Spotted this in dmesg while investigating why my wireless broke
sometime between next-20141201 and next-20141208. Probably not
related, as wireless has been broken on several boot attempts of -1208,
but this has popped only once....
Looks like the fault of
commit 46e5da40aec256155cfedee96dd21a75da941f2c
Author: John Fastabend <[email protected]>
Date: Fri Sep 12 20:04:52 2014 -0700
net: qdisc: use rcu prefix and silence sparse warnings
but I freely admit that exhausts my RCU debugging skills. :)
[ 104.743212] ===============================
[ 104.743214] [ INFO: suspicious RCU usage. ]
[ 104.743218] 3.18.0-next-20141208 #27 Tainted: G OE
[ 104.743219] -------------------------------
[ 104.743224] net/sched/sch_fq_codel.c:97 suspicious rcu_dereference_check() usage!
[ 104.743227]
other info that might help us debug this:
[ 104.743231]
rcu_scheduler_active = 1, debug_locks = 1
[ 104.743236] 4 locks held by unbound/1873:
[ 104.743239] #0: (rcu_read_lock_bh){......}, at: [<ffffffff8a537066>] rcu_lock_acquire+0x0/0x22
[ 104.743257] #1: (rcu_read_lock_bh){......}, at: [<ffffffff8a4ebd44>] rcu_lock_acquire+0x0/0x22
[ 104.743274] #2: (dev->qdisc_tx_busylock ?: &qdisc_tx_busylock){+.....}, at: [<ffffffff8a4f3ea4>] __dev_queue_xmit+0x32a/0x66a
[ 104.743292] #3: (&qdisc_tx_lock){+.-...}, at: [<ffffffff8a4f3ead>] __dev_queue_xmit+0x333/0x66a
[ 104.743302]
stack backtrace:
[ 104.743307] CPU: 2 PID: 1873 Comm: unbound Tainted: G OE 3.18.0-next-20141208 #27
[ 104.743309] Hardware name: Dell Inc. Latitude E6530/07Y85M, BIOS A15 06/20/2014
[ 104.743312] 0000000000000001 ffff8800b9f3f978 ffffffff8a6ba925 ffff880223bac210
[ 104.743317] 0000000000000001 ffff8800b9f3f9a8 ffffffff8a07f124 ffff8800b9ce4e00
[ 104.743323] ffff880223fea600 0000000000000000 ffff8800c6e2c000 ffff8800b9f3f9f8
[ 104.743329] Call Trace:
[ 104.743336] [<ffffffff8a6ba925>] dump_stack+0x50/0xa8
[ 104.743343] [<ffffffff8a07f124>] lockdep_rcu_suspicious+0xea/0xf3
[ 104.743348] [<ffffffff8a514903>] fq_codel_enqueue+0x79/0x21c
[ 104.743353] [<ffffffff8a6c5228>] ? _raw_spin_lock+0x37/0x3f
[ 104.743357] [<ffffffff8a4f3fb5>] __dev_queue_xmit+0x43b/0x66a
[ 104.743362] [<ffffffff8a4f41ef>] dev_queue_xmit+0xb/0xd
[ 104.743367] [<ffffffff8a537beb>] ip_finish_output2+0x42b/0x46c
[ 104.743371] [<ffffffff8a53838f>] ip_finish_output+0x13c/0x144
[ 104.743376] [<ffffffff8a5393f1>] ip_output+0x4c/0x70
[ 104.743381] [<ffffffff8a538d92>] ip_local_out_sk+0x54/0x5f
[ 104.743385] [<ffffffff8a539c2f>] ip_send_skb+0x12/0x37
[ 104.743390] [<ffffffff8a55d145>] udp_send_skb+0x176/0x1db
[ 104.743394] [<ffffffff8a55d839>] udp_sendmsg+0x63e/0x87c
[ 104.743398] [<ffffffff8a537379>] ? skb_set_owner_w+0x56/0x56
[ 104.743404] [<ffffffff8a09011d>] ? __rcu_read_unlock+0xf5/0xfd
[ 104.743408] [<ffffffff8a568135>] ? rcu_read_unlock+0x3e/0x5d
[ 104.743413] [<ffffffff8a568af9>] inet_sendmsg+0x3d/0x66
[ 104.743419] [<ffffffff8a4d6a4b>] __sock_sendmsg_nosec+0x25/0x27
[ 104.743422] [<ffffffff8a4d80da>] sock_sendmsg+0x5a/0x7b
[ 104.743429] [<ffffffff8a12581b>] ? might_fault+0xae/0xb0
[ 104.743433] [<ffffffff8a4d6c22>] ? audit_sockaddr+0x2b/0x44
[ 104.743437] [<ffffffff8a4da1ff>] SYSC_sendto+0x119/0x15c
[ 104.743443] [<ffffffff8a00e2b5>] ? do_audit_syscall_entry+0xb8/0xc7
[ 104.743447] [<ffffffff8a4da85d>] SyS_sendto+0x9/0xb
[ 104.743453] [<ffffffff8a6c5fd2>] system_call_fastpath+0x12/0x17
On Tue, Dec 9, 2014 at 9:51 AM, Valdis Kletnieks
<[email protected]> wrote:
> Spotted this in dmesg while investigating why my wireless broke
> sometime between next-20141201 and next-20141208. Probably not
> related, as wireless has been broken on several boot attempts of -1208,
> but this has popped only once....
>
> Looks like the fault of
>
> commit 46e5da40aec256155cfedee96dd21a75da941f2c
> Author: John Fastabend <[email protected]>
> Date: Fri Sep 12 20:04:52 2014 -0700
Well, its a (harmless) typo : this should really be an
rcu_dereference_bh() instead of rcu_dereference()
On Tue, 09 Dec 2014 10:07:31 -0800, Eric Dumazet said:
> On Tue, Dec 9, 2014 at 9:51 AM, Valdis Kletnieks
> > commit 46e5da40aec256155cfedee96dd21a75da941f2c
> > Author: John Fastabend <[email protected]>
> > Date: Fri Sep 12 20:04:52 2014 -0700
>
> Well, its a (harmless) typo : this should really be an
> rcu_dereference_bh() instead of rcu_dereference()
You want a patch, or you got it already?
On Tue, Dec 9, 2014 at 10:54 AM, <[email protected]> wrote:
> You want a patch, or you got it already?
>
If you could submit it formally, this would be great , thanks !
diff --git a/net/sched/sch_fq_codel.c b/net/sched/sch_fq_codel.c
index b9ca32ebc1de..1e52decb7b59 100644
--- a/net/sched/sch_fq_codel.c
+++ b/net/sched/sch_fq_codel.c
@@ -94,7 +94,7 @@ static unsigned int fq_codel_classify(struct sk_buff
*skb, struct Qdisc *sch,
TC_H_MIN(skb->priority) <= q->flows_cnt)
return TC_H_MIN(skb->priority);
- filter = rcu_dereference(q->filter_list);
+filter = rcu_dereference_bh(q->filter_list);
if (!filter)
return fq_codel_hash(q, skb) + 1;