[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20151027162047.GK9891@dhcp22.suse.cz>
Date: Tue, 27 Oct 2015 17:20:47 +0100
From: Michal Hocko <mhocko@...nel.org>
To: Aristeu Rozanski <arozansk@...hat.com>
Cc: linux-kernel@...r.kernel.org, Greg Thelen <gthelen@...gle.com>,
Johannes Weiner <hannes@...xchg.org>, linux-mm@...ck.org,
cgroups@...r.kernel.org
Subject: Re: [PATCH] oom_kill: add option to disable dump_stack()
On Tue 27-10-15 11:43:42, Aristeu Rozanski wrote:
> Hi Michal,
> On Tue, Oct 27, 2015 at 09:09:21AM +0100, Michal Hocko wrote:
> > On Mon 26-10-15 13:40:49, Aristeu Rozanski wrote:
> > > Hi Michal,
> > > On Mon, Oct 26, 2015 at 06:20:12PM +0100, Michal Hocko wrote:
> > [...]
> > > > Would it make more sense to distinguish different parts of the OOM
> > > > report by loglevel properly?
> > > > pr_err - killed task report
> > > > pr_warning - oom invocation + memory info
> > > > pr_notice - task list
> > > > pr_info - stack trace
> > >
> > > That'd work, yes, but I'd think the stack trace would be pr_debug. At a
> > > point that you suspect the OOM killer isn't doing the right thing picking
> > > up tasks and you need more information.
> >
> > Stack trace should be independent on the oom victim selection because
> > the selection should be as much deterministic as possible - so it should
> > only depend on the memory consumption. I do agree that the exact trace
> > is not very useful for the (maybe) majority of OOM reports. I am trying
> > to remember when it was really useful the last time and have trouble to
> > find an example. So I would tend to agree that pr_debug would me more
> > suitable.
>
> Only problem I see so far with this approach is that it'll require
> reworing show_stack() on all architectures in order to be able to pass
> and use log level and I'm wondering if it's something people will find
> useful for other uses.
Yes this is a mess. But I think it is worth cleaning up.
dump_stack_print_info (arch independent) has a log level parameter.
show_stack_log_lvl (x86) has a loglevel parameter which is unused.
I haven't checked other architectures but the transition doesn't have to
be all at once I guess.
--
Michal Hocko
SUSE Labs
--
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