]> git.kernelconcepts.de Git - karo-tx-linux.git/commitdiff
rcu: Fix soft lockup for rcu_nocb_kthread
authorDing Tianhong <dingtianhong@huawei.com>
Wed, 15 Jun 2016 07:27:36 +0000 (15:27 +0800)
committerPaul E. McKenney <paulmck@linux.vnet.ibm.com>
Mon, 22 Aug 2016 14:53:20 +0000 (07:53 -0700)
Carrying out the following steps results in a softlockup in the
RCU callback-offload (rcuo) kthreads:

1. Connect to ixgbevf, and set the speed to 10Gb/s.
2. Use ifconfig to bring the nic up and down repeatedly.

[  317.005148] IPv6: ADDRCONF(NETDEV_CHANGE): eth2: link becomes ready
[  368.106005] BUG: soft lockup - CPU#1 stuck for 22s! [rcuos/1:15]
[  368.106005] Hardware name: Bochs Bochs, BIOS Bochs 01/01/2011
[  368.106005] task: ffff88057dd8a220 ti: ffff88057dd9c000 task.ti: ffff88057dd9c000
[  368.106005] RIP: 0010:[<ffffffff81579e04>]  [<ffffffff81579e04>] fib_table_lookup+0x14/0x390
[  368.106005] RSP: 0018:ffff88061fc83ce8  EFLAGS: 00000286
[  368.106005] RAX: 0000000000000001 RBX: 00000000020155c0 RCX: 0000000000000001
[  368.106005] RDX: ffff88061fc83d50 RSI: ffff88061fc83d70 RDI: ffff880036d11a00
[  368.106005] RBP: ffff88061fc83d08 R08: 0000000000000001 R09: 0000000000000000
[  368.106005] R10: ffff880036d11a00 R11: ffffffff819e0900 R12: ffff88061fc83c58
[  368.106005] R13: ffffffff816154dd R14: ffff88061fc83d08 R15: 00000000020155c0
[  368.106005] FS:  0000000000000000(0000) GS:ffff88061fc80000(0000) knlGS:0000000000000000
[  368.106005] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[  368.106005] CR2: 00007f8c2aee9c40 CR3: 000000057b222000 CR4: 00000000000407e0
[  368.106005] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[  368.106005] DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
[  368.106005] Stack:
[  368.106005]  00000000010000c0 ffff88057b766000 ffff8802e380b000 ffff88057af03e00
[  368.106005]  ffff88061fc83dc0 ffffffff815349a6 ffff88061fc83d40 ffffffff814ee146
[  368.106005]  ffff8802e380af00 00000000e380af00 ffffffff819e0900 020155c0010000c0
[  368.106005] Call Trace:
[  368.106005]  <IRQ>
[  368.106005]
[  368.106005]  [<ffffffff815349a6>] ip_route_input_noref+0x516/0xbd0
[  368.106005]  [<ffffffff814ee146>] ? skb_release_data+0xd6/0x110
[  368.106005]  [<ffffffff814ee20a>] ? kfree_skb+0x3a/0xa0
[  368.106005]  [<ffffffff8153698f>] ip_rcv_finish+0x29f/0x350
[  368.106005]  [<ffffffff81537034>] ip_rcv+0x234/0x380
[  368.106005]  [<ffffffff814fd656>] __netif_receive_skb_core+0x676/0x870
[  368.106005]  [<ffffffff814fd868>] __netif_receive_skb+0x18/0x60
[  368.106005]  [<ffffffff814fe4de>] process_backlog+0xae/0x180
[  368.106005]  [<ffffffff814fdcb2>] net_rx_action+0x152/0x240
[  368.106005]  [<ffffffff81077b3f>] __do_softirq+0xef/0x280
[  368.106005]  [<ffffffff8161619c>] call_softirq+0x1c/0x30
[  368.106005]  <EOI>
[  368.106005]
[  368.106005]  [<ffffffff81015d95>] do_softirq+0x65/0xa0
[  368.106005]  [<ffffffff81077174>] local_bh_enable+0x94/0xa0
[  368.106005]  [<ffffffff81114922>] rcu_nocb_kthread+0x232/0x370
[  368.106005]  [<ffffffff81098250>] ? wake_up_bit+0x30/0x30
[  368.106005]  [<ffffffff811146f0>] ? rcu_start_gp+0x40/0x40
[  368.106005]  [<ffffffff8109728f>] kthread+0xcf/0xe0
[  368.106005]  [<ffffffff810971c0>] ? kthread_create_on_node+0x140/0x140
[  368.106005]  [<ffffffff816147d8>] ret_from_fork+0x58/0x90
[  368.106005]  [<ffffffff810971c0>] ? kthread_create_on_node+0x140/0x140

==================================cut here==============================

It turns out that the rcuos callback-offload kthread is busy processing
a very large quantity of RCU callbacks, and it is not reliquishing the
CPU while doing so.  This commit therefore adds an cond_resched_rcu_qs()
within the loop to allow other tasks to run.

Signed-off-by: Ding Tianhong <dingtianhong@huawei.com>
[ paulmck: Substituted cond_resched_rcu_qs for cond_resched. ]
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
kernel/rcu/tree_plugin.h

index 0082fce402a0c11c922648ae0d90a9d2bd85561f..85c5a883c6e31047194a8c74603ce71ab8381f67 100644 (file)
@@ -2173,6 +2173,7 @@ static int rcu_nocb_kthread(void *arg)
                                cl++;
                        c++;
                        local_bh_enable();
+                       cond_resched_rcu_qs();
                        list = next;
                }
                trace_rcu_batch_end(rdp->rsp->name, c, !!list, 0, 0, 1);