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
  38 enum {
  39         MLX5_LAG_FLAG_BONDED = 1 << 0,
  40 };
  41
  42 struct lag_func {
  43         struct mlx5_core_dev *dev;
  44         struct net_device    *netdev;
  45 };
  46
  47 /* Used for collection of netdev event info. */
  48 struct lag_tracker {
  49         enum   netdev_lag_tx_type           tx_type;
  50         struct netdev_lag_lower_state_info  netdev_state[MLX5_MAX_PORTS];
  51         bool is_bonded;
  52 };
  53
  54 /* LAG data of a ConnectX card.
  55  * It serves both its phys functions.
  56  */
  57 struct mlx5_lag {
  58         u8                        flags;
  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;
  64 };
  65
  66 /* General purpose, use for short periods of time.
  67  * Beware of lock dependencies (preferably, no locks should be acquired
  68  * under it).
  69  */
  70 static DEFINE_MUTEX(lag_mutex);
  71
  72 static int mlx5_cmd_create_lag(struct mlx5_core_dev *dev, u8 remap_port1,
  73                                u8 remap_port2)
  74 {
  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);
  78
  79         MLX5_SET(create_lag_in, in, opcode, MLX5_CMD_OP_CREATE_LAG);
  80
  81         MLX5_SET(lagc, lag_ctx, tx_remap_affinity_1, remap_port1);
  82         MLX5_SET(lagc, lag_ctx, tx_remap_affinity_2, remap_port2);
  83
  84         return mlx5_cmd_exec(dev, in, sizeof(in), out, sizeof(out));
  85 }
  86
  87 static int mlx5_cmd_modify_lag(struct mlx5_core_dev *dev, u8 remap_port1,
  88                                u8 remap_port2)
  89 {
  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);
  93
  94         MLX5_SET(modify_lag_in, in, opcode, MLX5_CMD_OP_MODIFY_LAG);
  95         MLX5_SET(modify_lag_in, in, field_select, 0x1);
  96
  97         MLX5_SET(lagc, lag_ctx, tx_remap_affinity_1, remap_port1);
  98         MLX5_SET(lagc, lag_ctx, tx_remap_affinity_2, remap_port2);
  99
 100         return mlx5_cmd_exec(dev, in, sizeof(in), out, sizeof(out));
 101 }
 102
 103 static int mlx5_cmd_destroy_lag(struct mlx5_core_dev *dev)
 104 {
 105         u32  in[MLX5_ST_SZ_DW(destroy_lag_in)]  = {0};
 106         u32 out[MLX5_ST_SZ_DW(destroy_lag_out)] = {0};
 107
 108         MLX5_SET(destroy_lag_in, in, opcode, MLX5_CMD_OP_DESTROY_LAG);
 109
 110         return mlx5_cmd_exec(dev, in, sizeof(in), out, sizeof(out));
 111 }
 112
 113 int mlx5_cmd_create_vport_lag(struct mlx5_core_dev *dev)
 114 {
 115         u32  in[MLX5_ST_SZ_DW(create_vport_lag_in)]  = {0};
 116         u32 out[MLX5_ST_SZ_DW(create_vport_lag_out)] = {0};
 117
 118         MLX5_SET(create_vport_lag_in, in, opcode, MLX5_CMD_OP_CREATE_VPORT_LAG);
 119
 120         return mlx5_cmd_exec(dev, in, sizeof(in), out, sizeof(out));
 121 }
 122 EXPORT_SYMBOL(mlx5_cmd_create_vport_lag);
 123
 124 int mlx5_cmd_destroy_vport_lag(struct mlx5_core_dev *dev)
 125 {
 126         u32  in[MLX5_ST_SZ_DW(destroy_vport_lag_in)]  = {0};
 127         u32 out[MLX5_ST_SZ_DW(destroy_vport_lag_out)] = {0};
 128
 129         MLX5_SET(destroy_vport_lag_in, in, opcode, MLX5_CMD_OP_DESTROY_VPORT_LAG);
 130
 131         return mlx5_cmd_exec(dev, in, sizeof(in), out, sizeof(out));
 132 }
 133 EXPORT_SYMBOL(mlx5_cmd_destroy_vport_lag);
 134
 135 static struct mlx5_lag *mlx5_lag_dev_get(struct mlx5_core_dev *dev)
 136 {
 137         return dev->priv.lag;
 138 }
 139
 140 static int mlx5_lag_dev_get_netdev_idx(struct mlx5_lag *ldev,
 141                                        struct net_device *ndev)
 142 {
 143         int i;
 144
 145         for (i = 0; i < MLX5_MAX_PORTS; i++)
 146                 if (ldev->pf[i].netdev == ndev)
 147                         return i;
 148
 149         return -1;
 150 }
 151
 152 static bool mlx5_lag_is_bonded(struct mlx5_lag *ldev)
 153 {
 154         return !!(ldev->flags & MLX5_LAG_FLAG_BONDED);
 155 }
 156
 157 static void mlx5_infer_tx_affinity_mapping(struct lag_tracker *tracker,
 158                                            u8 *port1, u8 *port2)
 159 {
 160         if (tracker->tx_type == NETDEV_LAG_TX_TYPE_ACTIVEBACKUP) {
 161                 if (tracker->netdev_state[0].tx_enabled) {
 162                         *port1 = 1;
 163                         *port2 = 1;
 164                 } else {
 165                         *port1 = 2;
 166                         *port2 = 2;
 167                 }
 168         } else {
 169                 *port1 = 1;
 170                 *port2 = 2;
 171                 if (!tracker->netdev_state[0].link_up)
 172                         *port1 = 2;
 173                 else if (!tracker->netdev_state[1].link_up)
 174                         *port2 = 1;
 175         }
 176 }
 177
 178 static void mlx5_activate_lag(struct mlx5_lag *ldev,
 179                               struct lag_tracker *tracker)
 180 {
 181         struct mlx5_core_dev *dev0 = ldev->pf[0].dev;
 182         int err;
 183
 184         ldev->flags |= MLX5_LAG_FLAG_BONDED;
 185
 186         mlx5_infer_tx_affinity_mapping(tracker, &ldev->v2p_map[0],
 187                                        &ldev->v2p_map[1]);
 188
 189         err = mlx5_cmd_create_lag(dev0, ldev->v2p_map[0], ldev->v2p_map[1]);
 190         if (err)
 191                 mlx5_core_err(dev0,
 192                               "Failed to create LAG (%d)\n",
 193                               err);
 194 }
 195
 196 static void mlx5_deactivate_lag(struct mlx5_lag *ldev)
 197 {
 198         struct mlx5_core_dev *dev0 = ldev->pf[0].dev;
 199         int err;
 200
 201         ldev->flags &= ~MLX5_LAG_FLAG_BONDED;
 202
 203         err = mlx5_cmd_destroy_lag(dev0);
 204         if (err)
 205                 mlx5_core_err(dev0,
 206                               "Failed to destroy LAG (%d)\n",
 207                               err);
 208 }
 209
 210 static void mlx5_do_bond(struct mlx5_lag *ldev)
 211 {
 212         struct mlx5_core_dev *dev0 = ldev->pf[0].dev;
 213         struct mlx5_core_dev *dev1 = ldev->pf[1].dev;
 214         struct lag_tracker tracker;
 215         u8 v2p_port1, v2p_port2;
 216         int i, err;
 217
 218         if (!dev0 || !dev1)
 219                 return;
 220
 221         mutex_lock(&lag_mutex);
 222         tracker = ldev->tracker;
 223         mutex_unlock(&lag_mutex);
 224
 225         if (tracker.is_bonded && !mlx5_lag_is_bonded(ldev)) {
 226                 if (mlx5_sriov_is_enabled(dev0) ||
 227                     mlx5_sriov_is_enabled(dev1)) {
 228                         mlx5_core_warn(dev0, "LAG is not supported with SRIOV");
 229                         return;
 230                 }
 231
 232                 for (i = 0; i < MLX5_MAX_PORTS; i++)
 233                         mlx5_remove_dev_by_protocol(ldev->pf[i].dev,
 234                                                     MLX5_INTERFACE_PROTOCOL_IB);
 235
 236                 mlx5_activate_lag(ldev, &tracker);
 237
 238                 mlx5_add_dev_by_protocol(dev0, MLX5_INTERFACE_PROTOCOL_IB);
 239                 mlx5_nic_vport_enable_roce(dev1);
 240         } else if (tracker.is_bonded && mlx5_lag_is_bonded(ldev)) {
 241                 mlx5_infer_tx_affinity_mapping(&tracker, &v2p_port1,
 242                                                &v2p_port2);
 243
 244                 if ((v2p_port1 != ldev->v2p_map[0]) ||
 245                     (v2p_port2 != ldev->v2p_map[1])) {
 246                         ldev->v2p_map[0] = v2p_port1;
 247                         ldev->v2p_map[1] = v2p_port2;
 248
 249                         err = mlx5_cmd_modify_lag(dev0, v2p_port1, v2p_port2);
 250                         if (err)
 251                                 mlx5_core_err(dev0,
 252                                               "Failed to modify LAG (%d)\n",
 253                                               err);
 254                 }
 255         } else if (!tracker.is_bonded && mlx5_lag_is_bonded(ldev)) {
 256                 mlx5_remove_dev_by_protocol(dev0, MLX5_INTERFACE_PROTOCOL_IB);
 257                 mlx5_nic_vport_disable_roce(dev1);
 258
 259                 mlx5_deactivate_lag(ldev);
 260
 261                 for (i = 0; i < MLX5_MAX_PORTS; i++)
 262                         if (ldev->pf[i].dev)
 263                                 mlx5_add_dev_by_protocol(ldev->pf[i].dev,
 264                                                          MLX5_INTERFACE_PROTOCOL_IB);
 265         }
 266 }
 267
 268 static void mlx5_queue_bond_work(struct mlx5_lag *ldev, unsigned long delay)
 269 {
 270         schedule_delayed_work(&ldev->bond_work, delay);
 271 }
 272
 273 static void mlx5_do_bond_work(struct work_struct *work)
 274 {
 275         struct delayed_work *delayed_work = to_delayed_work(work);
 276         struct mlx5_lag *ldev = container_of(delayed_work, struct mlx5_lag,
 277                                              bond_work);
 278         int status;
 279
 280         status = mlx5_dev_list_trylock();
 281         if (!status) {
 282                 /* 1 sec delay. */
 283                 mlx5_queue_bond_work(ldev, HZ);
 284                 return;
 285         }
 286
 287         mlx5_do_bond(ldev);
 288         mlx5_dev_list_unlock();
 289 }
 290
 291 static int mlx5_handle_changeupper_event(struct mlx5_lag *ldev,
 292                                          struct lag_tracker *tracker,
 293                                          struct net_device *ndev,
 294                                          struct netdev_notifier_changeupper_info *info)
 295 {
 296         struct net_device *upper = info->upper_dev, *ndev_tmp;
 297         struct netdev_lag_upper_info *lag_upper_info = NULL;
 298         bool is_bonded;
 299         int bond_status = 0;
 300         int num_slaves = 0;
 301         int idx;
 302
 303         if (!netif_is_lag_master(upper))
 304                 return 0;
 305
 306         if (info->linking)
 307                 lag_upper_info = info->upper_info;
 308
 309         /* The event may still be of interest if the slave does not belong to
 310          * us, but is enslaved to a master which has one or more of our netdevs
 311          * as slaves (e.g., if a new slave is added to a master that bonds two
 312          * of our netdevs, we should unbond).
 313          */
 314         rcu_read_lock();
 315         for_each_netdev_in_bond_rcu(upper, ndev_tmp) {
 316                 idx = mlx5_lag_dev_get_netdev_idx(ldev, ndev_tmp);
 317                 if (idx > -1)
 318                         bond_status |= (1 << idx);
 319
 320                 num_slaves++;
 321         }
 322         rcu_read_unlock();
 323
 324         /* None of this lagdev's netdevs are slaves of this master. */
 325         if (!(bond_status & 0x3))
 326                 return 0;
 327
 328         if (lag_upper_info)
 329                 tracker->tx_type = lag_upper_info->tx_type;
 330
 331         /* Determine bonding status:
 332          * A device is considered bonded if both its physical ports are slaves
 333          * of the same lag master, and only them.
 334          * Lag mode must be activebackup or hash.
 335          */
 336         is_bonded = (num_slaves == MLX5_MAX_PORTS) &&
 337                     (bond_status == 0x3) &&
 338                     ((tracker->tx_type == NETDEV_LAG_TX_TYPE_ACTIVEBACKUP) ||
 339                      (tracker->tx_type == NETDEV_LAG_TX_TYPE_HASH));
 340
 341         if (tracker->is_bonded != is_bonded) {
 342                 tracker->is_bonded = is_bonded;
 343                 return 1;
 344         }
 345
 346         return 0;
 347 }
 348
 349 static int mlx5_handle_changelowerstate_event(struct mlx5_lag *ldev,
 350                                               struct lag_tracker *tracker,
 351                                               struct net_device *ndev,
 352                                               struct netdev_notifier_changelowerstate_info *info)
 353 {
 354         struct netdev_lag_lower_state_info *lag_lower_info;
 355         int idx;
 356
 357         if (!netif_is_lag_port(ndev))
 358                 return 0;
 359
 360         idx = mlx5_lag_dev_get_netdev_idx(ldev, ndev);
 361         if (idx == -1)
 362                 return 0;
 363
 364         /* This information is used to determine virtual to physical
 365          * port mapping.
 366          */
 367         lag_lower_info = info->lower_state_info;
 368         if (!lag_lower_info)
 369                 return 0;
 370
 371         tracker->netdev_state[idx] = *lag_lower_info;
 372
 373         return 1;
 374 }
 375
 376 static int mlx5_lag_netdev_event(struct notifier_block *this,
 377                                  unsigned long event, void *ptr)
 378 {
 379         struct net_device *ndev = netdev_notifier_info_to_dev(ptr);
 380         struct lag_tracker tracker;
 381         struct mlx5_lag *ldev;
 382         int changed = 0;
 383
 384         if (!net_eq(dev_net(ndev), &init_net))
 385                 return NOTIFY_DONE;
 386
 387         if ((event != NETDEV_CHANGEUPPER) && (event != NETDEV_CHANGELOWERSTATE))
 388                 return NOTIFY_DONE;
 389
 390         ldev    = container_of(this, struct mlx5_lag, nb);
 391         tracker = ldev->tracker;
 392
 393         switch (event) {
 394         case NETDEV_CHANGEUPPER:
 395                 changed = mlx5_handle_changeupper_event(ldev, &tracker, ndev,
 396                                                         ptr);
 397                 break;
 398         case NETDEV_CHANGELOWERSTATE:
 399                 changed = mlx5_handle_changelowerstate_event(ldev, &tracker,
 400                                                              ndev, ptr);
 401                 break;
 402         }
 403
 404         mutex_lock(&lag_mutex);
 405         ldev->tracker = tracker;
 406         mutex_unlock(&lag_mutex);
 407
 408         if (changed)
 409                 mlx5_queue_bond_work(ldev, 0);
 410
 411         return NOTIFY_DONE;
 412 }
 413
 414 static struct mlx5_lag *mlx5_lag_dev_alloc(void)
 415 {
 416         struct mlx5_lag *ldev;
 417
 418         ldev = kzalloc(sizeof(*ldev), GFP_KERNEL);
 419         if (!ldev)
 420                 return NULL;
 421
 422         INIT_DELAYED_WORK(&ldev->bond_work, mlx5_do_bond_work);
 423
 424         return ldev;
 425 }
 426
 427 static void mlx5_lag_dev_free(struct mlx5_lag *ldev)
 428 {
 429         kfree(ldev);
 430 }
 431
 432 static void mlx5_lag_dev_add_pf(struct mlx5_lag *ldev,
 433                                 struct mlx5_core_dev *dev,
 434                                 struct net_device *netdev)
 435 {
 436         unsigned int fn = PCI_FUNC(dev->pdev->devfn);
 437
 438         if (fn >= MLX5_MAX_PORTS)
 439                 return;
 440
 441         mutex_lock(&lag_mutex);
 442         ldev->pf[fn].dev    = dev;
 443         ldev->pf[fn].netdev = netdev;
 444         ldev->tracker.netdev_state[fn].link_up = 0;
 445         ldev->tracker.netdev_state[fn].tx_enabled = 0;
 446
 447         dev->priv.lag = ldev;
 448         mutex_unlock(&lag_mutex);
 449 }
 450
 451 static void mlx5_lag_dev_remove_pf(struct mlx5_lag *ldev,
 452                                    struct mlx5_core_dev *dev)
 453 {
 454         int i;
 455
 456         for (i = 0; i < MLX5_MAX_PORTS; i++)
 457                 if (ldev->pf[i].dev == dev)
 458                         break;
 459
 460         if (i == MLX5_MAX_PORTS)
 461                 return;
 462
 463         mutex_lock(&lag_mutex);
 464         memset(&ldev->pf[i], 0, sizeof(*ldev->pf));
 465
 466         dev->priv.lag = NULL;
 467         mutex_unlock(&lag_mutex);
 468 }
 469
 470
 471 /* Must be called with intf_mutex held */
 472 void mlx5_lag_add(struct mlx5_core_dev *dev, struct net_device *netdev)
 473 {
 474         struct mlx5_lag *ldev = NULL;
 475         struct mlx5_core_dev *tmp_dev;
 476
 477         if (!MLX5_CAP_GEN(dev, vport_group_manager) ||
 478             !MLX5_CAP_GEN(dev, lag_master) ||
 479             (MLX5_CAP_GEN(dev, num_lag_ports) != MLX5_MAX_PORTS))
 480                 return;
 481
 482         tmp_dev = mlx5_get_next_phys_dev(dev);
 483         if (tmp_dev)
 484                 ldev = tmp_dev->priv.lag;
 485
 486         if (!ldev) {
 487                 ldev = mlx5_lag_dev_alloc();
 488                 if (!ldev) {
 489                         mlx5_core_err(dev, "Failed to alloc lag dev\n");
 490                         return;
 491                 }
 492         }
 493
 494         mlx5_lag_dev_add_pf(ldev, dev, netdev);
 495
 496         if (!ldev->nb.notifier_call) {
 497                 ldev->nb.notifier_call = mlx5_lag_netdev_event;
 498                 if (register_netdevice_notifier(&ldev->nb)) {
 499                         ldev->nb.notifier_call = NULL;
 500                         mlx5_core_err(dev, "Failed to register LAG netdev notifier\n");
 501                 }
 502         }
 503 }
 504
 505 /* Must be called with intf_mutex held */
 506 void mlx5_lag_remove(struct mlx5_core_dev *dev)
 507 {
 508         struct mlx5_lag *ldev;
 509         int i;
 510
 511         ldev = mlx5_lag_dev_get(dev);
 512         if (!ldev)
 513                 return;
 514
 515         if (mlx5_lag_is_bonded(ldev))
 516                 mlx5_deactivate_lag(ldev);
 517
 518         mlx5_lag_dev_remove_pf(ldev, dev);
 519
 520         for (i = 0; i < MLX5_MAX_PORTS; i++)
 521                 if (ldev->pf[i].dev)
 522                         break;
 523
 524         if (i == MLX5_MAX_PORTS) {
 525                 if (ldev->nb.notifier_call)
 526                         unregister_netdevice_notifier(&ldev->nb);
 527                 cancel_delayed_work_sync(&ldev->bond_work);
 528                 mlx5_lag_dev_free(ldev);
 529         }
 530 }
 531
 532 bool mlx5_lag_is_active(struct mlx5_core_dev *dev)
 533 {
 534         struct mlx5_lag *ldev;
 535         bool res;
 536
 537         mutex_lock(&lag_mutex);
 538         ldev = mlx5_lag_dev_get(dev);
 539         res  = ldev && mlx5_lag_is_bonded(ldev);
 540         mutex_unlock(&lag_mutex);
 541
 542         return res;
 543 }
 544 EXPORT_SYMBOL(mlx5_lag_is_active);
 545
 546 struct net_device *mlx5_lag_get_roce_netdev(struct mlx5_core_dev *dev)
 547 {
 548         struct net_device *ndev = NULL;
 549         struct mlx5_lag *ldev;
 550
 551         mutex_lock(&lag_mutex);
 552         ldev = mlx5_lag_dev_get(dev);
 553
 554         if (!(ldev && mlx5_lag_is_bonded(ldev)))
 555                 goto unlock;
 556
 557         if (ldev->tracker.tx_type == NETDEV_LAG_TX_TYPE_ACTIVEBACKUP) {
 558                 ndev = ldev->tracker.netdev_state[0].tx_enabled ?
 559                        ldev->pf[0].netdev : ldev->pf[1].netdev;
 560         } else {
 561                 ndev = ldev->pf[0].netdev;
 562         }
 563         if (ndev)
 564                 dev_hold(ndev);
 565
 566 unlock:
 567         mutex_unlock(&lag_mutex);
 568
 569         return ndev;
 570 }
 571 EXPORT_SYMBOL(mlx5_lag_get_roce_netdev);
 572
 573 bool mlx5_lag_intf_add(struct mlx5_interface *intf, struct mlx5_priv *priv)
 574 {
 575         struct mlx5_core_dev *dev = container_of(priv, struct mlx5_core_dev,
 576                                                  priv);
 577         struct mlx5_lag *ldev;
 578
 579         if (intf->protocol != MLX5_INTERFACE_PROTOCOL_IB)
 580                 return true;
 581
 582         ldev = mlx5_lag_dev_get(dev);
 583         if (!ldev || !mlx5_lag_is_bonded(ldev) || ldev->pf[0].dev == dev)
 584                 return true;
 585
 586         /* If bonded, we do not add an IB device for PF1. */
 587         return false;
 588 }
 589