Re: Suspicious RCU usage in mac80211

[Catalin Marinas <catalin.marinas@arm.com>]

On Tue, May 01, 2012 at 08:18:54PM +0100, Larry Finger wrote:
> On 05/01/2012 09:25 AM, Mohammed Shafi wrote:
> > i just looked into the code and compared it with
> > sta_addba_resp_timer_expired and thought whether we should do
> > this..
> 
> Your patch does not help. I still get the following dump in the log:
> 
> ===============================
> [ INFO: suspicious RCU usage. ]
> 3.4.0-rc5-wl+ #287 Not tainted
> -------------------------------
> net/mac80211/sta_info.h:456 suspicious rcu_dereference_protected() usage!
> 
> other info that might help us debug this:
> 
> rcu_scheduler_active = 1, debug_locks = 0
> 3 locks held by kmemleak/622:
> #0:  (scan_mutex){+.+...}, at: [<ffffffff8113b0d6>] kmemleak_scan_thread+0x56/0xd0
> #1:  (&tid_tx->session_timer){+.-...}, at: [<ffffffff8104853a>] 
> run_timer_softirq+0xfa/0x6e0
> #2:  (rcu_read_lock){.+.+..}, at: [<ffffffffa0449ff0>] 
> sta_tx_agg_session_timer_expired+0x0/0x2a0 [mac80211]
> 
> stack backtrace:
> Pid: 622, comm: kmemleak Not tainted 3.4.0-rc5-wl+ #287
> Call Trace:
>   <IRQ>  [<ffffffff8109309d>] lockdep_rcu_suspicious+0xfd/0x130
>   [<ffffffffa044a1cf>] sta_tx_agg_session_timer_expired+0x1df/0x2a0 [mac80211]
>   [<ffffffffa0449ff0>] ? ieee80211_start_tx_ba_session+0x450/0x450 [mac80211]
>   [<ffffffff810485c5>] run_timer_softirq+0x185/0x6e0
> 
> As kmemleak seems to be involved, I have added Catalin Marinas to the Cc list.

It doesn't look related to kmemleak. It just happens that
sta_tx_agg_session_timer_expired() was called from a timer when the
kmemleak scanning thread was running (pretty likely as this thread can
take minutes to complete).

Looking at the code and the logs, ieee80211_start_tx_ba_session() calls
rcu_dereference_protected_tid_tx() which calls
rcu_dereference_protected() with the (lockdep_is_held(&sta->lock) ||
lockdep_is_held(&sta->ampdu_mlme.mtx)) condition which is false. As the
kernel log says, none of these locks are held, hence the warning.

Either the rcu_dereference annotation is wrong (introduced by commit
40b275 "mac80211: sparse RCU annotations") or there is something wrong
in the code and the RCU pointer is dereferenced without any of those
locks held.

-- 
Catalin