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                 /* rss_key won't be empty if RSS is configured in bonded dev */
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
1710                 slave_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf =
1711                                 bonded_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf;
1712                 slave_eth_dev->data->dev_conf.rxmode.mq_mode =
1713                                 bonded_eth_dev->data->dev_conf.rxmode.mq_mode;
1714         }
1715
1716         if (bonded_eth_dev->data->dev_conf.rxmode.offloads &
1717                         DEV_RX_OFFLOAD_VLAN_FILTER)
1718                 slave_eth_dev->data->dev_conf.rxmode.offloads |=
1719                                 DEV_RX_OFFLOAD_VLAN_FILTER;
1720         else
1721                 slave_eth_dev->data->dev_conf.rxmode.offloads &=
1722                                 ~DEV_RX_OFFLOAD_VLAN_FILTER;
1723
1724         slave_eth_dev->data->dev_conf.rxmode.max_rx_pkt_len =
1725                         bonded_eth_dev->data->dev_conf.rxmode.max_rx_pkt_len;
1726
1727         if (bonded_eth_dev->data->dev_conf.rxmode.offloads &
1728                         DEV_RX_OFFLOAD_JUMBO_FRAME)
1729                 slave_eth_dev->data->dev_conf.rxmode.offloads |=
1730                                 DEV_RX_OFFLOAD_JUMBO_FRAME;
1731         else
1732                 slave_eth_dev->data->dev_conf.rxmode.offloads &=
1733                                 ~DEV_RX_OFFLOAD_JUMBO_FRAME;
1734
1735         nb_rx_queues = bonded_eth_dev->data->nb_rx_queues;
1736         nb_tx_queues = bonded_eth_dev->data->nb_tx_queues;
1737
1738         if (internals->mode == BONDING_MODE_8023AD) {
1739                 if (internals->mode4.dedicated_queues.enabled == 1) {
1740                         nb_rx_queues++;
1741                         nb_tx_queues++;
1742                 }
1743         }
1744
1745         errval = rte_eth_dev_set_mtu(slave_eth_dev->data->port_id,
1746                                      bonded_eth_dev->data->mtu);
1747         if (errval != 0 && errval != -ENOTSUP) {
1748                 RTE_BOND_LOG(ERR, "rte_eth_dev_set_mtu: port %u, err (%d)",
1749                                 slave_eth_dev->data->port_id, errval);
1750                 return errval;
1751         }
1752
1753         /* Configure device */
1754         errval = rte_eth_dev_configure(slave_eth_dev->data->port_id,
1755                         nb_rx_queues, nb_tx_queues,
1756                         &(slave_eth_dev->data->dev_conf));
1757         if (errval != 0) {
1758                 RTE_BOND_LOG(ERR, "Cannot configure slave device: port %u, err (%d)",
1759                                 slave_eth_dev->data->port_id, errval);
1760                 return errval;
1761         }
1762
1763         /* Setup Rx Queues */
1764         for (q_id = 0; q_id < bonded_eth_dev->data->nb_rx_queues; q_id++) {
1765                 bd_rx_q = (struct bond_rx_queue *)bonded_eth_dev->data->rx_queues[q_id];
1766
1767                 errval = rte_eth_rx_queue_setup(slave_eth_dev->data->port_id, q_id,
1768                                 bd_rx_q->nb_rx_desc,
1769                                 rte_eth_dev_socket_id(slave_eth_dev->data->port_id),
1770                                 &(bd_rx_q->rx_conf), bd_rx_q->mb_pool);
1771                 if (errval != 0) {
1772                         RTE_BOND_LOG(ERR,
1773                                         "rte_eth_rx_queue_setup: port=%d queue_id %d, err (%d)",
1774                                         slave_eth_dev->data->port_id, q_id, errval);
1775                         return errval;
1776                 }
1777         }
1778
1779         /* Setup Tx Queues */
1780         for (q_id = 0; q_id < bonded_eth_dev->data->nb_tx_queues; q_id++) {
1781                 bd_tx_q = (struct bond_tx_queue *)bonded_eth_dev->data->tx_queues[q_id];
1782
1783                 errval = rte_eth_tx_queue_setup(slave_eth_dev->data->port_id, q_id,
1784                                 bd_tx_q->nb_tx_desc,
1785                                 rte_eth_dev_socket_id(slave_eth_dev->data->port_id),
1786                                 &bd_tx_q->tx_conf);
1787                 if (errval != 0) {
1788                         RTE_BOND_LOG(ERR,
1789                                 "rte_eth_tx_queue_setup: port=%d queue_id %d, err (%d)",
1790                                 slave_eth_dev->data->port_id, q_id, errval);
1791                         return errval;
1792                 }
1793         }
1794
1795         if (internals->mode == BONDING_MODE_8023AD &&
1796                         internals->mode4.dedicated_queues.enabled == 1) {
1797                 if (slave_configure_slow_queue(bonded_eth_dev, slave_eth_dev)
1798                                 != 0)
1799                         return errval;
1800
1801                 errval = bond_ethdev_8023ad_flow_verify(bonded_eth_dev,
1802                                 slave_eth_dev->data->port_id);
1803                 if (errval != 0) {
1804                         RTE_BOND_LOG(ERR,
1805                                 "bond_ethdev_8023ad_flow_verify: port=%d, err (%d)",
1806                                 slave_eth_dev->data->port_id, errval);
1807                         return errval;
1808                 }
1809
1810                 if (internals->mode4.dedicated_queues.flow[slave_eth_dev->data->port_id] != NULL)
1811                         rte_flow_destroy(slave_eth_dev->data->port_id,
1812                                         internals->mode4.dedicated_queues.flow[slave_eth_dev->data->port_id],
1813                                         &flow_error);
1814
1815                 errval = bond_ethdev_8023ad_flow_set(bonded_eth_dev,
1816                                 slave_eth_dev->data->port_id);
1817                 if (errval != 0) {
1818                         RTE_BOND_LOG(ERR,
1819                                 "bond_ethdev_8023ad_flow_set: port=%d, err (%d)",
1820                                 slave_eth_dev->data->port_id, errval);
1821                         return errval;
1822                 }
1823         }
1824
1825         /* Start device */
1826         errval = rte_eth_dev_start(slave_eth_dev->data->port_id);
1827         if (errval != 0) {
1828                 RTE_BOND_LOG(ERR, "rte_eth_dev_start: port=%u, err (%d)",
1829                                 slave_eth_dev->data->port_id, errval);
1830                 return -1;
1831         }
1832
1833         /* If RSS is enabled for bonding, synchronize RETA */
1834         if (bonded_eth_dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS) {
1835                 int i;
1836                 struct bond_dev_private *internals;
1837
1838                 internals = bonded_eth_dev->data->dev_private;
1839
1840                 for (i = 0; i < internals->slave_count; i++) {
1841                         if (internals->slaves[i].port_id == slave_eth_dev->data->port_id) {
1842                                 errval = rte_eth_dev_rss_reta_update(
1843                                                 slave_eth_dev->data->port_id,
1844                                                 &internals->reta_conf[0],
1845                                                 internals->slaves[i].reta_size);
1846                                 if (errval != 0) {
1847                                         RTE_BOND_LOG(WARNING,
1848                                                      "rte_eth_dev_rss_reta_update on slave port %d fails (err %d)."
1849                                                      " RSS Configuration for bonding may be inconsistent.",
1850                                                      slave_eth_dev->data->port_id, errval);
1851                                 }
1852                                 break;
1853                         }
1854                 }
1855         }
1856
1857         /* If lsc interrupt is set, check initial slave's link status */
1858         if (slave_eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
1859                 slave_eth_dev->dev_ops->link_update(slave_eth_dev, 0);
1860                 bond_ethdev_lsc_event_callback(slave_eth_dev->data->port_id,
1861                         RTE_ETH_EVENT_INTR_LSC, &bonded_eth_dev->data->port_id,
1862                         NULL);
1863         }
1864
1865         return 0;
1866 }
1867
1868 void
1869 slave_remove(struct bond_dev_private *internals,
1870                 struct rte_eth_dev *slave_eth_dev)
1871 {
1872         uint16_t i;
1873
1874         for (i = 0; i < internals->slave_count; i++)
1875                 if (internals->slaves[i].port_id ==
1876                                 slave_eth_dev->data->port_id)
1877                         break;
1878
1879         if (i < (internals->slave_count - 1)) {
1880                 struct rte_flow *flow;
1881
1882                 memmove(&internals->slaves[i], &internals->slaves[i + 1],
1883                                 sizeof(internals->slaves[0]) *
1884                                 (internals->slave_count - i - 1));
1885                 TAILQ_FOREACH(flow, &internals->flow_list, next) {
1886                         memmove(&flow->flows[i], &flow->flows[i + 1],
1887                                 sizeof(flow->flows[0]) *
1888                                 (internals->slave_count - i - 1));
1889                         flow->flows[internals->slave_count - 1] = NULL;
1890                 }
1891         }
1892
1893         internals->slave_count--;
1894
1895         /* force reconfiguration of slave interfaces */
1896         rte_eth_dev_internal_reset(slave_eth_dev);
1897 }
1898
1899 static void
1900 bond_ethdev_slave_link_status_change_monitor(void *cb_arg);
1901
1902 void
1903 slave_add(struct bond_dev_private *internals,
1904                 struct rte_eth_dev *slave_eth_dev)
1905 {
1906         struct bond_slave_details *slave_details =
1907                         &internals->slaves[internals->slave_count];
1908
1909         slave_details->port_id = slave_eth_dev->data->port_id;
1910         slave_details->last_link_status = 0;
1911
1912         /* Mark slave devices that don't support interrupts so we can
1913          * compensate when we start the bond
1914          */
1915         if (!(slave_eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
1916                 slave_details->link_status_poll_enabled = 1;
1917         }
1918
1919         slave_details->link_status_wait_to_complete = 0;
1920         /* clean tlb_last_obytes when adding port for bonding device */
1921         memcpy(&(slave_details->persisted_mac_addr), slave_eth_dev->data->mac_addrs,
1922                         sizeof(struct rte_ether_addr));
1923 }
1924
1925 void
1926 bond_ethdev_primary_set(struct bond_dev_private *internals,
1927                 uint16_t slave_port_id)
1928 {
1929         int i;
1930
1931         if (internals->active_slave_count < 1)
1932                 internals->current_primary_port = slave_port_id;
1933         else
1934                 /* Search bonded device slave ports for new proposed primary port */
1935                 for (i = 0; i < internals->active_slave_count; i++) {
1936                         if (internals->active_slaves[i] == slave_port_id)
1937                                 internals->current_primary_port = slave_port_id;
1938                 }
1939 }
1940
1941 static int
1942 bond_ethdev_promiscuous_enable(struct rte_eth_dev *eth_dev);
1943
1944 static int
1945 bond_ethdev_start(struct rte_eth_dev *eth_dev)
1946 {
1947         struct bond_dev_private *internals;
1948         int i;
1949
1950         /* slave eth dev will be started by bonded device */
1951         if (check_for_bonded_ethdev(eth_dev)) {
1952                 RTE_BOND_LOG(ERR, "User tried to explicitly start a slave eth_dev (%d)",
1953                                 eth_dev->data->port_id);
1954                 return -1;
1955         }
1956
1957         eth_dev->data->dev_link.link_status = ETH_LINK_DOWN;
1958         eth_dev->data->dev_started = 1;
1959
1960         internals = eth_dev->data->dev_private;
1961
1962         if (internals->slave_count == 0) {
1963                 RTE_BOND_LOG(ERR, "Cannot start port since there are no slave devices");
1964                 goto out_err;
1965         }
1966
1967         if (internals->user_defined_mac == 0) {
1968                 struct rte_ether_addr *new_mac_addr = NULL;
1969
1970                 for (i = 0; i < internals->slave_count; i++)
1971                         if (internals->slaves[i].port_id == internals->primary_port)
1972                                 new_mac_addr = &internals->slaves[i].persisted_mac_addr;
1973
1974                 if (new_mac_addr == NULL)
1975                         goto out_err;
1976
1977                 if (mac_address_set(eth_dev, new_mac_addr) != 0) {
1978                         RTE_BOND_LOG(ERR, "bonded port (%d) failed to update MAC address",
1979                                         eth_dev->data->port_id);
1980                         goto out_err;
1981                 }
1982         }
1983
1984         if (internals->mode == BONDING_MODE_8023AD) {
1985                 if (internals->mode4.dedicated_queues.enabled == 1) {
1986                         internals->mode4.dedicated_queues.rx_qid =
1987                                         eth_dev->data->nb_rx_queues;
1988                         internals->mode4.dedicated_queues.tx_qid =
1989                                         eth_dev->data->nb_tx_queues;
1990                 }
1991         }
1992
1993
1994         /* Reconfigure each slave device if starting bonded device */
1995         for (i = 0; i < internals->slave_count; i++) {
1996                 struct rte_eth_dev *slave_ethdev =
1997                                 &(rte_eth_devices[internals->slaves[i].port_id]);
1998                 if (slave_configure(eth_dev, slave_ethdev) != 0) {
1999                         RTE_BOND_LOG(ERR,
2000                                 "bonded port (%d) failed to reconfigure slave device (%d)",
2001                                 eth_dev->data->port_id,
2002                                 internals->slaves[i].port_id);
2003                         goto out_err;
2004                 }
2005                 /* We will need to poll for link status if any slave doesn't
2006                  * support interrupts
2007                  */
2008                 if (internals->slaves[i].link_status_poll_enabled)
2009                         internals->link_status_polling_enabled = 1;
2010         }
2011
2012         /* start polling if needed */
2013         if (internals->link_status_polling_enabled) {
2014                 rte_eal_alarm_set(
2015                         internals->link_status_polling_interval_ms * 1000,
2016                         bond_ethdev_slave_link_status_change_monitor,
2017                         (void *)&rte_eth_devices[internals->port_id]);
2018         }
2019
2020         /* Update all slave devices MACs*/
2021         if (mac_address_slaves_update(eth_dev) != 0)
2022                 goto out_err;
2023
2024         if (internals->user_defined_primary_port)
2025                 bond_ethdev_primary_set(internals, internals->primary_port);
2026
2027         if (internals->mode == BONDING_MODE_8023AD)
2028                 bond_mode_8023ad_start(eth_dev);
2029
2030         if (internals->mode == BONDING_MODE_TLB ||
2031                         internals->mode == BONDING_MODE_ALB)
2032                 bond_tlb_enable(internals);
2033
2034         return 0;
2035
2036 out_err:
2037         eth_dev->data->dev_started = 0;
2038         return -1;
2039 }
2040
2041 static void
2042 bond_ethdev_free_queues(struct rte_eth_dev *dev)
2043 {
2044         uint16_t i;
2045
2046         if (dev->data->rx_queues != NULL) {
2047                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
2048                         rte_free(dev->data->rx_queues[i]);
2049                         dev->data->rx_queues[i] = NULL;
2050                 }
2051                 dev->data->nb_rx_queues = 0;
2052         }
2053
2054         if (dev->data->tx_queues != NULL) {
2055                 for (i = 0; i < dev->data->nb_tx_queues; i++) {
2056                         rte_free(dev->data->tx_queues[i]);
2057                         dev->data->tx_queues[i] = NULL;
2058                 }
2059                 dev->data->nb_tx_queues = 0;
2060         }
2061 }
2062
2063 int
2064 bond_ethdev_stop(struct rte_eth_dev *eth_dev)
2065 {
2066         struct bond_dev_private *internals = eth_dev->data->dev_private;
2067         uint16_t i;
2068         int ret;
2069
2070         if (internals->mode == BONDING_MODE_8023AD) {
2071                 struct port *port;
2072                 void *pkt = NULL;
2073
2074                 bond_mode_8023ad_stop(eth_dev);
2075
2076                 /* Discard all messages to/from mode 4 state machines */
2077                 for (i = 0; i < internals->active_slave_count; i++) {
2078                         port = &bond_mode_8023ad_ports[internals->active_slaves[i]];
2079
2080                         RTE_ASSERT(port->rx_ring != NULL);
2081                         while (rte_ring_dequeue(port->rx_ring, &pkt) != -ENOENT)
2082                                 rte_pktmbuf_free(pkt);
2083
2084                         RTE_ASSERT(port->tx_ring != NULL);
2085                         while (rte_ring_dequeue(port->tx_ring, &pkt) != -ENOENT)
2086                                 rte_pktmbuf_free(pkt);
2087                 }
2088         }
2089
2090         if (internals->mode == BONDING_MODE_TLB ||
2091                         internals->mode == BONDING_MODE_ALB) {
2092                 bond_tlb_disable(internals);
2093                 for (i = 0; i < internals->active_slave_count; i++)
2094                         tlb_last_obytets[internals->active_slaves[i]] = 0;
2095         }
2096
2097         eth_dev->data->dev_link.link_status = ETH_LINK_DOWN;
2098         eth_dev->data->dev_started = 0;
2099
2100         internals->link_status_polling_enabled = 0;
2101         for (i = 0; i < internals->slave_count; i++) {
2102                 uint16_t slave_id = internals->slaves[i].port_id;
2103                 if (find_slave_by_id(internals->active_slaves,
2104                                 internals->active_slave_count, slave_id) !=
2105                                                 internals->active_slave_count) {
2106                         internals->slaves[i].last_link_status = 0;
2107                         ret = rte_eth_dev_stop(slave_id);
2108                         if (ret != 0) {
2109                                 RTE_BOND_LOG(ERR, "Failed to stop device on port %u",
2110                                              slave_id);
2111                                 return ret;
2112                         }
2113                         deactivate_slave(eth_dev, slave_id);
2114                 }
2115         }
2116
2117         return 0;
2118 }
2119
2120 int
2121 bond_ethdev_close(struct rte_eth_dev *dev)
2122 {
2123         struct bond_dev_private *internals = dev->data->dev_private;
2124         uint16_t bond_port_id = internals->port_id;
2125         int skipped = 0;
2126         struct rte_flow_error ferror;
2127
2128         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2129                 return 0;
2130
2131         RTE_BOND_LOG(INFO, "Closing bonded device %s", dev->device->name);
2132         while (internals->slave_count != skipped) {
2133                 uint16_t port_id = internals->slaves[skipped].port_id;
2134
2135                 if (rte_eth_dev_stop(port_id) != 0) {
2136                         RTE_BOND_LOG(ERR, "Failed to stop device on port %u",
2137                                      port_id);
2138                         skipped++;
2139                 }
2140
2141                 if (rte_eth_bond_slave_remove(bond_port_id, port_id) != 0) {
2142                         RTE_BOND_LOG(ERR,
2143                                      "Failed to remove port %d from bonded device %s",
2144                                      port_id, dev->device->name);
2145                         skipped++;
2146                 }
2147         }
2148         bond_flow_ops.flush(dev, &ferror);
2149         bond_ethdev_free_queues(dev);
2150         rte_bitmap_reset(internals->vlan_filter_bmp);
2151         rte_bitmap_free(internals->vlan_filter_bmp);
2152         rte_free(internals->vlan_filter_bmpmem);
2153
2154         /* Try to release mempool used in mode6. If the bond
2155          * device is not mode6, free the NULL is not problem.
2156          */
2157         rte_mempool_free(internals->mode6.mempool);
2158
2159         if (internals->kvlist != NULL)
2160                 rte_kvargs_free(internals->kvlist);
2161
2162         return 0;
2163 }
2164
2165 /* forward declaration */
2166 static int bond_ethdev_configure(struct rte_eth_dev *dev);
2167
2168 static int
2169 bond_ethdev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2170 {
2171         struct bond_dev_private *internals = dev->data->dev_private;
2172         struct bond_slave_details slave;
2173         int ret;
2174
2175         uint16_t max_nb_rx_queues = UINT16_MAX;
2176         uint16_t max_nb_tx_queues = UINT16_MAX;
2177         uint16_t max_rx_desc_lim = UINT16_MAX;
2178         uint16_t max_tx_desc_lim = UINT16_MAX;
2179
2180         dev_info->max_mac_addrs = BOND_MAX_MAC_ADDRS;
2181
2182         dev_info->max_rx_pktlen = internals->candidate_max_rx_pktlen ?
2183                         internals->candidate_max_rx_pktlen :
2184                         RTE_ETHER_MAX_JUMBO_FRAME_LEN;
2185
2186         /* Max number of tx/rx queues that the bonded device can support is the
2187          * minimum values of the bonded slaves, as all slaves must be capable
2188          * of supporting the same number of tx/rx queues.
2189          */
2190         if (internals->slave_count > 0) {
2191                 struct rte_eth_dev_info slave_info;
2192                 uint16_t idx;
2193
2194                 for (idx = 0; idx < internals->slave_count; idx++) {
2195                         slave = internals->slaves[idx];
2196                         ret = rte_eth_dev_info_get(slave.port_id, &slave_info);
2197                         if (ret != 0) {
2198                                 RTE_BOND_LOG(ERR,
2199                                         "%s: Error during getting device (port %u) info: %s\n",
2200                                         __func__,
2201                                         slave.port_id,
2202                                         strerror(-ret));
2203
2204                                 return ret;
2205                         }
2206
2207                         if (slave_info.max_rx_queues < max_nb_rx_queues)
2208                                 max_nb_rx_queues = slave_info.max_rx_queues;
2209
2210                         if (slave_info.max_tx_queues < max_nb_tx_queues)
2211                                 max_nb_tx_queues = slave_info.max_tx_queues;
2212
2213                         if (slave_info.rx_desc_lim.nb_max < max_rx_desc_lim)
2214                                 max_rx_desc_lim = slave_info.rx_desc_lim.nb_max;
2215
2216                         if (slave_info.tx_desc_lim.nb_max < max_tx_desc_lim)
2217                                 max_tx_desc_lim = slave_info.tx_desc_lim.nb_max;
2218                 }
2219         }
2220
2221         dev_info->max_rx_queues = max_nb_rx_queues;
2222         dev_info->max_tx_queues = max_nb_tx_queues;
2223
2224         memcpy(&dev_info->default_rxconf, &internals->default_rxconf,
2225                sizeof(dev_info->default_rxconf));
2226         memcpy(&dev_info->default_txconf, &internals->default_txconf,
2227                sizeof(dev_info->default_txconf));
2228
2229         dev_info->rx_desc_lim.nb_max = max_rx_desc_lim;
2230         dev_info->tx_desc_lim.nb_max = max_tx_desc_lim;
2231
2232         /**
2233          * If dedicated hw queues enabled for link bonding device in LACP mode
2234          * then we need to reduce the maximum number of data path queues by 1.
2235          */
2236         if (internals->mode == BONDING_MODE_8023AD &&
2237                 internals->mode4.dedicated_queues.enabled == 1) {
2238                 dev_info->max_rx_queues--;
2239                 dev_info->max_tx_queues--;
2240         }
2241
2242         dev_info->min_rx_bufsize = 0;
2243
2244         dev_info->rx_offload_capa = internals->rx_offload_capa;
2245         dev_info->tx_offload_capa = internals->tx_offload_capa;
2246         dev_info->rx_queue_offload_capa = internals->rx_queue_offload_capa;
2247         dev_info->tx_queue_offload_capa = internals->tx_queue_offload_capa;
2248         dev_info->flow_type_rss_offloads = internals->flow_type_rss_offloads;
2249
2250         dev_info->reta_size = internals->reta_size;
2251         dev_info->hash_key_size = internals->rss_key_len;
2252
2253         return 0;
2254 }
2255
2256 static int
2257 bond_ethdev_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
2258 {
2259         int res;
2260         uint16_t i;
2261         struct bond_dev_private *internals = dev->data->dev_private;
2262
2263         /* don't do this while a slave is being added */
2264         rte_spinlock_lock(&internals->lock);
2265
2266         if (on)
2267                 rte_bitmap_set(internals->vlan_filter_bmp, vlan_id);
2268         else
2269                 rte_bitmap_clear(internals->vlan_filter_bmp, vlan_id);
2270
2271         for (i = 0; i < internals->slave_count; i++) {
2272                 uint16_t port_id = internals->slaves[i].port_id;
2273
2274                 res = rte_eth_dev_vlan_filter(port_id, vlan_id, on);
2275                 if (res == ENOTSUP)
2276                         RTE_BOND_LOG(WARNING,
2277                                      "Setting VLAN filter on slave port %u not supported.",
2278                                      port_id);
2279         }
2280
2281         rte_spinlock_unlock(&internals->lock);
2282         return 0;
2283 }
2284
2285 static int
2286 bond_ethdev_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
2287                 uint16_t nb_rx_desc, unsigned int socket_id __rte_unused,
2288                 const struct rte_eth_rxconf *rx_conf, struct rte_mempool *mb_pool)
2289 {
2290         struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)
2291                         rte_zmalloc_socket(NULL, sizeof(struct bond_rx_queue),
2292                                         0, dev->data->numa_node);
2293         if (bd_rx_q == NULL)
2294                 return -1;
2295
2296         bd_rx_q->queue_id = rx_queue_id;
2297         bd_rx_q->dev_private = dev->data->dev_private;
2298
2299         bd_rx_q->nb_rx_desc = nb_rx_desc;
2300
2301         memcpy(&(bd_rx_q->rx_conf), rx_conf, sizeof(struct rte_eth_rxconf));
2302         bd_rx_q->mb_pool = mb_pool;
2303
2304         dev->data->rx_queues[rx_queue_id] = bd_rx_q;
2305
2306         return 0;
2307 }
2308
2309 static int
2310 bond_ethdev_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
2311                 uint16_t nb_tx_desc, unsigned int socket_id __rte_unused,
2312                 const struct rte_eth_txconf *tx_conf)
2313 {
2314         struct bond_tx_queue *bd_tx_q  = (struct bond_tx_queue *)
2315                         rte_zmalloc_socket(NULL, sizeof(struct bond_tx_queue),
2316                                         0, dev->data->numa_node);
2317
2318         if (bd_tx_q == NULL)
2319                 return -1;
2320
2321         bd_tx_q->queue_id = tx_queue_id;
2322         bd_tx_q->dev_private = dev->data->dev_private;
2323
2324         bd_tx_q->nb_tx_desc = nb_tx_desc;
2325         memcpy(&(bd_tx_q->tx_conf), tx_conf, sizeof(bd_tx_q->tx_conf));
2326
2327         dev->data->tx_queues[tx_queue_id] = bd_tx_q;
2328
2329         return 0;
2330 }
2331
2332 static void
2333 bond_ethdev_rx_queue_release(struct rte_eth_dev *dev, uint16_t queue_id)
2334 {
2335         void *queue = dev->data->rx_queues[queue_id];
2336
2337         if (queue == NULL)
2338                 return;
2339
2340         rte_free(queue);
2341 }
2342
2343 static void
2344 bond_ethdev_tx_queue_release(struct rte_eth_dev *dev, uint16_t queue_id)
2345 {
2346         void *queue = dev->data->rx_queues[queue_id];
2347
2348         if (queue == NULL)
2349                 return;
2350
2351         rte_free(queue);
2352 }
2353
2354 static void
2355 bond_ethdev_slave_link_status_change_monitor(void *cb_arg)
2356 {
2357         struct rte_eth_dev *bonded_ethdev, *slave_ethdev;
2358         struct bond_dev_private *internals;
2359
2360         /* Default value for polling slave found is true as we don't want to
2361          * disable the polling thread if we cannot get the lock */
2362         int i, polling_slave_found = 1;
2363
2364         if (cb_arg == NULL)
2365                 return;
2366
2367         bonded_ethdev = cb_arg;
2368         internals = bonded_ethdev->data->dev_private;
2369
2370         if (!bonded_ethdev->data->dev_started ||
2371                 !internals->link_status_polling_enabled)
2372                 return;
2373
2374         /* If device is currently being configured then don't check slaves link
2375          * status, wait until next period */
2376         if (rte_spinlock_trylock(&internals->lock)) {
2377                 if (internals->slave_count > 0)
2378                         polling_slave_found = 0;
2379
2380                 for (i = 0; i < internals->slave_count; i++) {
2381                         if (!internals->slaves[i].link_status_poll_enabled)
2382                                 continue;
2383
2384                         slave_ethdev = &rte_eth_devices[internals->slaves[i].port_id];
2385                         polling_slave_found = 1;
2386
2387                         /* Update slave link status */
2388                         (*slave_ethdev->dev_ops->link_update)(slave_ethdev,
2389                                         internals->slaves[i].link_status_wait_to_complete);
2390
2391                         /* if link status has changed since last checked then call lsc
2392                          * event callback */
2393                         if (slave_ethdev->data->dev_link.link_status !=
2394                                         internals->slaves[i].last_link_status) {
2395                                 internals->slaves[i].last_link_status =
2396                                                 slave_ethdev->data->dev_link.link_status;
2397
2398                                 bond_ethdev_lsc_event_callback(internals->slaves[i].port_id,
2399                                                 RTE_ETH_EVENT_INTR_LSC,
2400                                                 &bonded_ethdev->data->port_id,
2401                                                 NULL);
2402                         }
2403                 }
2404                 rte_spinlock_unlock(&internals->lock);
2405         }
2406
2407         if (polling_slave_found)
2408                 /* Set alarm to continue monitoring link status of slave ethdev's */
2409                 rte_eal_alarm_set(internals->link_status_polling_interval_ms * 1000,
2410                                 bond_ethdev_slave_link_status_change_monitor, cb_arg);
2411 }
2412
2413 static int
2414 bond_ethdev_link_update(struct rte_eth_dev *ethdev, int wait_to_complete)
2415 {
2416         int (*link_update)(uint16_t port_id, struct rte_eth_link *eth_link);
2417
2418         struct bond_dev_private *bond_ctx;
2419         struct rte_eth_link slave_link;
2420
2421         bool one_link_update_succeeded;
2422         uint32_t idx;
2423         int ret;
2424
2425         bond_ctx = ethdev->data->dev_private;
2426
2427         ethdev->data->dev_link.link_speed = ETH_SPEED_NUM_NONE;
2428
2429         if (ethdev->data->dev_started == 0 ||
2430                         bond_ctx->active_slave_count == 0) {
2431                 ethdev->data->dev_link.link_status = ETH_LINK_DOWN;
2432                 return 0;
2433         }
2434
2435         ethdev->data->dev_link.link_status = ETH_LINK_UP;
2436
2437         if (wait_to_complete)
2438                 link_update = rte_eth_link_get;
2439         else
2440                 link_update = rte_eth_link_get_nowait;
2441
2442         switch (bond_ctx->mode) {
2443         case BONDING_MODE_BROADCAST:
2444                 /**
2445                  * Setting link speed to UINT32_MAX to ensure we pick up the
2446                  * value of the first active slave
2447                  */
2448                 ethdev->data->dev_link.link_speed = UINT32_MAX;
2449
2450                 /**
2451                  * link speed is minimum value of all the slaves link speed as
2452                  * packet loss will occur on this slave if transmission at rates
2453                  * greater than this are attempted
2454                  */
2455                 for (idx = 0; idx < bond_ctx->active_slave_count; idx++) {
2456                         ret = link_update(bond_ctx->active_slaves[idx],
2457                                           &slave_link);
2458                         if (ret < 0) {
2459                                 ethdev->data->dev_link.link_speed =
2460                                         ETH_SPEED_NUM_NONE;
2461                                 RTE_BOND_LOG(ERR,
2462                                         "Slave (port %u) link get failed: %s",
2463                                         bond_ctx->active_slaves[idx],
2464                                         rte_strerror(-ret));
2465                                 return 0;
2466                         }
2467
2468                         if (slave_link.link_speed <
2469                                         ethdev->data->dev_link.link_speed)
2470                                 ethdev->data->dev_link.link_speed =
2471                                                 slave_link.link_speed;
2472                 }
2473                 break;
2474         case BONDING_MODE_ACTIVE_BACKUP:
2475                 /* Current primary slave */
2476                 ret = link_update(bond_ctx->current_primary_port, &slave_link);
2477                 if (ret < 0) {
2478                         RTE_BOND_LOG(ERR, "Slave (port %u) link get failed: %s",
2479                                 bond_ctx->current_primary_port,
2480                                 rte_strerror(-ret));
2481                         return 0;
2482                 }
2483
2484                 ethdev->data->dev_link.link_speed = slave_link.link_speed;
2485                 break;
2486         case BONDING_MODE_8023AD:
2487                 ethdev->data->dev_link.link_autoneg =
2488                                 bond_ctx->mode4.slave_link.link_autoneg;
2489                 ethdev->data->dev_link.link_duplex =
2490                                 bond_ctx->mode4.slave_link.link_duplex;
2491                 /* fall through */
2492                 /* to update link speed */
2493         case BONDING_MODE_ROUND_ROBIN:
2494         case BONDING_MODE_BALANCE:
2495         case BONDING_MODE_TLB:
2496         case BONDING_MODE_ALB:
2497         default:
2498                 /**
2499                  * In theses mode the maximum theoretical link speed is the sum
2500                  * of all the slaves
2501                  */
2502                 ethdev->data->dev_link.link_speed = ETH_SPEED_NUM_NONE;
2503                 one_link_update_succeeded = false;
2504
2505                 for (idx = 0; idx < bond_ctx->active_slave_count; idx++) {
2506                         ret = link_update(bond_ctx->active_slaves[idx],
2507                                         &slave_link);
2508                         if (ret < 0) {
2509                                 RTE_BOND_LOG(ERR,
2510                                         "Slave (port %u) link get failed: %s",
2511                                         bond_ctx->active_slaves[idx],
2512                                         rte_strerror(-ret));
2513                                 continue;
2514                         }
2515
2516                         one_link_update_succeeded = true;
2517                         ethdev->data->dev_link.link_speed +=
2518                                         slave_link.link_speed;
2519                 }
2520
2521                 if (!one_link_update_succeeded) {
2522                         RTE_BOND_LOG(ERR, "All slaves link get failed");
2523                         return 0;
2524                 }
2525         }
2526
2527
2528         return 0;
2529 }
2530
2531
2532 static int
2533 bond_ethdev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
2534 {
2535         struct bond_dev_private *internals = dev->data->dev_private;
2536         struct rte_eth_stats slave_stats;
2537         int i, j;
2538
2539         for (i = 0; i < internals->slave_count; i++) {
2540                 rte_eth_stats_get(internals->slaves[i].port_id, &slave_stats);
2541
2542                 stats->ipackets += slave_stats.ipackets;
2543                 stats->opackets += slave_stats.opackets;
2544                 stats->ibytes += slave_stats.ibytes;
2545                 stats->obytes += slave_stats.obytes;
2546                 stats->imissed += slave_stats.imissed;
2547                 stats->ierrors += slave_stats.ierrors;
2548                 stats->oerrors += slave_stats.oerrors;
2549                 stats->rx_nombuf += slave_stats.rx_nombuf;
2550
2551                 for (j = 0; j < RTE_ETHDEV_QUEUE_STAT_CNTRS; j++) {
2552                         stats->q_ipackets[j] += slave_stats.q_ipackets[j];
2553                         stats->q_opackets[j] += slave_stats.q_opackets[j];
2554                         stats->q_ibytes[j] += slave_stats.q_ibytes[j];
2555                         stats->q_obytes[j] += slave_stats.q_obytes[j];
2556                         stats->q_errors[j] += slave_stats.q_errors[j];
2557                 }
2558
2559         }
2560
2561         return 0;
2562 }
2563
2564 static int
2565 bond_ethdev_stats_reset(struct rte_eth_dev *dev)
2566 {
2567         struct bond_dev_private *internals = dev->data->dev_private;
2568         int i;
2569         int err;
2570         int ret;
2571
2572         for (i = 0, err = 0; i < internals->slave_count; i++) {
2573                 ret = rte_eth_stats_reset(internals->slaves[i].port_id);
2574                 if (ret != 0)
2575                         err = ret;
2576         }
2577
2578         return err;
2579 }
2580
2581 static int
2582 bond_ethdev_promiscuous_enable(struct rte_eth_dev *eth_dev)
2583 {
2584         struct bond_dev_private *internals = eth_dev->data->dev_private;
2585         int i;
2586         int ret = 0;
2587         uint16_t port_id;
2588
2589         switch (internals->mode) {
2590         /* Promiscuous mode is propagated to all slaves */
2591         case BONDING_MODE_ROUND_ROBIN:
2592         case BONDING_MODE_BALANCE:
2593         case BONDING_MODE_BROADCAST:
2594         case BONDING_MODE_8023AD: {
2595                 unsigned int slave_ok = 0;
2596
2597                 for (i = 0; i < internals->slave_count; i++) {
2598                         port_id = internals->slaves[i].port_id;
2599
2600                         ret = rte_eth_promiscuous_enable(port_id);
2601                         if (ret != 0)
2602                                 RTE_BOND_LOG(ERR,
2603                                         "Failed to enable promiscuous mode for port %u: %s",
2604                                         port_id, rte_strerror(-ret));
2605                         else
2606                                 slave_ok++;
2607                 }
2608                 /*
2609                  * Report success if operation is successful on at least
2610                  * on one slave. Otherwise return last error code.
2611                  */
2612                 if (slave_ok > 0)
2613                         ret = 0;
2614                 break;
2615         }
2616         /* Promiscuous mode is propagated only to primary slave */
2617         case BONDING_MODE_ACTIVE_BACKUP:
2618         case BONDING_MODE_TLB:
2619         case BONDING_MODE_ALB:
2620         default:
2621                 /* Do not touch promisc when there cannot be primary ports */
2622                 if (internals->slave_count == 0)
2623                         break;
2624                 port_id = internals->current_primary_port;
2625                 ret = rte_eth_promiscuous_enable(port_id);
2626                 if (ret != 0)
2627                         RTE_BOND_LOG(ERR,
2628                                 "Failed to enable promiscuous mode for port %u: %s",
2629                                 port_id, rte_strerror(-ret));
2630         }
2631
2632         return ret;
2633 }
2634
2635 static int
2636 bond_ethdev_promiscuous_disable(struct rte_eth_dev *dev)
2637 {
2638         struct bond_dev_private *internals = dev->data->dev_private;
2639         int i;
2640         int ret = 0;
2641         uint16_t port_id;
2642
2643         switch (internals->mode) {
2644         /* Promiscuous mode is propagated to all slaves */
2645         case BONDING_MODE_ROUND_ROBIN:
2646         case BONDING_MODE_BALANCE:
2647         case BONDING_MODE_BROADCAST:
2648         case BONDING_MODE_8023AD: {
2649                 unsigned int slave_ok = 0;
2650
2651                 for (i = 0; i < internals->slave_count; i++) {
2652                         port_id = internals->slaves[i].port_id;
2653
2654                         if (internals->mode == BONDING_MODE_8023AD &&
2655                             bond_mode_8023ad_ports[port_id].forced_rx_flags ==
2656                                         BOND_8023AD_FORCED_PROMISC) {
2657                                 slave_ok++;
2658                                 continue;
2659                         }
2660                         ret = rte_eth_promiscuous_disable(port_id);
2661                         if (ret != 0)
2662                                 RTE_BOND_LOG(ERR,
2663                                         "Failed to disable promiscuous mode for port %u: %s",
2664                                         port_id, rte_strerror(-ret));
2665                         else
2666                                 slave_ok++;
2667                 }
2668                 /*
2669                  * Report success if operation is successful on at least
2670                  * on one slave. Otherwise return last error code.
2671                  */
2672                 if (slave_ok > 0)
2673                         ret = 0;
2674                 break;
2675         }
2676         /* Promiscuous mode is propagated only to primary slave */
2677         case BONDING_MODE_ACTIVE_BACKUP:
2678         case BONDING_MODE_TLB:
2679         case BONDING_MODE_ALB:
2680         default:
2681                 /* Do not touch promisc when there cannot be primary ports */
2682                 if (internals->slave_count == 0)
2683                         break;
2684                 port_id = internals->current_primary_port;
2685                 ret = rte_eth_promiscuous_disable(port_id);
2686                 if (ret != 0)
2687                         RTE_BOND_LOG(ERR,
2688                                 "Failed to disable promiscuous mode for port %u: %s",
2689                                 port_id, rte_strerror(-ret));
2690         }
2691
2692         return ret;
2693 }
2694
2695 static int
2696 bond_ethdev_allmulticast_enable(struct rte_eth_dev *eth_dev)
2697 {
2698         struct bond_dev_private *internals = eth_dev->data->dev_private;
2699         int i;
2700         int ret = 0;
2701         uint16_t port_id;
2702
2703         switch (internals->mode) {
2704         /* allmulti mode is propagated to all slaves */
2705         case BONDING_MODE_ROUND_ROBIN:
2706         case BONDING_MODE_BALANCE:
2707         case BONDING_MODE_BROADCAST:
2708         case BONDING_MODE_8023AD: {
2709                 unsigned int slave_ok = 0;
2710
2711                 for (i = 0; i < internals->slave_count; i++) {
2712                         port_id = internals->slaves[i].port_id;
2713
2714                         ret = rte_eth_allmulticast_enable(port_id);
2715                         if (ret != 0)
2716                                 RTE_BOND_LOG(ERR,
2717                                         "Failed to enable allmulti mode for port %u: %s",
2718                                         port_id, rte_strerror(-ret));
2719                         else
2720                                 slave_ok++;
2721                 }
2722                 /*
2723                  * Report success if operation is successful on at least
2724                  * on one slave. Otherwise return last error code.
2725                  */
2726                 if (slave_ok > 0)
2727                         ret = 0;
2728                 break;
2729         }
2730         /* allmulti mode is propagated only to primary slave */
2731         case BONDING_MODE_ACTIVE_BACKUP:
2732         case BONDING_MODE_TLB:
2733         case BONDING_MODE_ALB:
2734         default:
2735                 /* Do not touch allmulti when there cannot be primary ports */
2736                 if (internals->slave_count == 0)
2737                         break;
2738                 port_id = internals->current_primary_port;
2739                 ret = rte_eth_allmulticast_enable(port_id);
2740                 if (ret != 0)
2741                         RTE_BOND_LOG(ERR,
2742                                 "Failed to enable allmulti mode for port %u: %s",
2743                                 port_id, rte_strerror(-ret));
2744         }
2745
2746         return ret;
2747 }
2748
2749 static int
2750 bond_ethdev_allmulticast_disable(struct rte_eth_dev *eth_dev)
2751 {
2752         struct bond_dev_private *internals = eth_dev->data->dev_private;
2753         int i;
2754         int ret = 0;
2755         uint16_t port_id;
2756
2757         switch (internals->mode) {
2758         /* allmulti mode is propagated to all slaves */
2759         case BONDING_MODE_ROUND_ROBIN:
2760         case BONDING_MODE_BALANCE:
2761         case BONDING_MODE_BROADCAST:
2762         case BONDING_MODE_8023AD: {
2763                 unsigned int slave_ok = 0;
2764
2765                 for (i = 0; i < internals->slave_count; i++) {
2766                         uint16_t port_id = internals->slaves[i].port_id;
2767
2768                         if (internals->mode == BONDING_MODE_8023AD &&
2769                             bond_mode_8023ad_ports[port_id].forced_rx_flags ==
2770                                         BOND_8023AD_FORCED_ALLMULTI)
2771                                 continue;
2772
2773                         ret = rte_eth_allmulticast_disable(port_id);
2774                         if (ret != 0)
2775                                 RTE_BOND_LOG(ERR,
2776                                         "Failed to disable allmulti mode for port %u: %s",
2777                                         port_id, rte_strerror(-ret));
2778                         else
2779                                 slave_ok++;
2780                 }
2781                 /*
2782                  * Report success if operation is successful on at least
2783                  * on one slave. Otherwise return last error code.
2784                  */
2785                 if (slave_ok > 0)
2786                         ret = 0;
2787                 break;
2788         }
2789         /* allmulti mode is propagated only to primary slave */
2790         case BONDING_MODE_ACTIVE_BACKUP:
2791         case BONDING_MODE_TLB:
2792         case BONDING_MODE_ALB:
2793         default:
2794                 /* Do not touch allmulti when there cannot be primary ports */
2795                 if (internals->slave_count == 0)
2796                         break;
2797                 port_id = internals->current_primary_port;
2798                 ret = rte_eth_allmulticast_disable(port_id);
2799                 if (ret != 0)
2800                         RTE_BOND_LOG(ERR,
2801                                 "Failed to disable allmulti mode for port %u: %s",
2802                                 port_id, rte_strerror(-ret));
2803         }
2804
2805         return ret;
2806 }
2807
2808 static void
2809 bond_ethdev_delayed_lsc_propagation(void *arg)
2810 {
2811         if (arg == NULL)
2812                 return;
2813
2814         rte_eth_dev_callback_process((struct rte_eth_dev *)arg,
2815                         RTE_ETH_EVENT_INTR_LSC, NULL);
2816 }
2817
2818 int
2819 bond_ethdev_lsc_event_callback(uint16_t port_id, enum rte_eth_event_type type,
2820                 void *param, void *ret_param __rte_unused)
2821 {
2822         struct rte_eth_dev *bonded_eth_dev;
2823         struct bond_dev_private *internals;
2824         struct rte_eth_link link;
2825         int rc = -1;
2826         int ret;
2827
2828         uint8_t lsc_flag = 0;
2829         int valid_slave = 0;
2830         uint16_t active_pos;
2831         uint16_t i;
2832
2833         if (type != RTE_ETH_EVENT_INTR_LSC || param == NULL)
2834                 return rc;
2835
2836         bonded_eth_dev = &rte_eth_devices[*(uint16_t *)param];
2837
2838         if (check_for_bonded_ethdev(bonded_eth_dev))
2839                 return rc;
2840
2841         internals = bonded_eth_dev->data->dev_private;
2842
2843         /* If the device isn't started don't handle interrupts */
2844         if (!bonded_eth_dev->data->dev_started)
2845                 return rc;
2846
2847         /* verify that port_id is a valid slave of bonded port */
2848         for (i = 0; i < internals->slave_count; i++) {
2849                 if (internals->slaves[i].port_id == port_id) {
2850                         valid_slave = 1;
2851                         break;
2852                 }
2853         }
2854
2855         if (!valid_slave)
2856                 return rc;
2857
2858         /* Synchronize lsc callback parallel calls either by real link event
2859          * from the slaves PMDs or by the bonding PMD itself.
2860          */
2861         rte_spinlock_lock(&internals->lsc_lock);
2862
2863         /* Search for port in active port list */
2864         active_pos = find_slave_by_id(internals->active_slaves,
2865                         internals->active_slave_count, port_id);
2866
2867         ret = rte_eth_link_get_nowait(port_id, &link);
2868         if (ret < 0)
2869                 RTE_BOND_LOG(ERR, "Slave (port %u) link get failed", port_id);
2870
2871         if (ret == 0 && link.link_status) {
2872                 if (active_pos < internals->active_slave_count)
2873                         goto link_update;
2874
2875                 /* check link state properties if bonded link is up*/
2876                 if (bonded_eth_dev->data->dev_link.link_status == ETH_LINK_UP) {
2877                         if (link_properties_valid(bonded_eth_dev, &link) != 0)
2878                                 RTE_BOND_LOG(ERR, "Invalid link properties "
2879                                              "for slave %d in bonding mode %d",
2880                                              port_id, internals->mode);
2881                 } else {
2882                         /* inherit slave link properties */
2883                         link_properties_set(bonded_eth_dev, &link);
2884                 }
2885
2886                 /* If no active slave ports then set this port to be
2887                  * the primary port.
2888                  */
2889                 if (internals->active_slave_count < 1) {
2890                         /* If first active slave, then change link status */
2891                         bonded_eth_dev->data->dev_link.link_status =
2892                                                                 ETH_LINK_UP;
2893                         internals->current_primary_port = port_id;
2894                         lsc_flag = 1;
2895
2896                         mac_address_slaves_update(bonded_eth_dev);
2897                 }
2898
2899                 activate_slave(bonded_eth_dev, port_id);
2900
2901                 /* If the user has defined the primary port then default to
2902                  * using it.
2903                  */
2904                 if (internals->user_defined_primary_port &&
2905                                 internals->primary_port == port_id)
2906                         bond_ethdev_primary_set(internals, port_id);
2907         } else {
2908                 if (active_pos == internals->active_slave_count)
2909                         goto link_update;
2910
2911                 /* Remove from active slave list */
2912                 deactivate_slave(bonded_eth_dev, port_id);
2913
2914                 if (internals->active_slave_count < 1)
2915                         lsc_flag = 1;
2916
2917                 /* Update primary id, take first active slave from list or if none
2918                  * available set to -1 */
2919                 if (port_id == internals->current_primary_port) {
2920                         if (internals->active_slave_count > 0)
2921                                 bond_ethdev_primary_set(internals,
2922                                                 internals->active_slaves[0]);
2923                         else
2924                                 internals->current_primary_port = internals->primary_port;
2925                         mac_address_slaves_update(bonded_eth_dev);
2926                 }
2927         }
2928
2929 link_update:
2930         /**
2931          * Update bonded device link properties after any change to active
2932          * slaves
2933          */
2934         bond_ethdev_link_update(bonded_eth_dev, 0);
2935
2936         if (lsc_flag) {
2937                 /* Cancel any possible outstanding interrupts if delays are enabled */
2938                 if (internals->link_up_delay_ms > 0 ||
2939                         internals->link_down_delay_ms > 0)
2940                         rte_eal_alarm_cancel(bond_ethdev_delayed_lsc_propagation,
2941                                         bonded_eth_dev);
2942
2943                 if (bonded_eth_dev->data->dev_link.link_status) {
2944                         if (internals->link_up_delay_ms > 0)
2945                                 rte_eal_alarm_set(internals->link_up_delay_ms * 1000,
2946                                                 bond_ethdev_delayed_lsc_propagation,
2947                                                 (void *)bonded_eth_dev);
2948                         else
2949                                 rte_eth_dev_callback_process(bonded_eth_dev,
2950                                                 RTE_ETH_EVENT_INTR_LSC,
2951                                                 NULL);
2952
2953                 } else {
2954                         if (internals->link_down_delay_ms > 0)
2955                                 rte_eal_alarm_set(internals->link_down_delay_ms * 1000,
2956                                                 bond_ethdev_delayed_lsc_propagation,
2957                                                 (void *)bonded_eth_dev);
2958                         else
2959                                 rte_eth_dev_callback_process(bonded_eth_dev,
2960                                                 RTE_ETH_EVENT_INTR_LSC,
2961                                                 NULL);
2962                 }
2963         }
2964
2965         rte_spinlock_unlock(&internals->lsc_lock);
2966
2967         return rc;
2968 }
2969
2970 static int
2971 bond_ethdev_rss_reta_update(struct rte_eth_dev *dev,
2972                 struct rte_eth_rss_reta_entry64 *reta_conf, uint16_t reta_size)
2973 {
2974         unsigned i, j;
2975         int result = 0;
2976         int slave_reta_size;
2977         unsigned reta_count;
2978         struct bond_dev_private *internals = dev->data->dev_private;
2979
2980         if (reta_size != internals->reta_size)
2981                 return -EINVAL;
2982
2983          /* Copy RETA table */
2984         reta_count = (reta_size + RTE_RETA_GROUP_SIZE - 1) /
2985                         RTE_RETA_GROUP_SIZE;
2986
2987         for (i = 0; i < reta_count; i++) {
2988                 internals->reta_conf[i].mask = reta_conf[i].mask;
2989                 for (j = 0; j < RTE_RETA_GROUP_SIZE; j++)
2990                         if ((reta_conf[i].mask >> j) & 0x01)
2991                                 internals->reta_conf[i].reta[j] = reta_conf[i].reta[j];
2992         }
2993
2994         /* Fill rest of array */
2995         for (; i < RTE_DIM(internals->reta_conf); i += reta_count)
2996                 memcpy(&internals->reta_conf[i], &internals->reta_conf[0],
2997                                 sizeof(internals->reta_conf[0]) * reta_count);
2998
2999         /* Propagate RETA over slaves */
3000         for (i = 0; i < internals->slave_count; i++) {
3001                 slave_reta_size = internals->slaves[i].reta_size;
3002                 result = rte_eth_dev_rss_reta_update(internals->slaves[i].port_id,
3003                                 &internals->reta_conf[0], slave_reta_size);
3004                 if (result < 0)
3005                         return result;
3006         }
3007
3008         return 0;
3009 }
3010
3011 static int
3012 bond_ethdev_rss_reta_query(struct rte_eth_dev *dev,
3013                 struct rte_eth_rss_reta_entry64 *reta_conf, uint16_t reta_size)
3014 {
3015         int i, j;
3016         struct bond_dev_private *internals = dev->data->dev_private;
3017
3018         if (reta_size != internals->reta_size)
3019                 return -EINVAL;
3020
3021          /* Copy RETA table */
3022         for (i = 0; i < reta_size / RTE_RETA_GROUP_SIZE; i++)
3023                 for (j = 0; j < RTE_RETA_GROUP_SIZE; j++)
3024                         if ((reta_conf[i].mask >> j) & 0x01)
3025                                 reta_conf[i].reta[j] = internals->reta_conf[i].reta[j];
3026
3027         return 0;
3028 }
3029
3030 static int
3031 bond_ethdev_rss_hash_update(struct rte_eth_dev *dev,
3032                 struct rte_eth_rss_conf *rss_conf)
3033 {
3034         int i, result = 0;
3035         struct bond_dev_private *internals = dev->data->dev_private;
3036         struct rte_eth_rss_conf bond_rss_conf;
3037
3038         memcpy(&bond_rss_conf, rss_conf, sizeof(struct rte_eth_rss_conf));
3039
3040         bond_rss_conf.rss_hf &= internals->flow_type_rss_offloads;
3041
3042         if (bond_rss_conf.rss_hf != 0)
3043                 dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf = bond_rss_conf.rss_hf;
3044
3045         if (bond_rss_conf.rss_key) {
3046                 if (bond_rss_conf.rss_key_len < internals->rss_key_len)
3047                         return -EINVAL;
3048                 else if (bond_rss_conf.rss_key_len > internals->rss_key_len)
3049                         RTE_BOND_LOG(WARNING, "rss_key will be truncated");
3050
3051                 memcpy(internals->rss_key, bond_rss_conf.rss_key,
3052                                 internals->rss_key_len);
3053                 bond_rss_conf.rss_key_len = internals->rss_key_len;
3054         }
3055
3056         for (i = 0; i < internals->slave_count; i++) {
3057                 result = rte_eth_dev_rss_hash_update(internals->slaves[i].port_id,
3058                                 &bond_rss_conf);
3059                 if (result < 0)
3060                         return result;
3061         }
3062
3063         return 0;
3064 }
3065
3066 static int
3067 bond_ethdev_rss_hash_conf_get(struct rte_eth_dev *dev,
3068                 struct rte_eth_rss_conf *rss_conf)
3069 {
3070         struct bond_dev_private *internals = dev->data->dev_private;
3071
3072         rss_conf->rss_hf = dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf;
3073         rss_conf->rss_key_len = internals->rss_key_len;
3074         if (rss_conf->rss_key)
3075                 memcpy(rss_conf->rss_key, internals->rss_key, internals->rss_key_len);
3076
3077         return 0;
3078 }
3079
3080 static int
3081 bond_ethdev_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
3082 {
3083         struct rte_eth_dev *slave_eth_dev;
3084         struct bond_dev_private *internals = dev->data->dev_private;
3085         int ret, i;
3086
3087         rte_spinlock_lock(&internals->lock);
3088
3089         for (i = 0; i < internals->slave_count; i++) {
3090                 slave_eth_dev = &rte_eth_devices[internals->slaves[i].port_id];
3091                 if (*slave_eth_dev->dev_ops->mtu_set == NULL) {
3092                         rte_spinlock_unlock(&internals->lock);
3093                         return -ENOTSUP;
3094                 }
3095         }
3096         for (i = 0; i < internals->slave_count; i++) {
3097                 ret = rte_eth_dev_set_mtu(internals->slaves[i].port_id, mtu);
3098                 if (ret < 0) {
3099                         rte_spinlock_unlock(&internals->lock);
3100                         return ret;
3101                 }
3102         }
3103
3104         rte_spinlock_unlock(&internals->lock);
3105         return 0;
3106 }
3107
3108 static int
3109 bond_ethdev_mac_address_set(struct rte_eth_dev *dev,
3110                         struct rte_ether_addr *addr)
3111 {
3112         if (mac_address_set(dev, addr)) {
3113                 RTE_BOND_LOG(ERR, "Failed to update MAC address");
3114                 return -EINVAL;
3115         }
3116
3117         return 0;
3118 }
3119
3120 static int
3121 bond_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
3122                   const struct rte_flow_ops **ops)
3123 {
3124         *ops = &bond_flow_ops;
3125         return 0;
3126 }
3127
3128 static int
3129 bond_ethdev_mac_addr_add(struct rte_eth_dev *dev,
3130                         struct rte_ether_addr *mac_addr,
3131                         __rte_unused uint32_t index, uint32_t vmdq)
3132 {
3133         struct rte_eth_dev *slave_eth_dev;
3134         struct bond_dev_private *internals = dev->data->dev_private;
3135         int ret, i;
3136
3137         rte_spinlock_lock(&internals->lock);
3138
3139         for (i = 0; i < internals->slave_count; i++) {
3140                 slave_eth_dev = &rte_eth_devices[internals->slaves[i].port_id];
3141                 if (*slave_eth_dev->dev_ops->mac_addr_add == NULL ||
3142                          *slave_eth_dev->dev_ops->mac_addr_remove == NULL) {
3143                         ret = -ENOTSUP;
3144                         goto end;
3145                 }
3146         }
3147
3148         for (i = 0; i < internals->slave_count; i++) {
3149                 ret = rte_eth_dev_mac_addr_add(internals->slaves[i].port_id,
3150                                 mac_addr, vmdq);
3151                 if (ret < 0) {
3152                         /* rollback */
3153                         for (i--; i >= 0; i--)
3154                                 rte_eth_dev_mac_addr_remove(
3155                                         internals->slaves[i].port_id, mac_addr);
3156                         goto end;
3157                 }
3158         }
3159
3160         ret = 0;
3161 end:
3162         rte_spinlock_unlock(&internals->lock);
3163         return ret;
3164 }
3165
3166 static void
3167 bond_ethdev_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
3168 {
3169         struct rte_eth_dev *slave_eth_dev;
3170         struct bond_dev_private *internals = dev->data->dev_private;
3171         int i;
3172
3173         rte_spinlock_lock(&internals->lock);
3174
3175         for (i = 0; i < internals->slave_count; i++) {
3176                 slave_eth_dev = &rte_eth_devices[internals->slaves[i].port_id];
3177                 if (*slave_eth_dev->dev_ops->mac_addr_remove == NULL)
3178                         goto end;
3179         }
3180
3181         struct rte_ether_addr *mac_addr = &dev->data->mac_addrs[index];
3182
3183         for (i = 0; i < internals->slave_count; i++)
3184                 rte_eth_dev_mac_addr_remove(internals->slaves[i].port_id,
3185                                 mac_addr);
3186
3187 end:
3188         rte_spinlock_unlock(&internals->lock);
3189 }
3190
3191 const struct eth_dev_ops default_dev_ops = {
3192         .dev_start            = bond_ethdev_start,
3193         .dev_stop             = bond_ethdev_stop,
3194         .dev_close            = bond_ethdev_close,
3195         .dev_configure        = bond_ethdev_configure,
3196         .dev_infos_get        = bond_ethdev_info,
3197         .vlan_filter_set      = bond_ethdev_vlan_filter_set,
3198         .rx_queue_setup       = bond_ethdev_rx_queue_setup,
3199         .tx_queue_setup       = bond_ethdev_tx_queue_setup,
3200         .rx_queue_release     = bond_ethdev_rx_queue_release,
3201         .tx_queue_release     = bond_ethdev_tx_queue_release,
3202         .link_update          = bond_ethdev_link_update,
3203         .stats_get            = bond_ethdev_stats_get,
3204         .stats_reset          = bond_ethdev_stats_reset,
3205         .promiscuous_enable   = bond_ethdev_promiscuous_enable,
3206         .promiscuous_disable  = bond_ethdev_promiscuous_disable,
3207         .allmulticast_enable  = bond_ethdev_allmulticast_enable,
3208         .allmulticast_disable = bond_ethdev_allmulticast_disable,
3209         .reta_update          = bond_ethdev_rss_reta_update,
3210         .reta_query           = bond_ethdev_rss_reta_query,
3211         .rss_hash_update      = bond_ethdev_rss_hash_update,
3212         .rss_hash_conf_get    = bond_ethdev_rss_hash_conf_get,
3213         .mtu_set              = bond_ethdev_mtu_set,
3214         .mac_addr_set         = bond_ethdev_mac_address_set,
3215         .mac_addr_add         = bond_ethdev_mac_addr_add,
3216         .mac_addr_remove      = bond_ethdev_mac_addr_remove,
3217         .flow_ops_get         = bond_flow_ops_get
3218 };
3219
3220 static int
3221 bond_alloc(struct rte_vdev_device *dev, uint8_t mode)
3222 {
3223         const char *name = rte_vdev_device_name(dev);
3224         uint8_t socket_id = dev->device.numa_node;
3225         struct bond_dev_private *internals = NULL;
3226         struct rte_eth_dev *eth_dev = NULL;
3227         uint32_t vlan_filter_bmp_size;
3228
3229         /* now do all data allocation - for eth_dev structure, dummy pci driver
3230          * and internal (private) data
3231          */
3232
3233         /* reserve an ethdev entry */
3234         eth_dev = rte_eth_vdev_allocate(dev, sizeof(*internals));
3235         if (eth_dev == NULL) {
3236                 RTE_BOND_LOG(ERR, "Unable to allocate rte_eth_dev");
3237                 goto err;
3238         }
3239
3240         internals = eth_dev->data->dev_private;
3241         eth_dev->data->nb_rx_queues = (uint16_t)1;
3242         eth_dev->data->nb_tx_queues = (uint16_t)1;
3243
3244         /* Allocate memory for storing MAC addresses */
3245         eth_dev->data->mac_addrs = rte_zmalloc_socket(name, RTE_ETHER_ADDR_LEN *
3246                         BOND_MAX_MAC_ADDRS, 0, socket_id);
3247         if (eth_dev->data->mac_addrs == NULL) {
3248                 RTE_BOND_LOG(ERR,
3249                              "Failed to allocate %u bytes needed to store MAC addresses",
3250                              RTE_ETHER_ADDR_LEN * BOND_MAX_MAC_ADDRS);
3251                 goto err;
3252         }
3253
3254         eth_dev->dev_ops = &default_dev_ops;
3255         eth_dev->data->dev_flags = RTE_ETH_DEV_INTR_LSC |
3256                                         RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
3257
3258         rte_spinlock_init(&internals->lock);
3259         rte_spinlock_init(&internals->lsc_lock);
3260
3261         internals->port_id = eth_dev->data->port_id;
3262         internals->mode = BONDING_MODE_INVALID;
3263         internals->current_primary_port = RTE_MAX_ETHPORTS + 1;
3264         internals->balance_xmit_policy = BALANCE_XMIT_POLICY_LAYER2;
3265         internals->burst_xmit_hash = burst_xmit_l2_hash;
3266         internals->user_defined_mac = 0;
3267
3268         internals->link_status_polling_enabled = 0;
3269
3270         internals->link_status_polling_interval_ms =
3271                 DEFAULT_POLLING_INTERVAL_10_MS;
3272         internals->link_down_delay_ms = 0;
3273         internals->link_up_delay_ms = 0;
3274
3275         internals->slave_count = 0;
3276         internals->active_slave_count = 0;
3277         internals->rx_offload_capa = 0;
3278         internals->tx_offload_capa = 0;
3279         internals->rx_queue_offload_capa = 0;
3280         internals->tx_queue_offload_capa = 0;
3281         internals->candidate_max_rx_pktlen = 0;
3282         internals->max_rx_pktlen = 0;
3283
3284         /* Initially allow to choose any offload type */
3285         internals->flow_type_rss_offloads = ETH_RSS_PROTO_MASK;
3286
3287         memset(&internals->default_rxconf, 0,
3288                sizeof(internals->default_rxconf));
3289         memset(&internals->default_txconf, 0,
3290                sizeof(internals->default_txconf));
3291
3292         memset(&internals->rx_desc_lim, 0, sizeof(internals->rx_desc_lim));
3293         memset(&internals->tx_desc_lim, 0, sizeof(internals->tx_desc_lim));
3294
3295         memset(internals->active_slaves, 0, sizeof(internals->active_slaves));
3296         memset(internals->slaves, 0, sizeof(internals->slaves));
3297
3298         TAILQ_INIT(&internals->flow_list);
3299         internals->flow_isolated_valid = 0;
3300
3301         /* Set mode 4 default configuration */
3302         bond_mode_8023ad_setup(eth_dev, NULL);
3303         if (bond_ethdev_mode_set(eth_dev, mode)) {
3304                 RTE_BOND_LOG(ERR, "Failed to set bonded device %d mode to %d",
3305                                  eth_dev->data->port_id, mode);
3306                 goto err;
3307         }
3308
3309         vlan_filter_bmp_size =
3310                 rte_bitmap_get_memory_footprint(RTE_ETHER_MAX_VLAN_ID + 1);
3311         internals->vlan_filter_bmpmem = rte_malloc(name, vlan_filter_bmp_size,
3312                                                    RTE_CACHE_LINE_SIZE);
3313         if (internals->vlan_filter_bmpmem == NULL) {
3314                 RTE_BOND_LOG(ERR,
3315                              "Failed to allocate vlan bitmap for bonded device %u",
3316                              eth_dev->data->port_id);
3317                 goto err;
3318         }
3319
3320         internals->vlan_filter_bmp = rte_bitmap_init(RTE_ETHER_MAX_VLAN_ID + 1,
3321                         internals->vlan_filter_bmpmem, vlan_filter_bmp_size);
3322         if (internals->vlan_filter_bmp == NULL) {
3323                 RTE_BOND_LOG(ERR,
3324                              "Failed to init vlan bitmap for bonded device %u",
3325                              eth_dev->data->port_id);
3326                 rte_free(internals->vlan_filter_bmpmem);
3327                 goto err;
3328         }
3329
3330         return eth_dev->data->port_id;
3331
3332 err:
3333         rte_free(internals);
3334         if (eth_dev != NULL)
3335                 eth_dev->data->dev_private = NULL;
3336         rte_eth_dev_release_port(eth_dev);
3337         return -1;
3338 }
3339
3340 static int
3341 bond_probe(struct rte_vdev_device *dev)
3342 {
3343         const char *name;
3344         struct bond_dev_private *internals;
3345         struct rte_kvargs *kvlist;
3346         uint8_t bonding_mode;
3347         int arg_count, port_id;
3348         int socket_id;
3349         uint8_t agg_mode;
3350         struct rte_eth_dev *eth_dev;
3351
3352         if (!dev)
3353                 return -EINVAL;
3354
3355         name = rte_vdev_device_name(dev);
3356         RTE_BOND_LOG(INFO, "Initializing pmd_bond for %s", name);
3357
3358         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
3359                 eth_dev = rte_eth_dev_attach_secondary(name);
3360                 if (!eth_dev) {
3361                         RTE_BOND_LOG(ERR, "Failed to probe %s", name);
3362                         return -1;
3363                 }
3364                 /* TODO: request info from primary to set up Rx and Tx */
3365                 eth_dev->dev_ops = &default_dev_ops;
3366                 eth_dev->device = &dev->device;
3367                 rte_eth_dev_probing_finish(eth_dev);
3368                 return 0;
3369         }
3370
3371         kvlist = rte_kvargs_parse(rte_vdev_device_args(dev),
3372                 pmd_bond_init_valid_arguments);
3373         if (kvlist == NULL)
3374                 return -1;
3375
3376         /* Parse link bonding mode */
3377         if (rte_kvargs_count(kvlist, PMD_BOND_MODE_KVARG) == 1) {
3378                 if (rte_kvargs_process(kvlist, PMD_BOND_MODE_KVARG,
3379                                 &bond_ethdev_parse_slave_mode_kvarg,
3380                                 &bonding_mode) != 0) {
3381                         RTE_BOND_LOG(ERR, "Invalid mode for bonded device %s",
3382                                         name);
3383                         goto parse_error;
3384                 }
3385         } else {
3386                 RTE_BOND_LOG(ERR, "Mode must be specified only once for bonded "
3387                                 "device %s", name);
3388                 goto parse_error;
3389         }
3390
3391         /* Parse socket id to create bonding device on */
3392         arg_count = rte_kvargs_count(kvlist, PMD_BOND_SOCKET_ID_KVARG);
3393         if (arg_count == 1) {
3394                 if (rte_kvargs_process(kvlist, PMD_BOND_SOCKET_ID_KVARG,
3395                                 &bond_ethdev_parse_socket_id_kvarg, &socket_id)
3396                                 != 0) {
3397                         RTE_BOND_LOG(ERR, "Invalid socket Id specified for "
3398                                         "bonded device %s", name);
3399                         goto parse_error;
3400                 }
3401         } else if (arg_count > 1) {
3402                 RTE_BOND_LOG(ERR, "Socket Id can be specified only once for "
3403                                 "bonded device %s", name);
3404                 goto parse_error;
3405         } else {
3406                 socket_id = rte_socket_id();
3407         }
3408
3409         dev->device.numa_node = socket_id;
3410
3411         /* Create link bonding eth device */
3412         port_id = bond_alloc(dev, bonding_mode);
3413         if (port_id < 0) {
3414                 RTE_BOND_LOG(ERR, "Failed to create socket %s in mode %u on "
3415                                 "socket %u.",   name, bonding_mode, socket_id);
3416                 goto parse_error;
3417         }
3418         internals = rte_eth_devices[port_id].data->dev_private;
3419         internals->kvlist = kvlist;
3420
3421         if (rte_kvargs_count(kvlist, PMD_BOND_AGG_MODE_KVARG) == 1) {
3422                 if (rte_kvargs_process(kvlist,
3423                                 PMD_BOND_AGG_MODE_KVARG,
3424                                 &bond_ethdev_parse_slave_agg_mode_kvarg,
3425                                 &agg_mode) != 0) {
3426                         RTE_BOND_LOG(ERR,
3427                                         "Failed to parse agg selection mode for bonded device %s",
3428                                         name);
3429                         goto parse_error;
3430                 }
3431
3432                 if (internals->mode == BONDING_MODE_8023AD)
3433                         internals->mode4.agg_selection = agg_mode;
3434         } else {
3435                 internals->mode4.agg_selection = AGG_STABLE;
3436         }
3437
3438         rte_eth_dev_probing_finish(&rte_eth_devices[port_id]);
3439         RTE_BOND_LOG(INFO, "Create bonded device %s on port %d in mode %u on "
3440                         "socket %u.",   name, port_id, bonding_mode, socket_id);
3441         return 0;
3442
3443 parse_error:
3444         rte_kvargs_free(kvlist);
3445
3446         return -1;
3447 }
3448
3449 static int
3450 bond_remove(struct rte_vdev_device *dev)
3451 {
3452         struct rte_eth_dev *eth_dev;
3453         struct bond_dev_private *internals;
3454         const char *name;
3455         int ret = 0;
3456
3457         if (!dev)
3458                 return -EINVAL;
3459
3460         name = rte_vdev_device_name(dev);
3461         RTE_BOND_LOG(INFO, "Uninitializing pmd_bond for %s", name);
3462
3463         /* find an ethdev entry */
3464         eth_dev = rte_eth_dev_allocated(name);
3465         if (eth_dev == NULL)
3466                 return 0; /* port already released */
3467
3468         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
3469                 return rte_eth_dev_release_port(eth_dev);
3470
3471         RTE_ASSERT(eth_dev->device == &dev->device);
3472
3473         internals = eth_dev->data->dev_private;
3474         if (internals->slave_count != 0)
3475                 return -EBUSY;
3476
3477         if (eth_dev->data->dev_started == 1) {
3478                 ret = bond_ethdev_stop(eth_dev);
3479                 bond_ethdev_close(eth_dev);
3480         }
3481         rte_eth_dev_release_port(eth_dev);
3482
3483         return ret;
3484 }
3485
3486 /* this part will resolve the slave portids after all the other pdev and vdev
3487  * have been allocated */
3488 static int
3489 bond_ethdev_configure(struct rte_eth_dev *dev)
3490 {
3491         const char *name = dev->device->name;
3492         struct bond_dev_private *internals = dev->data->dev_private;
3493         struct rte_kvargs *kvlist = internals->kvlist;
3494         int arg_count;
3495         uint16_t port_id = dev - rte_eth_devices;
3496         uint8_t agg_mode;
3497
3498         static const uint8_t default_rss_key[40] = {
3499                 0x6D, 0x5A, 0x56, 0xDA, 0x25, 0x5B, 0x0E, 0xC2, 0x41, 0x67, 0x25, 0x3D,
3500                 0x43, 0xA3, 0x8F, 0xB0, 0xD0, 0xCA, 0x2B, 0xCB, 0xAE, 0x7B, 0x30, 0xB4,
3501                 0x77, 0xCB, 0x2D, 0xA3, 0x80, 0x30, 0xF2, 0x0C, 0x6A, 0x42, 0xB7, 0x3B,
3502                 0xBE, 0xAC, 0x01, 0xFA
3503         };
3504
3505         unsigned i, j;
3506
3507         /*
3508          * If RSS is enabled, fill table with default values and
3509          * set key to the the value specified in port RSS configuration.
3510          * Fall back to default RSS key if the key is not specified
3511          */
3512         if (dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS) {
3513                 struct rte_eth_rss_conf *rss_conf =
3514                         &dev->data->dev_conf.rx_adv_conf.rss_conf;
3515                 if (rss_conf->rss_key != NULL) {
3516                         if (internals->rss_key_len > rss_conf->rss_key_len) {
3517                                 RTE_BOND_LOG(ERR, "Invalid rss key length(%u)",
3518                                                 rss_conf->rss_key_len);
3519                                 return -EINVAL;
3520                         }
3521
3522                         memcpy(internals->rss_key, rss_conf->rss_key,
3523                                internals->rss_key_len);
3524                 } else {
3525                         if (internals->rss_key_len > sizeof(default_rss_key)) {
3526                                 RTE_BOND_LOG(ERR,
3527                                        "There is no suitable default hash key");
3528                                 return -EINVAL;
3529                         }
3530
3531                         memcpy(internals->rss_key, default_rss_key,
3532                                internals->rss_key_len);
3533                 }
3534
3535                 for (i = 0; i < RTE_DIM(internals->reta_conf); i++) {
3536                         internals->reta_conf[i].mask = ~0LL;
3537                         for (j = 0; j < RTE_RETA_GROUP_SIZE; j++)
3538                                 internals->reta_conf[i].reta[j] =
3539                                                 (i * RTE_RETA_GROUP_SIZE + j) %
3540                                                 dev->data->nb_rx_queues;
3541                 }
3542         }
3543
3544         /* set the max_rx_pktlen */
3545         internals->max_rx_pktlen = internals->candidate_max_rx_pktlen;
3546
3547         /*
3548          * if no kvlist, it means that this bonded device has been created
3549          * through the bonding api.
3550          */
3551         if (!kvlist)
3552                 return 0;
3553
3554         /* Parse MAC address for bonded device */
3555         arg_count = rte_kvargs_count(kvlist, PMD_BOND_MAC_ADDR_KVARG);
3556         if (arg_count == 1) {
3557                 struct rte_ether_addr bond_mac;
3558
3559                 if (rte_kvargs_process(kvlist, PMD_BOND_MAC_ADDR_KVARG,
3560                                        &bond_ethdev_parse_bond_mac_addr_kvarg, &bond_mac) < 0) {
3561                         RTE_BOND_LOG(INFO, "Invalid mac address for bonded device %s",
3562                                      name);
3563                         return -1;
3564                 }
3565
3566                 /* Set MAC address */
3567                 if (rte_eth_bond_mac_address_set(port_id, &bond_mac) != 0) {
3568                         RTE_BOND_LOG(ERR,
3569                                      "Failed to set mac address on bonded device %s",
3570                                      name);
3571                         return -1;
3572                 }
3573         } else if (arg_count > 1) {
3574                 RTE_BOND_LOG(ERR,
3575                              "MAC address can be specified only once for bonded device %s",
3576                              name);
3577                 return -1;
3578         }
3579
3580         /* Parse/set balance mode transmit policy */
3581         arg_count = rte_kvargs_count(kvlist, PMD_BOND_XMIT_POLICY_KVARG);
3582         if (arg_count == 1) {
3583                 uint8_t xmit_policy;
3584
3585                 if (rte_kvargs_process(kvlist, PMD_BOND_XMIT_POLICY_KVARG,
3586                                        &bond_ethdev_parse_balance_xmit_policy_kvarg, &xmit_policy) !=
3587                     0) {
3588                         RTE_BOND_LOG(INFO,
3589                                      "Invalid xmit policy specified for bonded device %s",
3590                                      name);
3591                         return -1;
3592                 }
3593
3594                 /* Set balance mode transmit policy*/
3595                 if (rte_eth_bond_xmit_policy_set(port_id, xmit_policy) != 0) {
3596                         RTE_BOND_LOG(ERR,
3597                                      "Failed to set balance xmit policy on bonded device %s",
3598                                      name);
3599                         return -1;
3600                 }
3601         } else if (arg_count > 1) {
3602                 RTE_BOND_LOG(ERR,
3603                              "Transmit policy can be specified only once for bonded device %s",
3604                              name);
3605                 return -1;
3606         }
3607
3608         if (rte_kvargs_count(kvlist, PMD_BOND_AGG_MODE_KVARG) == 1) {
3609                 if (rte_kvargs_process(kvlist,
3610                                        PMD_BOND_AGG_MODE_KVARG,
3611                                        &bond_ethdev_parse_slave_agg_mode_kvarg,
3612                                        &agg_mode) != 0) {
3613                         RTE_BOND_LOG(ERR,
3614                                      "Failed to parse agg selection mode for bonded device %s",
3615                                      name);
3616                 }
3617                 if (internals->mode == BONDING_MODE_8023AD) {
3618                         int ret = rte_eth_bond_8023ad_agg_selection_set(port_id,
3619                                         agg_mode);
3620                         if (ret < 0) {
3621                                 RTE_BOND_LOG(ERR,
3622                                         "Invalid args for agg selection set for bonded device %s",
3623                                         name);
3624                                 return -1;
3625                         }
3626                 }
3627         }
3628
3629         /* Parse/add slave ports to bonded device */
3630         if (rte_kvargs_count(kvlist, PMD_BOND_SLAVE_PORT_KVARG) > 0) {
3631                 struct bond_ethdev_slave_ports slave_ports;
3632                 unsigned i;
3633
3634                 memset(&slave_ports, 0, sizeof(slave_ports));
3635
3636                 if (rte_kvargs_process(kvlist, PMD_BOND_SLAVE_PORT_KVARG,
3637                                        &bond_ethdev_parse_slave_port_kvarg, &slave_ports) != 0) {
3638                         RTE_BOND_LOG(ERR,
3639                                      "Failed to parse slave ports for bonded device %s",
3640                                      name);
3641                         return -1;
3642                 }
3643
3644                 for (i = 0; i < slave_ports.slave_count; i++) {
3645                         if (rte_eth_bond_slave_add(port_id, slave_ports.slaves[i]) != 0) {
3646                                 RTE_BOND_LOG(ERR,
3647                                              "Failed to add port %d as slave to bonded device %s",
3648                                              slave_ports.slaves[i], name);
3649                         }
3650                 }
3651
3652         } else {
3653                 RTE_BOND_LOG(INFO, "No slaves specified for bonded device %s", name);
3654                 return -1;
3655         }
3656
3657         /* Parse/set primary slave port id*/
3658         arg_count = rte_kvargs_count(kvlist, PMD_BOND_PRIMARY_SLAVE_KVARG);
3659         if (arg_count == 1) {
3660                 uint16_t primary_slave_port_id;
3661
3662                 if (rte_kvargs_process(kvlist,
3663                                        PMD_BOND_PRIMARY_SLAVE_KVARG,
3664                                        &bond_ethdev_parse_primary_slave_port_id_kvarg,
3665                                        &primary_slave_port_id) < 0) {
3666                         RTE_BOND_LOG(INFO,
3667                                      "Invalid primary slave port id specified for bonded device %s",
3668                                      name);
3669                         return -1;
3670                 }
3671
3672                 /* Set balance mode transmit policy*/
3673                 if (rte_eth_bond_primary_set(port_id, primary_slave_port_id)
3674                     != 0) {
3675                         RTE_BOND_LOG(ERR,
3676                                      "Failed to set primary slave port %d on bonded device %s",
3677                                      primary_slave_port_id, name);
3678                         return -1;
3679                 }
3680         } else if (arg_count > 1) {
3681                 RTE_BOND_LOG(INFO,
3682                              "Primary slave can be specified only once for bonded device %s",
3683                              name);
3684                 return -1;
3685         }
3686
3687         /* Parse link status monitor polling interval */
3688         arg_count = rte_kvargs_count(kvlist, PMD_BOND_LSC_POLL_PERIOD_KVARG);
3689         if (arg_count == 1) {
3690                 uint32_t lsc_poll_interval_ms;
3691
3692                 if (rte_kvargs_process(kvlist,
3693                                        PMD_BOND_LSC_POLL_PERIOD_KVARG,
3694                                        &bond_ethdev_parse_time_ms_kvarg,
3695                                        &lsc_poll_interval_ms) < 0) {
3696                         RTE_BOND_LOG(INFO,
3697                                      "Invalid lsc polling interval value specified for bonded"
3698                                      " device %s", name);
3699                         return -1;
3700                 }
3701
3702                 if (rte_eth_bond_link_monitoring_set(port_id, lsc_poll_interval_ms)
3703                     != 0) {
3704                         RTE_BOND_LOG(ERR,
3705                                      "Failed to set lsc monitor polling interval (%u ms) on bonded device %s",
3706                                      lsc_poll_interval_ms, name);
3707                         return -1;
3708                 }
3709         } else if (arg_count > 1) {
3710                 RTE_BOND_LOG(INFO,
3711                              "LSC polling interval can be specified only once for bonded"
3712                              " device %s", name);
3713                 return -1;
3714         }
3715
3716         /* Parse link up interrupt propagation delay */
3717         arg_count = rte_kvargs_count(kvlist, PMD_BOND_LINK_UP_PROP_DELAY_KVARG);
3718         if (arg_count == 1) {
3719                 uint32_t link_up_delay_ms;
3720
3721                 if (rte_kvargs_process(kvlist,
3722                                        PMD_BOND_LINK_UP_PROP_DELAY_KVARG,
3723                                        &bond_ethdev_parse_time_ms_kvarg,
3724                                        &link_up_delay_ms) < 0) {
3725                         RTE_BOND_LOG(INFO,
3726                                      "Invalid link up propagation delay value specified for"
3727                                      " bonded device %s", name);
3728                         return -1;
3729                 }
3730
3731                 /* Set balance mode transmit policy*/
3732                 if (rte_eth_bond_link_up_prop_delay_set(port_id, link_up_delay_ms)
3733                     != 0) {
3734                         RTE_BOND_LOG(ERR,
3735                                      "Failed to set link up propagation delay (%u ms) on bonded"
3736                                      " device %s", link_up_delay_ms, name);
3737                         return -1;
3738                 }
3739         } else if (arg_count > 1) {
3740                 RTE_BOND_LOG(INFO,
3741                              "Link up propagation delay can be specified only once for"
3742                              " bonded device %s", name);
3743                 return -1;
3744         }
3745
3746         /* Parse link down interrupt propagation delay */
3747         arg_count = rte_kvargs_count(kvlist, PMD_BOND_LINK_DOWN_PROP_DELAY_KVARG);
3748         if (arg_count == 1) {
3749                 uint32_t link_down_delay_ms;
3750
3751                 if (rte_kvargs_process(kvlist,
3752                                        PMD_BOND_LINK_DOWN_PROP_DELAY_KVARG,
3753                                        &bond_ethdev_parse_time_ms_kvarg,
3754                                        &link_down_delay_ms) < 0) {
3755                         RTE_BOND_LOG(INFO,
3756                                      "Invalid link down propagation delay value specified for"
3757                                      " bonded device %s", name);
3758                         return -1;
3759                 }
3760
3761                 /* Set balance mode transmit policy*/
3762                 if (rte_eth_bond_link_down_prop_delay_set(port_id, link_down_delay_ms)
3763                     != 0) {
3764                         RTE_BOND_LOG(ERR,
3765                                      "Failed to set link down propagation delay (%u ms) on bonded device %s",
3766                                      link_down_delay_ms, name);
3767                         return -1;
3768                 }
3769         } else if (arg_count > 1) {
3770                 RTE_BOND_LOG(INFO,
3771                              "Link down propagation delay can be specified only once for  bonded device %s",
3772                              name);
3773                 return -1;
3774         }
3775
3776         return 0;
3777 }
3778
3779 struct rte_vdev_driver pmd_bond_drv = {
3780         .probe = bond_probe,
3781         .remove = bond_remove,
3782 };
3783
3784 RTE_PMD_REGISTER_VDEV(net_bonding, pmd_bond_drv);
3785 RTE_PMD_REGISTER_ALIAS(net_bonding, eth_bond);
3786
3787 RTE_PMD_REGISTER_PARAM_STRING(net_bonding,
3788         "slave=<ifc> "
3789         "primary=<ifc> "
3790         "mode=[0-6] "
3791         "xmit_policy=[l2 | l23 | l34] "
3792         "agg_mode=[count | stable | bandwidth] "
3793         "socket_id=<int> "
3794         "mac=<mac addr> "
3795         "lsc_poll_period_ms=<int> "
3796         "up_delay=<int> "
3797         "down_delay=<int>");
3798
3799 /* We can't use RTE_LOG_REGISTER_DEFAULT because of the forced name for
3800  * this library, see meson.build.
3801  */
3802 RTE_LOG_REGISTER(bond_logtype, pmd.net.bonding, NOTICE);