* walking this with preempt disabled. In all the failure paths, we
* call synchronize_sched(), but we don't want to slow down the success
* path, so use actual RCU here.
+ * Note that module_alloc() on most architectures creates W+X page
+ * mappings which won't be cleaned up until do_free_init() runs. Any
+ * code such as mark_rodata_ro() which depends on those mappings to
+ * be cleaned up needs to sync with the queued work - ie
+ * rcu_barrier_sched()
*/
call_rcu_sched(&freeinit->rcu, do_free_init);
mutex_unlock(&module_mutex);