Re: 2.6.31-rc4 - slab entry tak_delay_info leaking ???

[Paul Rolland]

Hello,

On Thu, 30 Jul 2009 12:07:15 +0200
Paul Rolland <rol@xxxxxxxxxx> wrote:

> Hello,
> 
> On Wed, 29 Jul 2009 08:20:47 +0200
> Paul Rolland <rol@xxxxxxxxxx> wrote:
> 
> > Hello,
> > 
> > Since I'm running 2.6.31-rc? (3 or 4 I think), I often find my machine
> > with my X session killed in the morning, and the kdm login screen
> > displayed.
> > 
> > This morning, I decided to save the /var/log/messages to check it before
> > rebooting, and saw that the OOM killer triggered during the night.
> > So, I quickly made a copy of it, and a slabtop output. There came the
> > surprise : 
> >   OBJS ACTIVE  USE OBJ SIZE  SLABS OBJ/SLAB CACHE SIZE NAME
> > 12899448 12899445  99%    0.11K 358318       36   1433272K
> > task_delay_info       
> > 
> > On a freshly booted machine, the same entry shows :
> >   3204   3191  99%    0.11K     89       36       356K task_delay_info
> > after the X session is started.
> > 
> > Please find attached the slabtop -o output.
> 
> Here it starts again :( The machine is up for a little bit more that 1 day
> [root@tux infocast]# uptime
>  12:04:42 up 1 day,  3:53, 12 users,  load average: 0.18, 0.37, 0.48
> 
> I feel it because the machine seems to be slower when it happens,
> probably because it does have less and less memory available...
> 
> And here what the slabtop output monitoring reports :
> Thu Jul 30 11:49:04 CEST 2009
> task_delay_info     3292   3312    112   36    1 : tunables    0    0
> 0 : slabdata     92     92      0 Thu Jul 30 11:50:04 CEST 2009
> task_delay_info     3292   3312    112   36    1 : tunables    0    0
> 0 : slabdata     92     92      0 Thu Jul 30 11:51:05 CEST 2009
> task_delay_info     3292   3312    112   36    1 : tunables    0    0
> 0 : slabdata     92     92      0 Thu Jul 30 11:52:05 CEST 2009
> task_delay_info     3292   3312    112   36    1 : tunables    0    0
> 0 : slabdata     92     92      0 Thu Jul 30 11:53:05 CEST 2009
> task_delay_info     3292   3312    112   36    1 : tunables    0    0
> 0 : slabdata     92     92      0 Thu Jul 30 11:54:05 CEST 2009
> task_delay_info     3289   3312    112   36    1 : tunables    0    0
> 0 : slabdata     92     92      0 Thu Jul 30 11:55:11 CEST 2009
> task_delay_info     3309   3312    112   36    1 : tunables    0    0
> 0 : slabdata     92     92      0 Thu Jul 30 11:56:15 CEST 2009
> task_delay_info    16631  16632    112   36    1 : tunables    0    0
> 0 : slabdata    462    462      0 Thu Jul 30 11:57:15 CEST 2009
> task_delay_info    58860  58860    112   36    1 : tunables    0    0
> 0 : slabdata   1635   1635      0 Thu Jul 30 11:58:15 CEST 2009
> task_delay_info    98422  98424    112   36    1 : tunables    0    0
> 0 : slabdata   2734   2734      0 Thu Jul 30 11:59:15 CEST 2009
> task_delay_info   138023 138024    112   36    1 : tunables    0    0
> 0 : slabdata   3834   3834      0 Thu Jul 30 12:00:15 CEST 2009
> task_delay_info   177587 177588    112   36    1 : tunables    0    0
> 0 : slabdata   4933   4933      0 Thu Jul 30 12:01:16 CEST 2009
> task_delay_info   217151 217152    112   36    1 : tunables    0    0
> 0 : slabdata   6032   6032      0 Thu Jul 30 12:02:17 CEST 2009
> task_delay_info   256715 256716    112   36    1 : tunables    0    0
> 0 : slabdata   7131   7131      0 Thu Jul 30 12:03:17 CEST 2009
> task_delay_info   296316 296316    112   36    1 : tunables    0    0
> 0 : slabdata   8231   8231      0
> 
> 
> Anyone with an idea ???

Well, I've killed my X session, and it stopped the "leak"... but didn't recover
the "lost" task_delay_info...

Attached is .config.

Video is :
00:02.0 VGA compatible controller: Intel Corporation Mobile 945GM/GMS, 943/940GML Express Integrated Graphics Controller (rev 03)
00:02.1 Display controller: Intel Corporation Mobile 945GM/GMS/GME,
943/940GML Express Integrated Graphics Controller (rev 03)

Paul

Attachment: .config
Description: Binary data