The following complaint is reported sometimes when I boot a 4.7.0-rc1 kernel: ========================================================= [ INFO: possible irq lock inversion dependency detected ] 4.7.0-rc1-dbg+ #1 Not tainted --------------------------------------------------------- kworker/u64:8/138 just changed the state of lock: (_xmit_INFINIBAND){+.+...}, at: [<ffffffffa02d360f>] ipoib_dev_addr_changed_valid+0x5f/0x1b0 [ib_ipoib] but this lock was taken by another, SOFTIRQ-safe lock in the past: (&(&mc->mca_lock)->rlock){+.-...} and interrupts could create inverse lock ordering between them. other info that might help us debug this: Possible interrupt unsafe locking scenario: CPU0 CPU1 ---- ---- lock(_xmit_INFINIBAND); local_irq_disable(); lock(&(&mc->mca_lock)->rlock); lock(_xmit_INFINIBAND); <Interrupt> lock(&(&mc->mca_lock)->rlock); *** DEADLOCK *** 2 locks held by kworker/u64:8/138: #0: ("%s""ipoib_flush"){.+.+.+}, at: [<ffffffff810736ed>] process_one_work+0x13d/0x490 #1: ((&priv->flush_light)){+.+.+.}, at: [<ffffffff810736ed>] process_one_work+0x13d/0x490 the shortest dependencies between 2nd lock and 1st lock: -> (&(&mc->mca_lock)->rlock){+.-...} ops: 91 { HARDIRQ-ON-W at: [<ffffffff810a86b6>] __lock_acquire+0x566/0x1480 [<ffffffff810a9990>] lock_acquire+0x60/0x80 [<ffffffff815b159a>] _raw_spin_lock_bh+0x3a/0x50 [<ffffffff815879e5>] igmp6_group_added+0x35/0x190 [<ffffffff815885bf>] ipv6_dev_mc_inc+0x33f/0x470 [<ffffffff815644e7>] ipv6_add_dev+0x3d7/0x4b0 [<ffffffff81b07278>] addrconf_init+0x8e/0x209 [<ffffffff81b07075>] inet6_init+0x1aa/0x30d [<ffffffff81000418>] do_one_initcall+0x38/0x150 [<ffffffff81ac7076>] kernel_init_freeable+0x1b1/0x239 [<ffffffff815a4729>] kernel_init+0x9/0x100 [<ffffffff815b20af>] ret_from_fork+0x1f/0x40 IN-SOFTIRQ-W at: [<ffffffff810a868a>] __lock_acquire+0x53a/0x1480 [<ffffffff810a9990>] lock_acquire+0x60/0x80 [<ffffffff815b159a>] _raw_spin_lock_bh+0x3a/0x50 [<ffffffff81586f18>] mld_ifc_timer_expire+0xe8/0x2b0 [<ffffffff810c05a6>] call_timer_fn+0x66/0xe0 [<ffffffff810c07cc>] run_timer_softirq+0x1ac/0x250 [<ffffffff8105e43f>] __do_softirq+0x10f/0x230 [<ffffffff8105e6e8>] irq_exit+0xa8/0xb0 [<ffffffff8103aa48>] smp_trace_apic_timer_interrupt+0x68/0x90 [<ffffffff8103aa79>] smp_apic_timer_interrupt+0x9/0x10 [<ffffffff815b2aa9>] apic_timer_interrupt+0x89/0x90 [<ffffffff8146ba38>] cpuidle_enter_state+0x138/0x210 [<ffffffff8146bb32>] cpuidle_enter+0x12/0x20 [<ffffffff8109d283>] cpu_startup_entry+0x1c3/0x210 [<ffffffff815a4710>] rest_init+0x130/0x140 [<ffffffff81ac6eb8>] start_kernel+0x3e9/0x3f6 [<ffffffff81ac6294>] x86_64_start_reservations+0x2f/0x31 [<ffffffff81ac63c0>] x86_64_start_kernel+0x12a/0x135 INITIAL USE at: [<ffffffff810a836c>] __lock_acquire+0x21c/0x1480 [<ffffffff810a9990>] lock_acquire+0x60/0x80 [<ffffffff815b159a>] _raw_spin_lock_bh+0x3a/0x50 [<ffffffff815879e5>] igmp6_group_added+0x35/0x190 [<ffffffff815885bf>] ipv6_dev_mc_inc+0x33f/0x470 [<ffffffff815644e7>] ipv6_add_dev+0x3d7/0x4b0 [<ffffffff81b07278>] addrconf_init+0x8e/0x209 [<ffffffff81b07075>] inet6_init+0x1aa/0x30d [<ffffffff81000418>] do_one_initcall+0x38/0x150 [<ffffffff81ac7076>] kernel_init_freeable+0x1b1/0x239 [<ffffffff815a4729>] kernel_init+0x9/0x100 [<ffffffff815b20af>] ret_from_fork+0x1f/0x40 } ... key at: [<ffffffff82940b20>] __key.57090+0x0/0x8 ... acquired at: [<ffffffff810a9990>] lock_acquire+0x60/0x80 [<ffffffff815b159a>] _raw_spin_lock_bh+0x3a/0x50 [<ffffffff814b6f07>] __dev_mc_add+0x27/0x80 [<ffffffff814b6f6b>] dev_mc_add+0xb/0x10 [<ffffffff81587b01>] igmp6_group_added+0x151/0x190 [<ffffffff815885bf>] ipv6_dev_mc_inc+0x33f/0x470 [<ffffffff815644e7>] ipv6_add_dev+0x3d7/0x4b0 [<ffffffff8156a8be>] addrconf_notify+0x3ae/0xd10 [<ffffffff8107b198>] notifier_call_chain+0x58/0x80 [<ffffffff8107b1e1>] raw_notifier_call_chain+0x11/0x20 [<ffffffff814a6220>] call_netdevice_notifiers_info+0x30/0x60 [<ffffffff814b1e35>] register_netdevice+0x455/0x510 [<ffffffff814b1f05>] register_netdev+0x15/0x30 [<ffffffffa02d3189>] ipoib_add_one+0x259/0x490 [ib_ipoib] [<ffffffffa0377485>] ib_register_client+0x65/0xb0 [ib_core] [<ffffffffa02e70f2>] srp_rport_set_state+0x42/0x80 [scsi_transport_srp] [<ffffffff81000418>] do_one_initcall+0x38/0x150 [<ffffffff8111acc9>] do_init_module+0x55/0x1cc [<ffffffff810db55b>] load_module+0x20ab/0x2660 [<ffffffff810dbd53>] SYSC_finit_module+0xc3/0xf0 [<ffffffff810dbd99>] SyS_finit_module+0x9/0x10 [<ffffffff815b1e69>] entry_SYSCALL_64_fastpath+0x1c/0xac > (_xmit_INFINIBAND){+.+...} ops: 5 { HARDIRQ-ON-W at: [<ffffffff810a86b6>] __lock_acquire+0x566/0x1480 [<ffffffff810a9990>] lock_acquire+0x60/0x80 [<ffffffff815b159a>] _raw_spin_lock_bh+0x3a/0x50 [<ffffffff814b6f07>] __dev_mc_add+0x27/0x80 [<ffffffff814b6f6b>] dev_mc_add+0xb/0x10 [<ffffffff81587b01>] igmp6_group_added+0x151/0x190 [<ffffffff815885bf>] ipv6_dev_mc_inc+0x33f/0x470 [<ffffffff815644e7>] ipv6_add_dev+0x3d7/0x4b0 [<ffffffff8156a8be>] addrconf_notify+0x3ae/0xd10 [<ffffffff8107b198>] notifier_call_chain+0x58/0x80 [<ffffffff8107b1e1>] raw_notifier_call_chain+0x11/0x20 [<ffffffff814a6220>] call_netdevice_notifiers_info+0x30/0x60 [<ffffffff814b1e35>] register_netdevice+0x455/0x510 [<ffffffff814b1f05>] register_netdev+0x15/0x30 [<ffffffffa02d3189>] ipoib_add_one+0x259/0x490 [ib_ipoib] [<ffffffffa0377485>] ib_register_client+0x65/0xb0 [ib_core] [<ffffffffa02e70f2>] srp_rport_set_state+0x42/0x80 [scsi_transport_srp] [<ffffffff81000418>] do_one_initcall+0x38/0x150 [<ffffffff8111acc9>] do_init_module+0x55/0x1cc [<ffffffff810db55b>] load_module+0x20ab/0x2660 [<ffffffff810dbd53>] SYSC_finit_module+0xc3/0xf0 [<ffffffff810dbd99>] SyS_finit_module+0x9/0x10 [<ffffffff815b1e69>] entry_SYSCALL_64_fastpath+0x1c/0xac SOFTIRQ-ON-W at: [<ffffffff810a86e8>] __lock_acquire+0x598/0x1480 [<ffffffff810a9990>] lock_acquire+0x60/0x80 [<ffffffff815b1453>] _raw_spin_lock+0x33/0x50 [<ffffffffa02d360f>] ipoib_dev_addr_changed_valid+0x5f/0x1b0 [ib_ipoib] [<ffffffffa02d5875>] __ipoib_ib_dev_flush+0x125/0x440 [ib_ipoib] [<ffffffffa02d5c54>] ipoib_ib_dev_flush_light+0x14/0x20 [ib_ipoib] [<ffffffff8107374d>] process_one_work+0x19d/0x490 [<ffffffff81073a89>] worker_thread+0x49/0x490 [<ffffffff8107a04a>] kthread+0xea/0x100 [<ffffffff815b20af>] ret_from_fork+0x1f/0x40 INITIAL USE at: [<ffffffff810a836c>] __lock_acquire+0x21c/0x1480 [<ffffffff810a9990>] lock_acquire+0x60/0x80 [<ffffffff815b159a>] _raw_spin_lock_bh+0x3a/0x50 [<ffffffff814b6f07>] __dev_mc_add+0x27/0x80 [<ffffffff814b6f6b>] dev_mc_add+0xb/0x10 [<ffffffff81587b01>] igmp6_group_added+0x151/0x190 [<ffffffff815885bf>] ipv6_dev_mc_inc+0x33f/0x470 [<ffffffff815644e7>] ipv6_add_dev+0x3d7/0x4b0 [<ffffffff8156a8be>] addrconf_notify+0x3ae/0xd10 [<ffffffff8107b198>] notifier_call_chain+0x58/0x80 [<ffffffff8107b1e1>] raw_notifier_call_chain+0x11/0x20 [<ffffffff814a6220>] call_netdevice_notifiers_info+0x30/0x60 [<ffffffff814b1e35>] register_netdevice+0x455/0x510 [<ffffffff814b1f05>] register_netdev+0x15/0x30 [<ffffffffa02d3189>] ipoib_add_one+0x259/0x490 [ib_ipoib] [<ffffffffa0377485>] ib_register_client+0x65/0xb0 [ib_core] [<ffffffffa02e70f2>] srp_rport_set_state+0x42/0x80 [scsi_transport_srp] [<ffffffff81000418>] do_one_initcall+0x38/0x150 [<ffffffff8111acc9>] do_init_module+0x55/0x1cc [<ffffffff810db55b>] load_module+0x20ab/0x2660 [<ffffffff810dbd53>] SYSC_finit_module+0xc3/0xf0 [<ffffffff810dbd99>] SyS_finit_module+0x9/0x10 [<ffffffff815b1e69>] entry_SYSCALL_64_fastpath+0x1c/0xac } ... key at: [<ffffffff8292b3d0>] netdev_addr_lock_key+0x70/0x1e0 ... acquired at: [<ffffffff810a6ba9>] check_usage_backwards+0x129/0x130 [<ffffffff810a7966>] mark_lock+0x336/0x610 [<ffffffff810a86e8>] __lock_acquire+0x598/0x1480 [<ffffffff810a9990>] lock_acquire+0x60/0x80 [<ffffffff815b1453>] _raw_spin_lock+0x33/0x50 [<ffffffffa02d360f>] ipoib_dev_addr_changed_valid+0x5f/0x1b0 [ib_ipoib] [<ffffffffa02d5875>] __ipoib_ib_dev_flush+0x125/0x440 [ib_ipoib] [<ffffffffa02d5c54>] ipoib_ib_dev_flush_light+0x14/0x20 [ib_ipoib] [<ffffffff8107374d>] process_one_work+0x19d/0x490 [<ffffffff81073a89>] worker_thread+0x49/0x490 [<ffffffff8107a04a>] kthread+0xea/0x100 [<ffffffff815b20af>] ret_from_fork+0x1f/0x40 stack backtrace: CPU: 5 PID: 138 Comm: kworker/u64:8 Not tainted 4.7.0-rc1-dbg+ #1 Hardware name: Dell Inc. PowerEdge R430/03XKDV, BIOS 1.0.2 11/17/2014 Workqueue: ipoib_flush ipoib_ib_dev_flush_light [ib_ipoib] 0000000000000000 ffff88045d53bac0 ffffffff812c71a5 ffffffff826d0720 ffff88045d53bb20 ffff88045d53bb00 ffffffff8111a93c ffff88045d5aea88 ffff88045d5aea88 ffff88045d5ae240 ffffffff81797209 0000000000000000 Call Trace: [<ffffffff812c71a5>] dump_stack+0x67/0x92 [<ffffffff8111a93c>] print_irq_inversion_bug.part.36+0x1a4/0x1b0 [<ffffffff810a6ba9>] check_usage_backwards+0x129/0x130 [<ffffffff810a7966>] mark_lock+0x336/0x610 [<ffffffff810a6a80>] ? print_shortest_lock_dependencies+0x1a0/0x1a0 [<ffffffff810a86e8>] __lock_acquire+0x598/0x1480 [<ffffffff81178509>] ? kfree+0xa9/0x150 [<ffffffffa03d5a81>] ? __mlx4_ib_query_gid+0x111/0x1b0 [mlx4_ib] [<ffffffff811785a4>] ? kfree+0x144/0x150 [<ffffffff810a9990>] lock_acquire+0x60/0x80 [<ffffffffa02d360f>] ? ipoib_dev_addr_changed_valid+0x5f/0x1b0 [ib_ipoib] [<ffffffff815b1453>] _raw_spin_lock+0x33/0x50 [<ffffffffa02d360f>] ? ipoib_dev_addr_changed_valid+0x5f/0x1b0 [ib_ipoib] [<ffffffffa02d360f>] ipoib_dev_addr_changed_valid+0x5f/0x1b0 [ib_ipoib] [<ffffffffa02d5875>] __ipoib_ib_dev_flush+0x125/0x440 [ib_ipoib] [<ffffffffa02d5c54>] ipoib_ib_dev_flush_light+0x14/0x20 [ib_ipoib] [<ffffffff8107374d>] process_one_work+0x19d/0x490 [<ffffffff810736ed>] ? process_one_work+0x13d/0x490 [<ffffffff81073a89>] worker_thread+0x49/0x490 [<ffffffff81073a40>] ? process_one_work+0x490/0x490 [<ffffffff8107a04a>] kthread+0xea/0x100 [<ffffffff815b20af>] ret_from_fork+0x1f/0x40 [<ffffffff81079f60>] ? kthread_create_on_node+0x1f0/0x1f0
Created attachment 218811 [details] Locking fix Hi, I have a fix for this issue, was planning to send it on Sunday. I've attached it here if you would like to have a look at it.
Thanks for having provided this patch! I will start testing it.
What's the latest status of this bug? Sincerely, your regression tracker for Linux 4.7 (http://bit.ly/28JRmJo )
The attached patch has been accepted upstream (commit 9b29953bf8ca) so I think this bug can be closed.
(In reply to Bart Van Assche from comment #4) > I think this bug can be closed. Please do that, you should be able to do it; I'm lacking permissions for it :-/