PM: runtime: Add safety net to supplier device release

Because refcount_dec_not_one() returns true if the target refcount
becomes saturated, it is generally unsafe to use its return value as
a loop termination condition, but that is what happens when a device
link's supplier device is released during runtime PM suspend
operations and on device link removal.

To address this, introduce pm_runtime_release_supplier() to be used
in the above cases which will check the supplier device's runtime
PM usage counter in addition to the refcount_dec_not_one() return
value, so the loop can be terminated in case the rpm_active refcount
value becomes invalid, and update the code in question to use it as
appropriate.

This change is not expected to have any visible functional impact.

Reported-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>

2 files changed, 31 insertions, 13 deletions

diff --git a/drivers/base/core.c b/drivers/base/core.c
index fd034d742447..b191bd17de89 100644
--- a/drivers/base/core.c
+++ b/drivers/base/core.c
@@ -485,8 +485,7 @@ static void device_link_release_fn(struct work_struct *work)
 	/* Ensure that all references to the link object have been dropped. */
 	device_link_synchronize_removal();
 
-	while (refcount_dec_not_one(&link->rpm_active))
-		pm_runtime_put(link->supplier);
+	pm_runtime_release_supplier(link, true);
 
 	put_device(link->consumer);
 	put_device(link->supplier);
diff --git a/drivers/base/power/runtime.c b/drivers/base/power/runtime.c
index 844b6b811610..f1b856ea8e10 100644
--- a/drivers/base/power/runtime.c
+++ b/drivers/base/power/runtime.c
@@ -305,19 +305,40 @@ static int rpm_get_suppliers(struct device *dev)
 	return 0;
 }
 
+/**
+ * pm_runtime_release_supplier - Drop references to device link's supplier.
+ * @link: Target device link.
+ * @check_idle: Whether or not to check if the supplier device is idle.
+ *
+ * Drop all runtime PM references associated with @link to its supplier device
+ * and if @check_idle is set, check if that device is idle (and so it can be
+ * suspended).
+ */
+void pm_runtime_release_supplier(struct device_link *link, bool check_idle)
+{
+	struct device *supplier = link->supplier;
+
+	/*
+	 * The additional power.usage_count check is a safety net in case
+	 * the rpm_active refcount becomes saturated, in which case
+	 * refcount_dec_not_one() would return true forever, but it is not
+	 * strictly necessary.
+	 */
+	while (refcount_dec_not_one(&link->rpm_active) &&
+	       atomic_read(&supplier->power.usage_count) > 0)
+		pm_runtime_put_noidle(supplier);
+
+	if (check_idle)
+		pm_request_idle(supplier);
+}
+
 static void __rpm_put_suppliers(struct device *dev, bool try_to_suspend)
 {
 	struct device_link *link;
 
 	list_for_each_entry_rcu(link, &dev->links.suppliers, c_node,
-				device_links_read_lock_held()) {
-
-		while (refcount_dec_not_one(&link->rpm_active))
-			pm_runtime_put_noidle(link->supplier);
-
-		if (try_to_suspend)
-			pm_request_idle(link->supplier);
-	}
+				device_links_read_lock_held())
+		pm_runtime_release_supplier(link, try_to_suspend);
 }
 
 static void rpm_put_suppliers(struct device *dev)
@@ -1777,9 +1798,7 @@ void pm_runtime_drop_link(struct device_link *link)
 		return;
 
 	pm_runtime_drop_link_count(link->consumer);
-
-	while (refcount_dec_not_one(&link->rpm_active))
-		pm_runtime_put(link->supplier);
+	pm_runtime_release_supplier(link, true);
 }
 
 static bool pm_runtime_need_not_resume(struct device *dev)