RDMA/cxgb4: Fix locking issue in process_mpa_request

Fix the following lockdep report:

    =============================================
    [ INFO: possible recursive locking detected ]
    3.17.0+ #3 Tainted: G            E
    ---------------------------------------------
    kworker/u64:3/299 is trying to acquire lock:
     (&epc->mutex){+.+.+.}, at: [<ffffffffa074e07a>]
    process_mpa_request+0x1aa/0x3e0 [iw_cxgb4]

    but task is already holding lock:
     (&epc->mutex){+.+.+.}, at: [<ffffffffa074e34e>] rx_data+0x9e/0x1f0 [iw_cxgb4]

    other info that might help us debug this:
     Possible unsafe locking scenario:

           CPU0
           ----
      lock(&epc->mutex);
      lock(&epc->mutex);

     *** DEADLOCK ***

     May be due to missing lock nesting notation

    3 locks held by kworker/u64:3/299:
     #0:  ("%s""iw_cxgb4"){.+.+.+}, at: [<ffffffff8106f14d>]
    process_one_work+0x13d/0x4d0
     #1:  (skb_work){+.+.+.}, at: [<ffffffff8106f14d>] process_one_work+0x13d/0x4d0
     #2:  (&epc->mutex){+.+.+.}, at: [<ffffffffa074e34e>] rx_data+0x9e/0x1f0
    [iw_cxgb4]

    stack backtrace:
    CPU: 2 PID: 299 Comm: kworker/u64:3 Tainted: G            E  3.17.0+ #3
    Hardware name: Dell Inc. PowerEdge T110/0X744K, BIOS 1.2.1 01/28/2010
    Workqueue: iw_cxgb4 process_work [iw_cxgb4]
     ffff8800b91593d0 ffff8800b8a2f9f8 ffffffff815df107 0000000000000001
     ffff8800b9158750 ffff8800b8a2fa28 ffffffff8109f0e2 ffff8800bb768a00
     ffff8800b91593d0 ffff8800b9158750 0000000000000000 ffff8800b8a2fa88
    Call Trace:
     [<ffffffff815df107>] dump_stack+0x49/0x62
     [<ffffffff8109f0e2>] print_deadlock_bug+0xf2/0x100
     [<ffffffff810a0f04>] validate_chain+0x454/0x700
     [<ffffffff810a1574>] __lock_acquire+0x3c4/0x580
     [<ffffffffa074e07a>] ? process_mpa_request+0x1aa/0x3e0 [iw_cxgb4]
     [<ffffffff810a17cc>] lock_acquire+0x9c/0x110
     [<ffffffffa074e07a>] ? process_mpa_request+0x1aa/0x3e0 [iw_cxgb4]
     [<ffffffff815e111b>] mutex_lock_nested+0x4b/0x360
     [<ffffffffa074e07a>] ? process_mpa_request+0x1aa/0x3e0 [iw_cxgb4]
     [<ffffffff810c181a>] ? del_timer_sync+0xaa/0xd0
     [<ffffffff810c1770>] ? try_to_del_timer_sync+0x70/0x70
     [<ffffffffa074e07a>] process_mpa_request+0x1aa/0x3e0 [iw_cxgb4]
     [<ffffffffa074a3ec>] ? update_rx_credits+0xec/0x140 [iw_cxgb4]
     [<ffffffffa074e381>] rx_data+0xd1/0x1f0 [iw_cxgb4]
     [<ffffffff8109ff23>] ? mark_held_locks+0x73/0xa0
     [<ffffffff815e4b90>] ? _raw_spin_unlock_irqrestore+0x40/0x70
     [<ffffffff810a020d>] ? trace_hardirqs_on_caller+0xfd/0x1c0
     [<ffffffff810a02dd>] ? trace_hardirqs_on+0xd/0x10
     [<ffffffffa074c931>] process_work+0x51/0x80 [iw_cxgb4]
     [<ffffffff8106f1c8>] process_one_work+0x1b8/0x4d0
     [<ffffffff8106f14d>] ? process_one_work+0x13d/0x4d0
     [<ffffffff8106f600>] worker_thread+0x120/0x3c0
     [<ffffffff8106f4e0>] ? process_one_work+0x4d0/0x4d0
     [<ffffffff81074a0e>] kthread+0xde/0x100
     [<ffffffff815e4b40>] ? _raw_spin_unlock_irq+0x30/0x40
     [<ffffffff81074930>] ? __init_kthread_worker+0x70/0x70
     [<ffffffff815e512c>] ret_from_fork+0x7c/0xb0
     [<ffffffff81074930>] ? __init_kthread_worker+0x70/0x70

Based on original work by Steve Wise <swise@opengridcomputing.com>.

Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Hariprasad Shenai <hariprasad@chelsio.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>

diff --git a/drivers/infiniband/hw/cxgb4/cm.c b/drivers/infiniband/hw/cxgb4/cm.c
index 4b8c611..0075621 100644 (file)
--- a/drivers/infiniband/hw/cxgb4/cm.c
+++ b/drivers/infiniband/hw/cxgb4/cm.c
@@ -1640,7 +1640,8 @@ static void process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
                __state_set(&ep->com, MPA_REQ_RCVD);
 
                /* drive upcall */
-               mutex_lock(&ep->parent_ep->com.mutex);
+               mutex_lock_nested(&ep->parent_ep->com.mutex,
+                                 SINGLE_DEPTH_NESTING);
                if (ep->parent_ep->com.state != DEAD) {
                        if (connect_request_upcall(ep))
                                abort_connection(ep, skb, GFP_KERNEL);