Re: Suspicious RCU usage in mac80211

[Johannes Berg <johannes@sipsolutions.net>]

On Wed, 2012-04-11 at 22:51 -0500, Larry Finger wrote:
> On 04/11/2012 10:31 PM, Johannes Berg wrote:
> > On Mon, 2012-04-09 at 22:19 -0500, Larry Finger wrote:
> >> When running kernel 3.4-rc2 from wireless testing, I got the following logged:
> >>
> >> [ 2299.344437] ===============================
> >> [ 2299.344443] [ INFO: suspicious RCU usage. ]
> >> [ 2299.344452] 3.4.0-rc2-wl+ #222 Not tainted
> >> [ 2299.344458] -------------------------------
> >> [ 2299.344466] net/mac80211/sta_info.h:449 suspicious rcu_dereference_protected() usage!
> >> [ 2299.344472]
> >> [ 2299.344474] other info that might help us debug this:
> >> [ 2299.344477]
> >> [ 2299.344484]
> >> [ 2299.344486] rcu_scheduler_active = 1, debug_locks = 0
> >> [ 2299.344493] 1 lock held by swapper/1/0:
> >> [ 2299.344499]  #0:  (&tid_tx->session_timer){+.-...}, at: [<ffffffff8104850a>]
> >> run_timer_softirq+0xfa/0x6e0
> >> [ 2299.344533]
> >> [ 2299.344535] stack backtrace:
> >> [ 2299.344544] Pid: 0, comm: swapper/1 Not tainted 3.4.0-rc2-wl+ #222
> >> [ 2299.344551] Call Trace:
> >> [ 2299.344557]<IRQ>   [<ffffffff81092fdd>] lockdep_rcu_suspicious+0xfd/0x130
> >> [ 2299.344651]  [<ffffffffa05db673>] sta_tx_agg_session_timer_expired+0xe3/0x100
> >> [mac80211]
> >
> > That's odd:
> >
> > static inline struct tid_ampdu_tx *
> > rcu_dereference_protected_tid_tx(struct sta_info *sta, int tid)
> > {
> >          return rcu_dereference_protected(sta->ampdu_mlme.tid_tx[tid],
> >                                           lockdep_is_held(&sta->lock) ||
> >                                           lockdep_is_held(&sta->ampdu_mlme.mtx));
> > }
> >
> > sta_tx_agg_session_timer_expired calls ieee80211_stop_tx_ba_session
> > which does:
> >
> >          spin_lock_bh(&sta->lock);
> >          tid_tx = rcu_dereference_protected_tid_tx(sta, tid);
> >
> >
> > So why would the message happen? Strange.
> 
> When I first saw the thread at 
> http://lkml.indiana.edu/hypermail/linux/kernel/1204.1/01446.html, I thought it 
> might be about the same problem, but that appears to be restricted to sparc64.
> 
> The problem does not happen every time - even more strangeness.

What do you mean by "every time"? I wouldn't expect the timer to fire
every time you set up aggregation etc., the timer firing is a special
case to start with.

johannes