Re: [patch -mm v2] mm: introduce oom_adj_child

[KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>]

On Mon, 3 Aug 2009 01:08:42 -0700 (PDT)
David Rientjes <rientjes@google.com> wrote:

> On Mon, 3 Aug 2009, KAMEZAWA Hiroyuki wrote:
> 
> > > You can't recalculate it if all the remaining threads have a different 
> > > oom_adj value than the effective oom_adj value from the thread that is now 
> > > exited.  
> > 
> > Then, crazy google apps pass different oom_adjs to each thread ?
> > And, threads other than thread-group-leader modifies its oom_adj.
> > 
> 
> Nope, but I'm afraid you've just made my point for me: it shows that 
> oom_adj really isn't sanely used as a per-thread attribute and actually 
> only represents a preference on oom killing a quantity of memory in all 
> other cases other than vfork() -> change /proc/pid-of-child/oom_adj -> 
> exec() for which we now appropriately have /proc/pid/oom_adj_child for.
> 
Maybe you're man I can't persuade. but making progress will be necessary.

The most ugly thing which annoies me is this part. 

@@ -679,6 +679,7 @@ good_mm:                    #  a label in copy_mm.
 
 	tsk->mm = mm;
 	tsk->active_mm = mm;
+	tsk->oom_adj_child = mm->oom_adj;
 	return 0;

Why ?

I wonder oom_adj_exec "change oom_adj to this value when execve() is called"
is much more straightforward, simple and easy to understand than oom_adj_child.

"just inherit at fork, change at exec" is an usual manner, I think.
If oom_adj_exec rather than oom_adj_child, I won't complain, more.


Thanks,
-Kame



--
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>