> On Mon, 9 Jul 2001, Mike Galbraith wrote:
>
> > On 9 Jul 2001, Christoph Rohland wrote:
> >
[snip]
> > > > So, did I fix it or just bust it in a convenient manner ;-)
> > >
> > > ... now you drop random pages. This of course helps reducing memory
> > > pressure ;-)
> >
> > (shoot. I figured that was too easy to be right)
> >
> > > But still this may be a hint. You are not running out of swap, aren't
> > > you?
> >
> > I'm running oom whether I have swap enabled or not. The inactive
> > dirty list starts growing forever, until it's full of (aparantly)
> > dirty pages and I'm utterly oom.
>
> We can make sure if this (inactive full of dirty pages) is really the case
> with the tracing code.
The problem turned out to be KDE. It opens/unlinks two files in /tmp
per terminal, and lseeks/writes terminal output to them continually.
With tmpfs/ramfs mounted on /tmp....
> The shmem fix in 2.4.7-pre5 is the solution for your problem ?
No. Now, the pages go to the active list. The solution is to not
use KDE-2.1's terminals if I fire up X ;-) It doesn't livelock at
oom anymore though.. thrashes uncontrollably until sysrq-e fixes it's
attitude problem.
(I never saw this problem before because I rarely run X, and never do
anything which generates enough terminal output to oom the box if I do
run it. I only ran the tar thing by chance, trying to figure out why
people were reporting updatedb causing massive swapping, when it didn't
do that here.. shrug)
> If not, I'll port the tracing code to the pre5 and hopefully we can
> actually figure out what is going on here.
No need. What looked like a really weird kernel bug turned out to be
a userland bug triggering oom livelock bug.
-Mike
-
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/