Re: [PATCH] mm,oom: use per signal_struct flag rather than clear TIF_MEMDIE

[Oleg Nesterov <oleg@redhat.com>]

On 06/28, Michal Hocko wrote:
>
> On Mon 27-06-16 19:55:55, Oleg Nesterov wrote:
> > On 06/27, Michal Hocko wrote:
> > >
> > > On Mon 27-06-16 17:51:20, Oleg Nesterov wrote:
> > > >
> > > > Yes I agree, it would be nice to remove find_lock_task_mm(). And in
> > > > fact it would be nice to kill task_struct->mm (but this needs a lot
> > > > of cleanups). We probably want signal_struct->mm, but this is a bit
> > > > complicated (locking).
> > >
> > > Is there any hard requirement to reset task_struct::mm in the first
> > > place?
> >
> > Well, at least the scheduler needs this.
>
> Could you point me to where it depends on that? I mean if we are past
> exit_mm then we have unmapped the address space most probably but why
> should we care about that in the scheduler? There shouldn't be any
> further access to the address space by that point. I can see that
> context_switch() checks task->mm but it should just work when it sees it
> non NULL, right?

But who will do the final mmdrop() then? I am not saying this is impossible
to change, say we do this in finish_task_switch(TASK_DEAD) or even in
free_task(), but we do not want this?

> Do you think this would be a way to go, though? We would have to special
> case this because the mm_struct is quite large (~900B with my config) so
> we would keep and pin it only for oom victims.

Plus page tables, so it is more than 900B. But as I said, personally I agree
with signal->oom_mm which can only be set by oom-killer.

Oleg.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>