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"
41 /* General purpose, use for short periods of time.
42 * Beware of lock dependencies (preferably, no locks should be acquired
45 static DEFINE_MUTEX(lag_mutex);
47 static int mlx5_cmd_create_lag(struct mlx5_core_dev *dev, u8 remap_port1,
50 u32 in[MLX5_ST_SZ_DW(create_lag_in)] = {0};
51 u32 out[MLX5_ST_SZ_DW(create_lag_out)] = {0};
52 void *lag_ctx = MLX5_ADDR_OF(create_lag_in, in, ctx);
54 MLX5_SET(create_lag_in, in, opcode, MLX5_CMD_OP_CREATE_LAG);
56 MLX5_SET(lagc, lag_ctx, tx_remap_affinity_1, remap_port1);
57 MLX5_SET(lagc, lag_ctx, tx_remap_affinity_2, remap_port2);
59 return mlx5_cmd_exec(dev, in, sizeof(in), out, sizeof(out));
62 static int mlx5_cmd_modify_lag(struct mlx5_core_dev *dev, u8 remap_port1,
65 u32 in[MLX5_ST_SZ_DW(modify_lag_in)] = {0};
66 u32 out[MLX5_ST_SZ_DW(modify_lag_out)] = {0};
67 void *lag_ctx = MLX5_ADDR_OF(modify_lag_in, in, ctx);
69 MLX5_SET(modify_lag_in, in, opcode, MLX5_CMD_OP_MODIFY_LAG);
70 MLX5_SET(modify_lag_in, in, field_select, 0x1);
72 MLX5_SET(lagc, lag_ctx, tx_remap_affinity_1, remap_port1);
73 MLX5_SET(lagc, lag_ctx, tx_remap_affinity_2, remap_port2);
75 return mlx5_cmd_exec(dev, in, sizeof(in), out, sizeof(out));
78 static int mlx5_cmd_destroy_lag(struct mlx5_core_dev *dev)
80 u32 in[MLX5_ST_SZ_DW(destroy_lag_in)] = {0};
81 u32 out[MLX5_ST_SZ_DW(destroy_lag_out)] = {0};
83 MLX5_SET(destroy_lag_in, in, opcode, MLX5_CMD_OP_DESTROY_LAG);
85 return mlx5_cmd_exec(dev, in, sizeof(in), out, sizeof(out));
88 int mlx5_cmd_create_vport_lag(struct mlx5_core_dev *dev)
90 u32 in[MLX5_ST_SZ_DW(create_vport_lag_in)] = {0};
91 u32 out[MLX5_ST_SZ_DW(create_vport_lag_out)] = {0};
93 MLX5_SET(create_vport_lag_in, in, opcode, MLX5_CMD_OP_CREATE_VPORT_LAG);
95 return mlx5_cmd_exec(dev, in, sizeof(in), out, sizeof(out));
97 EXPORT_SYMBOL(mlx5_cmd_create_vport_lag);
99 int mlx5_cmd_destroy_vport_lag(struct mlx5_core_dev *dev)
101 u32 in[MLX5_ST_SZ_DW(destroy_vport_lag_in)] = {0};
102 u32 out[MLX5_ST_SZ_DW(destroy_vport_lag_out)] = {0};
104 MLX5_SET(destroy_vport_lag_in, in, opcode, MLX5_CMD_OP_DESTROY_VPORT_LAG);
106 return mlx5_cmd_exec(dev, in, sizeof(in), out, sizeof(out));
108 EXPORT_SYMBOL(mlx5_cmd_destroy_vport_lag);
110 static int mlx5_cmd_query_cong_counter(struct mlx5_core_dev *dev,
111 bool reset, void *out, int out_size)
113 u32 in[MLX5_ST_SZ_DW(query_cong_statistics_in)] = { };
115 MLX5_SET(query_cong_statistics_in, in, opcode,
116 MLX5_CMD_OP_QUERY_CONG_STATISTICS);
117 MLX5_SET(query_cong_statistics_in, in, clear, reset);
118 return mlx5_cmd_exec(dev, in, sizeof(in), out, out_size);
121 int mlx5_lag_dev_get_netdev_idx(struct mlx5_lag *ldev,
122 struct net_device *ndev)
126 for (i = 0; i < MLX5_MAX_PORTS; i++)
127 if (ldev->pf[i].netdev == ndev)
133 static bool __mlx5_lag_is_roce(struct mlx5_lag *ldev)
135 return !!(ldev->flags & MLX5_LAG_FLAG_ROCE);
138 static bool __mlx5_lag_is_sriov(struct mlx5_lag *ldev)
140 return !!(ldev->flags & MLX5_LAG_FLAG_SRIOV);
143 static void mlx5_infer_tx_affinity_mapping(struct lag_tracker *tracker,
144 u8 *port1, u8 *port2)
148 if (!tracker->netdev_state[MLX5_LAG_P1].tx_enabled ||
149 !tracker->netdev_state[MLX5_LAG_P1].link_up) {
154 if (!tracker->netdev_state[MLX5_LAG_P2].tx_enabled ||
155 !tracker->netdev_state[MLX5_LAG_P2].link_up)
159 void mlx5_modify_lag(struct mlx5_lag *ldev,
160 struct lag_tracker *tracker)
162 struct mlx5_core_dev *dev0 = ldev->pf[MLX5_LAG_P1].dev;
163 u8 v2p_port1, v2p_port2;
166 mlx5_infer_tx_affinity_mapping(tracker, &v2p_port1,
169 if (v2p_port1 != ldev->v2p_map[MLX5_LAG_P1] ||
170 v2p_port2 != ldev->v2p_map[MLX5_LAG_P2]) {
171 ldev->v2p_map[MLX5_LAG_P1] = v2p_port1;
172 ldev->v2p_map[MLX5_LAG_P2] = v2p_port2;
174 mlx5_core_info(dev0, "modify lag map port 1:%d port 2:%d",
175 ldev->v2p_map[MLX5_LAG_P1],
176 ldev->v2p_map[MLX5_LAG_P2]);
178 err = mlx5_cmd_modify_lag(dev0, v2p_port1, v2p_port2);
181 "Failed to modify LAG (%d)\n",
186 static int mlx5_create_lag(struct mlx5_lag *ldev,
187 struct lag_tracker *tracker)
189 struct mlx5_core_dev *dev0 = ldev->pf[MLX5_LAG_P1].dev;
192 mlx5_infer_tx_affinity_mapping(tracker, &ldev->v2p_map[MLX5_LAG_P1],
193 &ldev->v2p_map[MLX5_LAG_P2]);
195 mlx5_core_info(dev0, "lag map port 1:%d port 2:%d",
196 ldev->v2p_map[MLX5_LAG_P1], ldev->v2p_map[MLX5_LAG_P2]);
198 err = mlx5_cmd_create_lag(dev0, ldev->v2p_map[MLX5_LAG_P1],
199 ldev->v2p_map[MLX5_LAG_P2]);
202 "Failed to create LAG (%d)\n",
207 int mlx5_activate_lag(struct mlx5_lag *ldev,
208 struct lag_tracker *tracker,
211 bool roce_lag = !!(flags & MLX5_LAG_FLAG_ROCE);
212 struct mlx5_core_dev *dev0 = ldev->pf[MLX5_LAG_P1].dev;
215 err = mlx5_create_lag(ldev, tracker);
219 "Failed to activate RoCE LAG\n");
222 "Failed to activate VF LAG\n"
223 "Make sure all VFs are unbound prior to VF LAG activation or deactivation\n");
228 ldev->flags |= flags;
232 static int mlx5_deactivate_lag(struct mlx5_lag *ldev)
234 struct mlx5_core_dev *dev0 = ldev->pf[MLX5_LAG_P1].dev;
235 bool roce_lag = __mlx5_lag_is_roce(ldev);
238 ldev->flags &= ~MLX5_LAG_MODE_FLAGS;
240 err = mlx5_cmd_destroy_lag(dev0);
244 "Failed to deactivate RoCE LAG; driver restart required\n");
247 "Failed to deactivate VF LAG; driver restart required\n"
248 "Make sure all VFs are unbound prior to VF LAG activation or deactivation\n");
255 static bool mlx5_lag_check_prereq(struct mlx5_lag *ldev)
257 if (!ldev->pf[MLX5_LAG_P1].dev || !ldev->pf[MLX5_LAG_P2].dev)
260 #ifdef CONFIG_MLX5_ESWITCH
261 return mlx5_esw_lag_prereq(ldev->pf[MLX5_LAG_P1].dev,
262 ldev->pf[MLX5_LAG_P2].dev);
264 return (!mlx5_sriov_is_enabled(ldev->pf[MLX5_LAG_P1].dev) &&
265 !mlx5_sriov_is_enabled(ldev->pf[MLX5_LAG_P2].dev));
269 static void mlx5_lag_add_ib_devices(struct mlx5_lag *ldev)
273 for (i = 0; i < MLX5_MAX_PORTS; i++)
275 mlx5_add_dev_by_protocol(ldev->pf[i].dev,
276 MLX5_INTERFACE_PROTOCOL_IB);
279 static void mlx5_lag_remove_ib_devices(struct mlx5_lag *ldev)
283 for (i = 0; i < MLX5_MAX_PORTS; i++)
285 mlx5_remove_dev_by_protocol(ldev->pf[i].dev,
286 MLX5_INTERFACE_PROTOCOL_IB);
289 static void mlx5_do_bond(struct mlx5_lag *ldev)
291 struct mlx5_core_dev *dev0 = ldev->pf[MLX5_LAG_P1].dev;
292 struct mlx5_core_dev *dev1 = ldev->pf[MLX5_LAG_P2].dev;
293 struct lag_tracker tracker;
294 bool do_bond, roce_lag;
300 mutex_lock(&lag_mutex);
301 tracker = ldev->tracker;
302 mutex_unlock(&lag_mutex);
304 do_bond = tracker.is_bonded && mlx5_lag_check_prereq(ldev);
306 if (do_bond && !__mlx5_lag_is_active(ldev)) {
307 roce_lag = !mlx5_sriov_is_enabled(dev0) &&
308 !mlx5_sriov_is_enabled(dev1);
310 #ifdef CONFIG_MLX5_ESWITCH
311 roce_lag &= dev0->priv.eswitch->mode == MLX5_ESWITCH_NONE &&
312 dev1->priv.eswitch->mode == MLX5_ESWITCH_NONE;
316 mlx5_lag_remove_ib_devices(ldev);
318 err = mlx5_activate_lag(ldev, &tracker,
319 roce_lag ? MLX5_LAG_FLAG_ROCE :
320 MLX5_LAG_FLAG_SRIOV);
323 mlx5_lag_add_ib_devices(ldev);
329 mlx5_add_dev_by_protocol(dev0, MLX5_INTERFACE_PROTOCOL_IB);
330 mlx5_nic_vport_enable_roce(dev1);
332 } else if (do_bond && __mlx5_lag_is_active(ldev)) {
333 mlx5_modify_lag(ldev, &tracker);
334 } else if (!do_bond && __mlx5_lag_is_active(ldev)) {
335 roce_lag = __mlx5_lag_is_roce(ldev);
338 mlx5_remove_dev_by_protocol(dev0, MLX5_INTERFACE_PROTOCOL_IB);
339 mlx5_nic_vport_disable_roce(dev1);
342 err = mlx5_deactivate_lag(ldev);
347 mlx5_lag_add_ib_devices(ldev);
351 static void mlx5_queue_bond_work(struct mlx5_lag *ldev, unsigned long delay)
353 queue_delayed_work(ldev->wq, &ldev->bond_work, delay);
356 static void mlx5_do_bond_work(struct work_struct *work)
358 struct delayed_work *delayed_work = to_delayed_work(work);
359 struct mlx5_lag *ldev = container_of(delayed_work, struct mlx5_lag,
363 status = mlx5_dev_list_trylock();
366 mlx5_queue_bond_work(ldev, HZ);
371 mlx5_dev_list_unlock();
374 static int mlx5_handle_changeupper_event(struct mlx5_lag *ldev,
375 struct lag_tracker *tracker,
376 struct net_device *ndev,
377 struct netdev_notifier_changeupper_info *info)
379 struct net_device *upper = info->upper_dev, *ndev_tmp;
380 struct netdev_lag_upper_info *lag_upper_info = NULL;
386 if (!netif_is_lag_master(upper))
390 lag_upper_info = info->upper_info;
392 /* The event may still be of interest if the slave does not belong to
393 * us, but is enslaved to a master which has one or more of our netdevs
394 * as slaves (e.g., if a new slave is added to a master that bonds two
395 * of our netdevs, we should unbond).
398 for_each_netdev_in_bond_rcu(upper, ndev_tmp) {
399 idx = mlx5_lag_dev_get_netdev_idx(ldev, ndev_tmp);
401 bond_status |= (1 << idx);
407 /* None of this lagdev's netdevs are slaves of this master. */
408 if (!(bond_status & 0x3))
412 tracker->tx_type = lag_upper_info->tx_type;
414 /* Determine bonding status:
415 * A device is considered bonded if both its physical ports are slaves
416 * of the same lag master, and only them.
417 * Lag mode must be activebackup or hash.
419 is_bonded = (num_slaves == MLX5_MAX_PORTS) &&
420 (bond_status == 0x3) &&
421 ((tracker->tx_type == NETDEV_LAG_TX_TYPE_ACTIVEBACKUP) ||
422 (tracker->tx_type == NETDEV_LAG_TX_TYPE_HASH));
424 if (tracker->is_bonded != is_bonded) {
425 tracker->is_bonded = is_bonded;
432 static int mlx5_handle_changelowerstate_event(struct mlx5_lag *ldev,
433 struct lag_tracker *tracker,
434 struct net_device *ndev,
435 struct netdev_notifier_changelowerstate_info *info)
437 struct netdev_lag_lower_state_info *lag_lower_info;
440 if (!netif_is_lag_port(ndev))
443 idx = mlx5_lag_dev_get_netdev_idx(ldev, ndev);
447 /* This information is used to determine virtual to physical
450 lag_lower_info = info->lower_state_info;
454 tracker->netdev_state[idx] = *lag_lower_info;
459 static int mlx5_lag_netdev_event(struct notifier_block *this,
460 unsigned long event, void *ptr)
462 struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
463 struct lag_tracker tracker;
464 struct mlx5_lag *ldev;
467 if ((event != NETDEV_CHANGEUPPER) && (event != NETDEV_CHANGELOWERSTATE))
470 ldev = container_of(this, struct mlx5_lag, nb);
471 tracker = ldev->tracker;
474 case NETDEV_CHANGEUPPER:
475 changed = mlx5_handle_changeupper_event(ldev, &tracker, ndev,
478 case NETDEV_CHANGELOWERSTATE:
479 changed = mlx5_handle_changelowerstate_event(ldev, &tracker,
484 mutex_lock(&lag_mutex);
485 ldev->tracker = tracker;
486 mutex_unlock(&lag_mutex);
489 mlx5_queue_bond_work(ldev, 0);
494 static struct mlx5_lag *mlx5_lag_dev_alloc(void)
496 struct mlx5_lag *ldev;
498 ldev = kzalloc(sizeof(*ldev), GFP_KERNEL);
502 ldev->wq = create_singlethread_workqueue("mlx5_lag");
508 INIT_DELAYED_WORK(&ldev->bond_work, mlx5_do_bond_work);
513 static void mlx5_lag_dev_free(struct mlx5_lag *ldev)
515 destroy_workqueue(ldev->wq);
519 static void mlx5_lag_dev_add_pf(struct mlx5_lag *ldev,
520 struct mlx5_core_dev *dev,
521 struct net_device *netdev)
523 unsigned int fn = PCI_FUNC(dev->pdev->devfn);
525 if (fn >= MLX5_MAX_PORTS)
528 mutex_lock(&lag_mutex);
529 ldev->pf[fn].dev = dev;
530 ldev->pf[fn].netdev = netdev;
531 ldev->tracker.netdev_state[fn].link_up = 0;
532 ldev->tracker.netdev_state[fn].tx_enabled = 0;
534 dev->priv.lag = ldev;
536 mutex_unlock(&lag_mutex);
539 static void mlx5_lag_dev_remove_pf(struct mlx5_lag *ldev,
540 struct mlx5_core_dev *dev)
544 for (i = 0; i < MLX5_MAX_PORTS; i++)
545 if (ldev->pf[i].dev == dev)
548 if (i == MLX5_MAX_PORTS)
551 mutex_lock(&lag_mutex);
552 memset(&ldev->pf[i], 0, sizeof(*ldev->pf));
554 dev->priv.lag = NULL;
555 mutex_unlock(&lag_mutex);
558 /* Must be called with intf_mutex held */
559 void mlx5_lag_add(struct mlx5_core_dev *dev, struct net_device *netdev)
561 struct mlx5_lag *ldev = NULL;
562 struct mlx5_core_dev *tmp_dev;
565 if (!MLX5_CAP_GEN(dev, vport_group_manager) ||
566 !MLX5_CAP_GEN(dev, lag_master) ||
567 (MLX5_CAP_GEN(dev, num_lag_ports) != MLX5_MAX_PORTS))
570 tmp_dev = mlx5_get_next_phys_dev(dev);
572 ldev = tmp_dev->priv.lag;
575 ldev = mlx5_lag_dev_alloc();
577 mlx5_core_err(dev, "Failed to alloc lag dev\n");
582 mlx5_lag_dev_add_pf(ldev, dev, netdev);
584 if (!ldev->nb.notifier_call) {
585 ldev->nb.notifier_call = mlx5_lag_netdev_event;
586 if (register_netdevice_notifier_net(&init_net, &ldev->nb)) {
587 ldev->nb.notifier_call = NULL;
588 mlx5_core_err(dev, "Failed to register LAG netdev notifier\n");
592 err = mlx5_lag_mp_init(ldev);
594 mlx5_core_err(dev, "Failed to init multipath lag err=%d\n",
598 /* Must be called with intf_mutex held */
599 void mlx5_lag_remove(struct mlx5_core_dev *dev)
601 struct mlx5_lag *ldev;
604 ldev = mlx5_lag_dev_get(dev);
608 if (__mlx5_lag_is_active(ldev))
609 mlx5_deactivate_lag(ldev);
611 mlx5_lag_dev_remove_pf(ldev, dev);
613 for (i = 0; i < MLX5_MAX_PORTS; i++)
617 if (i == MLX5_MAX_PORTS) {
618 if (ldev->nb.notifier_call)
619 unregister_netdevice_notifier_net(&init_net, &ldev->nb);
620 mlx5_lag_mp_cleanup(ldev);
621 cancel_delayed_work_sync(&ldev->bond_work);
622 mlx5_lag_dev_free(ldev);
626 bool mlx5_lag_is_roce(struct mlx5_core_dev *dev)
628 struct mlx5_lag *ldev;
631 mutex_lock(&lag_mutex);
632 ldev = mlx5_lag_dev_get(dev);
633 res = ldev && __mlx5_lag_is_roce(ldev);
634 mutex_unlock(&lag_mutex);
638 EXPORT_SYMBOL(mlx5_lag_is_roce);
640 bool mlx5_lag_is_active(struct mlx5_core_dev *dev)
642 struct mlx5_lag *ldev;
645 mutex_lock(&lag_mutex);
646 ldev = mlx5_lag_dev_get(dev);
647 res = ldev && __mlx5_lag_is_active(ldev);
648 mutex_unlock(&lag_mutex);
652 EXPORT_SYMBOL(mlx5_lag_is_active);
654 bool mlx5_lag_is_sriov(struct mlx5_core_dev *dev)
656 struct mlx5_lag *ldev;
659 mutex_lock(&lag_mutex);
660 ldev = mlx5_lag_dev_get(dev);
661 res = ldev && __mlx5_lag_is_sriov(ldev);
662 mutex_unlock(&lag_mutex);
666 EXPORT_SYMBOL(mlx5_lag_is_sriov);
668 void mlx5_lag_update(struct mlx5_core_dev *dev)
670 struct mlx5_lag *ldev;
672 mlx5_dev_list_lock();
673 ldev = mlx5_lag_dev_get(dev);
680 mlx5_dev_list_unlock();
683 struct net_device *mlx5_lag_get_roce_netdev(struct mlx5_core_dev *dev)
685 struct net_device *ndev = NULL;
686 struct mlx5_lag *ldev;
688 mutex_lock(&lag_mutex);
689 ldev = mlx5_lag_dev_get(dev);
691 if (!(ldev && __mlx5_lag_is_roce(ldev)))
694 if (ldev->tracker.tx_type == NETDEV_LAG_TX_TYPE_ACTIVEBACKUP) {
695 ndev = ldev->tracker.netdev_state[MLX5_LAG_P1].tx_enabled ?
696 ldev->pf[MLX5_LAG_P1].netdev :
697 ldev->pf[MLX5_LAG_P2].netdev;
699 ndev = ldev->pf[MLX5_LAG_P1].netdev;
705 mutex_unlock(&lag_mutex);
709 EXPORT_SYMBOL(mlx5_lag_get_roce_netdev);
711 bool mlx5_lag_intf_add(struct mlx5_interface *intf, struct mlx5_priv *priv)
713 struct mlx5_core_dev *dev = container_of(priv, struct mlx5_core_dev,
715 struct mlx5_lag *ldev;
717 if (intf->protocol != MLX5_INTERFACE_PROTOCOL_IB)
720 ldev = mlx5_lag_dev_get(dev);
721 if (!ldev || !__mlx5_lag_is_roce(ldev) ||
722 ldev->pf[MLX5_LAG_P1].dev == dev)
725 /* If bonded, we do not add an IB device for PF1. */
729 int mlx5_lag_query_cong_counters(struct mlx5_core_dev *dev,
734 int outlen = MLX5_ST_SZ_BYTES(query_cong_statistics_out);
735 struct mlx5_core_dev *mdev[MLX5_MAX_PORTS];
736 struct mlx5_lag *ldev;
741 out = kvzalloc(outlen, GFP_KERNEL);
745 memset(values, 0, sizeof(*values) * num_counters);
747 mutex_lock(&lag_mutex);
748 ldev = mlx5_lag_dev_get(dev);
749 if (ldev && __mlx5_lag_is_roce(ldev)) {
750 num_ports = MLX5_MAX_PORTS;
751 mdev[MLX5_LAG_P1] = ldev->pf[MLX5_LAG_P1].dev;
752 mdev[MLX5_LAG_P2] = ldev->pf[MLX5_LAG_P2].dev;
755 mdev[MLX5_LAG_P1] = dev;
758 for (i = 0; i < num_ports; ++i) {
759 ret = mlx5_cmd_query_cong_counter(mdev[i], false, out, outlen);
763 for (j = 0; j < num_counters; ++j)
764 values[j] += be64_to_cpup((__be64 *)(out + offsets[j]));
768 mutex_unlock(&lag_mutex);
772 EXPORT_SYMBOL(mlx5_lag_query_cong_counters);