net/sched: taprio: taprio_offload_config_changed() is protected by rtnl_mutex
authorVladimir Oltean <vladimir.oltean@nxp.com>
Thu, 15 Sep 2022 10:50:40 +0000 (13:50 +0300)
committerJakub Kicinski <kuba@kernel.org>
Tue, 20 Sep 2022 20:53:33 +0000 (13:53 -0700)
The locking in taprio_offload_config_changed() is wrong (but also
inconsequentially so). The current_entry_lock does not serialize changes
to the admin and oper schedules, only to the current entry. In fact, the
rtnl_mutex does that, and that is taken at the time when taprio_change()
is called.

Replace the rcu_dereference_protected() method with the proper RCU
annotation, and drop the unnecessary spin lock.

Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
net/sched/sch_taprio.c

index db88a69..e9f57ef 100644 (file)
@@ -1193,16 +1193,10 @@ static void taprio_offload_config_changed(struct taprio_sched *q)
 {
        struct sched_gate_list *oper, *admin;
 
-       spin_lock(&q->current_entry_lock);
-
-       oper = rcu_dereference_protected(q->oper_sched,
-                                        lockdep_is_held(&q->current_entry_lock));
-       admin = rcu_dereference_protected(q->admin_sched,
-                                         lockdep_is_held(&q->current_entry_lock));
+       oper = rtnl_dereference(q->oper_sched);
+       admin = rtnl_dereference(q->admin_sched);
 
        switch_schedules(q, &admin, &oper);
-
-       spin_unlock(&q->current_entry_lock);
 }
 
 static u32 tc_map_to_queue_mask(struct net_device *dev, u32 tc_mask)