Re: per-process shared information

[Marcelo Tosatti]

On Fri, Oct 15, 2004 at 01:58:45AM +0200, Andrea Arcangeli wrote:
> On Thu, Oct 14, 2004 at 06:47:11PM -0300, Marcelo Tosatti wrote:
> > 
> > Hi Andrea!
> > 
> > No useful comments on the statm reporting issue.
> > 
> > > Ps. if somebody like Hugh volunteers implementing it, you're very
> > > welcome, just let me know (I'll eventually want to work on the oom
> > > handling too, which is pretty screwed right now, 
> > 
> > Yes, we've got reports of bad OOM killing behaviour (is that what you're
> > talking about?) 
> > 
> > One thing is the removal of "if (nr_swap_pages > 0) goto out" from oom_kill() 
> > causes problems (spurious oom kill). 
> > 
> > We need to throttle more, on page reclaiming progress I think.
> > 
> > Take a look at 
> > 
> > http://marc.theaimsgroup.com/?l=linux-mm&m=109587921204602&w=2
> > 
> > What else you're seeing?
> > 
> > > I've plenty of bugs
> > > open on that area and the lowmem zone protection needs a rewrite too to
> > > be set to a sane default value no matter the pages_lows etc..).
> > 
> > Nick has been working on that lately I think. What is the problem?
> 
> things went worse with the switch from 2.6.8 to 2.6.9-rc, so that's not
> the nr_swap_pages > 0, likely the latest changes introduced regressions
> instead of fixing them.

Just FYI - removing the "nr_swap_pages > 0" fixes the problem at
the URL I posted above.

But having it creates hard locks on Oracle workloads (wli removed 
that line) due to pinned memory.

> I'm seeing both hard deadlocks and suprious oom kills, and that all
> makes sense, I can see the bugs, it's just I need to fix them, my plan
> is to forward port some code from 2.4 which works fine, objrmap will make
> it even better.

Ok, very nice!

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