On Tue 04-06-13 14:48:52, Johannes Weiner wrote:
> On Tue, Jun 04, 2013 at 11:17:49AM +0200, Michal Hocko wrote:
[...]
> > > diff --git a/mm/memory.c b/mm/memory.c
> > > index 6dc1882..ff5e2d7 100644
> > > --- a/mm/memory.c
> > > +++ b/mm/memory.c
> > > @@ -1815,7 +1815,7 @@ long __get_user_pages(struct task_struct *tsk,
> > > struct mm_struct *mm,
> > > while (!(page = follow_page_mask(vma, start,
> > > foll_flags, &page_mask))) {
> > > int ret;
> > > - unsigned int fault_flags = 0;
> > > + unsigned int fault_flags = FAULT_FLAG_KERNEL;
> > >
> > > /* For mlock, just skip the stack guard page. */
> > > if (foll_flags & FOLL_MLOCK) {
> >
> > This is also a bit tricky. Say there is an unlikely situation when a
> > task fails to charge because of memcg OOM, it couldn't lock the oom
> > so it ended up with current->memcg_oom set and __get_user_pages will
> > turn VM_FAULT_OOM into ENOMEM but memcg_oom is still there. Then the
> > following global OOM condition gets confused (well the oom will be
> > triggered by somebody else so it shouldn't end up in the endless loop
> > but still...), doesn't it?
>
> But current->memcg_oom is not set up unless current->in_userfault.
> And get_user_pages does not set this flag.
And my selective blindness strikes again :/ For some reason I have read
those places as they enable the fault flag. Which would make some sense
if there was a post handling...
Anyway, I will get back to the updated patch tomorrow with a clean and
fresh head.
--
Michal Hocko
SUSE Labs
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
