On Fri, Jan 03, 2014 at 03:54:04PM +0100, Ludovic Desroches wrote:

Hi,

On Tue, Dec 24, 2013 at 03:38:37PM +0900, Joonsoo Kim wrote:

[...]

quoted

quoted

quoted

quoted

quoted

quoted

I think that this commit may not introduce a bug. This patch remove one
variable on slab management structure and replace variable name. So there
is no functional change.

You are right, the commit given by git bisect was not the good one...
Since I removed other patches done on top of it, I thought it really was
this one but in fact it is 8456a64.

Okay. It seems more reasonable to me.
I guess that this is the same issue with following link.
http://lkml.org/lkml/2014/1/4/81

And, perhaps, that patch solves your problem. But I'm not sure that it is the
best solution for this problem. I should discuss with slab maintainers.

I will think about this problem more deeply and report the solution to you
as soon as possible.

Thanks.

 dd0f774  Fri Jan 3 12:33:55 2014 +0100  Revert "slab: remove useless
statement for checking pfmemalloc"  Ludovic Desroches 
 ff7487d  Fri Jan 3 12:32:33 2014 +0100  Revert "slab: rename
slab_bufctl to slab_freelist"  Ludovic Desroches 
 b963564  Fri Jan 3 12:32:13 2014 +0100  Revert "slab: fix to calm down
kmemleak warning"  Ludovic Desroches 
 3fcfe50  Fri Jan 3 12:30:32 2014 +0100  Revert "slab: replace
non-existing 'struct freelist *' with 'void *'"  Ludovic Desroches 
 750a795  Fri Jan 3 12:30:16 2014 +0100  Revert "memcg, kmem: rename
cache_from_memcg to cache_from_memcg_idx"  Ludovic Desroches 
 7e2de8a  Fri Jan 3 12:30:10 2014 +0100  mmc: atmel-mci: disable pdc
Ludovic Desroches

In this case I have the kernel oops. If I revert 8456a64 too, it
disappears.

I will try to test it on other devices because I couldn't reproduce it
with newer ones (but it's not the same ARM architecture so I would like
to see if it's also related to the device itself).

In attachment, there are the results of /proc/slabinfo before inserted the
sdio wifi module causing the oops.

Regards

Ludovic