Re: /proc/kcore has a unreasonable size(281474974617600) in x86_64 2.6.30-rc8.

[ebiederm@xmission.com (Eric W. Biederman)]

Amerigo Wang <xiyou.wangcong@gmail.com> writes:

> On Mon, Jun 08, 2009 at 09:10:10PM -0700, Eric W. Biederman wrote:
>>Américo Wang <xiyou.wangcong@gmail.com> writes:
>>
>>> On Mon, Jun 8, 2009 at 4:00 PM, Tao Ma<tao.ma@oracle.com> wrote:
>>>>>>
>>>>>> But the result is the same
>>>>>
>>>>> Yes?
>>>>> Your printk() shows kcore size is: 5301604352, and in your subject it is
>>>>> 281474974617600...
>>>>>
>>>>> Or they happened in the same time?
>>>>
>>>> yes. the same box and the same linux version.
>>>> A bit strange.
>>>>
>>>> [taoma@ocfs2-test2 ~]$ dmesg|grep "high memory"
>>>> high memory ffff88013c000000, size 5301604352
>>>> [taoma@ocfs2-test2 ~]$ ll /proc/kcore
>>>> -r-------- 1 root root 281474974617600 Jun  8 15:20 /proc/kcore
>>>
>>> Really weird...
>>> They should be the same. This means we have some problem in our procfs.
>>>
>>> And, we have no problem on i386, I, myself, even can't reproduce this on my
>>> x86_64 box...
>>>
>>> Drop Cc to x86 people, add some Cc to proc people. :)
>>>
>>> Eric, Alexey, any ideas?
>>>
>>> Tao, would you like to send us your .config? Thanks.
>>
>>Short of some strange patch applied I would guess that a non-sense /proc/kcore
>>size is related to a kernel memory stomp, stepping on the high_memory variable.
>
> Hello, Eric.
>
> I see the problem now, I think the documentation of /proc/kcore
> is wrong, the size of kcore can be more than the size of physical
> memory, because it also contains the info of kernel modules which
> stay above the mapping of phy memory, see arch/x86/mm/init_64.c.
>
> What do you think?

I think that doesn't make any sense.

I was reading the code.

I smell a nasty problem somewhere.

Eric