[bionic:linux-bluefield][PATCH 2/2] RDMA/mlx5: Cleanup rep when doing unload

Wen-chien Jesse Sung jesse.sung at canonical.com
Wed Nov 20 07:39:03 UTC 2019


From: Bodong Wang <bodong at mellanox.com>

BugLink: https://launchpad.net/bugs/1853245

When an IB rep is loaded, netdev for the same vport is saved for later
reference. However, it's not cleaned up when doing unload. For ECPF,
kernel crashes when driver is referring to the already removed netdev.

Following steps lead to a shown call trace:
1. Create n VFs from host PF
2. Distroy the VFs
3. Run "rdma link" from ARM

Call trace:
  mlx5_ib_get_netdev+0x9c/0xe8 [mlx5_ib]
  mlx5_query_port_roce+0x268/0x558 [mlx5_ib]
  mlx5_ib_rep_query_port+0x14/0x34 [mlx5_ib]
  ib_query_port+0x9c/0xfc [ib_core]
  fill_port_info+0x74/0x28c [ib_core]
  nldev_port_get_doit+0x1a8/0x1e8 [ib_core]
  rdma_nl_rcv_msg+0x16c/0x1c0 [ib_core]
  rdma_nl_rcv+0xe8/0x144 [ib_core]
  netlink_unicast+0x184/0x214
  netlink_sendmsg+0x288/0x354
  sock_sendmsg+0x18/0x2c
  __sys_sendto+0xbc/0x138
  __arm64_sys_sendto+0x28/0x34
  el0_svc_common+0xb0/0x100
  el0_svc_handler+0x6c/0x84
  el0_svc+0x8/0xc

Cleanup the rep and netdev reference when unloading IB rep.

Fixes: 26628e2d58c9 ("RDMA/mlx5: Move to single device multiport ports in switchdev mode")
Signed-off-by: Bodong Wang <bodong at mellanox.com>
Reviewed-by: Mark Bloch <markb at mellanox.com>
Reviewed-by: Parav Pandit <parav at mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm at mellanox.com>
(cherry picked from commit b8ca123860ee556a8d42ab8c5c2afa469817a813)
---
 drivers/infiniband/hw/mlx5/ib_rep.c | 18 +++++++++++-------
 1 file changed, 11 insertions(+), 7 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/ib_rep.c b/drivers/infiniband/hw/mlx5/ib_rep.c
index 0381031ad3ec..b200dfae4d1f 100644
--- a/drivers/infiniband/hw/mlx5/ib_rep.c
+++ b/drivers/infiniband/hw/mlx5/ib_rep.c
@@ -17,6 +17,7 @@ mlx5_ib_set_vport_rep(struct mlx5_core_dev *dev, struct mlx5_eswitch_rep *rep)
 	vport_index = rep->vport_index;
 
 	ibdev->port[vport_index].rep = rep;
+	rep->rep_data[REP_IB].priv = ibdev;
 	write_lock(&ibdev->port[vport_index].roce.netdev_lock);
 	ibdev->port[vport_index].roce.netdev =
 		mlx5_ib_get_rep_netdev(dev->priv.eswitch, rep->vport);
@@ -68,15 +69,18 @@ mlx5_ib_vport_rep_load(struct mlx5_core_dev *dev, struct mlx5_eswitch_rep *rep)
 static void
 mlx5_ib_vport_rep_unload(struct mlx5_eswitch_rep *rep)
 {
-	struct mlx5_ib_dev *dev;
-
-	if (!rep->rep_data[REP_IB].priv ||
-	    rep->vport != MLX5_VPORT_UPLINK)
-		return;
+	struct mlx5_ib_dev *dev = mlx5_ib_rep_to_dev(rep);
+	struct mlx5_ib_port *port;
 
-	dev = mlx5_ib_rep_to_dev(rep);
-	__mlx5_ib_remove(dev, dev->profile, MLX5_IB_STAGE_MAX);
+	port = &dev->port[rep->vport_index];
+	write_lock(&port->roce.netdev_lock);
+	port->roce.netdev = NULL;
+	write_unlock(&port->roce.netdev_lock);
 	rep->rep_data[REP_IB].priv = NULL;
+	port->rep = NULL;
+
+	if (rep->vport == MLX5_VPORT_UPLINK)
+		__mlx5_ib_remove(dev, dev->profile, MLX5_IB_STAGE_MAX);
 }
 
 static void *mlx5_ib_vport_get_proto_dev(struct mlx5_eswitch_rep *rep)
-- 
2.20.1




More information about the kernel-team mailing list