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