while (read_seqcount_retry(>->stats.lock, seq));
> > +
> > + return total;
> > +}
>
> I wish there was an easy way to extract some sort of struct stats
> between engine stats and this and have common helpers, but sadly structs
> are not exactly identical.
Nor do they get used from exactly the same locking contexts.
And context-stats are very similar but different again. :|
> > #if IS_ENABLED(CONFIG_DRM_I915_SELFTEST)
> > #include "selftest_gt_pm.c"
> > #endif
> > diff --git a/drivers/gpu/drm/i915/gt/intel_gt_pm.h b/drivers/gpu/drm/i915/gt/intel_gt_pm.h
> > index 60f0e2fbe55c..aa8f2cda946b 100644
> > --- a/drivers/gpu/drm/i915/gt/intel_gt_pm.h
> > +++ b/drivers/gpu/drm/i915/gt/intel_gt_pm.h
> > @@ -58,6 +58,8 @@ int intel_gt_resume(struct intel_gt *gt);
> > void intel_gt_runtime_suspend(struct intel_gt *gt);
> > int intel_gt_runtime_resume(struct intel_gt *gt);
> >
> > +ktime_t intel_gt_get_busy_time(const struct intel_gt *gt);
> > +
> > static inline bool is_mock_gt(const struct intel_gt *gt)
> > {
> > return I915_SELFTEST_ONLY(gt->awake == -ENODEV);
> > diff --git a/drivers/gpu/drm/i915/gt/intel_gt_types.h b/drivers/gpu/drm/i915/gt/intel_gt_types.h
> > index 6d39a4a11bf3..c7bde529feab 100644
> > --- a/drivers/gpu/drm/i915/gt/intel_gt_types.h
> > +++ b/drivers/gpu/drm/i915/gt/intel_gt_types.h
> > @@ -87,6 +87,30 @@ struct intel_gt {
> >
> > u32 pm_guc_events;
> >
> > + struct {
> > + bool active;
> > +
> > + /**
> > + * @lock: Lock protecting the below fields.
> > + */
> > + seqcount_mutex_t lock;
> > +
> > + /**
> > + * @total: Total time this engine was busy.
> > + *
> > + * Accumulated time not counting the most recent block in cases
> > + * where engine is currently busy (active > 0).
> > + */
> > + ktime_t total;
> > +
> > + /**
> > + * @start: Timestamp of the last idle to active transition.
> > + *
> > + * Idle is defined as active == 0, active is active > 0.
> > + */
> > + ktime_t start;
> > + } stats;
> > +
> > struct intel_engine_cs *engine[I915_NUM_ENGINES];
> > struct intel_engine_cs *engine_class[MAX_ENGINE_CLASS + 1]
> > [MAX_ENGINE_INSTANCE + 1];
> > diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c
> > index 77e76b665098..337293c7bb7d 100644
> > --- a/drivers/gpu/drm/i915/i915_debugfs.c
> > +++ b/drivers/gpu/drm/i915/i915_debugfs.c
> > @@ -1316,6 +1316,8 @@ static int i915_engine_info(struct seq_file *m, void *unused)
> > seq_printf(m, "GT awake? %s [%d]\n",
> > yesno(dev_priv->gt.awake),
> > atomic_read(&dev_priv->gt.wakeref.count));
> > + seq_printf(m, "GT busy: %llu ms\n",
> > + ktime_to_ms(intel_gt_get_busy_time(&dev_priv->gt)));
>
> Would it be worth putting something in debugfs_gt_pm.c as well?
Closest we have at the moment is rps_boost_show, and that's a bit stale.
I was expecting we would have another engines under gt/; maybe one day.
>
> > seq_printf(m, "CS timestamp frequency: %u Hz\n",
> > RUNTIME_INFO(dev_priv)->cs_timestamp_frequency_hz);
> >
> > diff --git a/drivers/gpu/drm/i915/i915_pmu.c b/drivers/gpu/drm/i915/i915_pmu.c
> > index cd786ad12be7..36fc60cf5725 100644
> > --- a/drivers/gpu/drm/i915/i915_pmu.c
> > +++ b/drivers/gpu/drm/i915/i915_pmu.c
> > @@ -488,6 +488,8 @@ config_status(struct drm_i915_private *i915, u64 config)
> > if (!HAS_RC6(i915))
> > return -ENODEV;
> > break;
> > + case I915_PMU_BUSY_TIME:
> > + break;
> > default:
> > return -ENOENT;
> > }
> > @@ -595,6 +597,9 @@ static u64 __i915_pmu_event_read(struct perf_event *event)
> > case I915_PMU_RC6_RESIDENCY:
> > val = get_rc6(&i915->gt);
> > break;
> > + case I915_PMU_BUSY_TIME:
> > + val = ktime_to_ns(intel_gt_get_busy_time(&i915->gt));
> > + break;
> > }
> > }
> >
> > @@ -898,6 +903,7 @@ create_event_attributes(struct i915_pmu *pmu)
> > __event(I915_PMU_REQUESTED_FREQUENCY, "requested-frequency", "M"),
> > __event(I915_PMU_INTERRUPTS, "interrupts", NULL),
> > __event(I915_PMU_RC6_RESIDENCY, "rc6-residency", "ns"),
> > + __event(I915_PMU_BUSY_TIME, "busy-time", "ns"),
> > };
> > static const struct {
> > enum drm_i915_pmu_engine_sample sample;
> > diff --git a/include/uapi/drm/i915_drm.h b/include/uapi/drm/i915_drm.h
> > index fa1f3d62f9a6..b66b7c1fd564 100644
> > --- a/include/uapi/drm/i915_drm.h
> > +++ b/include/uapi/drm/i915_drm.h
> > @@ -177,8 +177,9 @@ enum drm_i915_pmu_engine_sample {
> > #define I915_PMU_REQUESTED_FREQUENCY __I915_PMU_OTHER(1)
> > #define I915_PMU_INTERRUPTS __I915_PMU_OTHER(2)
> > #define I915_PMU_RC6_RESIDENCY __I915_PMU_OTHER(3)
> > +#define I915_PMU_BUSY_TIME __I915_PMU_OTHER(4)
>
> I'd be tempted to call this I915_PMU_GT_BUSY_TIME - or even better awake
> time?
awake is better than busy, but this is the same level as RC6_RESIDENCY,
so I was following its trend.
> > -#define I915_PMU_LAST I915_PMU_RC6_RESIDENCY
> > +#define I915_PMU_LAST I915_PMU_BUSY_TIME
> >
> > /* Each region is a minimum of 16k, and there are at most 255 of them.
> > */
> >
>
> Code looks fine, but should we be adding a pmu counter?
It's the inverse of rc6 residency; so I can already put it to use.
Yes, it's useful.
-Chris
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
