linux-kernel - Re: Improving OOM killer

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-Id: <20100203105236.b4a60754.kamezawa.hiroyu@jp.fujitsu.com>
Date:	Wed, 3 Feb 2010 10:52:36 +0900
From:	KAMEZAWA Hiroyuki <kamezawa.hiroyu@...fujitsu.com>
To:	David Rientjes <rientjes@...gle.com>
Cc:	Lubos Lunak <l.lunak@...e.cz>, linux-mm@...ck.org,
	linux-kernel@...r.kernel.org,
	Andrew Morton <akpm@...ux-foundation.org>,
	KOSAKI Motohiro <kosaki.motohiro@...fujitsu.com>,
	Balbir Singh <balbir@...ux.vnet.ibm.com>,
	Nick Piggin <npiggin@...e.de>, Jiri Kosina <jkosina@...e.cz>
Subject: Re: Improving OOM killer

On Tue, 2 Feb 2010 17:41:41 -0800 (PST)
David Rientjes <rientjes@...gle.com> wrote:

> On Tue, 2 Feb 2010, Lubos Lunak wrote:
> 
> > > > init
> > > >   |- kdeinit
> > > >   |  |- ksmserver
> > > >   |  |  |- kwin
> > > >   |  |- <other>
> > > >   |- konsole
> > > >      |- make
> > > >         |- sh
> > > >         |  |- meinproc4
> > > >         |- sh
> > > >         |  |- meinproc4
> > > >         |- <etc>
> > > >
> > > >  What happens is that OOM killer usually selects either ksmserver (KDE
> > > > session manager) or kdeinit (KDE master process that spawns most KDE
> > > > processes). Note that in either case OOM killer does not reach the point
> > > > of killing the actual offender - it will randomly kill in the tree under
> > > > kdeinit until it decides to kill ksmserver, which means terminating the
> > > > desktop session. As konsole is a KUniqueApplication, it forks into
> > > > background and gets reparented to init, thus getting away from the
> > > > kdeinit subtree. Since the memory pressure is distributed among several
> > > > meinproc4 processes, the badness does not get summed up in its make
> > > > grandparent, as badness() does this only for direct parents.
> > >
> > > There's no randomness involved in selecting a task to kill;
> > 
> >  That was rather a figure of speech, but even if you want to take it 
> > literally, then from the user's point of view it is random. Badness of 
> > kdeinit depends on the number of children it has spawned, badness of 
> > ksmserver depends for example on the number and size of windows open (as its 
> > child kwin is a window and compositing manager).
> > 
> 
> As I've mentioned, I believe Kame (now cc'd) is working on replacing the 
> heuristic that adds the VM size for children into the parent task's 
> badness score with a forkbomb detector. 

I stopped that as I mentioned. I'm heavily disappointed with myself and
would like not to touch oom-killer things for a while.

I'd like to conentrate on memcg for a while, which I've starved for these 3 months.

Then, you don't need to CC me.

Bye,
-Kame

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