drm/i915: Lift runtime-pm acquire callbacks out of intel_wakeref.mutex

When runtime pm is first woken, it will synchronously call the
registered callbacks for the device. These callbacks
may pull in their own forest of locks, which we do not want to
conflate with the intel_wakeref.mutex. A second minor benefit to
reducing the coverage of the mutex, is that it will reduce
contention for frequent sleeps and wakes (such as when being used
for soft-rc6).

v2: remove usage of fetch_and_zero() and other improvements(Jani)
Signed-off-by: Chris Wilson <chris.p.wilson@intel.com>
Signed-off-by: Nirmoy Das <nirmoy.das@intel.com>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230926083742.14740-2-nirmoy.das@intel.com

drivers/gpu/drm/i915/intel_wakeref.c

@@ -10,21 +10,12 @@
 #include "intel_wakeref.h"
 #include "i915_drv.h"
 
-static void rpm_get(struct intel_wakeref *wf)
-{
-	wf->wakeref = intel_runtime_pm_get(&wf->i915->runtime_pm);
-}
-
-static void rpm_put(struct intel_wakeref *wf)
-{
-	intel_wakeref_t wakeref = fetch_and_zero(&wf->wakeref);
-
-	intel_runtime_pm_put(&wf->i915->runtime_pm, wakeref);
-	INTEL_WAKEREF_BUG_ON(!wakeref);
-}
-
 int __intel_wakeref_get_first(struct intel_wakeref *wf)
 {
+	intel_wakeref_t wakeref;
+	int ret = 0;
+
+	wakeref = intel_runtime_pm_get(&wf->i915->runtime_pm);
 	/*
 	 * Treat get/put as different subclasses, as we may need to run
 	 * the put callback from under the shrinker and do not want to
@@ -32,41 +23,52 @@ int __intel_wakeref_get_first(struct intel_wakeref *wf)
 	 * upon acquiring the wakeref.
 	 */
 	mutex_lock_nested(&wf->mutex, SINGLE_DEPTH_NESTING);
-	if (!atomic_read(&wf->count)) {
-		int err;
-
-		rpm_get(wf);
 
-		err = wf->ops->get(wf);
-		if (unlikely(err)) {
-			rpm_put(wf);
-			mutex_unlock(&wf->mutex);
-			return err;
+	if (!atomic_read(&wf->count)) {
+		INTEL_WAKEREF_BUG_ON(wf->wakeref);
+		wf->wakeref = wakeref;
+		wakeref = 0;
+
+		ret = wf->ops->get(wf);
+		if (ret) {
+			wakeref = xchg(&wf->wakeref, 0);
+			wake_up_var(&wf->wakeref);
+			goto unlock;
 		}
 
 		smp_mb__before_atomic(); /* release wf->count */
 	}
+
 	atomic_inc(&wf->count);
+	INTEL_WAKEREF_BUG_ON(atomic_read(&wf->count) <= 0);
+
+unlock:
 	mutex_unlock(&wf->mutex);
+	if (unlikely(wakeref))
+		intel_runtime_pm_put(&wf->i915->runtime_pm, wakeref);
 
-	INTEL_WAKEREF_BUG_ON(atomic_read(&wf->count) <= 0);
-	return 0;
+	return ret;
 }
 
 static void ____intel_wakeref_put_last(struct intel_wakeref *wf)
 {
+	intel_wakeref_t wakeref = 0;
+
 	INTEL_WAKEREF_BUG_ON(atomic_read(&wf->count) <= 0);
 	if (unlikely(!atomic_dec_and_test(&wf->count)))
 		goto unlock;
 
 	/* ops->put() must reschedule its own release on error/deferral */
 	if (likely(!wf->ops->put(wf))) {
-		rpm_put(wf);
+		INTEL_WAKEREF_BUG_ON(!wf->wakeref);
+		wakeref = xchg(&wf->wakeref, 0);
 		wake_up_var(&wf->wakeref);
 	}
 
 unlock:
 	mutex_unlock(&wf->mutex);
+	if (wakeref)
+		intel_runtime_pm_put(&wf->i915->runtime_pm, wakeref);
 }
 
 void __intel_wakeref_put_last(struct intel_wakeref *wf, unsigned long flags)