Re: lockdep warning in ieee80211 rx path

[Christian Lamparter <chunkeey@googlemail.com>]

On Sunday 05 September 2010 11:32:26 Ming Lei wrote:
> Seems the warning does not affect use of wireless,  false positive?

No, it's a bug... but please read & test the attached patch.

> [  221.023116] =======================================================
> [  221.023164] [ INFO: possible circular locking dependency detected ]
> [  221.023195] 2.6.36-rc3-next-20100903+ #65
> [  221.023215] -------------------------------------------------------
> [  221.023246] X/2091 is trying to acquire lock:
> [  221.023268]  (slock-AF_INET/1){+.-...}, at: [<ffffffff81381151>]
> tcp_v4_rcv+0x290/0x6b7
> [  221.023323]
> [  221.023323] but task is already holding lock:
> [  221.023354]  (&(&sta->lock)->rlock){+.-...}, at:
> [<ffffffffa01edd37>] sta_rx_agg_reorder_timer_expired+0x61/0x9c
> [mac80211]
> [  221.023425]
> [  221.023426] which lock already depends on the new lock.
> [  221.023426]
> [  221.023469]
> [  221.023469] the existing dependency chain (in reverse order) is:
> [  221.023508]
> [  221.023509] -> #2 (&(&sta->lock)->rlock){+.-...}:
> [  221.023547]        [<ffffffff8107d7e3>] lock_acquire+0xe6/0x113
> [  221.023581]        [<ffffffff813ce6a4>] _raw_spin_lock_irqsave+0x5d/0x97
> [  221.024007]
> [  221.024007] -> #1 (_xmit_ETHER){+.-...}:
> [  221.024007]        [<ffffffff8107d7e3>] lock_acquire+0xe6/0x113
> [  221.024007]        [<ffffffff81340d9c>] netif_receive_skb+0x6c/0x73
> [  221.024007]        [<ffffffffa01fcf27>] ieee80211_rx+0x7b5/0x826 [mac80211]
> [  221.024007]
> [  221.024007] -> #0 (slock-AF_INET/1){+.-...}:
> [  221.024007]        [<ffffffff8107d406>] __lock_acquire+0xa2c/0xd23
> [  221.024007]        [<ffffffff8107d7e3>] lock_acquire+0xe6/0x113
> [  221.024007]        [<ffffffff813ce47b>] _raw_spin_lock_nested+0x43/0x76
> [  221.024007]        [<ffffffff81381151>] tcp_v4_rcv+0x290/0x6b7
> [  221.024007]        [<ffffffff81364424>] ip_local_deliver+0x130/0x1c0
> [  221.024007]        [<ffffffff81364161>] ip_rcv+0x4d9/0x519
> [  221.024007]        [<ffffffff8133fdd9>] __netif_receive_skb+0x292/0x2bf
> [  221.024007]        [<ffffffff81340d9c>] netif_receive_skb+0x6c/0x73
---
[PATCH] mac80211: hoist sta->lock from reorder release timer

The patch "mac80211: AMPDU rx reorder timeout timer" clashes
with "mac80211: use netif_receive_skb in ieee80211_rx callpath"

The timer itself is part of the station's private struct and
it gets killed whenever the station is removed. Therefore
the extra sta->lock protection (that can interferes with the
tx path) is not necessary.

Reported-by: Ming Lei <tom.leiming@gmail.com>
Signed-off-by: Christian Lamparter <chunkeey@googlemail.com>
---
diff --git a/net/mac80211/agg-rx.c b/net/mac80211/agg-rx.c
index 58eab9e..309ed70 100644
--- a/net/mac80211/agg-rx.c
+++ b/net/mac80211/agg-rx.c
@@ -129,9 +129,7 @@ static void sta_rx_agg_reorder_timer_expired(unsigned long data)
 			timer_to_tid[0]);
 
 	rcu_read_lock();
-	spin_lock(&sta->lock);
 	ieee80211_release_reorder_timeout(sta, *ptid);
-	spin_unlock(&sta->lock);
 	rcu_read_unlock();
 }