Re: mem_cgroup_page_lruvec: BUG: unable to handle kernel NULLpointer dereference at 00000000000001a8

[Michal Hocko]

On Thu 13-06-13 15:34:59, Richard Weinberger wrote:
> Am 13.06.2013 15:32, schrieb Michal Hocko:
> >Ohh and could you post the config please? Sorry should have asked
> >earlier.
> 
> See attachment.

Nothing unusual there. Could you enable CONFIG_DEBUG_VM maybe it will
help too catch the problem earlier.

> >On Thu 13-06-13 15:29:08, Michal Hocko wrote:
> >>
> >>On Thu 13-06-13 14:06:20, Richard Weinberger wrote:
> >>[...]
> >>>All code
> >>>========
> >>>    0:   89 50 08                mov    %edx,0x8(%rax)
> >>>    3:   48 89 d1                mov    %rdx,%rcx
> >>>    6:   0f 1f 40 00             nopl   0x0(%rax)
> >>>    a:   49 8b 04 24             mov    (%r12),%rax
> >>>    e:   48 89 c2                mov    %rax,%rdx
> >>>   11:   48 c1 e8 38             shr    $0x38,%rax
> >>>   15:   83 e0 03                and    $0x3,%eax
> >>					nid = page_to_nid
> >>>   18:   48 c1 ea 3a             shr    $0x3a,%rdx
> >>					zid = page_zonenum

Ohh, I am wrong here. rdx should be nid and eax the zid.

> >>
> >>>   1c:   48 69 c0 38 01 00 00    imul   $0x138,%rax,%rax
> >>>   23:   48 03 84 d1 e0 02 00    add    0x2e0(%rcx,%rdx,8),%rax
> >>					&memcg->nodeinfo[nid]->zoneinfo[zid]
> >>
> >>>   2a:   00
> >>>   2b:*  48 3b 58 70             cmp    0x70(%rax),%rbx     <-- trapping instruction
> >>
> >>OK, so this maps to:
> >>         if (unlikely(lruvec->zone != zone)) <<<
> >>                 lruvec->zone = zone;
> >>
> >>>[35355.883056] RSP: 0000:ffff88003d523aa8  EFLAGS: 00010002
> >>>[35355.883056] RAX: 0000000000000138 RBX: ffff88003fffa600 RCX: ffff88003e04a800
> >>>[35355.883056] RDX: 0000000000000020 RSI: 0000000000000000 RDI: 0000000000028500
> >>>[35355.883056] RBP: ffff88003d523ab8 R08: 0000000000000000 R09: 0000000000000000
> >>>[35355.883056] R10: 0000000000000000 R11: dead000000100100 R12: ffffea0000a14000
> >>>[35355.883056] R13: ffff88003e04b138 R14: ffff88003d523bb8 R15: ffffea0000a14020
> >>>[35355.883056] FS:  0000000000000000(0000) GS:ffff88003fd80000(0000)
> >>
> >>RAX (lruvec) is obviously incorrect and it doesn't make any sense. rax should
> >>contain an address at an offset from ffff88003e04a800 But there is 0x138 there
> >>instead.

Hmm, now that I am looking at the registers again. RDX which should be
nid seems to be quite big. It says this is node 32. Does the machine
have really so many NUMA nodes?
Also I think the trapping instruction was one instruction above:
IP: [<ffffffff811297d9>] mem_cgroup_page_lruvec+0x79/0x90

0x000000000004fb09 <+121>:   add    0x2e0(%rcx,%rdx,8),%rax
0x000000000004fb11 <+129>:   cmp    0x70(%rax),%rbx

rather than cmp marked above. This would explain why rax is 138 because
that would point the zid=1 and 138 is offset of mem_cgroup_per_zone
within mem_cgroup_per_node for that zone. This would mean that the
struct page contains a weird node id.

-- 
Michal Hocko
SUSE Labs
--
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/