lib/librte_pmd_af_packet/rte_eth_af_packet.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2014 John W. Linville <linville@tuxdriver.com>
   5  *
   6  *   Originally based upon librte_pmd_pcap code:
   7  *
   8  *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
   9  *   Copyright(c) 2014 6WIND S.A.
  10  *   All rights reserved.
  11  *
  12  *   Redistribution and use in source and binary forms, with or without
  13  *   modification, are permitted provided that the following conditions
  14  *   are met:
  15  *
  16  *     * Redistributions of source code must retain the above copyright
  17  *       notice, this list of conditions and the following disclaimer.
  18  *     * Redistributions in binary form must reproduce the above copyright
  19  *       notice, this list of conditions and the following disclaimer in
  20  *       the documentation and/or other materials provided with the
  21  *       distribution.
  22  *     * Neither the name of Intel Corporation nor the names of its
  23  *       contributors may be used to endorse or promote products derived
  24  *       from this software without specific prior written permission.
  25  *
  26  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  27  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  28  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  29  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  30  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  31  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  32  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  33  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  34  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  35  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  36  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  37  */
  38
  39 #include <rte_mbuf.h>
  40 #include <rte_ethdev.h>
  41 #include <rte_malloc.h>
  42 #include <rte_kvargs.h>
  43 #include <rte_dev.h>
  44
  45 #include <linux/if_ether.h>
  46 #include <linux/if_packet.h>
  47 #include <arpa/inet.h>
  48 #include <net/if.h>
  49 #include <sys/types.h>
  50 #include <sys/socket.h>
  51 #include <sys/ioctl.h>
  52 #include <sys/mman.h>
  53 #include <unistd.h>
  54 #include <poll.h>
  55
  56 #include "rte_eth_af_packet.h"
  57
  58 #define ETH_AF_PACKET_IFACE_ARG         "iface"
  59 #define ETH_AF_PACKET_NUM_Q_ARG         "qpairs"
  60 #define ETH_AF_PACKET_BLOCKSIZE_ARG     "blocksz"
  61 #define ETH_AF_PACKET_FRAMESIZE_ARG     "framesz"
  62 #define ETH_AF_PACKET_FRAMECOUNT_ARG    "framecnt"
  63
  64 #define DFLT_BLOCK_SIZE         (1 << 12)
  65 #define DFLT_FRAME_SIZE         (1 << 11)
  66 #define DFLT_FRAME_COUNT        (1 << 9)
  67
  68 struct pkt_rx_queue {
  69         int sockfd;
  70
  71         struct iovec *rd;
  72         uint8_t *map;
  73         unsigned int framecount;
  74         unsigned int framenum;
  75
  76         struct rte_mempool *mb_pool;
  77
  78         volatile unsigned long rx_pkts;
  79         volatile unsigned long err_pkts;
  80 };
  81
  82 struct pkt_tx_queue {
  83         int sockfd;
  84
  85         struct iovec *rd;
  86         uint8_t *map;
  87         unsigned int framecount;
  88         unsigned int framenum;
  89
  90         volatile unsigned long tx_pkts;
  91         volatile unsigned long err_pkts;
  92 };
  93
  94 struct pmd_internals {
  95         unsigned nb_queues;
  96
  97         int if_index;
  98         struct ether_addr eth_addr;
  99
 100         struct tpacket_req req;
 101
 102         struct pkt_rx_queue rx_queue[RTE_PMD_AF_PACKET_MAX_RINGS];
 103         struct pkt_tx_queue tx_queue[RTE_PMD_AF_PACKET_MAX_RINGS];
 104 };
 105
 106 static const char *valid_arguments[] = {
 107         ETH_AF_PACKET_IFACE_ARG,
 108         ETH_AF_PACKET_NUM_Q_ARG,
 109         ETH_AF_PACKET_BLOCKSIZE_ARG,
 110         ETH_AF_PACKET_FRAMESIZE_ARG,
 111         ETH_AF_PACKET_FRAMECOUNT_ARG,
 112         NULL
 113 };
 114
 115 static const char *drivername = "AF_PACKET PMD";
 116
 117 static struct rte_eth_link pmd_link = {
 118         .link_speed = 10000,
 119         .link_duplex = ETH_LINK_FULL_DUPLEX,
 120         .link_status = 0
 121 };
 122
 123 static uint16_t
 124 eth_af_packet_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 125 {
 126         unsigned i;
 127         struct tpacket2_hdr *ppd;
 128         struct rte_mbuf *mbuf;
 129         uint8_t *pbuf;
 130         struct pkt_rx_queue *pkt_q = queue;
 131         uint16_t num_rx = 0;
 132         unsigned int framecount, framenum;
 133
 134         if (unlikely(nb_pkts == 0))
 135                 return 0;
 136
 137         /*
 138          * Reads the given number of packets from the AF_PACKET socket one by
 139          * one and copies the packet data into a newly allocated mbuf.
 140          */
 141         framecount = pkt_q->framecount;
 142         framenum = pkt_q->framenum;
 143         for (i = 0; i < nb_pkts; i++) {
 144                 /* point at the next incoming frame */
 145                 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
 146                 if ((ppd->tp_status & TP_STATUS_USER) == 0)
 147                         break;
 148
 149                 /* allocate the next mbuf */
 150                 mbuf = rte_pktmbuf_alloc(pkt_q->mb_pool);
 151                 if (unlikely(mbuf == NULL))
 152                         break;
 153
 154                 /* packet will fit in the mbuf, go ahead and receive it */
 155                 rte_pktmbuf_pkt_len(mbuf) = rte_pktmbuf_data_len(mbuf) = ppd->tp_snaplen;
 156                 pbuf = (uint8_t *) ppd + ppd->tp_mac;
 157                 memcpy(rte_pktmbuf_mtod(mbuf, void *), pbuf, rte_pktmbuf_data_len(mbuf));
 158
 159                 /* release incoming frame and advance ring buffer */
 160                 ppd->tp_status = TP_STATUS_KERNEL;
 161                 if (++framenum >= framecount)
 162                         framenum = 0;
 163
 164                 /* account for the receive frame */
 165                 bufs[i] = mbuf;
 166                 num_rx++;
 167         }
 168         pkt_q->framenum = framenum;
 169         pkt_q->rx_pkts += num_rx;
 170         return num_rx;
 171 }
 172
 173 /*
 174  * Callback to handle sending packets through a real NIC.
 175  */
 176 static uint16_t
 177 eth_af_packet_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 178 {
 179         struct tpacket2_hdr *ppd;
 180         struct rte_mbuf *mbuf;
 181         uint8_t *pbuf;
 182         unsigned int framecount, framenum;
 183         struct pollfd pfd;
 184         struct pkt_tx_queue *pkt_q = queue;
 185         uint16_t num_tx = 0;
 186         int i;
 187
 188         if (unlikely(nb_pkts == 0))
 189                 return 0;
 190
 191         memset(&pfd, 0, sizeof(pfd));
 192         pfd.fd = pkt_q->sockfd;
 193         pfd.events = POLLOUT;
 194         pfd.revents = 0;
 195
 196         framecount = pkt_q->framecount;
 197         framenum = pkt_q->framenum;
 198         ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
 199         for (i = 0; i < nb_pkts; i++) {
 200                 /* point at the next incoming frame */
 201                 if ((ppd->tp_status != TP_STATUS_AVAILABLE) &&
 202                     (poll(&pfd, 1, -1) < 0))
 203                                 continue;
 204
 205                 /* copy the tx frame data */
 206                 mbuf = bufs[num_tx];
 207                 pbuf = (uint8_t *) ppd + TPACKET2_HDRLEN -
 208                         sizeof(struct sockaddr_ll);
 209                 memcpy(pbuf, rte_pktmbuf_mtod(mbuf, void*), rte_pktmbuf_data_len(mbuf));
 210                 ppd->tp_len = ppd->tp_snaplen = rte_pktmbuf_data_len(mbuf);
 211
 212                 /* release incoming frame and advance ring buffer */
 213                 ppd->tp_status = TP_STATUS_SEND_REQUEST;
 214                 if (++framenum >= framecount)
 215                         framenum = 0;
 216                 ppd = (struct tpacket2_hdr *) pkt_q->rd[framenum].iov_base;
 217
 218                 num_tx++;
 219                 rte_pktmbuf_free(mbuf);
 220         }
 221
 222         /* kick-off transmits */
 223         sendto(pkt_q->sockfd, NULL, 0, MSG_DONTWAIT, NULL, 0);
 224
 225         pkt_q->framenum = framenum;
 226         pkt_q->tx_pkts += num_tx;
 227         pkt_q->err_pkts += nb_pkts - num_tx;
 228         return num_tx;
 229 }
 230
 231 static int
 232 eth_dev_start(struct rte_eth_dev *dev)
 233 {
 234         dev->data->dev_link.link_status = 1;
 235         return 0;
 236 }
 237
 238 /*
 239  * This function gets called when the current port gets stopped.
 240  */
 241 static void
 242 eth_dev_stop(struct rte_eth_dev *dev)
 243 {
 244         unsigned i;
 245         int sockfd;
 246         struct pmd_internals *internals = dev->data->dev_private;
 247
 248         for (i = 0; i < internals->nb_queues; i++) {
 249                 sockfd = internals->rx_queue[i].sockfd;
 250                 if (sockfd != -1)
 251                         close(sockfd);
 252                 sockfd = internals->tx_queue[i].sockfd;
 253                 if (sockfd != -1)
 254                         close(sockfd);
 255         }
 256
 257         dev->data->dev_link.link_status = 0;
 258 }
 259
 260 static int
 261 eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
 262 {
 263         return 0;
 264 }
 265
 266 static void
 267 eth_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 268 {
 269         struct pmd_internals *internals = dev->data->dev_private;
 270
 271         dev_info->driver_name = drivername;
 272         dev_info->if_index = internals->if_index;
 273         dev_info->max_mac_addrs = 1;
 274         dev_info->max_rx_pktlen = (uint32_t)ETH_FRAME_LEN;
 275         dev_info->max_rx_queues = (uint16_t)internals->nb_queues;
 276         dev_info->max_tx_queues = (uint16_t)internals->nb_queues;
 277         dev_info->min_rx_bufsize = 0;
 278         dev_info->pci_dev = NULL;
 279 }
 280
 281 static void
 282 eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *igb_stats)
 283 {
 284         unsigned i, imax;
 285         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 286         const struct pmd_internals *internal = dev->data->dev_private;
 287
 288         memset(igb_stats, 0, sizeof(*igb_stats));
 289
 290         imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
 291                 internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
 292         for (i = 0; i < imax; i++) {
 293                 igb_stats->q_ipackets[i] = internal->rx_queue[i].rx_pkts;
 294                 rx_total += igb_stats->q_ipackets[i];
 295         }
 296
 297         imax = (internal->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS ?
 298                 internal->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS);
 299         for (i = 0; i < imax; i++) {
 300                 igb_stats->q_opackets[i] = internal->tx_queue[i].tx_pkts;
 301                 igb_stats->q_errors[i] = internal->tx_queue[i].err_pkts;
 302                 tx_total += igb_stats->q_opackets[i];
 303                 tx_err_total += igb_stats->q_errors[i];
 304         }
 305
 306         igb_stats->ipackets = rx_total;
 307         igb_stats->opackets = tx_total;
 308         igb_stats->oerrors = tx_err_total;
 309 }
 310
 311 static void
 312 eth_stats_reset(struct rte_eth_dev *dev)
 313 {
 314         unsigned i;
 315         struct pmd_internals *internal = dev->data->dev_private;
 316
 317         for (i = 0; i < internal->nb_queues; i++)
 318                 internal->rx_queue[i].rx_pkts = 0;
 319
 320         for (i = 0; i < internal->nb_queues; i++) {
 321                 internal->tx_queue[i].tx_pkts = 0;
 322                 internal->tx_queue[i].err_pkts = 0;
 323         }
 324 }
 325
 326 static void
 327 eth_dev_close(struct rte_eth_dev *dev __rte_unused)
 328 {
 329 }
 330
 331 static void
 332 eth_queue_release(void *q __rte_unused)
 333 {
 334 }
 335
 336 static int
 337 eth_link_update(struct rte_eth_dev *dev __rte_unused,
 338                 int wait_to_complete __rte_unused)
 339 {
 340         return 0;
 341 }
 342
 343 static int
 344 eth_rx_queue_setup(struct rte_eth_dev *dev,
 345                    uint16_t rx_queue_id,
 346                    uint16_t nb_rx_desc __rte_unused,
 347                    unsigned int socket_id __rte_unused,
 348                    const struct rte_eth_rxconf *rx_conf __rte_unused,
 349                    struct rte_mempool *mb_pool)
 350 {
 351         struct pmd_internals *internals = dev->data->dev_private;
 352         struct pkt_rx_queue *pkt_q = &internals->rx_queue[rx_queue_id];
 353         struct rte_pktmbuf_pool_private *mbp_priv;
 354         uint16_t buf_size;
 355
 356         pkt_q->mb_pool = mb_pool;
 357
 358         /* Now get the space available for data in the mbuf */
 359         mbp_priv = rte_mempool_get_priv(pkt_q->mb_pool);
 360         buf_size = (uint16_t) (mbp_priv->mbuf_data_room_size -
 361                                RTE_PKTMBUF_HEADROOM);
 362
 363         if (ETH_FRAME_LEN > buf_size) {
 364                 RTE_LOG(ERR, PMD,
 365                         "%s: %d bytes will not fit in mbuf (%d bytes)\n",
 366                         dev->data->name, ETH_FRAME_LEN, buf_size);
 367                 return -ENOMEM;
 368         }
 369
 370         dev->data->rx_queues[rx_queue_id] = pkt_q;
 371
 372         return 0;
 373 }
 374
 375 static int
 376 eth_tx_queue_setup(struct rte_eth_dev *dev,
 377                    uint16_t tx_queue_id,
 378                    uint16_t nb_tx_desc __rte_unused,
 379                    unsigned int socket_id __rte_unused,
 380                    const struct rte_eth_txconf *tx_conf __rte_unused)
 381 {
 382
 383         struct pmd_internals *internals = dev->data->dev_private;
 384
 385         dev->data->tx_queues[tx_queue_id] = &internals->tx_queue[tx_queue_id];
 386         return 0;
 387 }
 388
 389 static struct eth_dev_ops ops = {
 390         .dev_start = eth_dev_start,
 391         .dev_stop = eth_dev_stop,
 392         .dev_close = eth_dev_close,
 393         .dev_configure = eth_dev_configure,
 394         .dev_infos_get = eth_dev_info,
 395         .rx_queue_setup = eth_rx_queue_setup,
 396         .tx_queue_setup = eth_tx_queue_setup,
 397         .rx_queue_release = eth_queue_release,
 398         .tx_queue_release = eth_queue_release,
 399         .link_update = eth_link_update,
 400         .stats_get = eth_stats_get,
 401         .stats_reset = eth_stats_reset,
 402 };
 403
 404 /*
 405  * Opens an AF_PACKET socket
 406  */
 407 static int
 408 open_packet_iface(const char *key __rte_unused,
 409                   const char *value __rte_unused,
 410                   void *extra_args)
 411 {
 412         int *sockfd = extra_args;
 413
 414         /* Open an AF_PACKET socket... */
 415         *sockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
 416         if (*sockfd == -1) {
 417                 RTE_LOG(ERR, PMD, "Could not open AF_PACKET socket\n");
 418                 return -1;
 419         }
 420
 421         return 0;
 422 }
 423
 424 static int
 425 rte_pmd_init_internals(const char *name,
 426                        const int sockfd,
 427                        const unsigned nb_queues,
 428                        unsigned int blocksize,
 429                        unsigned int blockcnt,
 430                        unsigned int framesize,
 431                        unsigned int framecnt,
 432                        const unsigned numa_node,
 433                        struct pmd_internals **internals,
 434                        struct rte_eth_dev **eth_dev,
 435                        struct rte_kvargs *kvlist)
 436 {
 437         struct rte_eth_dev_data *data = NULL;
 438         struct rte_pci_device *pci_dev = NULL;
 439         struct rte_kvargs_pair *pair = NULL;
 440         struct ifreq ifr;
 441         size_t ifnamelen;
 442         unsigned k_idx;
 443         struct sockaddr_ll sockaddr;
 444         struct tpacket_req *req;
 445         struct pkt_rx_queue *rx_queue;
 446         struct pkt_tx_queue *tx_queue;
 447         int rc, tpver, discard, bypass;
 448         unsigned int i, q, rdsize;
 449         int qsockfd, fanout_arg;
 450
 451         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 452                 pair = &kvlist->pairs[k_idx];
 453                 if (strstr(pair->key, ETH_AF_PACKET_IFACE_ARG) != NULL)
 454                         break;
 455         }
 456         if (pair == NULL) {
 457                 RTE_LOG(ERR, PMD,
 458                         "%s: no interface specified for AF_PACKET ethdev\n",
 459                         name);
 460                 goto error;
 461         }
 462
 463         RTE_LOG(INFO, PMD,
 464                 "%s: creating AF_PACKET-backed ethdev on numa socket %u\n",
 465                 name, numa_node);
 466
 467         /*
 468          * now do all data allocation - for eth_dev structure, dummy pci driver
 469          * and internal (private) data
 470          */
 471         data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
 472         if (data == NULL)
 473                 goto error;
 474
 475         pci_dev = rte_zmalloc_socket(name, sizeof(*pci_dev), 0, numa_node);
 476         if (pci_dev == NULL)
 477                 goto error;
 478
 479         *internals = rte_zmalloc_socket(name, sizeof(**internals),
 480                                         0, numa_node);
 481         if (*internals == NULL)
 482                 goto error;
 483
 484         req = &((*internals)->req);
 485
 486         req->tp_block_size = blocksize;
 487         req->tp_block_nr = blockcnt;
 488         req->tp_frame_size = framesize;
 489         req->tp_frame_nr = framecnt;
 490
 491         ifnamelen = strlen(pair->value);
 492         if (ifnamelen < sizeof(ifr.ifr_name)) {
 493                 memcpy(ifr.ifr_name, pair->value, ifnamelen);
 494                 ifr.ifr_name[ifnamelen] = '\0';
 495         } else {
 496                 RTE_LOG(ERR, PMD,
 497                         "%s: I/F name too long (%s)\n",
 498                         name, pair->value);
 499                 goto error;
 500         }
 501         if (ioctl(sockfd, SIOCGIFINDEX, &ifr) == -1) {
 502                 RTE_LOG(ERR, PMD,
 503                         "%s: ioctl failed (SIOCGIFINDEX)\n",
 504                         name);
 505                 goto error;
 506         }
 507         (*internals)->if_index = ifr.ifr_ifindex;
 508
 509         if (ioctl(sockfd, SIOCGIFHWADDR, &ifr) == -1) {
 510                 RTE_LOG(ERR, PMD,
 511                         "%s: ioctl failed (SIOCGIFHWADDR)\n",
 512                         name);
 513                 goto error;
 514         }
 515         memcpy(&(*internals)->eth_addr, ifr.ifr_hwaddr.sa_data, ETH_ALEN);
 516
 517         memset(&sockaddr, 0, sizeof(sockaddr));
 518         sockaddr.sll_family = AF_PACKET;
 519         sockaddr.sll_protocol = htons(ETH_P_ALL);
 520         sockaddr.sll_ifindex = (*internals)->if_index;
 521
 522         fanout_arg = (getpid() ^ (*internals)->if_index) & 0xffff;
 523         fanout_arg |= (PACKET_FANOUT_HASH | PACKET_FANOUT_FLAG_DEFRAG |
 524                        PACKET_FANOUT_FLAG_ROLLOVER) << 16;
 525
 526         for (q = 0; q < nb_queues; q++) {
 527                 /* Open an AF_PACKET socket for this queue... */
 528                 qsockfd = socket(AF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
 529                 if (qsockfd == -1) {
 530                         RTE_LOG(ERR, PMD,
 531                                 "%s: could not open AF_PACKET socket\n",
 532                                 name);
 533                         return -1;
 534                 }
 535
 536                 tpver = TPACKET_V2;
 537                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_VERSION,
 538                                 &tpver, sizeof(tpver));
 539                 if (rc == -1) {
 540                         RTE_LOG(ERR, PMD,
 541                                 "%s: could not set PACKET_VERSION on AF_PACKET "
 542                                 "socket for %s\n", name, pair->value);
 543                         goto error;
 544                 }
 545
 546                 discard = 1;
 547                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_LOSS,
 548                                 &discard, sizeof(discard));
 549                 if (rc == -1) {
 550                         RTE_LOG(ERR, PMD,
 551                                 "%s: could not set PACKET_LOSS on "
 552                                 "AF_PACKET socket for %s\n", name, pair->value);
 553                         goto error;
 554                 }
 555
 556                 bypass = 1;
 557                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_QDISC_BYPASS,
 558                                 &bypass, sizeof(bypass));
 559                 if (rc == -1) {
 560                         RTE_LOG(ERR, PMD,
 561                                 "%s: could not set PACKET_QDISC_BYPASS "
 562                                 "on AF_PACKET socket for %s\n", name,
 563                                 pair->value);
 564                         goto error;
 565                 }
 566
 567                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_RX_RING, req, sizeof(*req));
 568                 if (rc == -1) {
 569                         RTE_LOG(ERR, PMD,
 570                                 "%s: could not set PACKET_RX_RING on AF_PACKET "
 571                                 "socket for %s\n", name, pair->value);
 572                         goto error;
 573                 }
 574
 575                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_TX_RING, req, sizeof(*req));
 576                 if (rc == -1) {
 577                         RTE_LOG(ERR, PMD,
 578                                 "%s: could not set PACKET_TX_RING on AF_PACKET "
 579                                 "socket for %s\n", name, pair->value);
 580                         goto error;
 581                 }
 582
 583                 rx_queue = &((*internals)->rx_queue[q]);
 584                 rx_queue->framecount = req->tp_frame_nr;
 585
 586                 rx_queue->map = mmap(NULL, 2 * req->tp_block_size * req->tp_block_nr,
 587                                     PROT_READ | PROT_WRITE, MAP_SHARED | MAP_LOCKED,
 588                                     qsockfd, 0);
 589                 if (rx_queue->map == MAP_FAILED) {
 590                         RTE_LOG(ERR, PMD,
 591                                 "%s: call to mmap failed on AF_PACKET socket for %s\n",
 592                                 name, pair->value);
 593                         goto error;
 594                 }
 595
 596                 /* rdsize is same for both Tx and Rx */
 597                 rdsize = req->tp_frame_nr * sizeof(*(rx_queue->rd));
 598
 599                 rx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
 600                 for (i = 0; i < req->tp_frame_nr; ++i) {
 601                         rx_queue->rd[i].iov_base = rx_queue->map + (i * framesize);
 602                         rx_queue->rd[i].iov_len = req->tp_frame_size;
 603                 }
 604                 rx_queue->sockfd = qsockfd;
 605
 606                 tx_queue = &((*internals)->tx_queue[q]);
 607                 tx_queue->framecount = req->tp_frame_nr;
 608
 609                 tx_queue->map = rx_queue->map + req->tp_block_size * req->tp_block_nr;
 610
 611                 tx_queue->rd = rte_zmalloc_socket(name, rdsize, 0, numa_node);
 612                 for (i = 0; i < req->tp_frame_nr; ++i) {
 613                         tx_queue->rd[i].iov_base = tx_queue->map + (i * framesize);
 614                         tx_queue->rd[i].iov_len = req->tp_frame_size;
 615                 }
 616                 tx_queue->sockfd = qsockfd;
 617
 618                 rc = bind(qsockfd, (const struct sockaddr*)&sockaddr, sizeof(sockaddr));
 619                 if (rc == -1) {
 620                         RTE_LOG(ERR, PMD,
 621                                 "%s: could not bind AF_PACKET socket to %s\n",
 622                                 name, pair->value);
 623                         goto error;
 624                 }
 625
 626                 rc = setsockopt(qsockfd, SOL_PACKET, PACKET_FANOUT,
 627                                 &fanout_arg, sizeof(fanout_arg));
 628                 if (rc == -1) {
 629                         RTE_LOG(ERR, PMD,
 630                                 "%s: could not set PACKET_FANOUT on AF_PACKET socket "
 631                                 "for %s\n", name, pair->value);
 632                         goto error;
 633                 }
 634         }
 635
 636         /* reserve an ethdev entry */
 637         *eth_dev = rte_eth_dev_allocate(name);
 638         if (*eth_dev == NULL)
 639                 goto error;
 640
 641         /*
 642          * now put it all together
 643          * - store queue data in internals,
 644          * - store numa_node info in pci_driver
 645          * - point eth_dev_data to internals and pci_driver
 646          * - and point eth_dev structure to new eth_dev_data structure
 647          */
 648
 649         (*internals)->nb_queues = nb_queues;
 650
 651         data->dev_private = *internals;
 652         data->port_id = (*eth_dev)->data->port_id;
 653         data->nb_rx_queues = (uint16_t)nb_queues;
 654         data->nb_tx_queues = (uint16_t)nb_queues;
 655         data->dev_link = pmd_link;
 656         data->mac_addrs = &(*internals)->eth_addr;
 657
 658         pci_dev->numa_node = numa_node;
 659
 660         (*eth_dev)->data = data;
 661         (*eth_dev)->dev_ops = &ops;
 662         (*eth_dev)->pci_dev = pci_dev;
 663
 664         return 0;
 665
 666 error:
 667         if (data)
 668                 rte_free(data);
 669         if (pci_dev)
 670                 rte_free(pci_dev);
 671         for (q = 0; q < nb_queues; q++) {
 672                 if ((*internals)->rx_queue[q].rd)
 673                         rte_free((*internals)->rx_queue[q].rd);
 674                 if ((*internals)->tx_queue[q].rd)
 675                         rte_free((*internals)->tx_queue[q].rd);
 676         }
 677         if (*internals)
 678                 rte_free(*internals);
 679         return -1;
 680 }
 681
 682 static int
 683 rte_eth_from_packet(const char *name,
 684                     int const *sockfd,
 685                     const unsigned numa_node,
 686                     struct rte_kvargs *kvlist)
 687 {
 688         struct pmd_internals *internals = NULL;
 689         struct rte_eth_dev *eth_dev = NULL;
 690         struct rte_kvargs_pair *pair = NULL;
 691         unsigned k_idx;
 692         unsigned int blockcount;
 693         unsigned int blocksize = DFLT_BLOCK_SIZE;
 694         unsigned int framesize = DFLT_FRAME_SIZE;
 695         unsigned int framecount = DFLT_FRAME_COUNT;
 696         unsigned int qpairs = 1;
 697
 698         /* do some parameter checking */
 699         if (*sockfd < 0)
 700                 return -1;
 701
 702         /*
 703          * Walk arguments for configurable settings
 704          */
 705         for (k_idx = 0; k_idx < kvlist->count; k_idx++) {
 706                 pair = &kvlist->pairs[k_idx];
 707                 if (strstr(pair->key, ETH_AF_PACKET_NUM_Q_ARG) != NULL) {
 708                         qpairs = atoi(pair->value);
 709                         if (qpairs < 1 ||
 710                             qpairs > RTE_PMD_AF_PACKET_MAX_RINGS) {
 711                                 RTE_LOG(ERR, PMD,
 712                                         "%s: invalid qpairs value\n",
 713                                         name);
 714                                 return -1;
 715                         }
 716                         continue;
 717                 }
 718                 if (strstr(pair->key, ETH_AF_PACKET_BLOCKSIZE_ARG) != NULL) {
 719                         blocksize = atoi(pair->value);
 720                         if (!blocksize) {
 721                                 RTE_LOG(ERR, PMD,
 722                                         "%s: invalid blocksize value\n",
 723                                         name);
 724                                 return -1;
 725                         }
 726                         continue;
 727                 }
 728                 if (strstr(pair->key, ETH_AF_PACKET_FRAMESIZE_ARG) != NULL) {
 729                         framesize = atoi(pair->value);
 730                         if (!framesize) {
 731                                 RTE_LOG(ERR, PMD,
 732                                         "%s: invalid framesize value\n",
 733                                         name);
 734                                 return -1;
 735                         }
 736                         continue;
 737                 }
 738                 if (strstr(pair->key, ETH_AF_PACKET_FRAMECOUNT_ARG) != NULL) {
 739                         framecount = atoi(pair->value);
 740                         if (!framecount) {
 741                                 RTE_LOG(ERR, PMD,
 742                                         "%s: invalid framecount value\n",
 743                                         name);
 744                                 return -1;
 745                         }
 746                         continue;
 747                 }
 748         }
 749
 750         if (framesize > blocksize) {
 751                 RTE_LOG(ERR, PMD,
 752                         "%s: AF_PACKET MMAP frame size exceeds block size!\n",
 753                         name);
 754                 return -1;
 755         }
 756
 757         blockcount = framecount / (blocksize / framesize);
 758         if (!blockcount) {
 759                 RTE_LOG(ERR, PMD,
 760                         "%s: invalid AF_PACKET MMAP parameters\n", name);
 761                 return -1;
 762         }
 763
 764         RTE_LOG(INFO, PMD, "%s: AF_PACKET MMAP parameters:\n", name);
 765         RTE_LOG(INFO, PMD, "%s:\tblock size %d\n", name, blocksize);
 766         RTE_LOG(INFO, PMD, "%s:\tblock count %d\n", name, blockcount);
 767         RTE_LOG(INFO, PMD, "%s:\tframe size %d\n", name, framesize);
 768         RTE_LOG(INFO, PMD, "%s:\tframe count %d\n", name, framecount);
 769
 770         if (rte_pmd_init_internals(name, *sockfd, qpairs,
 771                                    blocksize, blockcount,
 772                                    framesize, framecount,
 773                                    numa_node, &internals, &eth_dev,
 774                                    kvlist) < 0)
 775                 return -1;
 776
 777         eth_dev->rx_pkt_burst = eth_af_packet_rx;
 778         eth_dev->tx_pkt_burst = eth_af_packet_tx;
 779
 780         return 0;
 781 }
 782
 783 int
 784 rte_pmd_af_packet_devinit(const char *name, const char *params)
 785 {
 786         unsigned numa_node;
 787         int ret;
 788         struct rte_kvargs *kvlist;
 789         int sockfd = -1;
 790
 791         RTE_LOG(INFO, PMD, "Initializing pmd_af_packet for %s\n", name);
 792
 793         numa_node = rte_socket_id();
 794
 795         kvlist = rte_kvargs_parse(params, valid_arguments);
 796         if (kvlist == NULL)
 797                 return -1;
 798
 799         /*
 800          * If iface argument is passed we open the NICs and use them for
 801          * reading / writing
 802          */
 803         if (rte_kvargs_count(kvlist, ETH_AF_PACKET_IFACE_ARG) == 1) {
 804
 805                 ret = rte_kvargs_process(kvlist, ETH_AF_PACKET_IFACE_ARG,
 806                                          &open_packet_iface, &sockfd);
 807                 if (ret < 0)
 808                         return -1;
 809         }
 810
 811         ret = rte_eth_from_packet(name, &sockfd, numa_node, kvlist);
 812         close(sockfd); /* no longer needed */
 813
 814         if (ret < 0)
 815                 return -1;
 816
 817         return 0;
 818 }
 819
 820 static struct rte_driver pmd_af_packet_drv = {
 821         .name = "eth_af_packet",
 822         .type = PMD_VDEV,
 823         .init = rte_pmd_af_packet_devinit,
 824 };
 825
 826 PMD_REGISTER_DRIVER(pmd_af_packet_drv);