tipc: fix lockdep warning when intra-node messages are delivered
authorYing Xue <ying.xue@windriver.com>
Mon, 20 Oct 2014 06:46:35 +0000 (14:46 +0800)
committerDavid S. Miller <davem@davemloft.net>
Tue, 21 Oct 2014 19:28:15 +0000 (15:28 -0400)
When running tipcTC&tipcTS test suite, below lockdep unsafe locking
scenario is reported:

[ 1109.997854]
[ 1109.997988] =================================
[ 1109.998290] [ INFO: inconsistent lock state ]
[ 1109.998575] 3.17.0-rc1+ #113 Not tainted
[ 1109.998762] ---------------------------------
[ 1109.998762] inconsistent {SOFTIRQ-ON-W} -> {IN-SOFTIRQ-W} usage.
[ 1109.998762] swapper/7/0 [HC0[0]:SC1[1]:HE1:SE0] takes:
[ 1109.998762]  (slock-AF_TIPC){+.?...}, at: [<ffffffffa0011969>] tipc_sk_rcv+0x49/0x2b0 [tipc]
[ 1109.998762] {SOFTIRQ-ON-W} state was registered at:
[ 1109.998762]   [<ffffffff810a4770>] __lock_acquire+0x6a0/0x1d80
[ 1109.998762]   [<ffffffff810a6555>] lock_acquire+0x95/0x1e0
[ 1109.998762]   [<ffffffff81a2d1ce>] _raw_spin_lock+0x3e/0x80
[ 1109.998762]   [<ffffffffa0011969>] tipc_sk_rcv+0x49/0x2b0 [tipc]
[ 1109.998762]   [<ffffffffa0004fe8>] tipc_link_xmit+0xa8/0xc0 [tipc]
[ 1109.998762]   [<ffffffffa000ec6f>] tipc_sendmsg+0x15f/0x550 [tipc]
[ 1109.998762]   [<ffffffffa000f165>] tipc_connect+0x105/0x140 [tipc]
[ 1109.998762]   [<ffffffff817676ee>] SYSC_connect+0xae/0xc0
[ 1109.998762]   [<ffffffff81767b7e>] SyS_connect+0xe/0x10
[ 1109.998762]   [<ffffffff817a9788>] compat_SyS_socketcall+0xb8/0x200
[ 1109.998762]   [<ffffffff81a306e5>] sysenter_dispatch+0x7/0x1f
[ 1109.998762] irq event stamp: 241060
[ 1109.998762] hardirqs last  enabled at (241060): [<ffffffff8105a4ad>] __local_bh_enable_ip+0x6d/0xd0
[ 1109.998762] hardirqs last disabled at (241059): [<ffffffff8105a46f>] __local_bh_enable_ip+0x2f/0xd0
[ 1109.998762] softirqs last  enabled at (241020): [<ffffffff81059a52>] _local_bh_enable+0x22/0x50
[ 1109.998762] softirqs last disabled at (241021): [<ffffffff8105a626>] irq_exit+0x96/0xc0
[ 1109.998762]
[ 1109.998762] other info that might help us debug this:
[ 1109.998762]  Possible unsafe locking scenario:
[ 1109.998762]
[ 1109.998762]        CPU0
[ 1109.998762]        ----
[ 1109.998762]   lock(slock-AF_TIPC);
[ 1109.998762]   <Interrupt>
[ 1109.998762]     lock(slock-AF_TIPC);
[ 1109.998762]
[ 1109.998762]  *** DEADLOCK ***
[ 1109.998762]
[ 1109.998762] 2 locks held by swapper/7/0:
[ 1109.998762]  #0:  (rcu_read_lock){......}, at: [<ffffffff81782dc9>] __netif_receive_skb_core+0x69/0xb70
[ 1109.998762]  #1:  (rcu_read_lock){......}, at: [<ffffffffa0001c90>] tipc_l2_rcv_msg+0x40/0x260 [tipc]
[ 1109.998762]
[ 1109.998762] stack backtrace:
[ 1109.998762] CPU: 7 PID: 0 Comm: swapper/7 Not tainted 3.17.0-rc1+ #113
[ 1109.998762] Hardware name: Bochs Bochs, BIOS Bochs 01/01/2007
[ 1109.998762]  ffffffff82745830 ffff880016c03828 ffffffff81a209eb 0000000000000007
[ 1109.998762]  ffff880017b3cac0 ffff880016c03888 ffffffff81a1c5ef 0000000000000001
[ 1109.998762]  ffff880000000001 ffff880000000000 ffffffff81012d4f 0000000000000000
[ 1109.998762] Call Trace:
[ 1109.998762]  <IRQ>  [<ffffffff81a209eb>] dump_stack+0x4e/0x68
[ 1109.998762]  [<ffffffff81a1c5ef>] print_usage_bug+0x1f1/0x202
[ 1109.998762]  [<ffffffff81012d4f>] ? save_stack_trace+0x2f/0x50
[ 1109.998762]  [<ffffffff810a406c>] mark_lock+0x28c/0x2f0
[ 1109.998762]  [<ffffffff810a3440>] ? print_irq_inversion_bug.part.46+0x1f0/0x1f0
[ 1109.998762]  [<ffffffff810a467d>] __lock_acquire+0x5ad/0x1d80
[ 1109.998762]  [<ffffffff810a70dd>] ? trace_hardirqs_on+0xd/0x10
[ 1109.998762]  [<ffffffff8108ace8>] ? sched_clock_cpu+0x98/0xc0
[ 1109.998762]  [<ffffffff8108ad2b>] ? local_clock+0x1b/0x30
[ 1109.998762]  [<ffffffff810a10dc>] ? lock_release_holdtime.part.29+0x1c/0x1a0
[ 1109.998762]  [<ffffffff8108aa05>] ? sched_clock_local+0x25/0x90
[ 1109.998762]  [<ffffffffa000dec0>] ? tipc_sk_get+0x60/0x80 [tipc]
[ 1109.998762]  [<ffffffff810a6555>] lock_acquire+0x95/0x1e0
[ 1109.998762]  [<ffffffffa0011969>] ? tipc_sk_rcv+0x49/0x2b0 [tipc]
[ 1109.998762]  [<ffffffff810a6fb6>] ? trace_hardirqs_on_caller+0xa6/0x1c0
[ 1109.998762]  [<ffffffff81a2d1ce>] _raw_spin_lock+0x3e/0x80
[ 1109.998762]  [<ffffffffa0011969>] ? tipc_sk_rcv+0x49/0x2b0 [tipc]
[ 1109.998762]  [<ffffffffa000dec0>] ? tipc_sk_get+0x60/0x80 [tipc]
[ 1109.998762]  [<ffffffffa0011969>] tipc_sk_rcv+0x49/0x2b0 [tipc]
[ 1109.998762]  [<ffffffffa00076bd>] tipc_rcv+0x5ed/0x960 [tipc]
[ 1109.998762]  [<ffffffffa0001d1c>] tipc_l2_rcv_msg+0xcc/0x260 [tipc]
[ 1109.998762]  [<ffffffffa0001c90>] ? tipc_l2_rcv_msg+0x40/0x260 [tipc]
[ 1109.998762]  [<ffffffff81783345>] __netif_receive_skb_core+0x5e5/0xb70
[ 1109.998762]  [<ffffffff81782dc9>] ? __netif_receive_skb_core+0x69/0xb70
[ 1109.998762]  [<ffffffff81784eb9>] ? dev_gro_receive+0x259/0x4e0
[ 1109.998762]  [<ffffffff817838f6>] __netif_receive_skb+0x26/0x70
[ 1109.998762]  [<ffffffff81783acd>] netif_receive_skb_internal+0x2d/0x1f0
[ 1109.998762]  [<ffffffff81785518>] napi_gro_receive+0xd8/0x240
[ 1109.998762]  [<ffffffff815bf854>] e1000_clean_rx_irq+0x2c4/0x530
[ 1109.998762]  [<ffffffff815c1a46>] e1000_clean+0x266/0x9c0
[ 1109.998762]  [<ffffffff8108ad2b>] ? local_clock+0x1b/0x30
[ 1109.998762]  [<ffffffff8108aa05>] ? sched_clock_local+0x25/0x90
[ 1109.998762]  [<ffffffff817842b1>] net_rx_action+0x141/0x310
[ 1109.998762]  [<ffffffff810bd710>] ? handle_fasteoi_irq+0xe0/0x150
[ 1109.998762]  [<ffffffff81059fa6>] __do_softirq+0x116/0x4d0
[ 1109.998762]  [<ffffffff8105a626>] irq_exit+0x96/0xc0
[ 1109.998762]  [<ffffffff81a30d07>] do_IRQ+0x67/0x110
[ 1109.998762]  [<ffffffff81a2ee2f>] common_interrupt+0x6f/0x6f
[ 1109.998762]  <EOI>  [<ffffffff8100d2b7>] ? default_idle+0x37/0x250
[ 1109.998762]  [<ffffffff8100d2b5>] ? default_idle+0x35/0x250
[ 1109.998762]  [<ffffffff8100dd1f>] arch_cpu_idle+0xf/0x20
[ 1109.998762]  [<ffffffff810999fd>] cpu_startup_entry+0x27d/0x4d0
[ 1109.998762]  [<ffffffff81034c78>] start_secondary+0x188/0x1f0

When intra-node messages are delivered from one process to another
process, tipc_link_xmit() doesn't disable BH before it directly calls
tipc_sk_rcv() on process context to forward messages to destination
socket. Meanwhile, if messages delivered by remote node arrive at the
node and their destinations are also the same socket, tipc_sk_rcv()
running on process context might be preempted by tipc_sk_rcv() running
BH context. As a result, the latter cannot obtain the socket lock as
the lock was obtained by the former, however, the former has no chance
to be run as the latter is owning the CPU now, so headlock happens. To
avoid it, BH should be always disabled in tipc_sk_rcv().

Signed-off-by: Ying Xue <ying.xue@windriver.com>
Reviewed-by: Jon Maloy <jon.maloy@ericsson.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/tipc/socket.c

index 3043f10dc32804d98bcedd7db27868ea883db2be..51bddc236a15582b1a1dd0deb0b1a733c1d40070 100644 (file)
@@ -1776,7 +1776,7 @@ int tipc_sk_rcv(struct sk_buff *buf)
        sk = &tsk->sk;
 
        /* Queue message */
-       bh_lock_sock(sk);
+       spin_lock_bh(&sk->sk_lock.slock);
 
        if (!sock_owned_by_user(sk)) {
                rc = filter_rcv(sk, buf);
@@ -1787,7 +1787,7 @@ int tipc_sk_rcv(struct sk_buff *buf)
                if (sk_add_backlog(sk, buf, limit))
                        rc = -TIPC_ERR_OVERLOAD;
        }
-       bh_unlock_sock(sk);
+       spin_unlock_bh(&sk->sk_lock.slock);
        tipc_sk_put(tsk);
        if (likely(!rc))
                return 0;