Re: Fix for wrong OOM killer trigger?

[Andrea Arcangeli <andrea@suse.de>]

On Mon, Sep 22, 2003 at 12:11:40PM +0200, Roland Bless wrote:
> was a fix for the particular bug. My suggestion is that the log entry
> below describes the bug fix for it:

the kmem cleanup wasn't a bug. So in theory it could be even the leaking
of pagetables that went from -aa to mainline in 23pre1, but I think it
really was the removal of the oom killer with the -aa VM merges that
went into 2.4.23pre[2-5] that really fixed your problem (if it's true
that you had no swap, which I understood it's the case, and no swap puts
at the light the brokeness of the oom killer), that leak is a minor one,
many other places shrinks the per-cpu queues, so it's unlikely to be
able to leak lots of ram in a misc workload.

It's good to hear that pre5 is fixed. thanks.


> 
> Summary of changes from v2.4.22 to v2.4.23-pre1
> ============================================
> ...
> Marc-Christian Petersen:
>   o Cleanup kmem_cache_reap()
> or was it related to this one:
>   o Avoid potentially leaking pagetables into the per-cpu queues
> 
> I hope that it was also fixed in 2.6, or is there a different mechanism
> used?

dunno, but the oom killer certainly has not enough information in 2.6
either to be able to do a reliable decision.

Andrea - If you prefer relying on open source software, check these links:
	    rsync.kernel.org::pub/scm/linux/kernel/bkcvs/linux-2.[45]/
	    http://www.cobite.com/cvsps/
	    svn://svn.kernel.org/linux-2.[46]/trunk