Re: [BUG]: Possibe recursive locking detected in sysfs

["Américo Wang" <xiyou.wangcong@gmail.com>]

On Mon, Feb 15, 2010 at 01:56:45AM -0800, Eric Biederman wrote:
>On Sun, Feb 14, 2010 at 6:22 PM, John Kacur <jkacur@redhat.com> wrote:
>> On Sun, Feb 14, 2010 at 4:50 PM, Eric Biederman
>> <ebiederm@aristanetworks.com> wrote:
>>>> Sure, are you referring to the patch-set that begins with
>>>> "[PATCH 1/6] sysfs: Serialize updates to the vfs inode"?
>>>
>>> Sorry no.
>>>
>>> http://lkml.org/lkml/2010/2/11/329
>>>
>>
>> I applied your patch, and yes, it removed the possible recursive
>> locking detected message, but everything still froze.
>> I don't think I really have any good info from the crash to report.
>> Your patch seems to have added the symptom of a huge number of
>> BUG: key ffff880126269e40 not in .data!
>> BUG: key ffff880136fc03f0 not in .data!
>
>Those are from dynamic sysfs entries that I have not yet annoted
>with sysfs_attr_init, and are generally harmless.  If you happen
>to see the first one.  I would appreciate having the backtrace so I
>can see about fixing it.
>
>With respect to your problem the important point is that lockdep does
>not throw a warning and disable itself.  Can you verify that?
>
>Assuming that lockdep has not complained and disabled itself than
>my patches are successful at disabling the sysfs lockdep false positives
>(except those BUG: key ... not in .data messages). and the lockdep
>warnings are just a coincidence in your case.
>
>I believe the cause of your hang is somewhere else entirely.  Perhaps
>a driver regression.
>

Right, we got a real deadlock here:

http://lkml.org/lkml/2010/1/28/320