lockdep complains use of uninitialized spinlock at ieee80211_do_stop() [1],
for commit f856373e2f31ffd3 ("wifi: mac80211: do not wake queues on a vif
that is being stopped") guards clear_bit() using fq.lock even before
fq_init() from ieee80211_txq_setup_flows() initializes this spinlock.
According to discussion [2], Toke was not happy with expanding usage of
fq.lock. Since __ieee80211_wake_txqs() is called under RCU read lock, we
can instead use synchronize_rcu() for flushing ieee80211_wake_txqs().
Link: https://syzkaller.appspot.com/bug?extid=eceab52db7c4b961e9d6 [1]
Link: https://lkml.kernel.org/r/874k0zowh2.fsf@xxxxxxx [2]
Reported-by: syzbot <syzbot+eceab52db7c4b961e9d6@xxxxxxxxxxxxxxxxxxxxxxxxx>
Signed-off-by: Tetsuo Handa <penguin-kernel@xxxxxxxxxxxxxxxxxxx>
Fixes: f856373e2f31ffd3 ("wifi: mac80211: do not wake queues on a vif that is being stopped")
Tested-by: syzbot <syzbot+eceab52db7c4b961e9d6@xxxxxxxxxxxxxxxxxxxxxxxxx>
---
Changes in v2:
Use synchronize_rcu() instead of initializing fq.lock early.
This bug is current top crasher for syzbot. Please fix as soon as possible.
net/mac80211/iface.c | 3 +--
1 file changed, 1 insertion(+), 2 deletions(-)
diff --git a/net/mac80211/iface.c b/net/mac80211/iface.c
index 15a73b7fdd75..1a9ada411879 100644
--- a/net/mac80211/iface.c
+++ b/net/mac80211/iface.c
@@ -377,9 +377,8 @@ static void ieee80211_do_stop(struct ieee80211_sub_if_data *sdata, bool going_do
bool cancel_scan;
struct cfg80211_nan_func *func;
- spin_lock_bh(&local->fq.lock);
clear_bit(SDATA_STATE_RUNNING, &sdata->state);
- spin_unlock_bh(&local->fq.lock);
+ synchronize_rcu(); /* flush _ieee80211_wake_txqs() */
cancel_scan = rcu_access_pointer(local->scan_sdata) == sdata;
if (cancel_scan)
--
2.18.4
