Re: [patch v2 1/2] oom: avoid killing a task if a thread sharing its mm cannot be killed

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

On Tue, 17 Aug 2010 19:36:02 -0700 (PDT)
David Rientjes <rientjes@google.com> wrote:

> On Wed, 18 Aug 2010, KAMEZAWA Hiroyuki wrote:
> 
> > > The oom killer's goal is to kill a memory-hogging task so that it may
> > > exit, free its memory, and allow the current context to allocate the
> > > memory that triggered it in the first place.  Thus, killing a task is
> > > pointless if other threads sharing its mm cannot be killed because of its
> > > /proc/pid/oom_adj or /proc/pid/oom_score_adj value.
> > > 
> > > This patch checks all user threads on the system to determine whether
> > > oom_badness(p) should return 0 for p, which means it should not be killed.
> > > If a thread shares p's mm and is unkillable, p is considered to be
> > > unkillable as well.
> > > 
> > > Kthreads are not considered toward this rule since they only temporarily
> > > assume a task's mm via use_mm().
> > > 
> > > Signed-off-by: David Rientjes <rientjes@google.com>
> > 
> > Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
> > 
> 
> Thanks!
> 
> > Thank you. BTW, do you have good idea for speed-up ?
> > This code seems terribly slow when a system has many processes.
> > 
> 
> I was thinking about adding an "unsinged long oom_kill_disable_count" to 
> struct mm_struct that would atomically increment anytime a task attached 
> to it had a signal->oom_score_adj of OOM_SCORE_ADJ_MIN.
> 
> The proc handler when changing /proc/pid/oom_score_adj would inc or dec 
> the counter depending on the new value, and exit_mm() would dec the 
> counter if current->signal->oom_score_adj is OOM_SCORE_ADJ_MIN.
> 
> What do you think?
> 

Hmm. I want to make hooks to "exit" small. 

One idea is.

add a new member
		mm->unkiilable_by_oom_jiffies.

And add
> +static bool is_mm_unfreeable(struct mm_struct *mm)
> +{
> +	struct task_struct *p;
> +
	if (mm->unkillable_by_oom_jiffies < jiffies)
		return true;

> +	for_each_process(p)
> +		if (p->mm == mm && !(p->flags & PF_KTHREAD) &&
> +		    p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) 

			mm->unkillable_by_oom_jiffies = jiffies + HZ;

> +			return true;
> +	return false;
> +}+static bool is_mm_unfreeable(struct mm_struct *mm)


Maybe no new lock is required and this not-accurate one will work enough.

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>