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 = mlx5_ifindex(dev);
 229         if (!ifindex) {
 230                 if (!priv->representor)
 231                         return mlx5_get_master_ifname(priv->sh->ibdev_path,
 232                                                       ifname);
 233                 rte_errno = ENXIO;
 234                 return -rte_errno;
 235         }
 236         if (if_indextoname(ifindex, &(*ifname)[0]))
 237                 return 0;
 238         rte_errno = errno;
 239         return -rte_errno;
 240 }
 241
 242 /**
 243  * Get the interface index from device name.
 244  *
 245  * @param[in] dev
 246  *   Pointer to Ethernet device.
 247  *
 248  * @return
 249  *   Nonzero interface index on success, zero otherwise and rte_errno is set.
 250  */
 251 unsigned int
 252 mlx5_ifindex(const struct rte_eth_dev *dev)
 253 {
 254         struct mlx5_priv *priv = dev->data->dev_private;
 255         unsigned int ifindex;
 256
 257         assert(priv);
 258         assert(priv->if_index);
 259         ifindex = priv->if_index;
 260         if (!ifindex)
 261                 rte_errno = ENXIO;
 262         return ifindex;
 263 }
 264
 265 /**
 266  * Perform ifreq ioctl() on associated Ethernet device.
 267  *
 268  * @param[in] dev
 269  *   Pointer to Ethernet device.
 270  * @param req
 271  *   Request number to pass to ioctl().
 272  * @param[out] ifr
 273  *   Interface request structure output buffer.
 274  *
 275  * @return
 276  *   0 on success, a negative errno value otherwise and rte_errno is set.
 277  */
 278 int
 279 mlx5_ifreq(const struct rte_eth_dev *dev, int req, struct ifreq *ifr)
 280 {
 281         int sock = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
 282         int ret = 0;
 283
 284         if (sock == -1) {
 285                 rte_errno = errno;
 286                 return -rte_errno;
 287         }
 288         ret = mlx5_get_ifname(dev, &ifr->ifr_name);
 289         if (ret)
 290                 goto error;
 291         ret = ioctl(sock, req, ifr);
 292         if (ret == -1) {
 293                 rte_errno = errno;
 294                 goto error;
 295         }
 296         close(sock);
 297         return 0;
 298 error:
 299         close(sock);
 300         return -rte_errno;
 301 }
 302
 303 /**
 304  * Get device MTU.
 305  *
 306  * @param dev
 307  *   Pointer to Ethernet device.
 308  * @param[out] mtu
 309  *   MTU value output buffer.
 310  *
 311  * @return
 312  *   0 on success, a negative errno value otherwise and rte_errno is set.
 313  */
 314 int
 315 mlx5_get_mtu(struct rte_eth_dev *dev, uint16_t *mtu)
 316 {
 317         struct ifreq request;
 318         int ret = mlx5_ifreq(dev, SIOCGIFMTU, &request);
 319
 320         if (ret)
 321                 return ret;
 322         *mtu = request.ifr_mtu;
 323         return 0;
 324 }
 325
 326 /**
 327  * Set device MTU.
 328  *
 329  * @param dev
 330  *   Pointer to Ethernet device.
 331  * @param mtu
 332  *   MTU value to set.
 333  *
 334  * @return
 335  *   0 on success, a negative errno value otherwise and rte_errno is set.
 336  */
 337 static int
 338 mlx5_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
 339 {
 340         struct ifreq request = { .ifr_mtu = mtu, };
 341
 342         return mlx5_ifreq(dev, SIOCSIFMTU, &request);
 343 }
 344
 345 /**
 346  * Set device flags.
 347  *
 348  * @param dev
 349  *   Pointer to Ethernet device.
 350  * @param keep
 351  *   Bitmask for flags that must remain untouched.
 352  * @param flags
 353  *   Bitmask for flags to modify.
 354  *
 355  * @return
 356  *   0 on success, a negative errno value otherwise and rte_errno is set.
 357  */
 358 int
 359 mlx5_set_flags(struct rte_eth_dev *dev, unsigned int keep, unsigned int flags)
 360 {
 361         struct ifreq request;
 362         int ret = mlx5_ifreq(dev, SIOCGIFFLAGS, &request);
 363
 364         if (ret)
 365                 return ret;
 366         request.ifr_flags &= keep;
 367         request.ifr_flags |= flags & ~keep;
 368         return mlx5_ifreq(dev, SIOCSIFFLAGS, &request);
 369 }
 370
 371 /**
 372  * DPDK callback for Ethernet device configuration.
 373  *
 374  * @param dev
 375  *   Pointer to Ethernet device structure.
 376  *
 377  * @return
 378  *   0 on success, a negative errno value otherwise and rte_errno is set.
 379  */
 380 int
 381 mlx5_dev_configure(struct rte_eth_dev *dev)
 382 {
 383         struct mlx5_priv *priv = dev->data->dev_private;
 384         unsigned int rxqs_n = dev->data->nb_rx_queues;
 385         unsigned int txqs_n = dev->data->nb_tx_queues;
 386         unsigned int i;
 387         unsigned int j;
 388         unsigned int reta_idx_n;
 389         const uint8_t use_app_rss_key =
 390                 !!dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key;
 391         int ret = 0;
 392         unsigned int lro_on = mlx5_lro_on(dev);
 393
 394         if (use_app_rss_key &&
 395             (dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key_len !=
 396              MLX5_RSS_HASH_KEY_LEN)) {
 397                 DRV_LOG(ERR, "port %u RSS key len must be %s Bytes long",
 398                         dev->data->port_id, RTE_STR(MLX5_RSS_HASH_KEY_LEN));
 399                 rte_errno = EINVAL;
 400                 return -rte_errno;
 401         }
 402         priv->rss_conf.rss_key =
 403                 rte_realloc(priv->rss_conf.rss_key,
 404                             MLX5_RSS_HASH_KEY_LEN, 0);
 405         if (!priv->rss_conf.rss_key) {
 406                 DRV_LOG(ERR, "port %u cannot allocate RSS hash key memory (%u)",
 407                         dev->data->port_id, rxqs_n);
 408                 rte_errno = ENOMEM;
 409                 return -rte_errno;
 410         }
 411         memcpy(priv->rss_conf.rss_key,
 412                use_app_rss_key ?
 413                dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key :
 414                rss_hash_default_key,
 415                MLX5_RSS_HASH_KEY_LEN);
 416         priv->rss_conf.rss_key_len = MLX5_RSS_HASH_KEY_LEN;
 417         priv->rss_conf.rss_hf = dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf;
 418         priv->rxqs = (void *)dev->data->rx_queues;
 419         priv->txqs = (void *)dev->data->tx_queues;
 420         if (txqs_n != priv->txqs_n) {
 421                 DRV_LOG(INFO, "port %u Tx queues number update: %u -> %u",
 422                         dev->data->port_id, priv->txqs_n, txqs_n);
 423                 priv->txqs_n = txqs_n;
 424         }
 425         if (rxqs_n > priv->config.ind_table_max_size) {
 426                 DRV_LOG(ERR, "port %u cannot handle this many Rx queues (%u)",
 427                         dev->data->port_id, rxqs_n);
 428                 rte_errno = EINVAL;
 429                 return -rte_errno;
 430         }
 431         if (rxqs_n != priv->rxqs_n) {
 432                 DRV_LOG(INFO, "port %u Rx queues number update: %u -> %u",
 433                         dev->data->port_id, priv->rxqs_n, rxqs_n);
 434                 priv->rxqs_n = rxqs_n;
 435                 /*
 436                  * If the requested number of RX queues is not a power of two,
 437                  * use the maximum indirection table size for better balancing.
 438                  * The result is always rounded to the next power of two.
 439                  */
 440                 reta_idx_n = (1 << log2above((rxqs_n & (rxqs_n - 1)) ?
 441                                              priv->config.ind_table_max_size :
 442                                              rxqs_n));
 443                 ret = mlx5_rss_reta_index_resize(dev, reta_idx_n);
 444                 if (ret)
 445                         return ret;
 446                 /*
 447                  * When the number of RX queues is not a power of two,
 448                  * the remaining table entries are padded with reused WQs
 449                  * and hashes are not spread uniformly.
 450                  */
 451                 for (i = 0, j = 0; (i != reta_idx_n); ++i) {
 452                         (*priv->reta_idx)[i] = j;
 453                         if (++j == rxqs_n)
 454                                 j = 0;
 455                 }
 456         }
 457         if (lro_on && priv->config.cqe_comp) {
 458                 /* CQE compressing is not supported for LRO CQEs. */
 459                 DRV_LOG(WARNING, "Rx CQE compression isn't supported with LRO");
 460                 priv->config.cqe_comp = 0;
 461         }
 462         ret = mlx5_proc_priv_init(dev);
 463         if (ret)
 464                 return ret;
 465         return 0;
 466 }
 467
 468 /**
 469  * Sets default tuning parameters.
 470  *
 471  * @param dev
 472  *   Pointer to Ethernet device.
 473  * @param[out] info
 474  *   Info structure output buffer.
 475  */
 476 static void
 477 mlx5_set_default_params(struct rte_eth_dev *dev, struct rte_eth_dev_info *info)
 478 {
 479         struct mlx5_priv *priv = dev->data->dev_private;
 480
 481         /* Minimum CPU utilization. */
 482         info->default_rxportconf.ring_size = 256;
 483         info->default_txportconf.ring_size = 256;
 484         info->default_rxportconf.burst_size = 64;
 485         info->default_txportconf.burst_size = 64;
 486         if (priv->link_speed_capa & ETH_LINK_SPEED_100G) {
 487                 info->default_rxportconf.nb_queues = 16;
 488                 info->default_txportconf.nb_queues = 16;
 489                 if (dev->data->nb_rx_queues > 2 ||
 490                     dev->data->nb_tx_queues > 2) {
 491                         /* Max Throughput. */
 492                         info->default_rxportconf.ring_size = 2048;
 493                         info->default_txportconf.ring_size = 2048;
 494                 }
 495         } else {
 496                 info->default_rxportconf.nb_queues = 8;
 497                 info->default_txportconf.nb_queues = 8;
 498                 if (dev->data->nb_rx_queues > 2 ||
 499                     dev->data->nb_tx_queues > 2) {
 500                         /* Max Throughput. */
 501                         info->default_rxportconf.ring_size = 4096;
 502                         info->default_txportconf.ring_size = 4096;
 503                 }
 504         }
 505 }
 506
 507 /**
 508  * Sets tx mbuf limiting parameters.
 509  *
 510  * @param dev
 511  *   Pointer to Ethernet device.
 512  * @param[out] info
 513  *   Info structure output buffer.
 514  */
 515 static void
 516 mlx5_set_txlimit_params(struct rte_eth_dev *dev, struct rte_eth_dev_info *info)
 517 {
 518         struct mlx5_priv *priv = dev->data->dev_private;
 519         struct mlx5_dev_config *config = &priv->config;
 520         unsigned int inlen;
 521         uint16_t nb_max;
 522
 523         inlen = (config->txq_inline_max == MLX5_ARG_UNSET) ?
 524                 MLX5_SEND_DEF_INLINE_LEN :
 525                 (unsigned int)config->txq_inline_max;
 526         assert(config->txq_inline_min >= 0);
 527         inlen = RTE_MAX(inlen, (unsigned int)config->txq_inline_min);
 528         inlen = RTE_MIN(inlen, MLX5_WQE_SIZE_MAX +
 529                                MLX5_ESEG_MIN_INLINE_SIZE -
 530                                MLX5_WQE_CSEG_SIZE -
 531                                MLX5_WQE_ESEG_SIZE -
 532                                MLX5_WQE_DSEG_SIZE * 2);
 533         nb_max = (MLX5_WQE_SIZE_MAX +
 534                   MLX5_ESEG_MIN_INLINE_SIZE -
 535                   MLX5_WQE_CSEG_SIZE -
 536                   MLX5_WQE_ESEG_SIZE -
 537                   MLX5_WQE_DSEG_SIZE -
 538                   inlen) / MLX5_WSEG_SIZE;
 539         info->tx_desc_lim.nb_seg_max = nb_max;
 540         info->tx_desc_lim.nb_mtu_seg_max = nb_max;
 541 }
 542
 543 /**
 544  * DPDK callback to get information about the device.
 545  *
 546  * @param dev
 547  *   Pointer to Ethernet device structure.
 548  * @param[out] info
 549  *   Info structure output buffer.
 550  */
 551 void
 552 mlx5_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *info)
 553 {
 554         struct mlx5_priv *priv = dev->data->dev_private;
 555         struct mlx5_dev_config *config = &priv->config;
 556         unsigned int max;
 557
 558         /* FIXME: we should ask the device for these values. */
 559         info->min_rx_bufsize = 32;
 560         info->max_rx_pktlen = 65536;
 561         /*
 562          * Since we need one CQ per QP, the limit is the minimum number
 563          * between the two values.
 564          */
 565         max = RTE_MIN(priv->sh->device_attr.orig_attr.max_cq,
 566                       priv->sh->device_attr.orig_attr.max_qp);
 567         /* If max >= 65535 then max = 0, max_rx_queues is uint16_t. */
 568         if (max >= 65535)
 569                 max = 65535;
 570         info->max_rx_queues = max;
 571         info->max_tx_queues = max;
 572         info->max_mac_addrs = MLX5_MAX_UC_MAC_ADDRESSES;
 573         info->rx_queue_offload_capa = mlx5_get_rx_queue_offloads(dev);
 574         info->rx_offload_capa = (mlx5_get_rx_port_offloads(dev) |
 575                                  info->rx_queue_offload_capa);
 576         info->tx_offload_capa = mlx5_get_tx_port_offloads(dev);
 577         info->if_index = mlx5_ifindex(dev);
 578         info->reta_size = priv->reta_idx_n ?
 579                 priv->reta_idx_n : config->ind_table_max_size;
 580         info->hash_key_size = MLX5_RSS_HASH_KEY_LEN;
 581         info->speed_capa = priv->link_speed_capa;
 582         info->flow_type_rss_offloads = ~MLX5_RSS_HF_MASK;
 583         mlx5_set_default_params(dev, info);
 584         mlx5_set_txlimit_params(dev, info);
 585         info->switch_info.name = dev->data->name;
 586         info->switch_info.domain_id = priv->domain_id;
 587         info->switch_info.port_id = priv->representor_id;
 588         if (priv->representor) {
 589                 unsigned int i = mlx5_dev_to_port_id(dev->device, NULL, 0);
 590                 uint16_t port_id[i];
 591
 592                 i = RTE_MIN(mlx5_dev_to_port_id(dev->device, port_id, i), i);
 593                 while (i--) {
 594                         struct mlx5_priv *opriv =
 595                                 rte_eth_devices[port_id[i]].data->dev_private;
 596
 597                         if (!opriv ||
 598                             opriv->representor ||
 599                             opriv->domain_id != priv->domain_id)
 600                                 continue;
 601                         /*
 602                          * Override switch name with that of the master
 603                          * device.
 604                          */
 605                         info->switch_info.name = opriv->dev_data->name;
 606                         break;
 607                 }
 608         }
 609 }
 610
 611 /**
 612  * Get device current raw clock counter
 613  *
 614  * @param dev
 615  *   Pointer to Ethernet device structure.
 616  * @param[out] time
 617  *   Current raw clock counter of the device.
 618  *
 619  * @return
 620  *   0 if the clock has correctly been read
 621  *   The value of errno in case of error
 622  */
 623 int
 624 mlx5_read_clock(struct rte_eth_dev *dev, uint64_t *clock)
 625 {
 626         struct mlx5_priv *priv = dev->data->dev_private;
 627         struct ibv_context *ctx = priv->sh->ctx;
 628         struct ibv_values_ex values;
 629         int err = 0;
 630
 631         values.comp_mask = IBV_VALUES_MASK_RAW_CLOCK;
 632         err = mlx5_glue->query_rt_values_ex(ctx, &values);
 633         if (err != 0) {
 634                 DRV_LOG(WARNING, "Could not query the clock !");
 635                 return err;
 636         }
 637         *clock = values.raw_clock.tv_nsec;
 638         return 0;
 639 }
 640
 641 /**
 642  * Get firmware version of a device.
 643  *
 644  * @param dev
 645  *   Ethernet device port.
 646  * @param fw_ver
 647  *   String output allocated by caller.
 648  * @param fw_size
 649  *   Size of the output string, including terminating null byte.
 650  *
 651  * @return
 652  *   0 on success, or the size of the non truncated string if too big.
 653  */
 654 int mlx5_fw_version_get(struct rte_eth_dev *dev, char *fw_ver, size_t fw_size)
 655 {
 656         struct mlx5_priv *priv = dev->data->dev_private;
 657         struct ibv_device_attr *attr = &priv->sh->device_attr.orig_attr;
 658         size_t size = strnlen(attr->fw_ver, sizeof(attr->fw_ver)) + 1;
 659
 660         if (fw_size < size)
 661                 return size;
 662         if (fw_ver != NULL)
 663                 strlcpy(fw_ver, attr->fw_ver, fw_size);
 664         return 0;
 665 }
 666
 667 /**
 668  * Get supported packet types.
 669  *
 670  * @param dev
 671  *   Pointer to Ethernet device structure.
 672  *
 673  * @return
 674  *   A pointer to the supported Packet types array.
 675  */
 676 const uint32_t *
 677 mlx5_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 678 {
 679         static const uint32_t ptypes[] = {
 680                 /* refers to rxq_cq_to_pkt_type() */
 681                 RTE_PTYPE_L2_ETHER,
 682                 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
 683                 RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
 684                 RTE_PTYPE_L4_NONFRAG,
 685                 RTE_PTYPE_L4_FRAG,
 686                 RTE_PTYPE_L4_TCP,
 687                 RTE_PTYPE_L4_UDP,
 688                 RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN,
 689                 RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN,
 690                 RTE_PTYPE_INNER_L4_NONFRAG,
 691                 RTE_PTYPE_INNER_L4_FRAG,
 692                 RTE_PTYPE_INNER_L4_TCP,
 693                 RTE_PTYPE_INNER_L4_UDP,
 694                 RTE_PTYPE_UNKNOWN
 695         };
 696
 697         if (dev->rx_pkt_burst == mlx5_rx_burst ||
 698             dev->rx_pkt_burst == mlx5_rx_burst_mprq ||
 699             dev->rx_pkt_burst == mlx5_rx_burst_vec)
 700                 return ptypes;
 701         return NULL;
 702 }
 703
 704 /**
 705  * Retrieve the master device for representor in the same switch domain.
 706  *
 707  * @param dev
 708  *   Pointer to representor Ethernet device structure.
 709  *
 710  * @return
 711  *   Master device structure  on success, NULL otherwise.
 712  */
 713
 714 static struct rte_eth_dev *
 715 mlx5_find_master_dev(struct rte_eth_dev *dev)
 716 {
 717         struct mlx5_priv *priv;
 718         uint16_t port_id;
 719         uint16_t domain_id;
 720
 721         priv = dev->data->dev_private;
 722         domain_id = priv->domain_id;
 723         assert(priv->representor);
 724         RTE_ETH_FOREACH_DEV_OF(port_id, dev->device) {
 725                 priv = rte_eth_devices[port_id].data->dev_private;
 726                 if (priv &&
 727                     priv->master &&
 728                     priv->domain_id == domain_id)
 729                         return &rte_eth_devices[port_id];
 730         }
 731         return NULL;
 732 }
 733
 734 /**
 735  * DPDK callback to retrieve physical link information.
 736  *
 737  * @param dev
 738  *   Pointer to Ethernet device structure.
 739  * @param[out] link
 740  *   Storage for current link status.
 741  *
 742  * @return
 743  *   0 on success, a negative errno value otherwise and rte_errno is set.
 744  */
 745 static int
 746 mlx5_link_update_unlocked_gset(struct rte_eth_dev *dev,
 747                                struct rte_eth_link *link)
 748 {
 749         struct mlx5_priv *priv = dev->data->dev_private;
 750         struct ethtool_cmd edata = {
 751                 .cmd = ETHTOOL_GSET /* Deprecated since Linux v4.5. */
 752         };
 753         struct ifreq ifr;
 754         struct rte_eth_link dev_link;
 755         int link_speed = 0;
 756         int ret;
 757
 758         ret = mlx5_ifreq(dev, SIOCGIFFLAGS, &ifr);
 759         if (ret) {
 760                 DRV_LOG(WARNING, "port %u ioctl(SIOCGIFFLAGS) failed: %s",
 761                         dev->data->port_id, strerror(rte_errno));
 762                 return ret;
 763         }
 764         dev_link = (struct rte_eth_link) {
 765                 .link_status = ((ifr.ifr_flags & IFF_UP) &&
 766                                 (ifr.ifr_flags & IFF_RUNNING)),
 767         };
 768         ifr = (struct ifreq) {
 769                 .ifr_data = (void *)&edata,
 770         };
 771         ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
 772         if (ret) {
 773                 if (ret == -ENOTSUP && priv->representor) {
 774                         struct rte_eth_dev *master;
 775
 776                         /*
 777                          * For representors we can try to inherit link
 778                          * settings from the master device. Actually
 779                          * link settings do not make a lot of sense
 780                          * for representors due to missing physical
 781                          * link. The old kernel drivers supported
 782                          * emulated settings query for representors,
 783                          * the new ones do not, so we have to add
 784                          * this code for compatibility issues.
 785                          */
 786                         master = mlx5_find_master_dev(dev);
 787                         if (master) {
 788                                 ifr = (struct ifreq) {
 789                                         .ifr_data = (void *)&edata,
 790                                 };
 791                                 ret = mlx5_ifreq(master, SIOCETHTOOL, &ifr);
 792                         }
 793                 }
 794                 if (ret) {
 795                         DRV_LOG(WARNING,
 796                                 "port %u ioctl(SIOCETHTOOL,"
 797                                 " ETHTOOL_GSET) failed: %s",
 798                                 dev->data->port_id, strerror(rte_errno));
 799                         return ret;
 800                 }
 801         }
 802         link_speed = ethtool_cmd_speed(&edata);
 803         if (link_speed == -1)
 804                 dev_link.link_speed = ETH_SPEED_NUM_NONE;
 805         else
 806                 dev_link.link_speed = link_speed;
 807         priv->link_speed_capa = 0;
 808         if (edata.supported & SUPPORTED_Autoneg)
 809                 priv->link_speed_capa |= ETH_LINK_SPEED_AUTONEG;
 810         if (edata.supported & (SUPPORTED_1000baseT_Full |
 811                                SUPPORTED_1000baseKX_Full))
 812                 priv->link_speed_capa |= ETH_LINK_SPEED_1G;
 813         if (edata.supported & SUPPORTED_10000baseKR_Full)
 814                 priv->link_speed_capa |= ETH_LINK_SPEED_10G;
 815         if (edata.supported & (SUPPORTED_40000baseKR4_Full |
 816                                SUPPORTED_40000baseCR4_Full |
 817                                SUPPORTED_40000baseSR4_Full |
 818                                SUPPORTED_40000baseLR4_Full))
 819                 priv->link_speed_capa |= ETH_LINK_SPEED_40G;
 820         dev_link.link_duplex = ((edata.duplex == DUPLEX_HALF) ?
 821                                 ETH_LINK_HALF_DUPLEX : ETH_LINK_FULL_DUPLEX);
 822         dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds &
 823                         ETH_LINK_SPEED_FIXED);
 824         if (((dev_link.link_speed && !dev_link.link_status) ||
 825              (!dev_link.link_speed && dev_link.link_status))) {
 826                 rte_errno = EAGAIN;
 827                 return -rte_errno;
 828         }
 829         *link = dev_link;
 830         return 0;
 831 }
 832
 833 /**
 834  * Retrieve physical link information (unlocked version using new ioctl).
 835  *
 836  * @param dev
 837  *   Pointer to Ethernet device structure.
 838  * @param[out] link
 839  *   Storage for current link status.
 840  *
 841  * @return
 842  *   0 on success, a negative errno value otherwise and rte_errno is set.
 843  */
 844 static int
 845 mlx5_link_update_unlocked_gs(struct rte_eth_dev *dev,
 846                              struct rte_eth_link *link)
 847
 848 {
 849         struct mlx5_priv *priv = dev->data->dev_private;
 850         struct ethtool_link_settings gcmd = { .cmd = ETHTOOL_GLINKSETTINGS };
 851         struct ifreq ifr;
 852         struct rte_eth_link dev_link;
 853         struct rte_eth_dev *master = NULL;
 854         uint64_t sc;
 855         int ret;
 856
 857         ret = mlx5_ifreq(dev, SIOCGIFFLAGS, &ifr);
 858         if (ret) {
 859                 DRV_LOG(WARNING, "port %u ioctl(SIOCGIFFLAGS) failed: %s",
 860                         dev->data->port_id, strerror(rte_errno));
 861                 return ret;
 862         }
 863         dev_link = (struct rte_eth_link) {
 864                 .link_status = ((ifr.ifr_flags & IFF_UP) &&
 865                                 (ifr.ifr_flags & IFF_RUNNING)),
 866         };
 867         ifr = (struct ifreq) {
 868                 .ifr_data = (void *)&gcmd,
 869         };
 870         ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
 871         if (ret) {
 872                 if (ret == -ENOTSUP && priv->representor) {
 873                         /*
 874                          * For representors we can try to inherit link
 875                          * settings from the master device. Actually
 876                          * link settings do not make a lot of sense
 877                          * for representors due to missing physical
 878                          * link. The old kernel drivers supported
 879                          * emulated settings query for representors,
 880                          * the new ones do not, so we have to add
 881                          * this code for compatibility issues.
 882                          */
 883                         master = mlx5_find_master_dev(dev);
 884                         if (master) {
 885                                 ifr = (struct ifreq) {
 886                                         .ifr_data = (void *)&gcmd,
 887                                 };
 888                                 ret = mlx5_ifreq(master, SIOCETHTOOL, &ifr);
 889                         }
 890                 }
 891                 if (ret) {
 892                         DRV_LOG(DEBUG,
 893                                 "port %u ioctl(SIOCETHTOOL,"
 894                                 " ETHTOOL_GLINKSETTINGS) failed: %s",
 895                                 dev->data->port_id, strerror(rte_errno));
 896                         return ret;
 897                 }
 898
 899         }
 900         gcmd.link_mode_masks_nwords = -gcmd.link_mode_masks_nwords;
 901
 902         alignas(struct ethtool_link_settings)
 903         uint8_t data[offsetof(struct ethtool_link_settings, link_mode_masks) +
 904                      sizeof(uint32_t) * gcmd.link_mode_masks_nwords * 3];
 905         struct ethtool_link_settings *ecmd = (void *)data;
 906
 907         *ecmd = gcmd;
 908         ifr.ifr_data = (void *)ecmd;
 909         ret = mlx5_ifreq(master ? master : dev, SIOCETHTOOL, &ifr);
 910         if (ret) {
 911                 DRV_LOG(DEBUG,
 912                         "port %u ioctl(SIOCETHTOOL,"
 913                         "ETHTOOL_GLINKSETTINGS) failed: %s",
 914                         dev->data->port_id, strerror(rte_errno));
 915                 return ret;
 916         }
 917         dev_link.link_speed = ecmd->speed;
 918         sc = ecmd->link_mode_masks[0] |
 919                 ((uint64_t)ecmd->link_mode_masks[1] << 32);
 920         priv->link_speed_capa = 0;
 921         if (sc & MLX5_BITSHIFT(ETHTOOL_LINK_MODE_Autoneg_BIT))
 922                 priv->link_speed_capa |= ETH_LINK_SPEED_AUTONEG;
 923         if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_1000baseT_Full_BIT) |
 924                   MLX5_BITSHIFT(ETHTOOL_LINK_MODE_1000baseKX_Full_BIT)))
 925                 priv->link_speed_capa |= ETH_LINK_SPEED_1G;
 926         if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_10000baseKX4_Full_BIT) |
 927                   MLX5_BITSHIFT(ETHTOOL_LINK_MODE_10000baseKR_Full_BIT) |
 928                   MLX5_BITSHIFT(ETHTOOL_LINK_MODE_10000baseR_FEC_BIT)))
 929                 priv->link_speed_capa |= ETH_LINK_SPEED_10G;
 930         if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_20000baseMLD2_Full_BIT) |
 931                   MLX5_BITSHIFT(ETHTOOL_LINK_MODE_20000baseKR2_Full_BIT)))
 932                 priv->link_speed_capa |= ETH_LINK_SPEED_20G;
 933         if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_40000baseKR4_Full_BIT) |
 934                   MLX5_BITSHIFT(ETHTOOL_LINK_MODE_40000baseCR4_Full_BIT) |
 935                   MLX5_BITSHIFT(ETHTOOL_LINK_MODE_40000baseSR4_Full_BIT) |
 936                   MLX5_BITSHIFT(ETHTOOL_LINK_MODE_40000baseLR4_Full_BIT)))
 937                 priv->link_speed_capa |= ETH_LINK_SPEED_40G;
 938         if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_56000baseKR4_Full_BIT) |
 939                   MLX5_BITSHIFT(ETHTOOL_LINK_MODE_56000baseCR4_Full_BIT) |
 940                   MLX5_BITSHIFT(ETHTOOL_LINK_MODE_56000baseSR4_Full_BIT) |
 941                   MLX5_BITSHIFT(ETHTOOL_LINK_MODE_56000baseLR4_Full_BIT)))
 942                 priv->link_speed_capa |= ETH_LINK_SPEED_56G;
 943         if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_25000baseCR_Full_BIT) |
 944                   MLX5_BITSHIFT(ETHTOOL_LINK_MODE_25000baseKR_Full_BIT) |
 945                   MLX5_BITSHIFT(ETHTOOL_LINK_MODE_25000baseSR_Full_BIT)))
 946                 priv->link_speed_capa |= ETH_LINK_SPEED_25G;
 947         if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_50000baseCR2_Full_BIT) |
 948                   MLX5_BITSHIFT(ETHTOOL_LINK_MODE_50000baseKR2_Full_BIT)))
 949                 priv->link_speed_capa |= ETH_LINK_SPEED_50G;
 950         if (sc & (MLX5_BITSHIFT(ETHTOOL_LINK_MODE_100000baseKR4_Full_BIT) |
 951                   MLX5_BITSHIFT(ETHTOOL_LINK_MODE_100000baseSR4_Full_BIT) |
 952                   MLX5_BITSHIFT(ETHTOOL_LINK_MODE_100000baseCR4_Full_BIT) |
 953                   MLX5_BITSHIFT(ETHTOOL_LINK_MODE_100000baseLR4_ER4_Full_BIT)))
 954                 priv->link_speed_capa |= ETH_LINK_SPEED_100G;
 955         dev_link.link_duplex = ((ecmd->duplex == DUPLEX_HALF) ?
 956                                 ETH_LINK_HALF_DUPLEX : ETH_LINK_FULL_DUPLEX);
 957         dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds &
 958                                   ETH_LINK_SPEED_FIXED);
 959         if (((dev_link.link_speed && !dev_link.link_status) ||
 960              (!dev_link.link_speed && dev_link.link_status))) {
 961                 rte_errno = EAGAIN;
 962                 return -rte_errno;
 963         }
 964         *link = dev_link;
 965         return 0;
 966 }
 967
 968 /**
 969  * DPDK callback to retrieve physical link information.
 970  *
 971  * @param dev
 972  *   Pointer to Ethernet device structure.
 973  * @param wait_to_complete
 974  *   Wait for request completion.
 975  *
 976  * @return
 977  *   0 if link status was not updated, positive if it was, a negative errno
 978  *   value otherwise and rte_errno is set.
 979  */
 980 int
 981 mlx5_link_update(struct rte_eth_dev *dev, int wait_to_complete)
 982 {
 983         int ret;
 984         struct rte_eth_link dev_link;
 985         time_t start_time = time(NULL);
 986
 987         do {
 988                 ret = mlx5_link_update_unlocked_gs(dev, &dev_link);
 989                 if (ret == -ENOTSUP)
 990                         ret = mlx5_link_update_unlocked_gset(dev, &dev_link);
 991                 if (ret == 0)
 992                         break;
 993                 /* Handle wait to complete situation. */
 994                 if (wait_to_complete && ret == -EAGAIN) {
 995                         if (abs((int)difftime(time(NULL), start_time)) <
 996                             MLX5_LINK_STATUS_TIMEOUT) {
 997                                 usleep(0);
 998                                 continue;
 999                         } else {
1000                                 rte_errno = EBUSY;
1001                                 return -rte_errno;
1002                         }
1003                 } else if (ret < 0) {
1004                         return ret;
1005                 }
1006         } while (wait_to_complete);
1007         ret = !!memcmp(&dev->data->dev_link, &dev_link,
1008                        sizeof(struct rte_eth_link));
1009         dev->data->dev_link = dev_link;
1010         return ret;
1011 }
1012
1013 /**
1014  * DPDK callback to change the MTU.
1015  *
1016  * @param dev
1017  *   Pointer to Ethernet device structure.
1018  * @param in_mtu
1019  *   New MTU.
1020  *
1021  * @return
1022  *   0 on success, a negative errno value otherwise and rte_errno is set.
1023  */
1024 int
1025 mlx5_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
1026 {
1027         struct mlx5_priv *priv = dev->data->dev_private;
1028         uint16_t kern_mtu = 0;
1029         int ret;
1030
1031         ret = mlx5_get_mtu(dev, &kern_mtu);
1032         if (ret)
1033                 return ret;
1034         /* Set kernel interface MTU first. */
1035         ret = mlx5_set_mtu(dev, mtu);
1036         if (ret)
1037                 return ret;
1038         ret = mlx5_get_mtu(dev, &kern_mtu);
1039         if (ret)
1040                 return ret;
1041         if (kern_mtu == mtu) {
1042                 priv->mtu = mtu;
1043                 DRV_LOG(DEBUG, "port %u adapter MTU set to %u",
1044                         dev->data->port_id, mtu);
1045                 return 0;
1046         }
1047         rte_errno = EAGAIN;
1048         return -rte_errno;
1049 }
1050
1051 /**
1052  * DPDK callback to get flow control status.
1053  *
1054  * @param dev
1055  *   Pointer to Ethernet device structure.
1056  * @param[out] fc_conf
1057  *   Flow control output buffer.
1058  *
1059  * @return
1060  *   0 on success, a negative errno value otherwise and rte_errno is set.
1061  */
1062 int
1063 mlx5_dev_get_flow_ctrl(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1064 {
1065         struct ifreq ifr;
1066         struct ethtool_pauseparam ethpause = {
1067                 .cmd = ETHTOOL_GPAUSEPARAM
1068         };
1069         int ret;
1070
1071         ifr.ifr_data = (void *)&ethpause;
1072         ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
1073         if (ret) {
1074                 DRV_LOG(WARNING,
1075                         "port %u ioctl(SIOCETHTOOL, ETHTOOL_GPAUSEPARAM) failed:"
1076                         " %s",
1077                         dev->data->port_id, strerror(rte_errno));
1078                 return ret;
1079         }
1080         fc_conf->autoneg = ethpause.autoneg;
1081         if (ethpause.rx_pause && ethpause.tx_pause)
1082                 fc_conf->mode = RTE_FC_FULL;
1083         else if (ethpause.rx_pause)
1084                 fc_conf->mode = RTE_FC_RX_PAUSE;
1085         else if (ethpause.tx_pause)
1086                 fc_conf->mode = RTE_FC_TX_PAUSE;
1087         else
1088                 fc_conf->mode = RTE_FC_NONE;
1089         return 0;
1090 }
1091
1092 /**
1093  * DPDK callback to modify flow control parameters.
1094  *
1095  * @param dev
1096  *   Pointer to Ethernet device structure.
1097  * @param[in] fc_conf
1098  *   Flow control parameters.
1099  *
1100  * @return
1101  *   0 on success, a negative errno value otherwise and rte_errno is set.
1102  */
1103 int
1104 mlx5_dev_set_flow_ctrl(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1105 {
1106         struct ifreq ifr;
1107         struct ethtool_pauseparam ethpause = {
1108                 .cmd = ETHTOOL_SPAUSEPARAM
1109         };
1110         int ret;
1111
1112         ifr.ifr_data = (void *)&ethpause;
1113         ethpause.autoneg = fc_conf->autoneg;
1114         if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) ||
1115             (fc_conf->mode & RTE_FC_RX_PAUSE))
1116                 ethpause.rx_pause = 1;
1117         else
1118                 ethpause.rx_pause = 0;
1119
1120         if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) ||
1121             (fc_conf->mode & RTE_FC_TX_PAUSE))
1122                 ethpause.tx_pause = 1;
1123         else
1124                 ethpause.tx_pause = 0;
1125         ret = mlx5_ifreq(dev, SIOCETHTOOL, &ifr);
1126         if (ret) {
1127                 DRV_LOG(WARNING,
1128                         "port %u ioctl(SIOCETHTOOL, ETHTOOL_SPAUSEPARAM)"
1129                         " failed: %s",
1130                         dev->data->port_id, strerror(rte_errno));
1131                 return ret;
1132         }
1133         return 0;
1134 }
1135
1136 /**
1137  * Get PCI information from struct ibv_device.
1138  *
1139  * @param device
1140  *   Pointer to Ethernet device structure.
1141  * @param[out] pci_addr
1142  *   PCI bus address output buffer.
1143  *
1144  * @return
1145  *   0 on success, a negative errno value otherwise and rte_errno is set.
1146  */
1147 int
1148 mlx5_ibv_device_to_pci_addr(const struct ibv_device *device,
1149                             struct rte_pci_addr *pci_addr)
1150 {
1151         FILE *file;
1152         char line[32];
1153         MKSTR(path, "%s/device/uevent", device->ibdev_path);
1154
1155         file = fopen(path, "rb");
1156         if (file == NULL) {
1157                 rte_errno = errno;
1158                 return -rte_errno;
1159         }
1160         while (fgets(line, sizeof(line), file) == line) {
1161                 size_t len = strlen(line);
1162                 int ret;
1163
1164                 /* Truncate long lines. */
1165                 if (len == (sizeof(line) - 1))
1166                         while (line[(len - 1)] != '\n') {
1167                                 ret = fgetc(file);
1168                                 if (ret == EOF)
1169                                         break;
1170                                 line[(len - 1)] = ret;
1171                         }
1172                 /* Extract information. */
1173                 if (sscanf(line,
1174                            "PCI_SLOT_NAME="
1175                            "%" SCNx32 ":%" SCNx8 ":%" SCNx8 ".%" SCNx8 "\n",
1176                            &pci_addr->domain,
1177                            &pci_addr->bus,
1178                            &pci_addr->devid,
1179                            &pci_addr->function) == 4) {
1180                         ret = 0;
1181                         break;
1182                 }
1183         }
1184         fclose(file);
1185         return 0;
1186 }
1187
1188 /**
1189  * Handle asynchronous removal event for entire multiport device.
1190  *
1191  * @param sh
1192  *   Infiniband device shared context.
1193  */
1194 static void
1195 mlx5_dev_interrupt_device_fatal(struct mlx5_ibv_shared *sh)
1196 {
1197         uint32_t i;
1198
1199         for (i = 0; i < sh->max_port; ++i) {
1200                 struct rte_eth_dev *dev;
1201
1202                 if (sh->port[i].ih_port_id >= RTE_MAX_ETHPORTS) {
1203                         /*
1204                          * Or not existing port either no
1205                          * handler installed for this port.
1206                          */
1207                         continue;
1208                 }
1209                 dev = &rte_eth_devices[sh->port[i].ih_port_id];
1210                 assert(dev);
1211                 if (dev->data->dev_conf.intr_conf.rmv)
1212                         _rte_eth_dev_callback_process
1213                                 (dev, RTE_ETH_EVENT_INTR_RMV, NULL);
1214         }
1215 }
1216
1217 /**
1218  * Handle shared asynchronous events the NIC (removal event
1219  * and link status change). Supports multiport IB device.
1220  *
1221  * @param cb_arg
1222  *   Callback argument.
1223  */
1224 void
1225 mlx5_dev_interrupt_handler(void *cb_arg)
1226 {
1227         struct mlx5_ibv_shared *sh = cb_arg;
1228         struct ibv_async_event event;
1229
1230         /* Read all message from the IB device and acknowledge them. */
1231         for (;;) {
1232                 struct rte_eth_dev *dev;
1233                 uint32_t tmp;
1234
1235                 if (mlx5_glue->get_async_event(sh->ctx, &event))
1236                         break;
1237                 /* Retrieve and check IB port index. */
1238                 tmp = (uint32_t)event.element.port_num;
1239                 if (!tmp && event.event_type == IBV_EVENT_DEVICE_FATAL) {
1240                         /*
1241                          * The DEVICE_FATAL event is called once for
1242                          * entire device without port specifying.
1243                          * We should notify all existing ports.
1244                          */
1245                         mlx5_glue->ack_async_event(&event);
1246                         mlx5_dev_interrupt_device_fatal(sh);
1247                         continue;
1248                 }
1249                 assert(tmp && (tmp <= sh->max_port));
1250                 if (!tmp) {
1251                         /* Unsupported devive level event. */
1252                         mlx5_glue->ack_async_event(&event);
1253                         DRV_LOG(DEBUG,
1254                                 "unsupported common event (type %d)",
1255                                 event.event_type);
1256                         continue;
1257                 }
1258                 if (tmp > sh->max_port) {
1259                         /* Invalid IB port index. */
1260                         mlx5_glue->ack_async_event(&event);
1261                         DRV_LOG(DEBUG,
1262                                 "cannot handle an event (type %d)"
1263                                 "due to invalid IB port index (%u)",
1264                                 event.event_type, tmp);
1265                         continue;
1266                 }
1267                 if (sh->port[tmp - 1].ih_port_id >= RTE_MAX_ETHPORTS) {
1268                         /* No handler installed. */
1269                         mlx5_glue->ack_async_event(&event);
1270                         DRV_LOG(DEBUG,
1271                                 "cannot handle an event (type %d)"
1272                                 "due to no handler installed for port %u",
1273                                 event.event_type, tmp);
1274                         continue;
1275                 }
1276                 /* Retrieve ethernet device descriptor. */
1277                 tmp = sh->port[tmp - 1].ih_port_id;
1278                 dev = &rte_eth_devices[tmp];
1279                 assert(dev);
1280                 if ((event.event_type == IBV_EVENT_PORT_ACTIVE ||
1281                      event.event_type == IBV_EVENT_PORT_ERR) &&
1282                         dev->data->dev_conf.intr_conf.lsc) {
1283                         mlx5_glue->ack_async_event(&event);
1284                         if (mlx5_link_update(dev, 0) == -EAGAIN) {
1285                                 usleep(0);
1286                                 continue;
1287                         }
1288                         _rte_eth_dev_callback_process
1289                                 (dev, RTE_ETH_EVENT_INTR_LSC, NULL);
1290                         continue;
1291                 }
1292                 DRV_LOG(DEBUG,
1293                         "port %u cannot handle an unknown event (type %d)",
1294                         dev->data->port_id, event.event_type);
1295                 mlx5_glue->ack_async_event(&event);
1296         }
1297 }
1298
1299 /*
1300  * Unregister callback handler safely. The handler may be active
1301  * while we are trying to unregister it, in this case code -EAGAIN
1302  * is returned by rte_intr_callback_unregister(). This routine checks
1303  * the return code and tries to unregister handler again.
1304  *
1305  * @param handle
1306  *   interrupt handle
1307  * @param cb_fn
1308  *   pointer to callback routine
1309  * @cb_arg
1310  *   opaque callback parameter
1311  */
1312 void
1313 mlx5_intr_callback_unregister(const struct rte_intr_handle *handle,
1314                               rte_intr_callback_fn cb_fn, void *cb_arg)
1315 {
1316         /*
1317          * Try to reduce timeout management overhead by not calling
1318          * the timer related routines on the first iteration. If the
1319          * unregistering succeeds on first call there will be no
1320          * timer calls at all.
1321          */
1322         uint64_t twait = 0;
1323         uint64_t start = 0;
1324
1325         do {
1326                 int ret;
1327
1328                 ret = rte_intr_callback_unregister(handle, cb_fn, cb_arg);
1329                 if (ret >= 0)
1330                         return;
1331                 if (ret != -EAGAIN) {
1332                         DRV_LOG(INFO, "failed to unregister interrupt"
1333                                       " handler (error: %d)", ret);
1334                         assert(false);
1335                         return;
1336                 }
1337                 if (twait) {
1338                         struct timespec onems;
1339
1340                         /* Wait one millisecond and try again. */
1341                         onems.tv_sec = 0;
1342                         onems.tv_nsec = NS_PER_S / MS_PER_S;
1343                         nanosleep(&onems, 0);
1344                         /* Check whether one second elapsed. */
1345                         if ((rte_get_timer_cycles() - start) <= twait)
1346                                 continue;
1347                 } else {
1348                         /*
1349                          * We get the amount of timer ticks for one second.
1350                          * If this amount elapsed it means we spent one
1351                          * second in waiting. This branch is executed once
1352                          * on first iteration.
1353                          */
1354                         twait = rte_get_timer_hz();
1355                         assert(twait);
1356                 }
1357                 /*
1358                  * Timeout elapsed, show message (once a second) and retry.
1359                  * We have no other acceptable option here, if we ignore
1360                  * the unregistering return code the handler will not
1361                  * be unregistered, fd will be closed and we may get the
1362                  * crush. Hanging and messaging in the loop seems not to be
1363                  * the worst choice.
1364                  */
1365                 DRV_LOG(INFO, "Retrying to unregister interrupt handler");
1366                 start = rte_get_timer_cycles();
1367         } while (true);
1368 }
1369
1370 /**
1371  * Handle DEVX interrupts from the NIC.
1372  * This function is probably called from the DPDK host thread.
1373  *
1374  * @param cb_arg
1375  *   Callback argument.
1376  */
1377 void
1378 mlx5_dev_interrupt_handler_devx(void *cb_arg)
1379 {
1380 #ifndef HAVE_IBV_DEVX_ASYNC
1381         (void)cb_arg;
1382         return;
1383 #else
1384         struct mlx5_ibv_shared *sh = cb_arg;
1385         union {
1386                 struct mlx5dv_devx_async_cmd_hdr cmd_resp;
1387                 uint8_t buf[MLX5_ST_SZ_BYTES(query_flow_counter_out) +
1388                             MLX5_ST_SZ_BYTES(traffic_counter) +
1389                             sizeof(struct mlx5dv_devx_async_cmd_hdr)];
1390         } out;
1391         uint8_t *buf = out.buf + sizeof(out.cmd_resp);
1392
1393         while (!mlx5_glue->devx_get_async_cmd_comp(sh->devx_comp,
1394                                                    &out.cmd_resp,
1395                                                    sizeof(out.buf)))
1396                 mlx5_flow_async_pool_query_handle
1397                         (sh, (uint64_t)out.cmd_resp.wr_id,
1398                          mlx5_devx_get_out_command_status(buf));
1399 #endif /* HAVE_IBV_DEVX_ASYNC */
1400 }
1401
1402 /**
1403  * Uninstall shared asynchronous device events handler.
1404  * This function is implemented to support event sharing
1405  * between multiple ports of single IB device.
1406  *
1407  * @param dev
1408  *   Pointer to Ethernet device.
1409  */
1410 static void
1411 mlx5_dev_shared_handler_uninstall(struct rte_eth_dev *dev)
1412 {
1413         struct mlx5_priv *priv = dev->data->dev_private;
1414         struct mlx5_ibv_shared *sh = priv->sh;
1415
1416         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1417                 return;
1418         pthread_mutex_lock(&sh->intr_mutex);
1419         assert(priv->ibv_port);
1420         assert(priv->ibv_port <= sh->max_port);
1421         assert(dev->data->port_id < RTE_MAX_ETHPORTS);
1422         if (sh->port[priv->ibv_port - 1].ih_port_id >= RTE_MAX_ETHPORTS)
1423                 goto exit;
1424         assert(sh->port[priv->ibv_port - 1].ih_port_id ==
1425                                         (uint32_t)dev->data->port_id);
1426         assert(sh->intr_cnt);
1427         sh->port[priv->ibv_port - 1].ih_port_id = RTE_MAX_ETHPORTS;
1428         if (!sh->intr_cnt || --sh->intr_cnt)
1429                 goto exit;
1430         mlx5_intr_callback_unregister(&sh->intr_handle,
1431                                      mlx5_dev_interrupt_handler, sh);
1432         sh->intr_handle.fd = 0;
1433         sh->intr_handle.type = RTE_INTR_HANDLE_UNKNOWN;
1434         if (sh->intr_handle_devx.fd) {
1435                 rte_intr_callback_unregister(&sh->intr_handle_devx,
1436                                              mlx5_dev_interrupt_handler_devx,
1437                                              sh);
1438                 sh->intr_handle_devx.fd = 0;
1439                 sh->intr_handle_devx.type = RTE_INTR_HANDLE_UNKNOWN;
1440         }
1441         if (sh->devx_comp) {
1442                 mlx5_glue->devx_destroy_cmd_comp(sh->devx_comp);
1443                 sh->devx_comp = NULL;
1444         }
1445 exit:
1446         pthread_mutex_unlock(&sh->intr_mutex);
1447 }
1448
1449 /**
1450  * Install shared asynchronous device events handler.
1451  * This function is implemented to support event sharing
1452  * between multiple ports of single IB device.
1453  *
1454  * @param dev
1455  *   Pointer to Ethernet device.
1456  */
1457 static void
1458 mlx5_dev_shared_handler_install(struct rte_eth_dev *dev)
1459 {
1460         struct mlx5_priv *priv = dev->data->dev_private;
1461         struct mlx5_ibv_shared *sh = priv->sh;
1462         int ret;
1463         int flags;
1464
1465         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1466                 return;
1467         pthread_mutex_lock(&sh->intr_mutex);
1468         assert(priv->ibv_port);
1469         assert(priv->ibv_port <= sh->max_port);
1470         assert(dev->data->port_id < RTE_MAX_ETHPORTS);
1471         if (sh->port[priv->ibv_port - 1].ih_port_id < RTE_MAX_ETHPORTS) {
1472                 /* The handler is already installed for this port. */
1473                 assert(sh->intr_cnt);
1474                 goto exit;
1475         }
1476         sh->port[priv->ibv_port - 1].ih_port_id = (uint32_t)dev->data->port_id;
1477         if (sh->intr_cnt) {
1478                 sh->intr_cnt++;
1479                 goto exit;
1480         }
1481         /* No shared handler installed. */
1482         assert(sh->ctx->async_fd > 0);
1483         flags = fcntl(sh->ctx->async_fd, F_GETFL);
1484         ret = fcntl(sh->ctx->async_fd, F_SETFL, flags | O_NONBLOCK);
1485         if (ret) {
1486                 DRV_LOG(INFO, "failed to change file descriptor"
1487                               " async event queue");
1488                 goto error;
1489         }
1490         sh->intr_handle.fd = sh->ctx->async_fd;
1491         sh->intr_handle.type = RTE_INTR_HANDLE_EXT;
1492         rte_intr_callback_register(&sh->intr_handle,
1493                                    mlx5_dev_interrupt_handler, sh);
1494         if (priv->config.devx) {
1495 #ifndef HAVE_IBV_DEVX_ASYNC
1496                 goto error_unregister;
1497 #else
1498                 sh->devx_comp = mlx5_glue->devx_create_cmd_comp(sh->ctx);
1499                 if (sh->devx_comp) {
1500                         flags = fcntl(sh->devx_comp->fd, F_GETFL);
1501                         ret = fcntl(sh->devx_comp->fd, F_SETFL,
1502                                     flags | O_NONBLOCK);
1503                         if (ret) {
1504                                 DRV_LOG(INFO, "failed to change file descriptor"
1505                                               " devx async event queue");
1506                                 goto error_unregister;
1507                         }
1508                         sh->intr_handle_devx.fd = sh->devx_comp->fd;
1509                         sh->intr_handle_devx.type = RTE_INTR_HANDLE_EXT;
1510                         rte_intr_callback_register
1511                                 (&sh->intr_handle_devx,
1512                                  mlx5_dev_interrupt_handler_devx, sh);
1513                 } else {
1514                         DRV_LOG(INFO, "failed to create devx async command "
1515                                 "completion");
1516                         goto error_unregister;
1517                 }
1518 #endif /* HAVE_IBV_DEVX_ASYNC */
1519         }
1520         sh->intr_cnt++;
1521         goto exit;
1522 error_unregister:
1523         rte_intr_callback_unregister(&sh->intr_handle,
1524                                      mlx5_dev_interrupt_handler, sh);
1525 error:
1526         /* Indicate there will be no interrupts. */
1527         dev->data->dev_conf.intr_conf.lsc = 0;
1528         dev->data->dev_conf.intr_conf.rmv = 0;
1529         sh->intr_handle.fd = 0;
1530         sh->intr_handle.type = RTE_INTR_HANDLE_UNKNOWN;
1531         sh->port[priv->ibv_port - 1].ih_port_id = RTE_MAX_ETHPORTS;
1532 exit:
1533         pthread_mutex_unlock(&sh->intr_mutex);
1534 }
1535
1536 /**
1537  * Uninstall interrupt handler.
1538  *
1539  * @param dev
1540  *   Pointer to Ethernet device.
1541  */
1542 void
1543 mlx5_dev_interrupt_handler_uninstall(struct rte_eth_dev *dev)
1544 {
1545         mlx5_dev_shared_handler_uninstall(dev);
1546 }
1547
1548 /**
1549  * Install interrupt handler.
1550  *
1551  * @param dev
1552  *   Pointer to Ethernet device.
1553  */
1554 void
1555 mlx5_dev_interrupt_handler_install(struct rte_eth_dev *dev)
1556 {
1557         mlx5_dev_shared_handler_install(dev);
1558 }
1559
1560 /**
1561  * DPDK callback to bring the link DOWN.
1562  *
1563  * @param dev
1564  *   Pointer to Ethernet device structure.
1565  *
1566  * @return
1567  *   0 on success, a negative errno value otherwise and rte_errno is set.
1568  */
1569 int
1570 mlx5_set_link_down(struct rte_eth_dev *dev)
1571 {
1572         return mlx5_set_flags(dev, ~IFF_UP, ~IFF_UP);
1573 }
1574
1575 /**
1576  * DPDK callback to bring the link UP.
1577  *
1578  * @param dev
1579  *   Pointer to Ethernet device structure.
1580  *
1581  * @return
1582  *   0 on success, a negative errno value otherwise and rte_errno is set.
1583  */
1584 int
1585 mlx5_set_link_up(struct rte_eth_dev *dev)
1586 {
1587         return mlx5_set_flags(dev, ~IFF_UP, IFF_UP);
1588 }
1589
1590 /**
1591  * Configure the RX function to use.
1592  *
1593  * @param dev
1594  *   Pointer to private data structure.
1595  *
1596  * @return
1597  *   Pointer to selected Rx burst function.
1598  */
1599 eth_rx_burst_t
1600 mlx5_select_rx_function(struct rte_eth_dev *dev)
1601 {
1602         eth_rx_burst_t rx_pkt_burst = mlx5_rx_burst;
1603
1604         assert(dev != NULL);
1605         if (mlx5_check_vec_rx_support(dev) > 0) {
1606                 rx_pkt_burst = mlx5_rx_burst_vec;
1607                 DRV_LOG(DEBUG, "port %u selected Rx vectorized function",
1608                         dev->data->port_id);
1609         } else if (mlx5_mprq_enabled(dev)) {
1610                 rx_pkt_burst = mlx5_rx_burst_mprq;
1611         }
1612         return rx_pkt_burst;
1613 }
1614
1615 /**
1616  * Check if mlx5 device was removed.
1617  *
1618  * @param dev
1619  *   Pointer to Ethernet device structure.
1620  *
1621  * @return
1622  *   1 when device is removed, otherwise 0.
1623  */
1624 int
1625 mlx5_is_removed(struct rte_eth_dev *dev)
1626 {
1627         struct ibv_device_attr device_attr;
1628         struct mlx5_priv *priv = dev->data->dev_private;
1629
1630         if (mlx5_glue->query_device(priv->sh->ctx, &device_attr) == EIO)
1631                 return 1;
1632         return 0;
1633 }
1634
1635 /**
1636  * Get port ID list of mlx5 instances sharing a common device.
1637  *
1638  * @param[in] dev
1639  *   Device to look for.
1640  * @param[out] port_list
1641  *   Result buffer for collected port IDs.
1642  * @param port_list_n
1643  *   Maximum number of entries in result buffer. If 0, @p port_list can be
1644  *   NULL.
1645  *
1646  * @return
1647  *   Number of matching instances regardless of the @p port_list_n
1648  *   parameter, 0 if none were found.
1649  */
1650 unsigned int
1651 mlx5_dev_to_port_id(const struct rte_device *dev, uint16_t *port_list,
1652                     unsigned int port_list_n)
1653 {
1654         uint16_t id;
1655         unsigned int n = 0;
1656
1657         RTE_ETH_FOREACH_DEV_OF(id, dev) {
1658                 if (n < port_list_n)
1659                         port_list[n] = id;
1660                 n++;
1661         }
1662         return n;
1663 }
1664
1665 /**
1666  * Get the E-Switch domain id this port belongs to.
1667  *
1668  * @param[in] port
1669  *   Device port id.
1670  * @param[out] es_domain_id
1671  *   E-Switch domain id.
1672  * @param[out] es_port_id
1673  *   The port id of the port in the E-Switch.
1674  *
1675  * @return
1676  *   0 on success, a negative errno value otherwise and rte_errno is set.
1677  */
1678 int
1679 mlx5_port_to_eswitch_info(uint16_t port,
1680                           uint16_t *es_domain_id, uint16_t *es_port_id)
1681 {
1682         struct rte_eth_dev *dev;
1683         struct mlx5_priv *priv;
1684
1685         if (port >= RTE_MAX_ETHPORTS) {
1686                 rte_errno = EINVAL;
1687                 return -rte_errno;
1688         }
1689         if (!rte_eth_dev_is_valid_port(port)) {
1690                 rte_errno = ENODEV;
1691                 return -rte_errno;
1692         }
1693         dev = &rte_eth_devices[port];
1694         priv = dev->data->dev_private;
1695         if (!(priv->representor || priv->master)) {
1696                 rte_errno = EINVAL;
1697                 return -rte_errno;
1698         }
1699         if (es_domain_id)
1700                 *es_domain_id = priv->domain_id;
1701         if (es_port_id)
1702                 *es_port_id = priv->vport_id;
1703         return 0;
1704 }
1705
1706 /**
1707  * Get switch information associated with network interface.
1708  *
1709  * @param ifindex
1710  *   Network interface index.
1711  * @param[out] info
1712  *   Switch information object, populated in case of success.
1713  *
1714  * @return
1715  *   0 on success, a negative errno value otherwise and rte_errno is set.
1716  */
1717 int
1718 mlx5_sysfs_switch_info(unsigned int ifindex, struct mlx5_switch_info *info)
1719 {
1720         char ifname[IF_NAMESIZE];
1721         char port_name[IF_NAMESIZE];
1722         FILE *file;
1723         struct mlx5_switch_info data = {
1724                 .master = 0,
1725                 .representor = 0,
1726                 .name_type = MLX5_PHYS_PORT_NAME_TYPE_NOTSET,
1727                 .port_name = 0,
1728                 .switch_id = 0,
1729         };
1730         DIR *dir;
1731         bool port_switch_id_set = false;
1732         bool device_dir = false;
1733         char c;
1734         int ret;
1735
1736         if (!if_indextoname(ifindex, ifname)) {
1737                 rte_errno = errno;
1738                 return -rte_errno;
1739         }
1740
1741         MKSTR(phys_port_name, "/sys/class/net/%s/phys_port_name",
1742               ifname);
1743         MKSTR(phys_switch_id, "/sys/class/net/%s/phys_switch_id",
1744               ifname);
1745         MKSTR(pci_device, "/sys/class/net/%s/device",
1746               ifname);
1747
1748         file = fopen(phys_port_name, "rb");
1749         if (file != NULL) {
1750                 ret = fscanf(file, "%s", port_name);
1751                 fclose(file);
1752                 if (ret == 1)
1753                         mlx5_translate_port_name(port_name, &data);
1754         }
1755         file = fopen(phys_switch_id, "rb");
1756         if (file == NULL) {
1757                 rte_errno = errno;
1758                 return -rte_errno;
1759         }
1760         port_switch_id_set =
1761                 fscanf(file, "%" SCNx64 "%c", &data.switch_id, &c) == 2 &&
1762                 c == '\n';
1763         fclose(file);
1764         dir = opendir(pci_device);
1765         if (dir != NULL) {
1766                 closedir(dir);
1767                 device_dir = true;
1768         }
1769         if (port_switch_id_set) {
1770                 /* We have some E-Switch configuration. */
1771                 mlx5_sysfs_check_switch_info(device_dir, &data);
1772         }
1773         *info = data;
1774         assert(!(data.master && data.representor));
1775         if (data.master && data.representor) {
1776                 DRV_LOG(ERR, "ifindex %u device is recognized as master"
1777                              " and as representor", ifindex);
1778                 rte_errno = ENODEV;
1779                 return -rte_errno;
1780         }
1781         return 0;
1782 }
1783
1784 /**
1785  * Analyze gathered port parameters via Netlink to recognize master
1786  * and representor devices for E-Switch configuration.
1787  *
1788  * @param[in] num_vf_set
1789  *   flag of presence of number of VFs port attribute.
1790  * @param[inout] switch_info
1791  *   Port information, including port name as a number and port name
1792  *   type if recognized
1793  *
1794  * @return
1795  *   master and representor flags are set in switch_info according to
1796  *   recognized parameters (if any).
1797  */
1798 void
1799 mlx5_nl_check_switch_info(bool num_vf_set,
1800                           struct mlx5_switch_info *switch_info)
1801 {
1802         switch (switch_info->name_type) {
1803         case MLX5_PHYS_PORT_NAME_TYPE_UNKNOWN:
1804                 /*
1805                  * Name is not recognized, assume the master,
1806                  * check the number of VFs key presence.
1807                  */
1808                 switch_info->master = num_vf_set;
1809                 break;
1810         case MLX5_PHYS_PORT_NAME_TYPE_NOTSET:
1811                 /*
1812                  * Name is not set, this assumes the legacy naming
1813                  * schema for master, just check if there is a
1814                  * number of VFs key.
1815                  */
1816                 switch_info->master = num_vf_set;
1817                 break;
1818         case MLX5_PHYS_PORT_NAME_TYPE_UPLINK:
1819                 /* New uplink naming schema recognized. */
1820                 switch_info->master = 1;
1821                 break;
1822         case MLX5_PHYS_PORT_NAME_TYPE_LEGACY:
1823                 /* Legacy representors naming schema. */
1824                 switch_info->representor = !num_vf_set;
1825                 break;
1826         case MLX5_PHYS_PORT_NAME_TYPE_PFVF:
1827                 /* New representors naming schema. */
1828                 switch_info->representor = 1;
1829                 break;
1830         }
1831 }
1832
1833 /**
1834  * Analyze gathered port parameters via sysfs to recognize master
1835  * and representor devices for E-Switch configuration.
1836  *
1837  * @param[in] device_dir
1838  *   flag of presence of "device" directory under port device key.
1839  * @param[inout] switch_info
1840  *   Port information, including port name as a number and port name
1841  *   type if recognized
1842  *
1843  * @return
1844  *   master and representor flags are set in switch_info according to
1845  *   recognized parameters (if any).
1846  */
1847 void
1848 mlx5_sysfs_check_switch_info(bool device_dir,
1849                              struct mlx5_switch_info *switch_info)
1850 {
1851         switch (switch_info->name_type) {
1852         case MLX5_PHYS_PORT_NAME_TYPE_UNKNOWN:
1853                 /*
1854                  * Name is not recognized, assume the master,
1855                  * check the device directory presence.
1856                  */
1857                 switch_info->master = device_dir;
1858                 break;
1859         case MLX5_PHYS_PORT_NAME_TYPE_NOTSET:
1860                 /*
1861                  * Name is not set, this assumes the legacy naming
1862                  * schema for master, just check if there is
1863                  * a device directory.
1864                  */
1865                 switch_info->master = device_dir;
1866                 break;
1867         case MLX5_PHYS_PORT_NAME_TYPE_UPLINK:
1868                 /* New uplink naming schema recognized. */
1869                 switch_info->master = 1;
1870                 break;
1871         case MLX5_PHYS_PORT_NAME_TYPE_LEGACY:
1872                 /* Legacy representors naming schema. */
1873                 switch_info->representor = !device_dir;
1874                 break;
1875         case MLX5_PHYS_PORT_NAME_TYPE_PFVF:
1876                 /* New representors naming schema. */
1877                 switch_info->representor = 1;
1878                 break;
1879         }
1880 }
1881
1882 /**
1883  * Extract port name, as a number, from sysfs or netlink information.
1884  *
1885  * @param[in] port_name_in
1886  *   String representing the port name.
1887  * @param[out] port_info_out
1888  *   Port information, including port name as a number and port name
1889  *   type if recognized
1890  *
1891  * @return
1892  *   port_name field set according to recognized name format.
1893  */
1894 void
1895 mlx5_translate_port_name(const char *port_name_in,
1896                          struct mlx5_switch_info *port_info_out)
1897 {
1898         char pf_c1, pf_c2, vf_c1, vf_c2;
1899         char *end;
1900         int sc_items;
1901
1902         /*
1903          * Check for port-name as a string of the form pf0vf0
1904          * (support kernel ver >= 5.0 or OFED ver >= 4.6).
1905          */
1906         sc_items = sscanf(port_name_in, "%c%c%d%c%c%d",
1907                           &pf_c1, &pf_c2, &port_info_out->pf_num,
1908                           &vf_c1, &vf_c2, &port_info_out->port_name);
1909         if (sc_items == 6 &&
1910             pf_c1 == 'p' && pf_c2 == 'f' &&
1911             vf_c1 == 'v' && vf_c2 == 'f') {
1912                 port_info_out->name_type = MLX5_PHYS_PORT_NAME_TYPE_PFVF;
1913                 return;
1914         }
1915         /*
1916          * Check for port-name as a string of the form p0
1917          * (support kernel ver >= 5.0, or OFED ver >= 4.6).
1918          */
1919         sc_items = sscanf(port_name_in, "%c%d",
1920                           &pf_c1, &port_info_out->port_name);
1921         if (sc_items == 2 && pf_c1 == 'p') {
1922                 port_info_out->name_type = MLX5_PHYS_PORT_NAME_TYPE_UPLINK;
1923                 return;
1924         }
1925         /* Check for port-name as a number (support kernel ver < 5.0 */
1926         errno = 0;
1927         port_info_out->port_name = strtol(port_name_in, &end, 0);
1928         if (!errno &&
1929             (size_t)(end - port_name_in) == strlen(port_name_in)) {
1930                 port_info_out->name_type = MLX5_PHYS_PORT_NAME_TYPE_LEGACY;
1931                 return;
1932         }
1933         port_info_out->name_type = MLX5_PHYS_PORT_NAME_TYPE_UNKNOWN;
1934         return;
1935 }