drivers/net/mlx5/mlx5_ethdev.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright 2015 6WIND S.A.
   3  * Copyright 2015 Mellanox Technologies, Ltd
   4  */
   5
   6 #include <stddef.h>
   7 #include <assert.h>
   8 #include <inttypes.h>
   9 #include <unistd.h>
  10 #include <stdbool.h>
  11 #include <stdint.h>
  12 #include <stdio.h>
  13 #include <string.h>
  14 #include <stdlib.h>
  15 #include <errno.h>
  16 #include <dirent.h>
  17 #include <net/if.h>
  18 #include <sys/ioctl.h>
  19 #include <sys/socket.h>
  20 #include <netinet/in.h>
  21 #include <linux/ethtool.h>
  22 #include <linux/sockios.h>
  23 #include <fcntl.h>
  24 #include <stdalign.h>
  25 #include <sys/un.h>
  26 #include <time.h>
  27
  28 #include <rte_atomic.h>
  29 #include <rte_ethdev_driver.h>
  30 #include <rte_bus_pci.h>
  31 #include <rte_mbuf.h>
  32 #include <rte_common.h>
  33 #include <rte_interrupts.h>
  34 #include <rte_malloc.h>
  35 #include <rte_string_fns.h>
  36 #include <rte_rwlock.h>
  37 #include <rte_cycles.h>
  38
  39 #include "mlx5.h"
  40 #include "mlx5_glue.h"
  41 #include "mlx5_rxtx.h"
  42 #include "mlx5_utils.h"
  43
  44 /* Supported speed values found in /usr/include/linux/ethtool.h */
  45 #ifndef HAVE_SUPPORTED_40000baseKR4_Full
  46 #define SUPPORTED_40000baseKR4_Full (1 << 23)
  47 #endif
  48 #ifndef HAVE_SUPPORTED_40000baseCR4_Full
  49 #define SUPPORTED_40000baseCR4_Full (1 << 24)
  50 #endif
  51 #ifndef HAVE_SUPPORTED_40000baseSR4_Full
  52 #define SUPPORTED_40000baseSR4_Full (1 << 25)
  53 #endif
  54 #ifndef HAVE_SUPPORTED_40000baseLR4_Full
  55 #define SUPPORTED_40000baseLR4_Full (1 << 26)
  56 #endif
  57 #ifndef HAVE_SUPPORTED_56000baseKR4_Full
  58 #define SUPPORTED_56000baseKR4_Full (1 << 27)
  59 #endif
  60 #ifndef HAVE_SUPPORTED_56000baseCR4_Full
  61 #define SUPPORTED_56000baseCR4_Full (1 << 28)
  62 #endif
  63 #ifndef HAVE_SUPPORTED_56000baseSR4_Full
  64 #define SUPPORTED_56000baseSR4_Full (1 << 29)
  65 #endif
  66 #ifndef HAVE_SUPPORTED_56000baseLR4_Full
  67 #define SUPPORTED_56000baseLR4_Full (1 << 30)
  68 #endif
  69
  70 /* Add defines in case the running kernel is not the same as user headers. */
  71 #ifndef ETHTOOL_GLINKSETTINGS
  72 struct ethtool_link_settings {
  73         uint32_t cmd;
  74         uint32_t speed;
  75         uint8_t duplex;
  76         uint8_t port;
  77         uint8_t phy_address;
  78         uint8_t autoneg;
  79         uint8_t mdio_support;
  80         uint8_t eth_to_mdix;
  81         uint8_t eth_tp_mdix_ctrl;
  82         int8_t link_mode_masks_nwords;
  83         uint32_t reserved[8];
  84         uint32_t link_mode_masks[];
  85 };
  86
  87 #define ETHTOOL_GLINKSETTINGS 0x0000004c
  88 #define ETHTOOL_LINK_MODE_1000baseT_Full_BIT 5
  89 #define ETHTOOL_LINK_MODE_Autoneg_BIT 6
  90 #define ETHTOOL_LINK_MODE_1000baseKX_Full_BIT 17
  91 #define ETHTOOL_LINK_MODE_10000baseKX4_Full_BIT 18
  92 #define ETHTOOL_LINK_MODE_10000baseKR_Full_BIT 19
  93 #define ETHTOOL_LINK_MODE_10000baseR_FEC_BIT 20
  94 #define ETHTOOL_LINK_MODE_20000baseMLD2_Full_BIT 21
  95 #define ETHTOOL_LINK_MODE_20000baseKR2_Full_BIT 22
  96 #define ETHTOOL_LINK_MODE_40000baseKR4_Full_BIT 23
  97 #define ETHTOOL_LINK_MODE_40000baseCR4_Full_BIT 24
  98 #define ETHTOOL_LINK_MODE_40000baseSR4_Full_BIT 25
  99 #define ETHTOOL_LINK_MODE_40000baseLR4_Full_BIT 26
 100 #define ETHTOOL_LINK_MODE_56000baseKR4_Full_BIT 27
 101 #define ETHTOOL_LINK_MODE_56000baseCR4_Full_BIT 28
 102 #define ETHTOOL_LINK_MODE_56000baseSR4_Full_BIT 29
 103 #define ETHTOOL_LINK_MODE_56000baseLR4_Full_BIT 30
 104 #endif
 105 #ifndef HAVE_ETHTOOL_LINK_MODE_25G
 106 #define ETHTOOL_LINK_MODE_25000baseCR_Full_BIT 31
 107 #define ETHTOOL_LINK_MODE_25000baseKR_Full_BIT 32
 108 #define ETHTOOL_LINK_MODE_25000baseSR_Full_BIT 33
 109 #endif
 110 #ifndef HAVE_ETHTOOL_LINK_MODE_50G
 111 #define ETHTOOL_LINK_MODE_50000baseCR2_Full_BIT 34
 112 #define ETHTOOL_LINK_MODE_50000baseKR2_Full_BIT 35
 113 #endif
 114 #ifndef HAVE_ETHTOOL_LINK_MODE_100G
 115 #define ETHTOOL_LINK_MODE_100000baseKR4_Full_BIT 36
 116 #define ETHTOOL_LINK_MODE_100000baseSR4_Full_BIT 37
 117 #define ETHTOOL_LINK_MODE_100000baseCR4_Full_BIT 38
 118 #define ETHTOOL_LINK_MODE_100000baseLR4_ER4_Full_BIT 39
 119 #endif
 120
 121 /**
 122  * Get master interface name from private structure.
 123  *
 124  * @param[in] dev
 125  *   Pointer to Ethernet device.
 126  * @param[out] ifname
 127  *   Interface name output buffer.
 128  *
 129  * @return
 130  *   0 on success, a negative errno value otherwise and rte_errno is set.
 131  */
 132 int
 133 mlx5_get_master_ifname(const char *ibdev_path, char (*ifname)[IF_NAMESIZE])
 134 {
 135         DIR *dir;
 136         struct dirent *dent;
 137         unsigned int dev_type = 0;
 138         unsigned int dev_port_prev = ~0u;
 139         char match[IF_NAMESIZE] = "";
 140
 141         assert(ibdev_path);
 142         {
 143                 MKSTR(path, "%s/device/net", ibdev_path);
 144
 145                 dir = opendir(path);
 146                 if (dir == NULL) {
 147                         rte_errno = errno;
 148                         return -rte_errno;
 149                 }
 150         }
 151         while ((dent = readdir(dir)) != NULL) {
 152                 char *name = dent->d_name;
 153                 FILE *file;
 154                 unsigned int dev_port;
 155                 int r;
 156
 157                 if ((name[0] == '.') &&
 158                     ((name[1] == '\0') ||
 159                      ((name[1] == '.') && (name[2] == '\0'))))
 160                         continue;
 161
 162                 MKSTR(path, "%s/device/net/%s/%s",
 163                       ibdev_path, name,
 164                       (dev_type ? "dev_id" : "dev_port"));
 165
 166                 file = fopen(path, "rb");
 167                 if (file == NULL) {
 168                         if (errno != ENOENT)
 169                                 continue;
 170                         /*
 171                          * Switch to dev_id when dev_port does not exist as
 172                          * is the case with Linux kernel versions < 3.15.
 173                          */
 174 try_dev_id:
 175                         match[0] = '\0';
 176                         if (dev_type)
 177                                 break;
 178                         dev_type = 1;
 179                         dev_port_prev = ~0u;
 180                         rewinddir(dir);
 181                         continue;
 182                 }
 183                 r = fscanf(file, (dev_type ? "%x" : "%u"), &dev_port);
 184                 fclose(file);
 185                 if (r != 1)
 186                         continue;
 187                 /*
 188                  * Switch to dev_id when dev_port returns the same value for
 189                  * all ports. May happen when using a MOFED release older than
 190                  * 3.0 with a Linux kernel >= 3.15.
 191                  */
 192                 if (dev_port == dev_port_prev)
 193                         goto try_dev_id;
 194                 dev_port_prev = dev_port;
 195                 if (dev_port == 0)
 196                         strlcpy(match, name, sizeof(match));
 197         }
 198         closedir(dir);
 199         if (match[0] == '\0') {
 200                 rte_errno = ENOENT;
 201                 return -rte_errno;
 202         }
 203         strncpy(*ifname, match, sizeof(*ifname));
 204         return 0;
 205 }
 206
 207 /**
 208  * Get interface name from private structure.
 209  *
 210  * This is a port representor-aware version of mlx5_get_master_ifname().
 211  *
 212  * @param[in] dev
 213  *   Pointer to Ethernet device.
 214  * @param[out] ifname
 215  *   Interface name output buffer.
 216  *
 217  * @return
 218  *   0 on success, a negative errno value otherwise and rte_errno is set.
 219  */
 220 int
 221 mlx5_get_ifname(const struct rte_eth_dev *dev, char (*ifname)[IF_NAMESIZE])
 222 {
 223         struct mlx5_priv *priv = dev->data->dev_private;
 224         unsigned int ifindex;
 225
 226         assert(priv);
 227         assert(priv->sh);
 228         ifindex = priv->nl_socket_rdma >= 0 ?
 229                   mlx5_nl_ifindex(priv->nl_socket_rdma,
 230                                   priv->sh->ibdev_name,
 231                                   priv->ibv_port) : 0;
 232         if (!ifindex) {
 233                 if (!priv->representor)
 234                         return mlx5_get_master_ifname(priv->sh->ibdev_path,
 235                                                       ifname);
 236                 rte_errno = ENXIO;
 237                 return -rte_errno;
 238         }
 239         if (if_indextoname(ifindex, &(*ifname)[0]))
 240                 return 0;
 241         rte_errno = errno;
 242         return -rte_errno;
 243 }
 244
 245 /**
 246  * Get interface name for the specified device, uses the extra base
 247  * device resources to perform Netlink requests.
 248  *
 249  * This is a port representor-aware version of mlx5_get_master_ifname().
 250  *
 251  * @param[in] base
 252  *   Pointer to Ethernet device to use Netlink socket from
 253  *   to perfrom requests.
 254  * @param[in] dev
 255  *   Pointer to Ethernet device.
 256  * @param[out] ifname
 257  *   Interface name output buffer.
 258  *
 259  * @return
 260  *   0 on success, a negative errno value otherwise and rte_errno is set.
 261  */
 262 int
 263 mlx5_get_ifname_base(const struct rte_eth_dev *base,
 264                      const struct rte_eth_dev *dev,
 265                      char (*ifname)[IF_NAMESIZE])
 266 {
 267         struct mlx5_priv *priv = dev->data->dev_private;
 268         struct mlx5_priv *priv_base = base->data->dev_private;
 269         unsigned int ifindex;
 270
 271         assert(priv);
 272         assert(priv->sh);
 273         assert(priv_base);
 274         ifindex = priv_base->nl_socket_rdma >= 0 ?
 275                   mlx5_nl_ifindex(priv_base->nl_socket_rdma,
 276                                   priv->sh->ibdev_name,
 277                                   priv->ibv_port) : 0;
 278         if (!ifindex) {
 279                 if (!priv->representor)
 280                         return mlx5_get_master_ifname(priv->sh->ibdev_path,
 281                                                       ifname);
 282                 rte_errno = ENXIO;
 283                 return -rte_errno;
 284         }
 285         if (if_indextoname(ifindex, &(*ifname)[0]))
 286                 return 0;
 287         rte_errno = errno;
 288         return -rte_errno;
 289 }
 290 /**
 291  * Get the interface index from device name.
 292  *
 293  * @param[in] dev
 294  *   Pointer to Ethernet device.
 295  *
 296  * @return
 297  *   Nonzero interface index on success, zero otherwise and rte_errno is set.
 298  */
 299 unsigned int
 300 mlx5_ifindex(const struct rte_eth_dev *dev)
 301 {
 302         char ifname[IF_NAMESIZE];
 303         unsigned int ifindex;
 304
 305         if (mlx5_get_ifname(dev, &ifname))
 306                 return 0;
 307         ifindex = if_nametoindex(ifname);
 308         if (!ifindex)
 309                 rte_errno = errno;
 310         return ifindex;
 311 }
 312
 313 /**
 314  * Perform ifreq ioctl() on associated Ethernet device.
 315  *
 316  * @param[in] dev
 317  *   Pointer to Ethernet device.
 318  * @param req
 319  *   Request number to pass to ioctl().
 320  * @param[out] ifr
 321  *   Interface request structure output buffer.
 322  *
 323  * @return
 324  *   0 on success, a negative errno value otherwise and rte_errno is set.
 325  */
 326 int
 327 mlx5_ifreq(const struct rte_eth_dev *dev, int req, struct ifreq *ifr)
 328 {
 329         int sock = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
 330         int ret = 0;
 331
 332         if (sock == -1) {
 333                 rte_errno = errno;
 334                 return -rte_errno;
 335         }
 336         ret = mlx5_get_ifname(dev, &ifr->ifr_name);
 337         if (ret)
 338                 goto error;
 339         ret = ioctl(sock, req, ifr);
 340         if (ret == -1) {
 341                 rte_errno = errno;
 342                 goto error;
 343         }
 344         close(sock);
 345         return 0;
 346 error:
 347         close(sock);
 348         return -rte_errno;
 349 }
 350
 351 /**
 352  * Perform ifreq ioctl() on specified Ethernet device,
 353  * ifindex, name and other attributes are requested
 354  * on the base device to avoid specified device Netlink
 355  * socket sharing (this is not thread-safe).
 356  *
 357  * @param[in] base
 358  *   Pointer to Ethernet device to get dev attributes.
 359  * @param[in] dev
 360  *   Pointer to Ethernet device to perform ioctl.
 361  * @param req
 362  *   Request number to pass to ioctl().
 363  * @param[out] ifr
 364  *   Interface request structure output buffer.
 365  *
 366  * @return
 367  *   0 on success, a negative errno value otherwise and rte_errno is set.
 368  */
 369 int
 370 mlx5_ifreq_base(const struct rte_eth_dev *base,
 371                 const struct rte_eth_dev *dev,
 372                 int req, struct ifreq *ifr)
 373 {
 374         int sock = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
 375         int ret = 0;
 376
 377         if (sock == -1) {
 378                 rte_errno = errno;
 379                 return -rte_errno;
 380         }
 381         ret = mlx5_get_ifname_base(base, dev, &ifr->ifr_name);
 382         if (ret)
 383                 goto error;
 384         ret = ioctl(sock, req, ifr);
 385         if (ret == -1) {
 386                 rte_errno = errno;
 387                 goto error;
 388         }
 389         close(sock);
 390         return 0;
 391 error:
 392         close(sock);
 393         return -rte_errno;
 394 }
 395
 396 /**
 397  * Get device MTU.
 398  *
 399  * @param dev
 400  *   Pointer to Ethernet device.
 401  * @param[out] mtu
 402  *   MTU value output buffer.
 403  *
 404  * @return
 405  *   0 on success, a negative errno value otherwise and rte_errno is set.
 406  */
 407 int
 408 mlx5_get_mtu(struct rte_eth_dev *dev, uint16_t *mtu)
 409 {
 410         struct ifreq request;
 411         int ret = mlx5_ifreq(dev, SIOCGIFMTU, &request);
 412
 413         if (ret)
 414                 return ret;
 415         *mtu = request.ifr_mtu;
 416         return 0;
 417 }
 418
 419 /**
 420  * Set device MTU.
 421  *
 422  * @param dev
 423  *   Pointer to Ethernet device.
 424  * @param mtu
 425  *   MTU value to set.
 426  *
 427  * @return
 428  *   0 on success, a negative errno value otherwise and rte_errno is set.
 429  */
 430 static int
 431 mlx5_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
 432 {
 433         struct ifreq request = { .ifr_mtu = mtu, };
 434
 435         return mlx5_ifreq(dev, SIOCSIFMTU, &request);
 436 }
 437
 438 /**
 439  * Set device flags.
 440  *
 441  * @param dev
 442  *   Pointer to Ethernet device.
 443  * @param keep
 444  *   Bitmask for flags that must remain untouched.
 445  * @param flags
 446  *   Bitmask for flags to modify.
 447  *
 448  * @return
 449  *   0 on success, a negative errno value otherwise and rte_errno is set.
 450  */
 451 int
 452 mlx5_set_flags(struct rte_eth_dev *dev, unsigned int keep, unsigned int flags)
 453 {
 454         struct ifreq request;
 455         int ret = mlx5_ifreq(dev, SIOCGIFFLAGS, &request);
 456
 457         if (ret)
 458                 return ret;
 459         request.ifr_flags &= keep;
 460         request.ifr_flags |= flags & ~keep;
 461         return mlx5_ifreq(dev, SIOCSIFFLAGS, &request);
 462 }
 463
 464 /**
 465  * DPDK callback for Ethernet device configuration.
 466  *
 467  * @param dev
 468  *   Pointer to Ethernet device structure.
 469  *
 470  * @return
 471  *   0 on success, a negative errno value otherwise and rte_errno is set.
 472  */
 473 int
 474 mlx5_dev_configure(struct rte_eth_dev *dev)
 475 {
 476         struct mlx5_priv *priv = dev->data->dev_private;
 477         unsigned int rxqs_n = dev->data->nb_rx_queues;
 478         unsigned int txqs_n = dev->data->nb_tx_queues;
 479         unsigned int i;
 480         unsigned int j;
 481         unsigned int reta_idx_n;
 482         const uint8_t use_app_rss_key =
 483                 !!dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key;
 484         int ret = 0;
 485
 486         if (use_app_rss_key &&
 487             (dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key_len !=
 488              MLX5_RSS_HASH_KEY_LEN)) {
 489                 DRV_LOG(ERR, "port %u RSS key len must be %s Bytes long",
 490                         dev->data->port_id, RTE_STR(MLX5_RSS_HASH_KEY_LEN));
 491                 rte_errno = EINVAL;
 492                 return -rte_errno;
 493         }
 494         priv->rss_conf.rss_key =
 495                 rte_realloc(priv->rss_conf.rss_key,
 496                             MLX5_RSS_HASH_KEY_LEN, 0);
 497         if (!priv->rss_conf.rss_key) {
 498                 DRV_LOG(ERR, "port %u cannot allocate RSS hash key memory (%u)",
 499                         dev->data->port_id, rxqs_n);
 500                 rte_errno = ENOMEM;
 501                 return -rte_errno;
 502         }
 503         memcpy(priv->rss_conf.rss_key,
 504                use_app_rss_key ?
 505                dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key :
 506                rss_hash_default_key,
 507                MLX5_RSS_HASH_KEY_LEN);
 508         priv->rss_conf.rss_key_len = MLX5_RSS_HASH_KEY_LEN;
 509         priv->rss_conf.rss_hf = dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf;
 510         priv->rxqs = (void *)dev->data->rx_queues;
 511         priv->txqs = (void *)dev->data->tx_queues;
 512         if (txqs_n != priv->txqs_n) {
 513                 DRV_LOG(INFO, "port %u Tx queues number update: %u -> %u",
 514                         dev->data->port_id, priv->txqs_n, txqs_n);
 515                 priv->txqs_n = txqs_n;
 516         }
 517         if (rxqs_n > priv->config.ind_table_max_size) {
 518                 DRV_LOG(ERR, "port %u cannot handle this many Rx queues (%u)",
 519                         dev->data->port_id, rxqs_n);
 520                 rte_errno = EINVAL;
 521                 return -rte_errno;
 522         }
 523         if (rxqs_n != priv->rxqs_n) {
 524                 DRV_LOG(INFO, "port %u Rx queues number update: %u -> %u",
 525                         dev->data->port_id, priv->rxqs_n, rxqs_n);
 526                 priv->rxqs_n = rxqs_n;
 527                 /*
 528                  * If the requested number of RX queues is not a power of two,
 529                  * use the maximum indirection table size for better balancing.
 530                  * The result is always rounded to the next power of two.
 531                  */
 532                 reta_idx_n = (1 << log2above((rxqs_n & (rxqs_n - 1)) ?
 533                                              priv->config.ind_table_max_size :
 534                                              rxqs_n));
 535                 ret = mlx5_rss_reta_index_resize(dev, reta_idx_n);
 536                 if (ret)
 537                         return ret;
 538                 /*
 539                  * When the number of RX queues is not a power of two,
 540                  * the remaining table entries are padded with reused WQs
 541                  * and hashes are not spread uniformly.
 542                  */
 543                 for (i = 0, j = 0; (i != reta_idx_n); ++i) {
 544                         (*priv->reta_idx)[i] = j;
 545                         if (++j == rxqs_n)
 546                                 j = 0;
 547                 }
 548         }
 549         ret = mlx5_proc_priv_init(dev);
 550         if (ret)
 551                 return ret;
 552         return 0;
 553 }
 554
 555 /**
 556  * Sets default tuning parameters.
 557  *
 558  * @param dev
 559  *   Pointer to Ethernet device.
 560  * @param[out] info
 561  *   Info structure output buffer.
 562  */
 563 static void
 564 mlx5_set_default_params(struct rte_eth_dev *dev, struct rte_eth_dev_info *info)
 565 {
 566         struct mlx5_priv *priv = dev->data->dev_private;
 567
 568         /* Minimum CPU utilization. */
 569         info->default_rxportconf.ring_size = 256;
 570         info->default_txportconf.ring_size = 256;
 571         info->default_rxportconf.burst_size = 64;
 572         info->default_txportconf.burst_size = 64;
 573         if (priv->link_speed_capa & ETH_LINK_SPEED_100G) {
 574                 info->default_rxportconf.nb_queues = 16;
 575                 info->default_txportconf.nb_queues = 16;
 576                 if (dev->data->nb_rx_queues > 2 ||
 577                     dev->data->nb_tx_queues > 2) {
 578                         /* Max Throughput. */
 579                         info->default_rxportconf.ring_size = 2048;
 580                         info->default_txportconf.ring_size = 2048;
 581                 }
 582         } else {
 583                 info->default_rxportconf.nb_queues = 8;
 584                 info->default_txportconf.nb_queues = 8;
 585                 if (dev->data->nb_rx_queues > 2 ||
 586                     dev->data->nb_tx_queues > 2) {
 587                         /* Max Throughput. */
 588                         info->default_rxportconf.ring_size = 4096;
 589                         info->default_txportconf.ring_size = 4096;
 590                 }
 591         }
 592 }
 593
 594 /**
 595  * DPDK callback to get information about the device.
 596  *
 597  * @param dev
 598  *   Pointer to Ethernet device structure.
 599  * @param[out] info
 600  *   Info structure output buffer.
 601  */
 602 void
 603 mlx5_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *info)
 604 {
 605         struct mlx5_priv *priv = dev->data->dev_private;
 606         struct mlx5_dev_config *config = &priv->config;
 607         unsigned int max;
 608         char ifname[IF_NAMESIZE];
 609
 610         /* FIXME: we should ask the device for these values. */
 611         info->min_rx_bufsize = 32;
 612         info->max_rx_pktlen = 65536;
 613         /*
 614          * Since we need one CQ per QP, the limit is the minimum number
 615          * between the two values.
 616          */
 617         max = RTE_MIN(priv->sh->device_attr.orig_attr.max_cq,
 618                       priv->sh->device_attr.orig_attr.max_qp);
 619         /* If max >= 65535 then max = 0, max_rx_queues is uint16_t. */
 620         if (max >= 65535)
 621                 max = 65535;
 622         info->max_rx_queues = max;
 623         info->max_tx_queues = max;
 624         info->max_mac_addrs = MLX5_MAX_UC_MAC_ADDRESSES;
 625         info->rx_queue_offload_capa = mlx5_get_rx_queue_offloads(dev);
 626         info->rx_offload_capa = (mlx5_get_rx_port_offloads() |
 627                                  info->rx_queue_offload_capa);
 628         info->tx_offload_capa = mlx5_get_tx_port_offloads(dev);
 629         if (mlx5_get_ifname(dev, &ifname) == 0)
 630                 info->if_index = if_nametoindex(ifname);
 631         info->reta_size = priv->reta_idx_n ?
 632                 priv->reta_idx_n : config->ind_table_max_size;
 633         info->hash_key_size = MLX5_RSS_HASH_KEY_LEN;
 634         info->speed_capa = priv->link_speed_capa;
 635         info->flow_type_rss_offloads = ~MLX5_RSS_HF_MASK;
 636         mlx5_set_default_params(dev, info);
 637         info->switch_info.name = dev->data->name;
 638         info->switch_info.domain_id = priv->domain_id;
 639         info->switch_info.port_id = priv->representor_id;
 640         if (priv->representor) {
 641                 unsigned int i = mlx5_dev_to_port_id(dev->device, NULL, 0);
 642                 uint16_t port_id[i];
 643
 644                 i = RTE_MIN(mlx5_dev_to_port_id(dev->device, port_id, i), i);
 645                 while (i--) {
 646                         struct mlx5_priv *opriv =
 647                                 rte_eth_devices[port_id[i]].data->dev_private;
 648
 649                         if (!opriv ||
 650                             opriv->representor ||
 651                             opriv->domain_id != priv->domain_id)
 652                                 continue;
 653                         /*
 654                          * Override switch name with that of the master
 655                          * device.
 656                          */
 657                         info->switch_info.name = opriv->dev_data->name;
 658                         break;
 659                 }
 660         }
 661 }
 662
 663 /**
 664  * Get device current raw clock counter
 665  *
 666  * @param dev
 667  *   Pointer to Ethernet device structure.
 668  * @param[out] time
 669  *   Current raw clock counter of the device.
 670  *
 671  * @return
 672  *   0 if the clock has correctly been read
 673  *   The value of errno in case of error
 674  */
 675 int
 676 mlx5_read_clock(struct rte_eth_dev *dev, uint64_t *clock)
 677 {
 678         struct mlx5_priv *priv = dev->data->dev_private;
 679         struct ibv_context *ctx = priv->sh->ctx;
 680         struct ibv_values_ex values;
 681         int err = 0;
 682
 683         values.comp_mask = IBV_VALUES_MASK_RAW_CLOCK;
 684         err = mlx5_glue->query_rt_values_ex(ctx, &values);
 685         if (err != 0) {
 686                 DRV_LOG(WARNING, "Could not query the clock !");
 687                 return err;
 688         }
 689         *clock = values.raw_clock.tv_nsec;
 690         return 0;
 691 }
 692
 693 /**
 694  * Get firmware version of a device.
 695  *
 696  * @param dev
 697  *   Ethernet device port.
 698  * @param fw_ver
 699  *   String output allocated by caller.
 700  * @param fw_size
 701  *   Size of the output string, including terminating null byte.
 702  *
 703  * @return
 704  *   0 on success, or the size of the non truncated string if too big.
 705  */
 706 int mlx5_fw_version_get(struct rte_eth_dev *dev, char *fw_ver, size_t fw_size)
 707 {
 708         struct mlx5_priv *priv = dev->data->dev_private;
 709         struct ibv_device_attr *attr = &priv->sh->device_attr.orig_attr;
 710         size_t size = strnlen(attr->fw_ver, sizeof(attr->fw_ver)) + 1;
 711
 712         if (fw_size < size)
 713                 return size;
 714         if (fw_ver != NULL)
 715                 strlcpy(fw_ver, attr->fw_ver, fw_size);
 716         return 0;
 717 }
 718
 719 /**
 720  * Get supported packet types.
 721  *
 722  * @param dev
 723  *   Pointer to Ethernet device structure.
 724  *
 725  * @return
 726  *   A pointer to the supported Packet types array.
 727  */
 728 const uint32_t *
 729 mlx5_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 730 {
 731         static const uint32_t ptypes[] = {
 732                 /* refers to rxq_cq_to_pkt_type() */
 733                 RTE_PTYPE_L2_ETHER,
 734                 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
 735                 RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
 736                 RTE_PTYPE_L4_NONFRAG,
 737                 RTE_PTYPE_L4_FRAG,
 738                 RTE_PTYPE_L4_TCP,
 739                 RTE_PTYPE_L4_UDP,
 740                 RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN,
 741                 RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN,
 742                 RTE_PTYPE_INNER_L4_NONFRAG,
 743                 RTE_PTYPE_INNER_L4_FRAG,
 744                 RTE_PTYPE_INNER_L4_TCP,
 745                 RTE_PTYPE_INNER_L4_UDP,
 746                 RTE_PTYPE_UNKNOWN
 747         };
 748
 749         if (dev->rx_pkt_burst == mlx5_rx_burst ||
 750             dev->rx_pkt_burst == mlx5_rx_burst_mprq ||
 751             dev->rx_pkt_burst == mlx5_rx_burst_vec)
 752                 return ptypes;
 753         return NULL;
 754 }
 755
 756 /**
 757  * Retrieve the master device for representor in the same switch domain.
 758  *
 759  * @param dev
 760  *   Pointer to representor Ethernet device structure.
 761  *
 762  * @return
 763  *   Master device structure  on success, NULL otherwise.
 764  */
 765
 766 static struct rte_eth_dev *
 767 mlx5_find_master_dev(struct rte_eth_dev *dev)
 768 {
 769         struct mlx5_priv *priv;
 770         uint16_t port_id;
 771         uint16_t domain_id;
 772
 773         priv = dev->data->dev_private;
 774         domain_id = priv->domain_id;
 775         assert(priv->representor);
 776         RTE_ETH_FOREACH_DEV_OF(port_id, dev->device) {
 777                 priv = rte_eth_devices[port_id].data->dev_private;
 778                 if (priv &&
 779                     priv->master &&
 780                     priv->domain_id == domain_id)
 781                         return &rte_eth_devices[port_id];
 782         }
 783         return NULL;
 784 }
 785
 786 /**
 787  * DPDK callback to retrieve physical link information.
 788  *
 789  * @param dev
 790  *   Pointer to Ethernet device structure.
 791  * @param[out] link
 792  *   Storage for current link status.
 793  *
 794  * @return
 795  *   0 on success, a negative errno value otherwise and rte_errno is set.
 796  */
 797 static int
 798 mlx5_link_update_unlocked_gset(struct rte_eth_dev *dev,
 799                                struct rte_eth_link *link)
 800 {
 801         struct mlx5_priv *priv = dev->data->dev_private;
 802         struct ethtool_cmd edata = {
 803                 .cmd = ETHTOOL_GSET /* Deprecated since Linux v4.5. */
 804         };
 805         struct ifreq ifr;
 806         struct rte_eth_link dev_link;
 807         int link_speed = 0;
 808         int ret;
 809
 810         ret = mlx5_ifreq(dev, SIOCGIFFLAGS, &ifr);
 811         if (ret) {
 812                 DRV_LOG(WARNING, "port %u ioctl(SIOCGIFFLAGS) failed: %s",
 813                         dev->data->port_id, strerror(rte_errno));
 814                 return ret;
 815         }
 816         dev_link = (struct rte_eth_link) {
 817                 .link_status = ((ifr.ifr_flags & IFF_UP) &&
 818                                 (ifr.ifr_flags & IFF_RUNNING)),
 819         };
 820         ifr = (struct ifreq) {
 821                 .ifr_data = (void *)&edata,
 822         };
 823         ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
 824         if (ret) {
 825                 if (ret == -ENOTSUP && priv->representor) {
 826                         struct rte_eth_dev *master;
 827
 828                         /*
 829                          * For representors we can try to inherit link
 830                          * settings from the master device. Actually
 831                          * link settings do not make a lot of sense
 832                          * for representors due to missing physical
 833                          * link. The old kernel drivers supported
 834                          * emulated settings query for representors,
 835                          * the new ones do not, so we have to add
 836                          * this code for compatibility issues.
 837                          */
 838                         master = mlx5_find_master_dev(dev);
 839                         if (master) {
 840                                 ifr = (struct ifreq) {
 841                                         .ifr_data = (void *)&edata,
 842                                 };
 843                                 /*
 844                                  * Use special version of mlx5_ifreq()
 845                                  * to get master device name with local
 846                                  * device Netlink socket. Using master
 847                                  * device Netlink socket is not thread
 848                                  * safe.
 849                                  */
 850                                 ret = mlx5_ifreq_base(dev, master,
 851                                                       SIOCETHTOOL, &ifr);
 852                         }
 853                 }
 854                 if (ret) {
 855                         DRV_LOG(WARNING,
 856                                 "port %u ioctl(SIOCETHTOOL,"
 857                                 " ETHTOOL_GSET) failed: %s",
 858                                 dev->data->port_id, strerror(rte_errno));
 859                         return ret;
 860                 }
 861         }
 862         link_speed = ethtool_cmd_speed(&edata);
 863         if (link_speed == -1)
 864                 dev_link.link_speed = ETH_SPEED_NUM_NONE;
 865         else
 866                 dev_link.link_speed = link_speed;
 867         priv->link_speed_capa = 0;
 868         if (edata.supported & SUPPORTED_Autoneg)
 869                 priv->link_speed_capa |= ETH_LINK_SPEED_AUTONEG;
 870         if (edata.supported & (SUPPORTED_1000baseT_Full |
 871                                SUPPORTED_1000baseKX_Full))
 872                 priv->link_speed_capa |= ETH_LINK_SPEED_1G;
 873         if (edata.supported & SUPPORTED_10000baseKR_Full)
 874                 priv->link_speed_capa |= ETH_LINK_SPEED_10G;
 875         if (edata.supported & (SUPPORTED_40000baseKR4_Full |
 876                                SUPPORTED_40000baseCR4_Full |
 877                                SUPPORTED_40000baseSR4_Full |
 878                                SUPPORTED_40000baseLR4_Full))
 879                 priv->link_speed_capa |= ETH_LINK_SPEED_40G;
 880         dev_link.link_duplex = ((edata.duplex == DUPLEX_HALF) ?
 881                                 ETH_LINK_HALF_DUPLEX : ETH_LINK_FULL_DUPLEX);
 882         dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds &
 883                         ETH_LINK_SPEED_FIXED);
 884         if (((dev_link.link_speed && !dev_link.link_status) ||
 885              (!dev_link.link_speed && dev_link.link_status))) {
 886                 rte_errno = EAGAIN;
 887                 return -rte_errno;
 888         }
 889         *link = dev_link;
 890         return 0;
 891 }
 892
 893 /**
 894  * Retrieve physical link information (unlocked version using new ioctl).
 895  *
 896  * @param dev
 897  *   Pointer to Ethernet device structure.
 898  * @param[out] link
 899  *   Storage for current link status.
 900  *
 901  * @return
 902  *   0 on success, a negative errno value otherwise and rte_errno is set.
 903  */
 904 static int
 905 mlx5_link_update_unlocked_gs(struct rte_eth_dev *dev,
 906                              struct rte_eth_link *link)
 907
 908 {
 909         struct mlx5_priv *priv = dev->data->dev_private;
 910         struct ethtool_link_settings gcmd = { .cmd = ETHTOOL_GLINKSETTINGS };
 911         struct ifreq ifr;
 912         struct rte_eth_link dev_link;
 913         struct rte_eth_dev *master = NULL;
 914         uint64_t sc;
 915         int ret;
 916
 917         ret = mlx5_ifreq(dev, SIOCGIFFLAGS, &ifr);
 918         if (ret) {
 919                 DRV_LOG(WARNING, "port %u ioctl(SIOCGIFFLAGS) failed: %s",
 920                         dev->data->port_id, strerror(rte_errno));
 921                 return ret;
 922         }
 923         dev_link = (struct rte_eth_link) {
 924                 .link_status = ((ifr.ifr_flags & IFF_UP) &&
 925                                 (ifr.ifr_flags & IFF_RUNNING)),
 926         };
 927         ifr = (struct ifreq) {
 928                 .ifr_data = (void *)&gcmd,
 929         };
 930         ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
 931         if (ret) {
 932                 if (ret == -ENOTSUP && priv->representor) {
 933                         /*
 934                          * For representors we can try to inherit link
 935                          * settings from the master device. Actually
 936                          * link settings do not make a lot of sense
 937                          * for representors due to missing physical
 938                          * link. The old kernel drivers supported
 939                          * emulated settings query for representors,
 940                          * the new ones do not, so we have to add
 941                          * this code for compatibility issues.
 942                          */
 943                         master = mlx5_find_master_dev(dev);
 944                         if (master) {
 945                                 ifr = (struct ifreq) {
 946                                         .ifr_data = (void *)&gcmd,
 947                                 };
 948                                 /*
 949                                  * Avoid using master Netlink socket.
 950                                  * This is not thread-safe.
 951                                  */
 952                                 ret = mlx5_ifreq_base(dev, master,
 953                                                       SIOCETHTOOL, &ifr);
 954                         }
 955                 }
 956                 if (ret) {
 957                         DRV_LOG(DEBUG,
 958                                 "port %u ioctl(SIOCETHTOOL,"
 959                                 " ETHTOOL_GLINKSETTINGS) failed: %s",
 960                                 dev->data->port_id, strerror(rte_errno));
 961                         return ret;
 962                 }
 963
 964         }
 965         gcmd.link_mode_masks_nwords = -gcmd.link_mode_masks_nwords;
 966
 967         alignas(struct ethtool_link_settings)
 968         uint8_t data[offsetof(struct ethtool_link_settings, link_mode_masks) +
 969                      sizeof(uint32_t) * gcmd.link_mode_masks_nwords * 3];
 970         struct ethtool_link_settings *ecmd = (void *)data;
 971
 972         *ecmd = gcmd;
 973         ifr.ifr_data = (void *)ecmd;
 974         ret = mlx5_ifreq_base(dev, master ? master : dev, SIOCETHTOOL, &ifr);
 975         if (ret) {
 976                 DRV_LOG(DEBUG,
 977                         "port %u ioctl(SIOCETHTOOL,"
 978                         "ETHTOOL_GLINKSETTINGS) failed: %s",
 979                         dev->data->port_id, strerror(rte_errno));
 980                 return ret;
 981         }
 982         dev_link.link_speed = ecmd->speed;
 983         sc = ecmd->link_mode_masks[0] |
 984                 ((uint64_t)ecmd->link_mode_masks[1] << 32);
 985         priv->link_speed_capa = 0;
 986         if (sc & MLX5_BITSHIFT(ETHTOOL_LINK_MODE_Autoneg_BIT))
 987                 priv->link_speed_capa |= ETH_LINK_SPEED_AUTONEG;
 988         if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_1000baseT_Full_BIT) |
 989                   MLX5_BITSHIFT(ETHTOOL_LINK_MODE_1000baseKX_Full_BIT)))
 990                 priv->link_speed_capa |= ETH_LINK_SPEED_1G;
 991         if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_10000baseKX4_Full_BIT) |
 992                   MLX5_BITSHIFT(ETHTOOL_LINK_MODE_10000baseKR_Full_BIT) |
 993                   MLX5_BITSHIFT(ETHTOOL_LINK_MODE_10000baseR_FEC_BIT)))
 994                 priv->link_speed_capa |= ETH_LINK_SPEED_10G;
 995         if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_20000baseMLD2_Full_BIT) |
 996                   MLX5_BITSHIFT(ETHTOOL_LINK_MODE_20000baseKR2_Full_BIT)))
 997                 priv->link_speed_capa |= ETH_LINK_SPEED_20G;
 998         if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_40000baseKR4_Full_BIT) |
 999                   MLX5_BITSHIFT(ETHTOOL_LINK_MODE_40000baseCR4_Full_BIT) |
1000                   MLX5_BITSHIFT(ETHTOOL_LINK_MODE_40000baseSR4_Full_BIT) |
1001                   MLX5_BITSHIFT(ETHTOOL_LINK_MODE_40000baseLR4_Full_BIT)))
1002                 priv->link_speed_capa |= ETH_LINK_SPEED_40G;
1003         if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_56000baseKR4_Full_BIT) |
1004                   MLX5_BITSHIFT(ETHTOOL_LINK_MODE_56000baseCR4_Full_BIT) |
1005                   MLX5_BITSHIFT(ETHTOOL_LINK_MODE_56000baseSR4_Full_BIT) |
1006                   MLX5_BITSHIFT(ETHTOOL_LINK_MODE_56000baseLR4_Full_BIT)))
1007                 priv->link_speed_capa |= ETH_LINK_SPEED_56G;
1008         if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_25000baseCR_Full_BIT) |
1009                   MLX5_BITSHIFT(ETHTOOL_LINK_MODE_25000baseKR_Full_BIT) |
1010                   MLX5_BITSHIFT(ETHTOOL_LINK_MODE_25000baseSR_Full_BIT)))
1011                 priv->link_speed_capa |= ETH_LINK_SPEED_25G;
1012         if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_50000baseCR2_Full_BIT) |
1013                   MLX5_BITSHIFT(ETHTOOL_LINK_MODE_50000baseKR2_Full_BIT)))
1014                 priv->link_speed_capa |= ETH_LINK_SPEED_50G;
1015         if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_100000baseKR4_Full_BIT) |
1016                   MLX5_BITSHIFT(ETHTOOL_LINK_MODE_100000baseSR4_Full_BIT) |
1017                   MLX5_BITSHIFT(ETHTOOL_LINK_MODE_100000baseCR4_Full_BIT) |
1018                   MLX5_BITSHIFT(ETHTOOL_LINK_MODE_100000baseLR4_ER4_Full_BIT)))
1019                 priv->link_speed_capa |= ETH_LINK_SPEED_100G;
1020         dev_link.link_duplex = ((ecmd->duplex == DUPLEX_HALF) ?
1021                                 ETH_LINK_HALF_DUPLEX : ETH_LINK_FULL_DUPLEX);
1022         dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds &
1023                                   ETH_LINK_SPEED_FIXED);
1024         if (((dev_link.link_speed && !dev_link.link_status) ||
1025              (!dev_link.link_speed && dev_link.link_status))) {
1026                 rte_errno = EAGAIN;
1027                 return -rte_errno;
1028         }
1029         *link = dev_link;
1030         return 0;
1031 }
1032
1033 /**
1034  * DPDK callback to retrieve physical link information.
1035  *
1036  * @param dev
1037  *   Pointer to Ethernet device structure.
1038  * @param wait_to_complete
1039  *   Wait for request completion.
1040  *
1041  * @return
1042  *   0 if link status was not updated, positive if it was, a negative errno
1043  *   value otherwise and rte_errno is set.
1044  */
1045 int
1046 mlx5_link_update(struct rte_eth_dev *dev, int wait_to_complete)
1047 {
1048         int ret;
1049         struct rte_eth_link dev_link;
1050         time_t start_time = time(NULL);
1051
1052         do {
1053                 ret = mlx5_link_update_unlocked_gs(dev, &dev_link);
1054                 if (ret)
1055                         ret = mlx5_link_update_unlocked_gset(dev, &dev_link);
1056                 if (ret == 0)
1057                         break;
1058                 /* Handle wait to complete situation. */
1059                 if (wait_to_complete && ret == -EAGAIN) {
1060                         if (abs((int)difftime(time(NULL), start_time)) <
1061                             MLX5_LINK_STATUS_TIMEOUT) {
1062                                 usleep(0);
1063                                 continue;
1064                         } else {
1065                                 rte_errno = EBUSY;
1066                                 return -rte_errno;
1067                         }
1068                 } else if (ret < 0) {
1069                         return ret;
1070                 }
1071         } while (wait_to_complete);
1072         ret = !!memcmp(&dev->data->dev_link, &dev_link,
1073                        sizeof(struct rte_eth_link));
1074         dev->data->dev_link = dev_link;
1075         return ret;
1076 }
1077
1078 /**
1079  * DPDK callback to change the MTU.
1080  *
1081  * @param dev
1082  *   Pointer to Ethernet device structure.
1083  * @param in_mtu
1084  *   New MTU.
1085  *
1086  * @return
1087  *   0 on success, a negative errno value otherwise and rte_errno is set.
1088  */
1089 int
1090 mlx5_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
1091 {
1092         struct mlx5_priv *priv = dev->data->dev_private;
1093         uint16_t kern_mtu = 0;
1094         int ret;
1095
1096         ret = mlx5_get_mtu(dev, &kern_mtu);
1097         if (ret)
1098                 return ret;
1099         /* Set kernel interface MTU first. */
1100         ret = mlx5_set_mtu(dev, mtu);
1101         if (ret)
1102                 return ret;
1103         ret = mlx5_get_mtu(dev, &kern_mtu);
1104         if (ret)
1105                 return ret;
1106         if (kern_mtu == mtu) {
1107                 priv->mtu = mtu;
1108                 DRV_LOG(DEBUG, "port %u adapter MTU set to %u",
1109                         dev->data->port_id, mtu);
1110                 return 0;
1111         }
1112         rte_errno = EAGAIN;
1113         return -rte_errno;
1114 }
1115
1116 /**
1117  * DPDK callback to get flow control status.
1118  *
1119  * @param dev
1120  *   Pointer to Ethernet device structure.
1121  * @param[out] fc_conf
1122  *   Flow control output buffer.
1123  *
1124  * @return
1125  *   0 on success, a negative errno value otherwise and rte_errno is set.
1126  */
1127 int
1128 mlx5_dev_get_flow_ctrl(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1129 {
1130         struct ifreq ifr;
1131         struct ethtool_pauseparam ethpause = {
1132                 .cmd = ETHTOOL_GPAUSEPARAM
1133         };
1134         int ret;
1135
1136         ifr.ifr_data = (void *)&ethpause;
1137         ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
1138         if (ret) {
1139                 DRV_LOG(WARNING,
1140                         "port %u ioctl(SIOCETHTOOL, ETHTOOL_GPAUSEPARAM) failed:"
1141                         " %s",
1142                         dev->data->port_id, strerror(rte_errno));
1143                 return ret;
1144         }
1145         fc_conf->autoneg = ethpause.autoneg;
1146         if (ethpause.rx_pause && ethpause.tx_pause)
1147                 fc_conf->mode = RTE_FC_FULL;
1148         else if (ethpause.rx_pause)
1149                 fc_conf->mode = RTE_FC_RX_PAUSE;
1150         else if (ethpause.tx_pause)
1151                 fc_conf->mode = RTE_FC_TX_PAUSE;
1152         else
1153                 fc_conf->mode = RTE_FC_NONE;
1154         return 0;
1155 }
1156
1157 /**
1158  * DPDK callback to modify flow control parameters.
1159  *
1160  * @param dev
1161  *   Pointer to Ethernet device structure.
1162  * @param[in] fc_conf
1163  *   Flow control parameters.
1164  *
1165  * @return
1166  *   0 on success, a negative errno value otherwise and rte_errno is set.
1167  */
1168 int
1169 mlx5_dev_set_flow_ctrl(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1170 {
1171         struct ifreq ifr;
1172         struct ethtool_pauseparam ethpause = {
1173                 .cmd = ETHTOOL_SPAUSEPARAM
1174         };
1175         int ret;
1176
1177         ifr.ifr_data = (void *)&ethpause;
1178         ethpause.autoneg = fc_conf->autoneg;
1179         if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) ||
1180             (fc_conf->mode & RTE_FC_RX_PAUSE))
1181                 ethpause.rx_pause = 1;
1182         else
1183                 ethpause.rx_pause = 0;
1184
1185         if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) ||
1186             (fc_conf->mode & RTE_FC_TX_PAUSE))
1187                 ethpause.tx_pause = 1;
1188         else
1189                 ethpause.tx_pause = 0;
1190         ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
1191         if (ret) {
1192                 DRV_LOG(WARNING,
1193                         "port %u ioctl(SIOCETHTOOL, ETHTOOL_SPAUSEPARAM)"
1194                         " failed: %s",
1195                         dev->data->port_id, strerror(rte_errno));
1196                 return ret;
1197         }
1198         return 0;
1199 }
1200
1201 /**
1202  * Get PCI information from struct ibv_device.
1203  *
1204  * @param device
1205  *   Pointer to Ethernet device structure.
1206  * @param[out] pci_addr
1207  *   PCI bus address output buffer.
1208  *
1209  * @return
1210  *   0 on success, a negative errno value otherwise and rte_errno is set.
1211  */
1212 int
1213 mlx5_ibv_device_to_pci_addr(const struct ibv_device *device,
1214                             struct rte_pci_addr *pci_addr)
1215 {
1216         FILE *file;
1217         char line[32];
1218         MKSTR(path, "%s/device/uevent", device->ibdev_path);
1219
1220         file = fopen(path, "rb");
1221         if (file == NULL) {
1222                 rte_errno = errno;
1223                 return -rte_errno;
1224         }
1225         while (fgets(line, sizeof(line), file) == line) {
1226                 size_t len = strlen(line);
1227                 int ret;
1228
1229                 /* Truncate long lines. */
1230                 if (len == (sizeof(line) - 1))
1231                         while (line[(len - 1)] != '\n') {
1232                                 ret = fgetc(file);
1233                                 if (ret == EOF)
1234                                         break;
1235                                 line[(len - 1)] = ret;
1236                         }
1237                 /* Extract information. */
1238                 if (sscanf(line,
1239                            "PCI_SLOT_NAME="
1240                            "%" SCNx32 ":%" SCNx8 ":%" SCNx8 ".%" SCNx8 "\n",
1241                            &pci_addr->domain,
1242                            &pci_addr->bus,
1243                            &pci_addr->devid,
1244                            &pci_addr->function) == 4) {
1245                         ret = 0;
1246                         break;
1247                 }
1248         }
1249         fclose(file);
1250         return 0;
1251 }
1252
1253 /**
1254  * Handle asynchronous removal event for entire multiport device.
1255  *
1256  * @param sh
1257  *   Infiniband device shared context.
1258  */
1259 static void
1260 mlx5_dev_interrupt_device_fatal(struct mlx5_ibv_shared *sh)
1261 {
1262         uint32_t i;
1263
1264         for (i = 0; i < sh->max_port; ++i) {
1265                 struct rte_eth_dev *dev;
1266
1267                 if (sh->port[i].ih_port_id >= RTE_MAX_ETHPORTS) {
1268                         /*
1269                          * Or not existing port either no
1270                          * handler installed for this port.
1271                          */
1272                         continue;
1273                 }
1274                 dev = &rte_eth_devices[sh->port[i].ih_port_id];
1275                 assert(dev);
1276                 if (dev->data->dev_conf.intr_conf.rmv)
1277                         _rte_eth_dev_callback_process
1278                                 (dev, RTE_ETH_EVENT_INTR_RMV, NULL);
1279         }
1280 }
1281
1282 /**
1283  * Handle shared asynchronous events the NIC (removal event
1284  * and link status change). Supports multiport IB device.
1285  *
1286  * @param cb_arg
1287  *   Callback argument.
1288  */
1289 void
1290 mlx5_dev_interrupt_handler(void *cb_arg)
1291 {
1292         struct mlx5_ibv_shared *sh = cb_arg;
1293         struct ibv_async_event event;
1294
1295         /* Read all message from the IB device and acknowledge them. */
1296         for (;;) {
1297                 struct rte_eth_dev *dev;
1298                 uint32_t tmp;
1299
1300                 if (mlx5_glue->get_async_event(sh->ctx, &event))
1301                         break;
1302                 /* Retrieve and check IB port index. */
1303                 tmp = (uint32_t)event.element.port_num;
1304                 if (!tmp && event.event_type == IBV_EVENT_DEVICE_FATAL) {
1305                         /*
1306                          * The DEVICE_FATAL event is called once for
1307                          * entire device without port specifying.
1308                          * We should notify all existing ports.
1309                          */
1310                         mlx5_glue->ack_async_event(&event);
1311                         mlx5_dev_interrupt_device_fatal(sh);
1312                         continue;
1313                 }
1314                 assert(tmp && (tmp <= sh->max_port));
1315                 if (!tmp) {
1316                         /* Unsupported devive level event. */
1317                         mlx5_glue->ack_async_event(&event);
1318                         DRV_LOG(DEBUG,
1319                                 "unsupported common event (type %d)",
1320                                 event.event_type);
1321                         continue;
1322                 }
1323                 if (tmp > sh->max_port) {
1324                         /* Invalid IB port index. */
1325                         mlx5_glue->ack_async_event(&event);
1326                         DRV_LOG(DEBUG,
1327                                 "cannot handle an event (type %d)"
1328                                 "due to invalid IB port index (%u)",
1329                                 event.event_type, tmp);
1330                         continue;
1331                 }
1332                 if (sh->port[tmp - 1].ih_port_id >= RTE_MAX_ETHPORTS) {
1333                         /* No handler installed. */
1334                         mlx5_glue->ack_async_event(&event);
1335                         DRV_LOG(DEBUG,
1336                                 "cannot handle an event (type %d)"
1337                                 "due to no handler installed for port %u",
1338                                 event.event_type, tmp);
1339                         continue;
1340                 }
1341                 /* Retrieve ethernet device descriptor. */
1342                 tmp = sh->port[tmp - 1].ih_port_id;
1343                 dev = &rte_eth_devices[tmp];
1344                 assert(dev);
1345                 if ((event.event_type == IBV_EVENT_PORT_ACTIVE ||
1346                      event.event_type == IBV_EVENT_PORT_ERR) &&
1347                         dev->data->dev_conf.intr_conf.lsc) {
1348                         mlx5_glue->ack_async_event(&event);
1349                         if (mlx5_link_update(dev, 0) == -EAGAIN) {
1350                                 usleep(0);
1351                                 continue;
1352                         }
1353                         _rte_eth_dev_callback_process
1354                                 (dev, RTE_ETH_EVENT_INTR_LSC, NULL);
1355                         continue;
1356                 }
1357                 DRV_LOG(DEBUG,
1358                         "port %u cannot handle an unknown event (type %d)",
1359                         dev->data->port_id, event.event_type);
1360                 mlx5_glue->ack_async_event(&event);
1361         }
1362 }
1363
1364 /*
1365  * Unregister callback handler safely. The handler may be active
1366  * while we are trying to unregister it, in this case code -EAGAIN
1367  * is returned by rte_intr_callback_unregister(). This routine checks
1368  * the return code and tries to unregister handler again.
1369  *
1370  * @param handle
1371  *   interrupt handle
1372  * @param cb_fn
1373  *   pointer to callback routine
1374  * @cb_arg
1375  *   opaque callback parameter
1376  */
1377 void
1378 mlx5_intr_callback_unregister(const struct rte_intr_handle *handle,
1379                               rte_intr_callback_fn cb_fn, void *cb_arg)
1380 {
1381         /*
1382          * Try to reduce timeout management overhead by not calling
1383          * the timer related routines on the first iteration. If the
1384          * unregistering succeeds on first call there will be no
1385          * timer calls at all.
1386          */
1387         uint64_t twait = 0;
1388         uint64_t start = 0;
1389
1390         do {
1391                 int ret;
1392
1393                 ret = rte_intr_callback_unregister(handle, cb_fn, cb_arg);
1394                 if (ret >= 0)
1395                         return;
1396                 if (ret != -EAGAIN) {
1397                         DRV_LOG(INFO, "failed to unregister interrupt"
1398                                       " handler (error: %d)", ret);
1399                         assert(false);
1400                         return;
1401                 }
1402                 if (twait) {
1403                         struct timespec onems;
1404
1405                         /* Wait one millisecond and try again. */
1406                         onems.tv_sec = 0;
1407                         onems.tv_nsec = NS_PER_S / MS_PER_S;
1408                         nanosleep(&onems, 0);
1409                         /* Check whether one second elapsed. */
1410                         if ((rte_get_timer_cycles() - start) <= twait)
1411                                 continue;
1412                 } else {
1413                         /*
1414                          * We get the amount of timer ticks for one second.
1415                          * If this amount elapsed it means we spent one
1416                          * second in waiting. This branch is executed once
1417                          * on first iteration.
1418                          */
1419                         twait = rte_get_timer_hz();
1420                         assert(twait);
1421                 }
1422                 /*
1423                  * Timeout elapsed, show message (once a second) and retry.
1424                  * We have no other acceptable option here, if we ignore
1425                  * the unregistering return code the handler will not
1426                  * be unregistered, fd will be closed and we may get the
1427                  * crush. Hanging and messaging in the loop seems not to be
1428                  * the worst choice.
1429                  */
1430                 DRV_LOG(INFO, "Retrying to unregister interrupt handler");
1431                 start = rte_get_timer_cycles();
1432         } while (true);
1433 }
1434
1435 /**
1436  * Uninstall shared asynchronous device events handler.
1437  * This function is implemented to support event sharing
1438  * between multiple ports of single IB device.
1439  *
1440  * @param dev
1441  *   Pointer to Ethernet device.
1442  */
1443 static void
1444 mlx5_dev_shared_handler_uninstall(struct rte_eth_dev *dev)
1445 {
1446         struct mlx5_priv *priv = dev->data->dev_private;
1447         struct mlx5_ibv_shared *sh = priv->sh;
1448
1449         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1450                 return;
1451         pthread_mutex_lock(&sh->intr_mutex);
1452         assert(priv->ibv_port);
1453         assert(priv->ibv_port <= sh->max_port);
1454         assert(dev->data->port_id < RTE_MAX_ETHPORTS);
1455         if (sh->port[priv->ibv_port - 1].ih_port_id >= RTE_MAX_ETHPORTS)
1456                 goto exit;
1457         assert(sh->port[priv->ibv_port - 1].ih_port_id ==
1458                                         (uint32_t)dev->data->port_id);
1459         assert(sh->intr_cnt);
1460         sh->port[priv->ibv_port - 1].ih_port_id = RTE_MAX_ETHPORTS;
1461         if (!sh->intr_cnt || --sh->intr_cnt)
1462                 goto exit;
1463         mlx5_intr_callback_unregister(&sh->intr_handle,
1464                                      mlx5_dev_interrupt_handler, sh);
1465         sh->intr_handle.fd = 0;
1466         sh->intr_handle.type = RTE_INTR_HANDLE_UNKNOWN;
1467 exit:
1468         pthread_mutex_unlock(&sh->intr_mutex);
1469 }
1470
1471 /**
1472  * Install shared asynchronous device events handler.
1473  * This function is implemented to support event sharing
1474  * between multiple ports of single IB device.
1475  *
1476  * @param dev
1477  *   Pointer to Ethernet device.
1478  */
1479 static void
1480 mlx5_dev_shared_handler_install(struct rte_eth_dev *dev)
1481 {
1482         struct mlx5_priv *priv = dev->data->dev_private;
1483         struct mlx5_ibv_shared *sh = priv->sh;
1484         int ret;
1485         int flags;
1486
1487         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1488                 return;
1489         pthread_mutex_lock(&sh->intr_mutex);
1490         assert(priv->ibv_port);
1491         assert(priv->ibv_port <= sh->max_port);
1492         assert(dev->data->port_id < RTE_MAX_ETHPORTS);
1493         if (sh->port[priv->ibv_port - 1].ih_port_id < RTE_MAX_ETHPORTS) {
1494                 /* The handler is already installed for this port. */
1495                 assert(sh->intr_cnt);
1496                 goto exit;
1497         }
1498         sh->port[priv->ibv_port - 1].ih_port_id = (uint32_t)dev->data->port_id;
1499         if (sh->intr_cnt) {
1500                 sh->intr_cnt++;
1501                 goto exit;
1502         }
1503         /* No shared handler installed. */
1504         assert(sh->ctx->async_fd > 0);
1505         flags = fcntl(sh->ctx->async_fd, F_GETFL);
1506         ret = fcntl(sh->ctx->async_fd, F_SETFL, flags | O_NONBLOCK);
1507         if (ret) {
1508                 DRV_LOG(INFO, "failed to change file descriptor"
1509                               " async event queue");
1510                 /* Indicate there will be no interrupts. */
1511                 dev->data->dev_conf.intr_conf.lsc = 0;
1512                 dev->data->dev_conf.intr_conf.rmv = 0;
1513                 sh->port[priv->ibv_port - 1].ih_port_id = RTE_MAX_ETHPORTS;
1514                 goto exit;
1515         }
1516         sh->intr_handle.fd = sh->ctx->async_fd;
1517         sh->intr_handle.type = RTE_INTR_HANDLE_EXT;
1518         rte_intr_callback_register(&sh->intr_handle,
1519                                    mlx5_dev_interrupt_handler, sh);
1520         sh->intr_cnt++;
1521 exit:
1522         pthread_mutex_unlock(&sh->intr_mutex);
1523 }
1524
1525 /**
1526  * Uninstall interrupt handler.
1527  *
1528  * @param dev
1529  *   Pointer to Ethernet device.
1530  */
1531 void
1532 mlx5_dev_interrupt_handler_uninstall(struct rte_eth_dev *dev)
1533 {
1534         mlx5_dev_shared_handler_uninstall(dev);
1535 }
1536
1537 /**
1538  * Install interrupt handler.
1539  *
1540  * @param dev
1541  *   Pointer to Ethernet device.
1542  */
1543 void
1544 mlx5_dev_interrupt_handler_install(struct rte_eth_dev *dev)
1545 {
1546         mlx5_dev_shared_handler_install(dev);
1547 }
1548
1549 /**
1550  * DPDK callback to bring the link DOWN.
1551  *
1552  * @param dev
1553  *   Pointer to Ethernet device structure.
1554  *
1555  * @return
1556  *   0 on success, a negative errno value otherwise and rte_errno is set.
1557  */
1558 int
1559 mlx5_set_link_down(struct rte_eth_dev *dev)
1560 {
1561         return mlx5_set_flags(dev, ~IFF_UP, ~IFF_UP);
1562 }
1563
1564 /**
1565  * DPDK callback to bring the link UP.
1566  *
1567  * @param dev
1568  *   Pointer to Ethernet device structure.
1569  *
1570  * @return
1571  *   0 on success, a negative errno value otherwise and rte_errno is set.
1572  */
1573 int
1574 mlx5_set_link_up(struct rte_eth_dev *dev)
1575 {
1576         return mlx5_set_flags(dev, ~IFF_UP, IFF_UP);
1577 }
1578
1579 /**
1580  * Configure the TX function to use.
1581  *
1582  * @param dev
1583  *   Pointer to private data structure.
1584  *
1585  * @return
1586  *   Pointer to selected Tx burst function.
1587  */
1588 eth_tx_burst_t
1589 mlx5_select_tx_function(struct rte_eth_dev *dev)
1590 {
1591         struct mlx5_priv *priv = dev->data->dev_private;
1592         eth_tx_burst_t tx_pkt_burst = mlx5_tx_burst;
1593         struct mlx5_dev_config *config = &priv->config;
1594         uint64_t tx_offloads = dev->data->dev_conf.txmode.offloads;
1595         int tso = !!(tx_offloads & (DEV_TX_OFFLOAD_TCP_TSO |
1596                                     DEV_TX_OFFLOAD_VXLAN_TNL_TSO |
1597                                     DEV_TX_OFFLOAD_GRE_TNL_TSO |
1598                                     DEV_TX_OFFLOAD_IP_TNL_TSO |
1599                                     DEV_TX_OFFLOAD_UDP_TNL_TSO));
1600         int swp = !!(tx_offloads & (DEV_TX_OFFLOAD_IP_TNL_TSO |
1601                                     DEV_TX_OFFLOAD_UDP_TNL_TSO |
1602                                     DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM));
1603         int vlan_insert = !!(tx_offloads & DEV_TX_OFFLOAD_VLAN_INSERT);
1604
1605         assert(priv != NULL);
1606         /* Select appropriate TX function. */
1607         if (vlan_insert || tso || swp)
1608                 return tx_pkt_burst;
1609         if (config->mps == MLX5_MPW_ENHANCED) {
1610                 if (mlx5_check_vec_tx_support(dev) > 0) {
1611                         if (mlx5_check_raw_vec_tx_support(dev) > 0)
1612                                 tx_pkt_burst = mlx5_tx_burst_raw_vec;
1613                         else
1614                                 tx_pkt_burst = mlx5_tx_burst_vec;
1615                         DRV_LOG(DEBUG,
1616                                 "port %u selected enhanced MPW Tx vectorized"
1617                                 " function",
1618                                 dev->data->port_id);
1619                 } else {
1620                         tx_pkt_burst = mlx5_tx_burst_empw;
1621                         DRV_LOG(DEBUG,
1622                                 "port %u selected enhanced MPW Tx function",
1623                                 dev->data->port_id);
1624                 }
1625         } else if (config->mps && (config->txq_inline > 0)) {
1626                 tx_pkt_burst = mlx5_tx_burst_mpw_inline;
1627                 DRV_LOG(DEBUG, "port %u selected MPW inline Tx function",
1628                         dev->data->port_id);
1629         } else if (config->mps) {
1630                 tx_pkt_burst = mlx5_tx_burst_mpw;
1631                 DRV_LOG(DEBUG, "port %u selected MPW Tx function",
1632                         dev->data->port_id);
1633         }
1634         return tx_pkt_burst;
1635 }
1636
1637 /**
1638  * Configure the RX function to use.
1639  *
1640  * @param dev
1641  *   Pointer to private data structure.
1642  *
1643  * @return
1644  *   Pointer to selected Rx burst function.
1645  */
1646 eth_rx_burst_t
1647 mlx5_select_rx_function(struct rte_eth_dev *dev)
1648 {
1649         eth_rx_burst_t rx_pkt_burst = mlx5_rx_burst;
1650
1651         assert(dev != NULL);
1652         if (mlx5_check_vec_rx_support(dev) > 0) {
1653                 rx_pkt_burst = mlx5_rx_burst_vec;
1654                 DRV_LOG(DEBUG, "port %u selected Rx vectorized function",
1655                         dev->data->port_id);
1656         } else if (mlx5_mprq_enabled(dev)) {
1657                 rx_pkt_burst = mlx5_rx_burst_mprq;
1658         }
1659         return rx_pkt_burst;
1660 }
1661
1662 /**
1663  * Check if mlx5 device was removed.
1664  *
1665  * @param dev
1666  *   Pointer to Ethernet device structure.
1667  *
1668  * @return
1669  *   1 when device is removed, otherwise 0.
1670  */
1671 int
1672 mlx5_is_removed(struct rte_eth_dev *dev)
1673 {
1674         struct ibv_device_attr device_attr;
1675         struct mlx5_priv *priv = dev->data->dev_private;
1676
1677         if (mlx5_glue->query_device(priv->sh->ctx, &device_attr) == EIO)
1678                 return 1;
1679         return 0;
1680 }
1681
1682 /**
1683  * Get port ID list of mlx5 instances sharing a common device.
1684  *
1685  * @param[in] dev
1686  *   Device to look for.
1687  * @param[out] port_list
1688  *   Result buffer for collected port IDs.
1689  * @param port_list_n
1690  *   Maximum number of entries in result buffer. If 0, @p port_list can be
1691  *   NULL.
1692  *
1693  * @return
1694  *   Number of matching instances regardless of the @p port_list_n
1695  *   parameter, 0 if none were found.
1696  */
1697 unsigned int
1698 mlx5_dev_to_port_id(const struct rte_device *dev, uint16_t *port_list,
1699                     unsigned int port_list_n)
1700 {
1701         uint16_t id;
1702         unsigned int n = 0;
1703
1704         RTE_ETH_FOREACH_DEV_OF(id, dev) {
1705                 if (n < port_list_n)
1706                         port_list[n] = id;
1707                 n++;
1708         }
1709         return n;
1710 }
1711
1712 /**
1713  * Get the E-Switch domain id this port belongs to.
1714  *
1715  * @param[in] port
1716  *   Device port id.
1717  * @param[out] es_domain_id
1718  *   E-Switch domain id.
1719  * @param[out] es_port_id
1720  *   The port id of the port in the E-Switch.
1721  *
1722  * @return
1723  *   0 on success, a negative errno value otherwise and rte_errno is set.
1724  */
1725 int
1726 mlx5_port_to_eswitch_info(uint16_t port,
1727                           uint16_t *es_domain_id, uint16_t *es_port_id)
1728 {
1729         struct rte_eth_dev *dev;
1730         struct mlx5_priv *priv;
1731
1732         if (port >= RTE_MAX_ETHPORTS) {
1733                 rte_errno = EINVAL;
1734                 return -rte_errno;
1735         }
1736         if (!rte_eth_dev_is_valid_port(port)) {
1737                 rte_errno = ENODEV;
1738                 return -rte_errno;
1739         }
1740         dev = &rte_eth_devices[port];
1741         priv = dev->data->dev_private;
1742         if (!(priv->representor || priv->master)) {
1743                 rte_errno = EINVAL;
1744                 return -rte_errno;
1745         }
1746         if (es_domain_id)
1747                 *es_domain_id = priv->domain_id;
1748         if (es_port_id)
1749                 *es_port_id = priv->vport_id;
1750         return 0;
1751 }
1752
1753 /**
1754  * Get switch information associated with network interface.
1755  *
1756  * @param ifindex
1757  *   Network interface index.
1758  * @param[out] info
1759  *   Switch information object, populated in case of success.
1760  *
1761  * @return
1762  *   0 on success, a negative errno value otherwise and rte_errno is set.
1763  */
1764 int
1765 mlx5_sysfs_switch_info(unsigned int ifindex, struct mlx5_switch_info *info)
1766 {
1767         char ifname[IF_NAMESIZE];
1768         char port_name[IF_NAMESIZE];
1769         FILE *file;
1770         struct mlx5_switch_info data = {
1771                 .master = 0,
1772                 .representor = 0,
1773                 .name_type = MLX5_PHYS_PORT_NAME_TYPE_NOTSET,
1774                 .port_name = 0,
1775                 .switch_id = 0,
1776         };
1777         DIR *dir;
1778         bool port_switch_id_set = false;
1779         bool device_dir = false;
1780         char c;
1781         int ret;
1782
1783         if (!if_indextoname(ifindex, ifname)) {
1784                 rte_errno = errno;
1785                 return -rte_errno;
1786         }
1787
1788         MKSTR(phys_port_name, "/sys/class/net/%s/phys_port_name",
1789               ifname);
1790         MKSTR(phys_switch_id, "/sys/class/net/%s/phys_switch_id",
1791               ifname);
1792         MKSTR(pci_device, "/sys/class/net/%s/device",
1793               ifname);
1794
1795         file = fopen(phys_port_name, "rb");
1796         if (file != NULL) {
1797                 ret = fscanf(file, "%s", port_name);
1798                 fclose(file);
1799                 if (ret == 1)
1800                         mlx5_translate_port_name(port_name, &data);
1801         }
1802         file = fopen(phys_switch_id, "rb");
1803         if (file == NULL) {
1804                 rte_errno = errno;
1805                 return -rte_errno;
1806         }
1807         port_switch_id_set =
1808                 fscanf(file, "%" SCNx64 "%c", &data.switch_id, &c) == 2 &&
1809                 c == '\n';
1810         fclose(file);
1811         dir = opendir(pci_device);
1812         if (dir != NULL) {
1813                 closedir(dir);
1814                 device_dir = true;
1815         }
1816         if (port_switch_id_set) {
1817                 /* We have some E-Switch configuration. */
1818                 mlx5_sysfs_check_switch_info(device_dir, &data);
1819         }
1820         *info = data;
1821         assert(!(data.master && data.representor));
1822         if (data.master && data.representor) {
1823                 DRV_LOG(ERR, "ifindex %u device is recognized as master"
1824                              " and as representor", ifindex);
1825                 rte_errno = ENODEV;
1826                 return -rte_errno;
1827         }
1828         return 0;
1829 }
1830
1831 /**
1832  * Analyze gathered port parameters via Netlink to recognize master
1833  * and representor devices for E-Switch configuration.
1834  *
1835  * @param[in] num_vf_set
1836  *   flag of presence of number of VFs port attribute.
1837  * @param[inout] switch_info
1838  *   Port information, including port name as a number and port name
1839  *   type if recognized
1840  *
1841  * @return
1842  *   master and representor flags are set in switch_info according to
1843  *   recognized parameters (if any).
1844  */
1845 void
1846 mlx5_nl_check_switch_info(bool num_vf_set,
1847                           struct mlx5_switch_info *switch_info)
1848 {
1849         switch (switch_info->name_type) {
1850         case MLX5_PHYS_PORT_NAME_TYPE_UNKNOWN:
1851                 /*
1852                  * Name is not recognized, assume the master,
1853                  * check the number of VFs key presence.
1854                  */
1855                 switch_info->master = num_vf_set;
1856                 break;
1857         case MLX5_PHYS_PORT_NAME_TYPE_NOTSET:
1858                 /*
1859                  * Name is not set, this assumes the legacy naming
1860                  * schema for master, just check if there is a
1861                  * number of VFs key.
1862                  */
1863                 switch_info->master = num_vf_set;
1864                 break;
1865         case MLX5_PHYS_PORT_NAME_TYPE_UPLINK:
1866                 /* New uplink naming schema recognized. */
1867                 switch_info->master = 1;
1868                 break;
1869         case MLX5_PHYS_PORT_NAME_TYPE_LEGACY:
1870                 /* Legacy representors naming schema. */
1871                 switch_info->representor = !num_vf_set;
1872                 break;
1873         case MLX5_PHYS_PORT_NAME_TYPE_PFVF:
1874                 /* New representors naming schema. */
1875                 switch_info->representor = 1;
1876                 break;
1877         }
1878 }
1879
1880 /**
1881  * Analyze gathered port parameters via sysfs to recognize master
1882  * and representor devices for E-Switch configuration.
1883  *
1884  * @param[in] device_dir
1885  *   flag of presence of "device" directory under port device key.
1886  * @param[inout] switch_info
1887  *   Port information, including port name as a number and port name
1888  *   type if recognized
1889  *
1890  * @return
1891  *   master and representor flags are set in switch_info according to
1892  *   recognized parameters (if any).
1893  */
1894 void
1895 mlx5_sysfs_check_switch_info(bool device_dir,
1896                              struct mlx5_switch_info *switch_info)
1897 {
1898         switch (switch_info->name_type) {
1899         case MLX5_PHYS_PORT_NAME_TYPE_UNKNOWN:
1900                 /*
1901                  * Name is not recognized, assume the master,
1902                  * check the device directory presence.
1903                  */
1904                 switch_info->master = device_dir;
1905                 break;
1906         case MLX5_PHYS_PORT_NAME_TYPE_NOTSET:
1907                 /*
1908                  * Name is not set, this assumes the legacy naming
1909                  * schema for master, just check if there is
1910                  * a device directory.
1911                  */
1912                 switch_info->master = device_dir;
1913                 break;
1914         case MLX5_PHYS_PORT_NAME_TYPE_UPLINK:
1915                 /* New uplink naming schema recognized. */
1916                 switch_info->master = 1;
1917                 break;
1918         case MLX5_PHYS_PORT_NAME_TYPE_LEGACY:
1919                 /* Legacy representors naming schema. */
1920                 switch_info->representor = !device_dir;
1921                 break;
1922         case MLX5_PHYS_PORT_NAME_TYPE_PFVF:
1923                 /* New representors naming schema. */
1924                 switch_info->representor = 1;
1925                 break;
1926         }
1927 }
1928
1929 /**
1930  * Extract port name, as a number, from sysfs or netlink information.
1931  *
1932  * @param[in] port_name_in
1933  *   String representing the port name.
1934  * @param[out] port_info_out
1935  *   Port information, including port name as a number and port name
1936  *   type if recognized
1937  *
1938  * @return
1939  *   port_name field set according to recognized name format.
1940  */
1941 void
1942 mlx5_translate_port_name(const char *port_name_in,
1943                          struct mlx5_switch_info *port_info_out)
1944 {
1945         char pf_c1, pf_c2, vf_c1, vf_c2;
1946         char *end;
1947         int sc_items;
1948
1949         /*
1950          * Check for port-name as a string of the form pf0vf0
1951          * (support kernel ver >= 5.0 or OFED ver >= 4.6).
1952          */
1953         sc_items = sscanf(port_name_in, "%c%c%d%c%c%d",
1954                           &pf_c1, &pf_c2, &port_info_out->pf_num,
1955                           &vf_c1, &vf_c2, &port_info_out->port_name);
1956         if (sc_items == 6 &&
1957             pf_c1 == 'p' && pf_c2 == 'f' &&
1958             vf_c1 == 'v' && vf_c2 == 'f') {
1959                 port_info_out->name_type = MLX5_PHYS_PORT_NAME_TYPE_PFVF;
1960                 return;
1961         }
1962         /*
1963          * Check for port-name as a string of the form p0
1964          * (support kernel ver >= 5.0, or OFED ver >= 4.6).
1965          */
1966         sc_items = sscanf(port_name_in, "%c%d",
1967                           &pf_c1, &port_info_out->port_name);
1968         if (sc_items == 2 && pf_c1 == 'p') {
1969                 port_info_out->name_type = MLX5_PHYS_PORT_NAME_TYPE_UPLINK;
1970                 return;
1971         }
1972         /* Check for port-name as a number (support kernel ver < 5.0 */
1973         errno = 0;
1974         port_info_out->port_name = strtol(port_name_in, &end, 0);
1975         if (!errno &&
1976             (size_t)(end - port_name_in) == strlen(port_name_in)) {
1977                 port_info_out->name_type = MLX5_PHYS_PORT_NAME_TYPE_LEGACY;
1978                 return;
1979         }
1980         port_info_out->name_type = MLX5_PHYS_PORT_NAME_TYPE_UNKNOWN;
1981         return;
1982 }