[PATCH 2/4] drm/i915: Grab the spinlock whilst checking for a stuck waiter

[Chris Wilson]

If we allow ourselves to take the engine->breadcrumbs.lock, we can
reduce the race when declaring a stuck waiter (and so reduce the number
of false positives for "missed interrupts").

Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_irq.c | 17 +++++++++++++++--
 1 file changed, 15 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
index f4e7acb97731..636d8105ec63 100644
--- a/drivers/gpu/drm/i915/i915_irq.c
+++ b/drivers/gpu/drm/i915/i915_irq.c
@@ -3057,13 +3057,27 @@ ring_stuck(struct intel_engine_cs *engine, u64 acthd)
 	return HANGCHECK_HUNG;
 }
 
+static bool missed_breadcrumb(struct intel_engine_cs *engine)
+{
+	struct intel_breadcrumbs *b = &engine->breadcrumbs;
+	bool missed = false;
+
+	spin_lock(&b->lock);
+	if (b->first_wait)
+		missed = i915_seqno_passed(intel_engine_get_seqno(engine),
+					   b->first_wait->seqno);
+	spin_unlock(&b->lock);
+
+	return missed;
+}
+
 static unsigned long kick_waiters(struct intel_engine_cs *engine)
 {
 	struct drm_i915_private *i915 = engine->i915;
 	unsigned long irq_count = READ_ONCE(engine->breadcrumbs.irq_wakeups);
 
-	rcu_read_lock();
 	if (engine->hangcheck.user_interrupts == irq_count &&
+	    missed_breadcrumb(engine) &&
 	    intel_engine_wakeup(engine) &&
 	    !test_and_set_bit(engine->id, &i915->gpu_error.missed_irq_rings)) {
 		if (!test_bit(engine->id, &i915->gpu_error.test_irq_rings))
@@ -3072,7 +3086,6 @@ static unsigned long kick_waiters(struct intel_engine_cs *engine)
 
 		intel_engine_enable_fake_irq(engine);
 	}
-	rcu_read_unlock();
 
 	return irq_count;
 }
-- 
2.8.1