Re: [patch 3/5] oom: less memdie

[Nick Piggin <nickpiggin@yahoo.com.au>]

Andrew Morton wrote:

>On Thu, 12 Oct 2006 16:10:01 +0200 (CEST)
>Nick Piggin <npiggin@suse.de> wrote:
>
>
>>Don't cause all threads in all other thread groups to gain TIF_MEMDIE
>>otherwise we'll get a thundering herd eating out memory reserve. This
>>may not be the optimal scheme, but it fits our policy of allowing just
>>one TIF_MEMDIE in the system at once.
>>
>>Signed-off-by: Nick Piggin <npiggin@suse.de>
>>
>>Index: linux-2.6/mm/oom_kill.c
>>===================================================================
>>--- linux-2.6.orig/mm/oom_kill.c
>>+++ linux-2.6/mm/oom_kill.c
>>@@ -322,11 +322,12 @@ static int oom_kill_task(struct task_str
>> 
>> 	/*
>> 	 * kill all processes that share the ->mm (i.e. all threads),
>>-	 * but are in a different thread group.
>>+	 * but are in a different thread group. Don't let them have access
>>+	 * to memory reserves though, otherwise we might deplete all memory.
>> 	 */
>> 	do_each_thread(g, q) {
>> 		if (q->mm == mm && q->tgid != p->tgid)
>>-			__oom_kill_task(q, 1);
>>+			force_sig(SIGKILL, p);
>> 	} while_each_thread(g, q);
>> 
>>
>
>Curious.  How much testing did you do of this stuff?  I assume there were
>some observed problems.  What were they, and what was the observed effect
>of these changes?
>

This change I actually didn't really test because I don't have any apps 
to speak
of which use multiple thread groups.

I stumbled on it by inspection when trying to fix the killing of 
OOM_DISABLE tasks.
Basically -- we don't set TIF_MEMDIE or boost the priority of *any* 
other thread in
our same group, so we shouldn't do it for *all* other threds of all 
other groups.

Consider an OOM situation, where there will likely be a lot of threads 
stuck in
__alloc_pages. If a large number of these suddenly get a big timeslice 
and full
access to memory reserves, they'll eat into more than we'd like.

Now I'm not sure that our current OOM killing / memory reserving scheme is
perfect -- indeed if we only allow a single TIF_MEMDIE thread at once, 
we can
get deadlocks. However, that's the direction we've chosen, and it seems 
to work
reasonably well. Mostly.

This is just an enforcement of that policy rather than a change in 
direction.
Criticism is always welcome though.

--

Send instant messages to your online friends http://au.messenger.yahoo.com 

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>