drivers/net/bonding/rte_eth_bond_api.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2017 Intel Corporation
   3  */
   4
   5 #include <string.h>
   6
   7 #include <rte_mbuf.h>
   8 #include <rte_malloc.h>
   9 #include <rte_ethdev_driver.h>
  10 #include <rte_tcp.h>
  11 #include <rte_bus_vdev.h>
  12 #include <rte_kvargs.h>
  13
  14 #include "rte_eth_bond.h"
  15 #include "eth_bond_private.h"
  16 #include "eth_bond_8023ad_private.h"
  17
  18 int
  19 check_for_bonded_ethdev(const struct rte_eth_dev *eth_dev)
  20 {
  21         /* Check valid pointer */
  22         if (eth_dev == NULL ||
  23                 eth_dev->device == NULL ||
  24                 eth_dev->device->driver == NULL ||
  25                 eth_dev->device->driver->name == NULL)
  26                 return -1;
  27
  28         /* return 0 if driver name matches */
  29         return eth_dev->device->driver->name != pmd_bond_drv.driver.name;
  30 }
  31
  32 int
  33 valid_bonded_port_id(uint16_t port_id)
  34 {
  35         RTE_ETH_VALID_PORTID_OR_ERR_RET(port_id, -1);
  36         return check_for_bonded_ethdev(&rte_eth_devices[port_id]);
  37 }
  38
  39 int
  40 check_for_master_bonded_ethdev(const struct rte_eth_dev *eth_dev)
  41 {
  42         int i;
  43         struct bond_dev_private *internals;
  44
  45         if (check_for_bonded_ethdev(eth_dev) != 0)
  46                 return 0;
  47
  48         internals = eth_dev->data->dev_private;
  49
  50         /* Check if any of slave devices is a bonded device */
  51         for (i = 0; i < internals->slave_count; i++)
  52                 if (valid_bonded_port_id(internals->slaves[i].port_id) == 0)
  53                         return 1;
  54
  55         return 0;
  56 }
  57
  58 int
  59 valid_slave_port_id(uint16_t port_id, uint8_t mode)
  60 {
  61         RTE_ETH_VALID_PORTID_OR_ERR_RET(port_id, -1);
  62
  63         /* Verify that port_id refers to a non bonded port */
  64         if (check_for_bonded_ethdev(&rte_eth_devices[port_id]) == 0 &&
  65                         mode == BONDING_MODE_8023AD) {
  66                 RTE_BOND_LOG(ERR, "Cannot add slave to bonded device in 802.3ad"
  67                                 " mode as slave is also a bonded device, only "
  68                                 "physical devices can be support in this mode.");
  69                 return -1;
  70         }
  71
  72         return 0;
  73 }
  74
  75 void
  76 activate_slave(struct rte_eth_dev *eth_dev, uint16_t port_id)
  77 {
  78         struct bond_dev_private *internals = eth_dev->data->dev_private;
  79         uint16_t active_count = internals->active_slave_count;
  80
  81         if (internals->mode == BONDING_MODE_8023AD)
  82                 bond_mode_8023ad_activate_slave(eth_dev, port_id);
  83
  84         if (internals->mode == BONDING_MODE_TLB
  85                         || internals->mode == BONDING_MODE_ALB) {
  86
  87                 internals->tlb_slaves_order[active_count] = port_id;
  88         }
  89
  90         RTE_ASSERT(internals->active_slave_count <
  91                         (RTE_DIM(internals->active_slaves) - 1));
  92
  93         internals->active_slaves[internals->active_slave_count] = port_id;
  94         internals->active_slave_count++;
  95
  96         if (internals->mode == BONDING_MODE_TLB)
  97                 bond_tlb_activate_slave(internals);
  98         if (internals->mode == BONDING_MODE_ALB)
  99                 bond_mode_alb_client_list_upd(eth_dev);
 100 }
 101
 102 void
 103 deactivate_slave(struct rte_eth_dev *eth_dev, uint16_t port_id)
 104 {
 105         uint16_t slave_pos;
 106         struct bond_dev_private *internals = eth_dev->data->dev_private;
 107         uint16_t active_count = internals->active_slave_count;
 108
 109         if (internals->mode == BONDING_MODE_8023AD) {
 110                 bond_mode_8023ad_stop(eth_dev);
 111                 bond_mode_8023ad_deactivate_slave(eth_dev, port_id);
 112         } else if (internals->mode == BONDING_MODE_TLB
 113                         || internals->mode == BONDING_MODE_ALB)
 114                 bond_tlb_disable(internals);
 115
 116         slave_pos = find_slave_by_id(internals->active_slaves, active_count,
 117                         port_id);
 118
 119         /* If slave was not at the end of the list
 120          * shift active slaves up active array list */
 121         if (slave_pos < active_count) {
 122                 active_count--;
 123                 memmove(internals->active_slaves + slave_pos,
 124                                 internals->active_slaves + slave_pos + 1,
 125                                 (active_count - slave_pos) *
 126                                         sizeof(internals->active_slaves[0]));
 127         }
 128
 129         RTE_ASSERT(active_count < RTE_DIM(internals->active_slaves));
 130         internals->active_slave_count = active_count;
 131
 132         if (eth_dev->data->dev_started) {
 133                 if (internals->mode == BONDING_MODE_8023AD) {
 134                         bond_mode_8023ad_start(eth_dev);
 135                 } else if (internals->mode == BONDING_MODE_TLB) {
 136                         bond_tlb_enable(internals);
 137                 } else if (internals->mode == BONDING_MODE_ALB) {
 138                         bond_tlb_enable(internals);
 139                         bond_mode_alb_client_list_upd(eth_dev);
 140                 }
 141         }
 142 }
 143
 144 int
 145 rte_eth_bond_create(const char *name, uint8_t mode, uint8_t socket_id)
 146 {
 147         struct bond_dev_private *internals;
 148         char devargs[52];
 149         uint16_t port_id;
 150         int ret;
 151
 152         if (name == NULL) {
 153                 RTE_BOND_LOG(ERR, "Invalid name specified");
 154                 return -EINVAL;
 155         }
 156
 157         ret = snprintf(devargs, sizeof(devargs),
 158                 "driver=net_bonding,mode=%d,socket_id=%d", mode, socket_id);
 159         if (ret < 0 || ret >= (int)sizeof(devargs))
 160                 return -ENOMEM;
 161
 162         ret = rte_vdev_init(name, devargs);
 163         if (ret)
 164                 return ret;
 165
 166         ret = rte_eth_dev_get_port_by_name(name, &port_id);
 167         RTE_ASSERT(!ret);
 168
 169         /*
 170          * To make bond_ethdev_configure() happy we need to free the
 171          * internals->kvlist here.
 172          *
 173          * Also see comment in bond_ethdev_configure().
 174          */
 175         internals = rte_eth_devices[port_id].data->dev_private;
 176         rte_kvargs_free(internals->kvlist);
 177         internals->kvlist = NULL;
 178
 179         return port_id;
 180 }
 181
 182 int
 183 rte_eth_bond_free(const char *name)
 184 {
 185         return rte_vdev_uninit(name);
 186 }
 187
 188 static int
 189 slave_vlan_filter_set(uint16_t bonded_port_id, uint16_t slave_port_id)
 190 {
 191         struct rte_eth_dev *bonded_eth_dev;
 192         struct bond_dev_private *internals;
 193         int found;
 194         int res = 0;
 195         uint64_t slab = 0;
 196         uint32_t pos = 0;
 197         uint16_t first;
 198
 199         bonded_eth_dev = &rte_eth_devices[bonded_port_id];
 200         if ((bonded_eth_dev->data->dev_conf.rxmode.offloads &
 201                         DEV_RX_OFFLOAD_VLAN_FILTER) == 0)
 202                 return 0;
 203
 204         internals = bonded_eth_dev->data->dev_private;
 205         found = rte_bitmap_scan(internals->vlan_filter_bmp, &pos, &slab);
 206         first = pos;
 207
 208         if (!found)
 209                 return 0;
 210
 211         do {
 212                 uint32_t i;
 213                 uint64_t mask;
 214
 215                 for (i = 0, mask = 1;
 216                      i < RTE_BITMAP_SLAB_BIT_SIZE;
 217                      i ++, mask <<= 1) {
 218                         if (unlikely(slab & mask)) {
 219                                 uint16_t vlan_id = pos + i;
 220
 221                                 res = rte_eth_dev_vlan_filter(slave_port_id,
 222                                                               vlan_id, 1);
 223                         }
 224                 }
 225                 found = rte_bitmap_scan(internals->vlan_filter_bmp,
 226                                         &pos, &slab);
 227         } while (found && first != pos && res == 0);
 228
 229         return res;
 230 }
 231
 232 static int
 233 slave_rte_flow_prepare(uint16_t slave_id, struct bond_dev_private *internals)
 234 {
 235         struct rte_flow *flow;
 236         struct rte_flow_error ferror;
 237         uint16_t slave_port_id = internals->slaves[slave_id].port_id;
 238
 239         if (internals->flow_isolated_valid != 0) {
 240                 rte_eth_dev_stop(slave_port_id);
 241                 if (rte_flow_isolate(slave_port_id, internals->flow_isolated,
 242                     &ferror)) {
 243                         RTE_BOND_LOG(ERR, "rte_flow_isolate failed for slave"
 244                                      " %d: %s", slave_id, ferror.message ?
 245                                      ferror.message : "(no stated reason)");
 246                         return -1;
 247                 }
 248         }
 249         TAILQ_FOREACH(flow, &internals->flow_list, next) {
 250                 flow->flows[slave_id] = rte_flow_create(slave_port_id,
 251                                                         flow->rule.attr,
 252                                                         flow->rule.pattern,
 253                                                         flow->rule.actions,
 254                                                         &ferror);
 255                 if (flow->flows[slave_id] == NULL) {
 256                         RTE_BOND_LOG(ERR, "Cannot create flow for slave"
 257                                      " %d: %s", slave_id,
 258                                      ferror.message ? ferror.message :
 259                                      "(no stated reason)");
 260                         /* Destroy successful bond flows from the slave */
 261                         TAILQ_FOREACH(flow, &internals->flow_list, next) {
 262                                 if (flow->flows[slave_id] != NULL) {
 263                                         rte_flow_destroy(slave_port_id,
 264                                                          flow->flows[slave_id],
 265                                                          &ferror);
 266                                         flow->flows[slave_id] = NULL;
 267                                 }
 268                         }
 269                         return -1;
 270                 }
 271         }
 272         return 0;
 273 }
 274
 275 static void
 276 eth_bond_slave_inherit_dev_info_rx_first(struct bond_dev_private *internals,
 277                                          const struct rte_eth_dev_info *di)
 278 {
 279         struct rte_eth_rxconf *rxconf_i = &internals->default_rxconf;
 280
 281         internals->reta_size = di->reta_size;
 282
 283         /* Inherit Rx offload capabilities from the first slave device */
 284         internals->rx_offload_capa = di->rx_offload_capa;
 285         internals->rx_queue_offload_capa = di->rx_queue_offload_capa;
 286         internals->flow_type_rss_offloads = di->flow_type_rss_offloads;
 287
 288         /* Inherit maximum Rx packet size from the first slave device */
 289         internals->candidate_max_rx_pktlen = di->max_rx_pktlen;
 290
 291         /* Inherit default Rx queue settings from the first slave device */
 292         memcpy(rxconf_i, &di->default_rxconf, sizeof(*rxconf_i));
 293
 294         /*
 295          * Turn off descriptor prefetch and writeback by default for all
 296          * slave devices. Applications may tweak this setting if need be.
 297          */
 298         rxconf_i->rx_thresh.pthresh = 0;
 299         rxconf_i->rx_thresh.hthresh = 0;
 300         rxconf_i->rx_thresh.wthresh = 0;
 301
 302         /* Setting this to zero should effectively enable default values */
 303         rxconf_i->rx_free_thresh = 0;
 304
 305         /* Disable deferred start by default for all slave devices */
 306         rxconf_i->rx_deferred_start = 0;
 307 }
 308
 309 static void
 310 eth_bond_slave_inherit_dev_info_tx_first(struct bond_dev_private *internals,
 311                                          const struct rte_eth_dev_info *di)
 312 {
 313         struct rte_eth_txconf *txconf_i = &internals->default_txconf;
 314
 315         /* Inherit Tx offload capabilities from the first slave device */
 316         internals->tx_offload_capa = di->tx_offload_capa;
 317         internals->tx_queue_offload_capa = di->tx_queue_offload_capa;
 318
 319         /* Inherit default Tx queue settings from the first slave device */
 320         memcpy(txconf_i, &di->default_txconf, sizeof(*txconf_i));
 321
 322         /*
 323          * Turn off descriptor prefetch and writeback by default for all
 324          * slave devices. Applications may tweak this setting if need be.
 325          */
 326         txconf_i->tx_thresh.pthresh = 0;
 327         txconf_i->tx_thresh.hthresh = 0;
 328         txconf_i->tx_thresh.wthresh = 0;
 329
 330         /*
 331          * Setting these parameters to zero assumes that default
 332          * values will be configured implicitly by slave devices.
 333          */
 334         txconf_i->tx_free_thresh = 0;
 335         txconf_i->tx_rs_thresh = 0;
 336
 337         /* Disable deferred start by default for all slave devices */
 338         txconf_i->tx_deferred_start = 0;
 339 }
 340
 341 static void
 342 eth_bond_slave_inherit_dev_info_rx_next(struct bond_dev_private *internals,
 343                                         const struct rte_eth_dev_info *di)
 344 {
 345         struct rte_eth_rxconf *rxconf_i = &internals->default_rxconf;
 346         const struct rte_eth_rxconf *rxconf = &di->default_rxconf;
 347
 348         internals->rx_offload_capa &= di->rx_offload_capa;
 349         internals->rx_queue_offload_capa &= di->rx_queue_offload_capa;
 350         internals->flow_type_rss_offloads &= di->flow_type_rss_offloads;
 351
 352         /*
 353          * If at least one slave device suggests enabling this
 354          * setting by default, enable it for all slave devices
 355          * since disabling it may not be necessarily supported.
 356          */
 357         if (rxconf->rx_drop_en == 1)
 358                 rxconf_i->rx_drop_en = 1;
 359
 360         /*
 361          * Adding a new slave device may cause some of previously inherited
 362          * offloads to be withdrawn from the internal rx_queue_offload_capa
 363          * value. Thus, the new internal value of default Rx queue offloads
 364          * has to be masked by rx_queue_offload_capa to make sure that only
 365          * commonly supported offloads are preserved from both the previous
 366          * value and the value being inhereted from the new slave device.
 367          */
 368         rxconf_i->offloads = (rxconf_i->offloads | rxconf->offloads) &
 369                              internals->rx_queue_offload_capa;
 370
 371         /*
 372          * RETA size is GCD of all slaves RETA sizes, so, if all sizes will be
 373          * the power of 2, the lower one is GCD
 374          */
 375         if (internals->reta_size > di->reta_size)
 376                 internals->reta_size = di->reta_size;
 377
 378         if (!internals->max_rx_pktlen &&
 379             di->max_rx_pktlen < internals->candidate_max_rx_pktlen)
 380                 internals->candidate_max_rx_pktlen = di->max_rx_pktlen;
 381 }
 382
 383 static void
 384 eth_bond_slave_inherit_dev_info_tx_next(struct bond_dev_private *internals,
 385                                         const struct rte_eth_dev_info *di)
 386 {
 387         struct rte_eth_txconf *txconf_i = &internals->default_txconf;
 388         const struct rte_eth_txconf *txconf = &di->default_txconf;
 389
 390         internals->tx_offload_capa &= di->tx_offload_capa;
 391         internals->tx_queue_offload_capa &= di->tx_queue_offload_capa;
 392
 393         /*
 394          * Adding a new slave device may cause some of previously inherited
 395          * offloads to be withdrawn from the internal tx_queue_offload_capa
 396          * value. Thus, the new internal value of default Tx queue offloads
 397          * has to be masked by tx_queue_offload_capa to make sure that only
 398          * commonly supported offloads are preserved from both the previous
 399          * value and the value being inhereted from the new slave device.
 400          */
 401         txconf_i->offloads = (txconf_i->offloads | txconf->offloads) &
 402                              internals->tx_queue_offload_capa;
 403 }
 404
 405 static void
 406 eth_bond_slave_inherit_desc_lim_first(struct rte_eth_desc_lim *bond_desc_lim,
 407                 const struct rte_eth_desc_lim *slave_desc_lim)
 408 {
 409         memcpy(bond_desc_lim, slave_desc_lim, sizeof(*bond_desc_lim));
 410 }
 411
 412 static int
 413 eth_bond_slave_inherit_desc_lim_next(struct rte_eth_desc_lim *bond_desc_lim,
 414                 const struct rte_eth_desc_lim *slave_desc_lim)
 415 {
 416         bond_desc_lim->nb_max = RTE_MIN(bond_desc_lim->nb_max,
 417                                         slave_desc_lim->nb_max);
 418         bond_desc_lim->nb_min = RTE_MAX(bond_desc_lim->nb_min,
 419                                         slave_desc_lim->nb_min);
 420         bond_desc_lim->nb_align = RTE_MAX(bond_desc_lim->nb_align,
 421                                           slave_desc_lim->nb_align);
 422
 423         if (bond_desc_lim->nb_min > bond_desc_lim->nb_max ||
 424             bond_desc_lim->nb_align > bond_desc_lim->nb_max) {
 425                 RTE_BOND_LOG(ERR, "Failed to inherit descriptor limits");
 426                 return -EINVAL;
 427         }
 428
 429         /* Treat maximum number of segments equal to 0 as unspecified */
 430         if (slave_desc_lim->nb_seg_max != 0 &&
 431             (bond_desc_lim->nb_seg_max == 0 ||
 432              slave_desc_lim->nb_seg_max < bond_desc_lim->nb_seg_max))
 433                 bond_desc_lim->nb_seg_max = slave_desc_lim->nb_seg_max;
 434         if (slave_desc_lim->nb_mtu_seg_max != 0 &&
 435             (bond_desc_lim->nb_mtu_seg_max == 0 ||
 436              slave_desc_lim->nb_mtu_seg_max < bond_desc_lim->nb_mtu_seg_max))
 437                 bond_desc_lim->nb_mtu_seg_max = slave_desc_lim->nb_mtu_seg_max;
 438
 439         return 0;
 440 }
 441
 442 static int
 443 __eth_bond_slave_add_lock_free(uint16_t bonded_port_id, uint16_t slave_port_id)
 444 {
 445         struct rte_eth_dev *bonded_eth_dev, *slave_eth_dev;
 446         struct bond_dev_private *internals;
 447         struct rte_eth_link link_props;
 448         struct rte_eth_dev_info dev_info;
 449         int ret;
 450
 451         bonded_eth_dev = &rte_eth_devices[bonded_port_id];
 452         internals = bonded_eth_dev->data->dev_private;
 453
 454         if (valid_slave_port_id(slave_port_id, internals->mode) != 0)
 455                 return -1;
 456
 457         slave_eth_dev = &rte_eth_devices[slave_port_id];
 458         if (slave_eth_dev->data->dev_flags & RTE_ETH_DEV_BONDED_SLAVE) {
 459                 RTE_BOND_LOG(ERR, "Slave device is already a slave of a bonded device");
 460                 return -1;
 461         }
 462
 463         ret = rte_eth_dev_info_get(slave_port_id, &dev_info);
 464         if (ret != 0) {
 465                 RTE_BOND_LOG(ERR,
 466                         "%s: Error during getting device (port %u) info: %s\n",
 467                         __func__, slave_port_id, strerror(-ret));
 468
 469                 return ret;
 470         }
 471         if (dev_info.max_rx_pktlen < internals->max_rx_pktlen) {
 472                 RTE_BOND_LOG(ERR, "Slave (port %u) max_rx_pktlen too small",
 473                              slave_port_id);
 474                 return -1;
 475         }
 476
 477         slave_add(internals, slave_eth_dev);
 478
 479         /* We need to store slaves reta_size to be able to synchronize RETA for all
 480          * slave devices even if its sizes are different.
 481          */
 482         internals->slaves[internals->slave_count].reta_size = dev_info.reta_size;
 483
 484         if (internals->slave_count < 1) {
 485                 /* if MAC is not user defined then use MAC of first slave add to
 486                  * bonded device */
 487                 if (!internals->user_defined_mac) {
 488                         if (mac_address_set(bonded_eth_dev,
 489                                             slave_eth_dev->data->mac_addrs)) {
 490                                 RTE_BOND_LOG(ERR, "Failed to set MAC address");
 491                                 return -1;
 492                         }
 493                 }
 494
 495                 /* Make primary slave */
 496                 internals->primary_port = slave_port_id;
 497                 internals->current_primary_port = slave_port_id;
 498
 499                 /* Inherit queues settings from first slave */
 500                 internals->nb_rx_queues = slave_eth_dev->data->nb_rx_queues;
 501                 internals->nb_tx_queues = slave_eth_dev->data->nb_tx_queues;
 502
 503                 eth_bond_slave_inherit_dev_info_rx_first(internals, &dev_info);
 504                 eth_bond_slave_inherit_dev_info_tx_first(internals, &dev_info);
 505
 506                 eth_bond_slave_inherit_desc_lim_first(&internals->rx_desc_lim,
 507                                                       &dev_info.rx_desc_lim);
 508                 eth_bond_slave_inherit_desc_lim_first(&internals->tx_desc_lim,
 509                                                       &dev_info.tx_desc_lim);
 510         } else {
 511                 int ret;
 512
 513                 eth_bond_slave_inherit_dev_info_rx_next(internals, &dev_info);
 514                 eth_bond_slave_inherit_dev_info_tx_next(internals, &dev_info);
 515
 516                 ret = eth_bond_slave_inherit_desc_lim_next(
 517                                 &internals->rx_desc_lim, &dev_info.rx_desc_lim);
 518                 if (ret != 0)
 519                         return ret;
 520
 521                 ret = eth_bond_slave_inherit_desc_lim_next(
 522                                 &internals->tx_desc_lim, &dev_info.tx_desc_lim);
 523                 if (ret != 0)
 524                         return ret;
 525         }
 526
 527         bonded_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf &=
 528                         internals->flow_type_rss_offloads;
 529
 530         if (slave_rte_flow_prepare(internals->slave_count, internals) != 0) {
 531                 RTE_BOND_LOG(ERR, "Failed to prepare new slave flows: port=%d",
 532                              slave_port_id);
 533                 return -1;
 534         }
 535
 536         /* Add additional MAC addresses to the slave */
 537         if (slave_add_mac_addresses(bonded_eth_dev, slave_port_id) != 0) {
 538                 RTE_BOND_LOG(ERR, "Failed to add mac address(es) to slave %hu",
 539                                 slave_port_id);
 540                 return -1;
 541         }
 542
 543         internals->slave_count++;
 544
 545         if (bonded_eth_dev->data->dev_started) {
 546                 if (slave_configure(bonded_eth_dev, slave_eth_dev) != 0) {
 547                         internals->slave_count--;
 548                         RTE_BOND_LOG(ERR, "rte_bond_slaves_configure: port=%d",
 549                                         slave_port_id);
 550                         return -1;
 551                 }
 552         }
 553
 554         /* Update all slave devices MACs */
 555         mac_address_slaves_update(bonded_eth_dev);
 556
 557         /* Register link status change callback with bonded device pointer as
 558          * argument*/
 559         rte_eth_dev_callback_register(slave_port_id, RTE_ETH_EVENT_INTR_LSC,
 560                         bond_ethdev_lsc_event_callback, &bonded_eth_dev->data->port_id);
 561
 562         /* If bonded device is started then we can add the slave to our active
 563          * slave array */
 564         if (bonded_eth_dev->data->dev_started) {
 565                 ret = rte_eth_link_get_nowait(slave_port_id, &link_props);
 566                 if (ret < 0) {
 567                         rte_eth_dev_callback_unregister(slave_port_id,
 568                                         RTE_ETH_EVENT_INTR_LSC,
 569                                         bond_ethdev_lsc_event_callback,
 570                                         &bonded_eth_dev->data->port_id);
 571                         internals->slave_count--;
 572                         RTE_BOND_LOG(ERR,
 573                                 "Slave (port %u) link get failed: %s\n",
 574                                 slave_port_id, rte_strerror(-ret));
 575                         return -1;
 576                 }
 577
 578                  if (link_props.link_status == ETH_LINK_UP) {
 579                         if (internals->active_slave_count == 0 &&
 580                             !internals->user_defined_primary_port)
 581                                 bond_ethdev_primary_set(internals,
 582                                                         slave_port_id);
 583                 }
 584         }
 585
 586         /* Add slave details to bonded device */
 587         slave_eth_dev->data->dev_flags |= RTE_ETH_DEV_BONDED_SLAVE;
 588
 589         slave_vlan_filter_set(bonded_port_id, slave_port_id);
 590
 591         return 0;
 592
 593 }
 594
 595 int
 596 rte_eth_bond_slave_add(uint16_t bonded_port_id, uint16_t slave_port_id)
 597 {
 598         struct rte_eth_dev *bonded_eth_dev;
 599         struct bond_dev_private *internals;
 600
 601         int retval;
 602
 603         /* Verify that port id's are valid bonded and slave ports */
 604         if (valid_bonded_port_id(bonded_port_id) != 0)
 605                 return -1;
 606
 607         bonded_eth_dev = &rte_eth_devices[bonded_port_id];
 608         internals = bonded_eth_dev->data->dev_private;
 609
 610         rte_spinlock_lock(&internals->lock);
 611
 612         retval = __eth_bond_slave_add_lock_free(bonded_port_id, slave_port_id);
 613
 614         rte_spinlock_unlock(&internals->lock);
 615
 616         return retval;
 617 }
 618
 619 static int
 620 __eth_bond_slave_remove_lock_free(uint16_t bonded_port_id,
 621                                    uint16_t slave_port_id)
 622 {
 623         struct rte_eth_dev *bonded_eth_dev;
 624         struct bond_dev_private *internals;
 625         struct rte_eth_dev *slave_eth_dev;
 626         struct rte_flow_error flow_error;
 627         struct rte_flow *flow;
 628         int i, slave_idx;
 629
 630         bonded_eth_dev = &rte_eth_devices[bonded_port_id];
 631         internals = bonded_eth_dev->data->dev_private;
 632
 633         if (valid_slave_port_id(slave_port_id, internals->mode) < 0)
 634                 return -1;
 635
 636         /* first remove from active slave list */
 637         slave_idx = find_slave_by_id(internals->active_slaves,
 638                 internals->active_slave_count, slave_port_id);
 639
 640         if (slave_idx < internals->active_slave_count)
 641                 deactivate_slave(bonded_eth_dev, slave_port_id);
 642
 643         slave_idx = -1;
 644         /* now find in slave list */
 645         for (i = 0; i < internals->slave_count; i++)
 646                 if (internals->slaves[i].port_id == slave_port_id) {
 647                         slave_idx = i;
 648                         break;
 649                 }
 650
 651         if (slave_idx < 0) {
 652                 RTE_BOND_LOG(ERR, "Couldn't find slave in port list, slave count %d",
 653                                 internals->slave_count);
 654                 return -1;
 655         }
 656
 657         /* Un-register link status change callback with bonded device pointer as
 658          * argument*/
 659         rte_eth_dev_callback_unregister(slave_port_id, RTE_ETH_EVENT_INTR_LSC,
 660                         bond_ethdev_lsc_event_callback,
 661                         &rte_eth_devices[bonded_port_id].data->port_id);
 662
 663         /* Restore original MAC address of slave device */
 664         rte_eth_dev_default_mac_addr_set(slave_port_id,
 665                         &(internals->slaves[slave_idx].persisted_mac_addr));
 666
 667         /* remove additional MAC addresses from the slave */
 668         slave_remove_mac_addresses(bonded_eth_dev, slave_port_id);
 669
 670         /*
 671          * Remove bond device flows from slave device.
 672          * Note: don't restore flow isolate mode.
 673          */
 674         TAILQ_FOREACH(flow, &internals->flow_list, next) {
 675                 if (flow->flows[slave_idx] != NULL) {
 676                         rte_flow_destroy(slave_port_id, flow->flows[slave_idx],
 677                                          &flow_error);
 678                         flow->flows[slave_idx] = NULL;
 679                 }
 680         }
 681
 682         slave_eth_dev = &rte_eth_devices[slave_port_id];
 683         slave_remove(internals, slave_eth_dev);
 684         slave_eth_dev->data->dev_flags &= (~RTE_ETH_DEV_BONDED_SLAVE);
 685
 686         /*  first slave in the active list will be the primary by default,
 687          *  otherwise use first device in list */
 688         if (internals->current_primary_port == slave_port_id) {
 689                 if (internals->active_slave_count > 0)
 690                         internals->current_primary_port = internals->active_slaves[0];
 691                 else if (internals->slave_count > 0)
 692                         internals->current_primary_port = internals->slaves[0].port_id;
 693                 else
 694                         internals->primary_port = 0;
 695                 mac_address_slaves_update(bonded_eth_dev);
 696         }
 697
 698         if (internals->active_slave_count < 1) {
 699                 /* if no slaves are any longer attached to bonded device and MAC is not
 700                  * user defined then clear MAC of bonded device as it will be reset
 701                  * when a new slave is added */
 702                 if (internals->slave_count < 1 && !internals->user_defined_mac)
 703                         memset(rte_eth_devices[bonded_port_id].data->mac_addrs, 0,
 704                                         sizeof(*(rte_eth_devices[bonded_port_id].data->mac_addrs)));
 705         }
 706         if (internals->slave_count == 0) {
 707                 internals->rx_offload_capa = 0;
 708                 internals->tx_offload_capa = 0;
 709                 internals->rx_queue_offload_capa = 0;
 710                 internals->tx_queue_offload_capa = 0;
 711                 internals->flow_type_rss_offloads = ETH_RSS_PROTO_MASK;
 712                 internals->reta_size = 0;
 713                 internals->candidate_max_rx_pktlen = 0;
 714                 internals->max_rx_pktlen = 0;
 715         }
 716         return 0;
 717 }
 718
 719 int
 720 rte_eth_bond_slave_remove(uint16_t bonded_port_id, uint16_t slave_port_id)
 721 {
 722         struct rte_eth_dev *bonded_eth_dev;
 723         struct bond_dev_private *internals;
 724         int retval;
 725
 726         if (valid_bonded_port_id(bonded_port_id) != 0)
 727                 return -1;
 728
 729         bonded_eth_dev = &rte_eth_devices[bonded_port_id];
 730         internals = bonded_eth_dev->data->dev_private;
 731
 732         rte_spinlock_lock(&internals->lock);
 733
 734         retval = __eth_bond_slave_remove_lock_free(bonded_port_id, slave_port_id);
 735
 736         rte_spinlock_unlock(&internals->lock);
 737
 738         return retval;
 739 }
 740
 741 int
 742 rte_eth_bond_mode_set(uint16_t bonded_port_id, uint8_t mode)
 743 {
 744         struct rte_eth_dev *bonded_eth_dev;
 745
 746         if (valid_bonded_port_id(bonded_port_id) != 0)
 747                 return -1;
 748
 749         bonded_eth_dev = &rte_eth_devices[bonded_port_id];
 750
 751         if (check_for_master_bonded_ethdev(bonded_eth_dev) != 0 &&
 752                         mode == BONDING_MODE_8023AD)
 753                 return -1;
 754
 755         return bond_ethdev_mode_set(bonded_eth_dev, mode);
 756 }
 757
 758 int
 759 rte_eth_bond_mode_get(uint16_t bonded_port_id)
 760 {
 761         struct bond_dev_private *internals;
 762
 763         if (valid_bonded_port_id(bonded_port_id) != 0)
 764                 return -1;
 765
 766         internals = rte_eth_devices[bonded_port_id].data->dev_private;
 767
 768         return internals->mode;
 769 }
 770
 771 int
 772 rte_eth_bond_primary_set(uint16_t bonded_port_id, uint16_t slave_port_id)
 773 {
 774         struct bond_dev_private *internals;
 775
 776         if (valid_bonded_port_id(bonded_port_id) != 0)
 777                 return -1;
 778
 779         internals = rte_eth_devices[bonded_port_id].data->dev_private;
 780
 781         if (valid_slave_port_id(slave_port_id, internals->mode) != 0)
 782                 return -1;
 783
 784         internals->user_defined_primary_port = 1;
 785         internals->primary_port = slave_port_id;
 786
 787         bond_ethdev_primary_set(internals, slave_port_id);
 788
 789         return 0;
 790 }
 791
 792 int
 793 rte_eth_bond_primary_get(uint16_t bonded_port_id)
 794 {
 795         struct bond_dev_private *internals;
 796
 797         if (valid_bonded_port_id(bonded_port_id) != 0)
 798                 return -1;
 799
 800         internals = rte_eth_devices[bonded_port_id].data->dev_private;
 801
 802         if (internals->slave_count < 1)
 803                 return -1;
 804
 805         return internals->current_primary_port;
 806 }
 807
 808 int
 809 rte_eth_bond_slaves_get(uint16_t bonded_port_id, uint16_t slaves[],
 810                         uint16_t len)
 811 {
 812         struct bond_dev_private *internals;
 813         uint16_t i;
 814
 815         if (valid_bonded_port_id(bonded_port_id) != 0)
 816                 return -1;
 817
 818         if (slaves == NULL)
 819                 return -1;
 820
 821         internals = rte_eth_devices[bonded_port_id].data->dev_private;
 822
 823         if (internals->slave_count > len)
 824                 return -1;
 825
 826         for (i = 0; i < internals->slave_count; i++)
 827                 slaves[i] = internals->slaves[i].port_id;
 828
 829         return internals->slave_count;
 830 }
 831
 832 int
 833 rte_eth_bond_active_slaves_get(uint16_t bonded_port_id, uint16_t slaves[],
 834                 uint16_t len)
 835 {
 836         struct bond_dev_private *internals;
 837
 838         if (valid_bonded_port_id(bonded_port_id) != 0)
 839                 return -1;
 840
 841         if (slaves == NULL)
 842                 return -1;
 843
 844         internals = rte_eth_devices[bonded_port_id].data->dev_private;
 845
 846         if (internals->active_slave_count > len)
 847                 return -1;
 848
 849         memcpy(slaves, internals->active_slaves,
 850         internals->active_slave_count * sizeof(internals->active_slaves[0]));
 851
 852         return internals->active_slave_count;
 853 }
 854
 855 int
 856 rte_eth_bond_mac_address_set(uint16_t bonded_port_id,
 857                 struct rte_ether_addr *mac_addr)
 858 {
 859         struct rte_eth_dev *bonded_eth_dev;
 860         struct bond_dev_private *internals;
 861
 862         if (valid_bonded_port_id(bonded_port_id) != 0)
 863                 return -1;
 864
 865         bonded_eth_dev = &rte_eth_devices[bonded_port_id];
 866         internals = bonded_eth_dev->data->dev_private;
 867
 868         /* Set MAC Address of Bonded Device */
 869         if (mac_address_set(bonded_eth_dev, mac_addr))
 870                 return -1;
 871
 872         internals->user_defined_mac = 1;
 873
 874         /* Update all slave devices MACs*/
 875         if (internals->slave_count > 0)
 876                 return mac_address_slaves_update(bonded_eth_dev);
 877
 878         return 0;
 879 }
 880
 881 int
 882 rte_eth_bond_mac_address_reset(uint16_t bonded_port_id)
 883 {
 884         struct rte_eth_dev *bonded_eth_dev;
 885         struct bond_dev_private *internals;
 886
 887         if (valid_bonded_port_id(bonded_port_id) != 0)
 888                 return -1;
 889
 890         bonded_eth_dev = &rte_eth_devices[bonded_port_id];
 891         internals = bonded_eth_dev->data->dev_private;
 892
 893         internals->user_defined_mac = 0;
 894
 895         if (internals->slave_count > 0) {
 896                 int slave_port;
 897                 /* Get the primary slave location based on the primary port
 898                  * number as, while slave_add(), we will keep the primary
 899                  * slave based on slave_count,but not based on the primary port.
 900                  */
 901                 for (slave_port = 0; slave_port < internals->slave_count;
 902                      slave_port++) {
 903                         if (internals->slaves[slave_port].port_id ==
 904                             internals->primary_port)
 905                                 break;
 906                 }
 907
 908                 /* Set MAC Address of Bonded Device */
 909                 if (mac_address_set(bonded_eth_dev,
 910                         &internals->slaves[slave_port].persisted_mac_addr)
 911                                 != 0) {
 912                         RTE_BOND_LOG(ERR, "Failed to set MAC address on bonded device");
 913                         return -1;
 914                 }
 915                 /* Update all slave devices MAC addresses */
 916                 return mac_address_slaves_update(bonded_eth_dev);
 917         }
 918         /* No need to update anything as no slaves present */
 919         return 0;
 920 }
 921
 922 int
 923 rte_eth_bond_xmit_policy_set(uint16_t bonded_port_id, uint8_t policy)
 924 {
 925         struct bond_dev_private *internals;
 926
 927         if (valid_bonded_port_id(bonded_port_id) != 0)
 928                 return -1;
 929
 930         internals = rte_eth_devices[bonded_port_id].data->dev_private;
 931
 932         switch (policy) {
 933         case BALANCE_XMIT_POLICY_LAYER2:
 934                 internals->balance_xmit_policy = policy;
 935                 internals->burst_xmit_hash = burst_xmit_l2_hash;
 936                 break;
 937         case BALANCE_XMIT_POLICY_LAYER23:
 938                 internals->balance_xmit_policy = policy;
 939                 internals->burst_xmit_hash = burst_xmit_l23_hash;
 940                 break;
 941         case BALANCE_XMIT_POLICY_LAYER34:
 942                 internals->balance_xmit_policy = policy;
 943                 internals->burst_xmit_hash = burst_xmit_l34_hash;
 944                 break;
 945
 946         default:
 947                 return -1;
 948         }
 949         return 0;
 950 }
 951
 952 int
 953 rte_eth_bond_xmit_policy_get(uint16_t bonded_port_id)
 954 {
 955         struct bond_dev_private *internals;
 956
 957         if (valid_bonded_port_id(bonded_port_id) != 0)
 958                 return -1;
 959
 960         internals = rte_eth_devices[bonded_port_id].data->dev_private;
 961
 962         return internals->balance_xmit_policy;
 963 }
 964
 965 int
 966 rte_eth_bond_link_monitoring_set(uint16_t bonded_port_id, uint32_t internal_ms)
 967 {
 968         struct bond_dev_private *internals;
 969
 970         if (valid_bonded_port_id(bonded_port_id) != 0)
 971                 return -1;
 972
 973         internals = rte_eth_devices[bonded_port_id].data->dev_private;
 974         internals->link_status_polling_interval_ms = internal_ms;
 975
 976         return 0;
 977 }
 978
 979 int
 980 rte_eth_bond_link_monitoring_get(uint16_t bonded_port_id)
 981 {
 982         struct bond_dev_private *internals;
 983
 984         if (valid_bonded_port_id(bonded_port_id) != 0)
 985                 return -1;
 986
 987         internals = rte_eth_devices[bonded_port_id].data->dev_private;
 988
 989         return internals->link_status_polling_interval_ms;
 990 }
 991
 992 int
 993 rte_eth_bond_link_down_prop_delay_set(uint16_t bonded_port_id,
 994                                        uint32_t delay_ms)
 995
 996 {
 997         struct bond_dev_private *internals;
 998
 999         if (valid_bonded_port_id(bonded_port_id) != 0)
1000                 return -1;
1001
1002         internals = rte_eth_devices[bonded_port_id].data->dev_private;
1003         internals->link_down_delay_ms = delay_ms;
1004
1005         return 0;
1006 }
1007
1008 int
1009 rte_eth_bond_link_down_prop_delay_get(uint16_t bonded_port_id)
1010 {
1011         struct bond_dev_private *internals;
1012
1013         if (valid_bonded_port_id(bonded_port_id) != 0)
1014                 return -1;
1015
1016         internals = rte_eth_devices[bonded_port_id].data->dev_private;
1017
1018         return internals->link_down_delay_ms;
1019 }
1020
1021 int
1022 rte_eth_bond_link_up_prop_delay_set(uint16_t bonded_port_id, uint32_t delay_ms)
1023
1024 {
1025         struct bond_dev_private *internals;
1026
1027         if (valid_bonded_port_id(bonded_port_id) != 0)
1028                 return -1;
1029
1030         internals = rte_eth_devices[bonded_port_id].data->dev_private;
1031         internals->link_up_delay_ms = delay_ms;
1032
1033         return 0;
1034 }
1035
1036 int
1037 rte_eth_bond_link_up_prop_delay_get(uint16_t bonded_port_id)
1038 {
1039         struct bond_dev_private *internals;
1040
1041         if (valid_bonded_port_id(bonded_port_id) != 0)
1042                 return -1;
1043
1044         internals = rte_eth_devices[bonded_port_id].data->dev_private;
1045
1046         return internals->link_up_delay_ms;
1047 }