linux-kernel - RE: [PATCH] mm, oom: normalize the adj to ensure oom

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [thread-next>] [day] [month] [year] [list]

Message-ID: <CD6925E8781EFD4D8E11882D20FC406D01AEE4A8@SHSMSX103.ccr.corp.intel.com>
Date:	Tue, 4 Mar 2014 05:49:14 +0000
From:	"He, Bo" <bo.he@...el.com>
To:	David Rientjes <rientjes@...gle.com>
CC:	"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
	Ingo Molnar <mingo@...e.hu>,
	"hannes@...xchg.org" <hannes@...xchg.org>,
	"oleg@...hat.com" <oleg@...hat.com>,
	"kirill.shutemov@...ux.intel.com" <kirill.shutemov@...ux.intel.com>,
	"Zhang, Yanmin" <yanmin.zhang@...el.com>,
	"yanmin_zhang@...el.linux.com" <yanmin_zhang@...el.linux.com>,
	"Wang, Biao" <biao.wang@...el.com>
Subject: RE: [PATCH]  mm, oom: normalize the adj to ensure oom_badness
 return a positive number

Sorry, the title is confusing. Change it to: mm, oom: normalize the adj to ensure oom_badness returns a positive number
We are enabling android mobiles. When running stress memory test, there is a bad issue. Some critical processes such as Healthd and watchdogd are killed, while some other processes are still alive.
OOM should kill the tasks whose oom_score are biggest. Many processes use a minus oom_score_adj. oom_badness returns 1 for all of them and their oom_score are all 0.

The patch tries to convert the minus oom_score_adj to a positive number when calculating oom_score. oom_score can keep right process priority sequence.

Without the patch, almost all the processes' oom_score are equal to 0. (cat /proc/XXX/oom_*)
Healthd            -15 0 -941
Servicemanager     -15 0 -941
Vold               -15 0 -941
ia_watchdogd      -15 0 -941
surfaceflinger       -15 0 -941
zygote            -15 0 -941
system_server     -15 0 -941
ndroid.systemui    -11 0 -705

With the patch, we can get different oom_score:
Healthd            -15 59 -941
Servicemanager     -15 89 -941
Vold               -15 60 -941
ia_watchdogd      -15 89 -941
surfaceflinger       -15 145 -941
zygote            -15 122 -941
system_server     -15 166 -941
ndroid.systemui    -11 419 -705

-----Original Message-----
From: David Rientjes [mailto:rientjes@...gle.com] 
Sent: Tuesday, March 04, 2014 1:05 PM
To: He, Bo
Cc: linux-kernel@...r.kernel.org; Ingo Molnar; hannes@...xchg.org; oleg@...hat.com; kirill.shutemov@...ux.intel.com; Zhang, Yanmin; yanmin_zhang@...el.linux.com; Wang, Biao
Subject: Re: [PATCH] mm, oom: normalize the adj to ensure oom_badness return one

On Mon, 3 Mar 2014, He, Bo wrote:

> if oom_score_adj is a big negative number, such as -941, adj *= 
> totalpages / 1000 will be a big negative number, finally the 
> oom_badness will get 0 points, here normalize the oom_score_adj to 
> ensure oom_badness return the positive number.
> 

Sorry, I have no clue what you're talking about or trying to fix.

A /proc/pid/oom_score_adj of -941 would mean discounting 94.1% of system 
memory from pid's resident memory usage for a system oom condition.  This 
is the effect of adj *= totalpages / 1000 and is working as intended per 
Documentation/filesystems/proc.txt.

oom_badness() will then return the smallest integer possible, 1 (not 0), 
that still allows the process to be killed, that's the effect of 
"return points > 0 ? points : 1".  It never returns 0 as you state, you're 
either not reading the code correctly or not working on any recent kernel.

> Change-Id: I1c56a948ce48b65a1bb63b56ffef07d5d76d7ec8
> Signed-off-by: he, bo <bo.he@...el.com>
> Signed-off-by: wang, biao <biao.wang@...el.com>
> Reviewed-by: Yanmin Zhang <yanmin_zhang@...el.com>
> ---
>  mm/oom_kill.c |    3 +++ 
>  1 files changed, 3 insertions(+), 0 deletions(-)
> 
> diff --git a/mm/oom_kill.c b/mm/oom_kill.c
> index 3291e82..5a93986 100644
> --- a/mm/oom_kill.c
> +++ b/mm/oom_kill.c
> @@ -181,6 +181,9 @@ unsigned long oom_badness(struct task_struct *p, struct mem_cgroup *memcg,
>         points -= (points * 3) / 100;
>  
>     /* Normalize to oom_score_adj units */
> +   if(OOM_SCORE_ADJ_MIN < 0)
> +       adj -= OOM_SCORE_ADJ_MIN;
> +
>     adj *= totalpages / 1000;
>     points += adj;
>  
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/