If somebody does a hash lookup for cfq_queue while ioprio of an async queue
is elevated, they shouldn't end up stuck with lowered ioprio when we go back.
Fix is to use ->org_ioprio{,class} in hash lookups.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
 
        hlist_for_each_safe(entry, next, hash_list) {
                struct cfq_queue *__cfqq = list_entry_qhash(entry);
-               const unsigned short __p = IOPRIO_PRIO_VALUE(__cfqq->ioprio_class, __cfqq->ioprio);
+               const unsigned short __p = IOPRIO_PRIO_VALUE(__cfqq->org_ioprio_class, __cfqq->org_ioprio);
 
                if (__cfqq->key == key && (__p == prio || prio == CFQ_KEY_ANY))
                        return __cfqq;