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                 sdev->state = DEV_STARTED;
 217         }
 218         if (PRIV(dev)->state < DEV_STARTED)
 219                 PRIV(dev)->state = DEV_STARTED;
 220         fs_switch_dev(dev, NULL);
 221         return 0;
 222 }
 223
 224 static void
 225 fs_dev_stop(struct rte_eth_dev *dev)
 226 {
 227         struct sub_device *sdev;
 228         uint8_t i;
 229
 230         PRIV(dev)->state = DEV_STARTED - 1;
 231         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_STARTED) {
 232                 rte_eth_dev_stop(PORT_ID(sdev));
 233                 sdev->state = DEV_STARTED - 1;
 234         }
 235         failsafe_rx_intr_uninstall(dev);
 236 }
 237
 238 static int
 239 fs_dev_set_link_up(struct rte_eth_dev *dev)
 240 {
 241         struct sub_device *sdev;
 242         uint8_t i;
 243         int ret;
 244
 245         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 246                 DEBUG("Calling rte_eth_dev_set_link_up on sub_device %d", i);
 247                 ret = rte_eth_dev_set_link_up(PORT_ID(sdev));
 248                 if ((ret = fs_err(sdev, ret))) {
 249                         ERROR("Operation rte_eth_dev_set_link_up failed for sub_device %d"
 250                               " with error %d", i, ret);
 251                         return ret;
 252                 }
 253         }
 254         return 0;
 255 }
 256
 257 static int
 258 fs_dev_set_link_down(struct rte_eth_dev *dev)
 259 {
 260         struct sub_device *sdev;
 261         uint8_t i;
 262         int ret;
 263
 264         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 265                 DEBUG("Calling rte_eth_dev_set_link_down on sub_device %d", i);
 266                 ret = rte_eth_dev_set_link_down(PORT_ID(sdev));
 267                 if ((ret = fs_err(sdev, ret))) {
 268                         ERROR("Operation rte_eth_dev_set_link_down failed for sub_device %d"
 269                               " with error %d", i, ret);
 270                         return ret;
 271                 }
 272         }
 273         return 0;
 274 }
 275
 276 static void fs_dev_free_queues(struct rte_eth_dev *dev);
 277 static void
 278 fs_dev_close(struct rte_eth_dev *dev)
 279 {
 280         struct sub_device *sdev;
 281         uint8_t i;
 282
 283         failsafe_hotplug_alarm_cancel(dev);
 284         if (PRIV(dev)->state == DEV_STARTED)
 285                 dev->dev_ops->dev_stop(dev);
 286         PRIV(dev)->state = DEV_ACTIVE - 1;
 287         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 288                 DEBUG("Closing sub_device %d", i);
 289                 rte_eth_dev_close(PORT_ID(sdev));
 290                 sdev->state = DEV_ACTIVE - 1;
 291         }
 292         fs_dev_free_queues(dev);
 293 }
 294
 295 static bool
 296 fs_rxq_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads)
 297 {
 298         uint64_t port_offloads;
 299         uint64_t queue_supp_offloads;
 300         uint64_t port_supp_offloads;
 301
 302         port_offloads = dev->data->dev_conf.rxmode.offloads;
 303         queue_supp_offloads = PRIV(dev)->infos.rx_queue_offload_capa;
 304         port_supp_offloads = PRIV(dev)->infos.rx_offload_capa;
 305         if ((offloads & (queue_supp_offloads | port_supp_offloads)) !=
 306              offloads)
 307                 return false;
 308         /* Verify we have no conflict with port offloads */
 309         if ((port_offloads ^ offloads) & port_supp_offloads)
 310                 return false;
 311         return true;
 312 }
 313
 314 static void
 315 fs_rx_queue_release(void *queue)
 316 {
 317         struct rte_eth_dev *dev;
 318         struct sub_device *sdev;
 319         uint8_t i;
 320         struct rxq *rxq;
 321
 322         if (queue == NULL)
 323                 return;
 324         rxq = queue;
 325         if (rxq->event_fd > 0)
 326                 close(rxq->event_fd);
 327         dev = rxq->priv->dev;
 328         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
 329                 SUBOPS(sdev, rx_queue_release)
 330                         (ETH(sdev)->data->rx_queues[rxq->qid]);
 331         dev->data->rx_queues[rxq->qid] = NULL;
 332         rte_free(rxq);
 333 }
 334
 335 static int
 336 fs_rx_queue_setup(struct rte_eth_dev *dev,
 337                 uint16_t rx_queue_id,
 338                 uint16_t nb_rx_desc,
 339                 unsigned int socket_id,
 340                 const struct rte_eth_rxconf *rx_conf,
 341                 struct rte_mempool *mb_pool)
 342 {
 343         /*
 344          * FIXME: Add a proper interface in rte_eal_interrupts for
 345          * allocating eventfd as an interrupt vector.
 346          * For the time being, fake as if we are using MSIX interrupts,
 347          * this will cause rte_intr_efd_enable to allocate an eventfd for us.
 348          */
 349         struct rte_intr_handle intr_handle = {
 350                 .type = RTE_INTR_HANDLE_VFIO_MSIX,
 351                 .efds = { -1, },
 352         };
 353         struct sub_device *sdev;
 354         struct rxq *rxq;
 355         uint8_t i;
 356         int ret;
 357
 358         rxq = dev->data->rx_queues[rx_queue_id];
 359         if (rxq != NULL) {
 360                 fs_rx_queue_release(rxq);
 361                 dev->data->rx_queues[rx_queue_id] = NULL;
 362         }
 363         /* Verify application offloads are valid for our port and queue. */
 364         if (fs_rxq_offloads_valid(dev, rx_conf->offloads) == false) {
 365                 rte_errno = ENOTSUP;
 366                 ERROR("Rx queue offloads 0x%" PRIx64
 367                       " don't match port offloads 0x%" PRIx64
 368                       " or supported offloads 0x%" PRIx64,
 369                       rx_conf->offloads,
 370                       dev->data->dev_conf.rxmode.offloads,
 371                       PRIV(dev)->infos.rx_offload_capa |
 372                       PRIV(dev)->infos.rx_queue_offload_capa);
 373                 return -rte_errno;
 374         }
 375         rxq = rte_zmalloc(NULL,
 376                           sizeof(*rxq) +
 377                           sizeof(rte_atomic64_t) * PRIV(dev)->subs_tail,
 378                           RTE_CACHE_LINE_SIZE);
 379         if (rxq == NULL)
 380                 return -ENOMEM;
 381         FOREACH_SUBDEV(sdev, i, dev)
 382                 rte_atomic64_init(&rxq->refcnt[i]);
 383         rxq->qid = rx_queue_id;
 384         rxq->socket_id = socket_id;
 385         rxq->info.mp = mb_pool;
 386         rxq->info.conf = *rx_conf;
 387         rxq->info.nb_desc = nb_rx_desc;
 388         rxq->priv = PRIV(dev);
 389         rxq->sdev = PRIV(dev)->subs;
 390         ret = rte_intr_efd_enable(&intr_handle, 1);
 391         if (ret < 0)
 392                 return ret;
 393         rxq->event_fd = intr_handle.efds[0];
 394         dev->data->rx_queues[rx_queue_id] = rxq;
 395         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 396                 ret = rte_eth_rx_queue_setup(PORT_ID(sdev),
 397                                 rx_queue_id,
 398                                 nb_rx_desc, socket_id,
 399                                 rx_conf, mb_pool);
 400                 if ((ret = fs_err(sdev, ret))) {
 401                         ERROR("RX queue setup failed for sub_device %d", i);
 402                         goto free_rxq;
 403                 }
 404         }
 405         return 0;
 406 free_rxq:
 407         fs_rx_queue_release(rxq);
 408         return ret;
 409 }
 410
 411 static int
 412 fs_rx_intr_enable(struct rte_eth_dev *dev, uint16_t idx)
 413 {
 414         struct rxq *rxq;
 415
 416         if (idx >= dev->data->nb_rx_queues) {
 417                 rte_errno = EINVAL;
 418                 return -rte_errno;
 419         }
 420         rxq = dev->data->rx_queues[idx];
 421         if (rxq == NULL || rxq->event_fd <= 0) {
 422                 rte_errno = EINVAL;
 423                 return -rte_errno;
 424         }
 425         rxq->enable_events = 1;
 426         return 0;
 427 }
 428
 429 static int
 430 fs_rx_intr_disable(struct rte_eth_dev *dev, uint16_t idx)
 431 {
 432         struct rxq *rxq;
 433         uint64_t u64;
 434
 435         if (idx >= dev->data->nb_rx_queues) {
 436                 rte_errno = EINVAL;
 437                 return -rte_errno;
 438         }
 439         rxq = dev->data->rx_queues[idx];
 440         if (rxq == NULL || rxq->event_fd <= 0) {
 441                 rte_errno = EINVAL;
 442                 return -rte_errno;
 443         }
 444         rxq->enable_events = 0;
 445         /* Clear pending events */
 446         while (read(rxq->event_fd, &u64, sizeof(uint64_t)) >  0)
 447                 ;
 448         return 0;
 449 }
 450
 451 static bool
 452 fs_txq_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads)
 453 {
 454         uint64_t port_offloads;
 455         uint64_t queue_supp_offloads;
 456         uint64_t port_supp_offloads;
 457
 458         port_offloads = dev->data->dev_conf.txmode.offloads;
 459         queue_supp_offloads = PRIV(dev)->infos.tx_queue_offload_capa;
 460         port_supp_offloads = PRIV(dev)->infos.tx_offload_capa;
 461         if ((offloads & (queue_supp_offloads | port_supp_offloads)) !=
 462              offloads)
 463                 return false;
 464         /* Verify we have no conflict with port offloads */
 465         if ((port_offloads ^ offloads) & port_supp_offloads)
 466                 return false;
 467         return true;
 468 }
 469
 470 static void
 471 fs_tx_queue_release(void *queue)
 472 {
 473         struct rte_eth_dev *dev;
 474         struct sub_device *sdev;
 475         uint8_t i;
 476         struct txq *txq;
 477
 478         if (queue == NULL)
 479                 return;
 480         txq = queue;
 481         dev = txq->priv->dev;
 482         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
 483                 SUBOPS(sdev, tx_queue_release)
 484                         (ETH(sdev)->data->tx_queues[txq->qid]);
 485         dev->data->tx_queues[txq->qid] = NULL;
 486         rte_free(txq);
 487 }
 488
 489 static int
 490 fs_tx_queue_setup(struct rte_eth_dev *dev,
 491                 uint16_t tx_queue_id,
 492                 uint16_t nb_tx_desc,
 493                 unsigned int socket_id,
 494                 const struct rte_eth_txconf *tx_conf)
 495 {
 496         struct sub_device *sdev;
 497         struct txq *txq;
 498         uint8_t i;
 499         int ret;
 500
 501         txq = dev->data->tx_queues[tx_queue_id];
 502         if (txq != NULL) {
 503                 fs_tx_queue_release(txq);
 504                 dev->data->tx_queues[tx_queue_id] = NULL;
 505         }
 506         /*
 507          * Don't verify queue offloads for applications which
 508          * use the old API.
 509          */
 510         if (tx_conf != NULL &&
 511             (tx_conf->txq_flags & ETH_TXQ_FLAGS_IGNORE) &&
 512             fs_txq_offloads_valid(dev, tx_conf->offloads) == false) {
 513                 rte_errno = ENOTSUP;
 514                 ERROR("Tx queue offloads 0x%" PRIx64
 515                       " don't match port offloads 0x%" PRIx64
 516                       " or supported offloads 0x%" PRIx64,
 517                       tx_conf->offloads,
 518                       dev->data->dev_conf.txmode.offloads,
 519                       PRIV(dev)->infos.tx_offload_capa |
 520                       PRIV(dev)->infos.tx_queue_offload_capa);
 521                 return -rte_errno;
 522         }
 523         txq = rte_zmalloc("ethdev TX queue",
 524                           sizeof(*txq) +
 525                           sizeof(rte_atomic64_t) * PRIV(dev)->subs_tail,
 526                           RTE_CACHE_LINE_SIZE);
 527         if (txq == NULL)
 528                 return -ENOMEM;
 529         FOREACH_SUBDEV(sdev, i, dev)
 530                 rte_atomic64_init(&txq->refcnt[i]);
 531         txq->qid = tx_queue_id;
 532         txq->socket_id = socket_id;
 533         txq->info.conf = *tx_conf;
 534         txq->info.nb_desc = nb_tx_desc;
 535         txq->priv = PRIV(dev);
 536         dev->data->tx_queues[tx_queue_id] = txq;
 537         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 538                 ret = rte_eth_tx_queue_setup(PORT_ID(sdev),
 539                                 tx_queue_id,
 540                                 nb_tx_desc, socket_id,
 541                                 tx_conf);
 542                 if ((ret = fs_err(sdev, ret))) {
 543                         ERROR("TX queue setup failed for sub_device %d", i);
 544                         goto free_txq;
 545                 }
 546         }
 547         return 0;
 548 free_txq:
 549         fs_tx_queue_release(txq);
 550         return ret;
 551 }
 552
 553 static void
 554 fs_dev_free_queues(struct rte_eth_dev *dev)
 555 {
 556         uint16_t i;
 557
 558         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 559                 fs_rx_queue_release(dev->data->rx_queues[i]);
 560                 dev->data->rx_queues[i] = NULL;
 561         }
 562         dev->data->nb_rx_queues = 0;
 563         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 564                 fs_tx_queue_release(dev->data->tx_queues[i]);
 565                 dev->data->tx_queues[i] = NULL;
 566         }
 567         dev->data->nb_tx_queues = 0;
 568 }
 569
 570 static void
 571 fs_promiscuous_enable(struct rte_eth_dev *dev)
 572 {
 573         struct sub_device *sdev;
 574         uint8_t i;
 575
 576         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
 577                 rte_eth_promiscuous_enable(PORT_ID(sdev));
 578 }
 579
 580 static void
 581 fs_promiscuous_disable(struct rte_eth_dev *dev)
 582 {
 583         struct sub_device *sdev;
 584         uint8_t i;
 585
 586         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
 587                 rte_eth_promiscuous_disable(PORT_ID(sdev));
 588 }
 589
 590 static void
 591 fs_allmulticast_enable(struct rte_eth_dev *dev)
 592 {
 593         struct sub_device *sdev;
 594         uint8_t i;
 595
 596         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
 597                 rte_eth_allmulticast_enable(PORT_ID(sdev));
 598 }
 599
 600 static void
 601 fs_allmulticast_disable(struct rte_eth_dev *dev)
 602 {
 603         struct sub_device *sdev;
 604         uint8_t i;
 605
 606         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
 607                 rte_eth_allmulticast_disable(PORT_ID(sdev));
 608 }
 609
 610 static int
 611 fs_link_update(struct rte_eth_dev *dev,
 612                 int wait_to_complete)
 613 {
 614         struct sub_device *sdev;
 615         uint8_t i;
 616         int ret;
 617
 618         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 619                 DEBUG("Calling link_update on sub_device %d", i);
 620                 ret = (SUBOPS(sdev, link_update))(ETH(sdev), wait_to_complete);
 621                 if (ret && ret != -1 && sdev->remove == 0 &&
 622                     rte_eth_dev_is_removed(PORT_ID(sdev)) == 0) {
 623                         ERROR("Link update failed for sub_device %d with error %d",
 624                               i, ret);
 625                         return ret;
 626                 }
 627         }
 628         if (TX_SUBDEV(dev)) {
 629                 struct rte_eth_link *l1;
 630                 struct rte_eth_link *l2;
 631
 632                 l1 = &dev->data->dev_link;
 633                 l2 = &ETH(TX_SUBDEV(dev))->data->dev_link;
 634                 if (memcmp(l1, l2, sizeof(*l1))) {
 635                         *l1 = *l2;
 636                         return 0;
 637                 }
 638         }
 639         return -1;
 640 }
 641
 642 static int
 643 fs_stats_get(struct rte_eth_dev *dev,
 644              struct rte_eth_stats *stats)
 645 {
 646         struct rte_eth_stats backup;
 647         struct sub_device *sdev;
 648         uint8_t i;
 649         int ret;
 650
 651         rte_memcpy(stats, &PRIV(dev)->stats_accumulator, sizeof(*stats));
 652         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 653                 struct rte_eth_stats *snapshot = &sdev->stats_snapshot.stats;
 654                 uint64_t *timestamp = &sdev->stats_snapshot.timestamp;
 655
 656                 rte_memcpy(&backup, snapshot, sizeof(backup));
 657                 ret = rte_eth_stats_get(PORT_ID(sdev), snapshot);
 658                 if (ret) {
 659                         if (!fs_err(sdev, ret)) {
 660                                 rte_memcpy(snapshot, &backup, sizeof(backup));
 661                                 goto inc;
 662                         }
 663                         ERROR("Operation rte_eth_stats_get failed for sub_device %d with error %d",
 664                                   i, ret);
 665                         *timestamp = 0;
 666                         return ret;
 667                 }
 668                 *timestamp = rte_rdtsc();
 669 inc:
 670                 failsafe_stats_increment(stats, snapshot);
 671         }
 672         return 0;
 673 }
 674
 675 static void
 676 fs_stats_reset(struct rte_eth_dev *dev)
 677 {
 678         struct sub_device *sdev;
 679         uint8_t i;
 680
 681         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 682                 rte_eth_stats_reset(PORT_ID(sdev));
 683                 memset(&sdev->stats_snapshot, 0, sizeof(struct rte_eth_stats));
 684         }
 685         memset(&PRIV(dev)->stats_accumulator, 0, sizeof(struct rte_eth_stats));
 686 }
 687
 688 /**
 689  * Fail-safe dev_infos_get rules:
 690  *
 691  * No sub_device:
 692  *   Numerables:
 693  *      Use the maximum possible values for any field, so as not
 694  *      to impede any further configuration effort.
 695  *   Capabilities:
 696  *      Limits capabilities to those that are understood by the
 697  *      fail-safe PMD. This understanding stems from the fail-safe
 698  *      being capable of verifying that the related capability is
 699  *      expressed within the device configuration (struct rte_eth_conf).
 700  *
 701  * At least one probed sub_device:
 702  *   Numerables:
 703  *      Uses values from the active probed sub_device
 704  *      The rationale here is that if any sub_device is less capable
 705  *      (for example concerning the number of queues) than the active
 706  *      sub_device, then its subsequent configuration will fail.
 707  *      It is impossible to foresee this failure when the failing sub_device
 708  *      is supposed to be plugged-in later on, so the configuration process
 709  *      is the single point of failure and error reporting.
 710  *   Capabilities:
 711  *      Uses a logical AND of RX capabilities among
 712  *      all sub_devices and the default capabilities.
 713  *      Uses a logical AND of TX capabilities among
 714  *      the active probed sub_device and the default capabilities.
 715  *
 716  */
 717 static void
 718 fs_dev_infos_get(struct rte_eth_dev *dev,
 719                   struct rte_eth_dev_info *infos)
 720 {
 721         struct sub_device *sdev;
 722         uint8_t i;
 723
 724         sdev = TX_SUBDEV(dev);
 725         if (sdev == NULL) {
 726                 DEBUG("No probed device, using default infos");
 727                 rte_memcpy(&PRIV(dev)->infos, &default_infos,
 728                            sizeof(default_infos));
 729         } else {
 730                 uint64_t rx_offload_capa;
 731                 uint64_t rxq_offload_capa;
 732
 733                 rx_offload_capa = default_infos.rx_offload_capa;
 734                 rxq_offload_capa = default_infos.rx_queue_offload_capa;
 735                 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_PROBED) {
 736                         rte_eth_dev_info_get(PORT_ID(sdev),
 737                                         &PRIV(dev)->infos);
 738                         rx_offload_capa &= PRIV(dev)->infos.rx_offload_capa;
 739                         rxq_offload_capa &=
 740                                         PRIV(dev)->infos.rx_queue_offload_capa;
 741                 }
 742                 sdev = TX_SUBDEV(dev);
 743                 rte_eth_dev_info_get(PORT_ID(sdev), &PRIV(dev)->infos);
 744                 PRIV(dev)->infos.rx_offload_capa = rx_offload_capa;
 745                 PRIV(dev)->infos.rx_queue_offload_capa = rxq_offload_capa;
 746                 PRIV(dev)->infos.tx_offload_capa &=
 747                                         default_infos.tx_offload_capa;
 748                 PRIV(dev)->infos.tx_queue_offload_capa &=
 749                                         default_infos.tx_queue_offload_capa;
 750                 PRIV(dev)->infos.flow_type_rss_offloads &=
 751                                         default_infos.flow_type_rss_offloads;
 752         }
 753         rte_memcpy(infos, &PRIV(dev)->infos, sizeof(*infos));
 754 }
 755
 756 static const uint32_t *
 757 fs_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 758 {
 759         struct sub_device *sdev;
 760         struct rte_eth_dev *edev;
 761
 762         sdev = TX_SUBDEV(dev);
 763         if (sdev == NULL)
 764                 return NULL;
 765         edev = ETH(sdev);
 766         /* ENOTSUP: counts as no supported ptypes */
 767         if (SUBOPS(sdev, dev_supported_ptypes_get) == NULL)
 768                 return NULL;
 769         /*
 770          * The API does not permit to do a clean AND of all ptypes,
 771          * It is also incomplete by design and we do not really care
 772          * to have a best possible value in this context.
 773          * We just return the ptypes of the device of highest
 774          * priority, usually the PREFERRED device.
 775          */
 776         return SUBOPS(sdev, dev_supported_ptypes_get)(edev);
 777 }
 778
 779 static int
 780 fs_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 781 {
 782         struct sub_device *sdev;
 783         uint8_t i;
 784         int ret;
 785
 786         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 787                 DEBUG("Calling rte_eth_dev_set_mtu on sub_device %d", i);
 788                 ret = rte_eth_dev_set_mtu(PORT_ID(sdev), mtu);
 789                 if ((ret = fs_err(sdev, ret))) {
 790                         ERROR("Operation rte_eth_dev_set_mtu failed for sub_device %d with error %d",
 791                               i, ret);
 792                         return ret;
 793                 }
 794         }
 795         return 0;
 796 }
 797
 798 static int
 799 fs_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
 800 {
 801         struct sub_device *sdev;
 802         uint8_t i;
 803         int ret;
 804
 805         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 806                 DEBUG("Calling rte_eth_dev_vlan_filter on sub_device %d", i);
 807                 ret = rte_eth_dev_vlan_filter(PORT_ID(sdev), vlan_id, on);
 808                 if ((ret = fs_err(sdev, ret))) {
 809                         ERROR("Operation rte_eth_dev_vlan_filter failed for sub_device %d"
 810                               " with error %d", i, ret);
 811                         return ret;
 812                 }
 813         }
 814         return 0;
 815 }
 816
 817 static int
 818 fs_flow_ctrl_get(struct rte_eth_dev *dev,
 819                 struct rte_eth_fc_conf *fc_conf)
 820 {
 821         struct sub_device *sdev;
 822
 823         sdev = TX_SUBDEV(dev);
 824         if (sdev == NULL)
 825                 return 0;
 826         if (SUBOPS(sdev, flow_ctrl_get) == NULL)
 827                 return -ENOTSUP;
 828         return SUBOPS(sdev, flow_ctrl_get)(ETH(sdev), fc_conf);
 829 }
 830
 831 static int
 832 fs_flow_ctrl_set(struct rte_eth_dev *dev,
 833                 struct rte_eth_fc_conf *fc_conf)
 834 {
 835         struct sub_device *sdev;
 836         uint8_t i;
 837         int ret;
 838
 839         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 840                 DEBUG("Calling rte_eth_dev_flow_ctrl_set on sub_device %d", i);
 841                 ret = rte_eth_dev_flow_ctrl_set(PORT_ID(sdev), fc_conf);
 842                 if ((ret = fs_err(sdev, ret))) {
 843                         ERROR("Operation rte_eth_dev_flow_ctrl_set failed for sub_device %d"
 844                               " with error %d", i, ret);
 845                         return ret;
 846                 }
 847         }
 848         return 0;
 849 }
 850
 851 static void
 852 fs_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
 853 {
 854         struct sub_device *sdev;
 855         uint8_t i;
 856
 857         /* No check: already done within the rte_eth_dev_mac_addr_remove
 858          * call for the fail-safe device.
 859          */
 860         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
 861                 rte_eth_dev_mac_addr_remove(PORT_ID(sdev),
 862                                 &dev->data->mac_addrs[index]);
 863         PRIV(dev)->mac_addr_pool[index] = 0;
 864 }
 865
 866 static int
 867 fs_mac_addr_add(struct rte_eth_dev *dev,
 868                 struct ether_addr *mac_addr,
 869                 uint32_t index,
 870                 uint32_t vmdq)
 871 {
 872         struct sub_device *sdev;
 873         int ret;
 874         uint8_t i;
 875
 876         RTE_ASSERT(index < FAILSAFE_MAX_ETHADDR);
 877         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 878                 ret = rte_eth_dev_mac_addr_add(PORT_ID(sdev), mac_addr, vmdq);
 879                 if ((ret = fs_err(sdev, ret))) {
 880                         ERROR("Operation rte_eth_dev_mac_addr_add failed for sub_device %"
 881                               PRIu8 " with error %d", i, ret);
 882                         return ret;
 883                 }
 884         }
 885         if (index >= PRIV(dev)->nb_mac_addr) {
 886                 DEBUG("Growing mac_addrs array");
 887                 PRIV(dev)->nb_mac_addr = index;
 888         }
 889         PRIV(dev)->mac_addr_pool[index] = vmdq;
 890         return 0;
 891 }
 892
 893 static void
 894 fs_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 895 {
 896         struct sub_device *sdev;
 897         uint8_t i;
 898
 899         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE)
 900                 rte_eth_dev_default_mac_addr_set(PORT_ID(sdev), mac_addr);
 901 }
 902
 903 static int
 904 fs_filter_ctrl(struct rte_eth_dev *dev,
 905                 enum rte_filter_type type,
 906                 enum rte_filter_op op,
 907                 void *arg)
 908 {
 909         struct sub_device *sdev;
 910         uint8_t i;
 911         int ret;
 912
 913         if (type == RTE_ETH_FILTER_GENERIC &&
 914             op == RTE_ETH_FILTER_GET) {
 915                 *(const void **)arg = &fs_flow_ops;
 916                 return 0;
 917         }
 918         FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_ACTIVE) {
 919                 DEBUG("Calling rte_eth_dev_filter_ctrl on sub_device %d", i);
 920                 ret = rte_eth_dev_filter_ctrl(PORT_ID(sdev), type, op, arg);
 921                 if ((ret = fs_err(sdev, ret))) {
 922                         ERROR("Operation rte_eth_dev_filter_ctrl failed for sub_device %d"
 923                               " with error %d", i, ret);
 924                         return ret;
 925                 }
 926         }
 927         return 0;
 928 }
 929
 930 const struct eth_dev_ops failsafe_ops = {
 931         .dev_configure = fs_dev_configure,
 932         .dev_start = fs_dev_start,
 933         .dev_stop = fs_dev_stop,
 934         .dev_set_link_down = fs_dev_set_link_down,
 935         .dev_set_link_up = fs_dev_set_link_up,
 936         .dev_close = fs_dev_close,
 937         .promiscuous_enable = fs_promiscuous_enable,
 938         .promiscuous_disable = fs_promiscuous_disable,
 939         .allmulticast_enable = fs_allmulticast_enable,
 940         .allmulticast_disable = fs_allmulticast_disable,
 941         .link_update = fs_link_update,
 942         .stats_get = fs_stats_get,
 943         .stats_reset = fs_stats_reset,
 944         .dev_infos_get = fs_dev_infos_get,
 945         .dev_supported_ptypes_get = fs_dev_supported_ptypes_get,
 946         .mtu_set = fs_mtu_set,
 947         .vlan_filter_set = fs_vlan_filter_set,
 948         .rx_queue_setup = fs_rx_queue_setup,
 949         .tx_queue_setup = fs_tx_queue_setup,
 950         .rx_queue_release = fs_rx_queue_release,
 951         .tx_queue_release = fs_tx_queue_release,
 952         .rx_queue_intr_enable = fs_rx_intr_enable,
 953         .rx_queue_intr_disable = fs_rx_intr_disable,
 954         .flow_ctrl_get = fs_flow_ctrl_get,
 955         .flow_ctrl_set = fs_flow_ctrl_set,
 956         .mac_addr_remove = fs_mac_addr_remove,
 957         .mac_addr_add = fs_mac_addr_add,
 958         .mac_addr_set = fs_mac_addr_set,
 959         .filter_ctrl = fs_filter_ctrl,
 960 };