[SCSI] bnx2fc: Fixed scsi_remove_target soft lockup when rmmod bnx2x
authorEddie Wai <eddie.wai@broadcom.com>
Wed, 11 Dec 2013 23:30:21 +0000 (15:30 -0800)
committerJames Bottomley <JBottomley@Parallels.com>
Sat, 15 Mar 2014 17:17:47 +0000 (10:17 -0700)
The problem has been identified to be a change in the scsi_remove_device
path where a call to the pm_runtime_set_memalloc_noio was added when
del_gendisk is called in this path.  Note that the new pm routine
attempts to cycle through all parent devices from the FC target device
to set the memalloc_noio flag.  Because of this new change, a dependency
was created between the FC target device and the parent netdev device
in the destroy path.

In order to synchronized the destroy paths, bnx2fc has been modified
to flush all destroy workqueues in the NETDEV_UNREGISTER return path.

[    4.123584] BUG: soft lockup - CPU#8 stuck for 22s! [kworker/8:3:8082]
[    4.123713] Call Trace:
[    4.123719]  [<ffffffff815dfbe0>] klist_next+0x20/0xf0
[    4.123725]  [<ffffffff813e9220>] ? pm_save_wakeup_count+0x70/0x70
[    4.123731]  [<ffffffff813d9e4e>] device_for_each_child+0x4e/0x70
[    4.123735]  [<ffffffff813e9554>] pm_runtime_set_memalloc_noio+0x94/0xf0
[    4.123740]  [<ffffffff812d4d74>] del_gendisk+0x264/0x2a0
[    4.123747]  [<ffffffffa00c6dc9>] sd_remove+0x69/0xb0 [sd_mod]
[    4.123751]  [<ffffffff813de24f>] __device_release_driver+0x7f/0xf0
[    4.123754]  [<ffffffff813de2e3>] device_release_driver+0x23/0x30
[    4.123757]  [<ffffffff813ddab4>] bus_remove_device+0xf4/0x170
[    4.123760]  [<ffffffff813da475>] device_del+0x135/0x1d0
[    4.123765]  [<ffffffff81411b75>] __scsi_remove_device+0xc5/0xd0
[    4.123768]  [<ffffffff81411ba6>] scsi_remove_device+0x26/0x40
[    4.123770]  [<ffffffff81411d40>] scsi_remove_target+0x160/0x210
[    4.123775]  [<ffffffffa0420e4c>] fc_rport_final_delete+0xac/0x1f0 [scsi_transport_fc]
[    4.123780]  [<ffffffff810774ab>] process_one_work+0x17b/0x460
[    4.123783]  [<ffffffff8107825b>] worker_thread+0x11b/0x400
[    4.123786]  [<ffffffff81078140>] ? rescuer_thread+0x3e0/0x3e0
[    4.123791]  [<ffffffff8107e9c0>] kthread+0xc0/0xd0
[    4.123794]  [<ffffffff8107e900>] ? kthread_create_on_node+0x110/0x110
[    4.123798]  [<ffffffff8160ceec>] ret_from_fork+0x7c/0xb0
[    4.123801]  [<ffffffff8107e900>] ? kthread_create_on_node+0x110/0x110

Signed-off-by: Eddie Wai <eddie.wai@broadcom.com>
Signed-off-by: James Bottomley <JBottomley@Parallels.com>
drivers/scsi/bnx2fc/bnx2fc_fcoe.c

index 9b948505d118be31b78643d18099eb597bf14864..4cabd3ec67ecde3dba2f7b7bcd0c7afe1c0e711a 100644 (file)
@@ -850,6 +850,9 @@ static void bnx2fc_indicate_netevent(void *context, unsigned long event,
                                __bnx2fc_destroy(interface);
                }
                mutex_unlock(&bnx2fc_dev_lock);
+
+               /* Ensure ALL destroy work has been completed before return */
+               flush_workqueue(bnx2fc_wq);
                return;
 
        default:
@@ -2389,6 +2392,9 @@ static void bnx2fc_ulp_exit(struct cnic_dev *dev)
                        __bnx2fc_destroy(interface);
        mutex_unlock(&bnx2fc_dev_lock);
 
+       /* Ensure ALL destroy work has been completed before return */
+       flush_workqueue(bnx2fc_wq);
+
        bnx2fc_ulp_stop(hba);
        /* unregister cnic device */
        if (test_and_clear_bit(BNX2FC_CNIC_REGISTERED, &hba->reg_with_cnic))