[PATCH 3/3] drm/i915/pmu: Use a delayed wakeref put in get_rc6()

[Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>]

get_rc6() has the same runqueue lock deadlock as guc_engine_busyness():
it is called from the PMU callbacks, which perf can invoke from the
scheduler with the runqueue lock held, and intel_gt_pm_put_async() there
can wake a workqueue worker via try_to_wake_up().

Use intel_gt_pm_put_delay() so the deferred put arms a timer instead.

The remaining async puts in atomic contexts (hrtimer, tasklet, irq_work,
G2H handlers) cannot run with the runqueue lock held and are left alone.

Fixes: 07779a76ee1f ("drm/i915: Mark up the calling context for intel_wakeref_put()")
Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Assisted-by: Claude:claude-opus-5
---
v2: Reword comment in code (Vinay)
---
 drivers/gpu/drm/i915/i915_pmu.c | 11 ++++++++++-
 1 file changed, 10 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/i915/i915_pmu.c b/drivers/gpu/drm/i915/i915_pmu.c
index 9ea5626399e5..0ebf1620519b 100644
--- a/drivers/gpu/drm/i915/i915_pmu.c
+++ b/drivers/gpu/drm/i915/i915_pmu.c
@@ -229,7 +229,16 @@ static u64 get_rc6(struct intel_gt *gt)
 	wakeref = intel_gt_pm_get_if_awake(gt);
 	if (wakeref) {
 		val = __get_rc6(gt);
-		intel_gt_pm_put_async(gt, wakeref);
+		/*
+		 * Perf framework may invoke this from the scheduler with the
+		 * runqueue lock held. If we are the last to drop the reference
+		 * here, then we cannot use intel_gt_pm_put_async since that can
+		 * result in waking up a task from this context which would try
+		 * to re-acquire the runqueue lock. To avoid such a deadlock,
+		 * defer the put on a timer so that waking the task can happen
+		 * from a timer context.
+		 */
+		intel_gt_pm_put_delay(gt, wakeref, 1);
 	}
 
 	spin_lock_irqsave(&pmu->lock, flags);
-- 
2.55.0