lkml.org 
[lkml]   [2016]   [Jun]   [16]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: [PATCH 07/10] mm, oom: fortify task_will_free_mem
From
Date
Michal Hocko wrote:
> On Sat 11-06-16 17:10:03, Tetsuo Handa wrote:
> > Michal Hocko wrote:
> > > > Also, I think setting TIF_MEMDIE on p when find_lock_task_mm(p) != p is
> > > > wrong. While oom_reap_task() will anyway clear TIF_MEMDIE even if we set
> > > > TIF_MEMDIE on p when p->mm == NULL, it is not true for CONFIG_MMU=n case.
> > >
> > > Yes this would be racy for !CONFIG_MMU but does it actually matter?
> >
> > I don't know because I've never used CONFIG_MMU=n kernels. But I think it
> > actually matters. You fixed this race by commit 83363b917a2982dd ("oom:
> > make sure that TIF_MEMDIE is set under task_lock").
>
> Yes and that commit was trying to address a highly theoretical issue
> reported by you. Let me quote:
> :oom_kill_process is currently prone to a race condition when the OOM
> :victim is already exiting and TIF_MEMDIE is set after the task releases
> :its address space. This might theoretically lead to OOM livelock if the
> :OOM victim blocks on an allocation later during exiting because it
> :wouldn't kill any other process and the exiting one won't be able to exit.
> :The situation is highly unlikely because the OOM victim is expected to
> :release some memory which should help to sort out OOM situation.
>
> Even if such a race is possible it wouldn't be with the oom
> reaper. Regarding CONFIG_MMU=n I am even less sure it is possible and
> I would rather focus on CONFIG_MMU=y where we know that problems exist
> rather than speculating about something as special as nommu which even
> might not care at all.

I still don't like it. current->mm == NULL in

- if (current->mm &&
- (fatal_signal_pending(current) || task_will_free_mem(current))) {
+ if (task_will_free_mem(current)) {

is not highly unlikely. You obviously break commit d7a94e7e11badf84
("oom: don't count on mm-less current process") on CONFIG_MMU=n kernels.

Also, since commit f44666b04605d1c7 ("mm,oom: speed up select_bad_process() loop")
changed to iterate using thread group leaders, it is no longer highly unlikely
that p is a thread group leader which already released mm. What you call "a highly
theoretical issue" (which is true as of commit 83363b917a2982dd ("oom: make sure
that TIF_MEMDIE is set under task_lock") was proposed) may not be true any more.
Regarding CONFIG_MMU=n kernels, making sure that TIF_MEMDIE is set on a thread
with non-NULL mm does matter.

\
 
 \ /
  Last update: 2016-06-16 15:41    [W:0.134 / U:3.912 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site