Re: [PATCH] oom: add tracepoints for oom_score_adj

[KAMEZAWA Hiroyuki]

On Wed, 07 Dec 2011 11:52:02 -0500
KOSAKI Motohiro <kosaki.motohiro@xxxxxxxxxxxxxx> wrote:

> On 12/6/2011 7:54 PM, KAMEZAWA Hiroyuki wrote:
> >>From 28189e4622fd97324893a0b234183f64472a54d6 Mon Sep 17 00:00:00 2001
> > From: KAMEZAWA Hiroyuki <kamezawa.hiroyu@xxxxxxxxxxxxxx>
> > Date: Wed, 7 Dec 2011 09:58:16 +0900
> > Subject: [PATCH] oom: trace point for oom_score_adj
> > 
> > oom_score_adj is set to prevent a task from being killed by OOM-Killer.
> > Some daemons sets this value and their children inerit it sometimes.
> > Because inheritance of oom_score_adj is done automatically, users
> > can be confused at seeing the value and finds it's hard to debug.
> > 
> > This patch adds trace point for oom_score_adj. This adds 3 trace
> > points. at
> > 	- update oom_score_adj
> 
> 
> > 	- fork()
> > 	- rename task->comm(typically, exec())
> 
> I don't think they have oom specific thing. Can you please add generic fork and
> task rename tracepoint instead?
> 
I think it makes oom-targeted debug difficult.
This tracehook using task->signal->oom_score_adj as filter.
This reduces traces much and makes debugging easier.
 
If you need another trace point for other purpose, another trace point
should be better. For generic purpose, oom_socre_adj filtering will not
be necessary.





> > 
> > Outputs will be following.
> >    bash-2404  [006]   199.620841: oom_score_adj_update: task 2404[bash] updates oom_score_ad  j=-1000
> 
> "task 2404[bash]" don't look good to me.
> 
> In almost case, we use either
> 
>  - [pid] comm
>  - pid:comm
>  - comm:pid
>  - comm-pid    (ftrace specific)
> 
> Why do we need to introduce alternative printing style?
> 

No reason. ok, I'll fix.

Thanks,
-Kame

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/