Re: bcache: for-next unable to handle kernel NULL pointer dereference at 0000000000000019

[Stefan Priebe - Profihost AG <s.priebe@profihost.ag>]

Am 23.10.2017 um 16:39 schrieb Coly Li:
> On 2017/10/23 下午10:26, Stefan Priebe - Profihost AG wrote:
>> Hi,
>> Am 23.10.2017 um 16:00 schrieb Coly Li:
>>> On 2017/10/23 下午9:16, Stefan Priebe - Profihost AG wrote:
>>>> Hi,
>>>>
>>>> Am 23.10.2017 um 15:05 schrieb Coly Li:
>>>>> On 2017/10/23 下午8:59, Stefan Priebe - Profihost AG wrote:
>>>>>> Hi Coly,
>>>>>>
>>>>>>
>>>>>> Am 23.10.2017 um 14:56 schrieb Coly Li:
>>>>>>> On 2017/10/23 下午7:42, Stefan Priebe - Profihost AG wrote:
>>>>>>>> Hello,
>>>>>>>>
>>>>>>>> i picked all bcache patches from for-next to my 4.4 kernel to test the
>>>>>>>> new controller.
>>>>>>>>
>>>>>>>> After doing so i see random kernel panics with the following trace:
>>>>>>>
>>>>>>> Hi Stefan,
>>>>>>>
>>>>>>> Thanks for the report. This is the 3rd report I see recently for NULL
>>>>>>> pointer dereference, maybe they are related (or maybe not). Is it a
>>>>>>> panic when bcache starts to run, or during heavy workload ?
>>>>>>
>>>>>> It's during heavy / normal workload.
>>>>>>
>>>>>>> If I may have chance to trigger similar oops on my server, that will be
>>>>>>> much easier. So far I cannot reproduce any oops, neither by rebooting
>>>>>>> and assemble bcache device by udev rules, nor compose bcache device and
>>>>>>> run it by bash scripts...
>>>>>>
>>>>>> Do you need the line where this happens? It should be possible to get
>>>>>> the line from the IP: [<ffffffffc04ef62e>] output?
>>>>>>
>>>>> This is very helpful.
>>>>
>>>> May be i'm too stupid but it does not print anything useful:
>>>>
>>>> # addr2line -f -e
>>>> /usr/lib/debug/lib/modules/4.4.92+534-ph/kernel/drivers/md/bcache/bcache.ko
>>>> ffffffffc04ef62e closure_sub
>>>> ??
>>>> ??:0
>>>> bch_inc_gen
>>>> ??:?
>>>>
>>>>> Is it possible to get a kdump crash for the kernel
>>>>> oops, that will be much more informative :-)
>>>>
>>>> no idea how to archieve this for a remote Server.
>>>
>>> Hi Stefan,
>>>
>>> In code path of closure_wake_up(), I remember there are two patches in
>>> last run,
>>> - commit a5f3d8a5eaaf ("bcache: use llist_for_each_entry_safe() in
>>> __closure_wake_up()")
>>> - commit 09b3efec81de ("bcache: Don't reinvent the wheel but use
>>> existing llist API")
>>>
>>> Can you check whether you have all of these patches ? Or can we try to
>>> revoke these two patches and see whether oops still happens.
>>
>> It seems i'm missing a5f3d8a5eaaf but i have 09b3efec81de.
>>
>> I missed it because
>> git log ..linux-block/for-next -- drivers/md/bcache/
>>
>> does not show it. It seems linux-block/for-next does not contain it?
>> Which branch should i use?
>>
>> Only those contain the mentioned commit:
>>   remotes/linux-block/for-linus
>>   remotes/linux-block/master
>>   remotes/linux-block/wbt-odirect
>>
> 
> Hi Stefan,
> 
> These 2 patches are in 4.14 mainline kernel already. This is my fault to
> make commit 09b3efec81de buggy, and fix it in commit a5f3d8a5eaaf.
> 
> Could you please try again with the fixing patch ?
> 
> (And I guess maybe other 2 reports may also miss this fix).

OK indeed it seems it fixed it. Sorry about that.

Greets,
Stefan

> Thanks.
> 
> Coly Li
>