drivers/net/failsafe/failsafe_ops.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright 2017 6WIND S.A.
   5  *   Copyright 2017 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 #include <stdbool.h>
  35 #include <stdint.h>
  36
  37 #include <rte_debug.h>
  38 #include <rte_atomic.h>
  39 #include <rte_ethdev.h>
  40 #include <rte_malloc.h>
  41 #include <rte_flow.h>
  42 #include <rte_cycles.h>
  43
  44 #include "failsafe_private.h"
  45
  46 static struct rte_eth_dev_info default_infos = {
  47         /* Max possible number of elements */
  48         .max_rx_pktlen = UINT32_MAX,
  49         .max_rx_queues = RTE_MAX_QUEUES_PER_PORT,
  50         .max_tx_queues = RTE_MAX_QUEUES_PER_PORT,
  51         .max_mac_addrs = FAILSAFE_MAX_ETHADDR,
  52         .max_hash_mac_addrs = UINT32_MAX,
  53         .max_vfs = UINT16_MAX,
  54         .max_vmdq_pools = UINT16_MAX,
  55         .rx_desc_lim = {
  56                 .nb_max = UINT16_MAX,
  57                 .nb_min = 0,
  58                 .nb_align = 1,
  59                 .nb_seg_max = UINT16_MAX,
  60                 .nb_mtu_seg_max = UINT16_MAX,
  61         },
  62         .tx_desc_lim = {
  63                 .nb_max = UINT16_MAX,
  64                 .nb_min = 0,
  65                 .nb_align = 1,
  66                 .nb_seg_max = UINT16_MAX,
  67                 .nb_mtu_seg_max = UINT16_MAX,
  68         },
  69         /*
  70          * Set of capabilities that can be verified upon
  71          * configuring a sub-device.
  72          */
  73         .rx_offload_capa =
  74                 DEV_RX_OFFLOAD_VLAN_STRIP |
  75                 DEV_RX_OFFLOAD_IPV4_CKSUM |
  76                 DEV_RX_OFFLOAD_UDP_CKSUM |
  77                 DEV_RX_OFFLOAD_TCP_CKSUM |
  78                 DEV_RX_OFFLOAD_TCP_LRO |
  79                 DEV_RX_OFFLOAD_QINQ_STRIP |
  80                 DEV_RX_OFFLOAD_OUTER_IPV4_CKSUM |
  81                 DEV_RX_OFFLOAD_MACSEC_STRIP |
  82                 DEV_RX_OFFLOAD_HEADER_SPLIT |
  83                 DEV_RX_OFFLOAD_VLAN_FILTER |
  84                 DEV_RX_OFFLOAD_VLAN_EXTEND |
  85                 DEV_RX_OFFLOAD_JUMBO_FRAME |
  86                 DEV_RX_OFFLOAD_CRC_STRIP |
  87                 DEV_RX_OFFLOAD_SCATTER |
  88                 DEV_RX_OFFLOAD_TIMESTAMP |
  89                 DEV_RX_OFFLOAD_SECURITY,
  90         .rx_queue_offload_capa =
  91                 DEV_RX_OFFLOAD_VLAN_STRIP |
  92                 DEV_RX_OFFLOAD_IPV4_CKSUM |
  93                 DEV_RX_OFFLOAD_UDP_CKSUM |
  94                 DEV_RX_OFFLOAD_TCP_CKSUM |
  95                 DEV_RX_OFFLOAD_TCP_LRO |
  96                 DEV_RX_OFFLOAD_QINQ_STRIP |
  97                 DEV_RX_OFFLOAD_OUTER_IPV4_CKSUM |
  98                 DEV_RX_OFFLOAD_MACSEC_STRIP |
  99                 DEV_RX_OFFLOAD_HEADER_SPLIT |
 100                 DEV_RX_OFFLOAD_VLAN_FILTER |
 101                 DEV_RX_OFFLOAD_VLAN_EXTEND |
 102                 DEV_RX_OFFLOAD_JUMBO_FRAME |
 103                 DEV_RX_OFFLOAD_CRC_STRIP |
 104                 DEV_RX_OFFLOAD_SCATTER |
 105                 DEV_RX_OFFLOAD_TIMESTAMP |
 106                 DEV_RX_OFFLOAD_SECURITY,
 107         .tx_offload_capa = 0x0,
 108         .flow_type_rss_offloads = 0x0,
 109 };
 110
 111 static int
 112 fs_dev_configure(struct rte_eth_dev *dev)
 113 {
 114         struct sub_device *sdev;
 115         uint64_t supp_tx_offloads;
 116         uint64_t tx_offloads;
 117         uint8_t i;
 118         int ret;
 119
 120         supp_tx_offloads = PRIV(dev)->infos.tx_offload_capa;
 121         tx_offloads = dev->data->dev_conf.txmode.offloads;
 122         if ((tx_offloads & supp_tx_offloads) != tx_offloads) {
 123                 rte_errno = ENOTSUP;
 124                 ERROR("Some Tx offloads are not supported, "
 125                       "requested 0x%" PRIx64 " supported 0x%" PRIx64,
 126                       tx_offloads, supp_tx_offloads);
 127                 return -rte_errno;
 128         }
 129         FOREACH_SUBDEV(sdev, i, dev) {
 130                 int rmv_interrupt = 0;
 131                 int lsc_interrupt = 0;
 132                 int lsc_enabled;
 133
 134                 if (sdev->state != DEV_PROBED)
 135                         continue;
 136
 137                 rmv_interrupt = ETH(sdev)->data->dev_flags &
 138                                 RTE_ETH_DEV_INTR_RMV;
 139                 if (rmv_interrupt) {
 140                         DEBUG("Enabling RMV interrupts for sub_device %d", i);
 141                         dev->data->dev_conf.intr_conf.rmv = 1;
 142                 } else {
 143                         DEBUG("sub_device %d does not support RMV event", i);
 144                 }
 145                 lsc_enabled = dev->data->dev_conf.intr_conf.lsc;
 146                 lsc_interrupt = lsc_enabled &&
 147                                 (ETH(sdev)->data->dev_flags &
 148                                  RTE_ETH_DEV_INTR_LSC);
 149                 if (lsc_interrupt) {
 150                         DEBUG("Enabling LSC interrupts for sub_device %d", i);
 151                         dev->data->dev_conf.intr_conf.lsc = 1;
 152                 } else if (lsc_enabled && !lsc_interrupt) {
 153                         DEBUG("Disabling LSC interrupts for sub_device %d", i);
 154                         dev->data->dev_conf.intr_conf.lsc = 0;
 155                 }
 156                 DEBUG("Configuring sub-device %d", i);
 157                 sdev->remove = 0;
 158                 ret = rte_eth_dev_configure(PORT_ID(sdev),
 159                                         dev->data->nb_rx_queues,
 160                                         dev->data->nb_tx_queues,
 161                                         &dev->data->dev_conf);
 162                 if (ret) {
 163                         ERROR("Could not configure sub_device %d", i);
 164                         return ret;
 165                 }
 166                 if (rmv_interrupt) {
 167                         ret = rte_eth_dev_callback_register(PORT_ID(sdev),
 168                                         RTE_ETH_EVENT_INTR_RMV,
 169                                         failsafe_eth_rmv_event_callback,
 170                                         sdev);
 171                         if (ret)
 172                                 WARN("Failed to register RMV callback for sub_device %d",
 173                                      SUB_ID(sdev));
 174                 }
 175                 dev->data->dev_conf.intr_conf.rmv = 0;
 176                 if (lsc_interrupt) {
 177                         ret = rte_eth_dev_callback_register(PORT_ID(sdev),
 178                                                 RTE_ETH_EVENT_INTR_LSC,
 179                                                 failsafe_eth_lsc_event_callback,
 180                                                 dev);
 181                         if (ret)
 182                                 WARN("Failed to register LSC callback for sub_device %d",
 183                                      SUB_ID(sdev));
 184                 }
 185                 dev->data->dev_conf.intr_conf.lsc = lsc_enabled;
 186                 sdev->state = DEV_ACTIVE;
 187         }
 188         if (PRIV(dev)->state < DEV_ACTIVE)
 189                 PRIV(dev)->state = DEV_ACTIVE;
 190         return 0;
 191 }
 192
 193 static int
 194 fs_dev_start(struct rte_eth_dev *dev)
 195 {
 196         struct sub_device *sdev;
 197         uint8_t i;
 198         int ret;
 199
 200         FOREACH_SUBDEV(sdev, i, dev) {
 201                 if (sdev->state != DEV_ACTIVE)
 202                         continue;
 203                 DEBUG("Starting sub_device %d", i);
 204                 ret = rte_eth_dev_start(PORT_ID(sdev));
 205                 if (ret)
 206                         return ret;
 207                 sdev->state = DEV_STARTED;
 208         }
 209         if (PRIV(dev)->state < DEV_STARTED)
 210                 PRIV(dev)->state = DEV_STARTED;
 211         fs_switch_dev(dev, NULL);
 212         return 0;
 213 }
 214
 215 static void
 216 fs_dev_stop(struct rte_eth_dev *dev)
 217 {
 218         struct sub_device *sdev;
 219         uint8_t i;
 220
 221         PRIV(dev)->state = DEV_STARTED - 1;
 222         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_STARTED) {
 223                 rte_eth_dev_stop(PORT_ID(sdev));
 224                 sdev->state = DEV_STARTED - 1;
 225         }
 226 }
 227
 228 static int
 229 fs_dev_set_link_up(struct rte_eth_dev *dev)
 230 {
 231         struct sub_device *sdev;
 232         uint8_t i;
 233         int ret;
 234
 235         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 236                 DEBUG("Calling rte_eth_dev_set_link_up on sub_device %d", i);
 237                 ret = rte_eth_dev_set_link_up(PORT_ID(sdev));
 238                 if (ret) {
 239                         ERROR("Operation rte_eth_dev_set_link_up failed for sub_device %d"
 240                               " with error %d", i, ret);
 241                         return ret;
 242                 }
 243         }
 244         return 0;
 245 }
 246
 247 static int
 248 fs_dev_set_link_down(struct rte_eth_dev *dev)
 249 {
 250         struct sub_device *sdev;
 251         uint8_t i;
 252         int ret;
 253
 254         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 255                 DEBUG("Calling rte_eth_dev_set_link_down on sub_device %d", i);
 256                 ret = rte_eth_dev_set_link_down(PORT_ID(sdev));
 257                 if (ret) {
 258                         ERROR("Operation rte_eth_dev_set_link_down failed for sub_device %d"
 259                               " with error %d", i, ret);
 260                         return ret;
 261                 }
 262         }
 263         return 0;
 264 }
 265
 266 static void fs_dev_free_queues(struct rte_eth_dev *dev);
 267 static void
 268 fs_dev_close(struct rte_eth_dev *dev)
 269 {
 270         struct sub_device *sdev;
 271         uint8_t i;
 272
 273         failsafe_hotplug_alarm_cancel(dev);
 274         if (PRIV(dev)->state == DEV_STARTED)
 275                 dev->dev_ops->dev_stop(dev);
 276         PRIV(dev)->state = DEV_ACTIVE - 1;
 277         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 278                 DEBUG("Closing sub_device %d", i);
 279                 rte_eth_dev_close(PORT_ID(sdev));
 280                 sdev->state = DEV_ACTIVE - 1;
 281         }
 282         fs_dev_free_queues(dev);
 283 }
 284
 285 static bool
 286 fs_rxq_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads)
 287 {
 288         uint64_t port_offloads;
 289         uint64_t queue_supp_offloads;
 290         uint64_t port_supp_offloads;
 291
 292         port_offloads = dev->data->dev_conf.rxmode.offloads;
 293         queue_supp_offloads = PRIV(dev)->infos.rx_queue_offload_capa;
 294         port_supp_offloads = PRIV(dev)->infos.rx_offload_capa;
 295         if ((offloads & (queue_supp_offloads | port_supp_offloads)) !=
 296              offloads)
 297                 return false;
 298         /* Verify we have no conflict with port offloads */
 299         if ((port_offloads ^ offloads) & port_supp_offloads)
 300                 return false;
 301         return true;
 302 }
 303
 304 static void
 305 fs_rx_queue_release(void *queue)
 306 {
 307         struct rte_eth_dev *dev;
 308         struct sub_device *sdev;
 309         uint8_t i;
 310         struct rxq *rxq;
 311
 312         if (queue == NULL)
 313                 return;
 314         rxq = queue;
 315         dev = rxq->priv->dev;
 316         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
 317                 SUBOPS(sdev, rx_queue_release)
 318                         (ETH(sdev)->data->rx_queues[rxq->qid]);
 319         dev->data->rx_queues[rxq->qid] = NULL;
 320         rte_free(rxq);
 321 }
 322
 323 static int
 324 fs_rx_queue_setup(struct rte_eth_dev *dev,
 325                 uint16_t rx_queue_id,
 326                 uint16_t nb_rx_desc,
 327                 unsigned int socket_id,
 328                 const struct rte_eth_rxconf *rx_conf,
 329                 struct rte_mempool *mb_pool)
 330 {
 331         struct sub_device *sdev;
 332         struct rxq *rxq;
 333         uint8_t i;
 334         int ret;
 335
 336         rxq = dev->data->rx_queues[rx_queue_id];
 337         if (rxq != NULL) {
 338                 fs_rx_queue_release(rxq);
 339                 dev->data->rx_queues[rx_queue_id] = NULL;
 340         }
 341         /* Verify application offloads are valid for our port and queue. */
 342         if (fs_rxq_offloads_valid(dev, rx_conf->offloads) == false) {
 343                 rte_errno = ENOTSUP;
 344                 ERROR("Rx queue offloads 0x%" PRIx64
 345                       " don't match port offloads 0x%" PRIx64
 346                       " or supported offloads 0x%" PRIx64,
 347                       rx_conf->offloads,
 348                       dev->data->dev_conf.rxmode.offloads,
 349                       PRIV(dev)->infos.rx_offload_capa |
 350                       PRIV(dev)->infos.rx_queue_offload_capa);
 351                 return -rte_errno;
 352         }
 353         rxq = rte_zmalloc(NULL,
 354                           sizeof(*rxq) +
 355                           sizeof(rte_atomic64_t) * PRIV(dev)->subs_tail,
 356                           RTE_CACHE_LINE_SIZE);
 357         if (rxq == NULL)
 358                 return -ENOMEM;
 359         FOREACH_SUBDEV(sdev, i, dev)
 360                 rte_atomic64_init(&rxq->refcnt[i]);
 361         rxq->qid = rx_queue_id;
 362         rxq->socket_id = socket_id;
 363         rxq->info.mp = mb_pool;
 364         rxq->info.conf = *rx_conf;
 365         rxq->info.nb_desc = nb_rx_desc;
 366         rxq->priv = PRIV(dev);
 367         rxq->sdev = PRIV(dev)->subs;
 368         dev->data->rx_queues[rx_queue_id] = rxq;
 369         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 370                 ret = rte_eth_rx_queue_setup(PORT_ID(sdev),
 371                                 rx_queue_id,
 372                                 nb_rx_desc, socket_id,
 373                                 rx_conf, mb_pool);
 374                 if (ret) {
 375                         ERROR("RX queue setup failed for sub_device %d", i);
 376                         goto free_rxq;
 377                 }
 378         }
 379         return 0;
 380 free_rxq:
 381         fs_rx_queue_release(rxq);
 382         return ret;
 383 }
 384
 385 static bool
 386 fs_txq_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads)
 387 {
 388         uint64_t port_offloads;
 389         uint64_t queue_supp_offloads;
 390         uint64_t port_supp_offloads;
 391
 392         port_offloads = dev->data->dev_conf.txmode.offloads;
 393         queue_supp_offloads = PRIV(dev)->infos.tx_queue_offload_capa;
 394         port_supp_offloads = PRIV(dev)->infos.tx_offload_capa;
 395         if ((offloads & (queue_supp_offloads | port_supp_offloads)) !=
 396              offloads)
 397                 return false;
 398         /* Verify we have no conflict with port offloads */
 399         if ((port_offloads ^ offloads) & port_supp_offloads)
 400                 return false;
 401         return true;
 402 }
 403
 404 static void
 405 fs_tx_queue_release(void *queue)
 406 {
 407         struct rte_eth_dev *dev;
 408         struct sub_device *sdev;
 409         uint8_t i;
 410         struct txq *txq;
 411
 412         if (queue == NULL)
 413                 return;
 414         txq = queue;
 415         dev = txq->priv->dev;
 416         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
 417                 SUBOPS(sdev, tx_queue_release)
 418                         (ETH(sdev)->data->tx_queues[txq->qid]);
 419         dev->data->tx_queues[txq->qid] = NULL;
 420         rte_free(txq);
 421 }
 422
 423 static int
 424 fs_tx_queue_setup(struct rte_eth_dev *dev,
 425                 uint16_t tx_queue_id,
 426                 uint16_t nb_tx_desc,
 427                 unsigned int socket_id,
 428                 const struct rte_eth_txconf *tx_conf)
 429 {
 430         struct sub_device *sdev;
 431         struct txq *txq;
 432         uint8_t i;
 433         int ret;
 434
 435         txq = dev->data->tx_queues[tx_queue_id];
 436         if (txq != NULL) {
 437                 fs_tx_queue_release(txq);
 438                 dev->data->tx_queues[tx_queue_id] = NULL;
 439         }
 440         /*
 441          * Don't verify queue offloads for applications which
 442          * use the old API.
 443          */
 444         if (tx_conf != NULL &&
 445             (tx_conf->txq_flags & ETH_TXQ_FLAGS_IGNORE) &&
 446             fs_txq_offloads_valid(dev, tx_conf->offloads) == false) {
 447                 rte_errno = ENOTSUP;
 448                 ERROR("Tx queue offloads 0x%" PRIx64
 449                       " don't match port offloads 0x%" PRIx64
 450                       " or supported offloads 0x%" PRIx64,
 451                       tx_conf->offloads,
 452                       dev->data->dev_conf.txmode.offloads,
 453                       PRIV(dev)->infos.tx_offload_capa |
 454                       PRIV(dev)->infos.tx_queue_offload_capa);
 455                 return -rte_errno;
 456         }
 457         txq = rte_zmalloc("ethdev TX queue",
 458                           sizeof(*txq) +
 459                           sizeof(rte_atomic64_t) * PRIV(dev)->subs_tail,
 460                           RTE_CACHE_LINE_SIZE);
 461         if (txq == NULL)
 462                 return -ENOMEM;
 463         FOREACH_SUBDEV(sdev, i, dev)
 464                 rte_atomic64_init(&txq->refcnt[i]);
 465         txq->qid = tx_queue_id;
 466         txq->socket_id = socket_id;
 467         txq->info.conf = *tx_conf;
 468         txq->info.nb_desc = nb_tx_desc;
 469         txq->priv = PRIV(dev);
 470         dev->data->tx_queues[tx_queue_id] = txq;
 471         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 472                 ret = rte_eth_tx_queue_setup(PORT_ID(sdev),
 473                                 tx_queue_id,
 474                                 nb_tx_desc, socket_id,
 475                                 tx_conf);
 476                 if (ret) {
 477                         ERROR("TX queue setup failed for sub_device %d", i);
 478                         goto free_txq;
 479                 }
 480         }
 481         return 0;
 482 free_txq:
 483         fs_tx_queue_release(txq);
 484         return ret;
 485 }
 486
 487 static void
 488 fs_dev_free_queues(struct rte_eth_dev *dev)
 489 {
 490         uint16_t i;
 491
 492         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 493                 fs_rx_queue_release(dev->data->rx_queues[i]);
 494                 dev->data->rx_queues[i] = NULL;
 495         }
 496         dev->data->nb_rx_queues = 0;
 497         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 498                 fs_tx_queue_release(dev->data->tx_queues[i]);
 499                 dev->data->tx_queues[i] = NULL;
 500         }
 501         dev->data->nb_tx_queues = 0;
 502 }
 503
 504 static void
 505 fs_promiscuous_enable(struct rte_eth_dev *dev)
 506 {
 507         struct sub_device *sdev;
 508         uint8_t i;
 509
 510         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
 511                 rte_eth_promiscuous_enable(PORT_ID(sdev));
 512 }
 513
 514 static void
 515 fs_promiscuous_disable(struct rte_eth_dev *dev)
 516 {
 517         struct sub_device *sdev;
 518         uint8_t i;
 519
 520         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
 521                 rte_eth_promiscuous_disable(PORT_ID(sdev));
 522 }
 523
 524 static void
 525 fs_allmulticast_enable(struct rte_eth_dev *dev)
 526 {
 527         struct sub_device *sdev;
 528         uint8_t i;
 529
 530         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
 531                 rte_eth_allmulticast_enable(PORT_ID(sdev));
 532 }
 533
 534 static void
 535 fs_allmulticast_disable(struct rte_eth_dev *dev)
 536 {
 537         struct sub_device *sdev;
 538         uint8_t i;
 539
 540         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
 541                 rte_eth_allmulticast_disable(PORT_ID(sdev));
 542 }
 543
 544 static int
 545 fs_link_update(struct rte_eth_dev *dev,
 546                 int wait_to_complete)
 547 {
 548         struct sub_device *sdev;
 549         uint8_t i;
 550         int ret;
 551
 552         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 553                 DEBUG("Calling link_update on sub_device %d", i);
 554                 ret = (SUBOPS(sdev, link_update))(ETH(sdev), wait_to_complete);
 555                 if (ret && ret != -1) {
 556                         ERROR("Link update failed for sub_device %d with error %d",
 557                               i, ret);
 558                         return ret;
 559                 }
 560         }
 561         if (TX_SUBDEV(dev)) {
 562                 struct rte_eth_link *l1;
 563                 struct rte_eth_link *l2;
 564
 565                 l1 = &dev->data->dev_link;
 566                 l2 = &ETH(TX_SUBDEV(dev))->data->dev_link;
 567                 if (memcmp(l1, l2, sizeof(*l1))) {
 568                         *l1 = *l2;
 569                         return 0;
 570                 }
 571         }
 572         return -1;
 573 }
 574
 575 static int
 576 fs_stats_get(struct rte_eth_dev *dev,
 577              struct rte_eth_stats *stats)
 578 {
 579         struct sub_device *sdev;
 580         uint8_t i;
 581         int ret;
 582
 583         rte_memcpy(stats, &PRIV(dev)->stats_accumulator, sizeof(*stats));
 584         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 585                 struct rte_eth_stats *snapshot = &sdev->stats_snapshot.stats;
 586                 uint64_t *timestamp = &sdev->stats_snapshot.timestamp;
 587
 588                 ret = rte_eth_stats_get(PORT_ID(sdev), snapshot);
 589                 if (ret) {
 590                         ERROR("Operation rte_eth_stats_get failed for sub_device %d with error %d",
 591                                   i, ret);
 592                         *timestamp = 0;
 593                         return ret;
 594                 }
 595                 *timestamp = rte_rdtsc();
 596                 failsafe_stats_increment(stats, snapshot);
 597         }
 598         return 0;
 599 }
 600
 601 static void
 602 fs_stats_reset(struct rte_eth_dev *dev)
 603 {
 604         struct sub_device *sdev;
 605         uint8_t i;
 606
 607         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 608                 rte_eth_stats_reset(PORT_ID(sdev));
 609                 memset(&sdev->stats_snapshot, 0, sizeof(struct rte_eth_stats));
 610         }
 611         memset(&PRIV(dev)->stats_accumulator, 0, sizeof(struct rte_eth_stats));
 612 }
 613
 614 /**
 615  * Fail-safe dev_infos_get rules:
 616  *
 617  * No sub_device:
 618  *   Numerables:
 619  *      Use the maximum possible values for any field, so as not
 620  *      to impede any further configuration effort.
 621  *   Capabilities:
 622  *      Limits capabilities to those that are understood by the
 623  *      fail-safe PMD. This understanding stems from the fail-safe
 624  *      being capable of verifying that the related capability is
 625  *      expressed within the device configuration (struct rte_eth_conf).
 626  *
 627  * At least one probed sub_device:
 628  *   Numerables:
 629  *      Uses values from the active probed sub_device
 630  *      The rationale here is that if any sub_device is less capable
 631  *      (for example concerning the number of queues) than the active
 632  *      sub_device, then its subsequent configuration will fail.
 633  *      It is impossible to foresee this failure when the failing sub_device
 634  *      is supposed to be plugged-in later on, so the configuration process
 635  *      is the single point of failure and error reporting.
 636  *   Capabilities:
 637  *      Uses a logical AND of RX capabilities among
 638  *      all sub_devices and the default capabilities.
 639  *      Uses a logical AND of TX capabilities among
 640  *      the active probed sub_device and the default capabilities.
 641  *
 642  */
 643 static void
 644 fs_dev_infos_get(struct rte_eth_dev *dev,
 645                   struct rte_eth_dev_info *infos)
 646 {
 647         struct sub_device *sdev;
 648         uint8_t i;
 649
 650         sdev = TX_SUBDEV(dev);
 651         if (sdev == NULL) {
 652                 DEBUG("No probed device, using default infos");
 653                 rte_memcpy(&PRIV(dev)->infos, &default_infos,
 654                            sizeof(default_infos));
 655         } else {
 656                 uint64_t rx_offload_capa;
 657                 uint64_t rxq_offload_capa;
 658
 659                 rx_offload_capa = default_infos.rx_offload_capa;
 660                 rxq_offload_capa = default_infos.rx_queue_offload_capa;
 661                 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_PROBED) {
 662                         rte_eth_dev_info_get(PORT_ID(sdev),
 663                                         &PRIV(dev)->infos);
 664                         rx_offload_capa &= PRIV(dev)->infos.rx_offload_capa;
 665                         rxq_offload_capa &=
 666                                         PRIV(dev)->infos.rx_queue_offload_capa;
 667                 }
 668                 sdev = TX_SUBDEV(dev);
 669                 rte_eth_dev_info_get(PORT_ID(sdev), &PRIV(dev)->infos);
 670                 PRIV(dev)->infos.rx_offload_capa = rx_offload_capa;
 671                 PRIV(dev)->infos.rx_queue_offload_capa = rxq_offload_capa;
 672                 PRIV(dev)->infos.tx_offload_capa &=
 673                                         default_infos.tx_offload_capa;
 674                 PRIV(dev)->infos.tx_queue_offload_capa &=
 675                                         default_infos.tx_queue_offload_capa;
 676                 PRIV(dev)->infos.flow_type_rss_offloads &=
 677                                         default_infos.flow_type_rss_offloads;
 678         }
 679         rte_memcpy(infos, &PRIV(dev)->infos, sizeof(*infos));
 680 }
 681
 682 static const uint32_t *
 683 fs_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 684 {
 685         struct sub_device *sdev;
 686         struct rte_eth_dev *edev;
 687
 688         sdev = TX_SUBDEV(dev);
 689         if (sdev == NULL)
 690                 return NULL;
 691         edev = ETH(sdev);
 692         /* ENOTSUP: counts as no supported ptypes */
 693         if (SUBOPS(sdev, dev_supported_ptypes_get) == NULL)
 694                 return NULL;
 695         /*
 696          * The API does not permit to do a clean AND of all ptypes,
 697          * It is also incomplete by design and we do not really care
 698          * to have a best possible value in this context.
 699          * We just return the ptypes of the device of highest
 700          * priority, usually the PREFERRED device.
 701          */
 702         return SUBOPS(sdev, dev_supported_ptypes_get)(edev);
 703 }
 704
 705 static int
 706 fs_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 707 {
 708         struct sub_device *sdev;
 709         uint8_t i;
 710         int ret;
 711
 712         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 713                 DEBUG("Calling rte_eth_dev_set_mtu on sub_device %d", i);
 714                 ret = rte_eth_dev_set_mtu(PORT_ID(sdev), mtu);
 715                 if (ret) {
 716                         ERROR("Operation rte_eth_dev_set_mtu failed for sub_device %d with error %d",
 717                               i, ret);
 718                         return ret;
 719                 }
 720         }
 721         return 0;
 722 }
 723
 724 static int
 725 fs_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
 726 {
 727         struct sub_device *sdev;
 728         uint8_t i;
 729         int ret;
 730
 731         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 732                 DEBUG("Calling rte_eth_dev_vlan_filter on sub_device %d", i);
 733                 ret = rte_eth_dev_vlan_filter(PORT_ID(sdev), vlan_id, on);
 734                 if (ret) {
 735                         ERROR("Operation rte_eth_dev_vlan_filter failed for sub_device %d"
 736                               " with error %d", i, ret);
 737                         return ret;
 738                 }
 739         }
 740         return 0;
 741 }
 742
 743 static int
 744 fs_flow_ctrl_get(struct rte_eth_dev *dev,
 745                 struct rte_eth_fc_conf *fc_conf)
 746 {
 747         struct sub_device *sdev;
 748
 749         sdev = TX_SUBDEV(dev);
 750         if (sdev == NULL)
 751                 return 0;
 752         if (SUBOPS(sdev, flow_ctrl_get) == NULL)
 753                 return -ENOTSUP;
 754         return SUBOPS(sdev, flow_ctrl_get)(ETH(sdev), fc_conf);
 755 }
 756
 757 static int
 758 fs_flow_ctrl_set(struct rte_eth_dev *dev,
 759                 struct rte_eth_fc_conf *fc_conf)
 760 {
 761         struct sub_device *sdev;
 762         uint8_t i;
 763         int ret;
 764
 765         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 766                 DEBUG("Calling rte_eth_dev_flow_ctrl_set on sub_device %d", i);
 767                 ret = rte_eth_dev_flow_ctrl_set(PORT_ID(sdev), fc_conf);
 768                 if (ret) {
 769                         ERROR("Operation rte_eth_dev_flow_ctrl_set failed for sub_device %d"
 770                               " with error %d", i, ret);
 771                         return ret;
 772                 }
 773         }
 774         return 0;
 775 }
 776
 777 static void
 778 fs_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
 779 {
 780         struct sub_device *sdev;
 781         uint8_t i;
 782
 783         /* No check: already done within the rte_eth_dev_mac_addr_remove
 784          * call for the fail-safe device.
 785          */
 786         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
 787                 rte_eth_dev_mac_addr_remove(PORT_ID(sdev),
 788                                 &dev->data->mac_addrs[index]);
 789         PRIV(dev)->mac_addr_pool[index] = 0;
 790 }
 791
 792 static int
 793 fs_mac_addr_add(struct rte_eth_dev *dev,
 794                 struct ether_addr *mac_addr,
 795                 uint32_t index,
 796                 uint32_t vmdq)
 797 {
 798         struct sub_device *sdev;
 799         int ret;
 800         uint8_t i;
 801
 802         RTE_ASSERT(index < FAILSAFE_MAX_ETHADDR);
 803         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 804                 ret = rte_eth_dev_mac_addr_add(PORT_ID(sdev), mac_addr, vmdq);
 805                 if (ret) {
 806                         ERROR("Operation rte_eth_dev_mac_addr_add failed for sub_device %"
 807                               PRIu8 " with error %d", i, ret);
 808                         return ret;
 809                 }
 810         }
 811         if (index >= PRIV(dev)->nb_mac_addr) {
 812                 DEBUG("Growing mac_addrs array");
 813                 PRIV(dev)->nb_mac_addr = index;
 814         }
 815         PRIV(dev)->mac_addr_pool[index] = vmdq;
 816         return 0;
 817 }
 818
 819 static void
 820 fs_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 821 {
 822         struct sub_device *sdev;
 823         uint8_t i;
 824
 825         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
 826                 rte_eth_dev_default_mac_addr_set(PORT_ID(sdev), mac_addr);
 827 }
 828
 829 static int
 830 fs_filter_ctrl(struct rte_eth_dev *dev,
 831                 enum rte_filter_type type,
 832                 enum rte_filter_op op,
 833                 void *arg)
 834 {
 835         struct sub_device *sdev;
 836         uint8_t i;
 837         int ret;
 838
 839         if (type == RTE_ETH_FILTER_GENERIC &&
 840             op == RTE_ETH_FILTER_GET) {
 841                 *(const void **)arg = &fs_flow_ops;
 842                 return 0;
 843         }
 844         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 845                 DEBUG("Calling rte_eth_dev_filter_ctrl on sub_device %d", i);
 846                 ret = rte_eth_dev_filter_ctrl(PORT_ID(sdev), type, op, arg);
 847                 if (ret) {
 848                         ERROR("Operation rte_eth_dev_filter_ctrl failed for sub_device %d"
 849                               " with error %d", i, ret);
 850                         return ret;
 851                 }
 852         }
 853         return 0;
 854 }
 855
 856 const struct eth_dev_ops failsafe_ops = {
 857         .dev_configure = fs_dev_configure,
 858         .dev_start = fs_dev_start,
 859         .dev_stop = fs_dev_stop,
 860         .dev_set_link_down = fs_dev_set_link_down,
 861         .dev_set_link_up = fs_dev_set_link_up,
 862         .dev_close = fs_dev_close,
 863         .promiscuous_enable = fs_promiscuous_enable,
 864         .promiscuous_disable = fs_promiscuous_disable,
 865         .allmulticast_enable = fs_allmulticast_enable,
 866         .allmulticast_disable = fs_allmulticast_disable,
 867         .link_update = fs_link_update,
 868         .stats_get = fs_stats_get,
 869         .stats_reset = fs_stats_reset,
 870         .dev_infos_get = fs_dev_infos_get,
 871         .dev_supported_ptypes_get = fs_dev_supported_ptypes_get,
 872         .mtu_set = fs_mtu_set,
 873         .vlan_filter_set = fs_vlan_filter_set,
 874         .rx_queue_setup = fs_rx_queue_setup,
 875         .tx_queue_setup = fs_tx_queue_setup,
 876         .rx_queue_release = fs_rx_queue_release,
 877         .tx_queue_release = fs_tx_queue_release,
 878         .flow_ctrl_get = fs_flow_ctrl_get,
 879         .flow_ctrl_set = fs_flow_ctrl_set,
 880         .mac_addr_remove = fs_mac_addr_remove,
 881         .mac_addr_add = fs_mac_addr_add,
 882         .mac_addr_set = fs_mac_addr_set,
 883         .filter_ctrl = fs_filter_ctrl,
 884 };