Re: Kernel falls apart under light memory pressure (i.e. linking vmlinux)

[Minchan Kim]

On Thu, May 19, 2011 at 11:15 AM, Andrew Lutomirski <luto@xxxxxxx> wrote:
> On Wed, May 18, 2011 at 1:17 AM, Minchan Kim <minchan.kim@xxxxxxxxx> wrote:
>> On Wed, May 18, 2011 at 4:22 AM, Andrew Lutomirski <luto@xxxxxxx> wrote:
>>>> No, thanks. However it would be valuable if you can retry with this
>>>> patch _alone_ (without the "if (need_resched()) return false;" change,
>>>> as I don't see how it helps your case).
>>>>
>>>> @@ -2286,7 +2290,7 @@ static bool sleeping_prematurely(pg_data_t
>>>> *pgdat, int order, long remaining,
>>>> Â Â Â Â* must be balanced
>>>> Â Â Â Â*/
>>>> Â Â Â if (order)
>>>> - Â Â Â Â Â Â Â return pgdat_balanced(pgdat, balanced, classzone_idx);
>>>> + Â Â Â Â Â Â Â return !pgdat_balanced(pgdat, balanced, classzone_idx);
>>>> Â Â Â else
>>>> Â Â Â Â Â Â Â return !all_zones_ok;
>>>> Â}
>>>
>>> Done.
>>>
>>> I logged in, added swap, and ran a program that allocated 1900MB of
>>> RAM and memset it. ÂThe system lagged a bit but survived. Âkswapd
>>> showed 10% CPU (which is odd, IMO, since I'm using aesni-intel and I
>>> think that all the crypt happens in kworker when aesni-intel is in
>>> use).
>>
>> I think kswapd could use 10% enough for reclaim.
>>
>>>
>>> Then I started Firefox, loaded gmail, and ran test_mempressure.sh.
>>> Kaboom! Â(I.e. system was hung) ÂSysRq-F saved the system and produced
>>
>> Hang?
>> It means you see softhangup of kswapd? or mouse/keyboard doesn't move?
>
> Mouse and keyboard dead.
>
>> Andrew, Could you test this patch with !pgdat_balanced patch?
>> I think we shouldn't see OOM message if we have lots of free swap space.
>>
>> == CUT_HERE ==
>> diff --git a/mm/vmscan.c b/mm/vmscan.c
>> index f73b865..cc23f04 100644
>> --- a/mm/vmscan.c
>> +++ b/mm/vmscan.c
>> @@ -1341,10 +1341,6 @@ static inline bool
>> should_reclaim_stall(unsigned long nr_taken,
>> Â Â Â Âif (current_is_kswapd())
>> Â Â Â Â Â Â Â Âreturn false;
>>
>> - Â Â Â /* Only stall on lumpy reclaim */
>> - Â Â Â if (sc->reclaim_mode & RECLAIM_MODE_SINGLE)
>> - Â Â Â Â Â Â Â return false;
>> -
>> Â Â Â Â/* If we have relaimed everything on the isolated list, no stall */
>> Â Â Â Âif (nr_freed == nr_taken)
>> Â Â Â Â Â Â Â Âreturn false;
>>
>>
>>
>> Then, if you don't see any unnecessary OOM but still see the hangup,
>> could you apply this patch based on previous?
>
> With this patch, I started GNOME and Firefox, turned on swap, and ran
> test_mempressure.sh 1500 1400 1. ÂInstant panic (or OOPS and hang or
> something -- didn't get the top part). ÂPicture attached -- it looks
> like memcg might be involved. ÂI'm running F15, so it might even be
> doing something.

I cannot figure out why happens OOPS.
Let me know your kernel version and config.
Kame. Is there anything related to memcg you guess?

In addition, the patch I give was utterly stupid.
The goal is that we wait dirty page writeback in (order-0 | high
priority) reclaim.
(But I don't think it's ideal solution in this problem but just for
proving the problem)
But although we pass sync with 1 in set_reclaim_mode, it ignores.
So fix is following as. (NOTICE: It doesn't related to your OOPS. )
But before further experiment, let's fix your oops.

diff --git a/mm/vmscan.c b/mm/vmscan.c
index 292582c..69d317e 100644
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -311,7 +311,8 @@ static void set_reclaim_mode(int priority, struct
scan_control *sc,
         */
        if (sc->order > PAGE_ALLOC_COSTLY_ORDER)
                sc->reclaim_mode |= syncmode;
-       else if (sc->order && priority < DEF_PRIORITY - 2)
+       else if ((sc->order && priority < DEF_PRIORITY - 2) ||
+                               prioiry <= DEF_PRIORITY / 3)
                sc->reclaim_mode |= syncmode;
        else
                sc->reclaim_mode = RECLAIM_MODE_SINGLE | RECLAIM_MODE_ASYNC;
@@ -1349,10 +1350,6 @@ static inline bool
should_reclaim_stall(unsigned long nr_taken,
        if (current_is_kswapd())
                return false;

-       /* Only stall on lumpy reclaim */
-       if (sc->reclaim_mode & RECLAIM_MODE_SINGLE)
-               return false;
-
        /* If we have relaimed everything on the isolated list, no stall */
        if (nr_freed == nr_taken)
                return false;



>
> I won't be able to get netconsole dumps until next week because I'm
> out of town and only have this one computer here.

No problem. :)
We should avoid OOPS for the experiment.


>
> I haven't tried the other patch.
>
> Also, the !pgdat_balanced fix plus the if (need_resched()) return
> false patch just hung once on 2.6.37-rc9. ÂI don't know what triggered

Thanks for the good information.
It seems need_resched patch isn't good candidate to fix current problem.
We already weeded it out.

Thank you very much for the testing!

> it. ÂMaybe yum.
>
> --Andy
>



-- 
Kind regards,
Minchan Kim
--
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/