drivers/net/bonding/rte_eth_bond_pmd.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
   5  *   All rights reserved.
   6  *
   7  *   Redistribution and use in source and binary forms, with or without
   8  *   modification, are permitted provided that the following conditions
   9  *   are met:
  10  *
  11  *     * Redistributions of source code must retain the above copyright
  12  *       notice, this list of conditions and the following disclaimer.
  13  *     * Redistributions in binary form must reproduce the above copyright
  14  *       notice, this list of conditions and the following disclaimer in
  15  *       the documentation and/or other materials provided with the
  16  *       distribution.
  17  *     * Neither the name of Intel Corporation nor the names of its
  18  *       contributors may be used to endorse or promote products derived
  19  *       from this software without specific prior written permission.
  20  *
  21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  */
  33 #include <stdlib.h>
  34 #include <netinet/in.h>
  35
  36 #include <rte_mbuf.h>
  37 #include <rte_malloc.h>
  38 #include <rte_ethdev.h>
  39 #include <rte_tcp.h>
  40 #include <rte_udp.h>
  41 #include <rte_ip.h>
  42 #include <rte_devargs.h>
  43 #include <rte_kvargs.h>
  44 #include <rte_dev.h>
  45 #include <rte_alarm.h>
  46 #include <rte_cycles.h>
  47
  48 #include "rte_eth_bond.h"
  49 #include "rte_eth_bond_private.h"
  50 #include "rte_eth_bond_8023ad_private.h"
  51
  52 #define REORDER_PERIOD_MS 10
  53
  54 #define HASH_L4_PORTS(h) ((h)->src_port ^ (h)->dst_port)
  55
  56 /* Table for statistics in mode 5 TLB */
  57 static uint64_t tlb_last_obytets[RTE_MAX_ETHPORTS];
  58
  59 static inline size_t
  60 get_vlan_offset(struct ether_hdr *eth_hdr, uint16_t *proto)
  61 {
  62         size_t vlan_offset = 0;
  63
  64         if (rte_cpu_to_be_16(ETHER_TYPE_VLAN) == *proto) {
  65                 struct vlan_hdr *vlan_hdr = (struct vlan_hdr *)(eth_hdr + 1);
  66
  67                 vlan_offset = sizeof(struct vlan_hdr);
  68                 *proto = vlan_hdr->eth_proto;
  69
  70                 if (rte_cpu_to_be_16(ETHER_TYPE_VLAN) == *proto) {
  71                         vlan_hdr = vlan_hdr + 1;
  72                         *proto = vlan_hdr->eth_proto;
  73                         vlan_offset += sizeof(struct vlan_hdr);
  74                 }
  75         }
  76         return vlan_offset;
  77 }
  78
  79 static uint16_t
  80 bond_ethdev_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
  81 {
  82         struct bond_dev_private *internals;
  83
  84         uint16_t num_rx_slave = 0;
  85         uint16_t num_rx_total = 0;
  86
  87         int i;
  88
  89         /* Cast to structure, containing bonded device's port id and queue id */
  90         struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)queue;
  91
  92         internals = bd_rx_q->dev_private;
  93
  94
  95         for (i = 0; i < internals->active_slave_count && nb_pkts; i++) {
  96                 /* Offset of pointer to *bufs increases as packets are received
  97                  * from other slaves */
  98                 num_rx_slave = rte_eth_rx_burst(internals->active_slaves[i],
  99                                 bd_rx_q->queue_id, bufs + num_rx_total, nb_pkts);
 100                 if (num_rx_slave) {
 101                         num_rx_total += num_rx_slave;
 102                         nb_pkts -= num_rx_slave;
 103                 }
 104         }
 105
 106         return num_rx_total;
 107 }
 108
 109 static uint16_t
 110 bond_ethdev_rx_burst_active_backup(void *queue, struct rte_mbuf **bufs,
 111                 uint16_t nb_pkts)
 112 {
 113         struct bond_dev_private *internals;
 114
 115         /* Cast to structure, containing bonded device's port id and queue id */
 116         struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)queue;
 117
 118         internals = bd_rx_q->dev_private;
 119
 120         return rte_eth_rx_burst(internals->current_primary_port,
 121                         bd_rx_q->queue_id, bufs, nb_pkts);
 122 }
 123
 124 static uint16_t
 125 bond_ethdev_rx_burst_8023ad(void *queue, struct rte_mbuf **bufs,
 126                 uint16_t nb_pkts)
 127 {
 128         /* Cast to structure, containing bonded device's port id and queue id */
 129         struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)queue;
 130         struct bond_dev_private *internals = bd_rx_q->dev_private;
 131         struct ether_addr bond_mac;
 132
 133         struct ether_hdr *hdr;
 134
 135         const uint16_t ether_type_slow_be = rte_be_to_cpu_16(ETHER_TYPE_SLOW);
 136         uint16_t num_rx_total = 0;      /* Total number of received packets */
 137         uint8_t slaves[RTE_MAX_ETHPORTS];
 138         uint8_t slave_count;
 139
 140         uint8_t collecting;  /* current slave collecting status */
 141         const uint8_t promisc = internals->promiscuous_en;
 142         uint8_t i, j, k;
 143
 144         rte_eth_macaddr_get(internals->port_id, &bond_mac);
 145         /* Copy slave list to protect against slave up/down changes during tx
 146          * bursting */
 147         slave_count = internals->active_slave_count;
 148         memcpy(slaves, internals->active_slaves,
 149                         sizeof(internals->active_slaves[0]) * slave_count);
 150
 151         for (i = 0; i < slave_count && num_rx_total < nb_pkts; i++) {
 152                 j = num_rx_total;
 153                 collecting = ACTOR_STATE(&mode_8023ad_ports[slaves[i]], COLLECTING);
 154
 155                 /* Read packets from this slave */
 156                 num_rx_total += rte_eth_rx_burst(slaves[i], bd_rx_q->queue_id,
 157                                 &bufs[num_rx_total], nb_pkts - num_rx_total);
 158
 159                 for (k = j; k < 2 && k < num_rx_total; k++)
 160                         rte_prefetch0(rte_pktmbuf_mtod(bufs[k], void *));
 161
 162                 /* Handle slow protocol packets. */
 163                 while (j < num_rx_total) {
 164                         if (j + 3 < num_rx_total)
 165                                 rte_prefetch0(rte_pktmbuf_mtod(bufs[j + 3], void *));
 166
 167                         hdr = rte_pktmbuf_mtod(bufs[j], struct ether_hdr *);
 168                         /* Remove packet from array if it is slow packet or slave is not
 169                          * in collecting state or bondign interface is not in promiscus
 170                          * mode and packet address does not match. */
 171                         if (unlikely(hdr->ether_type == ether_type_slow_be ||
 172                                 !collecting || (!promisc &&
 173                                         !is_same_ether_addr(&bond_mac, &hdr->d_addr)))) {
 174
 175                                 if (hdr->ether_type == ether_type_slow_be) {
 176                                         bond_mode_8023ad_handle_slow_pkt(internals, slaves[i],
 177                                                 bufs[j]);
 178                                 } else
 179                                         rte_pktmbuf_free(bufs[j]);
 180
 181                                 /* Packet is managed by mode 4 or dropped, shift the array */
 182                                 num_rx_total--;
 183                                 if (j < num_rx_total) {
 184                                         memmove(&bufs[j], &bufs[j + 1], sizeof(bufs[0]) *
 185                                                 (num_rx_total - j));
 186                                 }
 187                         } else
 188                                 j++;
 189                 }
 190         }
 191
 192         return num_rx_total;
 193 }
 194
 195 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
 196 uint32_t burstnumberRX;
 197 uint32_t burstnumberTX;
 198
 199 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
 200
 201 static void
 202 arp_op_name(uint16_t arp_op, char *buf)
 203 {
 204         switch (arp_op) {
 205         case ARP_OP_REQUEST:
 206                 snprintf(buf, sizeof("ARP Request"), "%s", "ARP Request");
 207                 return;
 208         case ARP_OP_REPLY:
 209                 snprintf(buf, sizeof("ARP Reply"), "%s", "ARP Reply");
 210                 return;
 211         case ARP_OP_REVREQUEST:
 212                 snprintf(buf, sizeof("Reverse ARP Request"), "%s",
 213                                 "Reverse ARP Request");
 214                 return;
 215         case ARP_OP_REVREPLY:
 216                 snprintf(buf, sizeof("Reverse ARP Reply"), "%s",
 217                                 "Reverse ARP Reply");
 218                 return;
 219         case ARP_OP_INVREQUEST:
 220                 snprintf(buf, sizeof("Peer Identify Request"), "%s",
 221                                 "Peer Identify Request");
 222                 return;
 223         case ARP_OP_INVREPLY:
 224                 snprintf(buf, sizeof("Peer Identify Reply"), "%s",
 225                                 "Peer Identify Reply");
 226                 return;
 227         default:
 228                 break;
 229         }
 230         snprintf(buf, sizeof("Unknown"), "%s", "Unknown");
 231         return;
 232 }
 233 #endif
 234 #define MaxIPv4String   16
 235 static void
 236 ipv4_addr_to_dot(uint32_t be_ipv4_addr, char *buf, uint8_t buf_size)
 237 {
 238         uint32_t ipv4_addr;
 239
 240         ipv4_addr = rte_be_to_cpu_32(be_ipv4_addr);
 241         snprintf(buf, buf_size, "%d.%d.%d.%d", (ipv4_addr >> 24) & 0xFF,
 242                 (ipv4_addr >> 16) & 0xFF, (ipv4_addr >> 8) & 0xFF,
 243                 ipv4_addr & 0xFF);
 244 }
 245
 246 #define MAX_CLIENTS_NUMBER      128
 247 uint8_t active_clients;
 248 struct client_stats_t {
 249         uint8_t port;
 250         uint32_t ipv4_addr;
 251         uint32_t ipv4_rx_packets;
 252         uint32_t ipv4_tx_packets;
 253 };
 254 struct client_stats_t client_stats[MAX_CLIENTS_NUMBER];
 255
 256 static void
 257 update_client_stats(uint32_t addr, uint8_t port, uint32_t *TXorRXindicator)
 258 {
 259         int i = 0;
 260
 261         for (; i < MAX_CLIENTS_NUMBER; i++)     {
 262                 if ((client_stats[i].ipv4_addr == addr) && (client_stats[i].port == port))      {
 263                         /* Just update RX packets number for this client */
 264                         if (TXorRXindicator == &burstnumberRX)
 265                                 client_stats[i].ipv4_rx_packets++;
 266                         else
 267                                 client_stats[i].ipv4_tx_packets++;
 268                         return;
 269                 }
 270         }
 271         /* We have a new client. Insert him to the table, and increment stats */
 272         if (TXorRXindicator == &burstnumberRX)
 273                 client_stats[active_clients].ipv4_rx_packets++;
 274         else
 275                 client_stats[active_clients].ipv4_tx_packets++;
 276         client_stats[active_clients].ipv4_addr = addr;
 277         client_stats[active_clients].port = port;
 278         active_clients++;
 279
 280 }
 281
 282 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
 283 #define MODE6_DEBUG(info, src_ip, dst_ip, eth_h, arp_op, port, burstnumber)     \
 284                 RTE_LOG(DEBUG, PMD, \
 285                 "%s " \
 286                 "port:%d " \
 287                 "SrcMAC:%02X:%02X:%02X:%02X:%02X:%02X " \
 288                 "SrcIP:%s " \
 289                 "DstMAC:%02X:%02X:%02X:%02X:%02X:%02X " \
 290                 "DstIP:%s " \
 291                 "%s " \
 292                 "%d\n", \
 293                 info, \
 294                 port, \
 295                 eth_h->s_addr.addr_bytes[0], \
 296                 eth_h->s_addr.addr_bytes[1], \
 297                 eth_h->s_addr.addr_bytes[2], \
 298                 eth_h->s_addr.addr_bytes[3], \
 299                 eth_h->s_addr.addr_bytes[4], \
 300                 eth_h->s_addr.addr_bytes[5], \
 301                 src_ip, \
 302                 eth_h->d_addr.addr_bytes[0], \
 303                 eth_h->d_addr.addr_bytes[1], \
 304                 eth_h->d_addr.addr_bytes[2], \
 305                 eth_h->d_addr.addr_bytes[3], \
 306                 eth_h->d_addr.addr_bytes[4], \
 307                 eth_h->d_addr.addr_bytes[5], \
 308                 dst_ip, \
 309                 arp_op, \
 310                 ++burstnumber)
 311 #endif
 312
 313 static void
 314 mode6_debug(const char __attribute__((unused)) *info, struct ether_hdr *eth_h,
 315                 uint8_t port, uint32_t __attribute__((unused)) *burstnumber)
 316 {
 317         struct ipv4_hdr *ipv4_h;
 318 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
 319         struct arp_hdr *arp_h;
 320         char dst_ip[16];
 321         char ArpOp[24];
 322         char buf[16];
 323 #endif
 324         char src_ip[16];
 325
 326         uint16_t ether_type = eth_h->ether_type;
 327         uint16_t offset = get_vlan_offset(eth_h, &ether_type);
 328
 329 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
 330         snprintf(buf, 16, "%s", info);
 331 #endif
 332
 333         if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4)) {
 334                 ipv4_h = (struct ipv4_hdr *)((char *)(eth_h + 1) + offset);
 335                 ipv4_addr_to_dot(ipv4_h->src_addr, src_ip, MaxIPv4String);
 336 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
 337                 ipv4_addr_to_dot(ipv4_h->dst_addr, dst_ip, MaxIPv4String);
 338                 MODE6_DEBUG(buf, src_ip, dst_ip, eth_h, "", port, *burstnumber);
 339 #endif
 340                 update_client_stats(ipv4_h->src_addr, port, burstnumber);
 341         }
 342 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
 343         else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_ARP)) {
 344                 arp_h = (struct arp_hdr *)((char *)(eth_h + 1) + offset);
 345                 ipv4_addr_to_dot(arp_h->arp_data.arp_sip, src_ip, MaxIPv4String);
 346                 ipv4_addr_to_dot(arp_h->arp_data.arp_tip, dst_ip, MaxIPv4String);
 347                 arp_op_name(rte_be_to_cpu_16(arp_h->arp_op), ArpOp);
 348                 MODE6_DEBUG(buf, src_ip, dst_ip, eth_h, ArpOp, port, *burstnumber);
 349         }
 350 #endif
 351 }
 352 #endif
 353
 354 static uint16_t
 355 bond_ethdev_rx_burst_alb(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 356 {
 357         struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue;
 358         struct bond_dev_private *internals = bd_tx_q->dev_private;
 359         struct ether_hdr *eth_h;
 360         uint16_t ether_type, offset;
 361         uint16_t nb_recv_pkts;
 362         int i;
 363
 364         nb_recv_pkts = bond_ethdev_rx_burst(queue, bufs, nb_pkts);
 365
 366         for (i = 0; i < nb_recv_pkts; i++) {
 367                 eth_h = rte_pktmbuf_mtod(bufs[i], struct ether_hdr *);
 368                 ether_type = eth_h->ether_type;
 369                 offset = get_vlan_offset(eth_h, &ether_type);
 370
 371                 if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_ARP)) {
 372 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
 373                         mode6_debug("RX ARP:", eth_h, bufs[i]->port, &burstnumberRX);
 374 #endif
 375                         bond_mode_alb_arp_recv(eth_h, offset, internals);
 376                 }
 377 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
 378                 else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4))
 379                         mode6_debug("RX IPv4:", eth_h, bufs[i]->port, &burstnumberRX);
 380 #endif
 381         }
 382
 383         return nb_recv_pkts;
 384 }
 385
 386 static uint16_t
 387 bond_ethdev_tx_burst_round_robin(void *queue, struct rte_mbuf **bufs,
 388                 uint16_t nb_pkts)
 389 {
 390         struct bond_dev_private *internals;
 391         struct bond_tx_queue *bd_tx_q;
 392
 393         struct rte_mbuf *slave_bufs[RTE_MAX_ETHPORTS][nb_pkts];
 394         uint16_t slave_nb_pkts[RTE_MAX_ETHPORTS] = { 0 };
 395
 396         uint8_t num_of_slaves;
 397         uint8_t slaves[RTE_MAX_ETHPORTS];
 398
 399         uint16_t num_tx_total = 0, num_tx_slave;
 400
 401         static int slave_idx = 0;
 402         int i, cslave_idx = 0, tx_fail_total = 0;
 403
 404         bd_tx_q = (struct bond_tx_queue *)queue;
 405         internals = bd_tx_q->dev_private;
 406
 407         /* Copy slave list to protect against slave up/down changes during tx
 408          * bursting */
 409         num_of_slaves = internals->active_slave_count;
 410         memcpy(slaves, internals->active_slaves,
 411                         sizeof(internals->active_slaves[0]) * num_of_slaves);
 412
 413         if (num_of_slaves < 1)
 414                 return num_tx_total;
 415
 416         /* Populate slaves mbuf with which packets are to be sent on it  */
 417         for (i = 0; i < nb_pkts; i++) {
 418                 cslave_idx = (slave_idx + i) % num_of_slaves;
 419                 slave_bufs[cslave_idx][(slave_nb_pkts[cslave_idx])++] = bufs[i];
 420         }
 421
 422         /* increment current slave index so the next call to tx burst starts on the
 423          * next slave */
 424         slave_idx = ++cslave_idx;
 425
 426         /* Send packet burst on each slave device */
 427         for (i = 0; i < num_of_slaves; i++) {
 428                 if (slave_nb_pkts[i] > 0) {
 429                         num_tx_slave = rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
 430                                         slave_bufs[i], slave_nb_pkts[i]);
 431
 432                         /* if tx burst fails move packets to end of bufs */
 433                         if (unlikely(num_tx_slave < slave_nb_pkts[i])) {
 434                                 int tx_fail_slave = slave_nb_pkts[i] - num_tx_slave;
 435
 436                                 tx_fail_total += tx_fail_slave;
 437
 438                                 memcpy(&bufs[nb_pkts - tx_fail_total],
 439                                                 &slave_bufs[i][num_tx_slave],
 440                                                 tx_fail_slave * sizeof(bufs[0]));
 441                         }
 442                         num_tx_total += num_tx_slave;
 443                 }
 444         }
 445
 446         return num_tx_total;
 447 }
 448
 449 static uint16_t
 450 bond_ethdev_tx_burst_active_backup(void *queue,
 451                 struct rte_mbuf **bufs, uint16_t nb_pkts)
 452 {
 453         struct bond_dev_private *internals;
 454         struct bond_tx_queue *bd_tx_q;
 455
 456         bd_tx_q = (struct bond_tx_queue *)queue;
 457         internals = bd_tx_q->dev_private;
 458
 459         if (internals->active_slave_count < 1)
 460                 return 0;
 461
 462         return rte_eth_tx_burst(internals->current_primary_port, bd_tx_q->queue_id,
 463                         bufs, nb_pkts);
 464 }
 465
 466 static inline uint16_t
 467 ether_hash(struct ether_hdr *eth_hdr)
 468 {
 469         uint16_t *word_src_addr = (uint16_t *)eth_hdr->s_addr.addr_bytes;
 470         uint16_t *word_dst_addr = (uint16_t *)eth_hdr->d_addr.addr_bytes;
 471
 472         return (word_src_addr[0] ^ word_dst_addr[0]) ^
 473                         (word_src_addr[1] ^ word_dst_addr[1]) ^
 474                         (word_src_addr[2] ^ word_dst_addr[2]);
 475 }
 476
 477 static inline uint32_t
 478 ipv4_hash(struct ipv4_hdr *ipv4_hdr)
 479 {
 480         return (ipv4_hdr->src_addr ^ ipv4_hdr->dst_addr);
 481 }
 482
 483 static inline uint32_t
 484 ipv6_hash(struct ipv6_hdr *ipv6_hdr)
 485 {
 486         uint32_t *word_src_addr = (uint32_t *)&(ipv6_hdr->src_addr[0]);
 487         uint32_t *word_dst_addr = (uint32_t *)&(ipv6_hdr->dst_addr[0]);
 488
 489         return (word_src_addr[0] ^ word_dst_addr[0]) ^
 490                         (word_src_addr[1] ^ word_dst_addr[1]) ^
 491                         (word_src_addr[2] ^ word_dst_addr[2]) ^
 492                         (word_src_addr[3] ^ word_dst_addr[3]);
 493 }
 494
 495 uint16_t
 496 xmit_l2_hash(const struct rte_mbuf *buf, uint8_t slave_count)
 497 {
 498         struct ether_hdr *eth_hdr = rte_pktmbuf_mtod(buf, struct ether_hdr *);
 499
 500         uint32_t hash = ether_hash(eth_hdr);
 501
 502         return (hash ^= hash >> 8) % slave_count;
 503 }
 504
 505 uint16_t
 506 xmit_l23_hash(const struct rte_mbuf *buf, uint8_t slave_count)
 507 {
 508         struct ether_hdr *eth_hdr = rte_pktmbuf_mtod(buf, struct ether_hdr *);
 509         uint16_t proto = eth_hdr->ether_type;
 510         size_t vlan_offset = get_vlan_offset(eth_hdr, &proto);
 511         uint32_t hash, l3hash = 0;
 512
 513         hash = ether_hash(eth_hdr);
 514
 515         if (rte_cpu_to_be_16(ETHER_TYPE_IPv4) == proto) {
 516                 struct ipv4_hdr *ipv4_hdr = (struct ipv4_hdr *)
 517                                 ((char *)(eth_hdr + 1) + vlan_offset);
 518                 l3hash = ipv4_hash(ipv4_hdr);
 519
 520         } else if (rte_cpu_to_be_16(ETHER_TYPE_IPv6) == proto) {
 521                 struct ipv6_hdr *ipv6_hdr = (struct ipv6_hdr *)
 522                                 ((char *)(eth_hdr + 1) + vlan_offset);
 523                 l3hash = ipv6_hash(ipv6_hdr);
 524         }
 525
 526         hash = hash ^ l3hash;
 527         hash ^= hash >> 16;
 528         hash ^= hash >> 8;
 529
 530         return hash % slave_count;
 531 }
 532
 533 uint16_t
 534 xmit_l34_hash(const struct rte_mbuf *buf, uint8_t slave_count)
 535 {
 536         struct ether_hdr *eth_hdr = rte_pktmbuf_mtod(buf, struct ether_hdr *);
 537         uint16_t proto = eth_hdr->ether_type;
 538         size_t vlan_offset = get_vlan_offset(eth_hdr, &proto);
 539
 540         struct udp_hdr *udp_hdr = NULL;
 541         struct tcp_hdr *tcp_hdr = NULL;
 542         uint32_t hash, l3hash = 0, l4hash = 0;
 543
 544         if (rte_cpu_to_be_16(ETHER_TYPE_IPv4) == proto) {
 545                 struct ipv4_hdr *ipv4_hdr = (struct ipv4_hdr *)
 546                                 ((char *)(eth_hdr + 1) + vlan_offset);
 547                 size_t ip_hdr_offset;
 548
 549                 l3hash = ipv4_hash(ipv4_hdr);
 550
 551                 ip_hdr_offset = (ipv4_hdr->version_ihl & IPV4_HDR_IHL_MASK) *
 552                                 IPV4_IHL_MULTIPLIER;
 553
 554                 if (ipv4_hdr->next_proto_id == IPPROTO_TCP) {
 555                         tcp_hdr = (struct tcp_hdr *)((char *)ipv4_hdr +
 556                                         ip_hdr_offset);
 557                         l4hash = HASH_L4_PORTS(tcp_hdr);
 558                 } else if (ipv4_hdr->next_proto_id == IPPROTO_UDP) {
 559                         udp_hdr = (struct udp_hdr *)((char *)ipv4_hdr +
 560                                         ip_hdr_offset);
 561                         l4hash = HASH_L4_PORTS(udp_hdr);
 562                 }
 563         } else if  (rte_cpu_to_be_16(ETHER_TYPE_IPv6) == proto) {
 564                 struct ipv6_hdr *ipv6_hdr = (struct ipv6_hdr *)
 565                                 ((char *)(eth_hdr + 1) + vlan_offset);
 566                 l3hash = ipv6_hash(ipv6_hdr);
 567
 568                 if (ipv6_hdr->proto == IPPROTO_TCP) {
 569                         tcp_hdr = (struct tcp_hdr *)(ipv6_hdr + 1);
 570                         l4hash = HASH_L4_PORTS(tcp_hdr);
 571                 } else if (ipv6_hdr->proto == IPPROTO_UDP) {
 572                         udp_hdr = (struct udp_hdr *)(ipv6_hdr + 1);
 573                         l4hash = HASH_L4_PORTS(udp_hdr);
 574                 }
 575         }
 576
 577         hash = l3hash ^ l4hash;
 578         hash ^= hash >> 16;
 579         hash ^= hash >> 8;
 580
 581         return hash % slave_count;
 582 }
 583
 584 struct bwg_slave {
 585         uint64_t bwg_left_int;
 586         uint64_t bwg_left_remainder;
 587         uint8_t slave;
 588 };
 589
 590 void
 591 bond_tlb_activate_slave(struct bond_dev_private *internals) {
 592         int i;
 593
 594         for (i = 0; i < internals->active_slave_count; i++) {
 595                 tlb_last_obytets[internals->active_slaves[i]] = 0;
 596         }
 597 }
 598
 599 static int
 600 bandwidth_cmp(const void *a, const void *b)
 601 {
 602         const struct bwg_slave *bwg_a = a;
 603         const struct bwg_slave *bwg_b = b;
 604         int64_t diff = (int64_t)bwg_b->bwg_left_int - (int64_t)bwg_a->bwg_left_int;
 605         int64_t diff2 = (int64_t)bwg_b->bwg_left_remainder -
 606                         (int64_t)bwg_a->bwg_left_remainder;
 607         if (diff > 0)
 608                 return 1;
 609         else if (diff < 0)
 610                 return -1;
 611         else if (diff2 > 0)
 612                 return 1;
 613         else if (diff2 < 0)
 614                 return -1;
 615         else
 616                 return 0;
 617 }
 618
 619 static void
 620 bandwidth_left(uint8_t port_id, uint64_t load, uint8_t update_idx,
 621                 struct bwg_slave *bwg_slave)
 622 {
 623         struct rte_eth_link link_status;
 624
 625         rte_eth_link_get(port_id, &link_status);
 626         uint64_t link_bwg = link_status.link_speed * 1000000ULL / 8;
 627         if (link_bwg == 0)
 628                 return;
 629         link_bwg = link_bwg * (update_idx+1) * REORDER_PERIOD_MS;
 630         bwg_slave->bwg_left_int = (link_bwg - 1000*load) / link_bwg;
 631         bwg_slave->bwg_left_remainder = (link_bwg - 1000*load) % link_bwg;
 632 }
 633
 634 static void
 635 bond_ethdev_update_tlb_slave_cb(void *arg)
 636 {
 637         struct bond_dev_private *internals = arg;
 638         struct rte_eth_stats slave_stats;
 639         struct bwg_slave bwg_array[RTE_MAX_ETHPORTS];
 640         uint8_t slave_count;
 641         uint64_t tx_bytes;
 642
 643         uint8_t update_stats = 0;
 644         uint8_t i, slave_id;
 645
 646         internals->slave_update_idx++;
 647
 648
 649         if (internals->slave_update_idx >= REORDER_PERIOD_MS)
 650                 update_stats = 1;
 651
 652         for (i = 0; i < internals->active_slave_count; i++) {
 653                 slave_id = internals->active_slaves[i];
 654                 rte_eth_stats_get(slave_id, &slave_stats);
 655                 tx_bytes = slave_stats.obytes - tlb_last_obytets[slave_id];
 656                 bandwidth_left(slave_id, tx_bytes,
 657                                 internals->slave_update_idx, &bwg_array[i]);
 658                 bwg_array[i].slave = slave_id;
 659
 660                 if (update_stats) {
 661                         tlb_last_obytets[slave_id] = slave_stats.obytes;
 662                 }
 663         }
 664
 665         if (update_stats == 1)
 666                 internals->slave_update_idx = 0;
 667
 668         slave_count = i;
 669         qsort(bwg_array, slave_count, sizeof(bwg_array[0]), bandwidth_cmp);
 670         for (i = 0; i < slave_count; i++)
 671                 internals->tlb_slaves_order[i] = bwg_array[i].slave;
 672
 673         rte_eal_alarm_set(REORDER_PERIOD_MS * 1000, bond_ethdev_update_tlb_slave_cb,
 674                         (struct bond_dev_private *)internals);
 675 }
 676
 677 static uint16_t
 678 bond_ethdev_tx_burst_tlb(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 679 {
 680         struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue;
 681         struct bond_dev_private *internals = bd_tx_q->dev_private;
 682
 683         struct rte_eth_dev *primary_port =
 684                         &rte_eth_devices[internals->primary_port];
 685         uint16_t num_tx_total = 0;
 686         uint8_t i, j;
 687
 688         uint8_t num_of_slaves = internals->active_slave_count;
 689         uint8_t slaves[RTE_MAX_ETHPORTS];
 690
 691         struct ether_hdr *ether_hdr;
 692         struct ether_addr primary_slave_addr;
 693         struct ether_addr active_slave_addr;
 694
 695         if (num_of_slaves < 1)
 696                 return num_tx_total;
 697
 698         memcpy(slaves, internals->tlb_slaves_order,
 699                                 sizeof(internals->tlb_slaves_order[0]) * num_of_slaves);
 700
 701
 702         ether_addr_copy(primary_port->data->mac_addrs, &primary_slave_addr);
 703
 704         if (nb_pkts > 3) {
 705                 for (i = 0; i < 3; i++)
 706                         rte_prefetch0(rte_pktmbuf_mtod(bufs[i], void*));
 707         }
 708
 709         for (i = 0; i < num_of_slaves; i++) {
 710                 rte_eth_macaddr_get(slaves[i], &active_slave_addr);
 711                 for (j = num_tx_total; j < nb_pkts; j++) {
 712                         if (j + 3 < nb_pkts)
 713                                 rte_prefetch0(rte_pktmbuf_mtod(bufs[j+3], void*));
 714
 715                         ether_hdr = rte_pktmbuf_mtod(bufs[j], struct ether_hdr *);
 716                         if (is_same_ether_addr(&ether_hdr->s_addr, &primary_slave_addr))
 717                                 ether_addr_copy(&active_slave_addr, &ether_hdr->s_addr);
 718 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
 719                                         mode6_debug("TX IPv4:", ether_hdr, slaves[i], &burstnumberTX);
 720 #endif
 721                 }
 722
 723                 num_tx_total += rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
 724                                 bufs + num_tx_total, nb_pkts - num_tx_total);
 725
 726                 if (num_tx_total == nb_pkts)
 727                         break;
 728         }
 729
 730         return num_tx_total;
 731 }
 732
 733 void
 734 bond_tlb_disable(struct bond_dev_private *internals)
 735 {
 736         rte_eal_alarm_cancel(bond_ethdev_update_tlb_slave_cb, internals);
 737 }
 738
 739 void
 740 bond_tlb_enable(struct bond_dev_private *internals)
 741 {
 742         bond_ethdev_update_tlb_slave_cb(internals);
 743 }
 744
 745 static uint16_t
 746 bond_ethdev_tx_burst_alb(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 747 {
 748         struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue;
 749         struct bond_dev_private *internals = bd_tx_q->dev_private;
 750
 751         struct ether_hdr *eth_h;
 752         uint16_t ether_type, offset;
 753
 754         struct client_data *client_info;
 755
 756         /*
 757          * We create transmit buffers for every slave and one additional to send
 758          * through tlb. In worst case every packet will be send on one port.
 759          */
 760         struct rte_mbuf *slave_bufs[RTE_MAX_ETHPORTS + 1][nb_pkts];
 761         uint16_t slave_bufs_pkts[RTE_MAX_ETHPORTS + 1] = { 0 };
 762
 763         /*
 764          * We create separate transmit buffers for update packets as they wont be
 765          * counted in num_tx_total.
 766          */
 767         struct rte_mbuf *update_bufs[RTE_MAX_ETHPORTS][ALB_HASH_TABLE_SIZE];
 768         uint16_t update_bufs_pkts[RTE_MAX_ETHPORTS] = { 0 };
 769
 770         struct rte_mbuf *upd_pkt;
 771         size_t pkt_size;
 772
 773         uint16_t num_send, num_not_send = 0;
 774         uint16_t num_tx_total = 0;
 775         uint8_t slave_idx;
 776
 777         int i, j;
 778
 779         /* Search tx buffer for ARP packets and forward them to alb */
 780         for (i = 0; i < nb_pkts; i++) {
 781                 eth_h = rte_pktmbuf_mtod(bufs[i], struct ether_hdr *);
 782                 ether_type = eth_h->ether_type;
 783                 offset = get_vlan_offset(eth_h, &ether_type);
 784
 785                 if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_ARP)) {
 786                         slave_idx = bond_mode_alb_arp_xmit(eth_h, offset, internals);
 787
 788                         /* Change src mac in eth header */
 789                         rte_eth_macaddr_get(slave_idx, &eth_h->s_addr);
 790
 791                         /* Add packet to slave tx buffer */
 792                         slave_bufs[slave_idx][slave_bufs_pkts[slave_idx]] = bufs[i];
 793                         slave_bufs_pkts[slave_idx]++;
 794                 } else {
 795                         /* If packet is not ARP, send it with TLB policy */
 796                         slave_bufs[RTE_MAX_ETHPORTS][slave_bufs_pkts[RTE_MAX_ETHPORTS]] =
 797                                         bufs[i];
 798                         slave_bufs_pkts[RTE_MAX_ETHPORTS]++;
 799                 }
 800         }
 801
 802         /* Update connected client ARP tables */
 803         if (internals->mode6.ntt) {
 804                 for (i = 0; i < ALB_HASH_TABLE_SIZE; i++) {
 805                         client_info = &internals->mode6.client_table[i];
 806
 807                         if (client_info->in_use) {
 808                                 /* Allocate new packet to send ARP update on current slave */
 809                                 upd_pkt = rte_pktmbuf_alloc(internals->mode6.mempool);
 810                                 if (upd_pkt == NULL) {
 811                                         RTE_LOG(ERR, PMD, "Failed to allocate ARP packet from pool\n");
 812                                         continue;
 813                                 }
 814                                 pkt_size = sizeof(struct ether_hdr) + sizeof(struct arp_hdr)
 815                                                 + client_info->vlan_count * sizeof(struct vlan_hdr);
 816                                 upd_pkt->data_len = pkt_size;
 817                                 upd_pkt->pkt_len = pkt_size;
 818
 819                                 slave_idx = bond_mode_alb_arp_upd(client_info, upd_pkt,
 820                                                 internals);
 821
 822                                 /* Add packet to update tx buffer */
 823                                 update_bufs[slave_idx][update_bufs_pkts[slave_idx]] = upd_pkt;
 824                                 update_bufs_pkts[slave_idx]++;
 825                         }
 826                 }
 827                 internals->mode6.ntt = 0;
 828         }
 829
 830         /* Send ARP packets on proper slaves */
 831         for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
 832                 if (slave_bufs_pkts[i] > 0) {
 833                         num_send = rte_eth_tx_burst(i, bd_tx_q->queue_id,
 834                                         slave_bufs[i], slave_bufs_pkts[i]);
 835                         for (j = 0; j < slave_bufs_pkts[i] - num_send; j++) {
 836                                 bufs[nb_pkts - 1 - num_not_send - j] =
 837                                                 slave_bufs[i][nb_pkts - 1 - j];
 838                         }
 839
 840                         num_tx_total += num_send;
 841                         num_not_send += slave_bufs_pkts[i] - num_send;
 842
 843 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
 844         /* Print TX stats including update packets */
 845                         for (j = 0; j < slave_bufs_pkts[i]; j++) {
 846                                 eth_h = rte_pktmbuf_mtod(slave_bufs[i][j], struct ether_hdr *);
 847                                 mode6_debug("TX ARP:", eth_h, i, &burstnumberTX);
 848                         }
 849 #endif
 850                 }
 851         }
 852
 853         /* Send update packets on proper slaves */
 854         for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
 855                 if (update_bufs_pkts[i] > 0) {
 856                         num_send = rte_eth_tx_burst(i, bd_tx_q->queue_id, update_bufs[i],
 857                                         update_bufs_pkts[i]);
 858                         for (j = num_send; j < update_bufs_pkts[i]; j++) {
 859                                 rte_pktmbuf_free(update_bufs[i][j]);
 860                         }
 861 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
 862                         for (j = 0; j < update_bufs_pkts[i]; j++) {
 863                                 eth_h = rte_pktmbuf_mtod(update_bufs[i][j], struct ether_hdr *);
 864                                 mode6_debug("TX ARPupd:", eth_h, i, &burstnumberTX);
 865                         }
 866 #endif
 867                 }
 868         }
 869
 870         /* Send non-ARP packets using tlb policy */
 871         if (slave_bufs_pkts[RTE_MAX_ETHPORTS] > 0) {
 872                 num_send = bond_ethdev_tx_burst_tlb(queue,
 873                                 slave_bufs[RTE_MAX_ETHPORTS],
 874                                 slave_bufs_pkts[RTE_MAX_ETHPORTS]);
 875
 876                 for (j = 0; j < slave_bufs_pkts[RTE_MAX_ETHPORTS]; j++) {
 877                         bufs[nb_pkts - 1 - num_not_send - j] =
 878                                         slave_bufs[RTE_MAX_ETHPORTS][nb_pkts - 1 - j];
 879                 }
 880
 881                 num_tx_total += num_send;
 882                 num_not_send += slave_bufs_pkts[RTE_MAX_ETHPORTS] - num_send;
 883         }
 884
 885         return num_tx_total;
 886 }
 887
 888 static uint16_t
 889 bond_ethdev_tx_burst_balance(void *queue, struct rte_mbuf **bufs,
 890                 uint16_t nb_pkts)
 891 {
 892         struct bond_dev_private *internals;
 893         struct bond_tx_queue *bd_tx_q;
 894
 895         uint8_t num_of_slaves;
 896         uint8_t slaves[RTE_MAX_ETHPORTS];
 897
 898         uint16_t num_tx_total = 0, num_tx_slave = 0, tx_fail_total = 0;
 899
 900         int i, op_slave_id;
 901
 902         struct rte_mbuf *slave_bufs[RTE_MAX_ETHPORTS][nb_pkts];
 903         uint16_t slave_nb_pkts[RTE_MAX_ETHPORTS] = { 0 };
 904
 905         bd_tx_q = (struct bond_tx_queue *)queue;
 906         internals = bd_tx_q->dev_private;
 907
 908         /* Copy slave list to protect against slave up/down changes during tx
 909          * bursting */
 910         num_of_slaves = internals->active_slave_count;
 911         memcpy(slaves, internals->active_slaves,
 912                         sizeof(internals->active_slaves[0]) * num_of_slaves);
 913
 914         if (num_of_slaves < 1)
 915                 return num_tx_total;
 916
 917         /* Populate slaves mbuf with the packets which are to be sent on it  */
 918         for (i = 0; i < nb_pkts; i++) {
 919                 /* Select output slave using hash based on xmit policy */
 920                 op_slave_id = internals->xmit_hash(bufs[i], num_of_slaves);
 921
 922                 /* Populate slave mbuf arrays with mbufs for that slave */
 923                 slave_bufs[op_slave_id][slave_nb_pkts[op_slave_id]++] = bufs[i];
 924         }
 925
 926         /* Send packet burst on each slave device */
 927         for (i = 0; i < num_of_slaves; i++) {
 928                 if (slave_nb_pkts[i] > 0) {
 929                         num_tx_slave = rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
 930                                         slave_bufs[i], slave_nb_pkts[i]);
 931
 932                         /* if tx burst fails move packets to end of bufs */
 933                         if (unlikely(num_tx_slave < slave_nb_pkts[i])) {
 934                                 int slave_tx_fail_count = slave_nb_pkts[i] - num_tx_slave;
 935
 936                                 tx_fail_total += slave_tx_fail_count;
 937                                 memcpy(&bufs[nb_pkts - tx_fail_total],
 938                                                 &slave_bufs[i][num_tx_slave],
 939                                                 slave_tx_fail_count * sizeof(bufs[0]));
 940                         }
 941
 942                         num_tx_total += num_tx_slave;
 943                 }
 944         }
 945
 946         return num_tx_total;
 947 }
 948
 949 static uint16_t
 950 bond_ethdev_tx_burst_8023ad(void *queue, struct rte_mbuf **bufs,
 951                 uint16_t nb_pkts)
 952 {
 953         struct bond_dev_private *internals;
 954         struct bond_tx_queue *bd_tx_q;
 955
 956         uint8_t num_of_slaves;
 957         uint8_t slaves[RTE_MAX_ETHPORTS];
 958          /* positions in slaves, not ID */
 959         uint8_t distributing_offsets[RTE_MAX_ETHPORTS];
 960         uint8_t distributing_count;
 961
 962         uint16_t num_tx_slave, num_tx_total = 0, num_tx_fail_total = 0;
 963         uint16_t i, j, op_slave_idx;
 964         const uint16_t buffs_size = nb_pkts + BOND_MODE_8023AX_SLAVE_TX_PKTS + 1;
 965
 966         /* Allocate additional packets in case 8023AD mode. */
 967         struct rte_mbuf *slave_bufs[RTE_MAX_ETHPORTS][buffs_size];
 968         void *slow_pkts[BOND_MODE_8023AX_SLAVE_TX_PKTS] = { NULL };
 969
 970         /* Total amount of packets in slave_bufs */
 971         uint16_t slave_nb_pkts[RTE_MAX_ETHPORTS] = { 0 };
 972         /* Slow packets placed in each slave */
 973         uint8_t slave_slow_nb_pkts[RTE_MAX_ETHPORTS] = { 0 };
 974
 975         bd_tx_q = (struct bond_tx_queue *)queue;
 976         internals = bd_tx_q->dev_private;
 977
 978         /* Copy slave list to protect against slave up/down changes during tx
 979          * bursting */
 980         num_of_slaves = internals->active_slave_count;
 981         if (num_of_slaves < 1)
 982                 return num_tx_total;
 983
 984         memcpy(slaves, internals->active_slaves, sizeof(slaves[0]) * num_of_slaves);
 985
 986         distributing_count = 0;
 987         for (i = 0; i < num_of_slaves; i++) {
 988                 struct port *port = &mode_8023ad_ports[slaves[i]];
 989
 990                 slave_slow_nb_pkts[i] = rte_ring_dequeue_burst(port->tx_ring,
 991                                 slow_pkts, BOND_MODE_8023AX_SLAVE_TX_PKTS);
 992                 slave_nb_pkts[i] = slave_slow_nb_pkts[i];
 993
 994                 for (j = 0; j < slave_slow_nb_pkts[i]; j++)
 995                         slave_bufs[i][j] = slow_pkts[j];
 996
 997                 if (ACTOR_STATE(port, DISTRIBUTING))
 998                         distributing_offsets[distributing_count++] = i;
 999         }
1000
1001         if (likely(distributing_count > 0)) {
1002                 /* Populate slaves mbuf with the packets which are to be sent on it */
1003                 for (i = 0; i < nb_pkts; i++) {
1004                         /* Select output slave using hash based on xmit policy */
1005                         op_slave_idx = internals->xmit_hash(bufs[i], distributing_count);
1006
1007                         /* Populate slave mbuf arrays with mbufs for that slave. Use only
1008                          * slaves that are currently distributing. */
1009                         uint8_t slave_offset = distributing_offsets[op_slave_idx];
1010                         slave_bufs[slave_offset][slave_nb_pkts[slave_offset]] = bufs[i];
1011                         slave_nb_pkts[slave_offset]++;
1012                 }
1013         }
1014
1015         /* Send packet burst on each slave device */
1016         for (i = 0; i < num_of_slaves; i++) {
1017                 if (slave_nb_pkts[i] == 0)
1018                         continue;
1019
1020                 num_tx_slave = rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
1021                                 slave_bufs[i], slave_nb_pkts[i]);
1022
1023                 /* If tx burst fails drop slow packets */
1024                 for ( ; num_tx_slave < slave_slow_nb_pkts[i]; num_tx_slave++)
1025                         rte_pktmbuf_free(slave_bufs[i][num_tx_slave]);
1026
1027                 num_tx_total += num_tx_slave - slave_slow_nb_pkts[i];
1028                 num_tx_fail_total += slave_nb_pkts[i] - num_tx_slave;
1029
1030                 /* If tx burst fails move packets to end of bufs */
1031                 if (unlikely(num_tx_slave < slave_nb_pkts[i])) {
1032                         uint16_t j = nb_pkts - num_tx_fail_total;
1033                         for ( ; num_tx_slave < slave_nb_pkts[i]; j++, num_tx_slave++)
1034                                 bufs[j] = slave_bufs[i][num_tx_slave];
1035                 }
1036         }
1037
1038         return num_tx_total;
1039 }
1040
1041 static uint16_t
1042 bond_ethdev_tx_burst_broadcast(void *queue, struct rte_mbuf **bufs,
1043                 uint16_t nb_pkts)
1044 {
1045         struct bond_dev_private *internals;
1046         struct bond_tx_queue *bd_tx_q;
1047
1048         uint8_t tx_failed_flag = 0, num_of_slaves;
1049         uint8_t slaves[RTE_MAX_ETHPORTS];
1050
1051         uint16_t max_nb_of_tx_pkts = 0;
1052
1053         int slave_tx_total[RTE_MAX_ETHPORTS];
1054         int i, most_successful_tx_slave = -1;
1055
1056         bd_tx_q = (struct bond_tx_queue *)queue;
1057         internals = bd_tx_q->dev_private;
1058
1059         /* Copy slave list to protect against slave up/down changes during tx
1060          * bursting */
1061         num_of_slaves = internals->active_slave_count;
1062         memcpy(slaves, internals->active_slaves,
1063                         sizeof(internals->active_slaves[0]) * num_of_slaves);
1064
1065         if (num_of_slaves < 1)
1066                 return 0;
1067
1068         /* Increment reference count on mbufs */
1069         for (i = 0; i < nb_pkts; i++)
1070                 rte_mbuf_refcnt_update(bufs[i], num_of_slaves - 1);
1071
1072         /* Transmit burst on each active slave */
1073         for (i = 0; i < num_of_slaves; i++) {
1074                 slave_tx_total[i] = rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
1075                                         bufs, nb_pkts);
1076
1077                 if (unlikely(slave_tx_total[i] < nb_pkts))
1078                         tx_failed_flag = 1;
1079
1080                 /* record the value and slave index for the slave which transmits the
1081                  * maximum number of packets */
1082                 if (slave_tx_total[i] > max_nb_of_tx_pkts) {
1083                         max_nb_of_tx_pkts = slave_tx_total[i];
1084                         most_successful_tx_slave = i;
1085                 }
1086         }
1087
1088         /* if slaves fail to transmit packets from burst, the calling application
1089          * is not expected to know about multiple references to packets so we must
1090          * handle failures of all packets except those of the most successful slave
1091          */
1092         if (unlikely(tx_failed_flag))
1093                 for (i = 0; i < num_of_slaves; i++)
1094                         if (i != most_successful_tx_slave)
1095                                 while (slave_tx_total[i] < nb_pkts)
1096                                         rte_pktmbuf_free(bufs[slave_tx_total[i]++]);
1097
1098         return max_nb_of_tx_pkts;
1099 }
1100
1101 void
1102 link_properties_set(struct rte_eth_dev *bonded_eth_dev,
1103                 struct rte_eth_link *slave_dev_link)
1104 {
1105         struct rte_eth_link *bonded_dev_link = &bonded_eth_dev->data->dev_link;
1106         struct bond_dev_private *internals = bonded_eth_dev->data->dev_private;
1107
1108         if (slave_dev_link->link_status &&
1109                 bonded_eth_dev->data->dev_started) {
1110                 bonded_dev_link->link_duplex = slave_dev_link->link_duplex;
1111                 bonded_dev_link->link_speed = slave_dev_link->link_speed;
1112
1113                 internals->link_props_set = 1;
1114         }
1115 }
1116
1117 void
1118 link_properties_reset(struct rte_eth_dev *bonded_eth_dev)
1119 {
1120         struct bond_dev_private *internals = bonded_eth_dev->data->dev_private;
1121
1122         memset(&(bonded_eth_dev->data->dev_link), 0,
1123                         sizeof(bonded_eth_dev->data->dev_link));
1124
1125         internals->link_props_set = 0;
1126 }
1127
1128 int
1129 link_properties_valid(struct rte_eth_link *bonded_dev_link,
1130                 struct rte_eth_link *slave_dev_link)
1131 {
1132         if (bonded_dev_link->link_duplex != slave_dev_link->link_duplex ||
1133                 bonded_dev_link->link_speed !=  slave_dev_link->link_speed)
1134                 return -1;
1135
1136         return 0;
1137 }
1138
1139 int
1140 mac_address_get(struct rte_eth_dev *eth_dev, struct ether_addr *dst_mac_addr)
1141 {
1142         struct ether_addr *mac_addr;
1143
1144         if (eth_dev == NULL) {
1145                 RTE_LOG(ERR, PMD, "%s: NULL pointer eth_dev specified\n", __func__);
1146                 return -1;
1147         }
1148
1149         if (dst_mac_addr == NULL) {
1150                 RTE_LOG(ERR, PMD, "%s: NULL pointer MAC specified\n", __func__);
1151                 return -1;
1152         }
1153
1154         mac_addr = eth_dev->data->mac_addrs;
1155
1156         ether_addr_copy(mac_addr, dst_mac_addr);
1157         return 0;
1158 }
1159
1160 int
1161 mac_address_set(struct rte_eth_dev *eth_dev, struct ether_addr *new_mac_addr)
1162 {
1163         struct ether_addr *mac_addr;
1164
1165         if (eth_dev == NULL) {
1166                 RTE_BOND_LOG(ERR, "NULL pointer eth_dev specified");
1167                 return -1;
1168         }
1169
1170         if (new_mac_addr == NULL) {
1171                 RTE_BOND_LOG(ERR, "NULL pointer MAC specified");
1172                 return -1;
1173         }
1174
1175         mac_addr = eth_dev->data->mac_addrs;
1176
1177         /* If new MAC is different to current MAC then update */
1178         if (memcmp(mac_addr, new_mac_addr, sizeof(*mac_addr)) != 0)
1179                 memcpy(mac_addr, new_mac_addr, sizeof(*mac_addr));
1180
1181         return 0;
1182 }
1183
1184 int
1185 mac_address_slaves_update(struct rte_eth_dev *bonded_eth_dev)
1186 {
1187         struct bond_dev_private *internals = bonded_eth_dev->data->dev_private;
1188         int i;
1189
1190         /* Update slave devices MAC addresses */
1191         if (internals->slave_count < 1)
1192                 return -1;
1193
1194         switch (internals->mode) {
1195         case BONDING_MODE_ROUND_ROBIN:
1196         case BONDING_MODE_BALANCE:
1197         case BONDING_MODE_BROADCAST:
1198                 for (i = 0; i < internals->slave_count; i++) {
1199                         if (mac_address_set(&rte_eth_devices[internals->slaves[i].port_id],
1200                                         bonded_eth_dev->data->mac_addrs)) {
1201                                 RTE_BOND_LOG(ERR, "Failed to update port Id %d MAC address",
1202                                                 internals->slaves[i].port_id);
1203                                 return -1;
1204                         }
1205                 }
1206                 break;
1207         case BONDING_MODE_8023AD:
1208                 bond_mode_8023ad_mac_address_update(bonded_eth_dev);
1209                 break;
1210         case BONDING_MODE_ACTIVE_BACKUP:
1211         case BONDING_MODE_TLB:
1212         case BONDING_MODE_ALB:
1213         default:
1214                 for (i = 0; i < internals->slave_count; i++) {
1215                         if (internals->slaves[i].port_id ==
1216                                         internals->current_primary_port) {
1217                                 if (mac_address_set(&rte_eth_devices[internals->primary_port],
1218                                                 bonded_eth_dev->data->mac_addrs)) {
1219                                         RTE_BOND_LOG(ERR, "Failed to update port Id %d MAC address",
1220                                                         internals->current_primary_port);
1221                                         return -1;
1222                                 }
1223                         } else {
1224                                 if (mac_address_set(
1225                                                 &rte_eth_devices[internals->slaves[i].port_id],
1226                                                 &internals->slaves[i].persisted_mac_addr)) {
1227                                         RTE_BOND_LOG(ERR, "Failed to update port Id %d MAC address",
1228                                                         internals->slaves[i].port_id);
1229                                         return -1;
1230                                 }
1231                         }
1232                 }
1233         }
1234
1235         return 0;
1236 }
1237
1238 int
1239 bond_ethdev_mode_set(struct rte_eth_dev *eth_dev, int mode)
1240 {
1241         struct bond_dev_private *internals;
1242
1243         internals = eth_dev->data->dev_private;
1244
1245         switch (mode) {
1246         case BONDING_MODE_ROUND_ROBIN:
1247                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_round_robin;
1248                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst;
1249                 break;
1250         case BONDING_MODE_ACTIVE_BACKUP:
1251                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_active_backup;
1252                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_active_backup;
1253                 break;
1254         case BONDING_MODE_BALANCE:
1255                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_balance;
1256                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst;
1257                 break;
1258         case BONDING_MODE_BROADCAST:
1259                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_broadcast;
1260                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst;
1261                 break;
1262         case BONDING_MODE_8023AD:
1263                 if (bond_mode_8023ad_enable(eth_dev) != 0)
1264                         return -1;
1265
1266                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_8023ad;
1267                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_8023ad;
1268                 RTE_LOG(WARNING, PMD,
1269                                 "Using mode 4, it is necessary to do TX burst and RX burst "
1270                                 "at least every 100ms.\n");
1271                 break;
1272         case BONDING_MODE_TLB:
1273                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_tlb;
1274                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_active_backup;
1275                 break;
1276         case BONDING_MODE_ALB:
1277                 if (bond_mode_alb_enable(eth_dev) != 0)
1278                         return -1;
1279
1280                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_alb;
1281                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_alb;
1282                 break;
1283         default:
1284                 return -1;
1285         }
1286
1287         internals->mode = mode;
1288
1289         return 0;
1290 }
1291
1292 int
1293 slave_configure(struct rte_eth_dev *bonded_eth_dev,
1294                 struct rte_eth_dev *slave_eth_dev)
1295 {
1296         struct bond_rx_queue *bd_rx_q;
1297         struct bond_tx_queue *bd_tx_q;
1298
1299         int errval;
1300         uint16_t q_id;
1301
1302         /* Stop slave */
1303         rte_eth_dev_stop(slave_eth_dev->data->port_id);
1304
1305         /* Enable interrupts on slave device if supported */
1306         if (slave_eth_dev->driver->pci_drv.drv_flags & RTE_PCI_DRV_INTR_LSC)
1307                 slave_eth_dev->data->dev_conf.intr_conf.lsc = 1;
1308
1309         /* Configure device */
1310         errval = rte_eth_dev_configure(slave_eth_dev->data->port_id,
1311                         bonded_eth_dev->data->nb_rx_queues,
1312                         bonded_eth_dev->data->nb_tx_queues,
1313                         &(slave_eth_dev->data->dev_conf));
1314         if (errval != 0) {
1315                 RTE_BOND_LOG(ERR, "Cannot configure slave device: port %u , err (%d)",
1316                                 slave_eth_dev->data->port_id, errval);
1317                 return errval;
1318         }
1319
1320         /* Setup Rx Queues */
1321         for (q_id = 0; q_id < bonded_eth_dev->data->nb_rx_queues; q_id++) {
1322                 bd_rx_q = (struct bond_rx_queue *)bonded_eth_dev->data->rx_queues[q_id];
1323
1324                 errval = rte_eth_rx_queue_setup(slave_eth_dev->data->port_id, q_id,
1325                                 bd_rx_q->nb_rx_desc,
1326                                 rte_eth_dev_socket_id(slave_eth_dev->data->port_id),
1327                                 &(bd_rx_q->rx_conf), bd_rx_q->mb_pool);
1328                 if (errval != 0) {
1329                         RTE_BOND_LOG(ERR,
1330                                         "rte_eth_rx_queue_setup: port=%d queue_id %d, err (%d)",
1331                                         slave_eth_dev->data->port_id, q_id, errval);
1332                         return errval;
1333                 }
1334         }
1335
1336         /* Setup Tx Queues */
1337         for (q_id = 0; q_id < bonded_eth_dev->data->nb_tx_queues; q_id++) {
1338                 bd_tx_q = (struct bond_tx_queue *)bonded_eth_dev->data->tx_queues[q_id];
1339
1340                 errval = rte_eth_tx_queue_setup(slave_eth_dev->data->port_id, q_id,
1341                                 bd_tx_q->nb_tx_desc,
1342                                 rte_eth_dev_socket_id(slave_eth_dev->data->port_id),
1343                                 &bd_tx_q->tx_conf);
1344                 if (errval != 0) {
1345                         RTE_BOND_LOG(ERR,
1346                                         "rte_eth_tx_queue_setup: port=%d queue_id %d, err (%d)",
1347                                         slave_eth_dev->data->port_id, q_id, errval);
1348                         return errval;
1349                 }
1350         }
1351
1352         /* Start device */
1353         errval = rte_eth_dev_start(slave_eth_dev->data->port_id);
1354         if (errval != 0) {
1355                 RTE_BOND_LOG(ERR, "rte_eth_dev_start: port=%u, err (%d)",
1356                                 slave_eth_dev->data->port_id, errval);
1357                 return -1;
1358         }
1359
1360         return 0;
1361 }
1362
1363 void
1364 slave_remove(struct bond_dev_private *internals,
1365                 struct rte_eth_dev *slave_eth_dev)
1366 {
1367         uint8_t i;
1368
1369         for (i = 0; i < internals->slave_count; i++)
1370                 if (internals->slaves[i].port_id ==
1371                                 slave_eth_dev->data->port_id)
1372                         break;
1373
1374         if (i < (internals->slave_count - 1))
1375                 memmove(&internals->slaves[i], &internals->slaves[i + 1],
1376                                 sizeof(internals->slaves[0]) *
1377                                 (internals->slave_count - i - 1));
1378
1379         internals->slave_count--;
1380 }
1381
1382 static void
1383 bond_ethdev_slave_link_status_change_monitor(void *cb_arg);
1384
1385 void
1386 slave_add(struct bond_dev_private *internals,
1387                 struct rte_eth_dev *slave_eth_dev)
1388 {
1389         struct bond_slave_details *slave_details =
1390                         &internals->slaves[internals->slave_count];
1391
1392         slave_details->port_id = slave_eth_dev->data->port_id;
1393         slave_details->last_link_status = 0;
1394
1395         /* If slave device doesn't support interrupts then we need to enabled
1396          * polling to monitor link status */
1397         if (!(slave_eth_dev->pci_dev->driver->drv_flags & RTE_PCI_DRV_INTR_LSC)) {
1398                 slave_details->link_status_poll_enabled = 1;
1399
1400                 if (!internals->link_status_polling_enabled) {
1401                         internals->link_status_polling_enabled = 1;
1402
1403                         rte_eal_alarm_set(internals->link_status_polling_interval_ms * 1000,
1404                                         bond_ethdev_slave_link_status_change_monitor,
1405                                         (void *)&rte_eth_devices[internals->port_id]);
1406                 }
1407         }
1408
1409         slave_details->link_status_wait_to_complete = 0;
1410         /* clean tlb_last_obytes when adding port for bonding device */
1411         memcpy(&(slave_details->persisted_mac_addr), slave_eth_dev->data->mac_addrs,
1412                         sizeof(struct ether_addr));
1413 }
1414
1415 void
1416 bond_ethdev_primary_set(struct bond_dev_private *internals,
1417                 uint8_t slave_port_id)
1418 {
1419         int i;
1420
1421         if (internals->active_slave_count < 1)
1422                 internals->current_primary_port = slave_port_id;
1423         else
1424                 /* Search bonded device slave ports for new proposed primary port */
1425                 for (i = 0; i < internals->active_slave_count; i++) {
1426                         if (internals->active_slaves[i] == slave_port_id)
1427                                 internals->current_primary_port = slave_port_id;
1428                 }
1429 }
1430
1431 static void
1432 bond_ethdev_promiscuous_enable(struct rte_eth_dev *eth_dev);
1433
1434 static int
1435 bond_ethdev_start(struct rte_eth_dev *eth_dev)
1436 {
1437         struct bond_dev_private *internals;
1438         int i;
1439
1440         /* slave eth dev will be started by bonded device */
1441         if (valid_bonded_ethdev(eth_dev)) {
1442                 RTE_BOND_LOG(ERR, "User tried to explicitly start a slave eth_dev (%d)",
1443                                 eth_dev->data->port_id);
1444                 return -1;
1445         }
1446
1447         eth_dev->data->dev_link.link_status = 0;
1448         eth_dev->data->dev_started = 1;
1449
1450         internals = eth_dev->data->dev_private;
1451
1452         if (internals->slave_count == 0) {
1453                 RTE_BOND_LOG(ERR, "Cannot start port since there are no slave devices");
1454                 return -1;
1455         }
1456
1457         if (internals->user_defined_mac == 0) {
1458                 struct ether_addr *new_mac_addr = NULL;
1459
1460                 for (i = 0; i < internals->slave_count; i++)
1461                         if (internals->slaves[i].port_id == internals->primary_port)
1462                                 new_mac_addr = &internals->slaves[i].persisted_mac_addr;
1463
1464                 if (new_mac_addr == NULL)
1465                         return -1;
1466
1467                 if (mac_address_set(eth_dev, new_mac_addr) != 0) {
1468                         RTE_BOND_LOG(ERR, "bonded port (%d) failed to update MAC address",
1469                                         eth_dev->data->port_id);
1470                         return -1;
1471                 }
1472         }
1473
1474         /* Update all slave devices MACs*/
1475         if (mac_address_slaves_update(eth_dev) != 0)
1476                 return -1;
1477
1478         /* If bonded device is configure in promiscuous mode then re-apply config */
1479         if (internals->promiscuous_en)
1480                 bond_ethdev_promiscuous_enable(eth_dev);
1481
1482         /* Reconfigure each slave device if starting bonded device */
1483         for (i = 0; i < internals->slave_count; i++) {
1484                 if (slave_configure(eth_dev,
1485                                 &(rte_eth_devices[internals->slaves[i].port_id])) != 0) {
1486                         RTE_BOND_LOG(ERR,
1487                                         "bonded port (%d) failed to reconfigure slave device (%d)",
1488                                         eth_dev->data->port_id, internals->slaves[i].port_id);
1489                         return -1;
1490                 }
1491         }
1492
1493         if (internals->user_defined_primary_port)
1494                 bond_ethdev_primary_set(internals, internals->primary_port);
1495
1496         if (internals->mode == BONDING_MODE_8023AD)
1497                 bond_mode_8023ad_start(eth_dev);
1498
1499         if (internals->mode == BONDING_MODE_TLB ||
1500                         internals->mode == BONDING_MODE_ALB)
1501                 bond_tlb_enable(internals);
1502
1503         return 0;
1504 }
1505
1506 static void
1507 bond_ethdev_stop(struct rte_eth_dev *eth_dev)
1508 {
1509         struct bond_dev_private *internals = eth_dev->data->dev_private;
1510         uint8_t i;
1511
1512         if (internals->mode == BONDING_MODE_8023AD) {
1513                 struct port *port;
1514                 void *pkt = NULL;
1515
1516                 bond_mode_8023ad_stop(eth_dev);
1517
1518                 /* Discard all messages to/from mode 4 state machines */
1519                 for (i = 0; i < internals->slave_count; i++) {
1520                         port = &mode_8023ad_ports[internals->slaves[i].port_id];
1521
1522                         RTE_VERIFY(port->rx_ring != NULL);
1523                         while (rte_ring_dequeue(port->rx_ring, &pkt) != -ENOENT)
1524                                 rte_pktmbuf_free(pkt);
1525
1526                         RTE_VERIFY(port->tx_ring != NULL);
1527                         while (rte_ring_dequeue(port->tx_ring, &pkt) != -ENOENT)
1528                                 rte_pktmbuf_free(pkt);
1529                 }
1530         }
1531
1532         if (internals->mode == BONDING_MODE_TLB ||
1533                         internals->mode == BONDING_MODE_ALB) {
1534                 bond_tlb_disable(internals);
1535                 for (i = 0; i < internals->active_slave_count; i++)
1536                         tlb_last_obytets[internals->active_slaves[i]] = 0;
1537         }
1538
1539         internals->active_slave_count = 0;
1540         internals->link_status_polling_enabled = 0;
1541
1542         eth_dev->data->dev_link.link_status = 0;
1543         eth_dev->data->dev_started = 0;
1544 }
1545
1546 static void
1547 bond_ethdev_close(struct rte_eth_dev *dev __rte_unused)
1548 {
1549 }
1550
1551 /* forward declaration */
1552 static int bond_ethdev_configure(struct rte_eth_dev *dev);
1553
1554 static void
1555 bond_ethdev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1556 {
1557         struct bond_dev_private *internals = dev->data->dev_private;
1558
1559         dev_info->driver_name = driver_name;
1560         dev_info->max_mac_addrs = 1;
1561
1562         dev_info->max_rx_pktlen = (uint32_t)2048;
1563
1564         dev_info->max_rx_queues = (uint16_t)128;
1565         dev_info->max_tx_queues = (uint16_t)512;
1566
1567         dev_info->min_rx_bufsize = 0;
1568         dev_info->pci_dev = dev->pci_dev;
1569
1570         dev_info->rx_offload_capa = internals->rx_offload_capa;
1571         dev_info->tx_offload_capa = internals->tx_offload_capa;
1572 }
1573
1574 static int
1575 bond_ethdev_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1576                 uint16_t nb_rx_desc, unsigned int socket_id __rte_unused,
1577                 const struct rte_eth_rxconf *rx_conf, struct rte_mempool *mb_pool)
1578 {
1579         struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)
1580                         rte_zmalloc_socket(NULL, sizeof(struct bond_rx_queue),
1581                                         0, dev->pci_dev->numa_node);
1582         if (bd_rx_q == NULL)
1583                 return -1;
1584
1585         bd_rx_q->queue_id = rx_queue_id;
1586         bd_rx_q->dev_private = dev->data->dev_private;
1587
1588         bd_rx_q->nb_rx_desc = nb_rx_desc;
1589
1590         memcpy(&(bd_rx_q->rx_conf), rx_conf, sizeof(struct rte_eth_rxconf));
1591         bd_rx_q->mb_pool = mb_pool;
1592
1593         dev->data->rx_queues[rx_queue_id] = bd_rx_q;
1594
1595         return 0;
1596 }
1597
1598 static int
1599 bond_ethdev_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1600                 uint16_t nb_tx_desc, unsigned int socket_id __rte_unused,
1601                 const struct rte_eth_txconf *tx_conf)
1602 {
1603         struct bond_tx_queue *bd_tx_q  = (struct bond_tx_queue *)
1604                         rte_zmalloc_socket(NULL, sizeof(struct bond_tx_queue),
1605                                         0, dev->pci_dev->numa_node);
1606
1607         if (bd_tx_q == NULL)
1608                 return -1;
1609
1610         bd_tx_q->queue_id = tx_queue_id;
1611         bd_tx_q->dev_private = dev->data->dev_private;
1612
1613         bd_tx_q->nb_tx_desc = nb_tx_desc;
1614         memcpy(&(bd_tx_q->tx_conf), tx_conf, sizeof(bd_tx_q->tx_conf));
1615
1616         dev->data->tx_queues[tx_queue_id] = bd_tx_q;
1617
1618         return 0;
1619 }
1620
1621 static void
1622 bond_ethdev_rx_queue_release(void *queue)
1623 {
1624         if (queue == NULL)
1625                 return;
1626
1627         rte_free(queue);
1628 }
1629
1630 static void
1631 bond_ethdev_tx_queue_release(void *queue)
1632 {
1633         if (queue == NULL)
1634                 return;
1635
1636         rte_free(queue);
1637 }
1638
1639 static void
1640 bond_ethdev_slave_link_status_change_monitor(void *cb_arg)
1641 {
1642         struct rte_eth_dev *bonded_ethdev, *slave_ethdev;
1643         struct bond_dev_private *internals;
1644
1645         /* Default value for polling slave found is true as we don't want to
1646          * disable the polling thread if we cannot get the lock */
1647         int i, polling_slave_found = 1;
1648
1649         if (cb_arg == NULL)
1650                 return;
1651
1652         bonded_ethdev = (struct rte_eth_dev *)cb_arg;
1653         internals = (struct bond_dev_private *)bonded_ethdev->data->dev_private;
1654
1655         if (!bonded_ethdev->data->dev_started ||
1656                 !internals->link_status_polling_enabled)
1657                 return;
1658
1659         /* If device is currently being configured then don't check slaves link
1660          * status, wait until next period */
1661         if (rte_spinlock_trylock(&internals->lock)) {
1662                 if (internals->slave_count > 0)
1663                         polling_slave_found = 0;
1664
1665                 for (i = 0; i < internals->slave_count; i++) {
1666                         if (!internals->slaves[i].link_status_poll_enabled)
1667                                 continue;
1668
1669                         slave_ethdev = &rte_eth_devices[internals->slaves[i].port_id];
1670                         polling_slave_found = 1;
1671
1672                         /* Update slave link status */
1673                         (*slave_ethdev->dev_ops->link_update)(slave_ethdev,
1674                                         internals->slaves[i].link_status_wait_to_complete);
1675
1676                         /* if link status has changed since last checked then call lsc
1677                          * event callback */
1678                         if (slave_ethdev->data->dev_link.link_status !=
1679                                         internals->slaves[i].last_link_status) {
1680                                 internals->slaves[i].last_link_status =
1681                                                 slave_ethdev->data->dev_link.link_status;
1682
1683                                 bond_ethdev_lsc_event_callback(internals->slaves[i].port_id,
1684                                                 RTE_ETH_EVENT_INTR_LSC,
1685                                                 &bonded_ethdev->data->port_id);
1686                         }
1687                 }
1688                 rte_spinlock_unlock(&internals->lock);
1689         }
1690
1691         if (polling_slave_found)
1692                 /* Set alarm to continue monitoring link status of slave ethdev's */
1693                 rte_eal_alarm_set(internals->link_status_polling_interval_ms * 1000,
1694                                 bond_ethdev_slave_link_status_change_monitor, cb_arg);
1695 }
1696
1697 static int
1698 bond_ethdev_link_update(struct rte_eth_dev *bonded_eth_dev,
1699                 int wait_to_complete)
1700 {
1701         struct bond_dev_private *internals = bonded_eth_dev->data->dev_private;
1702
1703         if (!bonded_eth_dev->data->dev_started ||
1704                 internals->active_slave_count == 0) {
1705                 bonded_eth_dev->data->dev_link.link_status = 0;
1706                 return 0;
1707         } else {
1708                 struct rte_eth_dev *slave_eth_dev;
1709                 int i, link_up = 0;
1710
1711                 for (i = 0; i < internals->active_slave_count; i++) {
1712                         slave_eth_dev = &rte_eth_devices[internals->active_slaves[i]];
1713
1714                         (*slave_eth_dev->dev_ops->link_update)(slave_eth_dev,
1715                                         wait_to_complete);
1716                         if (slave_eth_dev->data->dev_link.link_status == 1) {
1717                                 link_up = 1;
1718                                 break;
1719                         }
1720                 }
1721
1722                 bonded_eth_dev->data->dev_link.link_status = link_up;
1723         }
1724
1725         return 0;
1726 }
1727
1728 static void
1729 bond_ethdev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1730 {
1731         struct bond_dev_private *internals = dev->data->dev_private;
1732         struct rte_eth_stats slave_stats;
1733         int i;
1734
1735         for (i = 0; i < internals->slave_count; i++) {
1736                 rte_eth_stats_get(internals->slaves[i].port_id, &slave_stats);
1737
1738                 stats->ipackets += slave_stats.ipackets;
1739                 stats->opackets += slave_stats.opackets;
1740                 stats->ibytes += slave_stats.ibytes;
1741                 stats->obytes += slave_stats.obytes;
1742                 stats->ierrors += slave_stats.ierrors;
1743                 stats->oerrors += slave_stats.oerrors;
1744                 stats->imcasts += slave_stats.imcasts;
1745                 stats->rx_nombuf += slave_stats.rx_nombuf;
1746                 stats->fdirmatch += slave_stats.fdirmatch;
1747                 stats->fdirmiss += slave_stats.fdirmiss;
1748                 stats->tx_pause_xon += slave_stats.tx_pause_xon;
1749                 stats->rx_pause_xon += slave_stats.rx_pause_xon;
1750                 stats->tx_pause_xoff += slave_stats.tx_pause_xoff;
1751                 stats->rx_pause_xoff += slave_stats.rx_pause_xoff;
1752         }
1753 }
1754
1755 static void
1756 bond_ethdev_stats_reset(struct rte_eth_dev *dev)
1757 {
1758         struct bond_dev_private *internals = dev->data->dev_private;
1759         int i;
1760
1761         for (i = 0; i < internals->slave_count; i++)
1762                 rte_eth_stats_reset(internals->slaves[i].port_id);
1763 }
1764
1765 static void
1766 bond_ethdev_promiscuous_enable(struct rte_eth_dev *eth_dev)
1767 {
1768         struct bond_dev_private *internals = eth_dev->data->dev_private;
1769         int i;
1770
1771         internals->promiscuous_en = 1;
1772
1773         switch (internals->mode) {
1774         /* Promiscuous mode is propagated to all slaves */
1775         case BONDING_MODE_ROUND_ROBIN:
1776         case BONDING_MODE_BALANCE:
1777         case BONDING_MODE_BROADCAST:
1778                 for (i = 0; i < internals->slave_count; i++)
1779                         rte_eth_promiscuous_enable(internals->slaves[i].port_id);
1780                 break;
1781         /* In mode4 promiscus mode is managed when slave is added/removed */
1782         case BONDING_MODE_8023AD:
1783                 break;
1784         /* Promiscuous mode is propagated only to primary slave */
1785         case BONDING_MODE_ACTIVE_BACKUP:
1786         case BONDING_MODE_TLB:
1787         case BONDING_MODE_ALB:
1788         default:
1789                 rte_eth_promiscuous_enable(internals->current_primary_port);
1790         }
1791 }
1792
1793 static void
1794 bond_ethdev_promiscuous_disable(struct rte_eth_dev *dev)
1795 {
1796         struct bond_dev_private *internals = dev->data->dev_private;
1797         int i;
1798
1799         internals->promiscuous_en = 0;
1800
1801         switch (internals->mode) {
1802         /* Promiscuous mode is propagated to all slaves */
1803         case BONDING_MODE_ROUND_ROBIN:
1804         case BONDING_MODE_BALANCE:
1805         case BONDING_MODE_BROADCAST:
1806                 for (i = 0; i < internals->slave_count; i++)
1807                         rte_eth_promiscuous_disable(internals->slaves[i].port_id);
1808                 break;
1809         /* In mode4 promiscus mode is set managed when slave is added/removed */
1810         case BONDING_MODE_8023AD:
1811                 break;
1812         /* Promiscuous mode is propagated only to primary slave */
1813         case BONDING_MODE_ACTIVE_BACKUP:
1814         case BONDING_MODE_TLB:
1815         case BONDING_MODE_ALB:
1816         default:
1817                 rte_eth_promiscuous_disable(internals->current_primary_port);
1818         }
1819 }
1820
1821 static void
1822 bond_ethdev_delayed_lsc_propagation(void *arg)
1823 {
1824         if (arg == NULL)
1825                 return;
1826
1827         _rte_eth_dev_callback_process((struct rte_eth_dev *)arg,
1828                         RTE_ETH_EVENT_INTR_LSC);
1829 }
1830
1831 void
1832 bond_ethdev_lsc_event_callback(uint8_t port_id, enum rte_eth_event_type type,
1833                 void *param)
1834 {
1835         struct rte_eth_dev *bonded_eth_dev, *slave_eth_dev;
1836         struct bond_dev_private *internals;
1837         struct rte_eth_link link;
1838
1839         int i, valid_slave = 0;
1840         uint8_t active_pos;
1841         uint8_t lsc_flag = 0;
1842
1843         if (type != RTE_ETH_EVENT_INTR_LSC || param == NULL)
1844                 return;
1845
1846         bonded_eth_dev = &rte_eth_devices[*(uint8_t *)param];
1847         slave_eth_dev = &rte_eth_devices[port_id];
1848
1849         if (valid_bonded_ethdev(bonded_eth_dev))
1850                 return;
1851
1852         internals = bonded_eth_dev->data->dev_private;
1853
1854         /* If the device isn't started don't handle interrupts */
1855         if (!bonded_eth_dev->data->dev_started)
1856                 return;
1857
1858         /* verify that port_id is a valid slave of bonded port */
1859         for (i = 0; i < internals->slave_count; i++) {
1860                 if (internals->slaves[i].port_id == port_id) {
1861                         valid_slave = 1;
1862                         break;
1863                 }
1864         }
1865
1866         if (!valid_slave)
1867                 return;
1868
1869         /* Search for port in active port list */
1870         active_pos = find_slave_by_id(internals->active_slaves,
1871                         internals->active_slave_count, port_id);
1872
1873         rte_eth_link_get_nowait(port_id, &link);
1874         if (link.link_status) {
1875                 if (active_pos < internals->active_slave_count)
1876                         return;
1877
1878                 /* if no active slave ports then set this port to be primary port */
1879                 if (internals->active_slave_count < 1) {
1880                         /* If first active slave, then change link status */
1881                         bonded_eth_dev->data->dev_link.link_status = 1;
1882                         internals->current_primary_port = port_id;
1883                         lsc_flag = 1;
1884
1885                         mac_address_slaves_update(bonded_eth_dev);
1886
1887                         /* Inherit eth dev link properties from first active slave */
1888                         link_properties_set(bonded_eth_dev,
1889                                         &(slave_eth_dev->data->dev_link));
1890                 }
1891
1892                 activate_slave(bonded_eth_dev, port_id);
1893
1894                 /* If user has defined the primary port then default to using it */
1895                 if (internals->user_defined_primary_port &&
1896                                 internals->primary_port == port_id)
1897                         bond_ethdev_primary_set(internals, port_id);
1898         } else {
1899                 if (active_pos == internals->active_slave_count)
1900                         return;
1901
1902                 /* Remove from active slave list */
1903                 deactivate_slave(bonded_eth_dev, port_id);
1904
1905                 /* No active slaves, change link status to down and reset other
1906                  * link properties */
1907                 if (internals->active_slave_count < 1) {
1908                         lsc_flag = 1;
1909                         bonded_eth_dev->data->dev_link.link_status = 0;
1910
1911                         link_properties_reset(bonded_eth_dev);
1912                 }
1913
1914                 /* Update primary id, take first active slave from list or if none
1915                  * available set to -1 */
1916                 if (port_id == internals->current_primary_port) {
1917                         if (internals->active_slave_count > 0)
1918                                 bond_ethdev_primary_set(internals,
1919                                                 internals->active_slaves[0]);
1920                         else
1921                                 internals->current_primary_port = internals->primary_port;
1922                 }
1923         }
1924
1925         if (lsc_flag) {
1926                 /* Cancel any possible outstanding interrupts if delays are enabled */
1927                 if (internals->link_up_delay_ms > 0 ||
1928                         internals->link_down_delay_ms > 0)
1929                         rte_eal_alarm_cancel(bond_ethdev_delayed_lsc_propagation,
1930                                         bonded_eth_dev);
1931
1932                 if (bonded_eth_dev->data->dev_link.link_status) {
1933                         if (internals->link_up_delay_ms > 0)
1934                                 rte_eal_alarm_set(internals->link_up_delay_ms * 1000,
1935                                                 bond_ethdev_delayed_lsc_propagation,
1936                                                 (void *)bonded_eth_dev);
1937                         else
1938                                 _rte_eth_dev_callback_process(bonded_eth_dev,
1939                                                 RTE_ETH_EVENT_INTR_LSC);
1940
1941                 } else {
1942                         if (internals->link_down_delay_ms > 0)
1943                                 rte_eal_alarm_set(internals->link_down_delay_ms * 1000,
1944                                                 bond_ethdev_delayed_lsc_propagation,
1945                                                 (void *)bonded_eth_dev);
1946                         else
1947                                 _rte_eth_dev_callback_process(bonded_eth_dev,
1948                                                 RTE_ETH_EVENT_INTR_LSC);
1949                 }
1950         }
1951 }
1952
1953 struct eth_dev_ops default_dev_ops = {
1954                 .dev_start = bond_ethdev_start,
1955                 .dev_stop = bond_ethdev_stop,
1956                 .dev_close = bond_ethdev_close,
1957                 .dev_configure = bond_ethdev_configure,
1958                 .dev_infos_get = bond_ethdev_info,
1959                 .rx_queue_setup = bond_ethdev_rx_queue_setup,
1960                 .tx_queue_setup = bond_ethdev_tx_queue_setup,
1961                 .rx_queue_release = bond_ethdev_rx_queue_release,
1962                 .tx_queue_release = bond_ethdev_tx_queue_release,
1963                 .link_update = bond_ethdev_link_update,
1964                 .stats_get = bond_ethdev_stats_get,
1965                 .stats_reset = bond_ethdev_stats_reset,
1966                 .promiscuous_enable = bond_ethdev_promiscuous_enable,
1967                 .promiscuous_disable = bond_ethdev_promiscuous_disable
1968 };
1969
1970 static int
1971 bond_init(const char *name, const char *params)
1972 {
1973         struct bond_dev_private *internals;
1974         struct rte_kvargs *kvlist;
1975         uint8_t bonding_mode, socket_id;
1976         int  arg_count, port_id;
1977
1978         RTE_LOG(INFO, EAL, "Initializing pmd_bond for %s\n", name);
1979
1980         kvlist = rte_kvargs_parse(params, pmd_bond_init_valid_arguments);
1981         if (kvlist == NULL)
1982                 return -1;
1983
1984         /* Parse link bonding mode */
1985         if (rte_kvargs_count(kvlist, PMD_BOND_MODE_KVARG) == 1) {
1986                 if (rte_kvargs_process(kvlist, PMD_BOND_MODE_KVARG,
1987                                 &bond_ethdev_parse_slave_mode_kvarg,
1988                                 &bonding_mode) != 0) {
1989                         RTE_LOG(ERR, EAL, "Invalid mode for bonded device %s\n",
1990                                         name);
1991                         goto parse_error;
1992                 }
1993         } else {
1994                 RTE_LOG(ERR, EAL, "Mode must be specified only once for bonded "
1995                                 "device %s\n", name);
1996                 goto parse_error;
1997         }
1998
1999         /* Parse socket id to create bonding device on */
2000         arg_count = rte_kvargs_count(kvlist, PMD_BOND_SOCKET_ID_KVARG);
2001         if (arg_count == 1) {
2002                 if (rte_kvargs_process(kvlist, PMD_BOND_SOCKET_ID_KVARG,
2003                                 &bond_ethdev_parse_socket_id_kvarg, &socket_id)
2004                                 != 0) {
2005                         RTE_LOG(ERR, EAL, "Invalid socket Id specified for "
2006                                         "bonded device %s\n", name);
2007                         goto parse_error;
2008                 }
2009         } else if (arg_count > 1) {
2010                 RTE_LOG(ERR, EAL, "Socket Id can be specified only once for "
2011                                 "bonded device %s\n", name);
2012                 goto parse_error;
2013         } else {
2014                 socket_id = rte_socket_id();
2015         }
2016
2017         /* Create link bonding eth device */
2018         port_id = rte_eth_bond_create(name, bonding_mode, socket_id);
2019         if (port_id < 0) {
2020                 RTE_LOG(ERR, EAL, "Failed to create socket %s in mode %u on "
2021                                 "socket %u.\n", name, bonding_mode, socket_id);
2022                 goto parse_error;
2023         }
2024         internals = rte_eth_devices[port_id].data->dev_private;
2025         internals->kvlist = kvlist;
2026
2027         RTE_LOG(INFO, EAL, "Create bonded device %s on port %d in mode %u on "
2028                         "socket %u.\n", name, port_id, bonding_mode, socket_id);
2029         return 0;
2030
2031 parse_error:
2032         rte_kvargs_free(kvlist);
2033
2034         return -1;
2035 }
2036
2037 /* this part will resolve the slave portids after all the other pdev and vdev
2038  * have been allocated */
2039 static int
2040 bond_ethdev_configure(struct rte_eth_dev *dev)
2041 {
2042         char *name = dev->data->name;
2043         struct bond_dev_private *internals = dev->data->dev_private;
2044         struct rte_kvargs *kvlist = internals->kvlist;
2045         int arg_count;
2046         uint8_t port_id = dev - rte_eth_devices;
2047
2048         /*
2049          * if no kvlist, it means that this bonded device has been created
2050          * through the bonding api.
2051          */
2052         if (!kvlist)
2053                 return 0;
2054
2055         /* Parse MAC address for bonded device */
2056         arg_count = rte_kvargs_count(kvlist, PMD_BOND_MAC_ADDR_KVARG);
2057         if (arg_count == 1) {
2058                 struct ether_addr bond_mac;
2059
2060                 if (rte_kvargs_process(kvlist, PMD_BOND_MAC_ADDR_KVARG,
2061                                 &bond_ethdev_parse_bond_mac_addr_kvarg, &bond_mac) < 0) {
2062                         RTE_LOG(INFO, EAL, "Invalid mac address for bonded device %s\n",
2063                                         name);
2064                         return -1;
2065                 }
2066
2067                 /* Set MAC address */
2068                 if (rte_eth_bond_mac_address_set(port_id, &bond_mac) != 0) {
2069                         RTE_LOG(ERR, EAL,
2070                                         "Failed to set mac address on bonded device %s\n",
2071                                         name);
2072                         return -1;
2073                 }
2074         } else if (arg_count > 1) {
2075                 RTE_LOG(ERR, EAL,
2076                                 "MAC address can be specified only once for bonded device %s\n",
2077                                 name);
2078                 return -1;
2079         }
2080
2081         /* Parse/set balance mode transmit policy */
2082         arg_count = rte_kvargs_count(kvlist, PMD_BOND_XMIT_POLICY_KVARG);
2083         if (arg_count == 1) {
2084                 uint8_t xmit_policy;
2085
2086                 if (rte_kvargs_process(kvlist, PMD_BOND_XMIT_POLICY_KVARG,
2087                                 &bond_ethdev_parse_balance_xmit_policy_kvarg, &xmit_policy) !=
2088                                                 0) {
2089                         RTE_LOG(INFO, EAL,
2090                                         "Invalid xmit policy specified for bonded device %s\n",
2091                                         name);
2092                         return -1;
2093                 }
2094
2095                 /* Set balance mode transmit policy*/
2096                 if (rte_eth_bond_xmit_policy_set(port_id, xmit_policy) != 0) {
2097                         RTE_LOG(ERR, EAL,
2098                                         "Failed to set balance xmit policy on bonded device %s\n",
2099                                         name);
2100                         return -1;
2101                 }
2102         } else if (arg_count > 1) {
2103                 RTE_LOG(ERR, EAL,
2104                                 "Transmit policy can be specified only once for bonded device"
2105                                 " %s\n", name);
2106                 return -1;
2107         }
2108
2109         /* Parse/add slave ports to bonded device */
2110         if (rte_kvargs_count(kvlist, PMD_BOND_SLAVE_PORT_KVARG) > 0) {
2111                 struct bond_ethdev_slave_ports slave_ports;
2112                 unsigned i;
2113
2114                 memset(&slave_ports, 0, sizeof(slave_ports));
2115
2116                 if (rte_kvargs_process(kvlist, PMD_BOND_SLAVE_PORT_KVARG,
2117                                 &bond_ethdev_parse_slave_port_kvarg, &slave_ports) != 0) {
2118                         RTE_LOG(ERR, EAL,
2119                                         "Failed to parse slave ports for bonded device %s\n",
2120                                         name);
2121                         return -1;
2122                 }
2123
2124                 for (i = 0; i < slave_ports.slave_count; i++) {
2125                         if (rte_eth_bond_slave_add(port_id, slave_ports.slaves[i]) != 0) {
2126                                 RTE_LOG(ERR, EAL,
2127                                                 "Failed to add port %d as slave to bonded device %s\n",
2128                                                 slave_ports.slaves[i], name);
2129                         }
2130                 }
2131
2132         } else {
2133                 RTE_LOG(INFO, EAL, "No slaves specified for bonded device %s\n", name);
2134                 return -1;
2135         }
2136
2137         /* Parse/set primary slave port id*/
2138         arg_count = rte_kvargs_count(kvlist, PMD_BOND_PRIMARY_SLAVE_KVARG);
2139         if (arg_count == 1) {
2140                 uint8_t primary_slave_port_id;
2141
2142                 if (rte_kvargs_process(kvlist,
2143                                 PMD_BOND_PRIMARY_SLAVE_KVARG,
2144                                 &bond_ethdev_parse_primary_slave_port_id_kvarg,
2145                                 &primary_slave_port_id) < 0) {
2146                         RTE_LOG(INFO, EAL,
2147                                         "Invalid primary slave port id specified for bonded device"
2148                                         " %s\n", name);
2149                         return -1;
2150                 }
2151
2152                 /* Set balance mode transmit policy*/
2153                 if (rte_eth_bond_primary_set(port_id, (uint8_t)primary_slave_port_id)
2154                                 != 0) {
2155                         RTE_LOG(ERR, EAL,
2156                                         "Failed to set primary slave port %d on bonded device %s\n",
2157                                         primary_slave_port_id, name);
2158                         return -1;
2159                 }
2160         } else if (arg_count > 1) {
2161                 RTE_LOG(INFO, EAL,
2162                                 "Primary slave can be specified only once for bonded device"
2163                                 " %s\n", name);
2164                 return -1;
2165         }
2166
2167         /* Parse link status monitor polling interval */
2168         arg_count = rte_kvargs_count(kvlist, PMD_BOND_LSC_POLL_PERIOD_KVARG);
2169         if (arg_count == 1) {
2170                 uint32_t lsc_poll_interval_ms;
2171
2172                 if (rte_kvargs_process(kvlist,
2173                                 PMD_BOND_LSC_POLL_PERIOD_KVARG,
2174                                 &bond_ethdev_parse_time_ms_kvarg,
2175                                 &lsc_poll_interval_ms) < 0) {
2176                         RTE_LOG(INFO, EAL,
2177                                         "Invalid lsc polling interval value specified for bonded"
2178                                         " device %s\n", name);
2179                         return -1;
2180                 }
2181
2182                 if (rte_eth_bond_link_monitoring_set(port_id, lsc_poll_interval_ms)
2183                                 != 0) {
2184                         RTE_LOG(ERR, EAL,
2185                                         "Failed to set lsc monitor polling interval (%u ms) on"
2186                                         " bonded device %s\n", lsc_poll_interval_ms, name);
2187                         return -1;
2188                 }
2189         } else if (arg_count > 1) {
2190                 RTE_LOG(INFO, EAL,
2191                                 "LSC polling interval can be specified only once for bonded"
2192                                 " device %s\n", name);
2193                 return -1;
2194         }
2195
2196         /* Parse link up interrupt propagation delay */
2197         arg_count = rte_kvargs_count(kvlist, PMD_BOND_LINK_UP_PROP_DELAY_KVARG);
2198         if (arg_count == 1) {
2199                 uint32_t link_up_delay_ms;
2200
2201                 if (rte_kvargs_process(kvlist,
2202                                 PMD_BOND_LINK_UP_PROP_DELAY_KVARG,
2203                                 &bond_ethdev_parse_time_ms_kvarg,
2204                                 &link_up_delay_ms) < 0) {
2205                         RTE_LOG(INFO, EAL,
2206                                         "Invalid link up propagation delay value specified for"
2207                                         " bonded device %s\n", name);
2208                         return -1;
2209                 }
2210
2211                 /* Set balance mode transmit policy*/
2212                 if (rte_eth_bond_link_up_prop_delay_set(port_id, link_up_delay_ms)
2213                                 != 0) {
2214                         RTE_LOG(ERR, EAL,
2215                                         "Failed to set link up propagation delay (%u ms) on bonded"
2216                                         " device %s\n", link_up_delay_ms, name);
2217                         return -1;
2218                 }
2219         } else if (arg_count > 1) {
2220                 RTE_LOG(INFO, EAL,
2221                                 "Link up propagation delay can be specified only once for"
2222                                 " bonded device %s\n", name);
2223                 return -1;
2224         }
2225
2226         /* Parse link down interrupt propagation delay */
2227         arg_count = rte_kvargs_count(kvlist, PMD_BOND_LINK_DOWN_PROP_DELAY_KVARG);
2228         if (arg_count == 1) {
2229                 uint32_t link_down_delay_ms;
2230
2231                 if (rte_kvargs_process(kvlist,
2232                                 PMD_BOND_LINK_DOWN_PROP_DELAY_KVARG,
2233                                 &bond_ethdev_parse_time_ms_kvarg,
2234                                 &link_down_delay_ms) < 0) {
2235                         RTE_LOG(INFO, EAL,
2236                                         "Invalid link down propagation delay value specified for"
2237                                         " bonded device %s\n", name);
2238                         return -1;
2239                 }
2240
2241                 /* Set balance mode transmit policy*/
2242                 if (rte_eth_bond_link_down_prop_delay_set(port_id, link_down_delay_ms)
2243                                 != 0) {
2244                         RTE_LOG(ERR, EAL,
2245                                         "Failed to set link down propagation delay (%u ms) on"
2246                                         " bonded device %s\n", link_down_delay_ms, name);
2247                         return -1;
2248                 }
2249         } else if (arg_count > 1) {
2250                 RTE_LOG(INFO, EAL,
2251                                 "Link down propagation delay can be specified only once for"
2252                                 " bonded device %s\n", name);
2253                 return -1;
2254         }
2255
2256         return 0;
2257 }
2258
2259 static struct rte_driver bond_drv = {
2260         .name = "eth_bond",
2261         .type = PMD_VDEV,
2262         .init = bond_init,
2263 };
2264
2265 PMD_REGISTER_DRIVER(bond_drv);