lkml.org 
[lkml]   [2006]   [Apr]   [27]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH 1/2 (repost)] mm: serialize OOM kill operations
Dave Peterson <dsp@llnl.gov> wrote:
>
> The patch below modifies the behavior of the OOM killer so that only
> one OOM kill operation can be in progress at a time. When running a
> test program that eats lots of memory, I was observing behavior where
> the OOM killer gets impatient and shoots one or more system daemons
> in addition to the program that is eating lots of memory. This fixes
> the problematic behavior.
>
> ...
>
> @@ -379,6 +380,15 @@ void mmput(struct mm_struct *mm)
> spin_unlock(&mmlist_lock);
> }
> put_swap_token(mm);
> +
> + if (unlikely(test_bit(MM_FLAG_OOM_NOTIFY, &mm->flags)))
> + /* Terminate a pending OOM kill operation. No tasks
> + * actually spin on the lock. Tasks only do
> + * spin_trylock() (and abort OOM kill operation if
> + * lock is already taken).
> + */
> + spin_unlock(&oom_kill_lock);
> +

Gad. I guess if we're going to do this then a better implementation would
be to use test_and_set_bit(some_unsigned_long). And perhaps call some
oom_kill.c interface function here rather than directly accessing
oom-killer data structures (could be an inlined function).

But the broader question is "what do we want to do here".

If we've picked a task and we've signalled it then the right thing to do
would appear to be just to block all tasks as they enter the oom-killer.
Send them to sleep until the killed task actually exits. But

a) memory can become free (or reclaimable) for other reasons, so those
now-sleeping tasks shouldn't be sleeping any more (this is a minor
problem).

b) one of the sleeping tasks may be holding a lock which prevents the
killed task from reaching do_exit(). This is a showstopper.

But I think b) stops your show as well:

- task A enters the oom-killer, decides to kill task Z.

- task A holds a lock which is preventing task Z from exitting

- but oom_kill_lock is now held. task A just keeps on trying to reclaim
memory and trying (and failing) to kill tasks.

- user hits reset button.


IOW: we just have to keep killing more tasks until something happens.

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2006-04-28 01:01    [from the cache]
©2003-2014 Jasper Spaans. Advertise on this site