2 * Copyright (c) 2016, Mellanox Technologies. All rights reserved.
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
33 #include <linux/netdevice.h>
34 #include <linux/mlx5/driver.h>
35 #include <linux/mlx5/vport.h>
36 #include "mlx5_core.h"
39 MLX5_LAG_FLAG_BONDED = 1 << 0,
43 struct mlx5_core_dev *dev;
44 struct net_device *netdev;
47 /* Used for collection of netdev event info. */
49 enum netdev_lag_tx_type tx_type;
50 struct netdev_lag_lower_state_info netdev_state[MLX5_MAX_PORTS];
54 /* LAG data of a ConnectX card.
55 * It serves both its phys functions.
59 u8 v2p_map[MLX5_MAX_PORTS];
60 struct lag_func pf[MLX5_MAX_PORTS];
61 struct lag_tracker tracker;
62 struct delayed_work bond_work;
63 struct notifier_block nb;
66 /* General purpose, use for short periods of time.
67 * Beware of lock dependencies (preferably, no locks should be acquired
70 static DEFINE_MUTEX(lag_mutex);
72 static int mlx5_cmd_create_lag(struct mlx5_core_dev *dev, u8 remap_port1,
75 u32 in[MLX5_ST_SZ_DW(create_lag_in)] = {0};
76 u32 out[MLX5_ST_SZ_DW(create_lag_out)] = {0};
77 void *lag_ctx = MLX5_ADDR_OF(create_lag_in, in, ctx);
79 MLX5_SET(create_lag_in, in, opcode, MLX5_CMD_OP_CREATE_LAG);
81 MLX5_SET(lagc, lag_ctx, tx_remap_affinity_1, remap_port1);
82 MLX5_SET(lagc, lag_ctx, tx_remap_affinity_2, remap_port2);
84 return mlx5_cmd_exec(dev, in, sizeof(in), out, sizeof(out));
87 static int mlx5_cmd_modify_lag(struct mlx5_core_dev *dev, u8 remap_port1,
90 u32 in[MLX5_ST_SZ_DW(modify_lag_in)] = {0};
91 u32 out[MLX5_ST_SZ_DW(modify_lag_out)] = {0};
92 void *lag_ctx = MLX5_ADDR_OF(modify_lag_in, in, ctx);
94 MLX5_SET(modify_lag_in, in, opcode, MLX5_CMD_OP_MODIFY_LAG);
95 MLX5_SET(modify_lag_in, in, field_select, 0x1);
97 MLX5_SET(lagc, lag_ctx, tx_remap_affinity_1, remap_port1);
98 MLX5_SET(lagc, lag_ctx, tx_remap_affinity_2, remap_port2);
100 return mlx5_cmd_exec(dev, in, sizeof(in), out, sizeof(out));
103 static int mlx5_cmd_destroy_lag(struct mlx5_core_dev *dev)
105 u32 in[MLX5_ST_SZ_DW(destroy_lag_in)] = {0};
106 u32 out[MLX5_ST_SZ_DW(destroy_lag_out)] = {0};
108 MLX5_SET(destroy_lag_in, in, opcode, MLX5_CMD_OP_DESTROY_LAG);
110 return mlx5_cmd_exec(dev, in, sizeof(in), out, sizeof(out));
113 static struct mlx5_lag *mlx5_lag_dev_get(struct mlx5_core_dev *dev)
115 return dev->priv.lag;
118 static int mlx5_lag_dev_get_netdev_idx(struct mlx5_lag *ldev,
119 struct net_device *ndev)
123 for (i = 0; i < MLX5_MAX_PORTS; i++)
124 if (ldev->pf[i].netdev == ndev)
130 static bool mlx5_lag_is_bonded(struct mlx5_lag *ldev)
132 return !!(ldev->flags & MLX5_LAG_FLAG_BONDED);
135 static void mlx5_infer_tx_affinity_mapping(struct lag_tracker *tracker,
136 u8 *port1, u8 *port2)
138 if (tracker->tx_type == NETDEV_LAG_TX_TYPE_ACTIVEBACKUP) {
139 if (tracker->netdev_state[0].tx_enabled) {
149 if (!tracker->netdev_state[0].link_up)
151 else if (!tracker->netdev_state[1].link_up)
156 static void mlx5_activate_lag(struct mlx5_lag *ldev,
157 struct lag_tracker *tracker)
159 struct mlx5_core_dev *dev0 = ldev->pf[0].dev;
162 ldev->flags |= MLX5_LAG_FLAG_BONDED;
164 mlx5_infer_tx_affinity_mapping(tracker, &ldev->v2p_map[0],
167 err = mlx5_cmd_create_lag(dev0, ldev->v2p_map[0], ldev->v2p_map[1]);
170 "Failed to create LAG (%d)\n",
174 static void mlx5_deactivate_lag(struct mlx5_lag *ldev)
176 struct mlx5_core_dev *dev0 = ldev->pf[0].dev;
179 ldev->flags &= ~MLX5_LAG_FLAG_BONDED;
181 err = mlx5_cmd_destroy_lag(dev0);
184 "Failed to destroy LAG (%d)\n",
188 static void mlx5_do_bond(struct mlx5_lag *ldev)
190 struct mlx5_core_dev *dev0 = ldev->pf[0].dev;
191 struct mlx5_core_dev *dev1 = ldev->pf[1].dev;
192 struct lag_tracker tracker;
193 u8 v2p_port1, v2p_port2;
199 mutex_lock(&lag_mutex);
200 tracker = ldev->tracker;
201 mutex_unlock(&lag_mutex);
203 if (tracker.is_bonded && !mlx5_lag_is_bonded(ldev)) {
204 for (i = 0; i < MLX5_MAX_PORTS; i++)
205 mlx5_remove_dev_by_protocol(ldev->pf[i].dev,
206 MLX5_INTERFACE_PROTOCOL_IB);
208 mlx5_activate_lag(ldev, &tracker);
210 mlx5_add_dev_by_protocol(dev0, MLX5_INTERFACE_PROTOCOL_IB);
211 mlx5_nic_vport_enable_roce(dev1);
212 } else if (tracker.is_bonded && mlx5_lag_is_bonded(ldev)) {
213 mlx5_infer_tx_affinity_mapping(&tracker, &v2p_port1,
216 if ((v2p_port1 != ldev->v2p_map[0]) ||
217 (v2p_port2 != ldev->v2p_map[1])) {
218 ldev->v2p_map[0] = v2p_port1;
219 ldev->v2p_map[1] = v2p_port2;
221 err = mlx5_cmd_modify_lag(dev0, v2p_port1, v2p_port2);
224 "Failed to modify LAG (%d)\n",
227 } else if (!tracker.is_bonded && mlx5_lag_is_bonded(ldev)) {
228 mlx5_remove_dev_by_protocol(dev0, MLX5_INTERFACE_PROTOCOL_IB);
229 mlx5_nic_vport_disable_roce(dev1);
231 mlx5_deactivate_lag(ldev);
233 for (i = 0; i < MLX5_MAX_PORTS; i++)
235 mlx5_add_dev_by_protocol(ldev->pf[i].dev,
236 MLX5_INTERFACE_PROTOCOL_IB);
240 static void mlx5_queue_bond_work(struct mlx5_lag *ldev, unsigned long delay)
242 schedule_delayed_work(&ldev->bond_work, delay);
245 static void mlx5_do_bond_work(struct work_struct *work)
247 struct delayed_work *delayed_work = to_delayed_work(work);
248 struct mlx5_lag *ldev = container_of(delayed_work, struct mlx5_lag,
252 status = mutex_trylock(&mlx5_intf_mutex);
255 mlx5_queue_bond_work(ldev, HZ);
260 mutex_unlock(&mlx5_intf_mutex);
263 static int mlx5_handle_changeupper_event(struct mlx5_lag *ldev,
264 struct lag_tracker *tracker,
265 struct net_device *ndev,
266 struct netdev_notifier_changeupper_info *info)
268 struct net_device *upper = info->upper_dev, *ndev_tmp;
269 struct netdev_lag_upper_info *lag_upper_info;
275 if (!netif_is_lag_master(upper))
278 lag_upper_info = info->upper_info;
280 /* The event may still be of interest if the slave does not belong to
281 * us, but is enslaved to a master which has one or more of our netdevs
282 * as slaves (e.g., if a new slave is added to a master that bonds two
283 * of our netdevs, we should unbond).
286 for_each_netdev_in_bond_rcu(upper, ndev_tmp) {
287 idx = mlx5_lag_dev_get_netdev_idx(ldev, ndev_tmp);
289 bond_status |= (1 << idx);
295 /* None of this lagdev's netdevs are slaves of this master. */
296 if (!(bond_status & 0x3))
300 tracker->tx_type = lag_upper_info->tx_type;
302 /* Determine bonding status:
303 * A device is considered bonded if both its physical ports are slaves
304 * of the same lag master, and only them.
305 * Lag mode must be activebackup or hash.
307 is_bonded = (num_slaves == MLX5_MAX_PORTS) &&
308 (bond_status == 0x3) &&
309 ((tracker->tx_type == NETDEV_LAG_TX_TYPE_ACTIVEBACKUP) ||
310 (tracker->tx_type == NETDEV_LAG_TX_TYPE_HASH));
312 if (tracker->is_bonded != is_bonded) {
313 tracker->is_bonded = is_bonded;
320 static int mlx5_handle_changelowerstate_event(struct mlx5_lag *ldev,
321 struct lag_tracker *tracker,
322 struct net_device *ndev,
323 struct netdev_notifier_changelowerstate_info *info)
325 struct netdev_lag_lower_state_info *lag_lower_info;
328 if (!netif_is_lag_port(ndev))
331 idx = mlx5_lag_dev_get_netdev_idx(ldev, ndev);
335 /* This information is used to determine virtual to physical
338 lag_lower_info = info->lower_state_info;
342 tracker->netdev_state[idx] = *lag_lower_info;
347 static int mlx5_lag_netdev_event(struct notifier_block *this,
348 unsigned long event, void *ptr)
350 struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
351 struct lag_tracker tracker;
352 struct mlx5_lag *ldev;
355 if (!net_eq(dev_net(ndev), &init_net))
358 if ((event != NETDEV_CHANGEUPPER) && (event != NETDEV_CHANGELOWERSTATE))
361 ldev = container_of(this, struct mlx5_lag, nb);
362 tracker = ldev->tracker;
365 case NETDEV_CHANGEUPPER:
366 changed = mlx5_handle_changeupper_event(ldev, &tracker, ndev,
369 case NETDEV_CHANGELOWERSTATE:
370 changed = mlx5_handle_changelowerstate_event(ldev, &tracker,
375 mutex_lock(&lag_mutex);
376 ldev->tracker = tracker;
377 mutex_unlock(&lag_mutex);
380 mlx5_queue_bond_work(ldev, 0);
385 static struct mlx5_lag *mlx5_lag_dev_alloc(void)
387 struct mlx5_lag *ldev;
389 ldev = kzalloc(sizeof(*ldev), GFP_KERNEL);
393 INIT_DELAYED_WORK(&ldev->bond_work, mlx5_do_bond_work);
398 static void mlx5_lag_dev_free(struct mlx5_lag *ldev)
403 static void mlx5_lag_dev_add_pf(struct mlx5_lag *ldev,
404 struct mlx5_core_dev *dev,
405 struct net_device *netdev)
407 unsigned int fn = PCI_FUNC(dev->pdev->devfn);
409 if (fn >= MLX5_MAX_PORTS)
412 mutex_lock(&lag_mutex);
413 ldev->pf[fn].dev = dev;
414 ldev->pf[fn].netdev = netdev;
415 ldev->tracker.netdev_state[fn].link_up = 0;
416 ldev->tracker.netdev_state[fn].tx_enabled = 0;
418 dev->priv.lag = ldev;
419 mutex_unlock(&lag_mutex);
422 static void mlx5_lag_dev_remove_pf(struct mlx5_lag *ldev,
423 struct mlx5_core_dev *dev)
427 for (i = 0; i < MLX5_MAX_PORTS; i++)
428 if (ldev->pf[i].dev == dev)
431 if (i == MLX5_MAX_PORTS)
434 mutex_lock(&lag_mutex);
435 memset(&ldev->pf[i], 0, sizeof(*ldev->pf));
437 dev->priv.lag = NULL;
438 mutex_unlock(&lag_mutex);
441 static u16 mlx5_gen_pci_id(struct mlx5_core_dev *dev)
443 return (u16)((dev->pdev->bus->number << 8) |
444 PCI_SLOT(dev->pdev->devfn));
447 /* Must be called with intf_mutex held */
448 void mlx5_lag_add(struct mlx5_core_dev *dev, struct net_device *netdev)
450 struct mlx5_lag *ldev = NULL;
451 struct mlx5_core_dev *tmp_dev;
452 struct mlx5_priv *priv;
455 if (!MLX5_CAP_GEN(dev, vport_group_manager) ||
456 !MLX5_CAP_GEN(dev, lag_master) ||
457 (MLX5_CAP_GEN(dev, num_lag_ports) != MLX5_MAX_PORTS))
460 pci_id = mlx5_gen_pci_id(dev);
462 mlx5_core_for_each_priv(priv) {
463 tmp_dev = container_of(priv, struct mlx5_core_dev, priv);
464 if ((dev != tmp_dev) &&
465 (mlx5_gen_pci_id(tmp_dev) == pci_id)) {
466 ldev = tmp_dev->priv.lag;
472 ldev = mlx5_lag_dev_alloc();
474 mlx5_core_err(dev, "Failed to alloc lag dev\n");
479 mlx5_lag_dev_add_pf(ldev, dev, netdev);
481 if (!ldev->nb.notifier_call) {
482 ldev->nb.notifier_call = mlx5_lag_netdev_event;
483 if (register_netdevice_notifier(&ldev->nb)) {
484 ldev->nb.notifier_call = NULL;
485 mlx5_core_err(dev, "Failed to register LAG netdev notifier\n");
490 /* Must be called with intf_mutex held */
491 void mlx5_lag_remove(struct mlx5_core_dev *dev)
493 struct mlx5_lag *ldev;
496 ldev = mlx5_lag_dev_get(dev);
500 if (mlx5_lag_is_bonded(ldev))
501 mlx5_deactivate_lag(ldev);
503 mlx5_lag_dev_remove_pf(ldev, dev);
505 for (i = 0; i < MLX5_MAX_PORTS; i++)
509 if (i == MLX5_MAX_PORTS) {
510 if (ldev->nb.notifier_call)
511 unregister_netdevice_notifier(&ldev->nb);
512 cancel_delayed_work_sync(&ldev->bond_work);
513 mlx5_lag_dev_free(ldev);
517 bool mlx5_lag_is_active(struct mlx5_core_dev *dev)
519 struct mlx5_lag *ldev;
522 mutex_lock(&lag_mutex);
523 ldev = mlx5_lag_dev_get(dev);
524 res = ldev && mlx5_lag_is_bonded(ldev);
525 mutex_unlock(&lag_mutex);
529 EXPORT_SYMBOL(mlx5_lag_is_active);