| 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
| ||
|
Message-ID: <CAAmzW4OGBBz5jOCuLOPt1fzMXi9JMxYQYRUOiE0483kWufp34Q@mail.gmail.com>
Date: Wed, 29 Oct 2014 01:00:36 +0900
From: Joonsoo Kim <js1304@...il.com>
To: Markos Chandras <Markos.Chandras@...tec.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@....com>,
Christoph Lameter <cl@...ux.com>,
Pekka Enberg <penberg@...nel.org>,
David Rientjes <rientjes@...gle.com>,
Andrew Morton <akpm@...ux-foundation.org>,
LKML <linux-kernel@...r.kernel.org>, linux-mips@...ux-mips.org
Subject: Re: Boot problems on Malta with EVA (bisected to 12220dea07f1
"mm/slab: support slab merge")
2014-10-29 0:45 GMT+09:00 Markos Chandras <Markos.Chandras@...tec.com>:
> On 10/28/2014 03:00 PM, Joonsoo Kim wrote:
>> 2014-10-28 23:32 GMT+09:00 Markos Chandras <Markos.Chandras@...tec.com>:
>>> On 10/28/2014 02:21 PM, Joonsoo Kim wrote:
>>>> 2014-10-28 22:48 GMT+09:00 Joonsoo Kim <js1304@...il.com>:
>>>>> 2014-10-28 22:24 GMT+09:00 Markos Chandras <Markos.Chandras@...tec.com>:
>>>>>> On 10/28/2014 01:19 PM, Markos Chandras wrote:
>>>>>>> On 10/28/2014 01:01 PM, Joonsoo Kim wrote:
>>>>>>>> 2014-10-28 19:45 GMT+09:00 Markos Chandras <Markos.Chandras@...tec.com>:
>>>>>>>>> Hi,
>>>>>>>>>
>>>>>>>>> It seems I am unable to boot my Malta with EVA. The problem appeared in
>>>>>>>>> the 3.18 merge window. I bisected the problem (between v3.17 and
>>>>>>>>> v3.18-rc1) and I found the following commit responsible for the broken boot.
>>>>>>>>
>>>>>>>> Hello,
>>>>>>>>
>>>>>>>> Did you start to bisect from v3.18-rc1?
>>>>>>>> I'd like to be sure that this is another bug which is fixed by following commit.
>>>>>>>>
>>>>>>>> commit 85c9f4b04a08f6bc770b77530c22d04103468b8f
>>>>>>>> Author: Joonsoo Kim <iamjoonsoo.kim@....com>
>>>>>>>> Date: Mon Oct 13 15:51:01 2014 -0700
>>>>>>>>
>>>>>>>> mm/slab: fix unaligned access on sparc64
>>>>>>>>
>>>>>>>> This fix is merged into v3.18-rc1 sometime later that
>>>>>>>> 'support slab merge' is merged.
>>>>>>>>
>>>>>>>> Thanks.
>>>>>>>>
>>>>>>> Hi,
>>>>>>>
>>>>>>> I bisected from v3.17 until 3.18-rc1. But 3.18-rc2 and the latest
>>>>>>> mainline (f7e87a44ef60ad379e39b45437604141453bf0ec) still have the same
>>>>>>> problem
>>>>>>>
>>>>>>> btw i did more tests and this is not EVA specific. A maltaup_defconfig
>>>>>>> fails in the same way. I suspect all malta*_defconfigs will fail in a
>>>>>>> similar way which makes it probably easier for you to reproduce it on a
>>>>>>> QEMU.
>>>>>>>
>>>>>>
>>>>>> sorry maltaup_defconfig does not fail. maltasmvp_defconfig does. So it
>>>>>> might be a similar problem like the one fixed in
>>>>>> 85c9f4b04a08f6bc770b77530c22d04103468b8f
>>>>>
>>>>> Oops. Sorry. Above commit ('mm/slab: fix unaligned access on sparc64')
>>>>> is irrelevant to this problem.
>>>>>
>>>>> Anyway, your problem would be related to merging with incompatible slab cache.
>>>>> Best way to debug is printing source/target slab cache's object size and
>>>>> alignment and find the problem. I will try to reproduce it using QEMU.
>>>>
>>>> I found that cross compile for MIPS isn't easy job. :)
>>>
>>> You could grab the following toolchain
>>>
>>> https://sourcery.mentor.com/GNUToolchain/release2791
>>>
>>> (get the IA32 linux tar)
>>>
>>> unpack it somewhere (eg /tmp) and then
>>>
>>> make ARCH=mips maltasmvp_defconfig
>>> make ARCH=mips CROSS_COMPILE=/tmp/mips-2014.05/bin/mips-linux-gnu- -j8
>>> or something :)
>>
>> Wow!! Really Thanks!
>> I will try it.
>>
>>>> Could you help me to debug the problem with below patch?
>>>
>>> (there are a few build warnings with your patch
>>> mm/slab.c:2065:4: warning: format '%lu' expects argument of type 'long
>>> unsigned int', but argument 5 has type 'size_t' [-Wformat=]
>>> mm/slab.c:2065:4: warning: format '%lu' expects argument of type 'long
>>> unsigned int', but argument 7 has type 'unsigned int' [-Wformat=]
>>> mm/slab.c:2065:4: warning: format '%lu' expects argument of type 'long
>>> unsigned int', but argument 8 has type 'int' [-Wformat=]
>>> mm/slab.c:2065:4: warning: format '%lu' expects argument of type 'long
>>> unsigned int', but argument 9 has type 'int' [-Wformat=]
>>> )
>>>
>>> but here is the output from a QEMU boot right before the crash
>>>
>>> CPU frequency 200.00 MHz
>>> Calibrating delay loop... 1087.89 BogoMIPS (lpj=5439488)
>>> pid_max: default: 32768 minimum: 301
>>> __kmem_cache_alias: (cred_jar 92 0) to (kmalloc-128 128 128 128)
>>> __kmem_cache_alias: (files_cache 256 0) to (kmalloc-256 256 128 256)
>>> __kmem_cache_alias: (fs_cache 36 0) to (pid 64 64 44)
>>> __kmem_cache_alias: (names_cache 4096 0) to (kmalloc-4096 4096 128 4096)
>>> __kmem_cache_alias: (mnt_cache 160 0) to (filp 192 64 160)
>>> Mount-cache hash table entries: 4096 (order: 0, 16384 bytes)
>>> Mountpoint-cache hash table entries: 4096 (order: 0, 16384 bytes)
>>> __kmem_cache_alias: (pool_workqueue 256 256) to (kmalloc-256 256 128 256)
>>
>> alignment is mismatch between pool_workqueue and kmalloc-256,
>> but, slab caches are merged, because they have same object size.
>> Perhaps, slab cache for pool_workqueue returns 128 byte aligned memory
>> and workqueue can't work well with it.
>>
>> Quick fix may be something like below, but, I will try it on QEMU MIPS.
>>
>> Thanks.
>>
>> diff --git a/mm/slab_common.c b/mm/slab_common.c
>> index 3a6e0cf..d57b1a2 100644
>> --- a/mm/slab_common.c
>> +++ b/mm/slab_common.c
>> @@ -269,6 +269,9 @@ struct kmem_cache *find_mergeable(size_t size, size_t align,
>> if (s->size - size >= sizeof(void *))
>> continue;
>>
>> + if (align > s->align)
>> + continue;
>> +
>> return s;
>> }
>> return NULL;
>>
> Hi,
>
> Yeah this makes the kernel boot again.
(Just resend due to missing cc list)
Thanks for testing!
I tried boot on QEMU MIPS and can reproduce this problem.
I found that my assumption is correct.
128 byte aligned memory is returned for pool_workqueue.
I need some time to think better way to fix this problem.
I will send it in a few days.
Really thanks for your help.
Thanks.
--
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/
Powered by blists - more mailing lists