drivers/net/bonding/rte_eth_bond_pmd.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2017 Intel Corporation
   3  */
   4 #include <stdlib.h>
   5 #include <stdbool.h>
   6 #include <netinet/in.h>
   7
   8 #include <rte_mbuf.h>
   9 #include <rte_malloc.h>
  10 #include <ethdev_driver.h>
  11 #include <ethdev_vdev.h>
  12 #include <rte_tcp.h>
  13 #include <rte_udp.h>
  14 #include <rte_ip.h>
  15 #include <rte_ip_frag.h>
  16 #include <rte_devargs.h>
  17 #include <rte_kvargs.h>
  18 #include <rte_bus_vdev.h>
  19 #include <rte_alarm.h>
  20 #include <rte_cycles.h>
  21 #include <rte_string_fns.h>
  22
  23 #include "rte_eth_bond.h"
  24 #include "eth_bond_private.h"
  25 #include "eth_bond_8023ad_private.h"
  26
  27 #define REORDER_PERIOD_MS 10
  28 #define DEFAULT_POLLING_INTERVAL_10_MS (10)
  29 #define BOND_MAX_MAC_ADDRS 16
  30
  31 #define HASH_L4_PORTS(h) ((h)->src_port ^ (h)->dst_port)
  32
  33 /* Table for statistics in mode 5 TLB */
  34 static uint64_t tlb_last_obytets[RTE_MAX_ETHPORTS];
  35
  36 static inline size_t
  37 get_vlan_offset(struct rte_ether_hdr *eth_hdr, uint16_t *proto)
  38 {
  39         size_t vlan_offset = 0;
  40
  41         if (rte_cpu_to_be_16(RTE_ETHER_TYPE_VLAN) == *proto ||
  42                 rte_cpu_to_be_16(RTE_ETHER_TYPE_QINQ) == *proto) {
  43                 struct rte_vlan_hdr *vlan_hdr =
  44                         (struct rte_vlan_hdr *)(eth_hdr + 1);
  45
  46                 vlan_offset = sizeof(struct rte_vlan_hdr);
  47                 *proto = vlan_hdr->eth_proto;
  48
  49                 if (rte_cpu_to_be_16(RTE_ETHER_TYPE_VLAN) == *proto) {
  50                         vlan_hdr = vlan_hdr + 1;
  51                         *proto = vlan_hdr->eth_proto;
  52                         vlan_offset += sizeof(struct rte_vlan_hdr);
  53                 }
  54         }
  55         return vlan_offset;
  56 }
  57
  58 static uint16_t
  59 bond_ethdev_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
  60 {
  61         struct bond_dev_private *internals;
  62
  63         uint16_t num_rx_total = 0;
  64         uint16_t slave_count;
  65         uint16_t active_slave;
  66         int i;
  67
  68         /* Cast to structure, containing bonded device's port id and queue id */
  69         struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)queue;
  70         internals = bd_rx_q->dev_private;
  71         slave_count = internals->active_slave_count;
  72         active_slave = bd_rx_q->active_slave;
  73
  74         for (i = 0; i < slave_count && nb_pkts; i++) {
  75                 uint16_t num_rx_slave;
  76
  77                 /* Offset of pointer to *bufs increases as packets are received
  78                  * from other slaves */
  79                 num_rx_slave =
  80                         rte_eth_rx_burst(internals->active_slaves[active_slave],
  81                                          bd_rx_q->queue_id,
  82                                          bufs + num_rx_total, nb_pkts);
  83                 num_rx_total += num_rx_slave;
  84                 nb_pkts -= num_rx_slave;
  85                 if (++active_slave == slave_count)
  86                         active_slave = 0;
  87         }
  88
  89         if (++bd_rx_q->active_slave >= slave_count)
  90                 bd_rx_q->active_slave = 0;
  91         return num_rx_total;
  92 }
  93
  94 static uint16_t
  95 bond_ethdev_rx_burst_active_backup(void *queue, struct rte_mbuf **bufs,
  96                 uint16_t nb_pkts)
  97 {
  98         struct bond_dev_private *internals;
  99
 100         /* Cast to structure, containing bonded device's port id and queue id */
 101         struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)queue;
 102
 103         internals = bd_rx_q->dev_private;
 104
 105         return rte_eth_rx_burst(internals->current_primary_port,
 106                         bd_rx_q->queue_id, bufs, nb_pkts);
 107 }
 108
 109 static inline uint8_t
 110 is_lacp_packets(uint16_t ethertype, uint8_t subtype, struct rte_mbuf *mbuf)
 111 {
 112         const uint16_t ether_type_slow_be =
 113                 rte_be_to_cpu_16(RTE_ETHER_TYPE_SLOW);
 114
 115         return !((mbuf->ol_flags & PKT_RX_VLAN) ? mbuf->vlan_tci : 0) &&
 116                 (ethertype == ether_type_slow_be &&
 117                 (subtype == SLOW_SUBTYPE_MARKER || subtype == SLOW_SUBTYPE_LACP));
 118 }
 119
 120 /*****************************************************************************
 121  * Flow director's setup for mode 4 optimization
 122  */
 123
 124 static struct rte_flow_item_eth flow_item_eth_type_8023ad = {
 125         .dst.addr_bytes = { 0 },
 126         .src.addr_bytes = { 0 },
 127         .type = RTE_BE16(RTE_ETHER_TYPE_SLOW),
 128 };
 129
 130 static struct rte_flow_item_eth flow_item_eth_mask_type_8023ad = {
 131         .dst.addr_bytes = { 0 },
 132         .src.addr_bytes = { 0 },
 133         .type = 0xFFFF,
 134 };
 135
 136 static struct rte_flow_item flow_item_8023ad[] = {
 137         {
 138                 .type = RTE_FLOW_ITEM_TYPE_ETH,
 139                 .spec = &flow_item_eth_type_8023ad,
 140                 .last = NULL,
 141                 .mask = &flow_item_eth_mask_type_8023ad,
 142         },
 143         {
 144                 .type = RTE_FLOW_ITEM_TYPE_END,
 145                 .spec = NULL,
 146                 .last = NULL,
 147                 .mask = NULL,
 148         }
 149 };
 150
 151 const struct rte_flow_attr flow_attr_8023ad = {
 152         .group = 0,
 153         .priority = 0,
 154         .ingress = 1,
 155         .egress = 0,
 156         .reserved = 0,
 157 };
 158
 159 int
 160 bond_ethdev_8023ad_flow_verify(struct rte_eth_dev *bond_dev,
 161                 uint16_t slave_port) {
 162         struct rte_eth_dev_info slave_info;
 163         struct rte_flow_error error;
 164         struct bond_dev_private *internals = bond_dev->data->dev_private;
 165
 166         const struct rte_flow_action_queue lacp_queue_conf = {
 167                 .index = 0,
 168         };
 169
 170         const struct rte_flow_action actions[] = {
 171                 {
 172                         .type = RTE_FLOW_ACTION_TYPE_QUEUE,
 173                         .conf = &lacp_queue_conf
 174                 },
 175                 {
 176                         .type = RTE_FLOW_ACTION_TYPE_END,
 177                 }
 178         };
 179
 180         int ret = rte_flow_validate(slave_port, &flow_attr_8023ad,
 181                         flow_item_8023ad, actions, &error);
 182         if (ret < 0) {
 183                 RTE_BOND_LOG(ERR, "%s: %s (slave_port=%d queue_id=%d)",
 184                                 __func__, error.message, slave_port,
 185                                 internals->mode4.dedicated_queues.rx_qid);
 186                 return -1;
 187         }
 188
 189         ret = rte_eth_dev_info_get(slave_port, &slave_info);
 190         if (ret != 0) {
 191                 RTE_BOND_LOG(ERR,
 192                         "%s: Error during getting device (port %u) info: %s\n",
 193                         __func__, slave_port, strerror(-ret));
 194
 195                 return ret;
 196         }
 197
 198         if (slave_info.max_rx_queues < bond_dev->data->nb_rx_queues ||
 199                         slave_info.max_tx_queues < bond_dev->data->nb_tx_queues) {
 200                 RTE_BOND_LOG(ERR,
 201                         "%s: Slave %d capabilities doesn't allow to allocate additional queues",
 202                         __func__, slave_port);
 203                 return -1;
 204         }
 205
 206         return 0;
 207 }
 208
 209 int
 210 bond_8023ad_slow_pkt_hw_filter_supported(uint16_t port_id) {
 211         struct rte_eth_dev *bond_dev = &rte_eth_devices[port_id];
 212         struct bond_dev_private *internals = bond_dev->data->dev_private;
 213         struct rte_eth_dev_info bond_info;
 214         uint16_t idx;
 215         int ret;
 216
 217         /* Verify if all slaves in bonding supports flow director and */
 218         if (internals->slave_count > 0) {
 219                 ret = rte_eth_dev_info_get(bond_dev->data->port_id, &bond_info);
 220                 if (ret != 0) {
 221                         RTE_BOND_LOG(ERR,
 222                                 "%s: Error during getting device (port %u) info: %s\n",
 223                                 __func__, bond_dev->data->port_id,
 224                                 strerror(-ret));
 225
 226                         return ret;
 227                 }
 228
 229                 internals->mode4.dedicated_queues.rx_qid = bond_info.nb_rx_queues;
 230                 internals->mode4.dedicated_queues.tx_qid = bond_info.nb_tx_queues;
 231
 232                 for (idx = 0; idx < internals->slave_count; idx++) {
 233                         if (bond_ethdev_8023ad_flow_verify(bond_dev,
 234                                         internals->slaves[idx].port_id) != 0)
 235                                 return -1;
 236                 }
 237         }
 238
 239         return 0;
 240 }
 241
 242 int
 243 bond_ethdev_8023ad_flow_set(struct rte_eth_dev *bond_dev, uint16_t slave_port) {
 244
 245         struct rte_flow_error error;
 246         struct bond_dev_private *internals = bond_dev->data->dev_private;
 247         struct rte_flow_action_queue lacp_queue_conf = {
 248                 .index = internals->mode4.dedicated_queues.rx_qid,
 249         };
 250
 251         const struct rte_flow_action actions[] = {
 252                 {
 253                         .type = RTE_FLOW_ACTION_TYPE_QUEUE,
 254                         .conf = &lacp_queue_conf
 255                 },
 256                 {
 257                         .type = RTE_FLOW_ACTION_TYPE_END,
 258                 }
 259         };
 260
 261         internals->mode4.dedicated_queues.flow[slave_port] = rte_flow_create(slave_port,
 262                         &flow_attr_8023ad, flow_item_8023ad, actions, &error);
 263         if (internals->mode4.dedicated_queues.flow[slave_port] == NULL) {
 264                 RTE_BOND_LOG(ERR, "bond_ethdev_8023ad_flow_set: %s "
 265                                 "(slave_port=%d queue_id=%d)",
 266                                 error.message, slave_port,
 267                                 internals->mode4.dedicated_queues.rx_qid);
 268                 return -1;
 269         }
 270
 271         return 0;
 272 }
 273
 274 static inline uint16_t
 275 rx_burst_8023ad(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts,
 276                 bool dedicated_rxq)
 277 {
 278         /* Cast to structure, containing bonded device's port id and queue id */
 279         struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)queue;
 280         struct bond_dev_private *internals = bd_rx_q->dev_private;
 281         struct rte_eth_dev *bonded_eth_dev =
 282                                         &rte_eth_devices[internals->port_id];
 283         struct rte_ether_addr *bond_mac = bonded_eth_dev->data->mac_addrs;
 284         struct rte_ether_hdr *hdr;
 285
 286         const uint16_t ether_type_slow_be =
 287                 rte_be_to_cpu_16(RTE_ETHER_TYPE_SLOW);
 288         uint16_t num_rx_total = 0;      /* Total number of received packets */
 289         uint16_t slaves[RTE_MAX_ETHPORTS];
 290         uint16_t slave_count, idx;
 291
 292         uint8_t collecting;  /* current slave collecting status */
 293         const uint8_t promisc = rte_eth_promiscuous_get(internals->port_id);
 294         const uint8_t allmulti = rte_eth_allmulticast_get(internals->port_id);
 295         uint8_t subtype;
 296         uint16_t i;
 297         uint16_t j;
 298         uint16_t k;
 299
 300         /* Copy slave list to protect against slave up/down changes during tx
 301          * bursting */
 302         slave_count = internals->active_slave_count;
 303         memcpy(slaves, internals->active_slaves,
 304                         sizeof(internals->active_slaves[0]) * slave_count);
 305
 306         idx = bd_rx_q->active_slave;
 307         if (idx >= slave_count) {
 308                 bd_rx_q->active_slave = 0;
 309                 idx = 0;
 310         }
 311         for (i = 0; i < slave_count && num_rx_total < nb_pkts; i++) {
 312                 j = num_rx_total;
 313                 collecting = ACTOR_STATE(&bond_mode_8023ad_ports[slaves[idx]],
 314                                          COLLECTING);
 315
 316                 /* Read packets from this slave */
 317                 num_rx_total += rte_eth_rx_burst(slaves[idx], bd_rx_q->queue_id,
 318                                 &bufs[num_rx_total], nb_pkts - num_rx_total);
 319
 320                 for (k = j; k < 2 && k < num_rx_total; k++)
 321                         rte_prefetch0(rte_pktmbuf_mtod(bufs[k], void *));
 322
 323                 /* Handle slow protocol packets. */
 324                 while (j < num_rx_total) {
 325                         if (j + 3 < num_rx_total)
 326                                 rte_prefetch0(rte_pktmbuf_mtod(bufs[j + 3], void *));
 327
 328                         hdr = rte_pktmbuf_mtod(bufs[j], struct rte_ether_hdr *);
 329                         subtype = ((struct slow_protocol_frame *)hdr)->slow_protocol.subtype;
 330
 331                         /* Remove packet from array if:
 332                          * - it is slow packet but no dedicated rxq is present,
 333                          * - slave is not in collecting state,
 334                          * - bonding interface is not in promiscuous mode:
 335                          *   - packet is unicast and address does not match,
 336                          *   - packet is multicast and bonding interface
 337                          *     is not in allmulti,
 338                          */
 339                         if (unlikely(
 340                                 (!dedicated_rxq &&
 341                                  is_lacp_packets(hdr->ether_type, subtype,
 342                                                  bufs[j])) ||
 343                                 !collecting ||
 344                                 (!promisc &&
 345                                  ((rte_is_unicast_ether_addr(&hdr->dst_addr) &&
 346                                    !rte_is_same_ether_addr(bond_mac,
 347                                                        &hdr->dst_addr)) ||
 348                                   (!allmulti &&
 349                                    rte_is_multicast_ether_addr(&hdr->dst_addr)))))) {
 350
 351                                 if (hdr->ether_type == ether_type_slow_be) {
 352                                         bond_mode_8023ad_handle_slow_pkt(
 353                                             internals, slaves[idx], bufs[j]);
 354                                 } else
 355                                         rte_pktmbuf_free(bufs[j]);
 356
 357                                 /* Packet is managed by mode 4 or dropped, shift the array */
 358                                 num_rx_total--;
 359                                 if (j < num_rx_total) {
 360                                         memmove(&bufs[j], &bufs[j + 1], sizeof(bufs[0]) *
 361                                                 (num_rx_total - j));
 362                                 }
 363                         } else
 364                                 j++;
 365                 }
 366                 if (unlikely(++idx == slave_count))
 367                         idx = 0;
 368         }
 369
 370         if (++bd_rx_q->active_slave >= slave_count)
 371                 bd_rx_q->active_slave = 0;
 372
 373         return num_rx_total;
 374 }
 375
 376 static uint16_t
 377 bond_ethdev_rx_burst_8023ad(void *queue, struct rte_mbuf **bufs,
 378                 uint16_t nb_pkts)
 379 {
 380         return rx_burst_8023ad(queue, bufs, nb_pkts, false);
 381 }
 382
 383 static uint16_t
 384 bond_ethdev_rx_burst_8023ad_fast_queue(void *queue, struct rte_mbuf **bufs,
 385                 uint16_t nb_pkts)
 386 {
 387         return rx_burst_8023ad(queue, bufs, nb_pkts, true);
 388 }
 389
 390 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
 391 uint32_t burstnumberRX;
 392 uint32_t burstnumberTX;
 393
 394 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
 395
 396 static void
 397 arp_op_name(uint16_t arp_op, char *buf, size_t buf_len)
 398 {
 399         switch (arp_op) {
 400         case RTE_ARP_OP_REQUEST:
 401                 strlcpy(buf, "ARP Request", buf_len);
 402                 return;
 403         case RTE_ARP_OP_REPLY:
 404                 strlcpy(buf, "ARP Reply", buf_len);
 405                 return;
 406         case RTE_ARP_OP_REVREQUEST:
 407                 strlcpy(buf, "Reverse ARP Request", buf_len);
 408                 return;
 409         case RTE_ARP_OP_REVREPLY:
 410                 strlcpy(buf, "Reverse ARP Reply", buf_len);
 411                 return;
 412         case RTE_ARP_OP_INVREQUEST:
 413                 strlcpy(buf, "Peer Identify Request", buf_len);
 414                 return;
 415         case RTE_ARP_OP_INVREPLY:
 416                 strlcpy(buf, "Peer Identify Reply", buf_len);
 417                 return;
 418         default:
 419                 break;
 420         }
 421         strlcpy(buf, "Unknown", buf_len);
 422         return;
 423 }
 424 #endif
 425 #define MaxIPv4String   16
 426 static void
 427 ipv4_addr_to_dot(uint32_t be_ipv4_addr, char *buf, uint8_t buf_size)
 428 {
 429         uint32_t ipv4_addr;
 430
 431         ipv4_addr = rte_be_to_cpu_32(be_ipv4_addr);
 432         snprintf(buf, buf_size, "%d.%d.%d.%d", (ipv4_addr >> 24) & 0xFF,
 433                 (ipv4_addr >> 16) & 0xFF, (ipv4_addr >> 8) & 0xFF,
 434                 ipv4_addr & 0xFF);
 435 }
 436
 437 #define MAX_CLIENTS_NUMBER      128
 438 uint8_t active_clients;
 439 struct client_stats_t {
 440         uint16_t port;
 441         uint32_t ipv4_addr;
 442         uint32_t ipv4_rx_packets;
 443         uint32_t ipv4_tx_packets;
 444 };
 445 struct client_stats_t client_stats[MAX_CLIENTS_NUMBER];
 446
 447 static void
 448 update_client_stats(uint32_t addr, uint16_t port, uint32_t *TXorRXindicator)
 449 {
 450         int i = 0;
 451
 452         for (; i < MAX_CLIENTS_NUMBER; i++)     {
 453                 if ((client_stats[i].ipv4_addr == addr) && (client_stats[i].port == port))      {
 454                         /* Just update RX packets number for this client */
 455                         if (TXorRXindicator == &burstnumberRX)
 456                                 client_stats[i].ipv4_rx_packets++;
 457                         else
 458                                 client_stats[i].ipv4_tx_packets++;
 459                         return;
 460                 }
 461         }
 462         /* We have a new client. Insert him to the table, and increment stats */
 463         if (TXorRXindicator == &burstnumberRX)
 464                 client_stats[active_clients].ipv4_rx_packets++;
 465         else
 466                 client_stats[active_clients].ipv4_tx_packets++;
 467         client_stats[active_clients].ipv4_addr = addr;
 468         client_stats[active_clients].port = port;
 469         active_clients++;
 470
 471 }
 472
 473 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
 474 #define MODE6_DEBUG(info, src_ip, dst_ip, eth_h, arp_op, port, burstnumber) \
 475         rte_log(RTE_LOG_DEBUG, bond_logtype,                            \
 476                 "%s port:%d SrcMAC:" RTE_ETHER_ADDR_PRT_FMT " SrcIP:%s " \
 477                 "DstMAC:" RTE_ETHER_ADDR_PRT_FMT " DstIP:%s %s %d\n", \
 478                 info,                                                   \
 479                 port,                                                   \
 480                 RTE_ETHER_ADDR_BYTES(&eth_h->src_addr),                  \
 481                 src_ip,                                                 \
 482                 RTE_ETHER_ADDR_BYTES(&eth_h->dst_addr),                  \
 483                 dst_ip,                                                 \
 484                 arp_op, ++burstnumber)
 485 #endif
 486
 487 static void
 488 mode6_debug(const char __rte_unused *info,
 489         struct rte_ether_hdr *eth_h, uint16_t port,
 490         uint32_t __rte_unused *burstnumber)
 491 {
 492         struct rte_ipv4_hdr *ipv4_h;
 493 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
 494         struct rte_arp_hdr *arp_h;
 495         char dst_ip[16];
 496         char ArpOp[24];
 497         char buf[16];
 498 #endif
 499         char src_ip[16];
 500
 501         uint16_t ether_type = eth_h->ether_type;
 502         uint16_t offset = get_vlan_offset(eth_h, &ether_type);
 503
 504 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
 505         strlcpy(buf, info, 16);
 506 #endif
 507
 508         if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4)) {
 509                 ipv4_h = (struct rte_ipv4_hdr *)((char *)(eth_h + 1) + offset);
 510                 ipv4_addr_to_dot(ipv4_h->src_addr, src_ip, MaxIPv4String);
 511 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
 512                 ipv4_addr_to_dot(ipv4_h->dst_addr, dst_ip, MaxIPv4String);
 513                 MODE6_DEBUG(buf, src_ip, dst_ip, eth_h, "", port, *burstnumber);
 514 #endif
 515                 update_client_stats(ipv4_h->src_addr, port, burstnumber);
 516         }
 517 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
 518         else if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_ARP)) {
 519                 arp_h = (struct rte_arp_hdr *)((char *)(eth_h + 1) + offset);
 520                 ipv4_addr_to_dot(arp_h->arp_data.arp_sip, src_ip, MaxIPv4String);
 521                 ipv4_addr_to_dot(arp_h->arp_data.arp_tip, dst_ip, MaxIPv4String);
 522                 arp_op_name(rte_be_to_cpu_16(arp_h->arp_opcode),
 523                                 ArpOp, sizeof(ArpOp));
 524                 MODE6_DEBUG(buf, src_ip, dst_ip, eth_h, ArpOp, port, *burstnumber);
 525         }
 526 #endif
 527 }
 528 #endif
 529
 530 static uint16_t
 531 bond_ethdev_rx_burst_alb(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 532 {
 533         struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)queue;
 534         struct bond_dev_private *internals = bd_rx_q->dev_private;
 535         struct rte_ether_hdr *eth_h;
 536         uint16_t ether_type, offset;
 537         uint16_t nb_recv_pkts;
 538         int i;
 539
 540         nb_recv_pkts = bond_ethdev_rx_burst(queue, bufs, nb_pkts);
 541
 542         for (i = 0; i < nb_recv_pkts; i++) {
 543                 eth_h = rte_pktmbuf_mtod(bufs[i], struct rte_ether_hdr *);
 544                 ether_type = eth_h->ether_type;
 545                 offset = get_vlan_offset(eth_h, &ether_type);
 546
 547                 if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_ARP)) {
 548 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
 549                         mode6_debug("RX ARP:", eth_h, bufs[i]->port, &burstnumberRX);
 550 #endif
 551                         bond_mode_alb_arp_recv(eth_h, offset, internals);
 552                 }
 553 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
 554                 else if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4))
 555                         mode6_debug("RX IPv4:", eth_h, bufs[i]->port, &burstnumberRX);
 556 #endif
 557         }
 558
 559         return nb_recv_pkts;
 560 }
 561
 562 static uint16_t
 563 bond_ethdev_tx_burst_round_robin(void *queue, struct rte_mbuf **bufs,
 564                 uint16_t nb_pkts)
 565 {
 566         struct bond_dev_private *internals;
 567         struct bond_tx_queue *bd_tx_q;
 568
 569         struct rte_mbuf *slave_bufs[RTE_MAX_ETHPORTS][nb_pkts];
 570         uint16_t slave_nb_pkts[RTE_MAX_ETHPORTS] = { 0 };
 571
 572         uint16_t num_of_slaves;
 573         uint16_t slaves[RTE_MAX_ETHPORTS];
 574
 575         uint16_t num_tx_total = 0, num_tx_slave;
 576
 577         static int slave_idx = 0;
 578         int i, cslave_idx = 0, tx_fail_total = 0;
 579
 580         bd_tx_q = (struct bond_tx_queue *)queue;
 581         internals = bd_tx_q->dev_private;
 582
 583         /* Copy slave list to protect against slave up/down changes during tx
 584          * bursting */
 585         num_of_slaves = internals->active_slave_count;
 586         memcpy(slaves, internals->active_slaves,
 587                         sizeof(internals->active_slaves[0]) * num_of_slaves);
 588
 589         if (num_of_slaves < 1)
 590                 return num_tx_total;
 591
 592         /* Populate slaves mbuf with which packets are to be sent on it  */
 593         for (i = 0; i < nb_pkts; i++) {
 594                 cslave_idx = (slave_idx + i) % num_of_slaves;
 595                 slave_bufs[cslave_idx][(slave_nb_pkts[cslave_idx])++] = bufs[i];
 596         }
 597
 598         /* increment current slave index so the next call to tx burst starts on the
 599          * next slave */
 600         slave_idx = ++cslave_idx;
 601
 602         /* Send packet burst on each slave device */
 603         for (i = 0; i < num_of_slaves; i++) {
 604                 if (slave_nb_pkts[i] > 0) {
 605                         num_tx_slave = rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
 606                                         slave_bufs[i], slave_nb_pkts[i]);
 607
 608                         /* if tx burst fails move packets to end of bufs */
 609                         if (unlikely(num_tx_slave < slave_nb_pkts[i])) {
 610                                 int tx_fail_slave = slave_nb_pkts[i] - num_tx_slave;
 611
 612                                 tx_fail_total += tx_fail_slave;
 613
 614                                 memcpy(&bufs[nb_pkts - tx_fail_total],
 615                                        &slave_bufs[i][num_tx_slave],
 616                                        tx_fail_slave * sizeof(bufs[0]));
 617                         }
 618                         num_tx_total += num_tx_slave;
 619                 }
 620         }
 621
 622         return num_tx_total;
 623 }
 624
 625 static uint16_t
 626 bond_ethdev_tx_burst_active_backup(void *queue,
 627                 struct rte_mbuf **bufs, uint16_t nb_pkts)
 628 {
 629         struct bond_dev_private *internals;
 630         struct bond_tx_queue *bd_tx_q;
 631
 632         bd_tx_q = (struct bond_tx_queue *)queue;
 633         internals = bd_tx_q->dev_private;
 634
 635         if (internals->active_slave_count < 1)
 636                 return 0;
 637
 638         return rte_eth_tx_burst(internals->current_primary_port, bd_tx_q->queue_id,
 639                         bufs, nb_pkts);
 640 }
 641
 642 static inline uint16_t
 643 ether_hash(struct rte_ether_hdr *eth_hdr)
 644 {
 645         unaligned_uint16_t *word_src_addr =
 646                 (unaligned_uint16_t *)eth_hdr->src_addr.addr_bytes;
 647         unaligned_uint16_t *word_dst_addr =
 648                 (unaligned_uint16_t *)eth_hdr->dst_addr.addr_bytes;
 649
 650         return (word_src_addr[0] ^ word_dst_addr[0]) ^
 651                         (word_src_addr[1] ^ word_dst_addr[1]) ^
 652                         (word_src_addr[2] ^ word_dst_addr[2]);
 653 }
 654
 655 static inline uint32_t
 656 ipv4_hash(struct rte_ipv4_hdr *ipv4_hdr)
 657 {
 658         return ipv4_hdr->src_addr ^ ipv4_hdr->dst_addr;
 659 }
 660
 661 static inline uint32_t
 662 ipv6_hash(struct rte_ipv6_hdr *ipv6_hdr)
 663 {
 664         unaligned_uint32_t *word_src_addr =
 665                 (unaligned_uint32_t *)&(ipv6_hdr->src_addr[0]);
 666         unaligned_uint32_t *word_dst_addr =
 667                 (unaligned_uint32_t *)&(ipv6_hdr->dst_addr[0]);
 668
 669         return (word_src_addr[0] ^ word_dst_addr[0]) ^
 670                         (word_src_addr[1] ^ word_dst_addr[1]) ^
 671                         (word_src_addr[2] ^ word_dst_addr[2]) ^
 672                         (word_src_addr[3] ^ word_dst_addr[3]);
 673 }
 674
 675
 676 void
 677 burst_xmit_l2_hash(struct rte_mbuf **buf, uint16_t nb_pkts,
 678                 uint16_t slave_count, uint16_t *slaves)
 679 {
 680         struct rte_ether_hdr *eth_hdr;
 681         uint32_t hash;
 682         int i;
 683
 684         for (i = 0; i < nb_pkts; i++) {
 685                 eth_hdr = rte_pktmbuf_mtod(buf[i], struct rte_ether_hdr *);
 686
 687                 hash = ether_hash(eth_hdr);
 688
 689                 slaves[i] = (hash ^= hash >> 8) % slave_count;
 690         }
 691 }
 692
 693 void
 694 burst_xmit_l23_hash(struct rte_mbuf **buf, uint16_t nb_pkts,
 695                 uint16_t slave_count, uint16_t *slaves)
 696 {
 697         uint16_t i;
 698         struct rte_ether_hdr *eth_hdr;
 699         uint16_t proto;
 700         size_t vlan_offset;
 701         uint32_t hash, l3hash;
 702
 703         for (i = 0; i < nb_pkts; i++) {
 704                 eth_hdr = rte_pktmbuf_mtod(buf[i], struct rte_ether_hdr *);
 705                 l3hash = 0;
 706
 707                 proto = eth_hdr->ether_type;
 708                 hash = ether_hash(eth_hdr);
 709
 710                 vlan_offset = get_vlan_offset(eth_hdr, &proto);
 711
 712                 if (rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4) == proto) {
 713                         struct rte_ipv4_hdr *ipv4_hdr = (struct rte_ipv4_hdr *)
 714                                         ((char *)(eth_hdr + 1) + vlan_offset);
 715                         l3hash = ipv4_hash(ipv4_hdr);
 716
 717                 } else if (rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6) == proto) {
 718                         struct rte_ipv6_hdr *ipv6_hdr = (struct rte_ipv6_hdr *)
 719                                         ((char *)(eth_hdr + 1) + vlan_offset);
 720                         l3hash = ipv6_hash(ipv6_hdr);
 721                 }
 722
 723                 hash = hash ^ l3hash;
 724                 hash ^= hash >> 16;
 725                 hash ^= hash >> 8;
 726
 727                 slaves[i] = hash % slave_count;
 728         }
 729 }
 730
 731 void
 732 burst_xmit_l34_hash(struct rte_mbuf **buf, uint16_t nb_pkts,
 733                 uint16_t slave_count, uint16_t *slaves)
 734 {
 735         struct rte_ether_hdr *eth_hdr;
 736         uint16_t proto;
 737         size_t vlan_offset;
 738         int i;
 739
 740         struct rte_udp_hdr *udp_hdr;
 741         struct rte_tcp_hdr *tcp_hdr;
 742         uint32_t hash, l3hash, l4hash;
 743
 744         for (i = 0; i < nb_pkts; i++) {
 745                 eth_hdr = rte_pktmbuf_mtod(buf[i], struct rte_ether_hdr *);
 746                 size_t pkt_end = (size_t)eth_hdr + rte_pktmbuf_data_len(buf[i]);
 747                 proto = eth_hdr->ether_type;
 748                 vlan_offset = get_vlan_offset(eth_hdr, &proto);
 749                 l3hash = 0;
 750                 l4hash = 0;
 751
 752                 if (rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4) == proto) {
 753                         struct rte_ipv4_hdr *ipv4_hdr = (struct rte_ipv4_hdr *)
 754                                         ((char *)(eth_hdr + 1) + vlan_offset);
 755                         size_t ip_hdr_offset;
 756
 757                         l3hash = ipv4_hash(ipv4_hdr);
 758
 759                         /* there is no L4 header in fragmented packet */
 760                         if (likely(rte_ipv4_frag_pkt_is_fragmented(ipv4_hdr)
 761                                                                 == 0)) {
 762                                 ip_hdr_offset = (ipv4_hdr->version_ihl
 763                                         & RTE_IPV4_HDR_IHL_MASK) *
 764                                         RTE_IPV4_IHL_MULTIPLIER;
 765
 766                                 if (ipv4_hdr->next_proto_id == IPPROTO_TCP) {
 767                                         tcp_hdr = (struct rte_tcp_hdr *)
 768                                                 ((char *)ipv4_hdr +
 769                                                         ip_hdr_offset);
 770                                         if ((size_t)tcp_hdr + sizeof(*tcp_hdr)
 771                                                         < pkt_end)
 772                                                 l4hash = HASH_L4_PORTS(tcp_hdr);
 773                                 } else if (ipv4_hdr->next_proto_id ==
 774                                                                 IPPROTO_UDP) {
 775                                         udp_hdr = (struct rte_udp_hdr *)
 776                                                 ((char *)ipv4_hdr +
 777                                                         ip_hdr_offset);
 778                                         if ((size_t)udp_hdr + sizeof(*udp_hdr)
 779                                                         < pkt_end)
 780                                                 l4hash = HASH_L4_PORTS(udp_hdr);
 781                                 }
 782                         }
 783                 } else if  (rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6) == proto) {
 784                         struct rte_ipv6_hdr *ipv6_hdr = (struct rte_ipv6_hdr *)
 785                                         ((char *)(eth_hdr + 1) + vlan_offset);
 786                         l3hash = ipv6_hash(ipv6_hdr);
 787
 788                         if (ipv6_hdr->proto == IPPROTO_TCP) {
 789                                 tcp_hdr = (struct rte_tcp_hdr *)(ipv6_hdr + 1);
 790                                 l4hash = HASH_L4_PORTS(tcp_hdr);
 791                         } else if (ipv6_hdr->proto == IPPROTO_UDP) {
 792                                 udp_hdr = (struct rte_udp_hdr *)(ipv6_hdr + 1);
 793                                 l4hash = HASH_L4_PORTS(udp_hdr);
 794                         }
 795                 }
 796
 797                 hash = l3hash ^ l4hash;
 798                 hash ^= hash >> 16;
 799                 hash ^= hash >> 8;
 800
 801                 slaves[i] = hash % slave_count;
 802         }
 803 }
 804
 805 struct bwg_slave {
 806         uint64_t bwg_left_int;
 807         uint64_t bwg_left_remainder;
 808         uint16_t slave;
 809 };
 810
 811 void
 812 bond_tlb_activate_slave(struct bond_dev_private *internals) {
 813         int i;
 814
 815         for (i = 0; i < internals->active_slave_count; i++) {
 816                 tlb_last_obytets[internals->active_slaves[i]] = 0;
 817         }
 818 }
 819
 820 static int
 821 bandwidth_cmp(const void *a, const void *b)
 822 {
 823         const struct bwg_slave *bwg_a = a;
 824         const struct bwg_slave *bwg_b = b;
 825         int64_t diff = (int64_t)bwg_b->bwg_left_int - (int64_t)bwg_a->bwg_left_int;
 826         int64_t diff2 = (int64_t)bwg_b->bwg_left_remainder -
 827                         (int64_t)bwg_a->bwg_left_remainder;
 828         if (diff > 0)
 829                 return 1;
 830         else if (diff < 0)
 831                 return -1;
 832         else if (diff2 > 0)
 833                 return 1;
 834         else if (diff2 < 0)
 835                 return -1;
 836         else
 837                 return 0;
 838 }
 839
 840 static void
 841 bandwidth_left(uint16_t port_id, uint64_t load, uint8_t update_idx,
 842                 struct bwg_slave *bwg_slave)
 843 {
 844         struct rte_eth_link link_status;
 845         int ret;
 846
 847         ret = rte_eth_link_get_nowait(port_id, &link_status);
 848         if (ret < 0) {
 849                 RTE_BOND_LOG(ERR, "Slave (port %u) link get failed: %s",
 850                              port_id, rte_strerror(-ret));
 851                 return;
 852         }
 853         uint64_t link_bwg = link_status.link_speed * 1000000ULL / 8;
 854         if (link_bwg == 0)
 855                 return;
 856         link_bwg = link_bwg * (update_idx+1) * REORDER_PERIOD_MS;
 857         bwg_slave->bwg_left_int = (link_bwg - 1000*load) / link_bwg;
 858         bwg_slave->bwg_left_remainder = (link_bwg - 1000*load) % link_bwg;
 859 }
 860
 861 static void
 862 bond_ethdev_update_tlb_slave_cb(void *arg)
 863 {
 864         struct bond_dev_private *internals = arg;
 865         struct rte_eth_stats slave_stats;
 866         struct bwg_slave bwg_array[RTE_MAX_ETHPORTS];
 867         uint16_t slave_count;
 868         uint64_t tx_bytes;
 869
 870         uint8_t update_stats = 0;
 871         uint16_t slave_id;
 872         uint16_t i;
 873
 874         internals->slave_update_idx++;
 875
 876
 877         if (internals->slave_update_idx >= REORDER_PERIOD_MS)
 878                 update_stats = 1;
 879
 880         for (i = 0; i < internals->active_slave_count; i++) {
 881                 slave_id = internals->active_slaves[i];
 882                 rte_eth_stats_get(slave_id, &slave_stats);
 883                 tx_bytes = slave_stats.obytes - tlb_last_obytets[slave_id];
 884                 bandwidth_left(slave_id, tx_bytes,
 885                                 internals->slave_update_idx, &bwg_array[i]);
 886                 bwg_array[i].slave = slave_id;
 887
 888                 if (update_stats) {
 889                         tlb_last_obytets[slave_id] = slave_stats.obytes;
 890                 }
 891         }
 892
 893         if (update_stats == 1)
 894                 internals->slave_update_idx = 0;
 895
 896         slave_count = i;
 897         qsort(bwg_array, slave_count, sizeof(bwg_array[0]), bandwidth_cmp);
 898         for (i = 0; i < slave_count; i++)
 899                 internals->tlb_slaves_order[i] = bwg_array[i].slave;
 900
 901         rte_eal_alarm_set(REORDER_PERIOD_MS * 1000, bond_ethdev_update_tlb_slave_cb,
 902                         (struct bond_dev_private *)internals);
 903 }
 904
 905 static uint16_t
 906 bond_ethdev_tx_burst_tlb(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 907 {
 908         struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue;
 909         struct bond_dev_private *internals = bd_tx_q->dev_private;
 910
 911         struct rte_eth_dev *primary_port =
 912                         &rte_eth_devices[internals->primary_port];
 913         uint16_t num_tx_total = 0;
 914         uint16_t i, j;
 915
 916         uint16_t num_of_slaves = internals->active_slave_count;
 917         uint16_t slaves[RTE_MAX_ETHPORTS];
 918
 919         struct rte_ether_hdr *ether_hdr;
 920         struct rte_ether_addr primary_slave_addr;
 921         struct rte_ether_addr active_slave_addr;
 922
 923         if (num_of_slaves < 1)
 924                 return num_tx_total;
 925
 926         memcpy(slaves, internals->tlb_slaves_order,
 927                                 sizeof(internals->tlb_slaves_order[0]) * num_of_slaves);
 928
 929
 930         rte_ether_addr_copy(primary_port->data->mac_addrs, &primary_slave_addr);
 931
 932         if (nb_pkts > 3) {
 933                 for (i = 0; i < 3; i++)
 934                         rte_prefetch0(rte_pktmbuf_mtod(bufs[i], void*));
 935         }
 936
 937         for (i = 0; i < num_of_slaves; i++) {
 938                 rte_eth_macaddr_get(slaves[i], &active_slave_addr);
 939                 for (j = num_tx_total; j < nb_pkts; j++) {
 940                         if (j + 3 < nb_pkts)
 941                                 rte_prefetch0(rte_pktmbuf_mtod(bufs[j+3], void*));
 942
 943                         ether_hdr = rte_pktmbuf_mtod(bufs[j],
 944                                                 struct rte_ether_hdr *);
 945                         if (rte_is_same_ether_addr(&ether_hdr->src_addr,
 946                                                         &primary_slave_addr))
 947                                 rte_ether_addr_copy(&active_slave_addr,
 948                                                 &ether_hdr->src_addr);
 949 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
 950                                         mode6_debug("TX IPv4:", ether_hdr, slaves[i], &burstnumberTX);
 951 #endif
 952                 }
 953
 954                 num_tx_total += rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
 955                                 bufs + num_tx_total, nb_pkts - num_tx_total);
 956
 957                 if (num_tx_total == nb_pkts)
 958                         break;
 959         }
 960
 961         return num_tx_total;
 962 }
 963
 964 void
 965 bond_tlb_disable(struct bond_dev_private *internals)
 966 {
 967         rte_eal_alarm_cancel(bond_ethdev_update_tlb_slave_cb, internals);
 968 }
 969
 970 void
 971 bond_tlb_enable(struct bond_dev_private *internals)
 972 {
 973         bond_ethdev_update_tlb_slave_cb(internals);
 974 }
 975
 976 static uint16_t
 977 bond_ethdev_tx_burst_alb(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 978 {
 979         struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue;
 980         struct bond_dev_private *internals = bd_tx_q->dev_private;
 981
 982         struct rte_ether_hdr *eth_h;
 983         uint16_t ether_type, offset;
 984
 985         struct client_data *client_info;
 986
 987         /*
 988          * We create transmit buffers for every slave and one additional to send
 989          * through tlb. In worst case every packet will be send on one port.
 990          */
 991         struct rte_mbuf *slave_bufs[RTE_MAX_ETHPORTS + 1][nb_pkts];
 992         uint16_t slave_bufs_pkts[RTE_MAX_ETHPORTS + 1] = { 0 };
 993
 994         /*
 995          * We create separate transmit buffers for update packets as they won't
 996          * be counted in num_tx_total.
 997          */
 998         struct rte_mbuf *update_bufs[RTE_MAX_ETHPORTS][ALB_HASH_TABLE_SIZE];
 999         uint16_t update_bufs_pkts[RTE_MAX_ETHPORTS] = { 0 };
1000
1001         struct rte_mbuf *upd_pkt;
1002         size_t pkt_size;
1003
1004         uint16_t num_send, num_not_send = 0;
1005         uint16_t num_tx_total = 0;
1006         uint16_t slave_idx;
1007
1008         int i, j;
1009
1010         /* Search tx buffer for ARP packets and forward them to alb */
1011         for (i = 0; i < nb_pkts; i++) {
1012                 eth_h = rte_pktmbuf_mtod(bufs[i], struct rte_ether_hdr *);
1013                 ether_type = eth_h->ether_type;
1014                 offset = get_vlan_offset(eth_h, &ether_type);
1015
1016                 if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_ARP)) {
1017                         slave_idx = bond_mode_alb_arp_xmit(eth_h, offset, internals);
1018
1019                         /* Change src mac in eth header */
1020                         rte_eth_macaddr_get(slave_idx, &eth_h->src_addr);
1021
1022                         /* Add packet to slave tx buffer */
1023                         slave_bufs[slave_idx][slave_bufs_pkts[slave_idx]] = bufs[i];
1024                         slave_bufs_pkts[slave_idx]++;
1025                 } else {
1026                         /* If packet is not ARP, send it with TLB policy */
1027                         slave_bufs[RTE_MAX_ETHPORTS][slave_bufs_pkts[RTE_MAX_ETHPORTS]] =
1028                                         bufs[i];
1029                         slave_bufs_pkts[RTE_MAX_ETHPORTS]++;
1030                 }
1031         }
1032
1033         /* Update connected client ARP tables */
1034         if (internals->mode6.ntt) {
1035                 for (i = 0; i < ALB_HASH_TABLE_SIZE; i++) {
1036                         client_info = &internals->mode6.client_table[i];
1037
1038                         if (client_info->in_use) {
1039                                 /* Allocate new packet to send ARP update on current slave */
1040                                 upd_pkt = rte_pktmbuf_alloc(internals->mode6.mempool);
1041                                 if (upd_pkt == NULL) {
1042                                         RTE_BOND_LOG(ERR,
1043                                                      "Failed to allocate ARP packet from pool");
1044                                         continue;
1045                                 }
1046                                 pkt_size = sizeof(struct rte_ether_hdr) +
1047                                         sizeof(struct rte_arp_hdr) +
1048                                         client_info->vlan_count *
1049                                         sizeof(struct rte_vlan_hdr);
1050                                 upd_pkt->data_len = pkt_size;
1051                                 upd_pkt->pkt_len = pkt_size;
1052
1053                                 slave_idx = bond_mode_alb_arp_upd(client_info, upd_pkt,
1054                                                 internals);
1055
1056                                 /* Add packet to update tx buffer */
1057                                 update_bufs[slave_idx][update_bufs_pkts[slave_idx]] = upd_pkt;
1058                                 update_bufs_pkts[slave_idx]++;
1059                         }
1060                 }
1061                 internals->mode6.ntt = 0;
1062         }
1063
1064         /* Send ARP packets on proper slaves */
1065         for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
1066                 if (slave_bufs_pkts[i] > 0) {
1067                         num_send = rte_eth_tx_burst(i, bd_tx_q->queue_id,
1068                                         slave_bufs[i], slave_bufs_pkts[i]);
1069                         for (j = 0; j < slave_bufs_pkts[i] - num_send; j++) {
1070                                 bufs[nb_pkts - 1 - num_not_send - j] =
1071                                                 slave_bufs[i][nb_pkts - 1 - j];
1072                         }
1073
1074                         num_tx_total += num_send;
1075                         num_not_send += slave_bufs_pkts[i] - num_send;
1076
1077 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
1078         /* Print TX stats including update packets */
1079                         for (j = 0; j < slave_bufs_pkts[i]; j++) {
1080                                 eth_h = rte_pktmbuf_mtod(slave_bufs[i][j],
1081                                                         struct rte_ether_hdr *);
1082                                 mode6_debug("TX ARP:", eth_h, i, &burstnumberTX);
1083                         }
1084 #endif
1085                 }
1086         }
1087
1088         /* Send update packets on proper slaves */
1089         for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
1090                 if (update_bufs_pkts[i] > 0) {
1091                         num_send = rte_eth_tx_burst(i, bd_tx_q->queue_id, update_bufs[i],
1092                                         update_bufs_pkts[i]);
1093                         for (j = num_send; j < update_bufs_pkts[i]; j++) {
1094                                 rte_pktmbuf_free(update_bufs[i][j]);
1095                         }
1096 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
1097                         for (j = 0; j < update_bufs_pkts[i]; j++) {
1098                                 eth_h = rte_pktmbuf_mtod(update_bufs[i][j],
1099                                                         struct rte_ether_hdr *);
1100                                 mode6_debug("TX ARPupd:", eth_h, i, &burstnumberTX);
1101                         }
1102 #endif
1103                 }
1104         }
1105
1106         /* Send non-ARP packets using tlb policy */
1107         if (slave_bufs_pkts[RTE_MAX_ETHPORTS] > 0) {
1108                 num_send = bond_ethdev_tx_burst_tlb(queue,
1109                                 slave_bufs[RTE_MAX_ETHPORTS],
1110                                 slave_bufs_pkts[RTE_MAX_ETHPORTS]);
1111
1112                 for (j = 0; j < slave_bufs_pkts[RTE_MAX_ETHPORTS]; j++) {
1113                         bufs[nb_pkts - 1 - num_not_send - j] =
1114                                         slave_bufs[RTE_MAX_ETHPORTS][nb_pkts - 1 - j];
1115                 }
1116
1117                 num_tx_total += num_send;
1118         }
1119
1120         return num_tx_total;
1121 }
1122
1123 static inline uint16_t
1124 tx_burst_balance(void *queue, struct rte_mbuf **bufs, uint16_t nb_bufs,
1125                  uint16_t *slave_port_ids, uint16_t slave_count)
1126 {
1127         struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue;
1128         struct bond_dev_private *internals = bd_tx_q->dev_private;
1129
1130         /* Array to sort mbufs for transmission on each slave into */
1131         struct rte_mbuf *slave_bufs[RTE_MAX_ETHPORTS][nb_bufs];
1132         /* Number of mbufs for transmission on each slave */
1133         uint16_t slave_nb_bufs[RTE_MAX_ETHPORTS] = { 0 };
1134         /* Mapping array generated by hash function to map mbufs to slaves */
1135         uint16_t bufs_slave_port_idxs[nb_bufs];
1136
1137         uint16_t slave_tx_count;
1138         uint16_t total_tx_count = 0, total_tx_fail_count = 0;
1139
1140         uint16_t i;
1141
1142         /*
1143          * Populate slaves mbuf with the packets which are to be sent on it
1144          * selecting output slave using hash based on xmit policy
1145          */
1146         internals->burst_xmit_hash(bufs, nb_bufs, slave_count,
1147                         bufs_slave_port_idxs);
1148
1149         for (i = 0; i < nb_bufs; i++) {
1150                 /* Populate slave mbuf arrays with mbufs for that slave. */
1151                 uint16_t slave_idx = bufs_slave_port_idxs[i];
1152
1153                 slave_bufs[slave_idx][slave_nb_bufs[slave_idx]++] = bufs[i];
1154         }
1155
1156         /* Send packet burst on each slave device */
1157         for (i = 0; i < slave_count; i++) {
1158                 if (slave_nb_bufs[i] == 0)
1159                         continue;
1160
1161                 slave_tx_count = rte_eth_tx_burst(slave_port_ids[i],
1162                                 bd_tx_q->queue_id, slave_bufs[i],
1163                                 slave_nb_bufs[i]);
1164
1165                 total_tx_count += slave_tx_count;
1166
1167                 /* If tx burst fails move packets to end of bufs */
1168                 if (unlikely(slave_tx_count < slave_nb_bufs[i])) {
1169                         int slave_tx_fail_count = slave_nb_bufs[i] -
1170                                         slave_tx_count;
1171                         total_tx_fail_count += slave_tx_fail_count;
1172                         memcpy(&bufs[nb_bufs - total_tx_fail_count],
1173                                &slave_bufs[i][slave_tx_count],
1174                                slave_tx_fail_count * sizeof(bufs[0]));
1175                 }
1176         }
1177
1178         return total_tx_count;
1179 }
1180
1181 static uint16_t
1182 bond_ethdev_tx_burst_balance(void *queue, struct rte_mbuf **bufs,
1183                 uint16_t nb_bufs)
1184 {
1185         struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue;
1186         struct bond_dev_private *internals = bd_tx_q->dev_private;
1187
1188         uint16_t slave_port_ids[RTE_MAX_ETHPORTS];
1189         uint16_t slave_count;
1190
1191         if (unlikely(nb_bufs == 0))
1192                 return 0;
1193
1194         /* Copy slave list to protect against slave up/down changes during tx
1195          * bursting
1196          */
1197         slave_count = internals->active_slave_count;
1198         if (unlikely(slave_count < 1))
1199                 return 0;
1200
1201         memcpy(slave_port_ids, internals->active_slaves,
1202                         sizeof(slave_port_ids[0]) * slave_count);
1203         return tx_burst_balance(queue, bufs, nb_bufs, slave_port_ids,
1204                                 slave_count);
1205 }
1206
1207 static inline uint16_t
1208 tx_burst_8023ad(void *queue, struct rte_mbuf **bufs, uint16_t nb_bufs,
1209                 bool dedicated_txq)
1210 {
1211         struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue;
1212         struct bond_dev_private *internals = bd_tx_q->dev_private;
1213
1214         uint16_t slave_port_ids[RTE_MAX_ETHPORTS];
1215         uint16_t slave_count;
1216
1217         uint16_t dist_slave_port_ids[RTE_MAX_ETHPORTS];
1218         uint16_t dist_slave_count;
1219
1220         uint16_t slave_tx_count;
1221
1222         uint16_t i;
1223
1224         /* Copy slave list to protect against slave up/down changes during tx
1225          * bursting */
1226         slave_count = internals->active_slave_count;
1227         if (unlikely(slave_count < 1))
1228                 return 0;
1229
1230         memcpy(slave_port_ids, internals->active_slaves,
1231                         sizeof(slave_port_ids[0]) * slave_count);
1232
1233         if (dedicated_txq)
1234                 goto skip_tx_ring;
1235
1236         /* Check for LACP control packets and send if available */
1237         for (i = 0; i < slave_count; i++) {
1238                 struct port *port = &bond_mode_8023ad_ports[slave_port_ids[i]];
1239                 struct rte_mbuf *ctrl_pkt = NULL;
1240
1241                 if (likely(rte_ring_empty(port->tx_ring)))
1242                         continue;
1243
1244                 if (rte_ring_dequeue(port->tx_ring,
1245                                      (void **)&ctrl_pkt) != -ENOENT) {
1246                         slave_tx_count = rte_eth_tx_burst(slave_port_ids[i],
1247                                         bd_tx_q->queue_id, &ctrl_pkt, 1);
1248                         /*
1249                          * re-enqueue LAG control plane packets to buffering
1250                          * ring if transmission fails so the packet isn't lost.
1251                          */
1252                         if (slave_tx_count != 1)
1253                                 rte_ring_enqueue(port->tx_ring, ctrl_pkt);
1254                 }
1255         }
1256
1257 skip_tx_ring:
1258         if (unlikely(nb_bufs == 0))
1259                 return 0;
1260
1261         dist_slave_count = 0;
1262         for (i = 0; i < slave_count; i++) {
1263                 struct port *port = &bond_mode_8023ad_ports[slave_port_ids[i]];
1264
1265                 if (ACTOR_STATE(port, DISTRIBUTING))
1266                         dist_slave_port_ids[dist_slave_count++] =
1267                                         slave_port_ids[i];
1268         }
1269
1270         if (unlikely(dist_slave_count < 1))
1271                 return 0;
1272
1273         return tx_burst_balance(queue, bufs, nb_bufs, dist_slave_port_ids,
1274                                 dist_slave_count);
1275 }
1276
1277 static uint16_t
1278 bond_ethdev_tx_burst_8023ad(void *queue, struct rte_mbuf **bufs,
1279                 uint16_t nb_bufs)
1280 {
1281         return tx_burst_8023ad(queue, bufs, nb_bufs, false);
1282 }
1283
1284 static uint16_t
1285 bond_ethdev_tx_burst_8023ad_fast_queue(void *queue, struct rte_mbuf **bufs,
1286                 uint16_t nb_bufs)
1287 {
1288         return tx_burst_8023ad(queue, bufs, nb_bufs, true);
1289 }
1290
1291 static uint16_t
1292 bond_ethdev_tx_burst_broadcast(void *queue, struct rte_mbuf **bufs,
1293                 uint16_t nb_pkts)
1294 {
1295         struct bond_dev_private *internals;
1296         struct bond_tx_queue *bd_tx_q;
1297
1298         uint16_t slaves[RTE_MAX_ETHPORTS];
1299         uint8_t tx_failed_flag = 0;
1300         uint16_t num_of_slaves;
1301
1302         uint16_t max_nb_of_tx_pkts = 0;
1303
1304         int slave_tx_total[RTE_MAX_ETHPORTS];
1305         int i, most_successful_tx_slave = -1;
1306
1307         bd_tx_q = (struct bond_tx_queue *)queue;
1308         internals = bd_tx_q->dev_private;
1309
1310         /* Copy slave list to protect against slave up/down changes during tx
1311          * bursting */
1312         num_of_slaves = internals->active_slave_count;
1313         memcpy(slaves, internals->active_slaves,
1314                         sizeof(internals->active_slaves[0]) * num_of_slaves);
1315
1316         if (num_of_slaves < 1)
1317                 return 0;
1318
1319         /* Increment reference count on mbufs */
1320         for (i = 0; i < nb_pkts; i++)
1321                 rte_mbuf_refcnt_update(bufs[i], num_of_slaves - 1);
1322
1323         /* Transmit burst on each active slave */
1324         for (i = 0; i < num_of_slaves; i++) {
1325                 slave_tx_total[i] = rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
1326                                         bufs, nb_pkts);
1327
1328                 if (unlikely(slave_tx_total[i] < nb_pkts))
1329                         tx_failed_flag = 1;
1330
1331                 /* record the value and slave index for the slave which transmits the
1332                  * maximum number of packets */
1333                 if (slave_tx_total[i] > max_nb_of_tx_pkts) {
1334                         max_nb_of_tx_pkts = slave_tx_total[i];
1335                         most_successful_tx_slave = i;
1336                 }
1337         }
1338
1339         /* if slaves fail to transmit packets from burst, the calling application
1340          * is not expected to know about multiple references to packets so we must
1341          * handle failures of all packets except those of the most successful slave
1342          */
1343         if (unlikely(tx_failed_flag))
1344                 for (i = 0; i < num_of_slaves; i++)
1345                         if (i != most_successful_tx_slave)
1346                                 while (slave_tx_total[i] < nb_pkts)
1347                                         rte_pktmbuf_free(bufs[slave_tx_total[i]++]);
1348
1349         return max_nb_of_tx_pkts;
1350 }
1351
1352 static void
1353 link_properties_set(struct rte_eth_dev *ethdev, struct rte_eth_link *slave_link)
1354 {
1355         struct bond_dev_private *bond_ctx = ethdev->data->dev_private;
1356
1357         if (bond_ctx->mode == BONDING_MODE_8023AD) {
1358                 /**
1359                  * If in mode 4 then save the link properties of the first
1360                  * slave, all subsequent slaves must match these properties
1361                  */
1362                 struct rte_eth_link *bond_link = &bond_ctx->mode4.slave_link;
1363
1364                 bond_link->link_autoneg = slave_link->link_autoneg;
1365                 bond_link->link_duplex = slave_link->link_duplex;
1366                 bond_link->link_speed = slave_link->link_speed;
1367         } else {
1368                 /**
1369                  * In any other mode the link properties are set to default
1370                  * values of AUTONEG/DUPLEX
1371                  */
1372                 ethdev->data->dev_link.link_autoneg = ETH_LINK_AUTONEG;
1373                 ethdev->data->dev_link.link_duplex = ETH_LINK_FULL_DUPLEX;
1374         }
1375 }
1376
1377 static int
1378 link_properties_valid(struct rte_eth_dev *ethdev,
1379                 struct rte_eth_link *slave_link)
1380 {
1381         struct bond_dev_private *bond_ctx = ethdev->data->dev_private;
1382
1383         if (bond_ctx->mode == BONDING_MODE_8023AD) {
1384                 struct rte_eth_link *bond_link = &bond_ctx->mode4.slave_link;
1385
1386                 if (bond_link->link_duplex != slave_link->link_duplex ||
1387                         bond_link->link_autoneg != slave_link->link_autoneg ||
1388                         bond_link->link_speed != slave_link->link_speed)
1389                         return -1;
1390         }
1391
1392         return 0;
1393 }
1394
1395 int
1396 mac_address_get(struct rte_eth_dev *eth_dev,
1397                 struct rte_ether_addr *dst_mac_addr)
1398 {
1399         struct rte_ether_addr *mac_addr;
1400
1401         if (eth_dev == NULL) {
1402                 RTE_BOND_LOG(ERR, "NULL pointer eth_dev specified");
1403                 return -1;
1404         }
1405
1406         if (dst_mac_addr == NULL) {
1407                 RTE_BOND_LOG(ERR, "NULL pointer MAC specified");
1408                 return -1;
1409         }
1410
1411         mac_addr = eth_dev->data->mac_addrs;
1412
1413         rte_ether_addr_copy(mac_addr, dst_mac_addr);
1414         return 0;
1415 }
1416
1417 int
1418 mac_address_set(struct rte_eth_dev *eth_dev,
1419                 struct rte_ether_addr *new_mac_addr)
1420 {
1421         struct rte_ether_addr *mac_addr;
1422
1423         if (eth_dev == NULL) {
1424                 RTE_BOND_LOG(ERR, "NULL pointer eth_dev specified");
1425                 return -1;
1426         }
1427
1428         if (new_mac_addr == NULL) {
1429                 RTE_BOND_LOG(ERR, "NULL pointer MAC specified");
1430                 return -1;
1431         }
1432
1433         mac_addr = eth_dev->data->mac_addrs;
1434
1435         /* If new MAC is different to current MAC then update */
1436         if (memcmp(mac_addr, new_mac_addr, sizeof(*mac_addr)) != 0)
1437                 memcpy(mac_addr, new_mac_addr, sizeof(*mac_addr));
1438
1439         return 0;
1440 }
1441
1442 static const struct rte_ether_addr null_mac_addr;
1443
1444 /*
1445  * Add additional MAC addresses to the slave
1446  */
1447 int
1448 slave_add_mac_addresses(struct rte_eth_dev *bonded_eth_dev,
1449                 uint16_t slave_port_id)
1450 {
1451         int i, ret;
1452         struct rte_ether_addr *mac_addr;
1453
1454         for (i = 1; i < BOND_MAX_MAC_ADDRS; i++) {
1455                 mac_addr = &bonded_eth_dev->data->mac_addrs[i];
1456                 if (rte_is_same_ether_addr(mac_addr, &null_mac_addr))
1457                         break;
1458
1459                 ret = rte_eth_dev_mac_addr_add(slave_port_id, mac_addr, 0);
1460                 if (ret < 0) {
1461                         /* rollback */
1462                         for (i--; i > 0; i--)
1463                                 rte_eth_dev_mac_addr_remove(slave_port_id,
1464                                         &bonded_eth_dev->data->mac_addrs[i]);
1465                         return ret;
1466                 }
1467         }
1468
1469         return 0;
1470 }
1471
1472 /*
1473  * Remove additional MAC addresses from the slave
1474  */
1475 int
1476 slave_remove_mac_addresses(struct rte_eth_dev *bonded_eth_dev,
1477                 uint16_t slave_port_id)
1478 {
1479         int i, rc, ret;
1480         struct rte_ether_addr *mac_addr;
1481
1482         rc = 0;
1483         for (i = 1; i < BOND_MAX_MAC_ADDRS; i++) {
1484                 mac_addr = &bonded_eth_dev->data->mac_addrs[i];
1485                 if (rte_is_same_ether_addr(mac_addr, &null_mac_addr))
1486                         break;
1487
1488                 ret = rte_eth_dev_mac_addr_remove(slave_port_id, mac_addr);
1489                 /* save only the first error */
1490                 if (ret < 0 && rc == 0)
1491                         rc = ret;
1492         }
1493
1494         return rc;
1495 }
1496
1497 int
1498 mac_address_slaves_update(struct rte_eth_dev *bonded_eth_dev)
1499 {
1500         struct bond_dev_private *internals = bonded_eth_dev->data->dev_private;
1501         bool set;
1502         int i;
1503
1504         /* Update slave devices MAC addresses */
1505         if (internals->slave_count < 1)
1506                 return -1;
1507
1508         switch (internals->mode) {
1509         case BONDING_MODE_ROUND_ROBIN:
1510         case BONDING_MODE_BALANCE:
1511         case BONDING_MODE_BROADCAST:
1512                 for (i = 0; i < internals->slave_count; i++) {
1513                         if (rte_eth_dev_default_mac_addr_set(
1514                                         internals->slaves[i].port_id,
1515                                         bonded_eth_dev->data->mac_addrs)) {
1516                                 RTE_BOND_LOG(ERR, "Failed to update port Id %d MAC address",
1517                                                 internals->slaves[i].port_id);
1518                                 return -1;
1519                         }
1520                 }
1521                 break;
1522         case BONDING_MODE_8023AD:
1523                 bond_mode_8023ad_mac_address_update(bonded_eth_dev);
1524                 break;
1525         case BONDING_MODE_ACTIVE_BACKUP:
1526         case BONDING_MODE_TLB:
1527         case BONDING_MODE_ALB:
1528         default:
1529                 set = true;
1530                 for (i = 0; i < internals->slave_count; i++) {
1531                         if (internals->slaves[i].port_id ==
1532                                         internals->current_primary_port) {
1533                                 if (rte_eth_dev_default_mac_addr_set(
1534                                                 internals->current_primary_port,
1535                                                 bonded_eth_dev->data->mac_addrs)) {
1536                                         RTE_BOND_LOG(ERR, "Failed to update port Id %d MAC address",
1537                                                         internals->current_primary_port);
1538                                         set = false;
1539                                 }
1540                         } else {
1541                                 if (rte_eth_dev_default_mac_addr_set(
1542                                                 internals->slaves[i].port_id,
1543                                                 &internals->slaves[i].persisted_mac_addr)) {
1544                                         RTE_BOND_LOG(ERR, "Failed to update port Id %d MAC address",
1545                                                         internals->slaves[i].port_id);
1546                                 }
1547                         }
1548                 }
1549                 if (!set)
1550                         return -1;
1551         }
1552
1553         return 0;
1554 }
1555
1556 int
1557 bond_ethdev_mode_set(struct rte_eth_dev *eth_dev, int mode)
1558 {
1559         struct bond_dev_private *internals;
1560
1561         internals = eth_dev->data->dev_private;
1562
1563         switch (mode) {
1564         case BONDING_MODE_ROUND_ROBIN:
1565                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_round_robin;
1566                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst;
1567                 break;
1568         case BONDING_MODE_ACTIVE_BACKUP:
1569                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_active_backup;
1570                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_active_backup;
1571                 break;
1572         case BONDING_MODE_BALANCE:
1573                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_balance;
1574                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst;
1575                 break;
1576         case BONDING_MODE_BROADCAST:
1577                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_broadcast;
1578                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst;
1579                 break;
1580         case BONDING_MODE_8023AD:
1581                 if (bond_mode_8023ad_enable(eth_dev) != 0)
1582                         return -1;
1583
1584                 if (internals->mode4.dedicated_queues.enabled == 0) {
1585                         eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_8023ad;
1586                         eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_8023ad;
1587                         RTE_BOND_LOG(WARNING,
1588                                 "Using mode 4, it is necessary to do TX burst "
1589                                 "and RX burst at least every 100ms.");
1590                 } else {
1591                         /* Use flow director's optimization */
1592                         eth_dev->rx_pkt_burst =
1593                                         bond_ethdev_rx_burst_8023ad_fast_queue;
1594                         eth_dev->tx_pkt_burst =
1595                                         bond_ethdev_tx_burst_8023ad_fast_queue;
1596                 }
1597                 break;
1598         case BONDING_MODE_TLB:
1599                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_tlb;
1600                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_active_backup;
1601                 break;
1602         case BONDING_MODE_ALB:
1603                 if (bond_mode_alb_enable(eth_dev) != 0)
1604                         return -1;
1605
1606                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_alb;
1607                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_alb;
1608                 break;
1609         default:
1610                 return -1;
1611         }
1612
1613         internals->mode = mode;
1614
1615         return 0;
1616 }
1617
1618
1619 static int
1620 slave_configure_slow_queue(struct rte_eth_dev *bonded_eth_dev,
1621                 struct rte_eth_dev *slave_eth_dev)
1622 {
1623         int errval = 0;
1624         struct bond_dev_private *internals = bonded_eth_dev->data->dev_private;
1625         struct port *port = &bond_mode_8023ad_ports[slave_eth_dev->data->port_id];
1626
1627         if (port->slow_pool == NULL) {
1628                 char mem_name[256];
1629                 int slave_id = slave_eth_dev->data->port_id;
1630
1631                 snprintf(mem_name, RTE_DIM(mem_name), "slave_port%u_slow_pool",
1632                                 slave_id);
1633                 port->slow_pool = rte_pktmbuf_pool_create(mem_name, 8191,
1634                         250, 0, RTE_MBUF_DEFAULT_BUF_SIZE,
1635                         slave_eth_dev->data->numa_node);
1636
1637                 /* Any memory allocation failure in initialization is critical because
1638                  * resources can't be free, so reinitialization is impossible. */
1639                 if (port->slow_pool == NULL) {
1640                         rte_panic("Slave %u: Failed to create memory pool '%s': %s\n",
1641                                 slave_id, mem_name, rte_strerror(rte_errno));
1642                 }
1643         }
1644
1645         if (internals->mode4.dedicated_queues.enabled == 1) {
1646                 /* Configure slow Rx queue */
1647
1648                 errval = rte_eth_rx_queue_setup(slave_eth_dev->data->port_id,
1649                                 internals->mode4.dedicated_queues.rx_qid, 128,
1650                                 rte_eth_dev_socket_id(slave_eth_dev->data->port_id),
1651                                 NULL, port->slow_pool);
1652                 if (errval != 0) {
1653                         RTE_BOND_LOG(ERR,
1654                                         "rte_eth_rx_queue_setup: port=%d queue_id %d, err (%d)",
1655                                         slave_eth_dev->data->port_id,
1656                                         internals->mode4.dedicated_queues.rx_qid,
1657                                         errval);
1658                         return errval;
1659                 }
1660
1661                 errval = rte_eth_tx_queue_setup(slave_eth_dev->data->port_id,
1662                                 internals->mode4.dedicated_queues.tx_qid, 512,
1663                                 rte_eth_dev_socket_id(slave_eth_dev->data->port_id),
1664                                 NULL);
1665                 if (errval != 0) {
1666                         RTE_BOND_LOG(ERR,
1667                                 "rte_eth_tx_queue_setup: port=%d queue_id %d, err (%d)",
1668                                 slave_eth_dev->data->port_id,
1669                                 internals->mode4.dedicated_queues.tx_qid,
1670                                 errval);
1671                         return errval;
1672                 }
1673         }
1674         return 0;
1675 }
1676
1677 int
1678 slave_configure(struct rte_eth_dev *bonded_eth_dev,
1679                 struct rte_eth_dev *slave_eth_dev)
1680 {
1681         struct bond_rx_queue *bd_rx_q;
1682         struct bond_tx_queue *bd_tx_q;
1683         uint16_t nb_rx_queues;
1684         uint16_t nb_tx_queues;
1685
1686         int errval;
1687         uint16_t q_id;
1688         struct rte_flow_error flow_error;
1689
1690         struct bond_dev_private *internals = bonded_eth_dev->data->dev_private;
1691
1692         /* Stop slave */
1693         errval = rte_eth_dev_stop(slave_eth_dev->data->port_id);
1694         if (errval != 0)
1695                 RTE_BOND_LOG(ERR, "rte_eth_dev_stop: port %u, err (%d)",
1696                              slave_eth_dev->data->port_id, errval);
1697
1698         /* Enable interrupts on slave device if supported */
1699         if (slave_eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1700                 slave_eth_dev->data->dev_conf.intr_conf.lsc = 1;
1701
1702         /* If RSS is enabled for bonding, try to enable it for slaves  */
1703         if (bonded_eth_dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG) {
1704                 if (internals->rss_key_len != 0) {
1705                         slave_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key_len =
1706                                         internals->rss_key_len;
1707                         slave_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key =
1708                                         internals->rss_key;
1709                 } else {
1710                         slave_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key = NULL;
1711                 }
1712
1713                 slave_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf =
1714                                 bonded_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf;
1715                 slave_eth_dev->data->dev_conf.rxmode.mq_mode =
1716                                 bonded_eth_dev->data->dev_conf.rxmode.mq_mode;
1717         }
1718
1719         if (bonded_eth_dev->data->dev_conf.rxmode.offloads &
1720                         DEV_RX_OFFLOAD_VLAN_FILTER)
1721                 slave_eth_dev->data->dev_conf.rxmode.offloads |=
1722                                 DEV_RX_OFFLOAD_VLAN_FILTER;
1723         else
1724                 slave_eth_dev->data->dev_conf.rxmode.offloads &=
1725                                 ~DEV_RX_OFFLOAD_VLAN_FILTER;
1726
1727         slave_eth_dev->data->dev_conf.rxmode.max_rx_pkt_len =
1728                         bonded_eth_dev->data->dev_conf.rxmode.max_rx_pkt_len;
1729
1730         if (bonded_eth_dev->data->dev_conf.rxmode.offloads &
1731                         DEV_RX_OFFLOAD_JUMBO_FRAME)
1732                 slave_eth_dev->data->dev_conf.rxmode.offloads |=
1733                                 DEV_RX_OFFLOAD_JUMBO_FRAME;
1734         else
1735                 slave_eth_dev->data->dev_conf.rxmode.offloads &=
1736                                 ~DEV_RX_OFFLOAD_JUMBO_FRAME;
1737
1738         nb_rx_queues = bonded_eth_dev->data->nb_rx_queues;
1739         nb_tx_queues = bonded_eth_dev->data->nb_tx_queues;
1740
1741         if (internals->mode == BONDING_MODE_8023AD) {
1742                 if (internals->mode4.dedicated_queues.enabled == 1) {
1743                         nb_rx_queues++;
1744                         nb_tx_queues++;
1745                 }
1746         }
1747
1748         errval = rte_eth_dev_set_mtu(slave_eth_dev->data->port_id,
1749                                      bonded_eth_dev->data->mtu);
1750         if (errval != 0 && errval != -ENOTSUP) {
1751                 RTE_BOND_LOG(ERR, "rte_eth_dev_set_mtu: port %u, err (%d)",
1752                                 slave_eth_dev->data->port_id, errval);
1753                 return errval;
1754         }
1755
1756         /* Configure device */
1757         errval = rte_eth_dev_configure(slave_eth_dev->data->port_id,
1758                         nb_rx_queues, nb_tx_queues,
1759                         &(slave_eth_dev->data->dev_conf));
1760         if (errval != 0) {
1761                 RTE_BOND_LOG(ERR, "Cannot configure slave device: port %u, err (%d)",
1762                                 slave_eth_dev->data->port_id, errval);
1763                 return errval;
1764         }
1765
1766         /* Setup Rx Queues */
1767         for (q_id = 0; q_id < bonded_eth_dev->data->nb_rx_queues; q_id++) {
1768                 bd_rx_q = (struct bond_rx_queue *)bonded_eth_dev->data->rx_queues[q_id];
1769
1770                 errval = rte_eth_rx_queue_setup(slave_eth_dev->data->port_id, q_id,
1771                                 bd_rx_q->nb_rx_desc,
1772                                 rte_eth_dev_socket_id(slave_eth_dev->data->port_id),
1773                                 &(bd_rx_q->rx_conf), bd_rx_q->mb_pool);
1774                 if (errval != 0) {
1775                         RTE_BOND_LOG(ERR,
1776                                         "rte_eth_rx_queue_setup: port=%d queue_id %d, err (%d)",
1777                                         slave_eth_dev->data->port_id, q_id, errval);
1778                         return errval;
1779                 }
1780         }
1781
1782         /* Setup Tx Queues */
1783         for (q_id = 0; q_id < bonded_eth_dev->data->nb_tx_queues; q_id++) {
1784                 bd_tx_q = (struct bond_tx_queue *)bonded_eth_dev->data->tx_queues[q_id];
1785
1786                 errval = rte_eth_tx_queue_setup(slave_eth_dev->data->port_id, q_id,
1787                                 bd_tx_q->nb_tx_desc,
1788                                 rte_eth_dev_socket_id(slave_eth_dev->data->port_id),
1789                                 &bd_tx_q->tx_conf);
1790                 if (errval != 0) {
1791                         RTE_BOND_LOG(ERR,
1792                                 "rte_eth_tx_queue_setup: port=%d queue_id %d, err (%d)",
1793                                 slave_eth_dev->data->port_id, q_id, errval);
1794                         return errval;
1795                 }
1796         }
1797
1798         if (internals->mode == BONDING_MODE_8023AD &&
1799                         internals->mode4.dedicated_queues.enabled == 1) {
1800                 if (slave_configure_slow_queue(bonded_eth_dev, slave_eth_dev)
1801                                 != 0)
1802                         return errval;
1803
1804                 errval = bond_ethdev_8023ad_flow_verify(bonded_eth_dev,
1805                                 slave_eth_dev->data->port_id);
1806                 if (errval != 0) {
1807                         RTE_BOND_LOG(ERR,
1808                                 "bond_ethdev_8023ad_flow_verify: port=%d, err (%d)",
1809                                 slave_eth_dev->data->port_id, errval);
1810                         return errval;
1811                 }
1812
1813                 if (internals->mode4.dedicated_queues.flow[slave_eth_dev->data->port_id] != NULL)
1814                         rte_flow_destroy(slave_eth_dev->data->port_id,
1815                                         internals->mode4.dedicated_queues.flow[slave_eth_dev->data->port_id],
1816                                         &flow_error);
1817
1818                 errval = bond_ethdev_8023ad_flow_set(bonded_eth_dev,
1819                                 slave_eth_dev->data->port_id);
1820                 if (errval != 0) {
1821                         RTE_BOND_LOG(ERR,
1822                                 "bond_ethdev_8023ad_flow_set: port=%d, err (%d)",
1823                                 slave_eth_dev->data->port_id, errval);
1824                         return errval;
1825                 }
1826         }
1827
1828         /* Start device */
1829         errval = rte_eth_dev_start(slave_eth_dev->data->port_id);
1830         if (errval != 0) {
1831                 RTE_BOND_LOG(ERR, "rte_eth_dev_start: port=%u, err (%d)",
1832                                 slave_eth_dev->data->port_id, errval);
1833                 return -1;
1834         }
1835
1836         /* If RSS is enabled for bonding, synchronize RETA */
1837         if (bonded_eth_dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS) {
1838                 int i;
1839                 struct bond_dev_private *internals;
1840
1841                 internals = bonded_eth_dev->data->dev_private;
1842
1843                 for (i = 0; i < internals->slave_count; i++) {
1844                         if (internals->slaves[i].port_id == slave_eth_dev->data->port_id) {
1845                                 errval = rte_eth_dev_rss_reta_update(
1846                                                 slave_eth_dev->data->port_id,
1847                                                 &internals->reta_conf[0],
1848                                                 internals->slaves[i].reta_size);
1849                                 if (errval != 0) {
1850                                         RTE_BOND_LOG(WARNING,
1851                                                      "rte_eth_dev_rss_reta_update on slave port %d fails (err %d)."
1852                                                      " RSS Configuration for bonding may be inconsistent.",
1853                                                      slave_eth_dev->data->port_id, errval);
1854                                 }
1855                                 break;
1856                         }
1857                 }
1858         }
1859
1860         /* If lsc interrupt is set, check initial slave's link status */
1861         if (slave_eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
1862                 slave_eth_dev->dev_ops->link_update(slave_eth_dev, 0);
1863                 bond_ethdev_lsc_event_callback(slave_eth_dev->data->port_id,
1864                         RTE_ETH_EVENT_INTR_LSC, &bonded_eth_dev->data->port_id,
1865                         NULL);
1866         }
1867
1868         return 0;
1869 }
1870
1871 void
1872 slave_remove(struct bond_dev_private *internals,
1873                 struct rte_eth_dev *slave_eth_dev)
1874 {
1875         uint16_t i;
1876
1877         for (i = 0; i < internals->slave_count; i++)
1878                 if (internals->slaves[i].port_id ==
1879                                 slave_eth_dev->data->port_id)
1880                         break;
1881
1882         if (i < (internals->slave_count - 1)) {
1883                 struct rte_flow *flow;
1884
1885                 memmove(&internals->slaves[i], &internals->slaves[i + 1],
1886                                 sizeof(internals->slaves[0]) *
1887                                 (internals->slave_count - i - 1));
1888                 TAILQ_FOREACH(flow, &internals->flow_list, next) {
1889                         memmove(&flow->flows[i], &flow->flows[i + 1],
1890                                 sizeof(flow->flows[0]) *
1891                                 (internals->slave_count - i - 1));
1892                         flow->flows[internals->slave_count - 1] = NULL;
1893                 }
1894         }
1895
1896         internals->slave_count--;
1897
1898         /* force reconfiguration of slave interfaces */
1899         rte_eth_dev_internal_reset(slave_eth_dev);
1900 }
1901
1902 static void
1903 bond_ethdev_slave_link_status_change_monitor(void *cb_arg);
1904
1905 void
1906 slave_add(struct bond_dev_private *internals,
1907                 struct rte_eth_dev *slave_eth_dev)
1908 {
1909         struct bond_slave_details *slave_details =
1910                         &internals->slaves[internals->slave_count];
1911
1912         slave_details->port_id = slave_eth_dev->data->port_id;
1913         slave_details->last_link_status = 0;
1914
1915         /* Mark slave devices that don't support interrupts so we can
1916          * compensate when we start the bond
1917          */
1918         if (!(slave_eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
1919                 slave_details->link_status_poll_enabled = 1;
1920         }
1921
1922         slave_details->link_status_wait_to_complete = 0;
1923         /* clean tlb_last_obytes when adding port for bonding device */
1924         memcpy(&(slave_details->persisted_mac_addr), slave_eth_dev->data->mac_addrs,
1925                         sizeof(struct rte_ether_addr));
1926 }
1927
1928 void
1929 bond_ethdev_primary_set(struct bond_dev_private *internals,
1930                 uint16_t slave_port_id)
1931 {
1932         int i;
1933
1934         if (internals->active_slave_count < 1)
1935                 internals->current_primary_port = slave_port_id;
1936         else
1937                 /* Search bonded device slave ports for new proposed primary port */
1938                 for (i = 0; i < internals->active_slave_count; i++) {
1939                         if (internals->active_slaves[i] == slave_port_id)
1940                                 internals->current_primary_port = slave_port_id;
1941                 }
1942 }
1943
1944 static int
1945 bond_ethdev_promiscuous_enable(struct rte_eth_dev *eth_dev);
1946
1947 static int
1948 bond_ethdev_start(struct rte_eth_dev *eth_dev)
1949 {
1950         struct bond_dev_private *internals;
1951         int i;
1952
1953         /* slave eth dev will be started by bonded device */
1954         if (check_for_bonded_ethdev(eth_dev)) {
1955                 RTE_BOND_LOG(ERR, "User tried to explicitly start a slave eth_dev (%d)",
1956                                 eth_dev->data->port_id);
1957                 return -1;
1958         }
1959
1960         eth_dev->data->dev_link.link_status = ETH_LINK_DOWN;
1961         eth_dev->data->dev_started = 1;
1962
1963         internals = eth_dev->data->dev_private;
1964
1965         if (internals->slave_count == 0) {
1966                 RTE_BOND_LOG(ERR, "Cannot start port since there are no slave devices");
1967                 goto out_err;
1968         }
1969
1970         if (internals->user_defined_mac == 0) {
1971                 struct rte_ether_addr *new_mac_addr = NULL;
1972
1973                 for (i = 0; i < internals->slave_count; i++)
1974                         if (internals->slaves[i].port_id == internals->primary_port)
1975                                 new_mac_addr = &internals->slaves[i].persisted_mac_addr;
1976
1977                 if (new_mac_addr == NULL)
1978                         goto out_err;
1979
1980                 if (mac_address_set(eth_dev, new_mac_addr) != 0) {
1981                         RTE_BOND_LOG(ERR, "bonded port (%d) failed to update MAC address",
1982                                         eth_dev->data->port_id);
1983                         goto out_err;
1984                 }
1985         }
1986
1987         if (internals->mode == BONDING_MODE_8023AD) {
1988                 if (internals->mode4.dedicated_queues.enabled == 1) {
1989                         internals->mode4.dedicated_queues.rx_qid =
1990                                         eth_dev->data->nb_rx_queues;
1991                         internals->mode4.dedicated_queues.tx_qid =
1992                                         eth_dev->data->nb_tx_queues;
1993                 }
1994         }
1995
1996
1997         /* Reconfigure each slave device if starting bonded device */
1998         for (i = 0; i < internals->slave_count; i++) {
1999                 struct rte_eth_dev *slave_ethdev =
2000                                 &(rte_eth_devices[internals->slaves[i].port_id]);
2001                 if (slave_configure(eth_dev, slave_ethdev) != 0) {
2002                         RTE_BOND_LOG(ERR,
2003                                 "bonded port (%d) failed to reconfigure slave device (%d)",
2004                                 eth_dev->data->port_id,
2005                                 internals->slaves[i].port_id);
2006                         goto out_err;
2007                 }
2008                 /* We will need to poll for link status if any slave doesn't
2009                  * support interrupts
2010                  */
2011                 if (internals->slaves[i].link_status_poll_enabled)
2012                         internals->link_status_polling_enabled = 1;
2013         }
2014
2015         /* start polling if needed */
2016         if (internals->link_status_polling_enabled) {
2017                 rte_eal_alarm_set(
2018                         internals->link_status_polling_interval_ms * 1000,
2019                         bond_ethdev_slave_link_status_change_monitor,
2020                         (void *)&rte_eth_devices[internals->port_id]);
2021         }
2022
2023         /* Update all slave devices MACs*/
2024         if (mac_address_slaves_update(eth_dev) != 0)
2025                 goto out_err;
2026
2027         if (internals->user_defined_primary_port)
2028                 bond_ethdev_primary_set(internals, internals->primary_port);
2029
2030         if (internals->mode == BONDING_MODE_8023AD)
2031                 bond_mode_8023ad_start(eth_dev);
2032
2033         if (internals->mode == BONDING_MODE_TLB ||
2034                         internals->mode == BONDING_MODE_ALB)
2035                 bond_tlb_enable(internals);
2036
2037         return 0;
2038
2039 out_err:
2040         eth_dev->data->dev_started = 0;
2041         return -1;
2042 }
2043
2044 static void
2045 bond_ethdev_free_queues(struct rte_eth_dev *dev)
2046 {
2047         uint16_t i;
2048
2049         if (dev->data->rx_queues != NULL) {
2050                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2051                         rte_free(dev->data->rx_queues[i]);
2052                         dev->data->rx_queues[i] = NULL;
2053                 }
2054                 dev->data->nb_rx_queues = 0;
2055         }
2056
2057         if (dev->data->tx_queues != NULL) {
2058                 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2059                         rte_free(dev->data->tx_queues[i]);
2060                         dev->data->tx_queues[i] = NULL;
2061                 }
2062                 dev->data->nb_tx_queues = 0;
2063         }
2064 }
2065
2066 int
2067 bond_ethdev_stop(struct rte_eth_dev *eth_dev)
2068 {
2069         struct bond_dev_private *internals = eth_dev->data->dev_private;
2070         uint16_t i;
2071         int ret;
2072
2073         if (internals->mode == BONDING_MODE_8023AD) {
2074                 struct port *port;
2075                 void *pkt = NULL;
2076
2077                 bond_mode_8023ad_stop(eth_dev);
2078
2079                 /* Discard all messages to/from mode 4 state machines */
2080                 for (i = 0; i < internals->active_slave_count; i++) {
2081                         port = &bond_mode_8023ad_ports[internals->active_slaves[i]];
2082
2083                         RTE_ASSERT(port->rx_ring != NULL);
2084                         while (rte_ring_dequeue(port->rx_ring, &pkt) != -ENOENT)
2085                                 rte_pktmbuf_free(pkt);
2086
2087                         RTE_ASSERT(port->tx_ring != NULL);
2088                         while (rte_ring_dequeue(port->tx_ring, &pkt) != -ENOENT)
2089                                 rte_pktmbuf_free(pkt);
2090                 }
2091         }
2092
2093         if (internals->mode == BONDING_MODE_TLB ||
2094                         internals->mode == BONDING_MODE_ALB) {
2095                 bond_tlb_disable(internals);
2096                 for (i = 0; i < internals->active_slave_count; i++)
2097                         tlb_last_obytets[internals->active_slaves[i]] = 0;
2098         }
2099
2100         eth_dev->data->dev_link.link_status = ETH_LINK_DOWN;
2101         eth_dev->data->dev_started = 0;
2102
2103         internals->link_status_polling_enabled = 0;
2104         for (i = 0; i < internals->slave_count; i++) {
2105                 uint16_t slave_id = internals->slaves[i].port_id;
2106                 if (find_slave_by_id(internals->active_slaves,
2107                                 internals->active_slave_count, slave_id) !=
2108                                                 internals->active_slave_count) {
2109                         internals->slaves[i].last_link_status = 0;
2110                         ret = rte_eth_dev_stop(slave_id);
2111                         if (ret != 0) {
2112                                 RTE_BOND_LOG(ERR, "Failed to stop device on port %u",
2113                                              slave_id);
2114                                 return ret;
2115                         }
2116                         deactivate_slave(eth_dev, slave_id);
2117                 }
2118         }
2119
2120         return 0;
2121 }
2122
2123 int
2124 bond_ethdev_close(struct rte_eth_dev *dev)
2125 {
2126         struct bond_dev_private *internals = dev->data->dev_private;
2127         uint16_t bond_port_id = internals->port_id;
2128         int skipped = 0;
2129         struct rte_flow_error ferror;
2130
2131         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2132                 return 0;
2133
2134         RTE_BOND_LOG(INFO, "Closing bonded device %s", dev->device->name);
2135         while (internals->slave_count != skipped) {
2136                 uint16_t port_id = internals->slaves[skipped].port_id;
2137
2138                 if (rte_eth_dev_stop(port_id) != 0) {
2139                         RTE_BOND_LOG(ERR, "Failed to stop device on port %u",
2140                                      port_id);
2141                         skipped++;
2142                 }
2143
2144                 if (rte_eth_bond_slave_remove(bond_port_id, port_id) != 0) {
2145                         RTE_BOND_LOG(ERR,
2146                                      "Failed to remove port %d from bonded device %s",
2147                                      port_id, dev->device->name);
2148                         skipped++;
2149                 }
2150         }
2151         bond_flow_ops.flush(dev, &ferror);
2152         bond_ethdev_free_queues(dev);
2153         rte_bitmap_reset(internals->vlan_filter_bmp);
2154         rte_bitmap_free(internals->vlan_filter_bmp);
2155         rte_free(internals->vlan_filter_bmpmem);
2156
2157         /* Try to release mempool used in mode6. If the bond
2158          * device is not mode6, free the NULL is not problem.
2159          */
2160         rte_mempool_free(internals->mode6.mempool);
2161
2162         if (internals->kvlist != NULL)
2163                 rte_kvargs_free(internals->kvlist);
2164
2165         return 0;
2166 }
2167
2168 /* forward declaration */
2169 static int bond_ethdev_configure(struct rte_eth_dev *dev);
2170
2171 static int
2172 bond_ethdev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2173 {
2174         struct bond_dev_private *internals = dev->data->dev_private;
2175         struct bond_slave_details slave;
2176         int ret;
2177
2178         uint16_t max_nb_rx_queues = UINT16_MAX;
2179         uint16_t max_nb_tx_queues = UINT16_MAX;
2180         uint16_t max_rx_desc_lim = UINT16_MAX;
2181         uint16_t max_tx_desc_lim = UINT16_MAX;
2182
2183         dev_info->max_mac_addrs = BOND_MAX_MAC_ADDRS;
2184
2185         dev_info->max_rx_pktlen = internals->candidate_max_rx_pktlen ?
2186                         internals->candidate_max_rx_pktlen :
2187                         RTE_ETHER_MAX_JUMBO_FRAME_LEN;
2188
2189         /* Max number of tx/rx queues that the bonded device can support is the
2190          * minimum values of the bonded slaves, as all slaves must be capable
2191          * of supporting the same number of tx/rx queues.
2192          */
2193         if (internals->slave_count > 0) {
2194                 struct rte_eth_dev_info slave_info;
2195                 uint16_t idx;
2196
2197                 for (idx = 0; idx < internals->slave_count; idx++) {
2198                         slave = internals->slaves[idx];
2199                         ret = rte_eth_dev_info_get(slave.port_id, &slave_info);
2200                         if (ret != 0) {
2201                                 RTE_BOND_LOG(ERR,
2202                                         "%s: Error during getting device (port %u) info: %s\n",
2203                                         __func__,
2204                                         slave.port_id,
2205                                         strerror(-ret));
2206
2207                                 return ret;
2208                         }
2209
2210                         if (slave_info.max_rx_queues < max_nb_rx_queues)
2211                                 max_nb_rx_queues = slave_info.max_rx_queues;
2212
2213                         if (slave_info.max_tx_queues < max_nb_tx_queues)
2214                                 max_nb_tx_queues = slave_info.max_tx_queues;
2215
2216                         if (slave_info.rx_desc_lim.nb_max < max_rx_desc_lim)
2217                                 max_rx_desc_lim = slave_info.rx_desc_lim.nb_max;
2218
2219                         if (slave_info.tx_desc_lim.nb_max < max_tx_desc_lim)
2220                                 max_tx_desc_lim = slave_info.tx_desc_lim.nb_max;
2221                 }
2222         }
2223
2224         dev_info->max_rx_queues = max_nb_rx_queues;
2225         dev_info->max_tx_queues = max_nb_tx_queues;
2226
2227         memcpy(&dev_info->default_rxconf, &internals->default_rxconf,
2228                sizeof(dev_info->default_rxconf));
2229         memcpy(&dev_info->default_txconf, &internals->default_txconf,
2230                sizeof(dev_info->default_txconf));
2231
2232         dev_info->rx_desc_lim.nb_max = max_rx_desc_lim;
2233         dev_info->tx_desc_lim.nb_max = max_tx_desc_lim;
2234
2235         /**
2236          * If dedicated hw queues enabled for link bonding device in LACP mode
2237          * then we need to reduce the maximum number of data path queues by 1.
2238          */
2239         if (internals->mode == BONDING_MODE_8023AD &&
2240                 internals->mode4.dedicated_queues.enabled == 1) {
2241                 dev_info->max_rx_queues--;
2242                 dev_info->max_tx_queues--;
2243         }
2244
2245         dev_info->min_rx_bufsize = 0;
2246
2247         dev_info->rx_offload_capa = internals->rx_offload_capa;
2248         dev_info->tx_offload_capa = internals->tx_offload_capa;
2249         dev_info->rx_queue_offload_capa = internals->rx_queue_offload_capa;
2250         dev_info->tx_queue_offload_capa = internals->tx_queue_offload_capa;
2251         dev_info->flow_type_rss_offloads = internals->flow_type_rss_offloads;
2252
2253         dev_info->reta_size = internals->reta_size;
2254
2255         return 0;
2256 }
2257
2258 static int
2259 bond_ethdev_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
2260 {
2261         int res;
2262         uint16_t i;
2263         struct bond_dev_private *internals = dev->data->dev_private;
2264
2265         /* don't do this while a slave is being added */
2266         rte_spinlock_lock(&internals->lock);
2267
2268         if (on)
2269                 rte_bitmap_set(internals->vlan_filter_bmp, vlan_id);
2270         else
2271                 rte_bitmap_clear(internals->vlan_filter_bmp, vlan_id);
2272
2273         for (i = 0; i < internals->slave_count; i++) {
2274                 uint16_t port_id = internals->slaves[i].port_id;
2275
2276                 res = rte_eth_dev_vlan_filter(port_id, vlan_id, on);
2277                 if (res == ENOTSUP)
2278                         RTE_BOND_LOG(WARNING,
2279                                      "Setting VLAN filter on slave port %u not supported.",
2280                                      port_id);
2281         }
2282
2283         rte_spinlock_unlock(&internals->lock);
2284         return 0;
2285 }
2286
2287 static int
2288 bond_ethdev_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
2289                 uint16_t nb_rx_desc, unsigned int socket_id __rte_unused,
2290                 const struct rte_eth_rxconf *rx_conf, struct rte_mempool *mb_pool)
2291 {
2292         struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)
2293                         rte_zmalloc_socket(NULL, sizeof(struct bond_rx_queue),
2294                                         0, dev->data->numa_node);
2295         if (bd_rx_q == NULL)
2296                 return -1;
2297
2298         bd_rx_q->queue_id = rx_queue_id;
2299         bd_rx_q->dev_private = dev->data->dev_private;
2300
2301         bd_rx_q->nb_rx_desc = nb_rx_desc;
2302
2303         memcpy(&(bd_rx_q->rx_conf), rx_conf, sizeof(struct rte_eth_rxconf));
2304         bd_rx_q->mb_pool = mb_pool;
2305
2306         dev->data->rx_queues[rx_queue_id] = bd_rx_q;
2307
2308         return 0;
2309 }
2310
2311 static int
2312 bond_ethdev_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
2313                 uint16_t nb_tx_desc, unsigned int socket_id __rte_unused,
2314                 const struct rte_eth_txconf *tx_conf)
2315 {
2316         struct bond_tx_queue *bd_tx_q  = (struct bond_tx_queue *)
2317                         rte_zmalloc_socket(NULL, sizeof(struct bond_tx_queue),
2318                                         0, dev->data->numa_node);
2319
2320         if (bd_tx_q == NULL)
2321                 return -1;
2322
2323         bd_tx_q->queue_id = tx_queue_id;
2324         bd_tx_q->dev_private = dev->data->dev_private;
2325
2326         bd_tx_q->nb_tx_desc = nb_tx_desc;
2327         memcpy(&(bd_tx_q->tx_conf), tx_conf, sizeof(bd_tx_q->tx_conf));
2328
2329         dev->data->tx_queues[tx_queue_id] = bd_tx_q;
2330
2331         return 0;
2332 }
2333
2334 static void
2335 bond_ethdev_rx_queue_release(void *queue)
2336 {
2337         if (queue == NULL)
2338                 return;
2339
2340         rte_free(queue);
2341 }
2342
2343 static void
2344 bond_ethdev_tx_queue_release(void *queue)
2345 {
2346         if (queue == NULL)
2347                 return;
2348
2349         rte_free(queue);
2350 }
2351
2352 static void
2353 bond_ethdev_slave_link_status_change_monitor(void *cb_arg)
2354 {
2355         struct rte_eth_dev *bonded_ethdev, *slave_ethdev;
2356         struct bond_dev_private *internals;
2357
2358         /* Default value for polling slave found is true as we don't want to
2359          * disable the polling thread if we cannot get the lock */
2360         int i, polling_slave_found = 1;
2361
2362         if (cb_arg == NULL)
2363                 return;
2364
2365         bonded_ethdev = cb_arg;
2366         internals = bonded_ethdev->data->dev_private;
2367
2368         if (!bonded_ethdev->data->dev_started ||
2369                 !internals->link_status_polling_enabled)
2370                 return;
2371
2372         /* If device is currently being configured then don't check slaves link
2373          * status, wait until next period */
2374         if (rte_spinlock_trylock(&internals->lock)) {
2375                 if (internals->slave_count > 0)
2376                         polling_slave_found = 0;
2377
2378                 for (i = 0; i < internals->slave_count; i++) {
2379                         if (!internals->slaves[i].link_status_poll_enabled)
2380                                 continue;
2381
2382                         slave_ethdev = &rte_eth_devices[internals->slaves[i].port_id];
2383                         polling_slave_found = 1;
2384
2385                         /* Update slave link status */
2386                         (*slave_ethdev->dev_ops->link_update)(slave_ethdev,
2387                                         internals->slaves[i].link_status_wait_to_complete);
2388
2389                         /* if link status has changed since last checked then call lsc
2390                          * event callback */
2391                         if (slave_ethdev->data->dev_link.link_status !=
2392                                         internals->slaves[i].last_link_status) {
2393                                 internals->slaves[i].last_link_status =
2394                                                 slave_ethdev->data->dev_link.link_status;
2395
2396                                 bond_ethdev_lsc_event_callback(internals->slaves[i].port_id,
2397                                                 RTE_ETH_EVENT_INTR_LSC,
2398                                                 &bonded_ethdev->data->port_id,
2399                                                 NULL);
2400                         }
2401                 }
2402                 rte_spinlock_unlock(&internals->lock);
2403         }
2404
2405         if (polling_slave_found)
2406                 /* Set alarm to continue monitoring link status of slave ethdev's */
2407                 rte_eal_alarm_set(internals->link_status_polling_interval_ms * 1000,
2408                                 bond_ethdev_slave_link_status_change_monitor, cb_arg);
2409 }
2410
2411 static int
2412 bond_ethdev_link_update(struct rte_eth_dev *ethdev, int wait_to_complete)
2413 {
2414         int (*link_update)(uint16_t port_id, struct rte_eth_link *eth_link);
2415
2416         struct bond_dev_private *bond_ctx;
2417         struct rte_eth_link slave_link;
2418
2419         bool one_link_update_succeeded;
2420         uint32_t idx;
2421         int ret;
2422
2423         bond_ctx = ethdev->data->dev_private;
2424
2425         ethdev->data->dev_link.link_speed = ETH_SPEED_NUM_NONE;
2426
2427         if (ethdev->data->dev_started == 0 ||
2428                         bond_ctx->active_slave_count == 0) {
2429                 ethdev->data->dev_link.link_status = ETH_LINK_DOWN;
2430                 return 0;
2431         }
2432
2433         ethdev->data->dev_link.link_status = ETH_LINK_UP;
2434
2435         if (wait_to_complete)
2436                 link_update = rte_eth_link_get;
2437         else
2438                 link_update = rte_eth_link_get_nowait;
2439
2440         switch (bond_ctx->mode) {
2441         case BONDING_MODE_BROADCAST:
2442                 /**
2443                  * Setting link speed to UINT32_MAX to ensure we pick up the
2444                  * value of the first active slave
2445                  */
2446                 ethdev->data->dev_link.link_speed = UINT32_MAX;
2447
2448                 /**
2449                  * link speed is minimum value of all the slaves link speed as
2450                  * packet loss will occur on this slave if transmission at rates
2451                  * greater than this are attempted
2452                  */
2453                 for (idx = 0; idx < bond_ctx->active_slave_count; idx++) {
2454                         ret = link_update(bond_ctx->active_slaves[idx],
2455                                           &slave_link);
2456                         if (ret < 0) {
2457                                 ethdev->data->dev_link.link_speed =
2458                                         ETH_SPEED_NUM_NONE;
2459                                 RTE_BOND_LOG(ERR,
2460                                         "Slave (port %u) link get failed: %s",
2461                                         bond_ctx->active_slaves[idx],
2462                                         rte_strerror(-ret));
2463                                 return 0;
2464                         }
2465
2466                         if (slave_link.link_speed <
2467                                         ethdev->data->dev_link.link_speed)
2468                                 ethdev->data->dev_link.link_speed =
2469                                                 slave_link.link_speed;
2470                 }
2471                 break;
2472         case BONDING_MODE_ACTIVE_BACKUP:
2473                 /* Current primary slave */
2474                 ret = link_update(bond_ctx->current_primary_port, &slave_link);
2475                 if (ret < 0) {
2476                         RTE_BOND_LOG(ERR, "Slave (port %u) link get failed: %s",
2477                                 bond_ctx->current_primary_port,
2478                                 rte_strerror(-ret));
2479                         return 0;
2480                 }
2481
2482                 ethdev->data->dev_link.link_speed = slave_link.link_speed;
2483                 break;
2484         case BONDING_MODE_8023AD:
2485                 ethdev->data->dev_link.link_autoneg =
2486                                 bond_ctx->mode4.slave_link.link_autoneg;
2487                 ethdev->data->dev_link.link_duplex =
2488                                 bond_ctx->mode4.slave_link.link_duplex;
2489                 /* fall through */
2490                 /* to update link speed */
2491         case BONDING_MODE_ROUND_ROBIN:
2492         case BONDING_MODE_BALANCE:
2493         case BONDING_MODE_TLB:
2494         case BONDING_MODE_ALB:
2495         default:
2496                 /**
2497                  * In theses mode the maximum theoretical link speed is the sum
2498                  * of all the slaves
2499                  */
2500                 ethdev->data->dev_link.link_speed = ETH_SPEED_NUM_NONE;
2501                 one_link_update_succeeded = false;
2502
2503                 for (idx = 0; idx < bond_ctx->active_slave_count; idx++) {
2504                         ret = link_update(bond_ctx->active_slaves[idx],
2505                                         &slave_link);
2506                         if (ret < 0) {
2507                                 RTE_BOND_LOG(ERR,
2508                                         "Slave (port %u) link get failed: %s",
2509                                         bond_ctx->active_slaves[idx],
2510                                         rte_strerror(-ret));
2511                                 continue;
2512                         }
2513
2514                         one_link_update_succeeded = true;
2515                         ethdev->data->dev_link.link_speed +=
2516                                         slave_link.link_speed;
2517                 }
2518
2519                 if (!one_link_update_succeeded) {
2520                         RTE_BOND_LOG(ERR, "All slaves link get failed");
2521                         return 0;
2522                 }
2523         }
2524
2525
2526         return 0;
2527 }
2528
2529
2530 static int
2531 bond_ethdev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
2532 {
2533         struct bond_dev_private *internals = dev->data->dev_private;
2534         struct rte_eth_stats slave_stats;
2535         int i, j;
2536
2537         for (i = 0; i < internals->slave_count; i++) {
2538                 rte_eth_stats_get(internals->slaves[i].port_id, &slave_stats);
2539
2540                 stats->ipackets += slave_stats.ipackets;
2541                 stats->opackets += slave_stats.opackets;
2542                 stats->ibytes += slave_stats.ibytes;
2543                 stats->obytes += slave_stats.obytes;
2544                 stats->imissed += slave_stats.imissed;
2545                 stats->ierrors += slave_stats.ierrors;
2546                 stats->oerrors += slave_stats.oerrors;
2547                 stats->rx_nombuf += slave_stats.rx_nombuf;
2548
2549                 for (j = 0; j < RTE_ETHDEV_QUEUE_STAT_CNTRS; j++) {
2550                         stats->q_ipackets[j] += slave_stats.q_ipackets[j];
2551                         stats->q_opackets[j] += slave_stats.q_opackets[j];
2552                         stats->q_ibytes[j] += slave_stats.q_ibytes[j];
2553                         stats->q_obytes[j] += slave_stats.q_obytes[j];
2554                         stats->q_errors[j] += slave_stats.q_errors[j];
2555                 }
2556
2557         }
2558
2559         return 0;
2560 }
2561
2562 static int
2563 bond_ethdev_stats_reset(struct rte_eth_dev *dev)
2564 {
2565         struct bond_dev_private *internals = dev->data->dev_private;
2566         int i;
2567         int err;
2568         int ret;
2569
2570         for (i = 0, err = 0; i < internals->slave_count; i++) {
2571                 ret = rte_eth_stats_reset(internals->slaves[i].port_id);
2572                 if (ret != 0)
2573                         err = ret;
2574         }
2575
2576         return err;
2577 }
2578
2579 static int
2580 bond_ethdev_promiscuous_enable(struct rte_eth_dev *eth_dev)
2581 {
2582         struct bond_dev_private *internals = eth_dev->data->dev_private;
2583         int i;
2584         int ret = 0;
2585         uint16_t port_id;
2586
2587         switch (internals->mode) {
2588         /* Promiscuous mode is propagated to all slaves */
2589         case BONDING_MODE_ROUND_ROBIN:
2590         case BONDING_MODE_BALANCE:
2591         case BONDING_MODE_BROADCAST:
2592         case BONDING_MODE_8023AD: {
2593                 unsigned int slave_ok = 0;
2594
2595                 for (i = 0; i < internals->slave_count; i++) {
2596                         port_id = internals->slaves[i].port_id;
2597
2598                         ret = rte_eth_promiscuous_enable(port_id);
2599                         if (ret != 0)
2600                                 RTE_BOND_LOG(ERR,
2601                                         "Failed to enable promiscuous mode for port %u: %s",
2602                                         port_id, rte_strerror(-ret));
2603                         else
2604                                 slave_ok++;
2605                 }
2606                 /*
2607                  * Report success if operation is successful on at least
2608                  * on one slave. Otherwise return last error code.
2609                  */
2610                 if (slave_ok > 0)
2611                         ret = 0;
2612                 break;
2613         }
2614         /* Promiscuous mode is propagated only to primary slave */
2615         case BONDING_MODE_ACTIVE_BACKUP:
2616         case BONDING_MODE_TLB:
2617         case BONDING_MODE_ALB:
2618         default:
2619                 /* Do not touch promisc when there cannot be primary ports */
2620                 if (internals->slave_count == 0)
2621                         break;
2622                 port_id = internals->current_primary_port;
2623                 ret = rte_eth_promiscuous_enable(port_id);
2624                 if (ret != 0)
2625                         RTE_BOND_LOG(ERR,
2626                                 "Failed to enable promiscuous mode for port %u: %s",
2627                                 port_id, rte_strerror(-ret));
2628         }
2629
2630         return ret;
2631 }
2632
2633 static int
2634 bond_ethdev_promiscuous_disable(struct rte_eth_dev *dev)
2635 {
2636         struct bond_dev_private *internals = dev->data->dev_private;
2637         int i;
2638         int ret = 0;
2639         uint16_t port_id;
2640
2641         switch (internals->mode) {
2642         /* Promiscuous mode is propagated to all slaves */
2643         case BONDING_MODE_ROUND_ROBIN:
2644         case BONDING_MODE_BALANCE:
2645         case BONDING_MODE_BROADCAST:
2646         case BONDING_MODE_8023AD: {
2647                 unsigned int slave_ok = 0;
2648
2649                 for (i = 0; i < internals->slave_count; i++) {
2650                         port_id = internals->slaves[i].port_id;
2651
2652                         if (internals->mode == BONDING_MODE_8023AD &&
2653                             bond_mode_8023ad_ports[port_id].forced_rx_flags ==
2654                                         BOND_8023AD_FORCED_PROMISC) {
2655                                 slave_ok++;
2656                                 continue;
2657                         }
2658                         ret = rte_eth_promiscuous_disable(port_id);
2659                         if (ret != 0)
2660                                 RTE_BOND_LOG(ERR,
2661                                         "Failed to disable promiscuous mode for port %u: %s",
2662                                         port_id, rte_strerror(-ret));
2663                         else
2664                                 slave_ok++;
2665                 }
2666                 /*
2667                  * Report success if operation is successful on at least
2668                  * on one slave. Otherwise return last error code.
2669                  */
2670                 if (slave_ok > 0)
2671                         ret = 0;
2672                 break;
2673         }
2674         /* Promiscuous mode is propagated only to primary slave */
2675         case BONDING_MODE_ACTIVE_BACKUP:
2676         case BONDING_MODE_TLB:
2677         case BONDING_MODE_ALB:
2678         default:
2679                 /* Do not touch promisc when there cannot be primary ports */
2680                 if (internals->slave_count == 0)
2681                         break;
2682                 port_id = internals->current_primary_port;
2683                 ret = rte_eth_promiscuous_disable(port_id);
2684                 if (ret != 0)
2685                         RTE_BOND_LOG(ERR,
2686                                 "Failed to disable promiscuous mode for port %u: %s",
2687                                 port_id, rte_strerror(-ret));
2688         }
2689
2690         return ret;
2691 }
2692
2693 static int
2694 bond_ethdev_allmulticast_enable(struct rte_eth_dev *eth_dev)
2695 {
2696         struct bond_dev_private *internals = eth_dev->data->dev_private;
2697         int i;
2698         int ret = 0;
2699         uint16_t port_id;
2700
2701         switch (internals->mode) {
2702         /* allmulti mode is propagated to all slaves */
2703         case BONDING_MODE_ROUND_ROBIN:
2704         case BONDING_MODE_BALANCE:
2705         case BONDING_MODE_BROADCAST:
2706         case BONDING_MODE_8023AD: {
2707                 unsigned int slave_ok = 0;
2708
2709                 for (i = 0; i < internals->slave_count; i++) {
2710                         port_id = internals->slaves[i].port_id;
2711
2712                         ret = rte_eth_allmulticast_enable(port_id);
2713                         if (ret != 0)
2714                                 RTE_BOND_LOG(ERR,
2715                                         "Failed to enable allmulti mode for port %u: %s",
2716                                         port_id, rte_strerror(-ret));
2717                         else
2718                                 slave_ok++;
2719                 }
2720                 /*
2721                  * Report success if operation is successful on at least
2722                  * on one slave. Otherwise return last error code.
2723                  */
2724                 if (slave_ok > 0)
2725                         ret = 0;
2726                 break;
2727         }
2728         /* allmulti mode is propagated only to primary slave */
2729         case BONDING_MODE_ACTIVE_BACKUP:
2730         case BONDING_MODE_TLB:
2731         case BONDING_MODE_ALB:
2732         default:
2733                 /* Do not touch allmulti when there cannot be primary ports */
2734                 if (internals->slave_count == 0)
2735                         break;
2736                 port_id = internals->current_primary_port;
2737                 ret = rte_eth_allmulticast_enable(port_id);
2738                 if (ret != 0)
2739                         RTE_BOND_LOG(ERR,
2740                                 "Failed to enable allmulti mode for port %u: %s",
2741                                 port_id, rte_strerror(-ret));
2742         }
2743
2744         return ret;
2745 }
2746
2747 static int
2748 bond_ethdev_allmulticast_disable(struct rte_eth_dev *eth_dev)
2749 {
2750         struct bond_dev_private *internals = eth_dev->data->dev_private;
2751         int i;
2752         int ret = 0;
2753         uint16_t port_id;
2754
2755         switch (internals->mode) {
2756         /* allmulti mode is propagated to all slaves */
2757         case BONDING_MODE_ROUND_ROBIN:
2758         case BONDING_MODE_BALANCE:
2759         case BONDING_MODE_BROADCAST:
2760         case BONDING_MODE_8023AD: {
2761                 unsigned int slave_ok = 0;
2762
2763                 for (i = 0; i < internals->slave_count; i++) {
2764                         uint16_t port_id = internals->slaves[i].port_id;
2765
2766                         if (internals->mode == BONDING_MODE_8023AD &&
2767                             bond_mode_8023ad_ports[port_id].forced_rx_flags ==
2768                                         BOND_8023AD_FORCED_ALLMULTI)
2769                                 continue;
2770
2771                         ret = rte_eth_allmulticast_disable(port_id);
2772                         if (ret != 0)
2773                                 RTE_BOND_LOG(ERR,
2774                                         "Failed to disable allmulti mode for port %u: %s",
2775                                         port_id, rte_strerror(-ret));
2776                         else
2777                                 slave_ok++;
2778                 }
2779                 /*
2780                  * Report success if operation is successful on at least
2781                  * on one slave. Otherwise return last error code.
2782                  */
2783                 if (slave_ok > 0)
2784                         ret = 0;
2785                 break;
2786         }
2787         /* allmulti mode is propagated only to primary slave */
2788         case BONDING_MODE_ACTIVE_BACKUP:
2789         case BONDING_MODE_TLB:
2790         case BONDING_MODE_ALB:
2791         default:
2792                 /* Do not touch allmulti when there cannot be primary ports */
2793                 if (internals->slave_count == 0)
2794                         break;
2795                 port_id = internals->current_primary_port;
2796                 ret = rte_eth_allmulticast_disable(port_id);
2797                 if (ret != 0)
2798                         RTE_BOND_LOG(ERR,
2799                                 "Failed to disable allmulti mode for port %u: %s",
2800                                 port_id, rte_strerror(-ret));
2801         }
2802
2803         return ret;
2804 }
2805
2806 static void
2807 bond_ethdev_delayed_lsc_propagation(void *arg)
2808 {
2809         if (arg == NULL)
2810                 return;
2811
2812         rte_eth_dev_callback_process((struct rte_eth_dev *)arg,
2813                         RTE_ETH_EVENT_INTR_LSC, NULL);
2814 }
2815
2816 int
2817 bond_ethdev_lsc_event_callback(uint16_t port_id, enum rte_eth_event_type type,
2818                 void *param, void *ret_param __rte_unused)
2819 {
2820         struct rte_eth_dev *bonded_eth_dev;
2821         struct bond_dev_private *internals;
2822         struct rte_eth_link link;
2823         int rc = -1;
2824         int ret;
2825
2826         uint8_t lsc_flag = 0;
2827         int valid_slave = 0;
2828         uint16_t active_pos;
2829         uint16_t i;
2830
2831         if (type != RTE_ETH_EVENT_INTR_LSC || param == NULL)
2832                 return rc;
2833
2834         bonded_eth_dev = &rte_eth_devices[*(uint16_t *)param];
2835
2836         if (check_for_bonded_ethdev(bonded_eth_dev))
2837                 return rc;
2838
2839         internals = bonded_eth_dev->data->dev_private;
2840
2841         /* If the device isn't started don't handle interrupts */
2842         if (!bonded_eth_dev->data->dev_started)
2843                 return rc;
2844
2845         /* verify that port_id is a valid slave of bonded port */
2846         for (i = 0; i < internals->slave_count; i++) {
2847                 if (internals->slaves[i].port_id == port_id) {
2848                         valid_slave = 1;
2849                         break;
2850                 }
2851         }
2852
2853         if (!valid_slave)
2854                 return rc;
2855
2856         /* Synchronize lsc callback parallel calls either by real link event
2857          * from the slaves PMDs or by the bonding PMD itself.
2858          */
2859         rte_spinlock_lock(&internals->lsc_lock);
2860
2861         /* Search for port in active port list */
2862         active_pos = find_slave_by_id(internals->active_slaves,
2863                         internals->active_slave_count, port_id);
2864
2865         ret = rte_eth_link_get_nowait(port_id, &link);
2866         if (ret < 0)
2867                 RTE_BOND_LOG(ERR, "Slave (port %u) link get failed", port_id);
2868
2869         if (ret == 0 && link.link_status) {
2870                 if (active_pos < internals->active_slave_count)
2871                         goto link_update;
2872
2873                 /* check link state properties if bonded link is up*/
2874                 if (bonded_eth_dev->data->dev_link.link_status == ETH_LINK_UP) {
2875                         if (link_properties_valid(bonded_eth_dev, &link) != 0)
2876                                 RTE_BOND_LOG(ERR, "Invalid link properties "
2877                                              "for slave %d in bonding mode %d",
2878                                              port_id, internals->mode);
2879                 } else {
2880                         /* inherit slave link properties */
2881                         link_properties_set(bonded_eth_dev, &link);
2882                 }
2883
2884                 /* If no active slave ports then set this port to be
2885                  * the primary port.
2886                  */
2887                 if (internals->active_slave_count < 1) {
2888                         /* If first active slave, then change link status */
2889                         bonded_eth_dev->data->dev_link.link_status =
2890                                                                 ETH_LINK_UP;
2891                         internals->current_primary_port = port_id;
2892                         lsc_flag = 1;
2893
2894                         mac_address_slaves_update(bonded_eth_dev);
2895                 }
2896
2897                 activate_slave(bonded_eth_dev, port_id);
2898
2899                 /* If the user has defined the primary port then default to
2900                  * using it.
2901                  */
2902                 if (internals->user_defined_primary_port &&
2903                                 internals->primary_port == port_id)
2904                         bond_ethdev_primary_set(internals, port_id);
2905         } else {
2906                 if (active_pos == internals->active_slave_count)
2907                         goto link_update;
2908
2909                 /* Remove from active slave list */
2910                 deactivate_slave(bonded_eth_dev, port_id);
2911
2912                 if (internals->active_slave_count < 1)
2913                         lsc_flag = 1;
2914
2915                 /* Update primary id, take first active slave from list or if none
2916                  * available set to -1 */
2917                 if (port_id == internals->current_primary_port) {
2918                         if (internals->active_slave_count > 0)
2919                                 bond_ethdev_primary_set(internals,
2920                                                 internals->active_slaves[0]);
2921                         else
2922                                 internals->current_primary_port = internals->primary_port;
2923                         mac_address_slaves_update(bonded_eth_dev);
2924                 }
2925         }
2926
2927 link_update:
2928         /**
2929          * Update bonded device link properties after any change to active
2930          * slaves
2931          */
2932         bond_ethdev_link_update(bonded_eth_dev, 0);
2933
2934         if (lsc_flag) {
2935                 /* Cancel any possible outstanding interrupts if delays are enabled */
2936                 if (internals->link_up_delay_ms > 0 ||
2937                         internals->link_down_delay_ms > 0)
2938                         rte_eal_alarm_cancel(bond_ethdev_delayed_lsc_propagation,
2939                                         bonded_eth_dev);
2940
2941                 if (bonded_eth_dev->data->dev_link.link_status) {
2942                         if (internals->link_up_delay_ms > 0)
2943                                 rte_eal_alarm_set(internals->link_up_delay_ms * 1000,
2944                                                 bond_ethdev_delayed_lsc_propagation,
2945                                                 (void *)bonded_eth_dev);
2946                         else
2947                                 rte_eth_dev_callback_process(bonded_eth_dev,
2948                                                 RTE_ETH_EVENT_INTR_LSC,
2949                                                 NULL);
2950
2951                 } else {
2952                         if (internals->link_down_delay_ms > 0)
2953                                 rte_eal_alarm_set(internals->link_down_delay_ms * 1000,
2954                                                 bond_ethdev_delayed_lsc_propagation,
2955                                                 (void *)bonded_eth_dev);
2956                         else
2957                                 rte_eth_dev_callback_process(bonded_eth_dev,
2958                                                 RTE_ETH_EVENT_INTR_LSC,
2959                                                 NULL);
2960                 }
2961         }
2962
2963         rte_spinlock_unlock(&internals->lsc_lock);
2964
2965         return rc;
2966 }
2967
2968 static int
2969 bond_ethdev_rss_reta_update(struct rte_eth_dev *dev,
2970                 struct rte_eth_rss_reta_entry64 *reta_conf, uint16_t reta_size)
2971 {
2972         unsigned i, j;
2973         int result = 0;
2974         int slave_reta_size;
2975         unsigned reta_count;
2976         struct bond_dev_private *internals = dev->data->dev_private;
2977
2978         if (reta_size != internals->reta_size)
2979                 return -EINVAL;
2980
2981          /* Copy RETA table */
2982         reta_count = (reta_size + RTE_RETA_GROUP_SIZE - 1) /
2983                         RTE_RETA_GROUP_SIZE;
2984
2985         for (i = 0; i < reta_count; i++) {
2986                 internals->reta_conf[i].mask = reta_conf[i].mask;
2987                 for (j = 0; j < RTE_RETA_GROUP_SIZE; j++)
2988                         if ((reta_conf[i].mask >> j) & 0x01)
2989                                 internals->reta_conf[i].reta[j] = reta_conf[i].reta[j];
2990         }
2991
2992         /* Fill rest of array */
2993         for (; i < RTE_DIM(internals->reta_conf); i += reta_count)
2994                 memcpy(&internals->reta_conf[i], &internals->reta_conf[0],
2995                                 sizeof(internals->reta_conf[0]) * reta_count);
2996
2997         /* Propagate RETA over slaves */
2998         for (i = 0; i < internals->slave_count; i++) {
2999                 slave_reta_size = internals->slaves[i].reta_size;
3000                 result = rte_eth_dev_rss_reta_update(internals->slaves[i].port_id,
3001                                 &internals->reta_conf[0], slave_reta_size);
3002                 if (result < 0)
3003                         return result;
3004         }
3005
3006         return 0;
3007 }
3008
3009 static int
3010 bond_ethdev_rss_reta_query(struct rte_eth_dev *dev,
3011                 struct rte_eth_rss_reta_entry64 *reta_conf, uint16_t reta_size)
3012 {
3013         int i, j;
3014         struct bond_dev_private *internals = dev->data->dev_private;
3015
3016         if (reta_size != internals->reta_size)
3017                 return -EINVAL;
3018
3019          /* Copy RETA table */
3020         for (i = 0; i < reta_size / RTE_RETA_GROUP_SIZE; i++)
3021                 for (j = 0; j < RTE_RETA_GROUP_SIZE; j++)
3022                         if ((reta_conf[i].mask >> j) & 0x01)
3023                                 reta_conf[i].reta[j] = internals->reta_conf[i].reta[j];
3024
3025         return 0;
3026 }
3027
3028 static int
3029 bond_ethdev_rss_hash_update(struct rte_eth_dev *dev,
3030                 struct rte_eth_rss_conf *rss_conf)
3031 {
3032         int i, result = 0;
3033         struct bond_dev_private *internals = dev->data->dev_private;
3034         struct rte_eth_rss_conf bond_rss_conf;
3035
3036         memcpy(&bond_rss_conf, rss_conf, sizeof(struct rte_eth_rss_conf));
3037
3038         bond_rss_conf.rss_hf &= internals->flow_type_rss_offloads;
3039
3040         if (bond_rss_conf.rss_hf != 0)
3041                 dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf = bond_rss_conf.rss_hf;
3042
3043         if (bond_rss_conf.rss_key && bond_rss_conf.rss_key_len <
3044                         sizeof(internals->rss_key)) {
3045                 if (bond_rss_conf.rss_key_len == 0)
3046                         bond_rss_conf.rss_key_len = 40;
3047                 internals->rss_key_len = bond_rss_conf.rss_key_len;
3048                 memcpy(internals->rss_key, bond_rss_conf.rss_key,
3049                                 internals->rss_key_len);
3050         }
3051
3052         for (i = 0; i < internals->slave_count; i++) {
3053                 result = rte_eth_dev_rss_hash_update(internals->slaves[i].port_id,
3054                                 &bond_rss_conf);
3055                 if (result < 0)
3056                         return result;
3057         }
3058
3059         return 0;
3060 }
3061
3062 static int
3063 bond_ethdev_rss_hash_conf_get(struct rte_eth_dev *dev,
3064                 struct rte_eth_rss_conf *rss_conf)
3065 {
3066         struct bond_dev_private *internals = dev->data->dev_private;
3067
3068         rss_conf->rss_hf = dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf;
3069         rss_conf->rss_key_len = internals->rss_key_len;
3070         if (rss_conf->rss_key)
3071                 memcpy(rss_conf->rss_key, internals->rss_key, internals->rss_key_len);
3072
3073         return 0;
3074 }
3075
3076 static int
3077 bond_ethdev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
3078 {
3079         struct rte_eth_dev *slave_eth_dev;
3080         struct bond_dev_private *internals = dev->data->dev_private;
3081         int ret, i;
3082
3083         rte_spinlock_lock(&internals->lock);
3084
3085         for (i = 0; i < internals->slave_count; i++) {
3086                 slave_eth_dev = &rte_eth_devices[internals->slaves[i].port_id];
3087                 if (*slave_eth_dev->dev_ops->mtu_set == NULL) {
3088                         rte_spinlock_unlock(&internals->lock);
3089                         return -ENOTSUP;
3090                 }
3091         }
3092         for (i = 0; i < internals->slave_count; i++) {
3093                 ret = rte_eth_dev_set_mtu(internals->slaves[i].port_id, mtu);
3094                 if (ret < 0) {
3095                         rte_spinlock_unlock(&internals->lock);
3096                         return ret;
3097                 }
3098         }
3099
3100         rte_spinlock_unlock(&internals->lock);
3101         return 0;
3102 }
3103
3104 static int
3105 bond_ethdev_mac_address_set(struct rte_eth_dev *dev,
3106                         struct rte_ether_addr *addr)
3107 {
3108         if (mac_address_set(dev, addr)) {
3109                 RTE_BOND_LOG(ERR, "Failed to update MAC address");
3110                 return -EINVAL;
3111         }
3112
3113         return 0;
3114 }
3115
3116 static int
3117 bond_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
3118                   const struct rte_flow_ops **ops)
3119 {
3120         *ops = &bond_flow_ops;
3121         return 0;
3122 }
3123
3124 static int
3125 bond_ethdev_mac_addr_add(struct rte_eth_dev *dev,
3126                         struct rte_ether_addr *mac_addr,
3127                         __rte_unused uint32_t index, uint32_t vmdq)
3128 {
3129         struct rte_eth_dev *slave_eth_dev;
3130         struct bond_dev_private *internals = dev->data->dev_private;
3131         int ret, i;
3132
3133         rte_spinlock_lock(&internals->lock);
3134
3135         for (i = 0; i < internals->slave_count; i++) {
3136                 slave_eth_dev = &rte_eth_devices[internals->slaves[i].port_id];
3137                 if (*slave_eth_dev->dev_ops->mac_addr_add == NULL ||
3138                          *slave_eth_dev->dev_ops->mac_addr_remove == NULL) {
3139                         ret = -ENOTSUP;
3140                         goto end;
3141                 }
3142         }
3143
3144         for (i = 0; i < internals->slave_count; i++) {
3145                 ret = rte_eth_dev_mac_addr_add(internals->slaves[i].port_id,
3146                                 mac_addr, vmdq);
3147                 if (ret < 0) {
3148                         /* rollback */
3149                         for (i--; i >= 0; i--)
3150                                 rte_eth_dev_mac_addr_remove(
3151                                         internals->slaves[i].port_id, mac_addr);
3152                         goto end;
3153                 }
3154         }
3155
3156         ret = 0;
3157 end:
3158         rte_spinlock_unlock(&internals->lock);
3159         return ret;
3160 }
3161
3162 static void
3163 bond_ethdev_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
3164 {
3165         struct rte_eth_dev *slave_eth_dev;
3166         struct bond_dev_private *internals = dev->data->dev_private;
3167         int i;
3168
3169         rte_spinlock_lock(&internals->lock);
3170
3171         for (i = 0; i < internals->slave_count; i++) {
3172                 slave_eth_dev = &rte_eth_devices[internals->slaves[i].port_id];
3173                 if (*slave_eth_dev->dev_ops->mac_addr_remove == NULL)
3174                         goto end;
3175         }
3176
3177         struct rte_ether_addr *mac_addr = &dev->data->mac_addrs[index];
3178
3179         for (i = 0; i < internals->slave_count; i++)
3180                 rte_eth_dev_mac_addr_remove(internals->slaves[i].port_id,
3181                                 mac_addr);
3182
3183 end:
3184         rte_spinlock_unlock(&internals->lock);
3185 }
3186
3187 const struct eth_dev_ops default_dev_ops = {
3188         .dev_start            = bond_ethdev_start,
3189         .dev_stop             = bond_ethdev_stop,
3190         .dev_close            = bond_ethdev_close,
3191         .dev_configure        = bond_ethdev_configure,
3192         .dev_infos_get        = bond_ethdev_info,
3193         .vlan_filter_set      = bond_ethdev_vlan_filter_set,
3194         .rx_queue_setup       = bond_ethdev_rx_queue_setup,
3195         .tx_queue_setup       = bond_ethdev_tx_queue_setup,
3196         .rx_queue_release     = bond_ethdev_rx_queue_release,
3197         .tx_queue_release     = bond_ethdev_tx_queue_release,
3198         .link_update          = bond_ethdev_link_update,
3199         .stats_get            = bond_ethdev_stats_get,
3200         .stats_reset          = bond_ethdev_stats_reset,
3201         .promiscuous_enable   = bond_ethdev_promiscuous_enable,
3202         .promiscuous_disable  = bond_ethdev_promiscuous_disable,
3203         .allmulticast_enable  = bond_ethdev_allmulticast_enable,
3204         .allmulticast_disable = bond_ethdev_allmulticast_disable,
3205         .reta_update          = bond_ethdev_rss_reta_update,
3206         .reta_query           = bond_ethdev_rss_reta_query,
3207         .rss_hash_update      = bond_ethdev_rss_hash_update,
3208         .rss_hash_conf_get    = bond_ethdev_rss_hash_conf_get,
3209         .mtu_set              = bond_ethdev_mtu_set,
3210         .mac_addr_set         = bond_ethdev_mac_address_set,
3211         .mac_addr_add         = bond_ethdev_mac_addr_add,
3212         .mac_addr_remove      = bond_ethdev_mac_addr_remove,
3213         .flow_ops_get         = bond_flow_ops_get
3214 };
3215
3216 static int
3217 bond_alloc(struct rte_vdev_device *dev, uint8_t mode)
3218 {
3219         const char *name = rte_vdev_device_name(dev);
3220         uint8_t socket_id = dev->device.numa_node;
3221         struct bond_dev_private *internals = NULL;
3222         struct rte_eth_dev *eth_dev = NULL;
3223         uint32_t vlan_filter_bmp_size;
3224
3225         /* now do all data allocation - for eth_dev structure, dummy pci driver
3226          * and internal (private) data
3227          */
3228
3229         /* reserve an ethdev entry */
3230         eth_dev = rte_eth_vdev_allocate(dev, sizeof(*internals));
3231         if (eth_dev == NULL) {
3232                 RTE_BOND_LOG(ERR, "Unable to allocate rte_eth_dev");
3233                 goto err;
3234         }
3235
3236         internals = eth_dev->data->dev_private;
3237         eth_dev->data->nb_rx_queues = (uint16_t)1;
3238         eth_dev->data->nb_tx_queues = (uint16_t)1;
3239
3240         /* Allocate memory for storing MAC addresses */
3241         eth_dev->data->mac_addrs = rte_zmalloc_socket(name, RTE_ETHER_ADDR_LEN *
3242                         BOND_MAX_MAC_ADDRS, 0, socket_id);
3243         if (eth_dev->data->mac_addrs == NULL) {
3244                 RTE_BOND_LOG(ERR,
3245                              "Failed to allocate %u bytes needed to store MAC addresses",
3246                              RTE_ETHER_ADDR_LEN * BOND_MAX_MAC_ADDRS);
3247                 goto err;
3248         }
3249
3250         eth_dev->dev_ops = &default_dev_ops;
3251         eth_dev->data->dev_flags = RTE_ETH_DEV_INTR_LSC |
3252                                         RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
3253
3254         rte_spinlock_init(&internals->lock);
3255         rte_spinlock_init(&internals->lsc_lock);
3256
3257         internals->port_id = eth_dev->data->port_id;
3258         internals->mode = BONDING_MODE_INVALID;
3259         internals->current_primary_port = RTE_MAX_ETHPORTS + 1;
3260         internals->balance_xmit_policy = BALANCE_XMIT_POLICY_LAYER2;
3261         internals->burst_xmit_hash = burst_xmit_l2_hash;
3262         internals->user_defined_mac = 0;
3263
3264         internals->link_status_polling_enabled = 0;
3265
3266         internals->link_status_polling_interval_ms =
3267                 DEFAULT_POLLING_INTERVAL_10_MS;
3268         internals->link_down_delay_ms = 0;
3269         internals->link_up_delay_ms = 0;
3270
3271         internals->slave_count = 0;
3272         internals->active_slave_count = 0;
3273         internals->rx_offload_capa = 0;
3274         internals->tx_offload_capa = 0;
3275         internals->rx_queue_offload_capa = 0;
3276         internals->tx_queue_offload_capa = 0;
3277         internals->candidate_max_rx_pktlen = 0;
3278         internals->max_rx_pktlen = 0;
3279
3280         /* Initially allow to choose any offload type */
3281         internals->flow_type_rss_offloads = ETH_RSS_PROTO_MASK;
3282
3283         memset(&internals->default_rxconf, 0,
3284                sizeof(internals->default_rxconf));
3285         memset(&internals->default_txconf, 0,
3286                sizeof(internals->default_txconf));
3287
3288         memset(&internals->rx_desc_lim, 0, sizeof(internals->rx_desc_lim));
3289         memset(&internals->tx_desc_lim, 0, sizeof(internals->tx_desc_lim));
3290
3291         memset(internals->active_slaves, 0, sizeof(internals->active_slaves));
3292         memset(internals->slaves, 0, sizeof(internals->slaves));
3293
3294         TAILQ_INIT(&internals->flow_list);
3295         internals->flow_isolated_valid = 0;
3296
3297         /* Set mode 4 default configuration */
3298         bond_mode_8023ad_setup(eth_dev, NULL);
3299         if (bond_ethdev_mode_set(eth_dev, mode)) {
3300                 RTE_BOND_LOG(ERR, "Failed to set bonded device %d mode to %d",
3301                                  eth_dev->data->port_id, mode);
3302                 goto err;
3303         }
3304
3305         vlan_filter_bmp_size =
3306                 rte_bitmap_get_memory_footprint(RTE_ETHER_MAX_VLAN_ID + 1);
3307         internals->vlan_filter_bmpmem = rte_malloc(name, vlan_filter_bmp_size,
3308                                                    RTE_CACHE_LINE_SIZE);
3309         if (internals->vlan_filter_bmpmem == NULL) {
3310                 RTE_BOND_LOG(ERR,
3311                              "Failed to allocate vlan bitmap for bonded device %u",
3312                              eth_dev->data->port_id);
3313                 goto err;
3314         }
3315
3316         internals->vlan_filter_bmp = rte_bitmap_init(RTE_ETHER_MAX_VLAN_ID + 1,
3317                         internals->vlan_filter_bmpmem, vlan_filter_bmp_size);
3318         if (internals->vlan_filter_bmp == NULL) {
3319                 RTE_BOND_LOG(ERR,
3320                              "Failed to init vlan bitmap for bonded device %u",
3321                              eth_dev->data->port_id);
3322                 rte_free(internals->vlan_filter_bmpmem);
3323                 goto err;
3324         }
3325
3326         return eth_dev->data->port_id;
3327
3328 err:
3329         rte_free(internals);
3330         if (eth_dev != NULL)
3331                 eth_dev->data->dev_private = NULL;
3332         rte_eth_dev_release_port(eth_dev);
3333         return -1;
3334 }
3335
3336 static int
3337 bond_probe(struct rte_vdev_device *dev)
3338 {
3339         const char *name;
3340         struct bond_dev_private *internals;
3341         struct rte_kvargs *kvlist;
3342         uint8_t bonding_mode;
3343         int arg_count, port_id;
3344         int socket_id;
3345         uint8_t agg_mode;
3346         struct rte_eth_dev *eth_dev;
3347
3348         if (!dev)
3349                 return -EINVAL;
3350
3351         name = rte_vdev_device_name(dev);
3352         RTE_BOND_LOG(INFO, "Initializing pmd_bond for %s", name);
3353
3354         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
3355                 eth_dev = rte_eth_dev_attach_secondary(name);
3356                 if (!eth_dev) {
3357                         RTE_BOND_LOG(ERR, "Failed to probe %s", name);
3358                         return -1;
3359                 }
3360                 /* TODO: request info from primary to set up Rx and Tx */
3361                 eth_dev->dev_ops = &default_dev_ops;
3362                 eth_dev->device = &dev->device;
3363                 rte_eth_dev_probing_finish(eth_dev);
3364                 return 0;
3365         }
3366
3367         kvlist = rte_kvargs_parse(rte_vdev_device_args(dev),
3368                 pmd_bond_init_valid_arguments);
3369         if (kvlist == NULL)
3370                 return -1;
3371
3372         /* Parse link bonding mode */
3373         if (rte_kvargs_count(kvlist, PMD_BOND_MODE_KVARG) == 1) {
3374                 if (rte_kvargs_process(kvlist, PMD_BOND_MODE_KVARG,
3375                                 &bond_ethdev_parse_slave_mode_kvarg,
3376                                 &bonding_mode) != 0) {
3377                         RTE_BOND_LOG(ERR, "Invalid mode for bonded device %s",
3378                                         name);
3379                         goto parse_error;
3380                 }
3381         } else {
3382                 RTE_BOND_LOG(ERR, "Mode must be specified only once for bonded "
3383                                 "device %s", name);
3384                 goto parse_error;
3385         }
3386
3387         /* Parse socket id to create bonding device on */
3388         arg_count = rte_kvargs_count(kvlist, PMD_BOND_SOCKET_ID_KVARG);
3389         if (arg_count == 1) {
3390                 if (rte_kvargs_process(kvlist, PMD_BOND_SOCKET_ID_KVARG,
3391                                 &bond_ethdev_parse_socket_id_kvarg, &socket_id)
3392                                 != 0) {
3393                         RTE_BOND_LOG(ERR, "Invalid socket Id specified for "
3394                                         "bonded device %s", name);
3395                         goto parse_error;
3396                 }
3397         } else if (arg_count > 1) {
3398                 RTE_BOND_LOG(ERR, "Socket Id can be specified only once for "
3399                                 "bonded device %s", name);
3400                 goto parse_error;
3401         } else {
3402                 socket_id = rte_socket_id();
3403         }
3404
3405         dev->device.numa_node = socket_id;
3406
3407         /* Create link bonding eth device */
3408         port_id = bond_alloc(dev, bonding_mode);
3409         if (port_id < 0) {
3410                 RTE_BOND_LOG(ERR, "Failed to create socket %s in mode %u on "
3411                                 "socket %u.",   name, bonding_mode, socket_id);
3412                 goto parse_error;
3413         }
3414         internals = rte_eth_devices[port_id].data->dev_private;
3415         internals->kvlist = kvlist;
3416
3417         if (rte_kvargs_count(kvlist, PMD_BOND_AGG_MODE_KVARG) == 1) {
3418                 if (rte_kvargs_process(kvlist,
3419                                 PMD_BOND_AGG_MODE_KVARG,
3420                                 &bond_ethdev_parse_slave_agg_mode_kvarg,
3421                                 &agg_mode) != 0) {
3422                         RTE_BOND_LOG(ERR,
3423                                         "Failed to parse agg selection mode for bonded device %s",
3424                                         name);
3425                         goto parse_error;
3426                 }
3427
3428                 if (internals->mode == BONDING_MODE_8023AD)
3429                         internals->mode4.agg_selection = agg_mode;
3430         } else {
3431                 internals->mode4.agg_selection = AGG_STABLE;
3432         }
3433
3434         rte_eth_dev_probing_finish(&rte_eth_devices[port_id]);
3435         RTE_BOND_LOG(INFO, "Create bonded device %s on port %d in mode %u on "
3436                         "socket %u.",   name, port_id, bonding_mode, socket_id);
3437         return 0;
3438
3439 parse_error:
3440         rte_kvargs_free(kvlist);
3441
3442         return -1;
3443 }
3444
3445 static int
3446 bond_remove(struct rte_vdev_device *dev)
3447 {
3448         struct rte_eth_dev *eth_dev;
3449         struct bond_dev_private *internals;
3450         const char *name;
3451         int ret = 0;
3452
3453         if (!dev)
3454                 return -EINVAL;
3455
3456         name = rte_vdev_device_name(dev);
3457         RTE_BOND_LOG(INFO, "Uninitializing pmd_bond for %s", name);
3458
3459         /* find an ethdev entry */
3460         eth_dev = rte_eth_dev_allocated(name);
3461         if (eth_dev == NULL)
3462                 return 0; /* port already released */
3463
3464         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
3465                 return rte_eth_dev_release_port(eth_dev);
3466
3467         RTE_ASSERT(eth_dev->device == &dev->device);
3468
3469         internals = eth_dev->data->dev_private;
3470         if (internals->slave_count != 0)
3471                 return -EBUSY;
3472
3473         if (eth_dev->data->dev_started == 1) {
3474                 ret = bond_ethdev_stop(eth_dev);
3475                 bond_ethdev_close(eth_dev);
3476         }
3477         rte_eth_dev_release_port(eth_dev);
3478
3479         return ret;
3480 }
3481
3482 /* this part will resolve the slave portids after all the other pdev and vdev
3483  * have been allocated */
3484 static int
3485 bond_ethdev_configure(struct rte_eth_dev *dev)
3486 {
3487         const char *name = dev->device->name;
3488         struct bond_dev_private *internals = dev->data->dev_private;
3489         struct rte_kvargs *kvlist = internals->kvlist;
3490         int arg_count;
3491         uint16_t port_id = dev - rte_eth_devices;
3492         uint8_t agg_mode;
3493
3494         static const uint8_t default_rss_key[40] = {
3495                 0x6D, 0x5A, 0x56, 0xDA, 0x25, 0x5B, 0x0E, 0xC2, 0x41, 0x67, 0x25, 0x3D,
3496                 0x43, 0xA3, 0x8F, 0xB0, 0xD0, 0xCA, 0x2B, 0xCB, 0xAE, 0x7B, 0x30, 0xB4,
3497                 0x77, 0xCB, 0x2D, 0xA3, 0x80, 0x30, 0xF2, 0x0C, 0x6A, 0x42, 0xB7, 0x3B,
3498                 0xBE, 0xAC, 0x01, 0xFA
3499         };
3500
3501         unsigned i, j;
3502
3503         /*
3504          * If RSS is enabled, fill table with default values and
3505          * set key to the the value specified in port RSS configuration.
3506          * Fall back to default RSS key if the key is not specified
3507          */
3508         if (dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS) {
3509                 if (dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key != NULL) {
3510                         internals->rss_key_len =
3511                                 dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key_len;
3512                         memcpy(internals->rss_key,
3513                                dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key,
3514                                internals->rss_key_len);
3515                 } else {
3516                         internals->rss_key_len = sizeof(default_rss_key);
3517                         memcpy(internals->rss_key, default_rss_key,
3518                                internals->rss_key_len);
3519                 }
3520
3521                 for (i = 0; i < RTE_DIM(internals->reta_conf); i++) {
3522                         internals->reta_conf[i].mask = ~0LL;
3523                         for (j = 0; j < RTE_RETA_GROUP_SIZE; j++)
3524                                 internals->reta_conf[i].reta[j] =
3525                                                 (i * RTE_RETA_GROUP_SIZE + j) %
3526                                                 dev->data->nb_rx_queues;
3527                 }
3528         }
3529
3530         /* set the max_rx_pktlen */
3531         internals->max_rx_pktlen = internals->candidate_max_rx_pktlen;
3532
3533         /*
3534          * if no kvlist, it means that this bonded device has been created
3535          * through the bonding api.
3536          */
3537         if (!kvlist)
3538                 return 0;
3539
3540         /* Parse MAC address for bonded device */
3541         arg_count = rte_kvargs_count(kvlist, PMD_BOND_MAC_ADDR_KVARG);
3542         if (arg_count == 1) {
3543                 struct rte_ether_addr bond_mac;
3544
3545                 if (rte_kvargs_process(kvlist, PMD_BOND_MAC_ADDR_KVARG,
3546                                        &bond_ethdev_parse_bond_mac_addr_kvarg, &bond_mac) < 0) {
3547                         RTE_BOND_LOG(INFO, "Invalid mac address for bonded device %s",
3548                                      name);
3549                         return -1;
3550                 }
3551
3552                 /* Set MAC address */
3553                 if (rte_eth_bond_mac_address_set(port_id, &bond_mac) != 0) {
3554                         RTE_BOND_LOG(ERR,
3555                                      "Failed to set mac address on bonded device %s",
3556                                      name);
3557                         return -1;
3558                 }
3559         } else if (arg_count > 1) {
3560                 RTE_BOND_LOG(ERR,
3561                              "MAC address can be specified only once for bonded device %s",
3562                              name);
3563                 return -1;
3564         }
3565
3566         /* Parse/set balance mode transmit policy */
3567         arg_count = rte_kvargs_count(kvlist, PMD_BOND_XMIT_POLICY_KVARG);
3568         if (arg_count == 1) {
3569                 uint8_t xmit_policy;
3570
3571                 if (rte_kvargs_process(kvlist, PMD_BOND_XMIT_POLICY_KVARG,
3572                                        &bond_ethdev_parse_balance_xmit_policy_kvarg, &xmit_policy) !=
3573                     0) {
3574                         RTE_BOND_LOG(INFO,
3575                                      "Invalid xmit policy specified for bonded device %s",
3576                                      name);
3577                         return -1;
3578                 }
3579
3580                 /* Set balance mode transmit policy*/
3581                 if (rte_eth_bond_xmit_policy_set(port_id, xmit_policy) != 0) {
3582                         RTE_BOND_LOG(ERR,
3583                                      "Failed to set balance xmit policy on bonded device %s",
3584                                      name);
3585                         return -1;
3586                 }
3587         } else if (arg_count > 1) {
3588                 RTE_BOND_LOG(ERR,
3589                              "Transmit policy can be specified only once for bonded device %s",
3590                              name);
3591                 return -1;
3592         }
3593
3594         if (rte_kvargs_count(kvlist, PMD_BOND_AGG_MODE_KVARG) == 1) {
3595                 if (rte_kvargs_process(kvlist,
3596                                        PMD_BOND_AGG_MODE_KVARG,
3597                                        &bond_ethdev_parse_slave_agg_mode_kvarg,
3598                                        &agg_mode) != 0) {
3599                         RTE_BOND_LOG(ERR,
3600                                      "Failed to parse agg selection mode for bonded device %s",
3601                                      name);
3602                 }
3603                 if (internals->mode == BONDING_MODE_8023AD) {
3604                         int ret = rte_eth_bond_8023ad_agg_selection_set(port_id,
3605                                         agg_mode);
3606                         if (ret < 0) {
3607                                 RTE_BOND_LOG(ERR,
3608                                         "Invalid args for agg selection set for bonded device %s",
3609                                         name);
3610                                 return -1;
3611                         }
3612                 }
3613         }
3614
3615         /* Parse/add slave ports to bonded device */
3616         if (rte_kvargs_count(kvlist, PMD_BOND_SLAVE_PORT_KVARG) > 0) {
3617                 struct bond_ethdev_slave_ports slave_ports;
3618                 unsigned i;
3619
3620                 memset(&slave_ports, 0, sizeof(slave_ports));
3621
3622                 if (rte_kvargs_process(kvlist, PMD_BOND_SLAVE_PORT_KVARG,
3623                                        &bond_ethdev_parse_slave_port_kvarg, &slave_ports) != 0) {
3624                         RTE_BOND_LOG(ERR,
3625                                      "Failed to parse slave ports for bonded device %s",
3626                                      name);
3627                         return -1;
3628                 }
3629
3630                 for (i = 0; i < slave_ports.slave_count; i++) {
3631                         if (rte_eth_bond_slave_add(port_id, slave_ports.slaves[i]) != 0) {
3632                                 RTE_BOND_LOG(ERR,
3633                                              "Failed to add port %d as slave to bonded device %s",
3634                                              slave_ports.slaves[i], name);
3635                         }
3636                 }
3637
3638         } else {
3639                 RTE_BOND_LOG(INFO, "No slaves specified for bonded device %s", name);
3640                 return -1;
3641         }
3642
3643         /* Parse/set primary slave port id*/
3644         arg_count = rte_kvargs_count(kvlist, PMD_BOND_PRIMARY_SLAVE_KVARG);
3645         if (arg_count == 1) {
3646                 uint16_t primary_slave_port_id;
3647
3648                 if (rte_kvargs_process(kvlist,
3649                                        PMD_BOND_PRIMARY_SLAVE_KVARG,
3650                                        &bond_ethdev_parse_primary_slave_port_id_kvarg,
3651                                        &primary_slave_port_id) < 0) {
3652                         RTE_BOND_LOG(INFO,
3653                                      "Invalid primary slave port id specified for bonded device %s",
3654                                      name);
3655                         return -1;
3656                 }
3657
3658                 /* Set balance mode transmit policy*/
3659                 if (rte_eth_bond_primary_set(port_id, primary_slave_port_id)
3660                     != 0) {
3661                         RTE_BOND_LOG(ERR,
3662                                      "Failed to set primary slave port %d on bonded device %s",
3663                                      primary_slave_port_id, name);
3664                         return -1;
3665                 }
3666         } else if (arg_count > 1) {
3667                 RTE_BOND_LOG(INFO,
3668                              "Primary slave can be specified only once for bonded device %s",
3669                              name);
3670                 return -1;
3671         }
3672
3673         /* Parse link status monitor polling interval */
3674         arg_count = rte_kvargs_count(kvlist, PMD_BOND_LSC_POLL_PERIOD_KVARG);
3675         if (arg_count == 1) {
3676                 uint32_t lsc_poll_interval_ms;
3677
3678                 if (rte_kvargs_process(kvlist,
3679                                        PMD_BOND_LSC_POLL_PERIOD_KVARG,
3680                                        &bond_ethdev_parse_time_ms_kvarg,
3681                                        &lsc_poll_interval_ms) < 0) {
3682                         RTE_BOND_LOG(INFO,
3683                                      "Invalid lsc polling interval value specified for bonded"
3684                                      " device %s", name);
3685                         return -1;
3686                 }
3687
3688                 if (rte_eth_bond_link_monitoring_set(port_id, lsc_poll_interval_ms)
3689                     != 0) {
3690                         RTE_BOND_LOG(ERR,
3691                                      "Failed to set lsc monitor polling interval (%u ms) on bonded device %s",
3692                                      lsc_poll_interval_ms, name);
3693                         return -1;
3694                 }
3695         } else if (arg_count > 1) {
3696                 RTE_BOND_LOG(INFO,
3697                              "LSC polling interval can be specified only once for bonded"
3698                              " device %s", name);
3699                 return -1;
3700         }
3701
3702         /* Parse link up interrupt propagation delay */
3703         arg_count = rte_kvargs_count(kvlist, PMD_BOND_LINK_UP_PROP_DELAY_KVARG);
3704         if (arg_count == 1) {
3705                 uint32_t link_up_delay_ms;
3706
3707                 if (rte_kvargs_process(kvlist,
3708                                        PMD_BOND_LINK_UP_PROP_DELAY_KVARG,
3709                                        &bond_ethdev_parse_time_ms_kvarg,
3710                                        &link_up_delay_ms) < 0) {
3711                         RTE_BOND_LOG(INFO,
3712                                      "Invalid link up propagation delay value specified for"
3713                                      " bonded device %s", name);
3714                         return -1;
3715                 }
3716
3717                 /* Set balance mode transmit policy*/
3718                 if (rte_eth_bond_link_up_prop_delay_set(port_id, link_up_delay_ms)
3719                     != 0) {
3720                         RTE_BOND_LOG(ERR,
3721                                      "Failed to set link up propagation delay (%u ms) on bonded"
3722                                      " device %s", link_up_delay_ms, name);
3723                         return -1;
3724                 }
3725         } else if (arg_count > 1) {
3726                 RTE_BOND_LOG(INFO,
3727                              "Link up propagation delay can be specified only once for"
3728                              " bonded device %s", name);
3729                 return -1;
3730         }
3731
3732         /* Parse link down interrupt propagation delay */
3733         arg_count = rte_kvargs_count(kvlist, PMD_BOND_LINK_DOWN_PROP_DELAY_KVARG);
3734         if (arg_count == 1) {
3735                 uint32_t link_down_delay_ms;
3736
3737                 if (rte_kvargs_process(kvlist,
3738                                        PMD_BOND_LINK_DOWN_PROP_DELAY_KVARG,
3739                                        &bond_ethdev_parse_time_ms_kvarg,
3740                                        &link_down_delay_ms) < 0) {
3741                         RTE_BOND_LOG(INFO,
3742                                      "Invalid link down propagation delay value specified for"
3743                                      " bonded device %s", name);
3744                         return -1;
3745                 }
3746
3747                 /* Set balance mode transmit policy*/
3748                 if (rte_eth_bond_link_down_prop_delay_set(port_id, link_down_delay_ms)
3749                     != 0) {
3750                         RTE_BOND_LOG(ERR,
3751                                      "Failed to set link down propagation delay (%u ms) on bonded device %s",
3752                                      link_down_delay_ms, name);
3753                         return -1;
3754                 }
3755         } else if (arg_count > 1) {
3756                 RTE_BOND_LOG(INFO,
3757                              "Link down propagation delay can be specified only once for  bonded device %s",
3758                              name);
3759                 return -1;
3760         }
3761
3762         return 0;
3763 }
3764
3765 struct rte_vdev_driver pmd_bond_drv = {
3766         .probe = bond_probe,
3767         .remove = bond_remove,
3768 };
3769
3770 RTE_PMD_REGISTER_VDEV(net_bonding, pmd_bond_drv);
3771 RTE_PMD_REGISTER_ALIAS(net_bonding, eth_bond);
3772
3773 RTE_PMD_REGISTER_PARAM_STRING(net_bonding,
3774         "slave=<ifc> "
3775         "primary=<ifc> "
3776         "mode=[0-6] "
3777         "xmit_policy=[l2 | l23 | l34] "
3778         "agg_mode=[count | stable | bandwidth] "
3779         "socket_id=<int> "
3780         "mac=<mac addr> "
3781         "lsc_poll_period_ms=<int> "
3782         "up_delay=<int> "
3783         "down_delay=<int>");
3784
3785 /* We can't use RTE_LOG_REGISTER_DEFAULT because of the forced name for
3786  * this library, see meson.build.
3787  */
3788 RTE_LOG_REGISTER(bond_logtype, pmd.net.bonding, NOTICE);