Re: [memcg BUG] unable to handle kernel NULL pointer derefence at 00000000

[Balbir Singh <balbir@linux.vnet.ibm.com>]

KAMEZAWA Hiroyuki wrote:
> On Tue, 21 Oct 2008 16:59:28 +0530
> Balbir Singh <balbir@linux.vnet.ibm.com> wrote:
> 
>> KAMEZAWA Hiroyuki wrote:
>>> On Tue, 21 Oct 2008 13:19:51 +0200
>>> Ingo Molnar <mingo@elte.hu> wrote:
>>>
>>>> * Li Zefan <lizf@cn.fujitsu.com> wrote:
>>>>
>>>>>> Oh! thanks...but it seems pc->page is NULL in the middle of ZONE_NORMAL..
>>>>>> ==
>>>>>>  Normal   0x00001000 -> 0x000373fe
>>>>>> ==
>>>>>> This is appearently in the range of page_cgroup initialization.
>>>>>> (if pgdat->node_page_cgroup is initalized correctly...)
>>>>>>
>>>>>> I think write to page_cgroup->page happens only at initialization.
>>>>>> Hmm ? not initilization failure but curruption ?
>>>>>>
>>>>> Yes, curruption. I didn't find informatation about initialization failure.
>>>>>
>>>>>> What happens if replacing __alloc_bootmem() with vmalloc() in page_cgroup.c init ?
>>>>>>
>>>>> So I did this change, and the box booted up without any problem.
>>>>>
>>>>> diff --git a/mm/page_cgroup.c b/mm/page_cgroup.c
>>>>> index 5d86550..82a30b1 100644
>>>>> --- a/mm/page_cgroup.c
>>>>> +++ b/mm/page_cgroup.c
>>>>> @@ -48,8 +48,7 @@ static int __init alloc_node_page_cgroup(int nid)
>>>>>  
>>>>>  	table_size = sizeof(struct page_cgroup) * nr_pages;
>>>>>  
>>>>> -	base = __alloc_bootmem_node_nopanic(NODE_DATA(nid),
>>>>> -			table_size, PAGE_SIZE, __pa(MAX_DMA_ADDRESS));
>>>>> +	base = vmalloc_node(table_size, nid);
>>>>>  	if (!base)
>>>>>  		return -ENOMEM;
>>>> i have this:
>>>>
>>>>   CONFIG_FAILSLAB=y
>>>>   CONFIG_FAIL_PAGE_ALLOC=y
>>>>   # CONFIG_FAIL_MAKE_REQUEST is not set
>>>>   CONFIG_FAIL_IO_TIMEOUT=y
>>>>
>>>> so the bug was perhaps that the __alloc_bootmem_node_nopanic() failed 
>>>> and this code continued silently? vmalloc_node() probably is more 
>>>> agressive about allocating memory.
>>>>
>>> Sorry. I think I cannot use alloc_bootmem() at this point because
>>> it's too late in init-path. (we can use usual page allocator)
>>> So, just replacing alloc_bootmem() with vmalloc_node() is a fix....
>> Kamezawa-San,
>>
>> I would prefer to use alloc_bootmem() instead of vmalloc_node(). May be we can
>> shift cgroups, so that we use early_init for allocating page_cgroups.  What do
>> you think?
>>
> I got an idea and maybe can send a patch soon. I'm now finding x86-32 box..

Please send it to me, I am able to reproduce the problem with my kvm setup on my
32 bit system. I can do a quick test/verification for you.

-- 
	Balbir

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>