Possible deadlock in cfg80211

Possible deadlock in cfg80211

[Markus Theil]

Hi,

while testing hostapd with hwsim, the following deadlock was reported:

[  500.865102] ============================================
[  500.865106] WARNING: possible recursive locking detected
[  500.865112] 5.4.8-arch1-1 #1 Not tainted
[  500.865117] --------------------------------------------
[  500.865122] kworker/3:0/36 is trying to acquire lock:
[  500.865127] ffff8881acdf0d08 (&wdev->mtx){+.+.}, at:
cfg80211_stop_ap+0x37/0x60 [cfg80211]
[  500.865249]
               but task is already holding lock:
[  500.865253] ffff8881acdf0d08 (&wdev->mtx){+.+.}, at:
cfg80211_autodisconnect_wk+0x3d/0x190 [cfg80211]
[  500.865354]
               other info that might help us debug this:
[  500.865358]  Possible unsafe locking scenario:

[  500.865363]        CPU0
[  500.865366]        ----
[  500.865369]   lock(&wdev->mtx);
[  500.865373]   lock(&wdev->mtx);
[  500.865378]
                *** DEADLOCK ***

[  500.865382]  May be due to missing lock nesting notation

[  500.865388] 3 locks held by kworker/3:0/36:
[  500.865391]  #0: ffff8881d4c22148 ((wq_completion)events){+.+.}, at:
process_one_work+0x4ee/0xac0
[  500.865407]  #1: ffff8881d4327dc8
((work_completion)(&wdev->disconnect_wk)){+.+.}, at:
process_one_work+0x4ee/0xac0
[  500.865418]  #2: ffff8881acdf0d08 (&wdev->mtx){+.+.}, at:
cfg80211_autodisconnect_wk+0x3d/0x190 [cfg80211]
[  500.865517]
               stack backtrace:
[  500.865524] CPU: 3 PID: 36 Comm: kworker/3:0 Not tainted 5.4.8-arch1-1 #1
[  500.865530] Hardware name: LENOVO 2349QM6/2349QM6, BIOS G1ETC2WW
(2.82 ) 08/07/2019
[  500.865628] Workqueue: events cfg80211_autodisconnect_wk [cfg80211]
[  500.865634] Call Trace:
[  500.865645]  dump_stack+0xb8/0x110
[  500.865654]  __lock_acquire.cold+0xe2/0x212
[  500.865672]  ? lockdep_hardirqs_on+0x260/0x260
[  500.865681]  ? find_held_lock+0x8c/0xa0
[  500.865693]  lock_acquire+0xe5/0x220
[  500.865772]  ? cfg80211_stop_ap+0x37/0x60 [cfg80211]
[  500.865783]  __mutex_lock+0x137/0xd00
[  500.865864]  ? cfg80211_stop_ap+0x37/0x60 [cfg80211]
[  500.865873]  ? preempt_count_sub+0x14/0xc0
[  500.865880]  ? __mutex_lock+0x71f/0xd00
[  500.865888]  ? find_first_zero_bit+0x28/0x50
[  500.865977]  ? cfg80211_stop_ap+0x37/0x60 [cfg80211]
[  500.865986]  ? mutex_lock_io_nested+0xc20/0xc20
[  500.866090]  ? cfg80211_autodisconnect_wk+0x3d/0x190 [cfg80211]
[  500.866098]  ? mutex_lock_io_nested+0xc20/0xc20
[  500.866109]  ? lockdep_hardirqs_on+0x260/0x260
[  500.866116]  ? match_held_lock+0x2e/0x240
[  500.866127]  ? preempt_count_sub+0x14/0xc0
[  500.866224]  ? cfg80211_stop_ap+0x37/0x60 [cfg80211]
[  500.866322]  cfg80211_stop_ap+0x37/0x60 [cfg80211]
[  500.866433]  cfg80211_autodisconnect_wk+0xd5/0x190 [cfg80211]
[  500.866445]  process_one_work+0x5c0/0xac0
[  500.866460]  ? pwq_dec_nr_in_flight+0x110/0x110
[  500.866480]  worker_thread+0x7a/0x690
[  500.866497]  kthread+0x1ab/0x200
[  500.866505]  ? process_one_work+0xac0/0xac0
[  500.866511]  ? kthread_create_on_node+0xd0/0xd0
[  500.866521]  ret_from_fork+0x3a/0x50

Markus

Re: Possible deadlock in cfg80211

[Johannes Berg]

On Wed, 2020-01-08 at 12:23 +0100, Markus Theil wrote:
> Hi,
> 
> while testing hostapd with hwsim, the following deadlock was reported:

I see you already sent a patch to fix this, thanks :)

johannes