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

   1 /*
   2  * Copyright (c) 2016, Mellanox Technologies. All rights reserved.
   3  *
   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:
   9  *
  10  *     Redistribution and use in source and binary forms, with or
  11  *     without modification, are permitted provided that the following
  12  *     conditions are met:
  13  *
  14  *      - Redistributions of source code must retain the above
  15  *        copyright notice, this list of conditions and the following
  16  *        disclaimer.
  17  *
  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.
  22  *
  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
  30  * SOFTWARE.
  31  */
  32
  33 #include <linux/netdevice.h>
  34 #include <linux/mlx5/driver.h>
  35 #include <linux/mlx5/vport.h>
  36 #include "mlx5_core.h"
  37 #include "eswitch.h"
  38 #include "lag.h"
  39 #include "lag_mp.h"
  40
  41 /* General purpose, use for short periods of time.
  42  * Beware of lock dependencies (preferably, no locks should be acquired
  43  * under it).
  44  */
  45 static DEFINE_MUTEX(lag_mutex);
  46
  47 static int mlx5_cmd_create_lag(struct mlx5_core_dev *dev, u8 remap_port1,
  48                                u8 remap_port2)
  49 {
  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);
  53
  54         MLX5_SET(create_lag_in, in, opcode, MLX5_CMD_OP_CREATE_LAG);
  55
  56         MLX5_SET(lagc, lag_ctx, tx_remap_affinity_1, remap_port1);
  57         MLX5_SET(lagc, lag_ctx, tx_remap_affinity_2, remap_port2);
  58
  59         return mlx5_cmd_exec(dev, in, sizeof(in), out, sizeof(out));
  60 }
  61
  62 static int mlx5_cmd_modify_lag(struct mlx5_core_dev *dev, u8 remap_port1,
  63                                u8 remap_port2)
  64 {
  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);
  68
  69         MLX5_SET(modify_lag_in, in, opcode, MLX5_CMD_OP_MODIFY_LAG);
  70         MLX5_SET(modify_lag_in, in, field_select, 0x1);
  71
  72         MLX5_SET(lagc, lag_ctx, tx_remap_affinity_1, remap_port1);
  73         MLX5_SET(lagc, lag_ctx, tx_remap_affinity_2, remap_port2);
  74
  75         return mlx5_cmd_exec(dev, in, sizeof(in), out, sizeof(out));
  76 }
  77
  78 static int mlx5_cmd_destroy_lag(struct mlx5_core_dev *dev)
  79 {
  80         u32  in[MLX5_ST_SZ_DW(destroy_lag_in)]  = {0};
  81         u32 out[MLX5_ST_SZ_DW(destroy_lag_out)] = {0};
  82
  83         MLX5_SET(destroy_lag_in, in, opcode, MLX5_CMD_OP_DESTROY_LAG);
  84
  85         return mlx5_cmd_exec(dev, in, sizeof(in), out, sizeof(out));
  86 }
  87
  88 int mlx5_cmd_create_vport_lag(struct mlx5_core_dev *dev)
  89 {
  90         u32  in[MLX5_ST_SZ_DW(create_vport_lag_in)]  = {0};
  91         u32 out[MLX5_ST_SZ_DW(create_vport_lag_out)] = {0};
  92
  93         MLX5_SET(create_vport_lag_in, in, opcode, MLX5_CMD_OP_CREATE_VPORT_LAG);
  94
  95         return mlx5_cmd_exec(dev, in, sizeof(in), out, sizeof(out));
  96 }
  97 EXPORT_SYMBOL(mlx5_cmd_create_vport_lag);
  98
  99 int mlx5_cmd_destroy_vport_lag(struct mlx5_core_dev *dev)
 100 {
 101         u32  in[MLX5_ST_SZ_DW(destroy_vport_lag_in)]  = {0};
 102         u32 out[MLX5_ST_SZ_DW(destroy_vport_lag_out)] = {0};
 103
 104         MLX5_SET(destroy_vport_lag_in, in, opcode, MLX5_CMD_OP_DESTROY_VPORT_LAG);
 105
 106         return mlx5_cmd_exec(dev, in, sizeof(in), out, sizeof(out));
 107 }
 108 EXPORT_SYMBOL(mlx5_cmd_destroy_vport_lag);
 109
 110 static int mlx5_cmd_query_cong_counter(struct mlx5_core_dev *dev,
 111                                        bool reset, void *out, int out_size)
 112 {
 113         u32 in[MLX5_ST_SZ_DW(query_cong_statistics_in)] = { };
 114
 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);
 119 }
 120
 121 int mlx5_lag_dev_get_netdev_idx(struct mlx5_lag *ldev,
 122                                 struct net_device *ndev)
 123 {
 124         int i;
 125
 126         for (i = 0; i < MLX5_MAX_PORTS; i++)
 127                 if (ldev->pf[i].netdev == ndev)
 128                         return i;
 129
 130         return -1;
 131 }
 132
 133 static bool __mlx5_lag_is_roce(struct mlx5_lag *ldev)
 134 {
 135         return !!(ldev->flags & MLX5_LAG_FLAG_ROCE);
 136 }
 137
 138 static bool __mlx5_lag_is_sriov(struct mlx5_lag *ldev)
 139 {
 140         return !!(ldev->flags & MLX5_LAG_FLAG_SRIOV);
 141 }
 142
 143 static void mlx5_infer_tx_affinity_mapping(struct lag_tracker *tracker,
 144                                            u8 *port1, u8 *port2)
 145 {
 146         *port1 = 1;
 147         *port2 = 2;
 148         if (!tracker->netdev_state[MLX5_LAG_P1].tx_enabled ||
 149             !tracker->netdev_state[MLX5_LAG_P1].link_up) {
 150                 *port1 = 2;
 151                 return;
 152         }
 153
 154         if (!tracker->netdev_state[MLX5_LAG_P2].tx_enabled ||
 155             !tracker->netdev_state[MLX5_LAG_P2].link_up)
 156                 *port2 = 1;
 157 }
 158
 159 void mlx5_modify_lag(struct mlx5_lag *ldev,
 160                      struct lag_tracker *tracker)
 161 {
 162         struct mlx5_core_dev *dev0 = ldev->pf[MLX5_LAG_P1].dev;
 163         u8 v2p_port1, v2p_port2;
 164         int err;
 165
 166         mlx5_infer_tx_affinity_mapping(tracker, &v2p_port1,
 167                                        &v2p_port2);
 168
 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;
 173
 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]);
 177
 178                 err = mlx5_cmd_modify_lag(dev0, v2p_port1, v2p_port2);
 179                 if (err)
 180                         mlx5_core_err(dev0,
 181                                       "Failed to modify LAG (%d)\n",
 182                                       err);
 183         }
 184 }
 185
 186 static int mlx5_create_lag(struct mlx5_lag *ldev,
 187                            struct lag_tracker *tracker)
 188 {
 189         struct mlx5_core_dev *dev0 = ldev->pf[MLX5_LAG_P1].dev;
 190         int err;
 191
 192         mlx5_infer_tx_affinity_mapping(tracker, &ldev->v2p_map[MLX5_LAG_P1],
 193                                        &ldev->v2p_map[MLX5_LAG_P2]);
 194
 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]);
 197
 198         err = mlx5_cmd_create_lag(dev0, ldev->v2p_map[MLX5_LAG_P1],
 199                                   ldev->v2p_map[MLX5_LAG_P2]);
 200         if (err)
 201                 mlx5_core_err(dev0,
 202                               "Failed to create LAG (%d)\n",
 203                               err);
 204         return err;
 205 }
 206
 207 int mlx5_activate_lag(struct mlx5_lag *ldev,
 208                       struct lag_tracker *tracker,
 209                       u8 flags)
 210 {
 211         bool roce_lag = !!(flags & MLX5_LAG_FLAG_ROCE);
 212         struct mlx5_core_dev *dev0 = ldev->pf[MLX5_LAG_P1].dev;
 213         int err;
 214
 215         err = mlx5_create_lag(ldev, tracker);
 216         if (err) {
 217                 if (roce_lag) {
 218                         mlx5_core_err(dev0,
 219                                       "Failed to activate RoCE LAG\n");
 220                 } else {
 221                         mlx5_core_err(dev0,
 222                                       "Failed to activate VF LAG\n"
 223                                       "Make sure all VFs are unbound prior to VF LAG activation or deactivation\n");
 224                 }
 225                 return err;
 226         }
 227
 228         ldev->flags |= flags;
 229         return 0;
 230 }
 231
 232 static int mlx5_deactivate_lag(struct mlx5_lag *ldev)
 233 {
 234         struct mlx5_core_dev *dev0 = ldev->pf[MLX5_LAG_P1].dev;
 235         bool roce_lag = __mlx5_lag_is_roce(ldev);
 236         int err;
 237
 238         ldev->flags &= ~MLX5_LAG_MODE_FLAGS;
 239
 240         err = mlx5_cmd_destroy_lag(dev0);
 241         if (err) {
 242                 if (roce_lag) {
 243                         mlx5_core_err(dev0,
 244                                       "Failed to deactivate RoCE LAG; driver restart required\n");
 245                 } else {
 246                         mlx5_core_err(dev0,
 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");
 249                 }
 250         }
 251
 252         return err;
 253 }
 254
 255 static bool mlx5_lag_check_prereq(struct mlx5_lag *ldev)
 256 {
 257         if (!ldev->pf[MLX5_LAG_P1].dev || !ldev->pf[MLX5_LAG_P2].dev)
 258                 return false;
 259
 260 #ifdef CONFIG_MLX5_ESWITCH
 261         return mlx5_esw_lag_prereq(ldev->pf[MLX5_LAG_P1].dev,
 262                                    ldev->pf[MLX5_LAG_P2].dev);
 263 #else
 264         return (!mlx5_sriov_is_enabled(ldev->pf[MLX5_LAG_P1].dev) &&
 265                 !mlx5_sriov_is_enabled(ldev->pf[MLX5_LAG_P2].dev));
 266 #endif
 267 }
 268
 269 static void mlx5_lag_add_ib_devices(struct mlx5_lag *ldev)
 270 {
 271         int i;
 272
 273         for (i = 0; i < MLX5_MAX_PORTS; i++)
 274                 if (ldev->pf[i].dev)
 275                         mlx5_add_dev_by_protocol(ldev->pf[i].dev,
 276                                                  MLX5_INTERFACE_PROTOCOL_IB);
 277 }
 278
 279 static void mlx5_lag_remove_ib_devices(struct mlx5_lag *ldev)
 280 {
 281         int i;
 282
 283         for (i = 0; i < MLX5_MAX_PORTS; i++)
 284                 if (ldev->pf[i].dev)
 285                         mlx5_remove_dev_by_protocol(ldev->pf[i].dev,
 286                                                     MLX5_INTERFACE_PROTOCOL_IB);
 287 }
 288
 289 static void mlx5_do_bond(struct mlx5_lag *ldev)
 290 {
 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;
 295         int err;
 296
 297         if (!dev0 || !dev1)
 298                 return;
 299
 300         mutex_lock(&lag_mutex);
 301         tracker = ldev->tracker;
 302         mutex_unlock(&lag_mutex);
 303
 304         do_bond = tracker.is_bonded && mlx5_lag_check_prereq(ldev);
 305
 306         if (do_bond && !__mlx5_lag_is_active(ldev)) {
 307                 roce_lag = !mlx5_sriov_is_enabled(dev0) &&
 308                            !mlx5_sriov_is_enabled(dev1);
 309
 310 #ifdef CONFIG_MLX5_ESWITCH
 311                 roce_lag &= dev0->priv.eswitch->mode == MLX5_ESWITCH_NONE &&
 312                             dev1->priv.eswitch->mode == MLX5_ESWITCH_NONE;
 313 #endif
 314
 315                 if (roce_lag)
 316                         mlx5_lag_remove_ib_devices(ldev);
 317
 318                 err = mlx5_activate_lag(ldev, &tracker,
 319                                         roce_lag ? MLX5_LAG_FLAG_ROCE :
 320                                         MLX5_LAG_FLAG_SRIOV);
 321                 if (err) {
 322                         if (roce_lag)
 323                                 mlx5_lag_add_ib_devices(ldev);
 324
 325                         return;
 326                 }
 327
 328                 if (roce_lag) {
 329                         mlx5_add_dev_by_protocol(dev0, MLX5_INTERFACE_PROTOCOL_IB);
 330                         mlx5_nic_vport_enable_roce(dev1);
 331                 }
 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);
 336
 337                 if (roce_lag) {
 338                         mlx5_remove_dev_by_protocol(dev0, MLX5_INTERFACE_PROTOCOL_IB);
 339                         mlx5_nic_vport_disable_roce(dev1);
 340                 }
 341
 342                 err = mlx5_deactivate_lag(ldev);
 343                 if (err)
 344                         return;
 345
 346                 if (roce_lag)
 347                         mlx5_lag_add_ib_devices(ldev);
 348         }
 349 }
 350
 351 static void mlx5_queue_bond_work(struct mlx5_lag *ldev, unsigned long delay)
 352 {
 353         queue_delayed_work(ldev->wq, &ldev->bond_work, delay);
 354 }
 355
 356 static void mlx5_do_bond_work(struct work_struct *work)
 357 {
 358         struct delayed_work *delayed_work = to_delayed_work(work);
 359         struct mlx5_lag *ldev = container_of(delayed_work, struct mlx5_lag,
 360                                              bond_work);
 361         int status;
 362
 363         status = mlx5_dev_list_trylock();
 364         if (!status) {
 365                 /* 1 sec delay. */
 366                 mlx5_queue_bond_work(ldev, HZ);
 367                 return;
 368         }
 369
 370         mlx5_do_bond(ldev);
 371         mlx5_dev_list_unlock();
 372 }
 373
 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)
 378 {
 379         struct net_device *upper = info->upper_dev, *ndev_tmp;
 380         struct netdev_lag_upper_info *lag_upper_info = NULL;
 381         bool is_bonded;
 382         int bond_status = 0;
 383         int num_slaves = 0;
 384         int idx;
 385
 386         if (!netif_is_lag_master(upper))
 387                 return 0;
 388
 389         if (info->linking)
 390                 lag_upper_info = info->upper_info;
 391
 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).
 396          */
 397         rcu_read_lock();
 398         for_each_netdev_in_bond_rcu(upper, ndev_tmp) {
 399                 idx = mlx5_lag_dev_get_netdev_idx(ldev, ndev_tmp);
 400                 if (idx > -1)
 401                         bond_status |= (1 << idx);
 402
 403                 num_slaves++;
 404         }
 405         rcu_read_unlock();
 406
 407         /* None of this lagdev's netdevs are slaves of this master. */
 408         if (!(bond_status & 0x3))
 409                 return 0;
 410
 411         if (lag_upper_info)
 412                 tracker->tx_type = lag_upper_info->tx_type;
 413
 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.
 418          */
 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));
 423
 424         if (tracker->is_bonded != is_bonded) {
 425                 tracker->is_bonded = is_bonded;
 426                 return 1;
 427         }
 428
 429         return 0;
 430 }
 431
 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)
 436 {
 437         struct netdev_lag_lower_state_info *lag_lower_info;
 438         int idx;
 439
 440         if (!netif_is_lag_port(ndev))
 441                 return 0;
 442
 443         idx = mlx5_lag_dev_get_netdev_idx(ldev, ndev);
 444         if (idx == -1)
 445                 return 0;
 446
 447         /* This information is used to determine virtual to physical
 448          * port mapping.
 449          */
 450         lag_lower_info = info->lower_state_info;
 451         if (!lag_lower_info)
 452                 return 0;
 453
 454         tracker->netdev_state[idx] = *lag_lower_info;
 455
 456         return 1;
 457 }
 458
 459 static int mlx5_lag_netdev_event(struct notifier_block *this,
 460                                  unsigned long event, void *ptr)
 461 {
 462         struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
 463         struct lag_tracker tracker;
 464         struct mlx5_lag *ldev;
 465         int changed = 0;
 466
 467         if (!net_eq(dev_net(ndev), &init_net))
 468                 return NOTIFY_DONE;
 469
 470         if ((event != NETDEV_CHANGEUPPER) && (event != NETDEV_CHANGELOWERSTATE))
 471                 return NOTIFY_DONE;
 472
 473         ldev    = container_of(this, struct mlx5_lag, nb);
 474         tracker = ldev->tracker;
 475
 476         switch (event) {
 477         case NETDEV_CHANGEUPPER:
 478                 changed = mlx5_handle_changeupper_event(ldev, &tracker, ndev,
 479                                                         ptr);
 480                 break;
 481         case NETDEV_CHANGELOWERSTATE:
 482                 changed = mlx5_handle_changelowerstate_event(ldev, &tracker,
 483                                                              ndev, ptr);
 484                 break;
 485         }
 486
 487         mutex_lock(&lag_mutex);
 488         ldev->tracker = tracker;
 489         mutex_unlock(&lag_mutex);
 490
 491         if (changed)
 492                 mlx5_queue_bond_work(ldev, 0);
 493
 494         return NOTIFY_DONE;
 495 }
 496
 497 static struct mlx5_lag *mlx5_lag_dev_alloc(void)
 498 {
 499         struct mlx5_lag *ldev;
 500
 501         ldev = kzalloc(sizeof(*ldev), GFP_KERNEL);
 502         if (!ldev)
 503                 return NULL;
 504
 505         ldev->wq = create_singlethread_workqueue("mlx5_lag");
 506         if (!ldev->wq) {
 507                 kfree(ldev);
 508                 return NULL;
 509         }
 510
 511         INIT_DELAYED_WORK(&ldev->bond_work, mlx5_do_bond_work);
 512
 513         return ldev;
 514 }
 515
 516 static void mlx5_lag_dev_free(struct mlx5_lag *ldev)
 517 {
 518         destroy_workqueue(ldev->wq);
 519         kfree(ldev);
 520 }
 521
 522 static void mlx5_lag_dev_add_pf(struct mlx5_lag *ldev,
 523                                 struct mlx5_core_dev *dev,
 524                                 struct net_device *netdev)
 525 {
 526         unsigned int fn = PCI_FUNC(dev->pdev->devfn);
 527
 528         if (fn >= MLX5_MAX_PORTS)
 529                 return;
 530
 531         mutex_lock(&lag_mutex);
 532         ldev->pf[fn].dev    = dev;
 533         ldev->pf[fn].netdev = netdev;
 534         ldev->tracker.netdev_state[fn].link_up = 0;
 535         ldev->tracker.netdev_state[fn].tx_enabled = 0;
 536
 537         dev->priv.lag = ldev;
 538
 539         mutex_unlock(&lag_mutex);
 540 }
 541
 542 static void mlx5_lag_dev_remove_pf(struct mlx5_lag *ldev,
 543                                    struct mlx5_core_dev *dev)
 544 {
 545         int i;
 546
 547         for (i = 0; i < MLX5_MAX_PORTS; i++)
 548                 if (ldev->pf[i].dev == dev)
 549                         break;
 550
 551         if (i == MLX5_MAX_PORTS)
 552                 return;
 553
 554         mutex_lock(&lag_mutex);
 555         memset(&ldev->pf[i], 0, sizeof(*ldev->pf));
 556
 557         dev->priv.lag = NULL;
 558         mutex_unlock(&lag_mutex);
 559 }
 560
 561 /* Must be called with intf_mutex held */
 562 void mlx5_lag_add(struct mlx5_core_dev *dev, struct net_device *netdev)
 563 {
 564         struct mlx5_lag *ldev = NULL;
 565         struct mlx5_core_dev *tmp_dev;
 566         int err;
 567
 568         if (!MLX5_CAP_GEN(dev, vport_group_manager) ||
 569             !MLX5_CAP_GEN(dev, lag_master) ||
 570             (MLX5_CAP_GEN(dev, num_lag_ports) != MLX5_MAX_PORTS))
 571                 return;
 572
 573         tmp_dev = mlx5_get_next_phys_dev(dev);
 574         if (tmp_dev)
 575                 ldev = tmp_dev->priv.lag;
 576
 577         if (!ldev) {
 578                 ldev = mlx5_lag_dev_alloc();
 579                 if (!ldev) {
 580                         mlx5_core_err(dev, "Failed to alloc lag dev\n");
 581                         return;
 582                 }
 583         }
 584
 585         mlx5_lag_dev_add_pf(ldev, dev, netdev);
 586
 587         if (!ldev->nb.notifier_call) {
 588                 ldev->nb.notifier_call = mlx5_lag_netdev_event;
 589                 if (register_netdevice_notifier(&ldev->nb)) {
 590                         ldev->nb.notifier_call = NULL;
 591                         mlx5_core_err(dev, "Failed to register LAG netdev notifier\n");
 592                 }
 593         }
 594
 595         err = mlx5_lag_mp_init(ldev);
 596         if (err)
 597                 mlx5_core_err(dev, "Failed to init multipath lag err=%d\n",
 598                               err);
 599 }
 600
 601 /* Must be called with intf_mutex held */
 602 void mlx5_lag_remove(struct mlx5_core_dev *dev)
 603 {
 604         struct mlx5_lag *ldev;
 605         int i;
 606
 607         ldev = mlx5_lag_dev_get(dev);
 608         if (!ldev)
 609                 return;
 610
 611         if (__mlx5_lag_is_active(ldev))
 612                 mlx5_deactivate_lag(ldev);
 613
 614         mlx5_lag_dev_remove_pf(ldev, dev);
 615
 616         for (i = 0; i < MLX5_MAX_PORTS; i++)
 617                 if (ldev->pf[i].dev)
 618                         break;
 619
 620         if (i == MLX5_MAX_PORTS) {
 621                 if (ldev->nb.notifier_call)
 622                         unregister_netdevice_notifier(&ldev->nb);
 623                 mlx5_lag_mp_cleanup(ldev);
 624                 cancel_delayed_work_sync(&ldev->bond_work);
 625                 mlx5_lag_dev_free(ldev);
 626         }
 627 }
 628
 629 bool mlx5_lag_is_roce(struct mlx5_core_dev *dev)
 630 {
 631         struct mlx5_lag *ldev;
 632         bool res;
 633
 634         mutex_lock(&lag_mutex);
 635         ldev = mlx5_lag_dev_get(dev);
 636         res  = ldev && __mlx5_lag_is_roce(ldev);
 637         mutex_unlock(&lag_mutex);
 638
 639         return res;
 640 }
 641 EXPORT_SYMBOL(mlx5_lag_is_roce);
 642
 643 bool mlx5_lag_is_active(struct mlx5_core_dev *dev)
 644 {
 645         struct mlx5_lag *ldev;
 646         bool res;
 647
 648         mutex_lock(&lag_mutex);
 649         ldev = mlx5_lag_dev_get(dev);
 650         res  = ldev && __mlx5_lag_is_active(ldev);
 651         mutex_unlock(&lag_mutex);
 652
 653         return res;
 654 }
 655 EXPORT_SYMBOL(mlx5_lag_is_active);
 656
 657 bool mlx5_lag_is_sriov(struct mlx5_core_dev *dev)
 658 {
 659         struct mlx5_lag *ldev;
 660         bool res;
 661
 662         mutex_lock(&lag_mutex);
 663         ldev = mlx5_lag_dev_get(dev);
 664         res  = ldev && __mlx5_lag_is_sriov(ldev);
 665         mutex_unlock(&lag_mutex);
 666
 667         return res;
 668 }
 669 EXPORT_SYMBOL(mlx5_lag_is_sriov);
 670
 671 void mlx5_lag_update(struct mlx5_core_dev *dev)
 672 {
 673         struct mlx5_lag *ldev;
 674
 675         mlx5_dev_list_lock();
 676         ldev = mlx5_lag_dev_get(dev);
 677         if (!ldev)
 678                 goto unlock;
 679
 680         mlx5_do_bond(ldev);
 681
 682 unlock:
 683         mlx5_dev_list_unlock();
 684 }
 685
 686 struct net_device *mlx5_lag_get_roce_netdev(struct mlx5_core_dev *dev)
 687 {
 688         struct net_device *ndev = NULL;
 689         struct mlx5_lag *ldev;
 690
 691         mutex_lock(&lag_mutex);
 692         ldev = mlx5_lag_dev_get(dev);
 693
 694         if (!(ldev && __mlx5_lag_is_roce(ldev)))
 695                 goto unlock;
 696
 697         if (ldev->tracker.tx_type == NETDEV_LAG_TX_TYPE_ACTIVEBACKUP) {
 698                 ndev = ldev->tracker.netdev_state[MLX5_LAG_P1].tx_enabled ?
 699                        ldev->pf[MLX5_LAG_P1].netdev :
 700                        ldev->pf[MLX5_LAG_P2].netdev;
 701         } else {
 702                 ndev = ldev->pf[MLX5_LAG_P1].netdev;
 703         }
 704         if (ndev)
 705                 dev_hold(ndev);
 706
 707 unlock:
 708         mutex_unlock(&lag_mutex);
 709
 710         return ndev;
 711 }
 712 EXPORT_SYMBOL(mlx5_lag_get_roce_netdev);
 713
 714 bool mlx5_lag_intf_add(struct mlx5_interface *intf, struct mlx5_priv *priv)
 715 {
 716         struct mlx5_core_dev *dev = container_of(priv, struct mlx5_core_dev,
 717                                                  priv);
 718         struct mlx5_lag *ldev;
 719
 720         if (intf->protocol != MLX5_INTERFACE_PROTOCOL_IB)
 721                 return true;
 722
 723         ldev = mlx5_lag_dev_get(dev);
 724         if (!ldev || !__mlx5_lag_is_roce(ldev) ||
 725             ldev->pf[MLX5_LAG_P1].dev == dev)
 726                 return true;
 727
 728         /* If bonded, we do not add an IB device for PF1. */
 729         return false;
 730 }
 731
 732 int mlx5_lag_query_cong_counters(struct mlx5_core_dev *dev,
 733                                  u64 *values,
 734                                  int num_counters,
 735                                  size_t *offsets)
 736 {
 737         int outlen = MLX5_ST_SZ_BYTES(query_cong_statistics_out);
 738         struct mlx5_core_dev *mdev[MLX5_MAX_PORTS];
 739         struct mlx5_lag *ldev;
 740         int num_ports;
 741         int ret, i, j;
 742         void *out;
 743
 744         out = kvzalloc(outlen, GFP_KERNEL);
 745         if (!out)
 746                 return -ENOMEM;
 747
 748         memset(values, 0, sizeof(*values) * num_counters);
 749
 750         mutex_lock(&lag_mutex);
 751         ldev = mlx5_lag_dev_get(dev);
 752         if (ldev && __mlx5_lag_is_roce(ldev)) {
 753                 num_ports = MLX5_MAX_PORTS;
 754                 mdev[MLX5_LAG_P1] = ldev->pf[MLX5_LAG_P1].dev;
 755                 mdev[MLX5_LAG_P2] = ldev->pf[MLX5_LAG_P2].dev;
 756         } else {
 757                 num_ports = 1;
 758                 mdev[MLX5_LAG_P1] = dev;
 759         }
 760
 761         for (i = 0; i < num_ports; ++i) {
 762                 ret = mlx5_cmd_query_cong_counter(mdev[i], false, out, outlen);
 763                 if (ret)
 764                         goto unlock;
 765
 766                 for (j = 0; j < num_counters; ++j)
 767                         values[j] += be64_to_cpup((__be64 *)(out + offsets[j]));
 768         }
 769
 770 unlock:
 771         mutex_unlock(&lag_mutex);
 772         kvfree(out);
 773         return ret;
 774 }
 775 EXPORT_SYMBOL(mlx5_lag_query_cong_counters);