[Intel-gfx] [PATCH 10/10] drm/i915/gt: Reduce GT runtime stats from seqlock to a latch

[Chris Wilson]

Since we can compute the elapsed time to add to the total, during the
PMU sample we only need to have a consistent view of the (start, total,
active) tuple to be able to locally determine the runtime. That can be
arrange by a pair of memory bariiers and carefully sequencing of the
writes and reads.

Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/gt/intel_gt_pm.c    | 34 ++++++------------------
 drivers/gpu/drm/i915/gt/intel_gt_types.h |  5 ----
 2 files changed, 8 insertions(+), 31 deletions(-)

diff --git a/drivers/gpu/drm/i915/gt/intel_gt_pm.c b/drivers/gpu/drm/i915/gt/intel_gt_pm.c
index c94e8ac884eb..49aba3588fe7 100644
--- a/drivers/gpu/drm/i915/gt/intel_gt_pm.c
+++ b/drivers/gpu/drm/i915/gt/intel_gt_pm.c
@@ -41,24 +41,18 @@ static void user_forcewake(struct intel_gt *gt, bool suspend)
 
 static void runtime_begin(struct intel_gt *gt)
 {
-	local_irq_disable();
-	write_seqcount_begin(&gt->stats.lock);
 	gt->stats.start = ktime_get();
-	gt->stats.active = true;
-	write_seqcount_end(&gt->stats.lock);
-	local_irq_enable();
+	smp_wmb(); /* pairs with intel_gt_get_busy_time() */
+	WRITE_ONCE(gt->stats.active, true);
 }
 
 static void runtime_end(struct intel_gt *gt)
 {
-	local_irq_disable();
-	write_seqcount_begin(&gt->stats.lock);
-	gt->stats.active = false;
 	gt->stats.total =
 		ktime_add(gt->stats.total,
 			  ktime_sub(ktime_get(), gt->stats.start));
-	write_seqcount_end(&gt->stats.lock);
-	local_irq_enable();
+	smp_wmb(); /* pairs with intel_gt_get_busy_time() */
+	WRITE_ONCE(gt->stats.active, false);
 }
 
 static int __gt_unpark(struct intel_wakeref *wf)
@@ -130,7 +124,6 @@ static const struct intel_wakeref_ops wf_ops = {
 void intel_gt_pm_init_early(struct intel_gt *gt)
 {
 	intel_wakeref_init(&gt->wakeref, gt->uncore->rpm, &wf_ops);
-	seqcount_mutex_init(&gt->stats.lock, &gt->wakeref.mutex);
 }
 
 void intel_gt_pm_init(struct intel_gt *gt)
@@ -364,26 +357,15 @@ int intel_gt_runtime_resume(struct intel_gt *gt)
 	return intel_uc_runtime_resume(&gt->uc);
 }
 
-static ktime_t __intel_gt_get_awake_time(const struct intel_gt *gt)
+ktime_t intel_gt_get_awake_time(const struct intel_gt *gt)
 {
 	ktime_t total = gt->stats.total;
 
-	if (gt->stats.active)
+	if (READ_ONCE(gt->stats.active)) {
+		smp_rmb(); /* pairs with runtime_begin/end */
 		total = ktime_add(total,
 				  ktime_sub(ktime_get(), gt->stats.start));
-
-	return total;
-}
-
-ktime_t intel_gt_get_awake_time(const struct intel_gt *gt)
-{
-	unsigned int seq;
-	ktime_t total;
-
-	do {
-		seq = read_seqcount_begin(&gt->stats.lock);
-		total = __intel_gt_get_awake_time(gt);
-	} while (read_seqcount_retry(&gt->stats.lock, seq));
+	}
 
 	return total;
 }
diff --git a/drivers/gpu/drm/i915/gt/intel_gt_types.h b/drivers/gpu/drm/i915/gt/intel_gt_types.h
index a83d3e18254d..4205c3f022ad 100644
--- a/drivers/gpu/drm/i915/gt/intel_gt_types.h
+++ b/drivers/gpu/drm/i915/gt/intel_gt_types.h
@@ -91,11 +91,6 @@ struct intel_gt {
 	struct {
 		bool active;
 
-		/**
-		 * @lock: Lock protecting the below fields.
-		 */
-		seqcount_mutex_t lock;
-
 		/**
 		 * @total: Total time this engine was busy.
 		 *
-- 
2.20.1