drivers/net/mlx5/mlx5_ethdev.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright 2015 6WIND S.A.
   5  *   Copyright 2015 Mellanox.
   6  *
   7  *   Redistribution and use in source and binary forms, with or without
   8  *   modification, are permitted provided that the following conditions
   9  *   are met:
  10  *
  11  *     * Redistributions of source code must retain the above copyright
  12  *       notice, this list of conditions and the following disclaimer.
  13  *     * Redistributions in binary form must reproduce the above copyright
  14  *       notice, this list of conditions and the following disclaimer in
  15  *       the documentation and/or other materials provided with the
  16  *       distribution.
  17  *     * Neither the name of 6WIND S.A. nor the names of its
  18  *       contributors may be used to endorse or promote products derived
  19  *       from this software without specific prior written permission.
  20  *
  21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  */
  33
  34 #define _GNU_SOURCE
  35
  36 #include <stddef.h>
  37 #include <assert.h>
  38 #include <unistd.h>
  39 #include <stdint.h>
  40 #include <stdio.h>
  41 #include <string.h>
  42 #include <stdlib.h>
  43 #include <errno.h>
  44 #include <dirent.h>
  45 #include <net/if.h>
  46 #include <sys/ioctl.h>
  47 #include <sys/socket.h>
  48 #include <sys/utsname.h>
  49 #include <netinet/in.h>
  50 #include <linux/ethtool.h>
  51 #include <linux/sockios.h>
  52 #include <linux/version.h>
  53 #include <fcntl.h>
  54 #include <stdalign.h>
  55 #include <sys/un.h>
  56
  57 #include <rte_atomic.h>
  58 #include <rte_ethdev.h>
  59 #include <rte_mbuf.h>
  60 #include <rte_common.h>
  61 #include <rte_interrupts.h>
  62 #include <rte_alarm.h>
  63 #include <rte_malloc.h>
  64
  65 #include "mlx5.h"
  66 #include "mlx5_rxtx.h"
  67 #include "mlx5_utils.h"
  68
  69 /* Add defines in case the running kernel is not the same as user headers. */
  70 #ifndef ETHTOOL_GLINKSETTINGS
  71 struct ethtool_link_settings {
  72         uint32_t cmd;
  73         uint32_t speed;
  74         uint8_t duplex;
  75         uint8_t port;
  76         uint8_t phy_address;
  77         uint8_t autoneg;
  78         uint8_t mdio_support;
  79         uint8_t eth_to_mdix;
  80         uint8_t eth_tp_mdix_ctrl;
  81         int8_t link_mode_masks_nwords;
  82         uint32_t reserved[8];
  83         uint32_t link_mode_masks[];
  84 };
  85
  86 #define ETHTOOL_GLINKSETTINGS 0x0000004c
  87 #define ETHTOOL_LINK_MODE_1000baseT_Full_BIT 5
  88 #define ETHTOOL_LINK_MODE_Autoneg_BIT 6
  89 #define ETHTOOL_LINK_MODE_1000baseKX_Full_BIT 17
  90 #define ETHTOOL_LINK_MODE_10000baseKX4_Full_BIT 18
  91 #define ETHTOOL_LINK_MODE_10000baseKR_Full_BIT 19
  92 #define ETHTOOL_LINK_MODE_10000baseR_FEC_BIT 20
  93 #define ETHTOOL_LINK_MODE_20000baseMLD2_Full_BIT 21
  94 #define ETHTOOL_LINK_MODE_20000baseKR2_Full_BIT 22
  95 #define ETHTOOL_LINK_MODE_40000baseKR4_Full_BIT 23
  96 #define ETHTOOL_LINK_MODE_40000baseCR4_Full_BIT 24
  97 #define ETHTOOL_LINK_MODE_40000baseSR4_Full_BIT 25
  98 #define ETHTOOL_LINK_MODE_40000baseLR4_Full_BIT 26
  99 #define ETHTOOL_LINK_MODE_56000baseKR4_Full_BIT 27
 100 #define ETHTOOL_LINK_MODE_56000baseCR4_Full_BIT 28
 101 #define ETHTOOL_LINK_MODE_56000baseSR4_Full_BIT 29
 102 #define ETHTOOL_LINK_MODE_56000baseLR4_Full_BIT 30
 103 #endif
 104 #ifndef HAVE_ETHTOOL_LINK_MODE_25G
 105 #define ETHTOOL_LINK_MODE_25000baseCR_Full_BIT 31
 106 #define ETHTOOL_LINK_MODE_25000baseKR_Full_BIT 32
 107 #define ETHTOOL_LINK_MODE_25000baseSR_Full_BIT 33
 108 #endif
 109 #ifndef HAVE_ETHTOOL_LINK_MODE_50G
 110 #define ETHTOOL_LINK_MODE_50000baseCR2_Full_BIT 34
 111 #define ETHTOOL_LINK_MODE_50000baseKR2_Full_BIT 35
 112 #endif
 113 #ifndef HAVE_ETHTOOL_LINK_MODE_100G
 114 #define ETHTOOL_LINK_MODE_100000baseKR4_Full_BIT 36
 115 #define ETHTOOL_LINK_MODE_100000baseSR4_Full_BIT 37
 116 #define ETHTOOL_LINK_MODE_100000baseCR4_Full_BIT 38
 117 #define ETHTOOL_LINK_MODE_100000baseLR4_ER4_Full_BIT 39
 118 #endif
 119
 120 /**
 121  * Return private structure associated with an Ethernet device.
 122  *
 123  * @param dev
 124  *   Pointer to Ethernet device structure.
 125  *
 126  * @return
 127  *   Pointer to private structure.
 128  */
 129 struct priv *
 130 mlx5_get_priv(struct rte_eth_dev *dev)
 131 {
 132         return dev->data->dev_private;
 133 }
 134
 135 /**
 136  * Check if running as a secondary process.
 137  *
 138  * @return
 139  *   Nonzero if running as a secondary process.
 140  */
 141 inline int
 142 mlx5_is_secondary(void)
 143 {
 144         return rte_eal_process_type() == RTE_PROC_SECONDARY;
 145 }
 146
 147 /**
 148  * Get interface name from private structure.
 149  *
 150  * @param[in] priv
 151  *   Pointer to private structure.
 152  * @param[out] ifname
 153  *   Interface name output buffer.
 154  *
 155  * @return
 156  *   0 on success, -1 on failure and errno is set.
 157  */
 158 int
 159 priv_get_ifname(const struct priv *priv, char (*ifname)[IF_NAMESIZE])
 160 {
 161         DIR *dir;
 162         struct dirent *dent;
 163         unsigned int dev_type = 0;
 164         unsigned int dev_port_prev = ~0u;
 165         char match[IF_NAMESIZE] = "";
 166
 167         {
 168                 MKSTR(path, "%s/device/net", priv->ibdev_path);
 169
 170                 dir = opendir(path);
 171                 if (dir == NULL)
 172                         return -1;
 173         }
 174         while ((dent = readdir(dir)) != NULL) {
 175                 char *name = dent->d_name;
 176                 FILE *file;
 177                 unsigned int dev_port;
 178                 int r;
 179
 180                 if ((name[0] == '.') &&
 181                     ((name[1] == '\0') ||
 182                      ((name[1] == '.') && (name[2] == '\0'))))
 183                         continue;
 184
 185                 MKSTR(path, "%s/device/net/%s/%s",
 186                       priv->ibdev_path, name,
 187                       (dev_type ? "dev_id" : "dev_port"));
 188
 189                 file = fopen(path, "rb");
 190                 if (file == NULL) {
 191                         if (errno != ENOENT)
 192                                 continue;
 193                         /*
 194                          * Switch to dev_id when dev_port does not exist as
 195                          * is the case with Linux kernel versions < 3.15.
 196                          */
 197 try_dev_id:
 198                         match[0] = '\0';
 199                         if (dev_type)
 200                                 break;
 201                         dev_type = 1;
 202                         dev_port_prev = ~0u;
 203                         rewinddir(dir);
 204                         continue;
 205                 }
 206                 r = fscanf(file, (dev_type ? "%x" : "%u"), &dev_port);
 207                 fclose(file);
 208                 if (r != 1)
 209                         continue;
 210                 /*
 211                  * Switch to dev_id when dev_port returns the same value for
 212                  * all ports. May happen when using a MOFED release older than
 213                  * 3.0 with a Linux kernel >= 3.15.
 214                  */
 215                 if (dev_port == dev_port_prev)
 216                         goto try_dev_id;
 217                 dev_port_prev = dev_port;
 218                 if (dev_port == (priv->port - 1u))
 219                         snprintf(match, sizeof(match), "%s", name);
 220         }
 221         closedir(dir);
 222         if (match[0] == '\0')
 223                 return -1;
 224         strncpy(*ifname, match, sizeof(*ifname));
 225         return 0;
 226 }
 227
 228 /**
 229  * Check if the counter is located on ib counters file.
 230  *
 231  * @param[in] cntr
 232  *   Counter name.
 233  *
 234  * @return
 235  *   1 if counter is located on ib counters file , 0 otherwise.
 236  */
 237 int
 238 priv_is_ib_cntr(const char *cntr)
 239 {
 240         if (!strcmp(cntr, "out_of_buffer"))
 241                 return 1;
 242         return 0;
 243 }
 244
 245 /**
 246  * Read from sysfs entry.
 247  *
 248  * @param[in] priv
 249  *   Pointer to private structure.
 250  * @param[in] entry
 251  *   Entry name relative to sysfs path.
 252  * @param[out] buf
 253  *   Data output buffer.
 254  * @param size
 255  *   Buffer size.
 256  *
 257  * @return
 258  *   0 on success, -1 on failure and errno is set.
 259  */
 260 static int
 261 priv_sysfs_read(const struct priv *priv, const char *entry,
 262                 char *buf, size_t size)
 263 {
 264         char ifname[IF_NAMESIZE];
 265         FILE *file;
 266         int ret;
 267         int err;
 268
 269         if (priv_get_ifname(priv, &ifname))
 270                 return -1;
 271
 272         if (priv_is_ib_cntr(entry)) {
 273                 MKSTR(path, "%s/ports/1/hw_counters/%s",
 274                       priv->ibdev_path, entry);
 275                 file = fopen(path, "rb");
 276         } else {
 277                 MKSTR(path, "%s/device/net/%s/%s",
 278                       priv->ibdev_path, ifname, entry);
 279                 file = fopen(path, "rb");
 280         }
 281         if (file == NULL)
 282                 return -1;
 283         ret = fread(buf, 1, size, file);
 284         err = errno;
 285         if (((size_t)ret < size) && (ferror(file)))
 286                 ret = -1;
 287         else
 288                 ret = size;
 289         fclose(file);
 290         errno = err;
 291         return ret;
 292 }
 293
 294 /**
 295  * Write to sysfs entry.
 296  *
 297  * @param[in] priv
 298  *   Pointer to private structure.
 299  * @param[in] entry
 300  *   Entry name relative to sysfs path.
 301  * @param[in] buf
 302  *   Data buffer.
 303  * @param size
 304  *   Buffer size.
 305  *
 306  * @return
 307  *   0 on success, -1 on failure and errno is set.
 308  */
 309 static int
 310 priv_sysfs_write(const struct priv *priv, const char *entry,
 311                  char *buf, size_t size)
 312 {
 313         char ifname[IF_NAMESIZE];
 314         FILE *file;
 315         int ret;
 316         int err;
 317
 318         if (priv_get_ifname(priv, &ifname))
 319                 return -1;
 320
 321         MKSTR(path, "%s/device/net/%s/%s", priv->ibdev_path, ifname, entry);
 322
 323         file = fopen(path, "wb");
 324         if (file == NULL)
 325                 return -1;
 326         ret = fwrite(buf, 1, size, file);
 327         err = errno;
 328         if (((size_t)ret < size) || (ferror(file)))
 329                 ret = -1;
 330         else
 331                 ret = size;
 332         fclose(file);
 333         errno = err;
 334         return ret;
 335 }
 336
 337 /**
 338  * Get unsigned long sysfs property.
 339  *
 340  * @param priv
 341  *   Pointer to private structure.
 342  * @param[in] name
 343  *   Entry name relative to sysfs path.
 344  * @param[out] value
 345  *   Value output buffer.
 346  *
 347  * @return
 348  *   0 on success, -1 on failure and errno is set.
 349  */
 350 static int
 351 priv_get_sysfs_ulong(struct priv *priv, const char *name, unsigned long *value)
 352 {
 353         int ret;
 354         unsigned long value_ret;
 355         char value_str[32];
 356
 357         ret = priv_sysfs_read(priv, name, value_str, (sizeof(value_str) - 1));
 358         if (ret == -1) {
 359                 DEBUG("cannot read %s value from sysfs: %s",
 360                       name, strerror(errno));
 361                 return -1;
 362         }
 363         value_str[ret] = '\0';
 364         errno = 0;
 365         value_ret = strtoul(value_str, NULL, 0);
 366         if (errno) {
 367                 DEBUG("invalid %s value `%s': %s", name, value_str,
 368                       strerror(errno));
 369                 return -1;
 370         }
 371         *value = value_ret;
 372         return 0;
 373 }
 374
 375 /**
 376  * Set unsigned long sysfs property.
 377  *
 378  * @param priv
 379  *   Pointer to private structure.
 380  * @param[in] name
 381  *   Entry name relative to sysfs path.
 382  * @param value
 383  *   Value to set.
 384  *
 385  * @return
 386  *   0 on success, -1 on failure and errno is set.
 387  */
 388 static int
 389 priv_set_sysfs_ulong(struct priv *priv, const char *name, unsigned long value)
 390 {
 391         int ret;
 392         MKSTR(value_str, "%lu", value);
 393
 394         ret = priv_sysfs_write(priv, name, value_str, (sizeof(value_str) - 1));
 395         if (ret == -1) {
 396                 DEBUG("cannot write %s `%s' (%lu) to sysfs: %s",
 397                       name, value_str, value, strerror(errno));
 398                 return -1;
 399         }
 400         return 0;
 401 }
 402
 403 /**
 404  * Perform ifreq ioctl() on associated Ethernet device.
 405  *
 406  * @param[in] priv
 407  *   Pointer to private structure.
 408  * @param req
 409  *   Request number to pass to ioctl().
 410  * @param[out] ifr
 411  *   Interface request structure output buffer.
 412  *
 413  * @return
 414  *   0 on success, -1 on failure and errno is set.
 415  */
 416 int
 417 priv_ifreq(const struct priv *priv, int req, struct ifreq *ifr)
 418 {
 419         int sock = socket(PF_INET, SOCK_DGRAM, IPPROTO_IP);
 420         int ret = -1;
 421
 422         if (sock == -1)
 423                 return ret;
 424         if (priv_get_ifname(priv, &ifr->ifr_name) == 0)
 425                 ret = ioctl(sock, req, ifr);
 426         close(sock);
 427         return ret;
 428 }
 429
 430 /**
 431  * Return the number of active VFs for the current device.
 432  *
 433  * @param[in] priv
 434  *   Pointer to private structure.
 435  * @param[out] num_vfs
 436  *   Number of active VFs.
 437  *
 438  * @return
 439  *   0 on success, -1 on failure and errno is set.
 440  */
 441 int
 442 priv_get_num_vfs(struct priv *priv, uint16_t *num_vfs)
 443 {
 444         /* The sysfs entry name depends on the operating system. */
 445         const char **name = (const char *[]){
 446                 "device/sriov_numvfs",
 447                 "device/mlx5_num_vfs",
 448                 NULL,
 449         };
 450         int ret;
 451
 452         do {
 453                 unsigned long ulong_num_vfs;
 454
 455                 ret = priv_get_sysfs_ulong(priv, *name, &ulong_num_vfs);
 456                 if (!ret)
 457                         *num_vfs = ulong_num_vfs;
 458         } while (*(++name) && ret);
 459         return ret;
 460 }
 461
 462 /**
 463  * Get device MTU.
 464  *
 465  * @param priv
 466  *   Pointer to private structure.
 467  * @param[out] mtu
 468  *   MTU value output buffer.
 469  *
 470  * @return
 471  *   0 on success, -1 on failure and errno is set.
 472  */
 473 int
 474 priv_get_mtu(struct priv *priv, uint16_t *mtu)
 475 {
 476         unsigned long ulong_mtu;
 477
 478         if (priv_get_sysfs_ulong(priv, "mtu", &ulong_mtu) == -1)
 479                 return -1;
 480         *mtu = ulong_mtu;
 481         return 0;
 482 }
 483
 484 /**
 485  * Read device counter from sysfs.
 486  *
 487  * @param priv
 488  *   Pointer to private structure.
 489  * @param name
 490  *   Counter name.
 491  * @param[out] cntr
 492  *   Counter output buffer.
 493  *
 494  * @return
 495  *   0 on success, -1 on failure and errno is set.
 496  */
 497 int
 498 priv_get_cntr_sysfs(struct priv *priv, const char *name, uint64_t *cntr)
 499 {
 500         unsigned long ulong_ctr;
 501
 502         if (priv_get_sysfs_ulong(priv, name, &ulong_ctr) == -1)
 503                 return -1;
 504         *cntr = ulong_ctr;
 505         return 0;
 506 }
 507
 508 /**
 509  * Set device MTU.
 510  *
 511  * @param priv
 512  *   Pointer to private structure.
 513  * @param mtu
 514  *   MTU value to set.
 515  *
 516  * @return
 517  *   0 on success, -1 on failure and errno is set.
 518  */
 519 static int
 520 priv_set_mtu(struct priv *priv, uint16_t mtu)
 521 {
 522         uint16_t new_mtu;
 523
 524         if (priv_set_sysfs_ulong(priv, "mtu", mtu) ||
 525             priv_get_mtu(priv, &new_mtu))
 526                 return -1;
 527         if (new_mtu == mtu)
 528                 return 0;
 529         errno = EINVAL;
 530         return -1;
 531 }
 532
 533 /**
 534  * Set device flags.
 535  *
 536  * @param priv
 537  *   Pointer to private structure.
 538  * @param keep
 539  *   Bitmask for flags that must remain untouched.
 540  * @param flags
 541  *   Bitmask for flags to modify.
 542  *
 543  * @return
 544  *   0 on success, -1 on failure and errno is set.
 545  */
 546 int
 547 priv_set_flags(struct priv *priv, unsigned int keep, unsigned int flags)
 548 {
 549         unsigned long tmp;
 550
 551         if (priv_get_sysfs_ulong(priv, "flags", &tmp) == -1)
 552                 return -1;
 553         tmp &= keep;
 554         tmp |= (flags & (~keep));
 555         return priv_set_sysfs_ulong(priv, "flags", tmp);
 556 }
 557
 558 /**
 559  * Ethernet device configuration.
 560  *
 561  * Prepare the driver for a given number of TX and RX queues.
 562  *
 563  * @param dev
 564  *   Pointer to Ethernet device structure.
 565  *
 566  * @return
 567  *   0 on success, errno value on failure.
 568  */
 569 static int
 570 dev_configure(struct rte_eth_dev *dev)
 571 {
 572         struct priv *priv = dev->data->dev_private;
 573         unsigned int rxqs_n = dev->data->nb_rx_queues;
 574         unsigned int txqs_n = dev->data->nb_tx_queues;
 575         unsigned int i;
 576         unsigned int j;
 577         unsigned int reta_idx_n;
 578         const uint8_t use_app_rss_key =
 579                 !!dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key_len;
 580
 581         if (use_app_rss_key &&
 582             (dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key_len !=
 583              rss_hash_default_key_len)) {
 584                 /* MLX5 RSS only support 40bytes key. */
 585                 return EINVAL;
 586         }
 587         priv->rss_conf.rss_key =
 588                 rte_realloc(priv->rss_conf.rss_key,
 589                             rss_hash_default_key_len, 0);
 590         if (!priv->rss_conf.rss_key) {
 591                 ERROR("cannot allocate RSS hash key memory (%u)", rxqs_n);
 592                 return ENOMEM;
 593         }
 594         memcpy(priv->rss_conf.rss_key,
 595                use_app_rss_key ?
 596                dev->data->dev_conf.rx_adv_conf.rss_conf.rss_key :
 597                rss_hash_default_key,
 598                rss_hash_default_key_len);
 599         priv->rss_conf.rss_key_len = rss_hash_default_key_len;
 600         priv->rss_conf.rss_hf = dev->data->dev_conf.rx_adv_conf.rss_conf.rss_hf;
 601         priv->rxqs = (void *)dev->data->rx_queues;
 602         priv->txqs = (void *)dev->data->tx_queues;
 603         if (txqs_n != priv->txqs_n) {
 604                 INFO("%p: TX queues number update: %u -> %u",
 605                      (void *)dev, priv->txqs_n, txqs_n);
 606                 priv->txqs_n = txqs_n;
 607         }
 608         if (rxqs_n > priv->ind_table_max_size) {
 609                 ERROR("cannot handle this many RX queues (%u)", rxqs_n);
 610                 return EINVAL;
 611         }
 612         if (rxqs_n == priv->rxqs_n)
 613                 return 0;
 614         INFO("%p: RX queues number update: %u -> %u",
 615              (void *)dev, priv->rxqs_n, rxqs_n);
 616         priv->rxqs_n = rxqs_n;
 617         /* If the requested number of RX queues is not a power of two, use the
 618          * maximum indirection table size for better balancing.
 619          * The result is always rounded to the next power of two. */
 620         reta_idx_n = (1 << log2above((rxqs_n & (rxqs_n - 1)) ?
 621                                      priv->ind_table_max_size :
 622                                      rxqs_n));
 623         if (priv_rss_reta_index_resize(priv, reta_idx_n))
 624                 return ENOMEM;
 625         /* When the number of RX queues is not a power of two, the remaining
 626          * table entries are padded with reused WQs and hashes are not spread
 627          * uniformly. */
 628         for (i = 0, j = 0; (i != reta_idx_n); ++i) {
 629                 (*priv->reta_idx)[i] = j;
 630                 if (++j == rxqs_n)
 631                         j = 0;
 632         }
 633         return 0;
 634 }
 635
 636 /**
 637  * DPDK callback for Ethernet device configuration.
 638  *
 639  * @param dev
 640  *   Pointer to Ethernet device structure.
 641  *
 642  * @return
 643  *   0 on success, negative errno value on failure.
 644  */
 645 int
 646 mlx5_dev_configure(struct rte_eth_dev *dev)
 647 {
 648         struct priv *priv = dev->data->dev_private;
 649         int ret;
 650
 651         if (mlx5_is_secondary())
 652                 return -E_RTE_SECONDARY;
 653
 654         priv_lock(priv);
 655         ret = dev_configure(dev);
 656         assert(ret >= 0);
 657         priv_unlock(priv);
 658         return -ret;
 659 }
 660
 661 /**
 662  * DPDK callback to get information about the device.
 663  *
 664  * @param dev
 665  *   Pointer to Ethernet device structure.
 666  * @param[out] info
 667  *   Info structure output buffer.
 668  */
 669 void
 670 mlx5_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *info)
 671 {
 672         struct priv *priv = mlx5_get_priv(dev);
 673         unsigned int max;
 674         char ifname[IF_NAMESIZE];
 675
 676         info->pci_dev = RTE_ETH_DEV_TO_PCI(dev);
 677
 678         priv_lock(priv);
 679         /* FIXME: we should ask the device for these values. */
 680         info->min_rx_bufsize = 32;
 681         info->max_rx_pktlen = 65536;
 682         /*
 683          * Since we need one CQ per QP, the limit is the minimum number
 684          * between the two values.
 685          */
 686         max = RTE_MIN(priv->device_attr.orig_attr.max_cq,
 687                       priv->device_attr.orig_attr.max_qp);
 688         /* If max >= 65535 then max = 0, max_rx_queues is uint16_t. */
 689         if (max >= 65535)
 690                 max = 65535;
 691         info->max_rx_queues = max;
 692         info->max_tx_queues = max;
 693         info->max_mac_addrs = RTE_DIM(priv->mac);
 694         info->rx_offload_capa =
 695                 (priv->hw_csum ?
 696                  (DEV_RX_OFFLOAD_IPV4_CKSUM |
 697                   DEV_RX_OFFLOAD_UDP_CKSUM |
 698                   DEV_RX_OFFLOAD_TCP_CKSUM) :
 699                  0) |
 700                 (priv->hw_vlan_strip ? DEV_RX_OFFLOAD_VLAN_STRIP : 0) |
 701                 DEV_RX_OFFLOAD_TIMESTAMP;
 702
 703         if (!priv->mps)
 704                 info->tx_offload_capa = DEV_TX_OFFLOAD_VLAN_INSERT;
 705         if (priv->hw_csum)
 706                 info->tx_offload_capa |=
 707                         (DEV_TX_OFFLOAD_IPV4_CKSUM |
 708                          DEV_TX_OFFLOAD_UDP_CKSUM |
 709                          DEV_TX_OFFLOAD_TCP_CKSUM);
 710         if (priv->tso)
 711                 info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
 712         if (priv->tunnel_en)
 713                 info->tx_offload_capa |= (DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM |
 714                                           DEV_TX_OFFLOAD_VXLAN_TNL_TSO |
 715                                           DEV_TX_OFFLOAD_GRE_TNL_TSO);
 716         if (priv_get_ifname(priv, &ifname) == 0)
 717                 info->if_index = if_nametoindex(ifname);
 718         info->reta_size = priv->reta_idx_n ?
 719                 priv->reta_idx_n : priv->ind_table_max_size;
 720         info->hash_key_size = priv->rss_conf.rss_key_len;
 721         info->speed_capa = priv->link_speed_capa;
 722         priv_unlock(priv);
 723 }
 724
 725 const uint32_t *
 726 mlx5_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 727 {
 728         static const uint32_t ptypes[] = {
 729                 /* refers to rxq_cq_to_pkt_type() */
 730                 RTE_PTYPE_L2_ETHER,
 731                 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
 732                 RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
 733                 RTE_PTYPE_L4_NONFRAG,
 734                 RTE_PTYPE_L4_FRAG,
 735                 RTE_PTYPE_L4_TCP,
 736                 RTE_PTYPE_L4_UDP,
 737                 RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN,
 738                 RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN,
 739                 RTE_PTYPE_INNER_L4_NONFRAG,
 740                 RTE_PTYPE_INNER_L4_FRAG,
 741                 RTE_PTYPE_INNER_L4_TCP,
 742                 RTE_PTYPE_INNER_L4_UDP,
 743                 RTE_PTYPE_UNKNOWN
 744         };
 745
 746         if (dev->rx_pkt_burst == mlx5_rx_burst ||
 747             dev->rx_pkt_burst == mlx5_rx_burst_vec)
 748                 return ptypes;
 749         return NULL;
 750 }
 751
 752 /**
 753  * DPDK callback to retrieve physical link information.
 754  *
 755  * @param dev
 756  *   Pointer to Ethernet device structure.
 757  * @param wait_to_complete
 758  *   Wait for request completion (ignored).
 759  */
 760 static int
 761 mlx5_link_update_unlocked_gset(struct rte_eth_dev *dev, int wait_to_complete)
 762 {
 763         struct priv *priv = mlx5_get_priv(dev);
 764         struct ethtool_cmd edata = {
 765                 .cmd = ETHTOOL_GSET /* Deprecated since Linux v4.5. */
 766         };
 767         struct ifreq ifr;
 768         struct rte_eth_link dev_link;
 769         int link_speed = 0;
 770
 771         /* priv_lock() is not taken to allow concurrent calls. */
 772
 773         (void)wait_to_complete;
 774         if (priv_ifreq(priv, SIOCGIFFLAGS, &ifr)) {
 775                 WARN("ioctl(SIOCGIFFLAGS) failed: %s", strerror(errno));
 776                 return -1;
 777         }
 778         memset(&dev_link, 0, sizeof(dev_link));
 779         dev_link.link_status = ((ifr.ifr_flags & IFF_UP) &&
 780                                 (ifr.ifr_flags & IFF_RUNNING));
 781         ifr.ifr_data = (void *)&edata;
 782         if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) {
 783                 WARN("ioctl(SIOCETHTOOL, ETHTOOL_GSET) failed: %s",
 784                      strerror(errno));
 785                 return -1;
 786         }
 787         link_speed = ethtool_cmd_speed(&edata);
 788         if (link_speed == -1)
 789                 dev_link.link_speed = 0;
 790         else
 791                 dev_link.link_speed = link_speed;
 792         priv->link_speed_capa = 0;
 793         if (edata.supported & SUPPORTED_Autoneg)
 794                 priv->link_speed_capa |= ETH_LINK_SPEED_AUTONEG;
 795         if (edata.supported & (SUPPORTED_1000baseT_Full |
 796                                SUPPORTED_1000baseKX_Full))
 797                 priv->link_speed_capa |= ETH_LINK_SPEED_1G;
 798         if (edata.supported & SUPPORTED_10000baseKR_Full)
 799                 priv->link_speed_capa |= ETH_LINK_SPEED_10G;
 800         if (edata.supported & (SUPPORTED_40000baseKR4_Full |
 801                                SUPPORTED_40000baseCR4_Full |
 802                                SUPPORTED_40000baseSR4_Full |
 803                                SUPPORTED_40000baseLR4_Full))
 804                 priv->link_speed_capa |= ETH_LINK_SPEED_40G;
 805         dev_link.link_duplex = ((edata.duplex == DUPLEX_HALF) ?
 806                                 ETH_LINK_HALF_DUPLEX : ETH_LINK_FULL_DUPLEX);
 807         dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds &
 808                         ETH_LINK_SPEED_FIXED);
 809         if (memcmp(&dev_link, &dev->data->dev_link, sizeof(dev_link))) {
 810                 /* Link status changed. */
 811                 dev->data->dev_link = dev_link;
 812                 return 0;
 813         }
 814         /* Link status is still the same. */
 815         return -1;
 816 }
 817
 818 /**
 819  * Retrieve physical link information (unlocked version using new ioctl).
 820  *
 821  * @param dev
 822  *   Pointer to Ethernet device structure.
 823  * @param wait_to_complete
 824  *   Wait for request completion (ignored).
 825  */
 826 static int
 827 mlx5_link_update_unlocked_gs(struct rte_eth_dev *dev, int wait_to_complete)
 828 {
 829         struct priv *priv = mlx5_get_priv(dev);
 830         struct ethtool_link_settings gcmd = { .cmd = ETHTOOL_GLINKSETTINGS };
 831         struct ifreq ifr;
 832         struct rte_eth_link dev_link;
 833         uint64_t sc;
 834
 835         (void)wait_to_complete;
 836         if (priv_ifreq(priv, SIOCGIFFLAGS, &ifr)) {
 837                 WARN("ioctl(SIOCGIFFLAGS) failed: %s", strerror(errno));
 838                 return -1;
 839         }
 840         memset(&dev_link, 0, sizeof(dev_link));
 841         dev_link.link_status = ((ifr.ifr_flags & IFF_UP) &&
 842                                 (ifr.ifr_flags & IFF_RUNNING));
 843         ifr.ifr_data = (void *)&gcmd;
 844         if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) {
 845                 DEBUG("ioctl(SIOCETHTOOL, ETHTOOL_GLINKSETTINGS) failed: %s",
 846                       strerror(errno));
 847                 return -1;
 848         }
 849         gcmd.link_mode_masks_nwords = -gcmd.link_mode_masks_nwords;
 850
 851         alignas(struct ethtool_link_settings)
 852         uint8_t data[offsetof(struct ethtool_link_settings, link_mode_masks) +
 853                      sizeof(uint32_t) * gcmd.link_mode_masks_nwords * 3];
 854         struct ethtool_link_settings *ecmd = (void *)data;
 855
 856         *ecmd = gcmd;
 857         ifr.ifr_data = (void *)ecmd;
 858         if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) {
 859                 DEBUG("ioctl(SIOCETHTOOL, ETHTOOL_GLINKSETTINGS) failed: %s",
 860                       strerror(errno));
 861                 return -1;
 862         }
 863         dev_link.link_speed = ecmd->speed;
 864         sc = ecmd->link_mode_masks[0] |
 865                 ((uint64_t)ecmd->link_mode_masks[1] << 32);
 866         priv->link_speed_capa = 0;
 867         if (sc & ETHTOOL_LINK_MODE_Autoneg_BIT)
 868                 priv->link_speed_capa |= ETH_LINK_SPEED_AUTONEG;
 869         if (sc & (ETHTOOL_LINK_MODE_1000baseT_Full_BIT |
 870                   ETHTOOL_LINK_MODE_1000baseKX_Full_BIT))
 871                 priv->link_speed_capa |= ETH_LINK_SPEED_1G;
 872         if (sc & (ETHTOOL_LINK_MODE_10000baseKX4_Full_BIT |
 873                   ETHTOOL_LINK_MODE_10000baseKR_Full_BIT |
 874                   ETHTOOL_LINK_MODE_10000baseR_FEC_BIT))
 875                 priv->link_speed_capa |= ETH_LINK_SPEED_10G;
 876         if (sc & (ETHTOOL_LINK_MODE_20000baseMLD2_Full_BIT |
 877                   ETHTOOL_LINK_MODE_20000baseKR2_Full_BIT))
 878                 priv->link_speed_capa |= ETH_LINK_SPEED_20G;
 879         if (sc & (ETHTOOL_LINK_MODE_40000baseKR4_Full_BIT |
 880                   ETHTOOL_LINK_MODE_40000baseCR4_Full_BIT |
 881                   ETHTOOL_LINK_MODE_40000baseSR4_Full_BIT |
 882                   ETHTOOL_LINK_MODE_40000baseLR4_Full_BIT))
 883                 priv->link_speed_capa |= ETH_LINK_SPEED_40G;
 884         if (sc & (ETHTOOL_LINK_MODE_56000baseKR4_Full_BIT |
 885                   ETHTOOL_LINK_MODE_56000baseCR4_Full_BIT |
 886                   ETHTOOL_LINK_MODE_56000baseSR4_Full_BIT |
 887                   ETHTOOL_LINK_MODE_56000baseLR4_Full_BIT))
 888                 priv->link_speed_capa |= ETH_LINK_SPEED_56G;
 889         if (sc & (ETHTOOL_LINK_MODE_25000baseCR_Full_BIT |
 890                   ETHTOOL_LINK_MODE_25000baseKR_Full_BIT |
 891                   ETHTOOL_LINK_MODE_25000baseSR_Full_BIT))
 892                 priv->link_speed_capa |= ETH_LINK_SPEED_25G;
 893         if (sc & (ETHTOOL_LINK_MODE_50000baseCR2_Full_BIT |
 894                   ETHTOOL_LINK_MODE_50000baseKR2_Full_BIT))
 895                 priv->link_speed_capa |= ETH_LINK_SPEED_50G;
 896         if (sc & (ETHTOOL_LINK_MODE_100000baseKR4_Full_BIT |
 897                   ETHTOOL_LINK_MODE_100000baseSR4_Full_BIT |
 898                   ETHTOOL_LINK_MODE_100000baseCR4_Full_BIT |
 899                   ETHTOOL_LINK_MODE_100000baseLR4_ER4_Full_BIT))
 900                 priv->link_speed_capa |= ETH_LINK_SPEED_100G;
 901         dev_link.link_duplex = ((ecmd->duplex == DUPLEX_HALF) ?
 902                                 ETH_LINK_HALF_DUPLEX : ETH_LINK_FULL_DUPLEX);
 903         dev_link.link_autoneg = !(dev->data->dev_conf.link_speeds &
 904                                   ETH_LINK_SPEED_FIXED);
 905         if (memcmp(&dev_link, &dev->data->dev_link, sizeof(dev_link))) {
 906                 /* Link status changed. */
 907                 dev->data->dev_link = dev_link;
 908                 return 0;
 909         }
 910         /* Link status is still the same. */
 911         return -1;
 912 }
 913
 914 /**
 915  * DPDK callback to retrieve physical link information.
 916  *
 917  * @param dev
 918  *   Pointer to Ethernet device structure.
 919  * @param wait_to_complete
 920  *   Wait for request completion (ignored).
 921  */
 922 int
 923 mlx5_link_update(struct rte_eth_dev *dev, int wait_to_complete)
 924 {
 925         struct utsname utsname;
 926         int ver[3];
 927
 928         if (uname(&utsname) == -1 ||
 929             sscanf(utsname.release, "%d.%d.%d",
 930                    &ver[0], &ver[1], &ver[2]) != 3 ||
 931             KERNEL_VERSION(ver[0], ver[1], ver[2]) < KERNEL_VERSION(4, 9, 0))
 932                 return mlx5_link_update_unlocked_gset(dev, wait_to_complete);
 933         return mlx5_link_update_unlocked_gs(dev, wait_to_complete);
 934 }
 935
 936 /**
 937  * DPDK callback to change the MTU.
 938  *
 939  * @param dev
 940  *   Pointer to Ethernet device structure.
 941  * @param in_mtu
 942  *   New MTU.
 943  *
 944  * @return
 945  *   0 on success, negative errno value on failure.
 946  */
 947 int
 948 mlx5_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
 949 {
 950         struct priv *priv = dev->data->dev_private;
 951         uint16_t kern_mtu;
 952         int ret = 0;
 953
 954         if (mlx5_is_secondary())
 955                 return -E_RTE_SECONDARY;
 956
 957         priv_lock(priv);
 958         ret = priv_get_mtu(priv, &kern_mtu);
 959         if (ret)
 960                 goto out;
 961         /* Set kernel interface MTU first. */
 962         ret = priv_set_mtu(priv, mtu);
 963         if (ret)
 964                 goto out;
 965         ret = priv_get_mtu(priv, &kern_mtu);
 966         if (ret)
 967                 goto out;
 968         if (kern_mtu == mtu) {
 969                 priv->mtu = mtu;
 970                 DEBUG("adapter port %u MTU set to %u", priv->port, mtu);
 971         }
 972         priv_unlock(priv);
 973         return 0;
 974 out:
 975         ret = errno;
 976         WARN("cannot set port %u MTU to %u: %s", priv->port, mtu,
 977              strerror(ret));
 978         priv_unlock(priv);
 979         assert(ret >= 0);
 980         return -ret;
 981 }
 982
 983 /**
 984  * DPDK callback to get flow control status.
 985  *
 986  * @param dev
 987  *   Pointer to Ethernet device structure.
 988  * @param[out] fc_conf
 989  *   Flow control output buffer.
 990  *
 991  * @return
 992  *   0 on success, negative errno value on failure.
 993  */
 994 int
 995 mlx5_dev_get_flow_ctrl(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 996 {
 997         struct priv *priv = dev->data->dev_private;
 998         struct ifreq ifr;
 999         struct ethtool_pauseparam ethpause = {
1000                 .cmd = ETHTOOL_GPAUSEPARAM
1001         };
1002         int ret;
1003
1004         if (mlx5_is_secondary())
1005                 return -E_RTE_SECONDARY;
1006
1007         ifr.ifr_data = (void *)&ethpause;
1008         priv_lock(priv);
1009         if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) {
1010                 ret = errno;
1011                 WARN("ioctl(SIOCETHTOOL, ETHTOOL_GPAUSEPARAM)"
1012                      " failed: %s",
1013                      strerror(ret));
1014                 goto out;
1015         }
1016
1017         fc_conf->autoneg = ethpause.autoneg;
1018         if (ethpause.rx_pause && ethpause.tx_pause)
1019                 fc_conf->mode = RTE_FC_FULL;
1020         else if (ethpause.rx_pause)
1021                 fc_conf->mode = RTE_FC_RX_PAUSE;
1022         else if (ethpause.tx_pause)
1023                 fc_conf->mode = RTE_FC_TX_PAUSE;
1024         else
1025                 fc_conf->mode = RTE_FC_NONE;
1026         ret = 0;
1027
1028 out:
1029         priv_unlock(priv);
1030         assert(ret >= 0);
1031         return -ret;
1032 }
1033
1034 /**
1035  * DPDK callback to modify flow control parameters.
1036  *
1037  * @param dev
1038  *   Pointer to Ethernet device structure.
1039  * @param[in] fc_conf
1040  *   Flow control parameters.
1041  *
1042  * @return
1043  *   0 on success, negative errno value on failure.
1044  */
1045 int
1046 mlx5_dev_set_flow_ctrl(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1047 {
1048         struct priv *priv = dev->data->dev_private;
1049         struct ifreq ifr;
1050         struct ethtool_pauseparam ethpause = {
1051                 .cmd = ETHTOOL_SPAUSEPARAM
1052         };
1053         int ret;
1054
1055         if (mlx5_is_secondary())
1056                 return -E_RTE_SECONDARY;
1057
1058         ifr.ifr_data = (void *)&ethpause;
1059         ethpause.autoneg = fc_conf->autoneg;
1060         if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) ||
1061             (fc_conf->mode & RTE_FC_RX_PAUSE))
1062                 ethpause.rx_pause = 1;
1063         else
1064                 ethpause.rx_pause = 0;
1065
1066         if (((fc_conf->mode & RTE_FC_FULL) == RTE_FC_FULL) ||
1067             (fc_conf->mode & RTE_FC_TX_PAUSE))
1068                 ethpause.tx_pause = 1;
1069         else
1070                 ethpause.tx_pause = 0;
1071
1072         priv_lock(priv);
1073         if (priv_ifreq(priv, SIOCETHTOOL, &ifr)) {
1074                 ret = errno;
1075                 WARN("ioctl(SIOCETHTOOL, ETHTOOL_SPAUSEPARAM)"
1076                      " failed: %s",
1077                      strerror(ret));
1078                 goto out;
1079         }
1080         ret = 0;
1081
1082 out:
1083         priv_unlock(priv);
1084         assert(ret >= 0);
1085         return -ret;
1086 }
1087
1088 /**
1089  * Get PCI information from struct ibv_device.
1090  *
1091  * @param device
1092  *   Pointer to Ethernet device structure.
1093  * @param[out] pci_addr
1094  *   PCI bus address output buffer.
1095  *
1096  * @return
1097  *   0 on success, -1 on failure and errno is set.
1098  */
1099 int
1100 mlx5_ibv_device_to_pci_addr(const struct ibv_device *device,
1101                             struct rte_pci_addr *pci_addr)
1102 {
1103         FILE *file;
1104         char line[32];
1105         MKSTR(path, "%s/device/uevent", device->ibdev_path);
1106
1107         file = fopen(path, "rb");
1108         if (file == NULL)
1109                 return -1;
1110         while (fgets(line, sizeof(line), file) == line) {
1111                 size_t len = strlen(line);
1112                 int ret;
1113
1114                 /* Truncate long lines. */
1115                 if (len == (sizeof(line) - 1))
1116                         while (line[(len - 1)] != '\n') {
1117                                 ret = fgetc(file);
1118                                 if (ret == EOF)
1119                                         break;
1120                                 line[(len - 1)] = ret;
1121                         }
1122                 /* Extract information. */
1123                 if (sscanf(line,
1124                            "PCI_SLOT_NAME="
1125                            "%" SCNx32 ":%" SCNx8 ":%" SCNx8 ".%" SCNx8 "\n",
1126                            &pci_addr->domain,
1127                            &pci_addr->bus,
1128                            &pci_addr->devid,
1129                            &pci_addr->function) == 4) {
1130                         ret = 0;
1131                         break;
1132                 }
1133         }
1134         fclose(file);
1135         return 0;
1136 }
1137
1138 /**
1139  * Update the link status.
1140  *
1141  * @param priv
1142  *   Pointer to private structure.
1143  *
1144  * @return
1145  *   Zero if the callback process can be called immediately.
1146  */
1147 static int
1148 priv_link_status_update(struct priv *priv)
1149 {
1150         struct rte_eth_link *link = &priv->dev->data->dev_link;
1151
1152         mlx5_link_update(priv->dev, 0);
1153         if (((link->link_speed == 0) && link->link_status) ||
1154                 ((link->link_speed != 0) && !link->link_status)) {
1155                 /*
1156                  * Inconsistent status. Event likely occurred before the
1157                  * kernel netdevice exposes the new status.
1158                  */
1159                 if (!priv->pending_alarm) {
1160                         priv->pending_alarm = 1;
1161                         rte_eal_alarm_set(MLX5_ALARM_TIMEOUT_US,
1162                                           mlx5_dev_link_status_handler,
1163                                           priv->dev);
1164                 }
1165                 return 1;
1166         } else if (unlikely(priv->pending_alarm)) {
1167                 /* Link interrupt occurred while alarm is already scheduled. */
1168                 priv->pending_alarm = 0;
1169                 rte_eal_alarm_cancel(mlx5_dev_link_status_handler, priv->dev);
1170         }
1171         return 0;
1172 }
1173
1174 /**
1175  * Device status handler.
1176  *
1177  * @param priv
1178  *   Pointer to private structure.
1179  * @param events
1180  *   Pointer to event flags holder.
1181  *
1182  * @return
1183  *   Events bitmap of callback process which can be called immediately.
1184  */
1185 static uint32_t
1186 priv_dev_status_handler(struct priv *priv)
1187 {
1188         struct ibv_async_event event;
1189         uint32_t ret = 0;
1190
1191         /* Read all message and acknowledge them. */
1192         for (;;) {
1193                 if (ibv_get_async_event(priv->ctx, &event))
1194                         break;
1195                 if ((event.event_type == IBV_EVENT_PORT_ACTIVE ||
1196                         event.event_type == IBV_EVENT_PORT_ERR) &&
1197                         (priv->dev->data->dev_conf.intr_conf.lsc == 1))
1198                         ret |= (1 << RTE_ETH_EVENT_INTR_LSC);
1199                 else if (event.event_type == IBV_EVENT_DEVICE_FATAL &&
1200                         priv->dev->data->dev_conf.intr_conf.rmv == 1)
1201                         ret |= (1 << RTE_ETH_EVENT_INTR_RMV);
1202                 else
1203                         DEBUG("event type %d on port %d not handled",
1204                               event.event_type, event.element.port_num);
1205                 ibv_ack_async_event(&event);
1206         }
1207         if (ret & (1 << RTE_ETH_EVENT_INTR_LSC))
1208                 if (priv_link_status_update(priv))
1209                         ret &= ~(1 << RTE_ETH_EVENT_INTR_LSC);
1210         return ret;
1211 }
1212
1213 /**
1214  * Handle delayed link status event.
1215  *
1216  * @param arg
1217  *   Registered argument.
1218  */
1219 void
1220 mlx5_dev_link_status_handler(void *arg)
1221 {
1222         struct rte_eth_dev *dev = arg;
1223         struct priv *priv = dev->data->dev_private;
1224         int ret;
1225
1226         priv_lock(priv);
1227         assert(priv->pending_alarm == 1);
1228         priv->pending_alarm = 0;
1229         ret = priv_link_status_update(priv);
1230         priv_unlock(priv);
1231         if (!ret)
1232                 _rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC, NULL,
1233                                               NULL);
1234 }
1235
1236 /**
1237  * Handle interrupts from the NIC.
1238  *
1239  * @param[in] intr_handle
1240  *   Interrupt handler.
1241  * @param cb_arg
1242  *   Callback argument.
1243  */
1244 void
1245 mlx5_dev_interrupt_handler(void *cb_arg)
1246 {
1247         struct rte_eth_dev *dev = cb_arg;
1248         struct priv *priv = dev->data->dev_private;
1249         uint32_t events;
1250
1251         priv_lock(priv);
1252         events = priv_dev_status_handler(priv);
1253         priv_unlock(priv);
1254         if (events & (1 << RTE_ETH_EVENT_INTR_LSC))
1255                 _rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC, NULL,
1256                                               NULL);
1257         if (events & (1 << RTE_ETH_EVENT_INTR_RMV))
1258                 _rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_RMV, NULL,
1259                                               NULL);
1260 }
1261
1262 /**
1263  * Handle interrupts from the socket.
1264  *
1265  * @param cb_arg
1266  *   Callback argument.
1267  */
1268 static void
1269 mlx5_dev_handler_socket(void *cb_arg)
1270 {
1271         struct rte_eth_dev *dev = cb_arg;
1272         struct priv *priv = dev->data->dev_private;
1273
1274         priv_lock(priv);
1275         priv_socket_handle(priv);
1276         priv_unlock(priv);
1277 }
1278
1279 /**
1280  * Uninstall interrupt handler.
1281  *
1282  * @param priv
1283  *   Pointer to private structure.
1284  * @param dev
1285  *   Pointer to the rte_eth_dev structure.
1286  */
1287 void
1288 priv_dev_interrupt_handler_uninstall(struct priv *priv, struct rte_eth_dev *dev)
1289 {
1290         if (dev->data->dev_conf.intr_conf.lsc ||
1291             dev->data->dev_conf.intr_conf.rmv)
1292                 rte_intr_callback_unregister(&priv->intr_handle,
1293                                              mlx5_dev_interrupt_handler, dev);
1294         if (priv->primary_socket)
1295                 rte_intr_callback_unregister(&priv->intr_handle_socket,
1296                                              mlx5_dev_handler_socket, dev);
1297         if (priv->pending_alarm)
1298                 rte_eal_alarm_cancel(mlx5_dev_link_status_handler, dev);
1299         priv->pending_alarm = 0;
1300         priv->intr_handle.fd = 0;
1301         priv->intr_handle.type = RTE_INTR_HANDLE_UNKNOWN;
1302         priv->intr_handle_socket.fd = 0;
1303         priv->intr_handle_socket.type = RTE_INTR_HANDLE_UNKNOWN;
1304 }
1305
1306 /**
1307  * Install interrupt handler.
1308  *
1309  * @param priv
1310  *   Pointer to private structure.
1311  * @param dev
1312  *   Pointer to the rte_eth_dev structure.
1313  */
1314 void
1315 priv_dev_interrupt_handler_install(struct priv *priv, struct rte_eth_dev *dev)
1316 {
1317         int rc, flags;
1318
1319         assert(!mlx5_is_secondary());
1320         assert(priv->ctx->async_fd > 0);
1321         flags = fcntl(priv->ctx->async_fd, F_GETFL);
1322         rc = fcntl(priv->ctx->async_fd, F_SETFL, flags | O_NONBLOCK);
1323         if (rc < 0) {
1324                 INFO("failed to change file descriptor async event queue");
1325                 dev->data->dev_conf.intr_conf.lsc = 0;
1326                 dev->data->dev_conf.intr_conf.rmv = 0;
1327         }
1328         if (dev->data->dev_conf.intr_conf.lsc ||
1329             dev->data->dev_conf.intr_conf.rmv) {
1330                 priv->intr_handle.fd = priv->ctx->async_fd;
1331                 priv->intr_handle.type = RTE_INTR_HANDLE_EXT;
1332                 rte_intr_callback_register(&priv->intr_handle,
1333                                            mlx5_dev_interrupt_handler, dev);
1334         }
1335
1336         rc = priv_socket_init(priv);
1337         if (!rc && priv->primary_socket) {
1338                 priv->intr_handle_socket.fd = priv->primary_socket;
1339                 priv->intr_handle_socket.type = RTE_INTR_HANDLE_EXT;
1340                 rte_intr_callback_register(&priv->intr_handle_socket,
1341                                            mlx5_dev_handler_socket, dev);
1342         }
1343 }
1344
1345 /**
1346  * Change the link state (UP / DOWN).
1347  *
1348  * @param priv
1349  *   Pointer to private data structure.
1350  * @param dev
1351  *   Pointer to rte_eth_dev structure.
1352  * @param up
1353  *   Nonzero for link up, otherwise link down.
1354  *
1355  * @return
1356  *   0 on success, errno value on failure.
1357  */
1358 static int
1359 priv_dev_set_link(struct priv *priv, struct rte_eth_dev *dev, int up)
1360 {
1361         int err;
1362
1363         if (up) {
1364                 err = priv_set_flags(priv, ~IFF_UP, IFF_UP);
1365                 if (err)
1366                         return err;
1367                 priv_dev_select_tx_function(priv, dev);
1368                 priv_dev_select_rx_function(priv, dev);
1369         } else {
1370                 err = priv_set_flags(priv, ~IFF_UP, ~IFF_UP);
1371                 if (err)
1372                         return err;
1373                 dev->rx_pkt_burst = removed_rx_burst;
1374                 dev->tx_pkt_burst = removed_tx_burst;
1375         }
1376         return 0;
1377 }
1378
1379 /**
1380  * DPDK callback to bring the link DOWN.
1381  *
1382  * @param dev
1383  *   Pointer to Ethernet device structure.
1384  *
1385  * @return
1386  *   0 on success, errno value on failure.
1387  */
1388 int
1389 mlx5_set_link_down(struct rte_eth_dev *dev)
1390 {
1391         struct priv *priv = dev->data->dev_private;
1392         int err;
1393
1394         priv_lock(priv);
1395         err = priv_dev_set_link(priv, dev, 0);
1396         priv_unlock(priv);
1397         return err;
1398 }
1399
1400 /**
1401  * DPDK callback to bring the link UP.
1402  *
1403  * @param dev
1404  *   Pointer to Ethernet device structure.
1405  *
1406  * @return
1407  *   0 on success, errno value on failure.
1408  */
1409 int
1410 mlx5_set_link_up(struct rte_eth_dev *dev)
1411 {
1412         struct priv *priv = dev->data->dev_private;
1413         int err;
1414
1415         priv_lock(priv);
1416         err = priv_dev_set_link(priv, dev, 1);
1417         priv_unlock(priv);
1418         return err;
1419 }
1420
1421 /**
1422  * Configure the TX function to use.
1423  *
1424  * @param priv
1425  *   Pointer to private data structure.
1426  * @param dev
1427  *   Pointer to rte_eth_dev structure.
1428  */
1429 void
1430 priv_dev_select_tx_function(struct priv *priv, struct rte_eth_dev *dev)
1431 {
1432         assert(priv != NULL);
1433         assert(dev != NULL);
1434         dev->tx_pkt_burst = mlx5_tx_burst;
1435         /* Select appropriate TX function. */
1436         if (priv->mps == MLX5_MPW_ENHANCED) {
1437                 if (priv_check_vec_tx_support(priv) > 0) {
1438                         if (priv_check_raw_vec_tx_support(priv) > 0)
1439                                 dev->tx_pkt_burst = mlx5_tx_burst_raw_vec;
1440                         else
1441                                 dev->tx_pkt_burst = mlx5_tx_burst_vec;
1442                         DEBUG("selected Enhanced MPW TX vectorized function");
1443                 } else {
1444                         dev->tx_pkt_burst = mlx5_tx_burst_empw;
1445                         DEBUG("selected Enhanced MPW TX function");
1446                 }
1447         } else if (priv->mps && priv->txq_inline) {
1448                 dev->tx_pkt_burst = mlx5_tx_burst_mpw_inline;
1449                 DEBUG("selected MPW inline TX function");
1450         } else if (priv->mps) {
1451                 dev->tx_pkt_burst = mlx5_tx_burst_mpw;
1452                 DEBUG("selected MPW TX function");
1453         }
1454 }
1455
1456 /**
1457  * Configure the RX function to use.
1458  *
1459  * @param priv
1460  *   Pointer to private data structure.
1461  * @param dev
1462  *   Pointer to rte_eth_dev structure.
1463  */
1464 void
1465 priv_dev_select_rx_function(struct priv *priv, struct rte_eth_dev *dev)
1466 {
1467         assert(priv != NULL);
1468         assert(dev != NULL);
1469         if (priv_check_vec_rx_support(priv) > 0) {
1470                 dev->rx_pkt_burst = mlx5_rx_burst_vec;
1471                 DEBUG("selected RX vectorized function");
1472         } else {
1473                 dev->rx_pkt_burst = mlx5_rx_burst;
1474         }
1475 }