[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <Pine.LNX.4.64.0710231305050.20095@schroedinger.engr.sgi.com>
Date: Tue, 23 Oct 2007 13:09:31 -0700 (PDT)
From: Christoph Lameter <clameter@....com>
To: Pekka Enberg <penberg@...helsinki.fi>
cc: Alexey Dobriyan <adobriyan@...il.com>,
linux-kernel@...r.kernel.org, linux-mm@...r.kernel.org
Subject: Re: SLUB 0:1 SLAB (OOM during massive parallel kernel builds)
On Tue, 23 Oct 2007, Pekka Enberg wrote:
> On 10/23/07, Christoph Lameter <clameter@....com> wrote:
> > Not sure what this is. Maybe the slowing SLUB solves the race.
>
> What kind of race are you thinking of? What I initially thought was
> that the problem is that SLUB messes up some other VM heuristics due
> to different object sizes, not holding on to empty slabs, and/or page
> allocator pass-through. I guess only object size is affected by
> debugging, though?
The number of objects per page is reduced by enabling full debugging. That
triggers a potential of more order 1 allocations but we are failing at
order 0 allocs. slub_debug=F does not increase object size with debugging
information but keeps things as they are while doing as much verification
of slab integrity as possible.
One--potentially far fetched theory--is that the difference in alloc
performance may trigger a race. But I guess it is more likely that we will
find something wrong with the gfp flags (that I recategorized for 2.6.24).
-
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/
Powered by blists - more mailing lists