Re: [Patch v2] sysfs: add lockdep class support to s_active

[Cong Wang <amwang@redhat.com>]

Xiaotian Feng wrote:
> On Fri, Feb 5, 2010 at 5:39 PM, Eric W. Biederman <ebiederm@xmission.com> wrote:
>> Xiaotian Feng <xtfeng@gmail.com> writes:
>>
>>> On Fri, Feb 5, 2010 at 2:42 PM, Amerigo Wang <amwang@redhat.com> wrote:
>>>> Recently we met a lockdep warning from sysfs during s2ram or cpu hotplug.
>>>> As reported by several people, it is something like:
>>>>
>>>> [ 6967.926563] ACPI: Preparing to enter system sleep state S3
>>>> [ 6967.956156] Disabling non-boot CPUs ...
>>>> [ 6967.970401]
>>>> [ 6967.970408] =============================================
>>>> [ 6967.970419] [ INFO: possible recursive locking detected ]
>>>> [ 6967.970431] 2.6.33-rc2-git6 #27
>>>> [ 6967.970439] ---------------------------------------------
>>>> [ 6967.970450] pm-suspend/22147 is trying to acquire lock:
>>>> [ 6967.970460]  (s_active){++++.+}, at: [<c10d2941>]
>>>> sysfs_hash_and_remove+0x3d/0x4f
>>>> [ 6967.970493]
>>>> [ 6967.970497] but task is already holding lock:
>>>> [ 6967.970506]  (s_active){++++.+}, at: [<c10d4110>]
>>>> sysfs_get_active_two+0x16/0x36
>>>> [...]
>>>>
>>>> Eric already provides a patch for this[1], but it still can't fix the
>>>> problem. Based on his work and Peter's suggestion, I write this patch,
>>>> hopefully we can fix the warning completely.
>>>>
>>>> This patch put sysfs s_active into two classes, one is for PM, the other
>>>> is for the rest, so lockdep will distinguish them.
>>> I think this patch does not hit the root cause, we have a similiar
>>> warning which is not related with PM.
>> The root cause is that our locking is crazy complicated.  No lockdep
>> changes are going to fix that.
>>
>> What we can do and what the patch does is teach lockdep to treat some
>> of the sysfs files as a different group (subclass) from other sysfs
>> files.  Which keeps us from overgeneralizing too much and having
>> a better signal to noise ratio.
>>
>> As for the block device problem goes, I can't easily say that
>> the block layer is correct.  I expect it is because changing
>> the scheduler is unlikely to delete block devices.  If the block layer
>> has bugs then adding another subclass as Amerigo suggests should simply
>> make lockdep warnings harder to trigger and more accurate so that
>> sounds like a path worth walking.
>>
>> In general I recommend that pieces of code that need to do a lot of
>> work in a sysfs attribute consider using a work queue or a kernel
>> thread, as that can be easier to analyze.
> 
> PM case
> store /sys/devices/system/cpu1/online
> remove /sys/devices/system/cpu1/cache/
> 
> iosched case
> store /sys/block/sdx/queue/scheduler
> remove /sys/block/sdx/queue/iosched/
> 
> So it looks like this is from sysfs layer ....
> 

Right, and both locks are s_active, so I think they are the
same problem, but I haven't check the iosched case carefully. ;)