drivers/net/bonding/rte_eth_bond_pmd.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2015 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_ip_frag.h>
  43 #include <rte_devargs.h>
  44 #include <rte_kvargs.h>
  45 #include <rte_vdev.h>
  46 #include <rte_alarm.h>
  47 #include <rte_cycles.h>
  48
  49 #include "rte_eth_bond.h"
  50 #include "rte_eth_bond_private.h"
  51 #include "rte_eth_bond_8023ad_private.h"
  52
  53 #define REORDER_PERIOD_MS 10
  54
  55 #define HASH_L4_PORTS(h) ((h)->src_port ^ (h)->dst_port)
  56
  57 /* Table for statistics in mode 5 TLB */
  58 static uint64_t tlb_last_obytets[RTE_MAX_ETHPORTS];
  59
  60 static inline size_t
  61 get_vlan_offset(struct ether_hdr *eth_hdr, uint16_t *proto)
  62 {
  63         size_t vlan_offset = 0;
  64
  65         if (rte_cpu_to_be_16(ETHER_TYPE_VLAN) == *proto) {
  66                 struct vlan_hdr *vlan_hdr = (struct vlan_hdr *)(eth_hdr + 1);
  67
  68                 vlan_offset = sizeof(struct vlan_hdr);
  69                 *proto = vlan_hdr->eth_proto;
  70
  71                 if (rte_cpu_to_be_16(ETHER_TYPE_VLAN) == *proto) {
  72                         vlan_hdr = vlan_hdr + 1;
  73                         *proto = vlan_hdr->eth_proto;
  74                         vlan_offset += sizeof(struct vlan_hdr);
  75                 }
  76         }
  77         return vlan_offset;
  78 }
  79
  80 static uint16_t
  81 bond_ethdev_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
  82 {
  83         struct bond_dev_private *internals;
  84
  85         uint16_t num_rx_slave = 0;
  86         uint16_t num_rx_total = 0;
  87
  88         int i;
  89
  90         /* Cast to structure, containing bonded device's port id and queue id */
  91         struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)queue;
  92
  93         internals = bd_rx_q->dev_private;
  94
  95
  96         for (i = 0; i < internals->active_slave_count && nb_pkts; i++) {
  97                 /* Offset of pointer to *bufs increases as packets are received
  98                  * from other slaves */
  99                 num_rx_slave = rte_eth_rx_burst(internals->active_slaves[i],
 100                                 bd_rx_q->queue_id, bufs + num_rx_total, nb_pkts);
 101                 if (num_rx_slave) {
 102                         num_rx_total += num_rx_slave;
 103                         nb_pkts -= num_rx_slave;
 104                 }
 105         }
 106
 107         return num_rx_total;
 108 }
 109
 110 static uint16_t
 111 bond_ethdev_rx_burst_active_backup(void *queue, struct rte_mbuf **bufs,
 112                 uint16_t nb_pkts)
 113 {
 114         struct bond_dev_private *internals;
 115
 116         /* Cast to structure, containing bonded device's port id and queue id */
 117         struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)queue;
 118
 119         internals = bd_rx_q->dev_private;
 120
 121         return rte_eth_rx_burst(internals->current_primary_port,
 122                         bd_rx_q->queue_id, bufs, nb_pkts);
 123 }
 124
 125 static inline uint8_t
 126 is_lacp_packets(uint16_t ethertype, uint8_t subtype, uint16_t vlan_tci)
 127 {
 128         const uint16_t ether_type_slow_be = rte_be_to_cpu_16(ETHER_TYPE_SLOW);
 129
 130         return !vlan_tci && (ethertype == ether_type_slow_be &&
 131                 (subtype == SLOW_SUBTYPE_MARKER || subtype == SLOW_SUBTYPE_LACP));
 132 }
 133
 134 static uint16_t
 135 bond_ethdev_rx_burst_8023ad(void *queue, struct rte_mbuf **bufs,
 136                 uint16_t nb_pkts)
 137 {
 138         /* Cast to structure, containing bonded device's port id and queue id */
 139         struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)queue;
 140         struct bond_dev_private *internals = bd_rx_q->dev_private;
 141         struct ether_addr bond_mac;
 142
 143         struct ether_hdr *hdr;
 144
 145         const uint16_t ether_type_slow_be = rte_be_to_cpu_16(ETHER_TYPE_SLOW);
 146         uint16_t num_rx_total = 0;      /* Total number of received packets */
 147         uint8_t slaves[RTE_MAX_ETHPORTS];
 148         uint8_t slave_count;
 149
 150         uint8_t collecting;  /* current slave collecting status */
 151         const uint8_t promisc = internals->promiscuous_en;
 152         uint8_t i, j, k;
 153         uint8_t subtype;
 154
 155         rte_eth_macaddr_get(internals->port_id, &bond_mac);
 156         /* Copy slave list to protect against slave up/down changes during tx
 157          * bursting */
 158         slave_count = internals->active_slave_count;
 159         memcpy(slaves, internals->active_slaves,
 160                         sizeof(internals->active_slaves[0]) * slave_count);
 161
 162         for (i = 0; i < slave_count && num_rx_total < nb_pkts; i++) {
 163                 j = num_rx_total;
 164                 collecting = ACTOR_STATE(&mode_8023ad_ports[slaves[i]], COLLECTING);
 165
 166                 /* Read packets from this slave */
 167                 num_rx_total += rte_eth_rx_burst(slaves[i], bd_rx_q->queue_id,
 168                                 &bufs[num_rx_total], nb_pkts - num_rx_total);
 169
 170                 for (k = j; k < 2 && k < num_rx_total; k++)
 171                         rte_prefetch0(rte_pktmbuf_mtod(bufs[k], void *));
 172
 173                 /* Handle slow protocol packets. */
 174                 while (j < num_rx_total) {
 175                         if (j + 3 < num_rx_total)
 176                                 rte_prefetch0(rte_pktmbuf_mtod(bufs[j + 3], void *));
 177
 178                         hdr = rte_pktmbuf_mtod(bufs[j], struct ether_hdr *);
 179                         subtype = ((struct slow_protocol_frame *)hdr)->slow_protocol.subtype;
 180
 181                         /* Remove packet from array if it is slow packet or slave is not
 182                          * in collecting state or bondign interface is not in promiscus
 183                          * mode and packet address does not match. */
 184                         if (unlikely(is_lacp_packets(hdr->ether_type, subtype, bufs[j]->vlan_tci) ||
 185                                 !collecting || (!promisc &&
 186                                         !is_multicast_ether_addr(&hdr->d_addr) &&
 187                                         !is_same_ether_addr(&bond_mac, &hdr->d_addr)))) {
 188
 189                                 if (hdr->ether_type == ether_type_slow_be) {
 190                                         bond_mode_8023ad_handle_slow_pkt(internals, slaves[i],
 191                                                 bufs[j]);
 192                                 } else
 193                                         rte_pktmbuf_free(bufs[j]);
 194
 195                                 /* Packet is managed by mode 4 or dropped, shift the array */
 196                                 num_rx_total--;
 197                                 if (j < num_rx_total) {
 198                                         memmove(&bufs[j], &bufs[j + 1], sizeof(bufs[0]) *
 199                                                 (num_rx_total - j));
 200                                 }
 201                         } else
 202                                 j++;
 203                 }
 204         }
 205
 206         return num_rx_total;
 207 }
 208
 209 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
 210 uint32_t burstnumberRX;
 211 uint32_t burstnumberTX;
 212
 213 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
 214
 215 static void
 216 arp_op_name(uint16_t arp_op, char *buf)
 217 {
 218         switch (arp_op) {
 219         case ARP_OP_REQUEST:
 220                 snprintf(buf, sizeof("ARP Request"), "%s", "ARP Request");
 221                 return;
 222         case ARP_OP_REPLY:
 223                 snprintf(buf, sizeof("ARP Reply"), "%s", "ARP Reply");
 224                 return;
 225         case ARP_OP_REVREQUEST:
 226                 snprintf(buf, sizeof("Reverse ARP Request"), "%s",
 227                                 "Reverse ARP Request");
 228                 return;
 229         case ARP_OP_REVREPLY:
 230                 snprintf(buf, sizeof("Reverse ARP Reply"), "%s",
 231                                 "Reverse ARP Reply");
 232                 return;
 233         case ARP_OP_INVREQUEST:
 234                 snprintf(buf, sizeof("Peer Identify Request"), "%s",
 235                                 "Peer Identify Request");
 236                 return;
 237         case ARP_OP_INVREPLY:
 238                 snprintf(buf, sizeof("Peer Identify Reply"), "%s",
 239                                 "Peer Identify Reply");
 240                 return;
 241         default:
 242                 break;
 243         }
 244         snprintf(buf, sizeof("Unknown"), "%s", "Unknown");
 245         return;
 246 }
 247 #endif
 248 #define MaxIPv4String   16
 249 static void
 250 ipv4_addr_to_dot(uint32_t be_ipv4_addr, char *buf, uint8_t buf_size)
 251 {
 252         uint32_t ipv4_addr;
 253
 254         ipv4_addr = rte_be_to_cpu_32(be_ipv4_addr);
 255         snprintf(buf, buf_size, "%d.%d.%d.%d", (ipv4_addr >> 24) & 0xFF,
 256                 (ipv4_addr >> 16) & 0xFF, (ipv4_addr >> 8) & 0xFF,
 257                 ipv4_addr & 0xFF);
 258 }
 259
 260 #define MAX_CLIENTS_NUMBER      128
 261 uint8_t active_clients;
 262 struct client_stats_t {
 263         uint8_t port;
 264         uint32_t ipv4_addr;
 265         uint32_t ipv4_rx_packets;
 266         uint32_t ipv4_tx_packets;
 267 };
 268 struct client_stats_t client_stats[MAX_CLIENTS_NUMBER];
 269
 270 static void
 271 update_client_stats(uint32_t addr, uint8_t port, uint32_t *TXorRXindicator)
 272 {
 273         int i = 0;
 274
 275         for (; i < MAX_CLIENTS_NUMBER; i++)     {
 276                 if ((client_stats[i].ipv4_addr == addr) && (client_stats[i].port == port))      {
 277                         /* Just update RX packets number for this client */
 278                         if (TXorRXindicator == &burstnumberRX)
 279                                 client_stats[i].ipv4_rx_packets++;
 280                         else
 281                                 client_stats[i].ipv4_tx_packets++;
 282                         return;
 283                 }
 284         }
 285         /* We have a new client. Insert him to the table, and increment stats */
 286         if (TXorRXindicator == &burstnumberRX)
 287                 client_stats[active_clients].ipv4_rx_packets++;
 288         else
 289                 client_stats[active_clients].ipv4_tx_packets++;
 290         client_stats[active_clients].ipv4_addr = addr;
 291         client_stats[active_clients].port = port;
 292         active_clients++;
 293
 294 }
 295
 296 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
 297 #define MODE6_DEBUG(info, src_ip, dst_ip, eth_h, arp_op, port, burstnumber)     \
 298                 RTE_LOG(DEBUG, PMD, \
 299                 "%s " \
 300                 "port:%d " \
 301                 "SrcMAC:%02X:%02X:%02X:%02X:%02X:%02X " \
 302                 "SrcIP:%s " \
 303                 "DstMAC:%02X:%02X:%02X:%02X:%02X:%02X " \
 304                 "DstIP:%s " \
 305                 "%s " \
 306                 "%d\n", \
 307                 info, \
 308                 port, \
 309                 eth_h->s_addr.addr_bytes[0], \
 310                 eth_h->s_addr.addr_bytes[1], \
 311                 eth_h->s_addr.addr_bytes[2], \
 312                 eth_h->s_addr.addr_bytes[3], \
 313                 eth_h->s_addr.addr_bytes[4], \
 314                 eth_h->s_addr.addr_bytes[5], \
 315                 src_ip, \
 316                 eth_h->d_addr.addr_bytes[0], \
 317                 eth_h->d_addr.addr_bytes[1], \
 318                 eth_h->d_addr.addr_bytes[2], \
 319                 eth_h->d_addr.addr_bytes[3], \
 320                 eth_h->d_addr.addr_bytes[4], \
 321                 eth_h->d_addr.addr_bytes[5], \
 322                 dst_ip, \
 323                 arp_op, \
 324                 ++burstnumber)
 325 #endif
 326
 327 static void
 328 mode6_debug(const char __attribute__((unused)) *info, struct ether_hdr *eth_h,
 329                 uint8_t port, uint32_t __attribute__((unused)) *burstnumber)
 330 {
 331         struct ipv4_hdr *ipv4_h;
 332 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
 333         struct arp_hdr *arp_h;
 334         char dst_ip[16];
 335         char ArpOp[24];
 336         char buf[16];
 337 #endif
 338         char src_ip[16];
 339
 340         uint16_t ether_type = eth_h->ether_type;
 341         uint16_t offset = get_vlan_offset(eth_h, &ether_type);
 342
 343 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
 344         snprintf(buf, 16, "%s", info);
 345 #endif
 346
 347         if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4)) {
 348                 ipv4_h = (struct ipv4_hdr *)((char *)(eth_h + 1) + offset);
 349                 ipv4_addr_to_dot(ipv4_h->src_addr, src_ip, MaxIPv4String);
 350 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
 351                 ipv4_addr_to_dot(ipv4_h->dst_addr, dst_ip, MaxIPv4String);
 352                 MODE6_DEBUG(buf, src_ip, dst_ip, eth_h, "", port, *burstnumber);
 353 #endif
 354                 update_client_stats(ipv4_h->src_addr, port, burstnumber);
 355         }
 356 #ifdef RTE_LIBRTE_BOND_DEBUG_ALB
 357         else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_ARP)) {
 358                 arp_h = (struct arp_hdr *)((char *)(eth_h + 1) + offset);
 359                 ipv4_addr_to_dot(arp_h->arp_data.arp_sip, src_ip, MaxIPv4String);
 360                 ipv4_addr_to_dot(arp_h->arp_data.arp_tip, dst_ip, MaxIPv4String);
 361                 arp_op_name(rte_be_to_cpu_16(arp_h->arp_op), ArpOp);
 362                 MODE6_DEBUG(buf, src_ip, dst_ip, eth_h, ArpOp, port, *burstnumber);
 363         }
 364 #endif
 365 }
 366 #endif
 367
 368 static uint16_t
 369 bond_ethdev_rx_burst_alb(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 370 {
 371         struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue;
 372         struct bond_dev_private *internals = bd_tx_q->dev_private;
 373         struct ether_hdr *eth_h;
 374         uint16_t ether_type, offset;
 375         uint16_t nb_recv_pkts;
 376         int i;
 377
 378         nb_recv_pkts = bond_ethdev_rx_burst(queue, bufs, nb_pkts);
 379
 380         for (i = 0; i < nb_recv_pkts; i++) {
 381                 eth_h = rte_pktmbuf_mtod(bufs[i], struct ether_hdr *);
 382                 ether_type = eth_h->ether_type;
 383                 offset = get_vlan_offset(eth_h, &ether_type);
 384
 385                 if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_ARP)) {
 386 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
 387                         mode6_debug("RX ARP:", eth_h, bufs[i]->port, &burstnumberRX);
 388 #endif
 389                         bond_mode_alb_arp_recv(eth_h, offset, internals);
 390                 }
 391 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
 392                 else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4))
 393                         mode6_debug("RX IPv4:", eth_h, bufs[i]->port, &burstnumberRX);
 394 #endif
 395         }
 396
 397         return nb_recv_pkts;
 398 }
 399
 400 static uint16_t
 401 bond_ethdev_tx_burst_round_robin(void *queue, struct rte_mbuf **bufs,
 402                 uint16_t nb_pkts)
 403 {
 404         struct bond_dev_private *internals;
 405         struct bond_tx_queue *bd_tx_q;
 406
 407         struct rte_mbuf *slave_bufs[RTE_MAX_ETHPORTS][nb_pkts];
 408         uint16_t slave_nb_pkts[RTE_MAX_ETHPORTS] = { 0 };
 409
 410         uint8_t num_of_slaves;
 411         uint8_t slaves[RTE_MAX_ETHPORTS];
 412
 413         uint16_t num_tx_total = 0, num_tx_slave;
 414
 415         static int slave_idx = 0;
 416         int i, cslave_idx = 0, tx_fail_total = 0;
 417
 418         bd_tx_q = (struct bond_tx_queue *)queue;
 419         internals = bd_tx_q->dev_private;
 420
 421         /* Copy slave list to protect against slave up/down changes during tx
 422          * bursting */
 423         num_of_slaves = internals->active_slave_count;
 424         memcpy(slaves, internals->active_slaves,
 425                         sizeof(internals->active_slaves[0]) * num_of_slaves);
 426
 427         if (num_of_slaves < 1)
 428                 return num_tx_total;
 429
 430         /* Populate slaves mbuf with which packets are to be sent on it  */
 431         for (i = 0; i < nb_pkts; i++) {
 432                 cslave_idx = (slave_idx + i) % num_of_slaves;
 433                 slave_bufs[cslave_idx][(slave_nb_pkts[cslave_idx])++] = bufs[i];
 434         }
 435
 436         /* increment current slave index so the next call to tx burst starts on the
 437          * next slave */
 438         slave_idx = ++cslave_idx;
 439
 440         /* Send packet burst on each slave device */
 441         for (i = 0; i < num_of_slaves; i++) {
 442                 if (slave_nb_pkts[i] > 0) {
 443                         num_tx_slave = rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
 444                                         slave_bufs[i], slave_nb_pkts[i]);
 445
 446                         /* if tx burst fails move packets to end of bufs */
 447                         if (unlikely(num_tx_slave < slave_nb_pkts[i])) {
 448                                 int tx_fail_slave = slave_nb_pkts[i] - num_tx_slave;
 449
 450                                 tx_fail_total += tx_fail_slave;
 451
 452                                 memcpy(&bufs[nb_pkts - tx_fail_total],
 453                                                 &slave_bufs[i][num_tx_slave],
 454                                                 tx_fail_slave * sizeof(bufs[0]));
 455                         }
 456                         num_tx_total += num_tx_slave;
 457                 }
 458         }
 459
 460         return num_tx_total;
 461 }
 462
 463 static uint16_t
 464 bond_ethdev_tx_burst_active_backup(void *queue,
 465                 struct rte_mbuf **bufs, uint16_t nb_pkts)
 466 {
 467         struct bond_dev_private *internals;
 468         struct bond_tx_queue *bd_tx_q;
 469
 470         bd_tx_q = (struct bond_tx_queue *)queue;
 471         internals = bd_tx_q->dev_private;
 472
 473         if (internals->active_slave_count < 1)
 474                 return 0;
 475
 476         return rte_eth_tx_burst(internals->current_primary_port, bd_tx_q->queue_id,
 477                         bufs, nb_pkts);
 478 }
 479
 480 static inline uint16_t
 481 ether_hash(struct ether_hdr *eth_hdr)
 482 {
 483         unaligned_uint16_t *word_src_addr =
 484                 (unaligned_uint16_t *)eth_hdr->s_addr.addr_bytes;
 485         unaligned_uint16_t *word_dst_addr =
 486                 (unaligned_uint16_t *)eth_hdr->d_addr.addr_bytes;
 487
 488         return (word_src_addr[0] ^ word_dst_addr[0]) ^
 489                         (word_src_addr[1] ^ word_dst_addr[1]) ^
 490                         (word_src_addr[2] ^ word_dst_addr[2]);
 491 }
 492
 493 static inline uint32_t
 494 ipv4_hash(struct ipv4_hdr *ipv4_hdr)
 495 {
 496         return ipv4_hdr->src_addr ^ ipv4_hdr->dst_addr;
 497 }
 498
 499 static inline uint32_t
 500 ipv6_hash(struct ipv6_hdr *ipv6_hdr)
 501 {
 502         unaligned_uint32_t *word_src_addr =
 503                 (unaligned_uint32_t *)&(ipv6_hdr->src_addr[0]);
 504         unaligned_uint32_t *word_dst_addr =
 505                 (unaligned_uint32_t *)&(ipv6_hdr->dst_addr[0]);
 506
 507         return (word_src_addr[0] ^ word_dst_addr[0]) ^
 508                         (word_src_addr[1] ^ word_dst_addr[1]) ^
 509                         (word_src_addr[2] ^ word_dst_addr[2]) ^
 510                         (word_src_addr[3] ^ word_dst_addr[3]);
 511 }
 512
 513 uint16_t
 514 xmit_l2_hash(const struct rte_mbuf *buf, uint8_t slave_count)
 515 {
 516         struct ether_hdr *eth_hdr = rte_pktmbuf_mtod(buf, struct ether_hdr *);
 517
 518         uint32_t hash = ether_hash(eth_hdr);
 519
 520         return (hash ^= hash >> 8) % slave_count;
 521 }
 522
 523 uint16_t
 524 xmit_l23_hash(const struct rte_mbuf *buf, uint8_t slave_count)
 525 {
 526         struct ether_hdr *eth_hdr = rte_pktmbuf_mtod(buf, struct ether_hdr *);
 527         uint16_t proto = eth_hdr->ether_type;
 528         size_t vlan_offset = get_vlan_offset(eth_hdr, &proto);
 529         uint32_t hash, l3hash = 0;
 530
 531         hash = ether_hash(eth_hdr);
 532
 533         if (rte_cpu_to_be_16(ETHER_TYPE_IPv4) == proto) {
 534                 struct ipv4_hdr *ipv4_hdr = (struct ipv4_hdr *)
 535                                 ((char *)(eth_hdr + 1) + vlan_offset);
 536                 l3hash = ipv4_hash(ipv4_hdr);
 537
 538         } else if (rte_cpu_to_be_16(ETHER_TYPE_IPv6) == proto) {
 539                 struct ipv6_hdr *ipv6_hdr = (struct ipv6_hdr *)
 540                                 ((char *)(eth_hdr + 1) + vlan_offset);
 541                 l3hash = ipv6_hash(ipv6_hdr);
 542         }
 543
 544         hash = hash ^ l3hash;
 545         hash ^= hash >> 16;
 546         hash ^= hash >> 8;
 547
 548         return hash % slave_count;
 549 }
 550
 551 uint16_t
 552 xmit_l34_hash(const struct rte_mbuf *buf, uint8_t slave_count)
 553 {
 554         struct ether_hdr *eth_hdr = rte_pktmbuf_mtod(buf, struct ether_hdr *);
 555         uint16_t proto = eth_hdr->ether_type;
 556         size_t vlan_offset = get_vlan_offset(eth_hdr, &proto);
 557
 558         struct udp_hdr *udp_hdr = NULL;
 559         struct tcp_hdr *tcp_hdr = NULL;
 560         uint32_t hash, l3hash = 0, l4hash = 0;
 561
 562         if (rte_cpu_to_be_16(ETHER_TYPE_IPv4) == proto) {
 563                 struct ipv4_hdr *ipv4_hdr = (struct ipv4_hdr *)
 564                                 ((char *)(eth_hdr + 1) + vlan_offset);
 565                 size_t ip_hdr_offset;
 566
 567                 l3hash = ipv4_hash(ipv4_hdr);
 568
 569                 /* there is no L4 header in fragmented packet */
 570                 if (likely(rte_ipv4_frag_pkt_is_fragmented(ipv4_hdr) == 0)) {
 571                         ip_hdr_offset = (ipv4_hdr->version_ihl & IPV4_HDR_IHL_MASK) *
 572                                         IPV4_IHL_MULTIPLIER;
 573
 574                         if (ipv4_hdr->next_proto_id == IPPROTO_TCP) {
 575                                 tcp_hdr = (struct tcp_hdr *)((char *)ipv4_hdr +
 576                                                 ip_hdr_offset);
 577                                 l4hash = HASH_L4_PORTS(tcp_hdr);
 578                         } else if (ipv4_hdr->next_proto_id == IPPROTO_UDP) {
 579                                 udp_hdr = (struct udp_hdr *)((char *)ipv4_hdr +
 580                                                 ip_hdr_offset);
 581                                 l4hash = HASH_L4_PORTS(udp_hdr);
 582                         }
 583                 }
 584         } else if  (rte_cpu_to_be_16(ETHER_TYPE_IPv6) == proto) {
 585                 struct ipv6_hdr *ipv6_hdr = (struct ipv6_hdr *)
 586                                 ((char *)(eth_hdr + 1) + vlan_offset);
 587                 l3hash = ipv6_hash(ipv6_hdr);
 588
 589                 if (ipv6_hdr->proto == IPPROTO_TCP) {
 590                         tcp_hdr = (struct tcp_hdr *)(ipv6_hdr + 1);
 591                         l4hash = HASH_L4_PORTS(tcp_hdr);
 592                 } else if (ipv6_hdr->proto == IPPROTO_UDP) {
 593                         udp_hdr = (struct udp_hdr *)(ipv6_hdr + 1);
 594                         l4hash = HASH_L4_PORTS(udp_hdr);
 595                 }
 596         }
 597
 598         hash = l3hash ^ l4hash;
 599         hash ^= hash >> 16;
 600         hash ^= hash >> 8;
 601
 602         return hash % slave_count;
 603 }
 604
 605 struct bwg_slave {
 606         uint64_t bwg_left_int;
 607         uint64_t bwg_left_remainder;
 608         uint8_t slave;
 609 };
 610
 611 void
 612 bond_tlb_activate_slave(struct bond_dev_private *internals) {
 613         int i;
 614
 615         for (i = 0; i < internals->active_slave_count; i++) {
 616                 tlb_last_obytets[internals->active_slaves[i]] = 0;
 617         }
 618 }
 619
 620 static int
 621 bandwidth_cmp(const void *a, const void *b)
 622 {
 623         const struct bwg_slave *bwg_a = a;
 624         const struct bwg_slave *bwg_b = b;
 625         int64_t diff = (int64_t)bwg_b->bwg_left_int - (int64_t)bwg_a->bwg_left_int;
 626         int64_t diff2 = (int64_t)bwg_b->bwg_left_remainder -
 627                         (int64_t)bwg_a->bwg_left_remainder;
 628         if (diff > 0)
 629                 return 1;
 630         else if (diff < 0)
 631                 return -1;
 632         else if (diff2 > 0)
 633                 return 1;
 634         else if (diff2 < 0)
 635                 return -1;
 636         else
 637                 return 0;
 638 }
 639
 640 static void
 641 bandwidth_left(uint8_t port_id, uint64_t load, uint8_t update_idx,
 642                 struct bwg_slave *bwg_slave)
 643 {
 644         struct rte_eth_link link_status;
 645
 646         rte_eth_link_get(port_id, &link_status);
 647         uint64_t link_bwg = link_status.link_speed * 1000000ULL / 8;
 648         if (link_bwg == 0)
 649                 return;
 650         link_bwg = link_bwg * (update_idx+1) * REORDER_PERIOD_MS;
 651         bwg_slave->bwg_left_int = (link_bwg - 1000*load) / link_bwg;
 652         bwg_slave->bwg_left_remainder = (link_bwg - 1000*load) % link_bwg;
 653 }
 654
 655 static void
 656 bond_ethdev_update_tlb_slave_cb(void *arg)
 657 {
 658         struct bond_dev_private *internals = arg;
 659         struct rte_eth_stats slave_stats;
 660         struct bwg_slave bwg_array[RTE_MAX_ETHPORTS];
 661         uint8_t slave_count;
 662         uint64_t tx_bytes;
 663
 664         uint8_t update_stats = 0;
 665         uint8_t i, slave_id;
 666
 667         internals->slave_update_idx++;
 668
 669
 670         if (internals->slave_update_idx >= REORDER_PERIOD_MS)
 671                 update_stats = 1;
 672
 673         for (i = 0; i < internals->active_slave_count; i++) {
 674                 slave_id = internals->active_slaves[i];
 675                 rte_eth_stats_get(slave_id, &slave_stats);
 676                 tx_bytes = slave_stats.obytes - tlb_last_obytets[slave_id];
 677                 bandwidth_left(slave_id, tx_bytes,
 678                                 internals->slave_update_idx, &bwg_array[i]);
 679                 bwg_array[i].slave = slave_id;
 680
 681                 if (update_stats) {
 682                         tlb_last_obytets[slave_id] = slave_stats.obytes;
 683                 }
 684         }
 685
 686         if (update_stats == 1)
 687                 internals->slave_update_idx = 0;
 688
 689         slave_count = i;
 690         qsort(bwg_array, slave_count, sizeof(bwg_array[0]), bandwidth_cmp);
 691         for (i = 0; i < slave_count; i++)
 692                 internals->tlb_slaves_order[i] = bwg_array[i].slave;
 693
 694         rte_eal_alarm_set(REORDER_PERIOD_MS * 1000, bond_ethdev_update_tlb_slave_cb,
 695                         (struct bond_dev_private *)internals);
 696 }
 697
 698 static uint16_t
 699 bond_ethdev_tx_burst_tlb(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 700 {
 701         struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue;
 702         struct bond_dev_private *internals = bd_tx_q->dev_private;
 703
 704         struct rte_eth_dev *primary_port =
 705                         &rte_eth_devices[internals->primary_port];
 706         uint16_t num_tx_total = 0;
 707         uint8_t i, j;
 708
 709         uint8_t num_of_slaves = internals->active_slave_count;
 710         uint8_t slaves[RTE_MAX_ETHPORTS];
 711
 712         struct ether_hdr *ether_hdr;
 713         struct ether_addr primary_slave_addr;
 714         struct ether_addr active_slave_addr;
 715
 716         if (num_of_slaves < 1)
 717                 return num_tx_total;
 718
 719         memcpy(slaves, internals->tlb_slaves_order,
 720                                 sizeof(internals->tlb_slaves_order[0]) * num_of_slaves);
 721
 722
 723         ether_addr_copy(primary_port->data->mac_addrs, &primary_slave_addr);
 724
 725         if (nb_pkts > 3) {
 726                 for (i = 0; i < 3; i++)
 727                         rte_prefetch0(rte_pktmbuf_mtod(bufs[i], void*));
 728         }
 729
 730         for (i = 0; i < num_of_slaves; i++) {
 731                 rte_eth_macaddr_get(slaves[i], &active_slave_addr);
 732                 for (j = num_tx_total; j < nb_pkts; j++) {
 733                         if (j + 3 < nb_pkts)
 734                                 rte_prefetch0(rte_pktmbuf_mtod(bufs[j+3], void*));
 735
 736                         ether_hdr = rte_pktmbuf_mtod(bufs[j], struct ether_hdr *);
 737                         if (is_same_ether_addr(&ether_hdr->s_addr, &primary_slave_addr))
 738                                 ether_addr_copy(&active_slave_addr, &ether_hdr->s_addr);
 739 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
 740                                         mode6_debug("TX IPv4:", ether_hdr, slaves[i], &burstnumberTX);
 741 #endif
 742                 }
 743
 744                 num_tx_total += rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
 745                                 bufs + num_tx_total, nb_pkts - num_tx_total);
 746
 747                 if (num_tx_total == nb_pkts)
 748                         break;
 749         }
 750
 751         return num_tx_total;
 752 }
 753
 754 void
 755 bond_tlb_disable(struct bond_dev_private *internals)
 756 {
 757         rte_eal_alarm_cancel(bond_ethdev_update_tlb_slave_cb, internals);
 758 }
 759
 760 void
 761 bond_tlb_enable(struct bond_dev_private *internals)
 762 {
 763         bond_ethdev_update_tlb_slave_cb(internals);
 764 }
 765
 766 static uint16_t
 767 bond_ethdev_tx_burst_alb(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 768 {
 769         struct bond_tx_queue *bd_tx_q = (struct bond_tx_queue *)queue;
 770         struct bond_dev_private *internals = bd_tx_q->dev_private;
 771
 772         struct ether_hdr *eth_h;
 773         uint16_t ether_type, offset;
 774
 775         struct client_data *client_info;
 776
 777         /*
 778          * We create transmit buffers for every slave and one additional to send
 779          * through tlb. In worst case every packet will be send on one port.
 780          */
 781         struct rte_mbuf *slave_bufs[RTE_MAX_ETHPORTS + 1][nb_pkts];
 782         uint16_t slave_bufs_pkts[RTE_MAX_ETHPORTS + 1] = { 0 };
 783
 784         /*
 785          * We create separate transmit buffers for update packets as they wont be
 786          * counted in num_tx_total.
 787          */
 788         struct rte_mbuf *update_bufs[RTE_MAX_ETHPORTS][ALB_HASH_TABLE_SIZE];
 789         uint16_t update_bufs_pkts[RTE_MAX_ETHPORTS] = { 0 };
 790
 791         struct rte_mbuf *upd_pkt;
 792         size_t pkt_size;
 793
 794         uint16_t num_send, num_not_send = 0;
 795         uint16_t num_tx_total = 0;
 796         uint8_t slave_idx;
 797
 798         int i, j;
 799
 800         /* Search tx buffer for ARP packets and forward them to alb */
 801         for (i = 0; i < nb_pkts; i++) {
 802                 eth_h = rte_pktmbuf_mtod(bufs[i], struct ether_hdr *);
 803                 ether_type = eth_h->ether_type;
 804                 offset = get_vlan_offset(eth_h, &ether_type);
 805
 806                 if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_ARP)) {
 807                         slave_idx = bond_mode_alb_arp_xmit(eth_h, offset, internals);
 808
 809                         /* Change src mac in eth header */
 810                         rte_eth_macaddr_get(slave_idx, &eth_h->s_addr);
 811
 812                         /* Add packet to slave tx buffer */
 813                         slave_bufs[slave_idx][slave_bufs_pkts[slave_idx]] = bufs[i];
 814                         slave_bufs_pkts[slave_idx]++;
 815                 } else {
 816                         /* If packet is not ARP, send it with TLB policy */
 817                         slave_bufs[RTE_MAX_ETHPORTS][slave_bufs_pkts[RTE_MAX_ETHPORTS]] =
 818                                         bufs[i];
 819                         slave_bufs_pkts[RTE_MAX_ETHPORTS]++;
 820                 }
 821         }
 822
 823         /* Update connected client ARP tables */
 824         if (internals->mode6.ntt) {
 825                 for (i = 0; i < ALB_HASH_TABLE_SIZE; i++) {
 826                         client_info = &internals->mode6.client_table[i];
 827
 828                         if (client_info->in_use) {
 829                                 /* Allocate new packet to send ARP update on current slave */
 830                                 upd_pkt = rte_pktmbuf_alloc(internals->mode6.mempool);
 831                                 if (upd_pkt == NULL) {
 832                                         RTE_LOG(ERR, PMD, "Failed to allocate ARP packet from pool\n");
 833                                         continue;
 834                                 }
 835                                 pkt_size = sizeof(struct ether_hdr) + sizeof(struct arp_hdr)
 836                                                 + client_info->vlan_count * sizeof(struct vlan_hdr);
 837                                 upd_pkt->data_len = pkt_size;
 838                                 upd_pkt->pkt_len = pkt_size;
 839
 840                                 slave_idx = bond_mode_alb_arp_upd(client_info, upd_pkt,
 841                                                 internals);
 842
 843                                 /* Add packet to update tx buffer */
 844                                 update_bufs[slave_idx][update_bufs_pkts[slave_idx]] = upd_pkt;
 845                                 update_bufs_pkts[slave_idx]++;
 846                         }
 847                 }
 848                 internals->mode6.ntt = 0;
 849         }
 850
 851         /* Send ARP packets on proper slaves */
 852         for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
 853                 if (slave_bufs_pkts[i] > 0) {
 854                         num_send = rte_eth_tx_burst(i, bd_tx_q->queue_id,
 855                                         slave_bufs[i], slave_bufs_pkts[i]);
 856                         for (j = 0; j < slave_bufs_pkts[i] - num_send; j++) {
 857                                 bufs[nb_pkts - 1 - num_not_send - j] =
 858                                                 slave_bufs[i][nb_pkts - 1 - j];
 859                         }
 860
 861                         num_tx_total += num_send;
 862                         num_not_send += slave_bufs_pkts[i] - num_send;
 863
 864 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
 865         /* Print TX stats including update packets */
 866                         for (j = 0; j < slave_bufs_pkts[i]; j++) {
 867                                 eth_h = rte_pktmbuf_mtod(slave_bufs[i][j], struct ether_hdr *);
 868                                 mode6_debug("TX ARP:", eth_h, i, &burstnumberTX);
 869                         }
 870 #endif
 871                 }
 872         }
 873
 874         /* Send update packets on proper slaves */
 875         for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
 876                 if (update_bufs_pkts[i] > 0) {
 877                         num_send = rte_eth_tx_burst(i, bd_tx_q->queue_id, update_bufs[i],
 878                                         update_bufs_pkts[i]);
 879                         for (j = num_send; j < update_bufs_pkts[i]; j++) {
 880                                 rte_pktmbuf_free(update_bufs[i][j]);
 881                         }
 882 #if defined(RTE_LIBRTE_BOND_DEBUG_ALB) || defined(RTE_LIBRTE_BOND_DEBUG_ALB_L1)
 883                         for (j = 0; j < update_bufs_pkts[i]; j++) {
 884                                 eth_h = rte_pktmbuf_mtod(update_bufs[i][j], struct ether_hdr *);
 885                                 mode6_debug("TX ARPupd:", eth_h, i, &burstnumberTX);
 886                         }
 887 #endif
 888                 }
 889         }
 890
 891         /* Send non-ARP packets using tlb policy */
 892         if (slave_bufs_pkts[RTE_MAX_ETHPORTS] > 0) {
 893                 num_send = bond_ethdev_tx_burst_tlb(queue,
 894                                 slave_bufs[RTE_MAX_ETHPORTS],
 895                                 slave_bufs_pkts[RTE_MAX_ETHPORTS]);
 896
 897                 for (j = 0; j < slave_bufs_pkts[RTE_MAX_ETHPORTS]; j++) {
 898                         bufs[nb_pkts - 1 - num_not_send - j] =
 899                                         slave_bufs[RTE_MAX_ETHPORTS][nb_pkts - 1 - j];
 900                 }
 901
 902                 num_tx_total += num_send;
 903                 num_not_send += slave_bufs_pkts[RTE_MAX_ETHPORTS] - num_send;
 904         }
 905
 906         return num_tx_total;
 907 }
 908
 909 static uint16_t
 910 bond_ethdev_tx_burst_balance(void *queue, struct rte_mbuf **bufs,
 911                 uint16_t nb_pkts)
 912 {
 913         struct bond_dev_private *internals;
 914         struct bond_tx_queue *bd_tx_q;
 915
 916         uint8_t num_of_slaves;
 917         uint8_t slaves[RTE_MAX_ETHPORTS];
 918
 919         uint16_t num_tx_total = 0, num_tx_slave = 0, tx_fail_total = 0;
 920
 921         int i, op_slave_id;
 922
 923         struct rte_mbuf *slave_bufs[RTE_MAX_ETHPORTS][nb_pkts];
 924         uint16_t slave_nb_pkts[RTE_MAX_ETHPORTS] = { 0 };
 925
 926         bd_tx_q = (struct bond_tx_queue *)queue;
 927         internals = bd_tx_q->dev_private;
 928
 929         /* Copy slave list to protect against slave up/down changes during tx
 930          * bursting */
 931         num_of_slaves = internals->active_slave_count;
 932         memcpy(slaves, internals->active_slaves,
 933                         sizeof(internals->active_slaves[0]) * num_of_slaves);
 934
 935         if (num_of_slaves < 1)
 936                 return num_tx_total;
 937
 938         /* Populate slaves mbuf with the packets which are to be sent on it  */
 939         for (i = 0; i < nb_pkts; i++) {
 940                 /* Select output slave using hash based on xmit policy */
 941                 op_slave_id = internals->xmit_hash(bufs[i], num_of_slaves);
 942
 943                 /* Populate slave mbuf arrays with mbufs for that slave */
 944                 slave_bufs[op_slave_id][slave_nb_pkts[op_slave_id]++] = bufs[i];
 945         }
 946
 947         /* Send packet burst on each slave device */
 948         for (i = 0; i < num_of_slaves; i++) {
 949                 if (slave_nb_pkts[i] > 0) {
 950                         num_tx_slave = rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
 951                                         slave_bufs[i], slave_nb_pkts[i]);
 952
 953                         /* if tx burst fails move packets to end of bufs */
 954                         if (unlikely(num_tx_slave < slave_nb_pkts[i])) {
 955                                 int slave_tx_fail_count = slave_nb_pkts[i] - num_tx_slave;
 956
 957                                 tx_fail_total += slave_tx_fail_count;
 958                                 memcpy(&bufs[nb_pkts - tx_fail_total],
 959                                                 &slave_bufs[i][num_tx_slave],
 960                                                 slave_tx_fail_count * sizeof(bufs[0]));
 961                         }
 962
 963                         num_tx_total += num_tx_slave;
 964                 }
 965         }
 966
 967         return num_tx_total;
 968 }
 969
 970 static uint16_t
 971 bond_ethdev_tx_burst_8023ad(void *queue, struct rte_mbuf **bufs,
 972                 uint16_t nb_pkts)
 973 {
 974         struct bond_dev_private *internals;
 975         struct bond_tx_queue *bd_tx_q;
 976
 977         uint8_t num_of_slaves;
 978         uint8_t slaves[RTE_MAX_ETHPORTS];
 979          /* positions in slaves, not ID */
 980         uint8_t distributing_offsets[RTE_MAX_ETHPORTS];
 981         uint8_t distributing_count;
 982
 983         uint16_t num_tx_slave, num_tx_total = 0, num_tx_fail_total = 0;
 984         uint16_t i, j, op_slave_idx;
 985         const uint16_t buffs_size = nb_pkts + BOND_MODE_8023AX_SLAVE_TX_PKTS + 1;
 986
 987         /* Allocate additional packets in case 8023AD mode. */
 988         struct rte_mbuf *slave_bufs[RTE_MAX_ETHPORTS][buffs_size];
 989         void *slow_pkts[BOND_MODE_8023AX_SLAVE_TX_PKTS] = { NULL };
 990
 991         /* Total amount of packets in slave_bufs */
 992         uint16_t slave_nb_pkts[RTE_MAX_ETHPORTS] = { 0 };
 993         /* Slow packets placed in each slave */
 994         uint8_t slave_slow_nb_pkts[RTE_MAX_ETHPORTS] = { 0 };
 995
 996         bd_tx_q = (struct bond_tx_queue *)queue;
 997         internals = bd_tx_q->dev_private;
 998
 999         /* Copy slave list to protect against slave up/down changes during tx
1000          * bursting */
1001         num_of_slaves = internals->active_slave_count;
1002         if (num_of_slaves < 1)
1003                 return num_tx_total;
1004
1005         memcpy(slaves, internals->active_slaves, sizeof(slaves[0]) * num_of_slaves);
1006
1007         distributing_count = 0;
1008         for (i = 0; i < num_of_slaves; i++) {
1009                 struct port *port = &mode_8023ad_ports[slaves[i]];
1010
1011                 slave_slow_nb_pkts[i] = rte_ring_dequeue_burst(port->tx_ring,
1012                                 slow_pkts, BOND_MODE_8023AX_SLAVE_TX_PKTS);
1013                 slave_nb_pkts[i] = slave_slow_nb_pkts[i];
1014
1015                 for (j = 0; j < slave_slow_nb_pkts[i]; j++)
1016                         slave_bufs[i][j] = slow_pkts[j];
1017
1018                 if (ACTOR_STATE(port, DISTRIBUTING))
1019                         distributing_offsets[distributing_count++] = i;
1020         }
1021
1022         if (likely(distributing_count > 0)) {
1023                 /* Populate slaves mbuf with the packets which are to be sent on it */
1024                 for (i = 0; i < nb_pkts; i++) {
1025                         /* Select output slave using hash based on xmit policy */
1026                         op_slave_idx = internals->xmit_hash(bufs[i], distributing_count);
1027
1028                         /* Populate slave mbuf arrays with mbufs for that slave. Use only
1029                          * slaves that are currently distributing. */
1030                         uint8_t slave_offset = distributing_offsets[op_slave_idx];
1031                         slave_bufs[slave_offset][slave_nb_pkts[slave_offset]] = bufs[i];
1032                         slave_nb_pkts[slave_offset]++;
1033                 }
1034         }
1035
1036         /* Send packet burst on each slave device */
1037         for (i = 0; i < num_of_slaves; i++) {
1038                 if (slave_nb_pkts[i] == 0)
1039                         continue;
1040
1041                 num_tx_slave = rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
1042                                 slave_bufs[i], slave_nb_pkts[i]);
1043
1044                 /* If tx burst fails drop slow packets */
1045                 for ( ; num_tx_slave < slave_slow_nb_pkts[i]; num_tx_slave++)
1046                         rte_pktmbuf_free(slave_bufs[i][num_tx_slave]);
1047
1048                 num_tx_total += num_tx_slave - slave_slow_nb_pkts[i];
1049                 num_tx_fail_total += slave_nb_pkts[i] - num_tx_slave;
1050
1051                 /* If tx burst fails move packets to end of bufs */
1052                 if (unlikely(num_tx_slave < slave_nb_pkts[i])) {
1053                         uint16_t j = nb_pkts - num_tx_fail_total;
1054                         for ( ; num_tx_slave < slave_nb_pkts[i]; j++, num_tx_slave++)
1055                                 bufs[j] = slave_bufs[i][num_tx_slave];
1056                 }
1057         }
1058
1059         return num_tx_total;
1060 }
1061
1062 static uint16_t
1063 bond_ethdev_tx_burst_broadcast(void *queue, struct rte_mbuf **bufs,
1064                 uint16_t nb_pkts)
1065 {
1066         struct bond_dev_private *internals;
1067         struct bond_tx_queue *bd_tx_q;
1068
1069         uint8_t tx_failed_flag = 0, num_of_slaves;
1070         uint8_t slaves[RTE_MAX_ETHPORTS];
1071
1072         uint16_t max_nb_of_tx_pkts = 0;
1073
1074         int slave_tx_total[RTE_MAX_ETHPORTS];
1075         int i, most_successful_tx_slave = -1;
1076
1077         bd_tx_q = (struct bond_tx_queue *)queue;
1078         internals = bd_tx_q->dev_private;
1079
1080         /* Copy slave list to protect against slave up/down changes during tx
1081          * bursting */
1082         num_of_slaves = internals->active_slave_count;
1083         memcpy(slaves, internals->active_slaves,
1084                         sizeof(internals->active_slaves[0]) * num_of_slaves);
1085
1086         if (num_of_slaves < 1)
1087                 return 0;
1088
1089         /* Increment reference count on mbufs */
1090         for (i = 0; i < nb_pkts; i++)
1091                 rte_mbuf_refcnt_update(bufs[i], num_of_slaves - 1);
1092
1093         /* Transmit burst on each active slave */
1094         for (i = 0; i < num_of_slaves; i++) {
1095                 slave_tx_total[i] = rte_eth_tx_burst(slaves[i], bd_tx_q->queue_id,
1096                                         bufs, nb_pkts);
1097
1098                 if (unlikely(slave_tx_total[i] < nb_pkts))
1099                         tx_failed_flag = 1;
1100
1101                 /* record the value and slave index for the slave which transmits the
1102                  * maximum number of packets */
1103                 if (slave_tx_total[i] > max_nb_of_tx_pkts) {
1104                         max_nb_of_tx_pkts = slave_tx_total[i];
1105                         most_successful_tx_slave = i;
1106                 }
1107         }
1108
1109         /* if slaves fail to transmit packets from burst, the calling application
1110          * is not expected to know about multiple references to packets so we must
1111          * handle failures of all packets except those of the most successful slave
1112          */
1113         if (unlikely(tx_failed_flag))
1114                 for (i = 0; i < num_of_slaves; i++)
1115                         if (i != most_successful_tx_slave)
1116                                 while (slave_tx_total[i] < nb_pkts)
1117                                         rte_pktmbuf_free(bufs[slave_tx_total[i]++]);
1118
1119         return max_nb_of_tx_pkts;
1120 }
1121
1122 void
1123 link_properties_set(struct rte_eth_dev *bonded_eth_dev,
1124                 struct rte_eth_link *slave_dev_link)
1125 {
1126         struct rte_eth_link *bonded_dev_link = &bonded_eth_dev->data->dev_link;
1127         struct bond_dev_private *internals = bonded_eth_dev->data->dev_private;
1128
1129         if (slave_dev_link->link_status &&
1130                 bonded_eth_dev->data->dev_started) {
1131                 bonded_dev_link->link_duplex = slave_dev_link->link_duplex;
1132                 bonded_dev_link->link_speed = slave_dev_link->link_speed;
1133
1134                 internals->link_props_set = 1;
1135         }
1136 }
1137
1138 void
1139 link_properties_reset(struct rte_eth_dev *bonded_eth_dev)
1140 {
1141         struct bond_dev_private *internals = bonded_eth_dev->data->dev_private;
1142
1143         memset(&(bonded_eth_dev->data->dev_link), 0,
1144                         sizeof(bonded_eth_dev->data->dev_link));
1145
1146         internals->link_props_set = 0;
1147 }
1148
1149 int
1150 link_properties_valid(struct rte_eth_link *bonded_dev_link,
1151                 struct rte_eth_link *slave_dev_link)
1152 {
1153         if (bonded_dev_link->link_duplex != slave_dev_link->link_duplex ||
1154                 bonded_dev_link->link_speed !=  slave_dev_link->link_speed)
1155                 return -1;
1156
1157         return 0;
1158 }
1159
1160 int
1161 mac_address_get(struct rte_eth_dev *eth_dev, struct ether_addr *dst_mac_addr)
1162 {
1163         struct ether_addr *mac_addr;
1164
1165         if (eth_dev == NULL) {
1166                 RTE_LOG(ERR, PMD, "%s: NULL pointer eth_dev specified\n", __func__);
1167                 return -1;
1168         }
1169
1170         if (dst_mac_addr == NULL) {
1171                 RTE_LOG(ERR, PMD, "%s: NULL pointer MAC specified\n", __func__);
1172                 return -1;
1173         }
1174
1175         mac_addr = eth_dev->data->mac_addrs;
1176
1177         ether_addr_copy(mac_addr, dst_mac_addr);
1178         return 0;
1179 }
1180
1181 int
1182 mac_address_set(struct rte_eth_dev *eth_dev, struct ether_addr *new_mac_addr)
1183 {
1184         struct ether_addr *mac_addr;
1185
1186         if (eth_dev == NULL) {
1187                 RTE_BOND_LOG(ERR, "NULL pointer eth_dev specified");
1188                 return -1;
1189         }
1190
1191         if (new_mac_addr == NULL) {
1192                 RTE_BOND_LOG(ERR, "NULL pointer MAC specified");
1193                 return -1;
1194         }
1195
1196         mac_addr = eth_dev->data->mac_addrs;
1197
1198         /* If new MAC is different to current MAC then update */
1199         if (memcmp(mac_addr, new_mac_addr, sizeof(*mac_addr)) != 0)
1200                 memcpy(mac_addr, new_mac_addr, sizeof(*mac_addr));
1201
1202         return 0;
1203 }
1204
1205 int
1206 mac_address_slaves_update(struct rte_eth_dev *bonded_eth_dev)
1207 {
1208         struct bond_dev_private *internals = bonded_eth_dev->data->dev_private;
1209         int i;
1210
1211         /* Update slave devices MAC addresses */
1212         if (internals->slave_count < 1)
1213                 return -1;
1214
1215         switch (internals->mode) {
1216         case BONDING_MODE_ROUND_ROBIN:
1217         case BONDING_MODE_BALANCE:
1218         case BONDING_MODE_BROADCAST:
1219                 for (i = 0; i < internals->slave_count; i++) {
1220                         if (mac_address_set(&rte_eth_devices[internals->slaves[i].port_id],
1221                                         bonded_eth_dev->data->mac_addrs)) {
1222                                 RTE_BOND_LOG(ERR, "Failed to update port Id %d MAC address",
1223                                                 internals->slaves[i].port_id);
1224                                 return -1;
1225                         }
1226                 }
1227                 break;
1228         case BONDING_MODE_8023AD:
1229                 bond_mode_8023ad_mac_address_update(bonded_eth_dev);
1230                 break;
1231         case BONDING_MODE_ACTIVE_BACKUP:
1232         case BONDING_MODE_TLB:
1233         case BONDING_MODE_ALB:
1234         default:
1235                 for (i = 0; i < internals->slave_count; i++) {
1236                         if (internals->slaves[i].port_id ==
1237                                         internals->current_primary_port) {
1238                                 if (mac_address_set(&rte_eth_devices[internals->primary_port],
1239                                                 bonded_eth_dev->data->mac_addrs)) {
1240                                         RTE_BOND_LOG(ERR, "Failed to update port Id %d MAC address",
1241                                                         internals->current_primary_port);
1242                                         return -1;
1243                                 }
1244                         } else {
1245                                 if (mac_address_set(
1246                                                 &rte_eth_devices[internals->slaves[i].port_id],
1247                                                 &internals->slaves[i].persisted_mac_addr)) {
1248                                         RTE_BOND_LOG(ERR, "Failed to update port Id %d MAC address",
1249                                                         internals->slaves[i].port_id);
1250                                         return -1;
1251                                 }
1252                         }
1253                 }
1254         }
1255
1256         return 0;
1257 }
1258
1259 int
1260 bond_ethdev_mode_set(struct rte_eth_dev *eth_dev, int mode)
1261 {
1262         struct bond_dev_private *internals;
1263
1264         internals = eth_dev->data->dev_private;
1265
1266         switch (mode) {
1267         case BONDING_MODE_ROUND_ROBIN:
1268                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_round_robin;
1269                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst;
1270                 break;
1271         case BONDING_MODE_ACTIVE_BACKUP:
1272                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_active_backup;
1273                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_active_backup;
1274                 break;
1275         case BONDING_MODE_BALANCE:
1276                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_balance;
1277                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst;
1278                 break;
1279         case BONDING_MODE_BROADCAST:
1280                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_broadcast;
1281                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst;
1282                 break;
1283         case BONDING_MODE_8023AD:
1284                 if (bond_mode_8023ad_enable(eth_dev) != 0)
1285                         return -1;
1286
1287                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_8023ad;
1288                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_8023ad;
1289                 RTE_LOG(WARNING, PMD,
1290                                 "Using mode 4, it is necessary to do TX burst and RX burst "
1291                                 "at least every 100ms.\n");
1292                 break;
1293         case BONDING_MODE_TLB:
1294                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_tlb;
1295                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_active_backup;
1296                 break;
1297         case BONDING_MODE_ALB:
1298                 if (bond_mode_alb_enable(eth_dev) != 0)
1299                         return -1;
1300
1301                 eth_dev->tx_pkt_burst = bond_ethdev_tx_burst_alb;
1302                 eth_dev->rx_pkt_burst = bond_ethdev_rx_burst_alb;
1303                 break;
1304         default:
1305                 return -1;
1306         }
1307
1308         internals->mode = mode;
1309
1310         return 0;
1311 }
1312
1313 int
1314 slave_configure(struct rte_eth_dev *bonded_eth_dev,
1315                 struct rte_eth_dev *slave_eth_dev)
1316 {
1317         struct bond_rx_queue *bd_rx_q;
1318         struct bond_tx_queue *bd_tx_q;
1319
1320         int errval;
1321         uint16_t q_id;
1322
1323         /* Stop slave */
1324         rte_eth_dev_stop(slave_eth_dev->data->port_id);
1325
1326         /* Enable interrupts on slave device if supported */
1327         if (slave_eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1328                 slave_eth_dev->data->dev_conf.intr_conf.lsc = 1;
1329
1330         /* If RSS is enabled for bonding, try to enable it for slaves  */
1331         if (bonded_eth_dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG) {
1332                 if (bonded_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key_len
1333                                 != 0) {
1334                         slave_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key_len =
1335                                         bonded_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key_len;
1336                         slave_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key =
1337                                         bonded_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key;
1338                 } else {
1339                         slave_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key = NULL;
1340                 }
1341
1342                 slave_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf =
1343                                 bonded_eth_dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf;
1344                 slave_eth_dev->data->dev_conf.rxmode.mq_mode =
1345                                 bonded_eth_dev->data->dev_conf.rxmode.mq_mode;
1346         }
1347
1348         slave_eth_dev->data->dev_conf.rxmode.hw_vlan_filter =
1349                         bonded_eth_dev->data->dev_conf.rxmode.hw_vlan_filter;
1350
1351         /* Configure device */
1352         errval = rte_eth_dev_configure(slave_eth_dev->data->port_id,
1353                         bonded_eth_dev->data->nb_rx_queues,
1354                         bonded_eth_dev->data->nb_tx_queues,
1355                         &(slave_eth_dev->data->dev_conf));
1356         if (errval != 0) {
1357                 RTE_BOND_LOG(ERR, "Cannot configure slave device: port %u , err (%d)",
1358                                 slave_eth_dev->data->port_id, errval);
1359                 return errval;
1360         }
1361
1362         /* Setup Rx Queues */
1363         for (q_id = 0; q_id < bonded_eth_dev->data->nb_rx_queues; q_id++) {
1364                 bd_rx_q = (struct bond_rx_queue *)bonded_eth_dev->data->rx_queues[q_id];
1365
1366                 errval = rte_eth_rx_queue_setup(slave_eth_dev->data->port_id, q_id,
1367                                 bd_rx_q->nb_rx_desc,
1368                                 rte_eth_dev_socket_id(slave_eth_dev->data->port_id),
1369                                 &(bd_rx_q->rx_conf), bd_rx_q->mb_pool);
1370                 if (errval != 0) {
1371                         RTE_BOND_LOG(ERR,
1372                                         "rte_eth_rx_queue_setup: port=%d queue_id %d, err (%d)",
1373                                         slave_eth_dev->data->port_id, q_id, errval);
1374                         return errval;
1375                 }
1376         }
1377
1378         /* Setup Tx Queues */
1379         for (q_id = 0; q_id < bonded_eth_dev->data->nb_tx_queues; q_id++) {
1380                 bd_tx_q = (struct bond_tx_queue *)bonded_eth_dev->data->tx_queues[q_id];
1381
1382                 errval = rte_eth_tx_queue_setup(slave_eth_dev->data->port_id, q_id,
1383                                 bd_tx_q->nb_tx_desc,
1384                                 rte_eth_dev_socket_id(slave_eth_dev->data->port_id),
1385                                 &bd_tx_q->tx_conf);
1386                 if (errval != 0) {
1387                         RTE_BOND_LOG(ERR,
1388                                         "rte_eth_tx_queue_setup: port=%d queue_id %d, err (%d)",
1389                                         slave_eth_dev->data->port_id, q_id, errval);
1390                         return errval;
1391                 }
1392         }
1393
1394         /* Start device */
1395         errval = rte_eth_dev_start(slave_eth_dev->data->port_id);
1396         if (errval != 0) {
1397                 RTE_BOND_LOG(ERR, "rte_eth_dev_start: port=%u, err (%d)",
1398                                 slave_eth_dev->data->port_id, errval);
1399                 return -1;
1400         }
1401
1402         /* If RSS is enabled for bonding, synchronize RETA */
1403         if (bonded_eth_dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS) {
1404                 int i;
1405                 struct bond_dev_private *internals;
1406
1407                 internals = bonded_eth_dev->data->dev_private;
1408
1409                 for (i = 0; i < internals->slave_count; i++) {
1410                         if (internals->slaves[i].port_id == slave_eth_dev->data->port_id) {
1411                                 errval = rte_eth_dev_rss_reta_update(
1412                                                 slave_eth_dev->data->port_id,
1413                                                 &internals->reta_conf[0],
1414                                                 internals->slaves[i].reta_size);
1415                                 if (errval != 0) {
1416                                         RTE_LOG(WARNING, PMD,
1417                                                         "rte_eth_dev_rss_reta_update on slave port %d fails (err %d)."
1418                                                         " RSS Configuration for bonding may be inconsistent.\n",
1419                                                         slave_eth_dev->data->port_id, errval);
1420                                 }
1421                                 break;
1422                         }
1423                 }
1424         }
1425
1426         /* If lsc interrupt is set, check initial slave's link status */
1427         if (slave_eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1428                 bond_ethdev_lsc_event_callback(slave_eth_dev->data->port_id,
1429                         RTE_ETH_EVENT_INTR_LSC, &bonded_eth_dev->data->port_id);
1430
1431         return 0;
1432 }
1433
1434 void
1435 slave_remove(struct bond_dev_private *internals,
1436                 struct rte_eth_dev *slave_eth_dev)
1437 {
1438         uint8_t i;
1439
1440         for (i = 0; i < internals->slave_count; i++)
1441                 if (internals->slaves[i].port_id ==
1442                                 slave_eth_dev->data->port_id)
1443                         break;
1444
1445         if (i < (internals->slave_count - 1))
1446                 memmove(&internals->slaves[i], &internals->slaves[i + 1],
1447                                 sizeof(internals->slaves[0]) *
1448                                 (internals->slave_count - i - 1));
1449
1450         internals->slave_count--;
1451
1452         /* force reconfiguration of slave interfaces */
1453         _rte_eth_dev_reset(slave_eth_dev);
1454 }
1455
1456 static void
1457 bond_ethdev_slave_link_status_change_monitor(void *cb_arg);
1458
1459 void
1460 slave_add(struct bond_dev_private *internals,
1461                 struct rte_eth_dev *slave_eth_dev)
1462 {
1463         struct bond_slave_details *slave_details =
1464                         &internals->slaves[internals->slave_count];
1465
1466         slave_details->port_id = slave_eth_dev->data->port_id;
1467         slave_details->last_link_status = 0;
1468
1469         /* Mark slave devices that don't support interrupts so we can
1470          * compensate when we start the bond
1471          */
1472         if (!(slave_eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
1473                 slave_details->link_status_poll_enabled = 1;
1474         }
1475
1476         slave_details->link_status_wait_to_complete = 0;
1477         /* clean tlb_last_obytes when adding port for bonding device */
1478         memcpy(&(slave_details->persisted_mac_addr), slave_eth_dev->data->mac_addrs,
1479                         sizeof(struct ether_addr));
1480 }
1481
1482 void
1483 bond_ethdev_primary_set(struct bond_dev_private *internals,
1484                 uint8_t slave_port_id)
1485 {
1486         int i;
1487
1488         if (internals->active_slave_count < 1)
1489                 internals->current_primary_port = slave_port_id;
1490         else
1491                 /* Search bonded device slave ports for new proposed primary port */
1492                 for (i = 0; i < internals->active_slave_count; i++) {
1493                         if (internals->active_slaves[i] == slave_port_id)
1494                                 internals->current_primary_port = slave_port_id;
1495                 }
1496 }
1497
1498 static void
1499 bond_ethdev_promiscuous_enable(struct rte_eth_dev *eth_dev);
1500
1501 static int
1502 bond_ethdev_start(struct rte_eth_dev *eth_dev)
1503 {
1504         struct bond_dev_private *internals;
1505         int i;
1506
1507         /* slave eth dev will be started by bonded device */
1508         if (check_for_bonded_ethdev(eth_dev)) {
1509                 RTE_BOND_LOG(ERR, "User tried to explicitly start a slave eth_dev (%d)",
1510                                 eth_dev->data->port_id);
1511                 return -1;
1512         }
1513
1514         eth_dev->data->dev_link.link_status = ETH_LINK_DOWN;
1515         eth_dev->data->dev_started = 1;
1516
1517         internals = eth_dev->data->dev_private;
1518
1519         if (internals->slave_count == 0) {
1520                 RTE_BOND_LOG(ERR, "Cannot start port since there are no slave devices");
1521                 return -1;
1522         }
1523
1524         if (internals->user_defined_mac == 0) {
1525                 struct ether_addr *new_mac_addr = NULL;
1526
1527                 for (i = 0; i < internals->slave_count; i++)
1528                         if (internals->slaves[i].port_id == internals->primary_port)
1529                                 new_mac_addr = &internals->slaves[i].persisted_mac_addr;
1530
1531                 if (new_mac_addr == NULL)
1532                         return -1;
1533
1534                 if (mac_address_set(eth_dev, new_mac_addr) != 0) {
1535                         RTE_BOND_LOG(ERR, "bonded port (%d) failed to update MAC address",
1536                                         eth_dev->data->port_id);
1537                         return -1;
1538                 }
1539         }
1540
1541         /* Update all slave devices MACs*/
1542         if (mac_address_slaves_update(eth_dev) != 0)
1543                 return -1;
1544
1545         /* If bonded device is configure in promiscuous mode then re-apply config */
1546         if (internals->promiscuous_en)
1547                 bond_ethdev_promiscuous_enable(eth_dev);
1548
1549         /* Reconfigure each slave device if starting bonded device */
1550         for (i = 0; i < internals->slave_count; i++) {
1551                 if (slave_configure(eth_dev,
1552                                 &(rte_eth_devices[internals->slaves[i].port_id])) != 0) {
1553                         RTE_BOND_LOG(ERR,
1554                                         "bonded port (%d) failed to reconfigure slave device (%d)",
1555                                         eth_dev->data->port_id, internals->slaves[i].port_id);
1556                         return -1;
1557                 }
1558                 /* We will need to poll for link status if any slave doesn't
1559                  * support interrupts
1560                  */
1561                 if (internals->slaves[i].link_status_poll_enabled)
1562                         internals->link_status_polling_enabled = 1;
1563         }
1564         /* start polling if needed */
1565         if (internals->link_status_polling_enabled) {
1566                 rte_eal_alarm_set(
1567                         internals->link_status_polling_interval_ms * 1000,
1568                         bond_ethdev_slave_link_status_change_monitor,
1569                         (void *)&rte_eth_devices[internals->port_id]);
1570         }
1571
1572         if (internals->user_defined_primary_port)
1573                 bond_ethdev_primary_set(internals, internals->primary_port);
1574
1575         if (internals->mode == BONDING_MODE_8023AD)
1576                 bond_mode_8023ad_start(eth_dev);
1577
1578         if (internals->mode == BONDING_MODE_TLB ||
1579                         internals->mode == BONDING_MODE_ALB)
1580                 bond_tlb_enable(internals);
1581
1582         return 0;
1583 }
1584
1585 static void
1586 bond_ethdev_free_queues(struct rte_eth_dev *dev)
1587 {
1588         uint8_t i;
1589
1590         if (dev->data->rx_queues != NULL) {
1591                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1592                         rte_free(dev->data->rx_queues[i]);
1593                         dev->data->rx_queues[i] = NULL;
1594                 }
1595                 dev->data->nb_rx_queues = 0;
1596         }
1597
1598         if (dev->data->tx_queues != NULL) {
1599                 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1600                         rte_free(dev->data->tx_queues[i]);
1601                         dev->data->tx_queues[i] = NULL;
1602                 }
1603                 dev->data->nb_tx_queues = 0;
1604         }
1605 }
1606
1607 void
1608 bond_ethdev_stop(struct rte_eth_dev *eth_dev)
1609 {
1610         struct bond_dev_private *internals = eth_dev->data->dev_private;
1611         uint8_t i;
1612
1613         if (internals->mode == BONDING_MODE_8023AD) {
1614                 struct port *port;
1615                 void *pkt = NULL;
1616
1617                 bond_mode_8023ad_stop(eth_dev);
1618
1619                 /* Discard all messages to/from mode 4 state machines */
1620                 for (i = 0; i < internals->active_slave_count; i++) {
1621                         port = &mode_8023ad_ports[internals->active_slaves[i]];
1622
1623                         RTE_ASSERT(port->rx_ring != NULL);
1624                         while (rte_ring_dequeue(port->rx_ring, &pkt) != -ENOENT)
1625                                 rte_pktmbuf_free(pkt);
1626
1627                         RTE_ASSERT(port->tx_ring != NULL);
1628                         while (rte_ring_dequeue(port->tx_ring, &pkt) != -ENOENT)
1629                                 rte_pktmbuf_free(pkt);
1630                 }
1631         }
1632
1633         if (internals->mode == BONDING_MODE_TLB ||
1634                         internals->mode == BONDING_MODE_ALB) {
1635                 bond_tlb_disable(internals);
1636                 for (i = 0; i < internals->active_slave_count; i++)
1637                         tlb_last_obytets[internals->active_slaves[i]] = 0;
1638         }
1639
1640         internals->active_slave_count = 0;
1641         internals->link_status_polling_enabled = 0;
1642         for (i = 0; i < internals->slave_count; i++)
1643                 internals->slaves[i].last_link_status = 0;
1644
1645         eth_dev->data->dev_link.link_status = ETH_LINK_DOWN;
1646         eth_dev->data->dev_started = 0;
1647 }
1648
1649 void
1650 bond_ethdev_close(struct rte_eth_dev *dev)
1651 {
1652         struct bond_dev_private *internals = dev->data->dev_private;
1653
1654         bond_ethdev_free_queues(dev);
1655         rte_bitmap_reset(internals->vlan_filter_bmp);
1656 }
1657
1658 /* forward declaration */
1659 static int bond_ethdev_configure(struct rte_eth_dev *dev);
1660
1661 static void
1662 bond_ethdev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1663 {
1664         struct bond_dev_private *internals = dev->data->dev_private;
1665
1666         dev_info->max_mac_addrs = 1;
1667
1668         dev_info->max_rx_pktlen = internals->candidate_max_rx_pktlen ?
1669                                   internals->candidate_max_rx_pktlen : 2048;
1670
1671         dev_info->max_rx_queues = (uint16_t)128;
1672         dev_info->max_tx_queues = (uint16_t)512;
1673
1674         dev_info->min_rx_bufsize = 0;
1675
1676         dev_info->rx_offload_capa = internals->rx_offload_capa;
1677         dev_info->tx_offload_capa = internals->tx_offload_capa;
1678         dev_info->flow_type_rss_offloads = internals->flow_type_rss_offloads;
1679
1680         dev_info->reta_size = internals->reta_size;
1681 }
1682
1683 static int
1684 bond_ethdev_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1685 {
1686         int res;
1687         uint8_t i;
1688         struct bond_dev_private *internals = dev->data->dev_private;
1689
1690         /* don't do this while a slave is being added */
1691         rte_spinlock_lock(&internals->lock);
1692
1693         if (on)
1694                 rte_bitmap_set(internals->vlan_filter_bmp, vlan_id);
1695         else
1696                 rte_bitmap_clear(internals->vlan_filter_bmp, vlan_id);
1697
1698         for (i = 0; i < internals->slave_count; i++) {
1699                 uint8_t port_id = internals->slaves[i].port_id;
1700
1701                 res = rte_eth_dev_vlan_filter(port_id, vlan_id, on);
1702                 if (res == ENOTSUP)
1703                         RTE_LOG(WARNING, PMD,
1704                                 "Setting VLAN filter on slave port %u not supported.\n",
1705                                 port_id);
1706         }
1707
1708         rte_spinlock_unlock(&internals->lock);
1709         return 0;
1710 }
1711
1712 static int
1713 bond_ethdev_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1714                 uint16_t nb_rx_desc, unsigned int socket_id __rte_unused,
1715                 const struct rte_eth_rxconf *rx_conf, struct rte_mempool *mb_pool)
1716 {
1717         struct bond_rx_queue *bd_rx_q = (struct bond_rx_queue *)
1718                         rte_zmalloc_socket(NULL, sizeof(struct bond_rx_queue),
1719                                         0, dev->data->numa_node);
1720         if (bd_rx_q == NULL)
1721                 return -1;
1722
1723         bd_rx_q->queue_id = rx_queue_id;
1724         bd_rx_q->dev_private = dev->data->dev_private;
1725
1726         bd_rx_q->nb_rx_desc = nb_rx_desc;
1727
1728         memcpy(&(bd_rx_q->rx_conf), rx_conf, sizeof(struct rte_eth_rxconf));
1729         bd_rx_q->mb_pool = mb_pool;
1730
1731         dev->data->rx_queues[rx_queue_id] = bd_rx_q;
1732
1733         return 0;
1734 }
1735
1736 static int
1737 bond_ethdev_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1738                 uint16_t nb_tx_desc, unsigned int socket_id __rte_unused,
1739                 const struct rte_eth_txconf *tx_conf)
1740 {
1741         struct bond_tx_queue *bd_tx_q  = (struct bond_tx_queue *)
1742                         rte_zmalloc_socket(NULL, sizeof(struct bond_tx_queue),
1743                                         0, dev->data->numa_node);
1744
1745         if (bd_tx_q == NULL)
1746                 return -1;
1747
1748         bd_tx_q->queue_id = tx_queue_id;
1749         bd_tx_q->dev_private = dev->data->dev_private;
1750
1751         bd_tx_q->nb_tx_desc = nb_tx_desc;
1752         memcpy(&(bd_tx_q->tx_conf), tx_conf, sizeof(bd_tx_q->tx_conf));
1753
1754         dev->data->tx_queues[tx_queue_id] = bd_tx_q;
1755
1756         return 0;
1757 }
1758
1759 static void
1760 bond_ethdev_rx_queue_release(void *queue)
1761 {
1762         if (queue == NULL)
1763                 return;
1764
1765         rte_free(queue);
1766 }
1767
1768 static void
1769 bond_ethdev_tx_queue_release(void *queue)
1770 {
1771         if (queue == NULL)
1772                 return;
1773
1774         rte_free(queue);
1775 }
1776
1777 static void
1778 bond_ethdev_slave_link_status_change_monitor(void *cb_arg)
1779 {
1780         struct rte_eth_dev *bonded_ethdev, *slave_ethdev;
1781         struct bond_dev_private *internals;
1782
1783         /* Default value for polling slave found is true as we don't want to
1784          * disable the polling thread if we cannot get the lock */
1785         int i, polling_slave_found = 1;
1786
1787         if (cb_arg == NULL)
1788                 return;
1789
1790         bonded_ethdev = (struct rte_eth_dev *)cb_arg;
1791         internals = (struct bond_dev_private *)bonded_ethdev->data->dev_private;
1792
1793         if (!bonded_ethdev->data->dev_started ||
1794                 !internals->link_status_polling_enabled)
1795                 return;
1796
1797         /* If device is currently being configured then don't check slaves link
1798          * status, wait until next period */
1799         if (rte_spinlock_trylock(&internals->lock)) {
1800                 if (internals->slave_count > 0)
1801                         polling_slave_found = 0;
1802
1803                 for (i = 0; i < internals->slave_count; i++) {
1804                         if (!internals->slaves[i].link_status_poll_enabled)
1805                                 continue;
1806
1807                         slave_ethdev = &rte_eth_devices[internals->slaves[i].port_id];
1808                         polling_slave_found = 1;
1809
1810                         /* Update slave link status */
1811                         (*slave_ethdev->dev_ops->link_update)(slave_ethdev,
1812                                         internals->slaves[i].link_status_wait_to_complete);
1813
1814                         /* if link status has changed since last checked then call lsc
1815                          * event callback */
1816                         if (slave_ethdev->data->dev_link.link_status !=
1817                                         internals->slaves[i].last_link_status) {
1818                                 internals->slaves[i].last_link_status =
1819                                                 slave_ethdev->data->dev_link.link_status;
1820
1821                                 bond_ethdev_lsc_event_callback(internals->slaves[i].port_id,
1822                                                 RTE_ETH_EVENT_INTR_LSC,
1823                                                 &bonded_ethdev->data->port_id);
1824                         }
1825                 }
1826                 rte_spinlock_unlock(&internals->lock);
1827         }
1828
1829         if (polling_slave_found)
1830                 /* Set alarm to continue monitoring link status of slave ethdev's */
1831                 rte_eal_alarm_set(internals->link_status_polling_interval_ms * 1000,
1832                                 bond_ethdev_slave_link_status_change_monitor, cb_arg);
1833 }
1834
1835 static int
1836 bond_ethdev_link_update(struct rte_eth_dev *bonded_eth_dev,
1837                 int wait_to_complete)
1838 {
1839         struct bond_dev_private *internals = bonded_eth_dev->data->dev_private;
1840
1841         if (!bonded_eth_dev->data->dev_started ||
1842                 internals->active_slave_count == 0) {
1843                 bonded_eth_dev->data->dev_link.link_status = ETH_LINK_DOWN;
1844                 return 0;
1845         } else {
1846                 struct rte_eth_dev *slave_eth_dev;
1847                 int i, link_up = 0;
1848
1849                 for (i = 0; i < internals->active_slave_count; i++) {
1850                         slave_eth_dev = &rte_eth_devices[internals->active_slaves[i]];
1851
1852                         (*slave_eth_dev->dev_ops->link_update)(slave_eth_dev,
1853                                         wait_to_complete);
1854                         if (slave_eth_dev->data->dev_link.link_status == ETH_LINK_UP) {
1855                                 link_up = 1;
1856                                 break;
1857                         }
1858                 }
1859
1860                 bonded_eth_dev->data->dev_link.link_status = link_up;
1861         }
1862
1863         return 0;
1864 }
1865
1866 static void
1867 bond_ethdev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1868 {
1869         struct bond_dev_private *internals = dev->data->dev_private;
1870         struct rte_eth_stats slave_stats;
1871         int i, j;
1872
1873         for (i = 0; i < internals->slave_count; i++) {
1874                 rte_eth_stats_get(internals->slaves[i].port_id, &slave_stats);
1875
1876                 stats->ipackets += slave_stats.ipackets;
1877                 stats->opackets += slave_stats.opackets;
1878                 stats->ibytes += slave_stats.ibytes;
1879                 stats->obytes += slave_stats.obytes;
1880                 stats->imissed += slave_stats.imissed;
1881                 stats->ierrors += slave_stats.ierrors;
1882                 stats->oerrors += slave_stats.oerrors;
1883                 stats->rx_nombuf += slave_stats.rx_nombuf;
1884
1885                 for (j = 0; j < RTE_ETHDEV_QUEUE_STAT_CNTRS; j++) {
1886                         stats->q_ipackets[j] += slave_stats.q_ipackets[j];
1887                         stats->q_opackets[j] += slave_stats.q_opackets[j];
1888                         stats->q_ibytes[j] += slave_stats.q_ibytes[j];
1889                         stats->q_obytes[j] += slave_stats.q_obytes[j];
1890                         stats->q_errors[j] += slave_stats.q_errors[j];
1891                 }
1892
1893         }
1894 }
1895
1896 static void
1897 bond_ethdev_stats_reset(struct rte_eth_dev *dev)
1898 {
1899         struct bond_dev_private *internals = dev->data->dev_private;
1900         int i;
1901
1902         for (i = 0; i < internals->slave_count; i++)
1903                 rte_eth_stats_reset(internals->slaves[i].port_id);
1904 }
1905
1906 static void
1907 bond_ethdev_promiscuous_enable(struct rte_eth_dev *eth_dev)
1908 {
1909         struct bond_dev_private *internals = eth_dev->data->dev_private;
1910         int i;
1911
1912         internals->promiscuous_en = 1;
1913
1914         switch (internals->mode) {
1915         /* Promiscuous mode is propagated to all slaves */
1916         case BONDING_MODE_ROUND_ROBIN:
1917         case BONDING_MODE_BALANCE:
1918         case BONDING_MODE_BROADCAST:
1919                 for (i = 0; i < internals->slave_count; i++)
1920                         rte_eth_promiscuous_enable(internals->slaves[i].port_id);
1921                 break;
1922         /* In mode4 promiscus mode is managed when slave is added/removed */
1923         case BONDING_MODE_8023AD:
1924                 break;
1925         /* Promiscuous mode is propagated only to primary slave */
1926         case BONDING_MODE_ACTIVE_BACKUP:
1927         case BONDING_MODE_TLB:
1928         case BONDING_MODE_ALB:
1929         default:
1930                 rte_eth_promiscuous_enable(internals->current_primary_port);
1931         }
1932 }
1933
1934 static void
1935 bond_ethdev_promiscuous_disable(struct rte_eth_dev *dev)
1936 {
1937         struct bond_dev_private *internals = dev->data->dev_private;
1938         int i;
1939
1940         internals->promiscuous_en = 0;
1941
1942         switch (internals->mode) {
1943         /* Promiscuous mode is propagated to all slaves */
1944         case BONDING_MODE_ROUND_ROBIN:
1945         case BONDING_MODE_BALANCE:
1946         case BONDING_MODE_BROADCAST:
1947                 for (i = 0; i < internals->slave_count; i++)
1948                         rte_eth_promiscuous_disable(internals->slaves[i].port_id);
1949                 break;
1950         /* In mode4 promiscus mode is set managed when slave is added/removed */
1951         case BONDING_MODE_8023AD:
1952                 break;
1953         /* Promiscuous mode is propagated only to primary slave */
1954         case BONDING_MODE_ACTIVE_BACKUP:
1955         case BONDING_MODE_TLB:
1956         case BONDING_MODE_ALB:
1957         default:
1958                 rte_eth_promiscuous_disable(internals->current_primary_port);
1959         }
1960 }
1961
1962 static void
1963 bond_ethdev_delayed_lsc_propagation(void *arg)
1964 {
1965         if (arg == NULL)
1966                 return;
1967
1968         _rte_eth_dev_callback_process((struct rte_eth_dev *)arg,
1969                         RTE_ETH_EVENT_INTR_LSC, NULL);
1970 }
1971
1972 void
1973 bond_ethdev_lsc_event_callback(uint8_t port_id, enum rte_eth_event_type type,
1974                 void *param)
1975 {
1976         struct rte_eth_dev *bonded_eth_dev, *slave_eth_dev;
1977         struct bond_dev_private *internals;
1978         struct rte_eth_link link;
1979
1980         int i, valid_slave = 0;
1981         uint8_t active_pos;
1982         uint8_t lsc_flag = 0;
1983
1984         if (type != RTE_ETH_EVENT_INTR_LSC || param == NULL)
1985                 return;
1986
1987         bonded_eth_dev = &rte_eth_devices[*(uint8_t *)param];
1988         slave_eth_dev = &rte_eth_devices[port_id];
1989
1990         if (check_for_bonded_ethdev(bonded_eth_dev))
1991                 return;
1992
1993         internals = bonded_eth_dev->data->dev_private;
1994
1995         /* If the device isn't started don't handle interrupts */
1996         if (!bonded_eth_dev->data->dev_started)
1997                 return;
1998
1999         /* verify that port_id is a valid slave of bonded port */
2000         for (i = 0; i < internals->slave_count; i++) {
2001                 if (internals->slaves[i].port_id == port_id) {
2002                         valid_slave = 1;
2003                         break;
2004                 }
2005         }
2006
2007         if (!valid_slave)
2008                 return;
2009
2010         /* Search for port in active port list */
2011         active_pos = find_slave_by_id(internals->active_slaves,
2012                         internals->active_slave_count, port_id);
2013
2014         rte_eth_link_get_nowait(port_id, &link);
2015         if (link.link_status) {
2016                 if (active_pos < internals->active_slave_count)
2017                         return;
2018
2019                 /* if no active slave ports then set this port to be primary port */
2020                 if (internals->active_slave_count < 1) {
2021                         /* If first active slave, then change link status */
2022                         bonded_eth_dev->data->dev_link.link_status = ETH_LINK_UP;
2023                         internals->current_primary_port = port_id;
2024                         lsc_flag = 1;
2025
2026                         mac_address_slaves_update(bonded_eth_dev);
2027
2028                         /* Inherit eth dev link properties from first active slave */
2029                         link_properties_set(bonded_eth_dev,
2030                                         &(slave_eth_dev->data->dev_link));
2031                 } else {
2032                         if (link_properties_valid(
2033                                 &bonded_eth_dev->data->dev_link, &link) != 0) {
2034                                 slave_eth_dev->data->dev_flags &=
2035                                         (~RTE_ETH_DEV_BONDED_SLAVE);
2036                                 RTE_LOG(ERR, PMD,
2037                                         "port %u invalid speed/duplex\n",
2038                                         port_id);
2039                                 return;
2040                         }
2041                 }
2042
2043                 activate_slave(bonded_eth_dev, port_id);
2044
2045                 /* If user has defined the primary port then default to using it */
2046                 if (internals->user_defined_primary_port &&
2047                                 internals->primary_port == port_id)
2048                         bond_ethdev_primary_set(internals, port_id);
2049         } else {
2050                 if (active_pos == internals->active_slave_count)
2051                         return;
2052
2053                 /* Remove from active slave list */
2054                 deactivate_slave(bonded_eth_dev, port_id);
2055
2056                 /* No active slaves, change link status to down and reset other
2057                  * link properties */
2058                 if (internals->active_slave_count < 1) {
2059                         lsc_flag = 1;
2060                         bonded_eth_dev->data->dev_link.link_status = ETH_LINK_DOWN;
2061
2062                         link_properties_reset(bonded_eth_dev);
2063                 }
2064
2065                 /* Update primary id, take first active slave from list or if none
2066                  * available set to -1 */
2067                 if (port_id == internals->current_primary_port) {
2068                         if (internals->active_slave_count > 0)
2069                                 bond_ethdev_primary_set(internals,
2070                                                 internals->active_slaves[0]);
2071                         else
2072                                 internals->current_primary_port = internals->primary_port;
2073                 }
2074         }
2075
2076         if (lsc_flag) {
2077                 /* Cancel any possible outstanding interrupts if delays are enabled */
2078                 if (internals->link_up_delay_ms > 0 ||
2079                         internals->link_down_delay_ms > 0)
2080                         rte_eal_alarm_cancel(bond_ethdev_delayed_lsc_propagation,
2081                                         bonded_eth_dev);
2082
2083                 if (bonded_eth_dev->data->dev_link.link_status) {
2084                         if (internals->link_up_delay_ms > 0)
2085                                 rte_eal_alarm_set(internals->link_up_delay_ms * 1000,
2086                                                 bond_ethdev_delayed_lsc_propagation,
2087                                                 (void *)bonded_eth_dev);
2088                         else
2089                                 _rte_eth_dev_callback_process(bonded_eth_dev,
2090                                                 RTE_ETH_EVENT_INTR_LSC, NULL);
2091
2092                 } else {
2093                         if (internals->link_down_delay_ms > 0)
2094                                 rte_eal_alarm_set(internals->link_down_delay_ms * 1000,
2095                                                 bond_ethdev_delayed_lsc_propagation,
2096                                                 (void *)bonded_eth_dev);
2097                         else
2098                                 _rte_eth_dev_callback_process(bonded_eth_dev,
2099                                                 RTE_ETH_EVENT_INTR_LSC, NULL);
2100                 }
2101         }
2102 }
2103
2104 static int
2105 bond_ethdev_rss_reta_update(struct rte_eth_dev *dev,
2106                 struct rte_eth_rss_reta_entry64 *reta_conf, uint16_t reta_size)
2107 {
2108         unsigned i, j;
2109         int result = 0;
2110         int slave_reta_size;
2111         unsigned reta_count;
2112         struct bond_dev_private *internals = dev->data->dev_private;
2113
2114         if (reta_size != internals->reta_size)
2115                 return -EINVAL;
2116
2117          /* Copy RETA table */
2118         reta_count = reta_size / RTE_RETA_GROUP_SIZE;
2119
2120         for (i = 0; i < reta_count; i++) {
2121                 internals->reta_conf[i].mask = reta_conf[i].mask;
2122                 for (j = 0; j < RTE_RETA_GROUP_SIZE; j++)
2123                         if ((reta_conf[i].mask >> j) & 0x01)
2124                                 internals->reta_conf[i].reta[j] = reta_conf[i].reta[j];
2125         }
2126
2127         /* Fill rest of array */
2128         for (; i < RTE_DIM(internals->reta_conf); i += reta_count)
2129                 memcpy(&internals->reta_conf[i], &internals->reta_conf[0],
2130                                 sizeof(internals->reta_conf[0]) * reta_count);
2131
2132         /* Propagate RETA over slaves */
2133         for (i = 0; i < internals->slave_count; i++) {
2134                 slave_reta_size = internals->slaves[i].reta_size;
2135                 result = rte_eth_dev_rss_reta_update(internals->slaves[i].port_id,
2136                                 &internals->reta_conf[0], slave_reta_size);
2137                 if (result < 0)
2138                         return result;
2139         }
2140
2141         return 0;
2142 }
2143
2144 static int
2145 bond_ethdev_rss_reta_query(struct rte_eth_dev *dev,
2146                 struct rte_eth_rss_reta_entry64 *reta_conf, uint16_t reta_size)
2147 {
2148         int i, j;
2149         struct bond_dev_private *internals = dev->data->dev_private;
2150
2151         if (reta_size != internals->reta_size)
2152                 return -EINVAL;
2153
2154          /* Copy RETA table */
2155         for (i = 0; i < reta_size / RTE_RETA_GROUP_SIZE; i++)
2156                 for (j = 0; j < RTE_RETA_GROUP_SIZE; j++)
2157                         if ((reta_conf[i].mask >> j) & 0x01)
2158                                 reta_conf[i].reta[j] = internals->reta_conf[i].reta[j];
2159
2160         return 0;
2161 }
2162
2163 static int
2164 bond_ethdev_rss_hash_update(struct rte_eth_dev *dev,
2165                 struct rte_eth_rss_conf *rss_conf)
2166 {
2167         int i, result = 0;
2168         struct bond_dev_private *internals = dev->data->dev_private;
2169         struct rte_eth_rss_conf bond_rss_conf;
2170
2171         memcpy(&bond_rss_conf, rss_conf, sizeof(struct rte_eth_rss_conf));
2172
2173         bond_rss_conf.rss_hf &= internals->flow_type_rss_offloads;
2174
2175         if (bond_rss_conf.rss_hf != 0)
2176                 dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf = bond_rss_conf.rss_hf;
2177
2178         if (bond_rss_conf.rss_key && bond_rss_conf.rss_key_len <
2179                         sizeof(internals->rss_key)) {
2180                 if (bond_rss_conf.rss_key_len == 0)
2181                         bond_rss_conf.rss_key_len = 40;
2182                 internals->rss_key_len = bond_rss_conf.rss_key_len;
2183                 memcpy(internals->rss_key, bond_rss_conf.rss_key,
2184                                 internals->rss_key_len);
2185         }
2186
2187         for (i = 0; i < internals->slave_count; i++) {
2188                 result = rte_eth_dev_rss_hash_update(internals->slaves[i].port_id,
2189                                 &bond_rss_conf);
2190                 if (result < 0)
2191                         return result;
2192         }
2193
2194         return 0;
2195 }
2196
2197 static int
2198 bond_ethdev_rss_hash_conf_get(struct rte_eth_dev *dev,
2199                 struct rte_eth_rss_conf *rss_conf)
2200 {
2201         struct bond_dev_private *internals = dev->data->dev_private;
2202
2203         rss_conf->rss_hf = dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf;
2204         rss_conf->rss_key_len = internals->rss_key_len;
2205         if (rss_conf->rss_key)
2206                 memcpy(rss_conf->rss_key, internals->rss_key, internals->rss_key_len);
2207
2208         return 0;
2209 }
2210
2211 const struct eth_dev_ops default_dev_ops = {
2212         .dev_start            = bond_ethdev_start,
2213         .dev_stop             = bond_ethdev_stop,
2214         .dev_close            = bond_ethdev_close,
2215         .dev_configure        = bond_ethdev_configure,
2216         .dev_infos_get        = bond_ethdev_info,
2217         .vlan_filter_set      = bond_ethdev_vlan_filter_set,
2218         .rx_queue_setup       = bond_ethdev_rx_queue_setup,
2219         .tx_queue_setup       = bond_ethdev_tx_queue_setup,
2220         .rx_queue_release     = bond_ethdev_rx_queue_release,
2221         .tx_queue_release     = bond_ethdev_tx_queue_release,
2222         .link_update          = bond_ethdev_link_update,
2223         .stats_get            = bond_ethdev_stats_get,
2224         .stats_reset          = bond_ethdev_stats_reset,
2225         .promiscuous_enable   = bond_ethdev_promiscuous_enable,
2226         .promiscuous_disable  = bond_ethdev_promiscuous_disable,
2227         .reta_update          = bond_ethdev_rss_reta_update,
2228         .reta_query           = bond_ethdev_rss_reta_query,
2229         .rss_hash_update      = bond_ethdev_rss_hash_update,
2230         .rss_hash_conf_get    = bond_ethdev_rss_hash_conf_get
2231 };
2232
2233 static int
2234 bond_probe(const char *name, const char *params)
2235 {
2236         struct bond_dev_private *internals;
2237         struct rte_kvargs *kvlist;
2238         uint8_t bonding_mode, socket_id;
2239         int  arg_count, port_id;
2240
2241         RTE_LOG(INFO, EAL, "Initializing pmd_bond for %s\n", name);
2242
2243         kvlist = rte_kvargs_parse(params, pmd_bond_init_valid_arguments);
2244         if (kvlist == NULL)
2245                 return -1;
2246
2247         /* Parse link bonding mode */
2248         if (rte_kvargs_count(kvlist, PMD_BOND_MODE_KVARG) == 1) {
2249                 if (rte_kvargs_process(kvlist, PMD_BOND_MODE_KVARG,
2250                                 &bond_ethdev_parse_slave_mode_kvarg,
2251                                 &bonding_mode) != 0) {
2252                         RTE_LOG(ERR, EAL, "Invalid mode for bonded device %s\n",
2253                                         name);
2254                         goto parse_error;
2255                 }
2256         } else {
2257                 RTE_LOG(ERR, EAL, "Mode must be specified only once for bonded "
2258                                 "device %s\n", name);
2259                 goto parse_error;
2260         }
2261
2262         /* Parse socket id to create bonding device on */
2263         arg_count = rte_kvargs_count(kvlist, PMD_BOND_SOCKET_ID_KVARG);
2264         if (arg_count == 1) {
2265                 if (rte_kvargs_process(kvlist, PMD_BOND_SOCKET_ID_KVARG,
2266                                 &bond_ethdev_parse_socket_id_kvarg, &socket_id)
2267                                 != 0) {
2268                         RTE_LOG(ERR, EAL, "Invalid socket Id specified for "
2269                                         "bonded device %s\n", name);
2270                         goto parse_error;
2271                 }
2272         } else if (arg_count > 1) {
2273                 RTE_LOG(ERR, EAL, "Socket Id can be specified only once for "
2274                                 "bonded device %s\n", name);
2275                 goto parse_error;
2276         } else {
2277                 socket_id = rte_socket_id();
2278         }
2279
2280         /* Create link bonding eth device */
2281         port_id = rte_eth_bond_create(name, bonding_mode, socket_id);
2282         if (port_id < 0) {
2283                 RTE_LOG(ERR, EAL, "Failed to create socket %s in mode %u on "
2284                                 "socket %u.\n", name, bonding_mode, socket_id);
2285                 goto parse_error;
2286         }
2287         internals = rte_eth_devices[port_id].data->dev_private;
2288         internals->kvlist = kvlist;
2289
2290         RTE_LOG(INFO, EAL, "Create bonded device %s on port %d in mode %u on "
2291                         "socket %u.\n", name, port_id, bonding_mode, socket_id);
2292         return 0;
2293
2294 parse_error:
2295         rte_kvargs_free(kvlist);
2296
2297         return -1;
2298 }
2299
2300 static int
2301 bond_remove(const char *name)
2302 {
2303         int  ret;
2304
2305         if (name == NULL)
2306                 return -EINVAL;
2307
2308         RTE_LOG(INFO, EAL, "Uninitializing pmd_bond for %s\n", name);
2309
2310         /* free link bonding eth device */
2311         ret = rte_eth_bond_free(name);
2312         if (ret < 0)
2313                 RTE_LOG(ERR, EAL, "Failed to free %s\n", name);
2314
2315         return ret;
2316 }
2317
2318 /* this part will resolve the slave portids after all the other pdev and vdev
2319  * have been allocated */
2320 static int
2321 bond_ethdev_configure(struct rte_eth_dev *dev)
2322 {
2323         char *name = dev->data->name;
2324         struct bond_dev_private *internals = dev->data->dev_private;
2325         struct rte_kvargs *kvlist = internals->kvlist;
2326         int arg_count;
2327         uint8_t port_id = dev - rte_eth_devices;
2328
2329         static const uint8_t default_rss_key[40] = {
2330                 0x6D, 0x5A, 0x56, 0xDA, 0x25, 0x5B, 0x0E, 0xC2, 0x41, 0x67, 0x25, 0x3D,
2331                 0x43, 0xA3, 0x8F, 0xB0, 0xD0, 0xCA, 0x2B, 0xCB, 0xAE, 0x7B, 0x30, 0xB4,
2332                 0x77, 0xCB, 0x2D, 0xA3, 0x80, 0x30, 0xF2, 0x0C, 0x6A, 0x42, 0xB7, 0x3B,
2333                 0xBE, 0xAC, 0x01, 0xFA
2334         };
2335
2336         unsigned i, j;
2337
2338         /* If RSS is enabled, fill table and key with default values */
2339         if (dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS) {
2340                 dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key = internals->rss_key;
2341                 dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key_len = 0;
2342                 memcpy(internals->rss_key, default_rss_key, 40);
2343
2344                 for (i = 0; i < RTE_DIM(internals->reta_conf); i++) {
2345                         internals->reta_conf[i].mask = ~0LL;
2346                         for (j = 0; j < RTE_RETA_GROUP_SIZE; j++)
2347                                 internals->reta_conf[i].reta[j] = j % dev->data->nb_rx_queues;
2348                 }
2349         }
2350
2351         /* set the max_rx_pktlen */
2352         internals->max_rx_pktlen = internals->candidate_max_rx_pktlen;
2353
2354         /*
2355          * if no kvlist, it means that this bonded device has been created
2356          * through the bonding api.
2357          */
2358         if (!kvlist)
2359                 return 0;
2360
2361         /* Parse MAC address for bonded device */
2362         arg_count = rte_kvargs_count(kvlist, PMD_BOND_MAC_ADDR_KVARG);
2363         if (arg_count == 1) {
2364                 struct ether_addr bond_mac;
2365
2366                 if (rte_kvargs_process(kvlist, PMD_BOND_MAC_ADDR_KVARG,
2367                                 &bond_ethdev_parse_bond_mac_addr_kvarg, &bond_mac) < 0) {
2368                         RTE_LOG(INFO, EAL, "Invalid mac address for bonded device %s\n",
2369                                         name);
2370                         return -1;
2371                 }
2372
2373                 /* Set MAC address */
2374                 if (rte_eth_bond_mac_address_set(port_id, &bond_mac) != 0) {
2375                         RTE_LOG(ERR, EAL,
2376                                         "Failed to set mac address on bonded device %s\n",
2377                                         name);
2378                         return -1;
2379                 }
2380         } else if (arg_count > 1) {
2381                 RTE_LOG(ERR, EAL,
2382                                 "MAC address can be specified only once for bonded device %s\n",
2383                                 name);
2384                 return -1;
2385         }
2386
2387         /* Parse/set balance mode transmit policy */
2388         arg_count = rte_kvargs_count(kvlist, PMD_BOND_XMIT_POLICY_KVARG);
2389         if (arg_count == 1) {
2390                 uint8_t xmit_policy;
2391
2392                 if (rte_kvargs_process(kvlist, PMD_BOND_XMIT_POLICY_KVARG,
2393                                 &bond_ethdev_parse_balance_xmit_policy_kvarg, &xmit_policy) !=
2394                                                 0) {
2395                         RTE_LOG(INFO, EAL,
2396                                         "Invalid xmit policy specified for bonded device %s\n",
2397                                         name);
2398                         return -1;
2399                 }
2400
2401                 /* Set balance mode transmit policy*/
2402                 if (rte_eth_bond_xmit_policy_set(port_id, xmit_policy) != 0) {
2403                         RTE_LOG(ERR, EAL,
2404                                         "Failed to set balance xmit policy on bonded device %s\n",
2405                                         name);
2406                         return -1;
2407                 }
2408         } else if (arg_count > 1) {
2409                 RTE_LOG(ERR, EAL,
2410                                 "Transmit policy can be specified only once for bonded device"
2411                                 " %s\n", name);
2412                 return -1;
2413         }
2414
2415         /* Parse/add slave ports to bonded device */
2416         if (rte_kvargs_count(kvlist, PMD_BOND_SLAVE_PORT_KVARG) > 0) {
2417                 struct bond_ethdev_slave_ports slave_ports;
2418                 unsigned i;
2419
2420                 memset(&slave_ports, 0, sizeof(slave_ports));
2421
2422                 if (rte_kvargs_process(kvlist, PMD_BOND_SLAVE_PORT_KVARG,
2423                                 &bond_ethdev_parse_slave_port_kvarg, &slave_ports) != 0) {
2424                         RTE_LOG(ERR, EAL,
2425                                         "Failed to parse slave ports for bonded device %s\n",
2426                                         name);
2427                         return -1;
2428                 }
2429
2430                 for (i = 0; i < slave_ports.slave_count; i++) {
2431                         if (rte_eth_bond_slave_add(port_id, slave_ports.slaves[i]) != 0) {
2432                                 RTE_LOG(ERR, EAL,
2433                                                 "Failed to add port %d as slave to bonded device %s\n",
2434                                                 slave_ports.slaves[i], name);
2435                         }
2436                 }
2437
2438         } else {
2439                 RTE_LOG(INFO, EAL, "No slaves specified for bonded device %s\n", name);
2440                 return -1;
2441         }
2442
2443         /* Parse/set primary slave port id*/
2444         arg_count = rte_kvargs_count(kvlist, PMD_BOND_PRIMARY_SLAVE_KVARG);
2445         if (arg_count == 1) {
2446                 uint8_t primary_slave_port_id;
2447
2448                 if (rte_kvargs_process(kvlist,
2449                                 PMD_BOND_PRIMARY_SLAVE_KVARG,
2450                                 &bond_ethdev_parse_primary_slave_port_id_kvarg,
2451                                 &primary_slave_port_id) < 0) {
2452                         RTE_LOG(INFO, EAL,
2453                                         "Invalid primary slave port id specified for bonded device"
2454                                         " %s\n", name);
2455                         return -1;
2456                 }
2457
2458                 /* Set balance mode transmit policy*/
2459                 if (rte_eth_bond_primary_set(port_id, (uint8_t)primary_slave_port_id)
2460                                 != 0) {
2461                         RTE_LOG(ERR, EAL,
2462                                         "Failed to set primary slave port %d on bonded device %s\n",
2463                                         primary_slave_port_id, name);
2464                         return -1;
2465                 }
2466         } else if (arg_count > 1) {
2467                 RTE_LOG(INFO, EAL,
2468                                 "Primary slave can be specified only once for bonded device"
2469                                 " %s\n", name);
2470                 return -1;
2471         }
2472
2473         /* Parse link status monitor polling interval */
2474         arg_count = rte_kvargs_count(kvlist, PMD_BOND_LSC_POLL_PERIOD_KVARG);
2475         if (arg_count == 1) {
2476                 uint32_t lsc_poll_interval_ms;
2477
2478                 if (rte_kvargs_process(kvlist,
2479                                 PMD_BOND_LSC_POLL_PERIOD_KVARG,
2480                                 &bond_ethdev_parse_time_ms_kvarg,
2481                                 &lsc_poll_interval_ms) < 0) {
2482                         RTE_LOG(INFO, EAL,
2483                                         "Invalid lsc polling interval value specified for bonded"
2484                                         " device %s\n", name);
2485                         return -1;
2486                 }
2487
2488                 if (rte_eth_bond_link_monitoring_set(port_id, lsc_poll_interval_ms)
2489                                 != 0) {
2490                         RTE_LOG(ERR, EAL,
2491                                         "Failed to set lsc monitor polling interval (%u ms) on"
2492                                         " bonded device %s\n", lsc_poll_interval_ms, name);
2493                         return -1;
2494                 }
2495         } else if (arg_count > 1) {
2496                 RTE_LOG(INFO, EAL,
2497                                 "LSC polling interval can be specified only once for bonded"
2498                                 " device %s\n", name);
2499                 return -1;
2500         }
2501
2502         /* Parse link up interrupt propagation delay */
2503         arg_count = rte_kvargs_count(kvlist, PMD_BOND_LINK_UP_PROP_DELAY_KVARG);
2504         if (arg_count == 1) {
2505                 uint32_t link_up_delay_ms;
2506
2507                 if (rte_kvargs_process(kvlist,
2508                                 PMD_BOND_LINK_UP_PROP_DELAY_KVARG,
2509                                 &bond_ethdev_parse_time_ms_kvarg,
2510                                 &link_up_delay_ms) < 0) {
2511                         RTE_LOG(INFO, EAL,
2512                                         "Invalid link up propagation delay value specified for"
2513                                         " bonded device %s\n", name);
2514                         return -1;
2515                 }
2516
2517                 /* Set balance mode transmit policy*/
2518                 if (rte_eth_bond_link_up_prop_delay_set(port_id, link_up_delay_ms)
2519                                 != 0) {
2520                         RTE_LOG(ERR, EAL,
2521                                         "Failed to set link up propagation delay (%u ms) on bonded"
2522                                         " device %s\n", link_up_delay_ms, name);
2523                         return -1;
2524                 }
2525         } else if (arg_count > 1) {
2526                 RTE_LOG(INFO, EAL,
2527                                 "Link up propagation delay can be specified only once for"
2528                                 " bonded device %s\n", name);
2529                 return -1;
2530         }
2531
2532         /* Parse link down interrupt propagation delay */
2533         arg_count = rte_kvargs_count(kvlist, PMD_BOND_LINK_DOWN_PROP_DELAY_KVARG);
2534         if (arg_count == 1) {
2535                 uint32_t link_down_delay_ms;
2536
2537                 if (rte_kvargs_process(kvlist,
2538                                 PMD_BOND_LINK_DOWN_PROP_DELAY_KVARG,
2539                                 &bond_ethdev_parse_time_ms_kvarg,
2540                                 &link_down_delay_ms) < 0) {
2541                         RTE_LOG(INFO, EAL,
2542                                         "Invalid link down propagation delay value specified for"
2543                                         " bonded device %s\n", name);
2544                         return -1;
2545                 }
2546
2547                 /* Set balance mode transmit policy*/
2548                 if (rte_eth_bond_link_down_prop_delay_set(port_id, link_down_delay_ms)
2549                                 != 0) {
2550                         RTE_LOG(ERR, EAL,
2551                                         "Failed to set link down propagation delay (%u ms) on"
2552                                         " bonded device %s\n", link_down_delay_ms, name);
2553                         return -1;
2554                 }
2555         } else if (arg_count > 1) {
2556                 RTE_LOG(INFO, EAL,
2557                                 "Link down propagation delay can be specified only once for"
2558                                 " bonded device %s\n", name);
2559                 return -1;
2560         }
2561
2562         return 0;
2563 }
2564
2565 static struct rte_vdev_driver bond_drv = {
2566         .probe = bond_probe,
2567         .remove = bond_remove,
2568 };
2569
2570 RTE_PMD_REGISTER_VDEV(net_bonding, bond_drv);
2571 RTE_PMD_REGISTER_ALIAS(net_bonding, eth_bond);
2572
2573 RTE_PMD_REGISTER_PARAM_STRING(net_bonding,
2574         "slave=<ifc> "
2575         "primary=<ifc> "
2576         "mode=[0-6] "
2577         "xmit_policy=[l2 | l23 | l34] "
2578         "socket_id=<int> "
2579         "mac=<mac addr> "
2580         "lsc_poll_period_ms=<int> "
2581         "up_delay=<int> "
2582         "down_delay=<int>");