On Mon, 2014-03-10 at 13:47 -0400, Theodore Ts'o wrote: 

On Fri, Feb 21, 2014 at 02:54:12PM +0100, Mike Galbraith wrote:

quoted

ext3: dbench1 queues a buffer, blocks on journal mutex, it's plug is not
pulled.  dbench2 mutex owner is waiting for kjournald, who is waiting for
the buffer queued by dbench1.  Game over.

Where is in ext3/4 are we calling some function which could end up
blocking on kjournald while we have the I/O queue plugged?  That
sounds suspicious and potentially wrong.

I don't have the crash dumps and analysis handy, this was quite some
time ago.  Problem is that..

static inline void sched_submit_work(struct task_struct *tsk)
{
        if (!tsk->state || tsk_is_pi_blocked(tsk))
                return;
        /*
         * If we are going to sleep and we have plugged IO queued,
         * make sure to submit it to avoid deadlocks.
         */
        if (blk_needs_flush_plug(tsk))
                blk_schedule_flush_plug(tsk);
}

..tsk_is_pi_blocked(tsk) leaves us with IO queued, dependency on which
can (_did_ for ext[34] and xfs that I recall) end up with our waker
waiting on our IO.  There were other deadlock scenarios, not only the
one in the quoted text.

-Mike