Re: INFO: possible circular locking dependency at cleanup_workqueue_thread

[Ming Lei <tom.leiming@gmail.com>]

2009/5/20 Johannes Berg <johannes@sipsolutions.net>:
> On Wed, 2009-05-20 at 15:09 +0800, Ming Lei wrote:
>> 2009/5/20 Johannes Berg <johannes@sipsolutions.net>:
>> > On Wed, 2009-05-20 at 11:36 +0800, Ming Lei wrote:
>> >
>> >> > Anyway, you can have a deadlock like this:
>> >> >
>> >> > CPU 3                   CPU 2                           CPU 1
>> >> >                                                        suspend/hibernate
>> >> >                        something:
>> >> >                        rtnl_lock()                     device_pm_lock()
>> >> >                                                        -> mutex_lock(&dpm_list_mtx)
>> >> >
>> >> >                        mutex_lock(&dpm_list_mtx)
>> >>
>> >> Would you give a explaination why mutex_lock(&dpm_list_mtx) runs in CPU2
>> >> and depends on rtnl_lock?
>> >
>> > Why not? Something is registering a hotplugged netdev.
>>
>> I see. I just feel a bit curious how lockdep may build the dependency
>> of dpm_list_mtx on rtnl_lock,  and it  is certainly related with
>> lockdep internal.
>
> No, it's just the way drivers/base/power/ works -- it acquires the lock
> when you register a new struct device.

For me, the real puzzle is that  how lockdep introduce  #3
(dpm_list_mtx){+.+.+.}

-> #3 (dpm_list_mtx){+.+.+.}:
      [<ffffffff80271a64>] __lock_acquire+0xc64/0x10a0
      [<ffffffff80271f38>] lock_acquire+0x98/0x140
      [<ffffffff8054e78c>] __mutex_lock_common+0x4c/0x3b0
      [<ffffffff8054ebf6>] mutex_lock_nested+0x46/0x60
      [<ffffffff804532ff>] device_pm_add+0x1f/0xe0
      [<ffffffff8044b9bf>] device_add+0x45f/0x570
      [<ffffffffa007c578>] wiphy_register+0x158/0x280 [cfg80211]
      [<ffffffffa017567c>] ieee80211_register_hw+0xbc/0x410 [mac80211]
      [<ffffffffa01f7c5c>] iwl3945_pci_probe+0xa1c/0x1080 [iwl3945]
      [<ffffffff803c4307>] local_pci_probe+0x17/0x20
      [<ffffffff803c5458>] pci_device_probe+0x88/0xb0
      [<ffffffff8044e1e9>] driver_probe_device+0x89/0x180
      [<ffffffff8044e37b>] __driver_attach+0x9b/0xa0
      [<ffffffff8044d67c>] bus_for_each_dev+0x6c/0xa0
      [<ffffffff8044e03e>] driver_attach+0x1e/0x20
      [<ffffffff8044d955>] bus_add_driver+0xd5/0x290
      [<ffffffff8044e668>] driver_register+0x78/0x140
      [<ffffffff803c56f6>] __pci_register_driver+0x66/0xe0
      [<ffffffffa00bd05c>] 0xffffffffa00bd05c
      [<ffffffff8020904f>] do_one_initcall+0x3f/0x1c0
      [<ffffffff8027d071>] sys_init_module+0xb1/0x200
      [<ffffffff8020c15b>] system_call_fastpath+0x16/0x1b
      [<ffffffffffffffff>] 0xffffffffffffffff

into the lockdep graph?  in which process context?  and what is the
previous held lock?
After all, there is a path ( #0,#1,#2,...,#5 ) in the directed graph
and #3 is added by
add_lock_to_list().

Thanks.

-- 
Lei Ming