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 #include <unistd.h>
  37
  38 #include <rte_debug.h>
  39 #include <rte_atomic.h>
  40 #include <rte_ethdev_driver.h>
  41 #include <rte_malloc.h>
  42 #include <rte_flow.h>
  43 #include <rte_cycles.h>
  44
  45 #include "failsafe_private.h"
  46
  47 static struct rte_eth_dev_info default_infos = {
  48         /* Max possible number of elements */
  49         .max_rx_pktlen = UINT32_MAX,
  50         .max_rx_queues = RTE_MAX_QUEUES_PER_PORT,
  51         .max_tx_queues = RTE_MAX_QUEUES_PER_PORT,
  52         .max_mac_addrs = FAILSAFE_MAX_ETHADDR,
  53         .max_hash_mac_addrs = UINT32_MAX,
  54         .max_vfs = UINT16_MAX,
  55         .max_vmdq_pools = UINT16_MAX,
  56         .rx_desc_lim = {
  57                 .nb_max = UINT16_MAX,
  58                 .nb_min = 0,
  59                 .nb_align = 1,
  60                 .nb_seg_max = UINT16_MAX,
  61                 .nb_mtu_seg_max = UINT16_MAX,
  62         },
  63         .tx_desc_lim = {
  64                 .nb_max = UINT16_MAX,
  65                 .nb_min = 0,
  66                 .nb_align = 1,
  67                 .nb_seg_max = UINT16_MAX,
  68                 .nb_mtu_seg_max = UINT16_MAX,
  69         },
  70         /*
  71          * Set of capabilities that can be verified upon
  72          * configuring a sub-device.
  73          */
  74         .rx_offload_capa =
  75                 DEV_RX_OFFLOAD_VLAN_STRIP |
  76                 DEV_RX_OFFLOAD_IPV4_CKSUM |
  77                 DEV_RX_OFFLOAD_UDP_CKSUM |
  78                 DEV_RX_OFFLOAD_TCP_CKSUM |
  79                 DEV_RX_OFFLOAD_TCP_LRO |
  80                 DEV_RX_OFFLOAD_QINQ_STRIP |
  81                 DEV_RX_OFFLOAD_OUTER_IPV4_CKSUM |
  82                 DEV_RX_OFFLOAD_MACSEC_STRIP |
  83                 DEV_RX_OFFLOAD_HEADER_SPLIT |
  84                 DEV_RX_OFFLOAD_VLAN_FILTER |
  85                 DEV_RX_OFFLOAD_VLAN_EXTEND |
  86                 DEV_RX_OFFLOAD_JUMBO_FRAME |
  87                 DEV_RX_OFFLOAD_CRC_STRIP |
  88                 DEV_RX_OFFLOAD_SCATTER |
  89                 DEV_RX_OFFLOAD_TIMESTAMP |
  90                 DEV_RX_OFFLOAD_SECURITY,
  91         .rx_queue_offload_capa =
  92                 DEV_RX_OFFLOAD_VLAN_STRIP |
  93                 DEV_RX_OFFLOAD_IPV4_CKSUM |
  94                 DEV_RX_OFFLOAD_UDP_CKSUM |
  95                 DEV_RX_OFFLOAD_TCP_CKSUM |
  96                 DEV_RX_OFFLOAD_TCP_LRO |
  97                 DEV_RX_OFFLOAD_QINQ_STRIP |
  98                 DEV_RX_OFFLOAD_OUTER_IPV4_CKSUM |
  99                 DEV_RX_OFFLOAD_MACSEC_STRIP |
 100                 DEV_RX_OFFLOAD_HEADER_SPLIT |
 101                 DEV_RX_OFFLOAD_VLAN_FILTER |
 102                 DEV_RX_OFFLOAD_VLAN_EXTEND |
 103                 DEV_RX_OFFLOAD_JUMBO_FRAME |
 104                 DEV_RX_OFFLOAD_CRC_STRIP |
 105                 DEV_RX_OFFLOAD_SCATTER |
 106                 DEV_RX_OFFLOAD_TIMESTAMP |
 107                 DEV_RX_OFFLOAD_SECURITY,
 108         .tx_offload_capa = 0x0,
 109         .flow_type_rss_offloads = 0x0,
 110 };
 111
 112 static int
 113 fs_dev_configure(struct rte_eth_dev *dev)
 114 {
 115         struct sub_device *sdev;
 116         uint64_t supp_tx_offloads;
 117         uint64_t tx_offloads;
 118         uint8_t i;
 119         int ret;
 120
 121         supp_tx_offloads = PRIV(dev)->infos.tx_offload_capa;
 122         tx_offloads = dev->data->dev_conf.txmode.offloads;
 123         if ((tx_offloads & supp_tx_offloads) != tx_offloads) {
 124                 rte_errno = ENOTSUP;
 125                 ERROR("Some Tx offloads are not supported, "
 126                       "requested 0x%" PRIx64 " supported 0x%" PRIx64,
 127                       tx_offloads, supp_tx_offloads);
 128                 return -rte_errno;
 129         }
 130         FOREACH_SUBDEV(sdev, i, dev) {
 131                 int rmv_interrupt = 0;
 132                 int lsc_interrupt = 0;
 133                 int lsc_enabled;
 134
 135                 if (sdev->state != DEV_PROBED)
 136                         continue;
 137
 138                 rmv_interrupt = ETH(sdev)->data->dev_flags &
 139                                 RTE_ETH_DEV_INTR_RMV;
 140                 if (rmv_interrupt) {
 141                         DEBUG("Enabling RMV interrupts for sub_device %d", i);
 142                         dev->data->dev_conf.intr_conf.rmv = 1;
 143                 } else {
 144                         DEBUG("sub_device %d does not support RMV event", i);
 145                 }
 146                 lsc_enabled = dev->data->dev_conf.intr_conf.lsc;
 147                 lsc_interrupt = lsc_enabled &&
 148                                 (ETH(sdev)->data->dev_flags &
 149                                  RTE_ETH_DEV_INTR_LSC);
 150                 if (lsc_interrupt) {
 151                         DEBUG("Enabling LSC interrupts for sub_device %d", i);
 152                         dev->data->dev_conf.intr_conf.lsc = 1;
 153                 } else if (lsc_enabled && !lsc_interrupt) {
 154                         DEBUG("Disabling LSC interrupts for sub_device %d", i);
 155                         dev->data->dev_conf.intr_conf.lsc = 0;
 156                 }
 157                 DEBUG("Configuring sub-device %d", i);
 158                 sdev->remove = 0;
 159                 ret = rte_eth_dev_configure(PORT_ID(sdev),
 160                                         dev->data->nb_rx_queues,
 161                                         dev->data->nb_tx_queues,
 162                                         &dev->data->dev_conf);
 163                 if (ret) {
 164                         if (!fs_err(sdev, ret))
 165                                 continue;
 166                         ERROR("Could not configure sub_device %d", i);
 167                         return ret;
 168                 }
 169                 if (rmv_interrupt) {
 170                         ret = rte_eth_dev_callback_register(PORT_ID(sdev),
 171                                         RTE_ETH_EVENT_INTR_RMV,
 172                                         failsafe_eth_rmv_event_callback,
 173                                         sdev);
 174                         if (ret)
 175                                 WARN("Failed to register RMV callback for sub_device %d",
 176                                      SUB_ID(sdev));
 177                 }
 178                 dev->data->dev_conf.intr_conf.rmv = 0;
 179                 if (lsc_interrupt) {
 180                         ret = rte_eth_dev_callback_register(PORT_ID(sdev),
 181                                                 RTE_ETH_EVENT_INTR_LSC,
 182                                                 failsafe_eth_lsc_event_callback,
 183                                                 dev);
 184                         if (ret)
 185                                 WARN("Failed to register LSC callback for sub_device %d",
 186                                      SUB_ID(sdev));
 187                 }
 188                 dev->data->dev_conf.intr_conf.lsc = lsc_enabled;
 189                 sdev->state = DEV_ACTIVE;
 190         }
 191         if (PRIV(dev)->state < DEV_ACTIVE)
 192                 PRIV(dev)->state = DEV_ACTIVE;
 193         return 0;
 194 }
 195
 196 static int
 197 fs_dev_start(struct rte_eth_dev *dev)
 198 {
 199         struct sub_device *sdev;
 200         uint8_t i;
 201         int ret;
 202
 203         ret = failsafe_rx_intr_install(dev);
 204         if (ret)
 205                 return ret;
 206         FOREACH_SUBDEV(sdev, i, dev) {
 207                 if (sdev->state != DEV_ACTIVE)
 208                         continue;
 209                 DEBUG("Starting sub_device %d", i);
 210                 ret = rte_eth_dev_start(PORT_ID(sdev));
 211                 if (ret) {
 212                         if (!fs_err(sdev, ret))
 213                                 continue;
 214                         return ret;
 215                 }
 216                 ret = failsafe_rx_intr_install_subdevice(sdev);
 217                 if (ret) {
 218                         if (!fs_err(sdev, ret))
 219                                 continue;
 220                         rte_eth_dev_stop(PORT_ID(sdev));
 221                         return ret;
 222                 }
 223                 sdev->state = DEV_STARTED;
 224         }
 225         if (PRIV(dev)->state < DEV_STARTED)
 226                 PRIV(dev)->state = DEV_STARTED;
 227         fs_switch_dev(dev, NULL);
 228         return 0;
 229 }
 230
 231 static void
 232 fs_dev_stop(struct rte_eth_dev *dev)
 233 {
 234         struct sub_device *sdev;
 235         uint8_t i;
 236
 237         PRIV(dev)->state = DEV_STARTED - 1;
 238         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_STARTED) {
 239                 rte_eth_dev_stop(PORT_ID(sdev));
 240                 failsafe_rx_intr_uninstall_subdevice(sdev);
 241                 sdev->state = DEV_STARTED - 1;
 242         }
 243         failsafe_rx_intr_uninstall(dev);
 244 }
 245
 246 static int
 247 fs_dev_set_link_up(struct rte_eth_dev *dev)
 248 {
 249         struct sub_device *sdev;
 250         uint8_t i;
 251         int ret;
 252
 253         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 254                 DEBUG("Calling rte_eth_dev_set_link_up on sub_device %d", i);
 255                 ret = rte_eth_dev_set_link_up(PORT_ID(sdev));
 256                 if ((ret = fs_err(sdev, ret))) {
 257                         ERROR("Operation rte_eth_dev_set_link_up failed for sub_device %d"
 258                               " with error %d", i, ret);
 259                         return ret;
 260                 }
 261         }
 262         return 0;
 263 }
 264
 265 static int
 266 fs_dev_set_link_down(struct rte_eth_dev *dev)
 267 {
 268         struct sub_device *sdev;
 269         uint8_t i;
 270         int ret;
 271
 272         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 273                 DEBUG("Calling rte_eth_dev_set_link_down on sub_device %d", i);
 274                 ret = rte_eth_dev_set_link_down(PORT_ID(sdev));
 275                 if ((ret = fs_err(sdev, ret))) {
 276                         ERROR("Operation rte_eth_dev_set_link_down failed for sub_device %d"
 277                               " with error %d", i, ret);
 278                         return ret;
 279                 }
 280         }
 281         return 0;
 282 }
 283
 284 static void fs_dev_free_queues(struct rte_eth_dev *dev);
 285 static void
 286 fs_dev_close(struct rte_eth_dev *dev)
 287 {
 288         struct sub_device *sdev;
 289         uint8_t i;
 290
 291         failsafe_hotplug_alarm_cancel(dev);
 292         if (PRIV(dev)->state == DEV_STARTED)
 293                 dev->dev_ops->dev_stop(dev);
 294         PRIV(dev)->state = DEV_ACTIVE - 1;
 295         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 296                 DEBUG("Closing sub_device %d", i);
 297                 rte_eth_dev_close(PORT_ID(sdev));
 298                 sdev->state = DEV_ACTIVE - 1;
 299         }
 300         fs_dev_free_queues(dev);
 301 }
 302
 303 static bool
 304 fs_rxq_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads)
 305 {
 306         uint64_t port_offloads;
 307         uint64_t queue_supp_offloads;
 308         uint64_t port_supp_offloads;
 309
 310         port_offloads = dev->data->dev_conf.rxmode.offloads;
 311         queue_supp_offloads = PRIV(dev)->infos.rx_queue_offload_capa;
 312         port_supp_offloads = PRIV(dev)->infos.rx_offload_capa;
 313         if ((offloads & (queue_supp_offloads | port_supp_offloads)) !=
 314              offloads)
 315                 return false;
 316         /* Verify we have no conflict with port offloads */
 317         if ((port_offloads ^ offloads) & port_supp_offloads)
 318                 return false;
 319         return true;
 320 }
 321
 322 static void
 323 fs_rx_queue_release(void *queue)
 324 {
 325         struct rte_eth_dev *dev;
 326         struct sub_device *sdev;
 327         uint8_t i;
 328         struct rxq *rxq;
 329
 330         if (queue == NULL)
 331                 return;
 332         rxq = queue;
 333         if (rxq->event_fd > 0)
 334                 close(rxq->event_fd);
 335         dev = rxq->priv->dev;
 336         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
 337                 SUBOPS(sdev, rx_queue_release)
 338                         (ETH(sdev)->data->rx_queues[rxq->qid]);
 339         dev->data->rx_queues[rxq->qid] = NULL;
 340         rte_free(rxq);
 341 }
 342
 343 static int
 344 fs_rx_queue_setup(struct rte_eth_dev *dev,
 345                 uint16_t rx_queue_id,
 346                 uint16_t nb_rx_desc,
 347                 unsigned int socket_id,
 348                 const struct rte_eth_rxconf *rx_conf,
 349                 struct rte_mempool *mb_pool)
 350 {
 351         /*
 352          * FIXME: Add a proper interface in rte_eal_interrupts for
 353          * allocating eventfd as an interrupt vector.
 354          * For the time being, fake as if we are using MSIX interrupts,
 355          * this will cause rte_intr_efd_enable to allocate an eventfd for us.
 356          */
 357         struct rte_intr_handle intr_handle = {
 358                 .type = RTE_INTR_HANDLE_VFIO_MSIX,
 359                 .efds = { -1, },
 360         };
 361         struct sub_device *sdev;
 362         struct rxq *rxq;
 363         uint8_t i;
 364         int ret;
 365
 366         rxq = dev->data->rx_queues[rx_queue_id];
 367         if (rxq != NULL) {
 368                 fs_rx_queue_release(rxq);
 369                 dev->data->rx_queues[rx_queue_id] = NULL;
 370         }
 371         /* Verify application offloads are valid for our port and queue. */
 372         if (fs_rxq_offloads_valid(dev, rx_conf->offloads) == false) {
 373                 rte_errno = ENOTSUP;
 374                 ERROR("Rx queue offloads 0x%" PRIx64
 375                       " don't match port offloads 0x%" PRIx64
 376                       " or supported offloads 0x%" PRIx64,
 377                       rx_conf->offloads,
 378                       dev->data->dev_conf.rxmode.offloads,
 379                       PRIV(dev)->infos.rx_offload_capa |
 380                       PRIV(dev)->infos.rx_queue_offload_capa);
 381                 return -rte_errno;
 382         }
 383         rxq = rte_zmalloc(NULL,
 384                           sizeof(*rxq) +
 385                           sizeof(rte_atomic64_t) * PRIV(dev)->subs_tail,
 386                           RTE_CACHE_LINE_SIZE);
 387         if (rxq == NULL)
 388                 return -ENOMEM;
 389         FOREACH_SUBDEV(sdev, i, dev)
 390                 rte_atomic64_init(&rxq->refcnt[i]);
 391         rxq->qid = rx_queue_id;
 392         rxq->socket_id = socket_id;
 393         rxq->info.mp = mb_pool;
 394         rxq->info.conf = *rx_conf;
 395         rxq->info.nb_desc = nb_rx_desc;
 396         rxq->priv = PRIV(dev);
 397         rxq->sdev = PRIV(dev)->subs;
 398         ret = rte_intr_efd_enable(&intr_handle, 1);
 399         if (ret < 0)
 400                 return ret;
 401         rxq->event_fd = intr_handle.efds[0];
 402         dev->data->rx_queues[rx_queue_id] = rxq;
 403         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 404                 ret = rte_eth_rx_queue_setup(PORT_ID(sdev),
 405                                 rx_queue_id,
 406                                 nb_rx_desc, socket_id,
 407                                 rx_conf, mb_pool);
 408                 if ((ret = fs_err(sdev, ret))) {
 409                         ERROR("RX queue setup failed for sub_device %d", i);
 410                         goto free_rxq;
 411                 }
 412         }
 413         return 0;
 414 free_rxq:
 415         fs_rx_queue_release(rxq);
 416         return ret;
 417 }
 418
 419 static int
 420 fs_rx_intr_enable(struct rte_eth_dev *dev, uint16_t idx)
 421 {
 422         struct rxq *rxq;
 423         struct sub_device *sdev;
 424         uint8_t i;
 425         int ret;
 426         int rc = 0;
 427
 428         if (idx >= dev->data->nb_rx_queues) {
 429                 rte_errno = EINVAL;
 430                 return -rte_errno;
 431         }
 432         rxq = dev->data->rx_queues[idx];
 433         if (rxq == NULL || rxq->event_fd <= 0) {
 434                 rte_errno = EINVAL;
 435                 return -rte_errno;
 436         }
 437         rxq->enable_events = 1;
 438         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 439                 ret = rte_eth_dev_rx_intr_enable(PORT_ID(sdev), idx);
 440                 ret = fs_err(sdev, ret);
 441                 if (ret)
 442                         rc = ret;
 443         }
 444         if (rc)
 445                 rte_errno = -rc;
 446         return rc;
 447 }
 448
 449 static int
 450 fs_rx_intr_disable(struct rte_eth_dev *dev, uint16_t idx)
 451 {
 452         struct rxq *rxq;
 453         struct sub_device *sdev;
 454         uint64_t u64;
 455         uint8_t i;
 456         int rc = 0;
 457         int ret;
 458
 459         if (idx >= dev->data->nb_rx_queues) {
 460                 rte_errno = EINVAL;
 461                 return -rte_errno;
 462         }
 463         rxq = dev->data->rx_queues[idx];
 464         if (rxq == NULL || rxq->event_fd <= 0) {
 465                 rte_errno = EINVAL;
 466                 return -rte_errno;
 467         }
 468         rxq->enable_events = 0;
 469         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 470                 ret = rte_eth_dev_rx_intr_disable(PORT_ID(sdev), idx);
 471                 ret = fs_err(sdev, ret);
 472                 if (ret)
 473                         rc = ret;
 474         }
 475         /* Clear pending events */
 476         while (read(rxq->event_fd, &u64, sizeof(uint64_t)) >  0)
 477                 ;
 478         if (rc)
 479                 rte_errno = -rc;
 480         return rc;
 481 }
 482
 483 static bool
 484 fs_txq_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads)
 485 {
 486         uint64_t port_offloads;
 487         uint64_t queue_supp_offloads;
 488         uint64_t port_supp_offloads;
 489
 490         port_offloads = dev->data->dev_conf.txmode.offloads;
 491         queue_supp_offloads = PRIV(dev)->infos.tx_queue_offload_capa;
 492         port_supp_offloads = PRIV(dev)->infos.tx_offload_capa;
 493         if ((offloads & (queue_supp_offloads | port_supp_offloads)) !=
 494              offloads)
 495                 return false;
 496         /* Verify we have no conflict with port offloads */
 497         if ((port_offloads ^ offloads) & port_supp_offloads)
 498                 return false;
 499         return true;
 500 }
 501
 502 static void
 503 fs_tx_queue_release(void *queue)
 504 {
 505         struct rte_eth_dev *dev;
 506         struct sub_device *sdev;
 507         uint8_t i;
 508         struct txq *txq;
 509
 510         if (queue == NULL)
 511                 return;
 512         txq = queue;
 513         dev = txq->priv->dev;
 514         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
 515                 SUBOPS(sdev, tx_queue_release)
 516                         (ETH(sdev)->data->tx_queues[txq->qid]);
 517         dev->data->tx_queues[txq->qid] = NULL;
 518         rte_free(txq);
 519 }
 520
 521 static int
 522 fs_tx_queue_setup(struct rte_eth_dev *dev,
 523                 uint16_t tx_queue_id,
 524                 uint16_t nb_tx_desc,
 525                 unsigned int socket_id,
 526                 const struct rte_eth_txconf *tx_conf)
 527 {
 528         struct sub_device *sdev;
 529         struct txq *txq;
 530         uint8_t i;
 531         int ret;
 532
 533         txq = dev->data->tx_queues[tx_queue_id];
 534         if (txq != NULL) {
 535                 fs_tx_queue_release(txq);
 536                 dev->data->tx_queues[tx_queue_id] = NULL;
 537         }
 538         /*
 539          * Don't verify queue offloads for applications which
 540          * use the old API.
 541          */
 542         if (tx_conf != NULL &&
 543             (tx_conf->txq_flags & ETH_TXQ_FLAGS_IGNORE) &&
 544             fs_txq_offloads_valid(dev, tx_conf->offloads) == false) {
 545                 rte_errno = ENOTSUP;
 546                 ERROR("Tx queue offloads 0x%" PRIx64
 547                       " don't match port offloads 0x%" PRIx64
 548                       " or supported offloads 0x%" PRIx64,
 549                       tx_conf->offloads,
 550                       dev->data->dev_conf.txmode.offloads,
 551                       PRIV(dev)->infos.tx_offload_capa |
 552                       PRIV(dev)->infos.tx_queue_offload_capa);
 553                 return -rte_errno;
 554         }
 555         txq = rte_zmalloc("ethdev TX queue",
 556                           sizeof(*txq) +
 557                           sizeof(rte_atomic64_t) * PRIV(dev)->subs_tail,
 558                           RTE_CACHE_LINE_SIZE);
 559         if (txq == NULL)
 560                 return -ENOMEM;
 561         FOREACH_SUBDEV(sdev, i, dev)
 562                 rte_atomic64_init(&txq->refcnt[i]);
 563         txq->qid = tx_queue_id;
 564         txq->socket_id = socket_id;
 565         txq->info.conf = *tx_conf;
 566         txq->info.nb_desc = nb_tx_desc;
 567         txq->priv = PRIV(dev);
 568         dev->data->tx_queues[tx_queue_id] = txq;
 569         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 570                 ret = rte_eth_tx_queue_setup(PORT_ID(sdev),
 571                                 tx_queue_id,
 572                                 nb_tx_desc, socket_id,
 573                                 tx_conf);
 574                 if ((ret = fs_err(sdev, ret))) {
 575                         ERROR("TX queue setup failed for sub_device %d", i);
 576                         goto free_txq;
 577                 }
 578         }
 579         return 0;
 580 free_txq:
 581         fs_tx_queue_release(txq);
 582         return ret;
 583 }
 584
 585 static void
 586 fs_dev_free_queues(struct rte_eth_dev *dev)
 587 {
 588         uint16_t i;
 589
 590         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 591                 fs_rx_queue_release(dev->data->rx_queues[i]);
 592                 dev->data->rx_queues[i] = NULL;
 593         }
 594         dev->data->nb_rx_queues = 0;
 595         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 596                 fs_tx_queue_release(dev->data->tx_queues[i]);
 597                 dev->data->tx_queues[i] = NULL;
 598         }
 599         dev->data->nb_tx_queues = 0;
 600 }
 601
 602 static void
 603 fs_promiscuous_enable(struct rte_eth_dev *dev)
 604 {
 605         struct sub_device *sdev;
 606         uint8_t i;
 607
 608         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
 609                 rte_eth_promiscuous_enable(PORT_ID(sdev));
 610 }
 611
 612 static void
 613 fs_promiscuous_disable(struct rte_eth_dev *dev)
 614 {
 615         struct sub_device *sdev;
 616         uint8_t i;
 617
 618         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
 619                 rte_eth_promiscuous_disable(PORT_ID(sdev));
 620 }
 621
 622 static void
 623 fs_allmulticast_enable(struct rte_eth_dev *dev)
 624 {
 625         struct sub_device *sdev;
 626         uint8_t i;
 627
 628         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
 629                 rte_eth_allmulticast_enable(PORT_ID(sdev));
 630 }
 631
 632 static void
 633 fs_allmulticast_disable(struct rte_eth_dev *dev)
 634 {
 635         struct sub_device *sdev;
 636         uint8_t i;
 637
 638         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
 639                 rte_eth_allmulticast_disable(PORT_ID(sdev));
 640 }
 641
 642 static int
 643 fs_link_update(struct rte_eth_dev *dev,
 644                 int wait_to_complete)
 645 {
 646         struct sub_device *sdev;
 647         uint8_t i;
 648         int ret;
 649
 650         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 651                 DEBUG("Calling link_update on sub_device %d", i);
 652                 ret = (SUBOPS(sdev, link_update))(ETH(sdev), wait_to_complete);
 653                 if (ret && ret != -1 && sdev->remove == 0 &&
 654                     rte_eth_dev_is_removed(PORT_ID(sdev)) == 0) {
 655                         ERROR("Link update failed for sub_device %d with error %d",
 656                               i, ret);
 657                         return ret;
 658                 }
 659         }
 660         if (TX_SUBDEV(dev)) {
 661                 struct rte_eth_link *l1;
 662                 struct rte_eth_link *l2;
 663
 664                 l1 = &dev->data->dev_link;
 665                 l2 = &ETH(TX_SUBDEV(dev))->data->dev_link;
 666                 if (memcmp(l1, l2, sizeof(*l1))) {
 667                         *l1 = *l2;
 668                         return 0;
 669                 }
 670         }
 671         return -1;
 672 }
 673
 674 static int
 675 fs_stats_get(struct rte_eth_dev *dev,
 676              struct rte_eth_stats *stats)
 677 {
 678         struct rte_eth_stats backup;
 679         struct sub_device *sdev;
 680         uint8_t i;
 681         int ret;
 682
 683         rte_memcpy(stats, &PRIV(dev)->stats_accumulator, sizeof(*stats));
 684         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 685                 struct rte_eth_stats *snapshot = &sdev->stats_snapshot.stats;
 686                 uint64_t *timestamp = &sdev->stats_snapshot.timestamp;
 687
 688                 rte_memcpy(&backup, snapshot, sizeof(backup));
 689                 ret = rte_eth_stats_get(PORT_ID(sdev), snapshot);
 690                 if (ret) {
 691                         if (!fs_err(sdev, ret)) {
 692                                 rte_memcpy(snapshot, &backup, sizeof(backup));
 693                                 goto inc;
 694                         }
 695                         ERROR("Operation rte_eth_stats_get failed for sub_device %d with error %d",
 696                                   i, ret);
 697                         *timestamp = 0;
 698                         return ret;
 699                 }
 700                 *timestamp = rte_rdtsc();
 701 inc:
 702                 failsafe_stats_increment(stats, snapshot);
 703         }
 704         return 0;
 705 }
 706
 707 static void
 708 fs_stats_reset(struct rte_eth_dev *dev)
 709 {
 710         struct sub_device *sdev;
 711         uint8_t i;
 712
 713         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 714                 rte_eth_stats_reset(PORT_ID(sdev));
 715                 memset(&sdev->stats_snapshot, 0, sizeof(struct rte_eth_stats));
 716         }
 717         memset(&PRIV(dev)->stats_accumulator, 0, sizeof(struct rte_eth_stats));
 718 }
 719
 720 /**
 721  * Fail-safe dev_infos_get rules:
 722  *
 723  * No sub_device:
 724  *   Numerables:
 725  *      Use the maximum possible values for any field, so as not
 726  *      to impede any further configuration effort.
 727  *   Capabilities:
 728  *      Limits capabilities to those that are understood by the
 729  *      fail-safe PMD. This understanding stems from the fail-safe
 730  *      being capable of verifying that the related capability is
 731  *      expressed within the device configuration (struct rte_eth_conf).
 732  *
 733  * At least one probed sub_device:
 734  *   Numerables:
 735  *      Uses values from the active probed sub_device
 736  *      The rationale here is that if any sub_device is less capable
 737  *      (for example concerning the number of queues) than the active
 738  *      sub_device, then its subsequent configuration will fail.
 739  *      It is impossible to foresee this failure when the failing sub_device
 740  *      is supposed to be plugged-in later on, so the configuration process
 741  *      is the single point of failure and error reporting.
 742  *   Capabilities:
 743  *      Uses a logical AND of RX capabilities among
 744  *      all sub_devices and the default capabilities.
 745  *      Uses a logical AND of TX capabilities among
 746  *      the active probed sub_device and the default capabilities.
 747  *
 748  */
 749 static void
 750 fs_dev_infos_get(struct rte_eth_dev *dev,
 751                   struct rte_eth_dev_info *infos)
 752 {
 753         struct sub_device *sdev;
 754         uint8_t i;
 755
 756         sdev = TX_SUBDEV(dev);
 757         if (sdev == NULL) {
 758                 DEBUG("No probed device, using default infos");
 759                 rte_memcpy(&PRIV(dev)->infos, &default_infos,
 760                            sizeof(default_infos));
 761         } else {
 762                 uint64_t rx_offload_capa;
 763                 uint64_t rxq_offload_capa;
 764
 765                 rx_offload_capa = default_infos.rx_offload_capa;
 766                 rxq_offload_capa = default_infos.rx_queue_offload_capa;
 767                 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_PROBED) {
 768                         rte_eth_dev_info_get(PORT_ID(sdev),
 769                                         &PRIV(dev)->infos);
 770                         rx_offload_capa &= PRIV(dev)->infos.rx_offload_capa;
 771                         rxq_offload_capa &=
 772                                         PRIV(dev)->infos.rx_queue_offload_capa;
 773                 }
 774                 sdev = TX_SUBDEV(dev);
 775                 rte_eth_dev_info_get(PORT_ID(sdev), &PRIV(dev)->infos);
 776                 PRIV(dev)->infos.rx_offload_capa = rx_offload_capa;
 777                 PRIV(dev)->infos.rx_queue_offload_capa = rxq_offload_capa;
 778                 PRIV(dev)->infos.tx_offload_capa &=
 779                                         default_infos.tx_offload_capa;
 780                 PRIV(dev)->infos.tx_queue_offload_capa &=
 781                                         default_infos.tx_queue_offload_capa;
 782                 PRIV(dev)->infos.flow_type_rss_offloads &=
 783                                         default_infos.flow_type_rss_offloads;
 784         }
 785         rte_memcpy(infos, &PRIV(dev)->infos, sizeof(*infos));
 786 }
 787
 788 static const uint32_t *
 789 fs_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 790 {
 791         struct sub_device *sdev;
 792         struct rte_eth_dev *edev;
 793
 794         sdev = TX_SUBDEV(dev);
 795         if (sdev == NULL)
 796                 return NULL;
 797         edev = ETH(sdev);
 798         /* ENOTSUP: counts as no supported ptypes */
 799         if (SUBOPS(sdev, dev_supported_ptypes_get) == NULL)
 800                 return NULL;
 801         /*
 802          * The API does not permit to do a clean AND of all ptypes,
 803          * It is also incomplete by design and we do not really care
 804          * to have a best possible value in this context.
 805          * We just return the ptypes of the device of highest
 806          * priority, usually the PREFERRED device.
 807          */
 808         return SUBOPS(sdev, dev_supported_ptypes_get)(edev);
 809 }
 810
 811 static int
 812 fs_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 813 {
 814         struct sub_device *sdev;
 815         uint8_t i;
 816         int ret;
 817
 818         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 819                 DEBUG("Calling rte_eth_dev_set_mtu on sub_device %d", i);
 820                 ret = rte_eth_dev_set_mtu(PORT_ID(sdev), mtu);
 821                 if ((ret = fs_err(sdev, ret))) {
 822                         ERROR("Operation rte_eth_dev_set_mtu failed for sub_device %d with error %d",
 823                               i, ret);
 824                         return ret;
 825                 }
 826         }
 827         return 0;
 828 }
 829
 830 static int
 831 fs_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
 832 {
 833         struct sub_device *sdev;
 834         uint8_t i;
 835         int ret;
 836
 837         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 838                 DEBUG("Calling rte_eth_dev_vlan_filter on sub_device %d", i);
 839                 ret = rte_eth_dev_vlan_filter(PORT_ID(sdev), vlan_id, on);
 840                 if ((ret = fs_err(sdev, ret))) {
 841                         ERROR("Operation rte_eth_dev_vlan_filter failed for sub_device %d"
 842                               " with error %d", i, ret);
 843                         return ret;
 844                 }
 845         }
 846         return 0;
 847 }
 848
 849 static int
 850 fs_flow_ctrl_get(struct rte_eth_dev *dev,
 851                 struct rte_eth_fc_conf *fc_conf)
 852 {
 853         struct sub_device *sdev;
 854
 855         sdev = TX_SUBDEV(dev);
 856         if (sdev == NULL)
 857                 return 0;
 858         if (SUBOPS(sdev, flow_ctrl_get) == NULL)
 859                 return -ENOTSUP;
 860         return SUBOPS(sdev, flow_ctrl_get)(ETH(sdev), fc_conf);
 861 }
 862
 863 static int
 864 fs_flow_ctrl_set(struct rte_eth_dev *dev,
 865                 struct rte_eth_fc_conf *fc_conf)
 866 {
 867         struct sub_device *sdev;
 868         uint8_t i;
 869         int ret;
 870
 871         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 872                 DEBUG("Calling rte_eth_dev_flow_ctrl_set on sub_device %d", i);
 873                 ret = rte_eth_dev_flow_ctrl_set(PORT_ID(sdev), fc_conf);
 874                 if ((ret = fs_err(sdev, ret))) {
 875                         ERROR("Operation rte_eth_dev_flow_ctrl_set failed for sub_device %d"
 876                               " with error %d", i, ret);
 877                         return ret;
 878                 }
 879         }
 880         return 0;
 881 }
 882
 883 static void
 884 fs_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
 885 {
 886         struct sub_device *sdev;
 887         uint8_t i;
 888
 889         /* No check: already done within the rte_eth_dev_mac_addr_remove
 890          * call for the fail-safe device.
 891          */
 892         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
 893                 rte_eth_dev_mac_addr_remove(PORT_ID(sdev),
 894                                 &dev->data->mac_addrs[index]);
 895         PRIV(dev)->mac_addr_pool[index] = 0;
 896 }
 897
 898 static int
 899 fs_mac_addr_add(struct rte_eth_dev *dev,
 900                 struct ether_addr *mac_addr,
 901                 uint32_t index,
 902                 uint32_t vmdq)
 903 {
 904         struct sub_device *sdev;
 905         int ret;
 906         uint8_t i;
 907
 908         RTE_ASSERT(index < FAILSAFE_MAX_ETHADDR);
 909         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 910                 ret = rte_eth_dev_mac_addr_add(PORT_ID(sdev), mac_addr, vmdq);
 911                 if ((ret = fs_err(sdev, ret))) {
 912                         ERROR("Operation rte_eth_dev_mac_addr_add failed for sub_device %"
 913                               PRIu8 " with error %d", i, ret);
 914                         return ret;
 915                 }
 916         }
 917         if (index >= PRIV(dev)->nb_mac_addr) {
 918                 DEBUG("Growing mac_addrs array");
 919                 PRIV(dev)->nb_mac_addr = index;
 920         }
 921         PRIV(dev)->mac_addr_pool[index] = vmdq;
 922         return 0;
 923 }
 924
 925 static void
 926 fs_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 927 {
 928         struct sub_device *sdev;
 929         uint8_t i;
 930
 931         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
 932                 rte_eth_dev_default_mac_addr_set(PORT_ID(sdev), mac_addr);
 933 }
 934
 935 static int
 936 fs_filter_ctrl(struct rte_eth_dev *dev,
 937                 enum rte_filter_type type,
 938                 enum rte_filter_op op,
 939                 void *arg)
 940 {
 941         struct sub_device *sdev;
 942         uint8_t i;
 943         int ret;
 944
 945         if (type == RTE_ETH_FILTER_GENERIC &&
 946             op == RTE_ETH_FILTER_GET) {
 947                 *(const void **)arg = &fs_flow_ops;
 948                 return 0;
 949         }
 950         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 951                 DEBUG("Calling rte_eth_dev_filter_ctrl on sub_device %d", i);
 952                 ret = rte_eth_dev_filter_ctrl(PORT_ID(sdev), type, op, arg);
 953                 if ((ret = fs_err(sdev, ret))) {
 954                         ERROR("Operation rte_eth_dev_filter_ctrl failed for sub_device %d"
 955                               " with error %d", i, ret);
 956                         return ret;
 957                 }
 958         }
 959         return 0;
 960 }
 961
 962 const struct eth_dev_ops failsafe_ops = {
 963         .dev_configure = fs_dev_configure,
 964         .dev_start = fs_dev_start,
 965         .dev_stop = fs_dev_stop,
 966         .dev_set_link_down = fs_dev_set_link_down,
 967         .dev_set_link_up = fs_dev_set_link_up,
 968         .dev_close = fs_dev_close,
 969         .promiscuous_enable = fs_promiscuous_enable,
 970         .promiscuous_disable = fs_promiscuous_disable,
 971         .allmulticast_enable = fs_allmulticast_enable,
 972         .allmulticast_disable = fs_allmulticast_disable,
 973         .link_update = fs_link_update,
 974         .stats_get = fs_stats_get,
 975         .stats_reset = fs_stats_reset,
 976         .dev_infos_get = fs_dev_infos_get,
 977         .dev_supported_ptypes_get = fs_dev_supported_ptypes_get,
 978         .mtu_set = fs_mtu_set,
 979         .vlan_filter_set = fs_vlan_filter_set,
 980         .rx_queue_setup = fs_rx_queue_setup,
 981         .tx_queue_setup = fs_tx_queue_setup,
 982         .rx_queue_release = fs_rx_queue_release,
 983         .tx_queue_release = fs_tx_queue_release,
 984         .rx_queue_intr_enable = fs_rx_intr_enable,
 985         .rx_queue_intr_disable = fs_rx_intr_disable,
 986         .flow_ctrl_get = fs_flow_ctrl_get,
 987         .flow_ctrl_set = fs_flow_ctrl_set,
 988         .mac_addr_remove = fs_mac_addr_remove,
 989         .mac_addr_add = fs_mac_addr_add,
 990         .mac_addr_set = fs_mac_addr_set,
 991         .filter_ctrl = fs_filter_ctrl,
 992 };