net/mlx5: Move SF dev table notifier registration outside the PF devlink lock

This completes the previous patches by moving notifier registration for
SF dev tables outside the devlink locked critical section in
mlx5_init_one() / mlx5_uninit_one() and into the mlx5_mdev_init() /
mlx5_mdev_uninit() functions.

This is only done for non-SFs, since SFs do not have a SF HW table
themselves.

After this patch, notifiers can grab the PF devlink lock (soon to be
necessary) without creating a locking cycle.

Signed-off-by: Cosmin Ratiu <[e-mail]>
Reviewed-by: Carolina Jubran <[e-mail]>
Signed-off-by: Tariq Toukan <[e-mail]>
Link: https://patch.msgid.link/[e-mail]
Signed-off-by: Jakub Kicinski <[e-mail]>

drivers/net/ethernet/mellanox/mlx5/core/main.c

···
 	if (err)
 		goto err_sf_notifiers;
 
+	err = mlx5_sf_dev_notifier_init(dev);
+	if (err)
+		goto err_sf_dev_notifier;
+
 	return 0;
 
+err_sf_dev_notifier:
+	mlx5_sf_notifiers_cleanup(dev);
 err_sf_notifiers:
 	mlx5_sf_hw_notifier_cleanup(dev);
 err_sf_hw_notifier:
···
 
 static void mlx5_notifiers_cleanup(struct mlx5_core_dev *dev)
 {
+	mlx5_sf_dev_notifier_cleanup(dev);
 	mlx5_sf_notifiers_cleanup(dev);
 	mlx5_sf_hw_notifier_cleanup(dev);
 	mlx5_events_cleanup(dev);

drivers/net/ethernet/mellanox/mlx5/core/sf/dev/dev.c

···
 	struct xarray devices;
 	phys_addr_t base_address;
 	u64 sf_bar_length;
-	struct notifier_block nb;
 	struct workqueue_struct *active_wq;
 	struct work_struct work;
 	u8 stop_active_wq:1;
···
 static int
 mlx5_sf_dev_state_change_handler(struct notifier_block *nb, unsigned long event_code, void *data)
 {
-	struct mlx5_sf_dev_table *table = container_of(nb, struct mlx5_sf_dev_table, nb);
+	struct mlx5_core_dev *dev = container_of(nb, struct mlx5_core_dev,
+						 priv.sf_dev_nb);
+	struct mlx5_sf_dev_table *table = dev->priv.sf_dev_table;
 	const struct mlx5_vhca_state_event *event = data;
 	struct mlx5_sf_dev *sf_dev;
 	u16 max_functions;
 	u16 sf_index;
 	u16 base_id;
 
-	max_functions = mlx5_sf_max_functions(table->dev);
+	if (!table)
+		return 0;
+
+	max_functions = mlx5_sf_max_functions(dev);
 	if (!max_functions)
 		return 0;
 
-	base_id = mlx5_sf_start_function_id(table->dev);
+	base_id = mlx5_sf_start_function_id(dev);
 	if (event->function_id < base_id || event->function_id >= (base_id + max_functions))
 		return 0;
 
···
 	case MLX5_VHCA_STATE_INVALID:
 	case MLX5_VHCA_STATE_ALLOCATED:
 		if (sf_dev)
-			mlx5_sf_dev_del(table->dev, sf_dev, sf_index);
+			mlx5_sf_dev_del(dev, sf_dev, sf_index);
 		break;
 	case MLX5_VHCA_STATE_TEARDOWN_REQUEST:
 		if (sf_dev)
-			mlx5_sf_dev_del(table->dev, sf_dev, sf_index);
+			mlx5_sf_dev_del(dev, sf_dev, sf_index);
 		else
-			mlx5_core_err(table->dev,
+			mlx5_core_err(dev,
 				      "SF DEV: teardown state for invalid dev index=%d sfnum=0x%x\n",
 				      sf_index, event->sw_function_id);
 		break;
 	case MLX5_VHCA_STATE_ACTIVE:
 		if (!sf_dev)
-			mlx5_sf_dev_add(table->dev, sf_index, event->function_id,
+			mlx5_sf_dev_add(dev, sf_index, event->function_id,
 					event->sw_function_id);
 		break;
 	default:
···
 	}
 }
 
+int mlx5_sf_dev_notifier_init(struct mlx5_core_dev *dev)
+{
+	if (mlx5_core_is_sf(dev))
+		return 0;
+
+	dev->priv.sf_dev_nb.notifier_call = mlx5_sf_dev_state_change_handler;
+	return mlx5_vhca_event_notifier_register(dev, &dev->priv.sf_dev_nb);
+}
+
 void mlx5_sf_dev_table_create(struct mlx5_core_dev *dev)
 {
 	struct mlx5_sf_dev_table *table;
···
 		goto table_err;
 	}
 
-	table->nb.notifier_call = mlx5_sf_dev_state_change_handler;
 	table->dev = dev;
 	table->sf_bar_length = 1 << (MLX5_CAP_GEN(dev, log_min_sf_size) + 12);
 	table->base_address = pci_resource_start(dev->pdev, 2);
 	xa_init(&table->devices);
 	dev->priv.sf_dev_table = table;
-
-	err = mlx5_vhca_event_notifier_register(dev, &table->nb);
-	if (err)
-		goto vhca_err;
 
 	err = mlx5_sf_dev_create_active_works(table);
 	if (err)
···
 
 arm_err:
 	mlx5_sf_dev_destroy_active_works(table);
-add_active_err:
-	mlx5_vhca_event_notifier_unregister(dev, &table->nb);
 	mlx5_vhca_event_work_queues_flush(dev);
-vhca_err:
+add_active_err:
 	kfree(table);
 	dev->priv.sf_dev_table = NULL;
 table_err:
···
 	}
 }
 
+void mlx5_sf_dev_notifier_cleanup(struct mlx5_core_dev *dev)
+{
+	if (mlx5_core_is_sf(dev))
+		return;
+
+	mlx5_vhca_event_notifier_unregister(dev, &dev->priv.sf_dev_nb);
+}
+
 void mlx5_sf_dev_table_destroy(struct mlx5_core_dev *dev)
 {
 	struct mlx5_sf_dev_table *table = dev->priv.sf_dev_table;
···
 		return;
 
 	mlx5_sf_dev_destroy_active_works(table);
-	mlx5_vhca_event_notifier_unregister(dev, &table->nb);
 
 	/* Now that event handler is not running, it is safe to destroy
 	 * the sf device without race.

drivers/net/ethernet/mellanox/mlx5/core/sf/dev/dev.h

···
 	int err;
 };
 
+int mlx5_sf_dev_notifier_init(struct mlx5_core_dev *dev);
 void mlx5_sf_dev_table_create(struct mlx5_core_dev *dev);
+void mlx5_sf_dev_notifier_cleanup(struct mlx5_core_dev *dev);
 void mlx5_sf_dev_table_destroy(struct mlx5_core_dev *dev);
 
 int mlx5_sf_driver_register(void);
···
 
 #else
 
+static inline int mlx5_sf_dev_notifier_init(struct mlx5_core_dev *dev)
+{
+	return 0;
+}
+
 static inline void mlx5_sf_dev_table_create(struct mlx5_core_dev *dev)
+{
+}
+
+static inline void mlx5_sf_dev_notifier_cleanup(struct mlx5_core_dev *dev)
 {
 }
 

include/linux/mlx5/driver.h

···
 #ifdef CONFIG_MLX5_SF
 	struct mlx5_nb vhca_state_nb;
 	struct blocking_notifier_head vhca_state_n_head;
+	struct notifier_block sf_dev_nb;
 	struct mlx5_sf_dev_table *sf_dev_table;
 	struct mlx5_core_dev *parent_mdev;
 #endif