On Mon, 19 Aug 2013 11:35:32 -0400
Steven Rostedt <rostedt@xxxxxxxxxxx> wrote:
> From: Steven Rostedt <rostedt@xxxxxxxxxxx>
>
> There's a race condition with swait wakeups and adding to the list. The
> __swait_wake() does a check for swait_head_has_waiters(), and if it is
> empty it will exit without doing any wake ups. The problem is that the
> check does not include any memory barriers before it makes a decision
> to wake up or not.
>
> CPU0 CPU1
> ---- ----
>
> condition = 1
>
> load h->list (is empty)
> raw_spin_lock(hlist->lock)
> hlist_add();
> __set_current_state();
> raw_spin_unlock(hlist->lock)
> swait_wake()
> swait_head_has_waiters()
> (sees h->list as empty and returns)
>
BTW, the race still exists if you move the raw_spin_unlock(hlist->lock)
above to here. That is:
swait_wake()
swait_head_has_waiters()
(sees h->list as empty and returns)
raw_spin_unlock(hlist->lock)
Maybe this will help to understand it more.
-- Steve
> check_condition (sees condition = 0)
>
> store condition = 1
>
> schedule()
>
> Now the task on CPU1 has just missed its wakeup. By adding a memory
> barrier before the list empty check, we fix the problem of miss seeing
> the list not empty as well as pushing out the condition for the other
> task to see.
>
> Reviewed-by: Paul E. McKenney <paulmck@xxxxxxxxxxxxxxxxxx>
> Signed-off-by: Steven Rostedt <rostedt@xxxxxxxxxxx>
--
To unsubscribe from this list: send the line "unsubscribe linux-rt-users" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
