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_SPINLOCK(lag_lock);
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)] = {};
51 void *lag_ctx = MLX5_ADDR_OF(create_lag_in, in, ctx);
53 MLX5_SET(create_lag_in, in, opcode, MLX5_CMD_OP_CREATE_LAG);
55 MLX5_SET(lagc, lag_ctx, tx_remap_affinity_1, remap_port1);
56 MLX5_SET(lagc, lag_ctx, tx_remap_affinity_2, remap_port2);
58 return mlx5_cmd_exec_in(dev, create_lag, in);
61 static int mlx5_cmd_modify_lag(struct mlx5_core_dev *dev, u8 remap_port1,
64 u32 in[MLX5_ST_SZ_DW(modify_lag_in)] = {};
65 void *lag_ctx = MLX5_ADDR_OF(modify_lag_in, in, ctx);
67 MLX5_SET(modify_lag_in, in, opcode, MLX5_CMD_OP_MODIFY_LAG);
68 MLX5_SET(modify_lag_in, in, field_select, 0x1);
70 MLX5_SET(lagc, lag_ctx, tx_remap_affinity_1, remap_port1);
71 MLX5_SET(lagc, lag_ctx, tx_remap_affinity_2, remap_port2);
73 return mlx5_cmd_exec_in(dev, modify_lag, in);
76 int mlx5_cmd_create_vport_lag(struct mlx5_core_dev *dev)
78 u32 in[MLX5_ST_SZ_DW(create_vport_lag_in)] = {};
80 MLX5_SET(create_vport_lag_in, in, opcode, MLX5_CMD_OP_CREATE_VPORT_LAG);
82 return mlx5_cmd_exec_in(dev, create_vport_lag, in);
84 EXPORT_SYMBOL(mlx5_cmd_create_vport_lag);
86 int mlx5_cmd_destroy_vport_lag(struct mlx5_core_dev *dev)
88 u32 in[MLX5_ST_SZ_DW(destroy_vport_lag_in)] = {};
90 MLX5_SET(destroy_vport_lag_in, in, opcode, MLX5_CMD_OP_DESTROY_VPORT_LAG);
92 return mlx5_cmd_exec_in(dev, destroy_vport_lag, in);
94 EXPORT_SYMBOL(mlx5_cmd_destroy_vport_lag);
96 int mlx5_lag_dev_get_netdev_idx(struct mlx5_lag *ldev,
97 struct net_device *ndev)
101 for (i = 0; i < MLX5_MAX_PORTS; i++)
102 if (ldev->pf[i].netdev == ndev)
108 static bool __mlx5_lag_is_roce(struct mlx5_lag *ldev)
110 return !!(ldev->flags & MLX5_LAG_FLAG_ROCE);
113 static bool __mlx5_lag_is_sriov(struct mlx5_lag *ldev)
115 return !!(ldev->flags & MLX5_LAG_FLAG_SRIOV);
118 static void mlx5_infer_tx_affinity_mapping(struct lag_tracker *tracker,
119 u8 *port1, u8 *port2)
123 if (!tracker->netdev_state[MLX5_LAG_P1].tx_enabled ||
124 !tracker->netdev_state[MLX5_LAG_P1].link_up) {
129 if (!tracker->netdev_state[MLX5_LAG_P2].tx_enabled ||
130 !tracker->netdev_state[MLX5_LAG_P2].link_up)
134 void mlx5_modify_lag(struct mlx5_lag *ldev,
135 struct lag_tracker *tracker)
137 struct mlx5_core_dev *dev0 = ldev->pf[MLX5_LAG_P1].dev;
138 u8 v2p_port1, v2p_port2;
141 mlx5_infer_tx_affinity_mapping(tracker, &v2p_port1,
144 if (v2p_port1 != ldev->v2p_map[MLX5_LAG_P1] ||
145 v2p_port2 != ldev->v2p_map[MLX5_LAG_P2]) {
146 ldev->v2p_map[MLX5_LAG_P1] = v2p_port1;
147 ldev->v2p_map[MLX5_LAG_P2] = v2p_port2;
149 mlx5_core_info(dev0, "modify lag map port 1:%d port 2:%d",
150 ldev->v2p_map[MLX5_LAG_P1],
151 ldev->v2p_map[MLX5_LAG_P2]);
153 err = mlx5_cmd_modify_lag(dev0, v2p_port1, v2p_port2);
156 "Failed to modify LAG (%d)\n",
161 static int mlx5_create_lag(struct mlx5_lag *ldev,
162 struct lag_tracker *tracker)
164 struct mlx5_core_dev *dev0 = ldev->pf[MLX5_LAG_P1].dev;
167 mlx5_infer_tx_affinity_mapping(tracker, &ldev->v2p_map[MLX5_LAG_P1],
168 &ldev->v2p_map[MLX5_LAG_P2]);
170 mlx5_core_info(dev0, "lag map port 1:%d port 2:%d",
171 ldev->v2p_map[MLX5_LAG_P1], ldev->v2p_map[MLX5_LAG_P2]);
173 err = mlx5_cmd_create_lag(dev0, ldev->v2p_map[MLX5_LAG_P1],
174 ldev->v2p_map[MLX5_LAG_P2]);
177 "Failed to create LAG (%d)\n",
182 int mlx5_activate_lag(struct mlx5_lag *ldev,
183 struct lag_tracker *tracker,
186 bool roce_lag = !!(flags & MLX5_LAG_FLAG_ROCE);
187 struct mlx5_core_dev *dev0 = ldev->pf[MLX5_LAG_P1].dev;
190 err = mlx5_create_lag(ldev, tracker);
194 "Failed to activate RoCE LAG\n");
197 "Failed to activate VF LAG\n"
198 "Make sure all VFs are unbound prior to VF LAG activation or deactivation\n");
203 ldev->flags |= flags;
207 static int mlx5_deactivate_lag(struct mlx5_lag *ldev)
209 struct mlx5_core_dev *dev0 = ldev->pf[MLX5_LAG_P1].dev;
210 u32 in[MLX5_ST_SZ_DW(destroy_lag_in)] = {};
211 bool roce_lag = __mlx5_lag_is_roce(ldev);
214 ldev->flags &= ~MLX5_LAG_MODE_FLAGS;
216 MLX5_SET(destroy_lag_in, in, opcode, MLX5_CMD_OP_DESTROY_LAG);
217 err = mlx5_cmd_exec_in(dev0, destroy_lag, in);
221 "Failed to deactivate RoCE LAG; driver restart required\n");
224 "Failed to deactivate VF LAG; driver restart required\n"
225 "Make sure all VFs are unbound prior to VF LAG activation or deactivation\n");
232 static bool mlx5_lag_check_prereq(struct mlx5_lag *ldev)
234 if (!ldev->pf[MLX5_LAG_P1].dev || !ldev->pf[MLX5_LAG_P2].dev)
237 #ifdef CONFIG_MLX5_ESWITCH
238 return mlx5_esw_lag_prereq(ldev->pf[MLX5_LAG_P1].dev,
239 ldev->pf[MLX5_LAG_P2].dev);
241 return (!mlx5_sriov_is_enabled(ldev->pf[MLX5_LAG_P1].dev) &&
242 !mlx5_sriov_is_enabled(ldev->pf[MLX5_LAG_P2].dev));
246 static void mlx5_lag_add_devices(struct mlx5_lag *ldev)
250 for (i = 0; i < MLX5_MAX_PORTS; i++) {
251 if (!ldev->pf[i].dev)
254 ldev->pf[i].dev->priv.flags &= ~MLX5_PRIV_FLAGS_DISABLE_IB_ADEV;
255 mlx5_rescan_drivers_locked(ldev->pf[i].dev);
259 static void mlx5_lag_remove_devices(struct mlx5_lag *ldev)
263 for (i = 0; i < MLX5_MAX_PORTS; i++) {
264 if (!ldev->pf[i].dev)
267 ldev->pf[i].dev->priv.flags |= MLX5_PRIV_FLAGS_DISABLE_IB_ADEV;
268 mlx5_rescan_drivers_locked(ldev->pf[i].dev);
272 static void mlx5_do_bond(struct mlx5_lag *ldev)
274 struct mlx5_core_dev *dev0 = ldev->pf[MLX5_LAG_P1].dev;
275 struct mlx5_core_dev *dev1 = ldev->pf[MLX5_LAG_P2].dev;
276 struct lag_tracker tracker;
277 bool do_bond, roce_lag;
280 if (!mlx5_lag_is_ready(ldev))
283 spin_lock(&lag_lock);
284 tracker = ldev->tracker;
285 spin_unlock(&lag_lock);
287 do_bond = tracker.is_bonded && mlx5_lag_check_prereq(ldev);
289 if (do_bond && !__mlx5_lag_is_active(ldev)) {
290 roce_lag = !mlx5_sriov_is_enabled(dev0) &&
291 !mlx5_sriov_is_enabled(dev1);
293 #ifdef CONFIG_MLX5_ESWITCH
294 roce_lag &= dev0->priv.eswitch->mode == MLX5_ESWITCH_NONE &&
295 dev1->priv.eswitch->mode == MLX5_ESWITCH_NONE;
299 mlx5_lag_remove_devices(ldev);
301 err = mlx5_activate_lag(ldev, &tracker,
302 roce_lag ? MLX5_LAG_FLAG_ROCE :
303 MLX5_LAG_FLAG_SRIOV);
306 mlx5_lag_add_devices(ldev);
312 dev0->priv.flags &= ~MLX5_PRIV_FLAGS_DISABLE_IB_ADEV;
313 mlx5_rescan_drivers_locked(dev0);
314 mlx5_nic_vport_enable_roce(dev1);
316 } else if (do_bond && __mlx5_lag_is_active(ldev)) {
317 mlx5_modify_lag(ldev, &tracker);
318 } else if (!do_bond && __mlx5_lag_is_active(ldev)) {
319 roce_lag = __mlx5_lag_is_roce(ldev);
322 dev0->priv.flags |= MLX5_PRIV_FLAGS_DISABLE_IB_ADEV;
323 mlx5_rescan_drivers_locked(dev0);
324 mlx5_nic_vport_disable_roce(dev1);
327 err = mlx5_deactivate_lag(ldev);
332 mlx5_lag_add_devices(ldev);
336 static void mlx5_queue_bond_work(struct mlx5_lag *ldev, unsigned long delay)
338 queue_delayed_work(ldev->wq, &ldev->bond_work, delay);
341 static void mlx5_do_bond_work(struct work_struct *work)
343 struct delayed_work *delayed_work = to_delayed_work(work);
344 struct mlx5_lag *ldev = container_of(delayed_work, struct mlx5_lag,
348 status = mlx5_dev_list_trylock();
351 mlx5_queue_bond_work(ldev, HZ);
356 mlx5_dev_list_unlock();
359 static int mlx5_handle_changeupper_event(struct mlx5_lag *ldev,
360 struct lag_tracker *tracker,
361 struct net_device *ndev,
362 struct netdev_notifier_changeupper_info *info)
364 struct net_device *upper = info->upper_dev, *ndev_tmp;
365 struct netdev_lag_upper_info *lag_upper_info = NULL;
366 bool is_bonded, is_in_lag, mode_supported;
371 if (!netif_is_lag_master(upper))
375 lag_upper_info = info->upper_info;
377 /* The event may still be of interest if the slave does not belong to
378 * us, but is enslaved to a master which has one or more of our netdevs
379 * as slaves (e.g., if a new slave is added to a master that bonds two
380 * of our netdevs, we should unbond).
383 for_each_netdev_in_bond_rcu(upper, ndev_tmp) {
384 idx = mlx5_lag_dev_get_netdev_idx(ldev, ndev_tmp);
386 bond_status |= (1 << idx);
392 /* None of this lagdev's netdevs are slaves of this master. */
393 if (!(bond_status & 0x3))
397 tracker->tx_type = lag_upper_info->tx_type;
399 /* Determine bonding status:
400 * A device is considered bonded if both its physical ports are slaves
401 * of the same lag master, and only them.
403 is_in_lag = num_slaves == MLX5_MAX_PORTS && bond_status == 0x3;
405 if (!mlx5_lag_is_ready(ldev) && is_in_lag) {
406 NL_SET_ERR_MSG_MOD(info->info.extack,
407 "Can't activate LAG offload, PF is configured with more than 64 VFs");
411 /* Lag mode must be activebackup or hash. */
412 mode_supported = tracker->tx_type == NETDEV_LAG_TX_TYPE_ACTIVEBACKUP ||
413 tracker->tx_type == NETDEV_LAG_TX_TYPE_HASH;
415 if (is_in_lag && !mode_supported)
416 NL_SET_ERR_MSG_MOD(info->info.extack,
417 "Can't activate LAG offload, TX type isn't supported");
419 is_bonded = is_in_lag && mode_supported;
420 if (tracker->is_bonded != is_bonded) {
421 tracker->is_bonded = is_bonded;
428 static int mlx5_handle_changelowerstate_event(struct mlx5_lag *ldev,
429 struct lag_tracker *tracker,
430 struct net_device *ndev,
431 struct netdev_notifier_changelowerstate_info *info)
433 struct netdev_lag_lower_state_info *lag_lower_info;
436 if (!netif_is_lag_port(ndev))
439 idx = mlx5_lag_dev_get_netdev_idx(ldev, ndev);
443 /* This information is used to determine virtual to physical
446 lag_lower_info = info->lower_state_info;
450 tracker->netdev_state[idx] = *lag_lower_info;
455 static int mlx5_lag_netdev_event(struct notifier_block *this,
456 unsigned long event, void *ptr)
458 struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
459 struct lag_tracker tracker;
460 struct mlx5_lag *ldev;
463 if ((event != NETDEV_CHANGEUPPER) && (event != NETDEV_CHANGELOWERSTATE))
466 ldev = container_of(this, struct mlx5_lag, nb);
468 if (!mlx5_lag_is_ready(ldev) && event == NETDEV_CHANGELOWERSTATE)
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 spin_lock(&lag_lock);
485 ldev->tracker = tracker;
486 spin_unlock(&lag_lock);
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 int 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 spin_lock(&lag_lock);
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 spin_unlock(&lag_lock);
541 static void mlx5_lag_dev_remove_pf(struct mlx5_lag *ldev,
542 struct mlx5_core_dev *dev)
546 for (i = 0; i < MLX5_MAX_PORTS; i++)
547 if (ldev->pf[i].dev == dev)
550 if (i == MLX5_MAX_PORTS)
553 spin_lock(&lag_lock);
554 memset(&ldev->pf[i], 0, sizeof(*ldev->pf));
556 dev->priv.lag = NULL;
557 spin_unlock(&lag_lock);
560 /* Must be called with intf_mutex held */
561 void mlx5_lag_add(struct mlx5_core_dev *dev, struct net_device *netdev)
563 struct mlx5_lag *ldev = NULL;
564 struct mlx5_core_dev *tmp_dev;
567 if (!MLX5_CAP_GEN(dev, vport_group_manager))
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 if (mlx5_lag_dev_add_pf(ldev, dev, netdev) < 0)
585 for (i = 0; i < MLX5_MAX_PORTS; i++) {
586 tmp_dev = ldev->pf[i].dev;
587 if (!tmp_dev || !MLX5_CAP_GEN(tmp_dev, lag_master) ||
588 MLX5_CAP_GEN(tmp_dev, num_lag_ports) != MLX5_MAX_PORTS)
592 if (i >= MLX5_MAX_PORTS)
593 ldev->flags |= MLX5_LAG_FLAG_READY;
595 if (!ldev->nb.notifier_call) {
596 ldev->nb.notifier_call = mlx5_lag_netdev_event;
597 if (register_netdevice_notifier_net(&init_net, &ldev->nb)) {
598 ldev->nb.notifier_call = NULL;
599 mlx5_core_err(dev, "Failed to register LAG netdev notifier\n");
603 err = mlx5_lag_mp_init(ldev);
605 mlx5_core_err(dev, "Failed to init multipath lag err=%d\n",
611 /* Must be called with intf_mutex held */
612 void mlx5_lag_remove(struct mlx5_core_dev *dev)
614 struct mlx5_lag *ldev;
617 ldev = mlx5_lag_dev_get(dev);
621 if (__mlx5_lag_is_active(ldev))
622 mlx5_deactivate_lag(ldev);
624 mlx5_lag_dev_remove_pf(ldev, dev);
626 ldev->flags &= ~MLX5_LAG_FLAG_READY;
628 for (i = 0; i < MLX5_MAX_PORTS; i++)
632 if (i == MLX5_MAX_PORTS) {
633 if (ldev->nb.notifier_call) {
634 unregister_netdevice_notifier_net(&init_net, &ldev->nb);
635 ldev->nb.notifier_call = NULL;
637 mlx5_lag_mp_cleanup(ldev);
638 cancel_delayed_work_sync(&ldev->bond_work);
639 mlx5_lag_dev_free(ldev);
643 bool mlx5_lag_is_roce(struct mlx5_core_dev *dev)
645 struct mlx5_lag *ldev;
648 spin_lock(&lag_lock);
649 ldev = mlx5_lag_dev_get(dev);
650 res = ldev && __mlx5_lag_is_roce(ldev);
651 spin_unlock(&lag_lock);
655 EXPORT_SYMBOL(mlx5_lag_is_roce);
657 bool mlx5_lag_is_active(struct mlx5_core_dev *dev)
659 struct mlx5_lag *ldev;
662 spin_lock(&lag_lock);
663 ldev = mlx5_lag_dev_get(dev);
664 res = ldev && __mlx5_lag_is_active(ldev);
665 spin_unlock(&lag_lock);
669 EXPORT_SYMBOL(mlx5_lag_is_active);
671 bool mlx5_lag_is_sriov(struct mlx5_core_dev *dev)
673 struct mlx5_lag *ldev;
676 spin_lock(&lag_lock);
677 ldev = mlx5_lag_dev_get(dev);
678 res = ldev && __mlx5_lag_is_sriov(ldev);
679 spin_unlock(&lag_lock);
683 EXPORT_SYMBOL(mlx5_lag_is_sriov);
685 void mlx5_lag_update(struct mlx5_core_dev *dev)
687 struct mlx5_lag *ldev;
689 mlx5_dev_list_lock();
690 ldev = mlx5_lag_dev_get(dev);
697 mlx5_dev_list_unlock();
700 struct net_device *mlx5_lag_get_roce_netdev(struct mlx5_core_dev *dev)
702 struct net_device *ndev = NULL;
703 struct mlx5_lag *ldev;
705 spin_lock(&lag_lock);
706 ldev = mlx5_lag_dev_get(dev);
708 if (!(ldev && __mlx5_lag_is_roce(ldev)))
711 if (ldev->tracker.tx_type == NETDEV_LAG_TX_TYPE_ACTIVEBACKUP) {
712 ndev = ldev->tracker.netdev_state[MLX5_LAG_P1].tx_enabled ?
713 ldev->pf[MLX5_LAG_P1].netdev :
714 ldev->pf[MLX5_LAG_P2].netdev;
716 ndev = ldev->pf[MLX5_LAG_P1].netdev;
722 spin_unlock(&lag_lock);
726 EXPORT_SYMBOL(mlx5_lag_get_roce_netdev);
728 u8 mlx5_lag_get_slave_port(struct mlx5_core_dev *dev,
729 struct net_device *slave)
731 struct mlx5_lag *ldev;
734 spin_lock(&lag_lock);
735 ldev = mlx5_lag_dev_get(dev);
736 if (!(ldev && __mlx5_lag_is_roce(ldev)))
739 if (ldev->pf[MLX5_LAG_P1].netdev == slave)
744 port = ldev->v2p_map[port];
747 spin_unlock(&lag_lock);
750 EXPORT_SYMBOL(mlx5_lag_get_slave_port);
752 int mlx5_lag_query_cong_counters(struct mlx5_core_dev *dev,
757 int outlen = MLX5_ST_SZ_BYTES(query_cong_statistics_out);
758 struct mlx5_core_dev *mdev[MLX5_MAX_PORTS];
759 struct mlx5_lag *ldev;
764 out = kvzalloc(outlen, GFP_KERNEL);
768 memset(values, 0, sizeof(*values) * num_counters);
770 spin_lock(&lag_lock);
771 ldev = mlx5_lag_dev_get(dev);
772 if (ldev && __mlx5_lag_is_roce(ldev)) {
773 num_ports = MLX5_MAX_PORTS;
774 mdev[MLX5_LAG_P1] = ldev->pf[MLX5_LAG_P1].dev;
775 mdev[MLX5_LAG_P2] = ldev->pf[MLX5_LAG_P2].dev;
778 mdev[MLX5_LAG_P1] = dev;
780 spin_unlock(&lag_lock);
782 for (i = 0; i < num_ports; ++i) {
783 u32 in[MLX5_ST_SZ_DW(query_cong_statistics_in)] = {};
785 MLX5_SET(query_cong_statistics_in, in, opcode,
786 MLX5_CMD_OP_QUERY_CONG_STATISTICS);
787 ret = mlx5_cmd_exec_inout(mdev[i], query_cong_statistics, in,
792 for (j = 0; j < num_counters; ++j)
793 values[j] += be64_to_cpup((__be64 *)(out + offsets[j]));
800 EXPORT_SYMBOL(mlx5_lag_query_cong_counters);