The introduction of Jens Axboe's explicit i/o plugging patches introduced a
deadlock in jfs.  This was caused by the process initiating I/O not
unplugging the queue before waiting on the commit thread.  The commit
thread itself was waiting for that I/O to complete.  Calling io_schedule()
rather than schedule() unplugs the I/O queue avoiding the deadlock, and it
appears to be the right function to call in any case.
Signed-off-by: Dave Kleikamp <shaggy@austin.ibm.com>
                if (cond)                               \
                        break;                          \
                unlock_cmd;                             \
-               schedule();                             \
+               io_schedule();                          \
                lock_cmd;                               \
        }                                               \
        current->state = TASK_RUNNING;                  \
 
                set_current_state(TASK_UNINTERRUPTIBLE);
                if (metapage_locked(mp)) {
                        unlock_page(mp->page);
-                       schedule();
+                       io_schedule();
                        lock_page(mp->page);
                }
        } while (trylock_metapage(mp));
 
        add_wait_queue(event, &wait);
        set_current_state(TASK_UNINTERRUPTIBLE);
        TXN_UNLOCK();
-       schedule();
+       io_schedule();
        current->state = TASK_RUNNING;
        remove_wait_queue(event, &wait);
 }