net/mlx5: Lag, properly lock eswitch if needed

Currently when doing hardware lag we check the eswitch mode
but as this isn't done under a lock the check isn't valid.

As the code needs to sync between two different devices an extra
care is needed.

- When going to change eswitch mode, if hardware lag is active destroy it.
- While changing eswitch modes block any hardware bond creation.
- Delay handling bonding events until there are no mode changes in
  progress.
- When attaching a new mdev to lag, block until there is no mode change
  in progress. In order for the mode change to finish the interface lock
  will have to be taken. Release the lock and sleep for 100ms to
  allow forward progress. As this is a very rare condition (can happen if
  the user unbinds and binds a PCI function while also changing eswitch
  mode of the other PCI function) it has no real world impact.

As taking multiple eswitch mode locks is now required lockdep will
complain about a possible deadlock. Register a key per eswitch to make
lockdep happy.

Signed-off-by: Mark Bloch <mbloch@nvidia.com>
Reviewed-by: Mark Zhang <markzhang@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>

1 files changed, 72 insertions, 11 deletions

diff --git a/drivers/net/ethernet/mellanox/mlx5/core/lag.c b/drivers/net/ethernet/mellanox/mlx5/core/lag.c
index 3049de648256..459e3e5ef13f 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/lag.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/lag.c
@@ -418,21 +418,48 @@ static void mlx5_queue_bond_work(struct mlx5_lag *ldev, unsigned long delay)
 	queue_delayed_work(ldev->wq, &ldev->bond_work, delay);
 }
 
+static void mlx5_lag_lock_eswitches(struct mlx5_core_dev *dev0,
+				    struct mlx5_core_dev *dev1)
+{
+	if (dev0)
+		mlx5_esw_lock(dev0->priv.eswitch);
+	if (dev1)
+		mlx5_esw_lock(dev1->priv.eswitch);
+}
+
+static void mlx5_lag_unlock_eswitches(struct mlx5_core_dev *dev0,
+				      struct mlx5_core_dev *dev1)
+{
+	if (dev1)
+		mlx5_esw_unlock(dev1->priv.eswitch);
+	if (dev0)
+		mlx5_esw_unlock(dev0->priv.eswitch);
+}
+
 static void mlx5_do_bond_work(struct work_struct *work)
 {
 	struct delayed_work *delayed_work = to_delayed_work(work);
 	struct mlx5_lag *ldev = container_of(delayed_work, struct mlx5_lag,
 					     bond_work);
+	struct mlx5_core_dev *dev0 = ldev->pf[MLX5_LAG_P1].dev;
+	struct mlx5_core_dev *dev1 = ldev->pf[MLX5_LAG_P2].dev;
 	int status;
 
 	status = mlx5_dev_list_trylock();
 	if (!status) {
-		/* 1 sec delay. */
 		mlx5_queue_bond_work(ldev, HZ);
 		return;
 	}
 
+	if (ldev->mode_changes_in_progress) {
+		mlx5_dev_list_unlock();
+		mlx5_queue_bond_work(ldev, HZ);
+		return;
+	}
+
+	mlx5_lag_lock_eswitches(dev0, dev1);
 	mlx5_do_bond(ldev);
+	mlx5_lag_unlock_eswitches(dev0, dev1);
 	mlx5_dev_list_unlock();
 }
 
@@ -630,7 +657,7 @@ static void mlx5_ldev_remove_mdev(struct mlx5_lag *ldev,
 }
 
 /* Must be called with intf_mutex held */
-static void __mlx5_lag_dev_add_mdev(struct mlx5_core_dev *dev)
+static int __mlx5_lag_dev_add_mdev(struct mlx5_core_dev *dev)
 {
 	struct mlx5_lag *ldev = NULL;
 	struct mlx5_core_dev *tmp_dev;
@@ -638,7 +665,7 @@ static void __mlx5_lag_dev_add_mdev(struct mlx5_core_dev *dev)
 	if (!MLX5_CAP_GEN(dev, vport_group_manager) ||
 	    !MLX5_CAP_GEN(dev, lag_master) ||
 	    MLX5_CAP_GEN(dev, num_lag_ports) != MLX5_MAX_PORTS)
-		return;
+		return 0;
 
 	tmp_dev = mlx5_get_next_phys_dev(dev);
 	if (tmp_dev)
@@ -648,15 +675,17 @@ static void __mlx5_lag_dev_add_mdev(struct mlx5_core_dev *dev)
 		ldev = mlx5_lag_dev_alloc(dev);
 		if (!ldev) {
 			mlx5_core_err(dev, "Failed to alloc lag dev\n");
-			return;
+			return 0;
 		}
 	} else {
+		if (ldev->mode_changes_in_progress)
+			return -EAGAIN;
 		mlx5_ldev_get(ldev);
 	}
 
 	mlx5_ldev_add_mdev(ldev, dev);
 
-	return;
+	return 0;
 }
 
 void mlx5_lag_remove_mdev(struct mlx5_core_dev *dev)
@@ -667,7 +696,13 @@ void mlx5_lag_remove_mdev(struct mlx5_core_dev *dev)
 	if (!ldev)
 		return;
 
+recheck:
 	mlx5_dev_list_lock();
+	if (ldev->mode_changes_in_progress) {
+		mlx5_dev_list_unlock();
+		msleep(100);
+		goto recheck;
+	}
 	mlx5_ldev_remove_mdev(ldev, dev);
 	mlx5_dev_list_unlock();
 	mlx5_ldev_put(ldev);
@@ -675,8 +710,16 @@ void mlx5_lag_remove_mdev(struct mlx5_core_dev *dev)
 
 void mlx5_lag_add_mdev(struct mlx5_core_dev *dev)
 {
+	int err;
+
+recheck:
 	mlx5_dev_list_lock();
-	__mlx5_lag_dev_add_mdev(dev);
+	err = __mlx5_lag_dev_add_mdev(dev);
+	if (err) {
+		mlx5_dev_list_unlock();
+		msleep(100);
+		goto recheck;
+	}
 	mlx5_dev_list_unlock();
 }
 
@@ -716,6 +759,7 @@ void mlx5_lag_add_netdev(struct mlx5_core_dev *dev,
 
 	if (i >= MLX5_MAX_PORTS)
 		ldev->flags |= MLX5_LAG_FLAG_READY;
+	mlx5_queue_bond_work(ldev, 0);
 }
 
 bool mlx5_lag_is_roce(struct mlx5_core_dev *dev)
@@ -789,19 +833,36 @@ bool mlx5_lag_is_shared_fdb(struct mlx5_core_dev *dev)
 }
 EXPORT_SYMBOL(mlx5_lag_is_shared_fdb);
 
-void mlx5_lag_update(struct mlx5_core_dev *dev)
+void mlx5_lag_disable_change(struct mlx5_core_dev *dev)
 {
+	struct mlx5_core_dev *dev0;
+	struct mlx5_core_dev *dev1;
 	struct mlx5_lag *ldev;
 
 	mlx5_dev_list_lock();
+
 	ldev = mlx5_lag_dev(dev);
-	if (!ldev)
-		goto unlock;
+	dev0 = ldev->pf[MLX5_LAG_P1].dev;
+	dev1 = ldev->pf[MLX5_LAG_P2].dev;
 
-	mlx5_do_bond(ldev);
+	ldev->mode_changes_in_progress++;
+	if (__mlx5_lag_is_active(ldev)) {
+		mlx5_lag_lock_eswitches(dev0, dev1);
+		mlx5_disable_lag(ldev);
+		mlx5_lag_unlock_eswitches(dev0, dev1);
+	}
+	mlx5_dev_list_unlock();
+}
 
-unlock:
+void mlx5_lag_enable_change(struct mlx5_core_dev *dev)
+{
+	struct mlx5_lag *ldev;
+
+	mlx5_dev_list_lock();
+	ldev = mlx5_lag_dev(dev);
+	ldev->mode_changes_in_progress--;
 	mlx5_dev_list_unlock();
+	mlx5_queue_bond_work(ldev, 0);
 }
 
 struct net_device *mlx5_lag_get_roce_netdev(struct mlx5_core_dev *dev)