drivers/net/failsafe/failsafe.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright 2017 6WIND S.A.
   3  * Copyright 2017 Mellanox Technologies, Ltd
   4  */
   5
   6 #include <stdbool.h>
   7
   8 #include <rte_alarm.h>
   9 #include <rte_malloc.h>
  10 #include <rte_ethdev_driver.h>
  11 #include <rte_ethdev_vdev.h>
  12 #include <rte_devargs.h>
  13 #include <rte_kvargs.h>
  14 #include <rte_bus_vdev.h>
  15
  16 #include "failsafe_private.h"
  17
  18 const char pmd_failsafe_driver_name[] = FAILSAFE_DRIVER_NAME;
  19 static const struct rte_eth_link eth_link = {
  20         .link_speed = ETH_SPEED_NUM_10G,
  21         .link_duplex = ETH_LINK_FULL_DUPLEX,
  22         .link_status = ETH_LINK_UP,
  23         .link_autoneg = ETH_LINK_AUTONEG,
  24 };
  25
  26 static int
  27 fs_sub_device_alloc(struct rte_eth_dev *dev,
  28                 const char *params)
  29 {
  30         uint8_t nb_subs;
  31         int ret;
  32         int i;
  33         struct sub_device *sdev;
  34         uint8_t sdev_iterator;
  35
  36         ret = failsafe_args_count_subdevice(dev, params);
  37         if (ret)
  38                 return ret;
  39         if (PRIV(dev)->subs_tail > FAILSAFE_MAX_ETHPORTS) {
  40                 ERROR("Cannot allocate more than %d ports",
  41                         FAILSAFE_MAX_ETHPORTS);
  42                 return -ENOSPC;
  43         }
  44         nb_subs = PRIV(dev)->subs_tail;
  45         PRIV(dev)->subs = rte_zmalloc(NULL,
  46                         sizeof(struct sub_device) * nb_subs,
  47                         RTE_CACHE_LINE_SIZE);
  48         if (PRIV(dev)->subs == NULL) {
  49                 ERROR("Could not allocate sub_devices");
  50                 return -ENOMEM;
  51         }
  52         /* Initiate static sub devices linked list. */
  53         for (i = 1; i < nb_subs; i++)
  54                 PRIV(dev)->subs[i - 1].next = PRIV(dev)->subs + i;
  55         PRIV(dev)->subs[i - 1].next = PRIV(dev)->subs;
  56
  57         FOREACH_SUBDEV(sdev, sdev_iterator, dev) {
  58                 sdev->sdev_port_id = RTE_MAX_ETHPORTS;
  59         }
  60         return 0;
  61 }
  62
  63 static void
  64 fs_sub_device_free(struct rte_eth_dev *dev)
  65 {
  66         rte_free(PRIV(dev)->subs);
  67 }
  68
  69 static void fs_hotplug_alarm(void *arg);
  70
  71 int
  72 failsafe_hotplug_alarm_install(struct rte_eth_dev *dev)
  73 {
  74         int ret;
  75
  76         if (dev == NULL)
  77                 return -EINVAL;
  78         if (PRIV(dev)->pending_alarm)
  79                 return 0;
  80         ret = rte_eal_alarm_set(failsafe_hotplug_poll * 1000,
  81                                 fs_hotplug_alarm,
  82                                 dev);
  83         if (ret) {
  84                 ERROR("Could not set up plug-in event detection");
  85                 return ret;
  86         }
  87         PRIV(dev)->pending_alarm = 1;
  88         return 0;
  89 }
  90
  91 int
  92 failsafe_hotplug_alarm_cancel(struct rte_eth_dev *dev)
  93 {
  94         int ret = 0;
  95
  96         rte_errno = 0;
  97         rte_eal_alarm_cancel(fs_hotplug_alarm, dev);
  98         if (rte_errno) {
  99                 ERROR("rte_eal_alarm_cancel failed (errno: %s)",
 100                       strerror(rte_errno));
 101                 ret = -rte_errno;
 102         } else {
 103                 PRIV(dev)->pending_alarm = 0;
 104         }
 105         return ret;
 106 }
 107
 108 static void
 109 fs_hotplug_alarm(void *arg)
 110 {
 111         struct rte_eth_dev *dev = arg;
 112         struct sub_device *sdev;
 113         int ret;
 114         uint8_t i;
 115
 116         if (!PRIV(dev)->pending_alarm)
 117                 return;
 118         PRIV(dev)->pending_alarm = 0;
 119         FOREACH_SUBDEV(sdev, i, dev)
 120                 if (sdev->state != PRIV(dev)->state)
 121                         break;
 122         /* if we have non-probed device */
 123         if (i != PRIV(dev)->subs_tail) {
 124                 if (fs_lock(dev, 1) != 0)
 125                         goto reinstall;
 126                 ret = failsafe_eth_dev_state_sync(dev);
 127                 fs_unlock(dev, 1);
 128                 if (ret)
 129                         ERROR("Unable to synchronize sub_device state");
 130         }
 131         failsafe_dev_remove(dev);
 132 reinstall:
 133         ret = failsafe_hotplug_alarm_install(dev);
 134         if (ret)
 135                 ERROR("Unable to set up next alarm");
 136 }
 137
 138 static int
 139 fs_mutex_init(struct fs_priv *priv)
 140 {
 141         int ret;
 142         pthread_mutexattr_t attr;
 143
 144         ret = pthread_mutexattr_init(&attr);
 145         if (ret) {
 146                 ERROR("Cannot initiate mutex attributes - %s", strerror(ret));
 147                 return ret;
 148         }
 149         /* Allow mutex relocks for the thread holding the mutex. */
 150         ret = pthread_mutexattr_settype(&attr, PTHREAD_MUTEX_RECURSIVE);
 151         if (ret) {
 152                 ERROR("Cannot set mutex type - %s", strerror(ret));
 153                 return ret;
 154         }
 155         ret = pthread_mutex_init(&priv->hotplug_mutex, &attr);
 156         if (ret) {
 157                 ERROR("Cannot initiate mutex - %s", strerror(ret));
 158                 return ret;
 159         }
 160         return 0;
 161 }
 162
 163 static int
 164 fs_eth_dev_create(struct rte_vdev_device *vdev)
 165 {
 166         struct rte_eth_dev *dev;
 167         struct rte_ether_addr *mac;
 168         struct fs_priv *priv;
 169         struct sub_device *sdev;
 170         const char *params;
 171         unsigned int socket_id;
 172         uint8_t i;
 173         int ret;
 174
 175         dev = NULL;
 176         priv = NULL;
 177         socket_id = rte_socket_id();
 178         INFO("Creating fail-safe device on NUMA socket %u", socket_id);
 179         params = rte_vdev_device_args(vdev);
 180         if (params == NULL) {
 181                 ERROR("This PMD requires sub-devices, none provided");
 182                 return -1;
 183         }
 184         dev = rte_eth_vdev_allocate(vdev, sizeof(*priv));
 185         if (dev == NULL) {
 186                 ERROR("Unable to allocate rte_eth_dev");
 187                 return -1;
 188         }
 189         priv = PRIV(dev);
 190         priv->data = dev->data;
 191         priv->rxp = FS_RX_PROXY_INIT;
 192         dev->dev_ops = &failsafe_ops;
 193         dev->data->mac_addrs = &PRIV(dev)->mac_addrs[0];
 194         dev->data->dev_link = eth_link;
 195         PRIV(dev)->nb_mac_addr = 1;
 196         TAILQ_INIT(&PRIV(dev)->flow_list);
 197         dev->rx_pkt_burst = (eth_rx_burst_t)&failsafe_rx_burst;
 198         dev->tx_pkt_burst = (eth_tx_burst_t)&failsafe_tx_burst;
 199         ret = fs_sub_device_alloc(dev, params);
 200         if (ret) {
 201                 ERROR("Could not allocate sub_devices");
 202                 goto free_dev;
 203         }
 204         ret = failsafe_args_parse(dev, params);
 205         if (ret)
 206                 goto free_subs;
 207         ret = rte_eth_dev_owner_new(&priv->my_owner.id);
 208         if (ret) {
 209                 ERROR("Failed to get unique owner identifier");
 210                 goto free_args;
 211         }
 212         snprintf(priv->my_owner.name, sizeof(priv->my_owner.name),
 213                  FAILSAFE_OWNER_NAME);
 214         DEBUG("Failsafe port %u owner info: %s_%016"PRIX64, dev->data->port_id,
 215               priv->my_owner.name, priv->my_owner.id);
 216         ret = rte_eth_dev_callback_register(RTE_ETH_ALL, RTE_ETH_EVENT_NEW,
 217                                             failsafe_eth_new_event_callback,
 218                                             dev);
 219         if (ret) {
 220                 ERROR("Failed to register NEW callback");
 221                 goto free_args;
 222         }
 223         ret = failsafe_eal_init(dev);
 224         if (ret)
 225                 goto unregister_new_callback;
 226         ret = fs_mutex_init(priv);
 227         if (ret)
 228                 goto unregister_new_callback;
 229         ret = failsafe_hotplug_alarm_install(dev);
 230         if (ret) {
 231                 ERROR("Could not set up plug-in event detection");
 232                 goto unregister_new_callback;
 233         }
 234         mac = &dev->data->mac_addrs[0];
 235         if (failsafe_mac_from_arg) {
 236                 /*
 237                  * If MAC address was provided as a parameter,
 238                  * apply to all probed slaves.
 239                  */
 240                 FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_PROBED) {
 241                         ret = rte_eth_dev_default_mac_addr_set(PORT_ID(sdev),
 242                                                                mac);
 243                         if (ret) {
 244                                 ERROR("Failed to set default MAC address");
 245                                 goto cancel_alarm;
 246                         }
 247                 }
 248         } else {
 249                 /*
 250                  * Use the ether_addr from first probed
 251                  * device, either preferred or fallback.
 252                  */
 253                 FOREACH_SUBDEV(sdev, i, dev)
 254                         if (sdev->state >= DEV_PROBED) {
 255                                 rte_ether_addr_copy(
 256                                         &ETH(sdev)->data->mac_addrs[0], mac);
 257                                 break;
 258                         }
 259                 /*
 260                  * If no device has been probed and no ether_addr
 261                  * has been provided on the command line, use a random
 262                  * valid one.
 263                  * It will be applied during future slave state syncs to
 264                  * probed slaves.
 265                  */
 266                 if (i == priv->subs_tail)
 267                         rte_eth_random_addr(&mac->addr_bytes[0]);
 268         }
 269         INFO("MAC address is %02x:%02x:%02x:%02x:%02x:%02x",
 270                 mac->addr_bytes[0], mac->addr_bytes[1],
 271                 mac->addr_bytes[2], mac->addr_bytes[3],
 272                 mac->addr_bytes[4], mac->addr_bytes[5]);
 273         dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
 274         PRIV(dev)->intr_handle = (struct rte_intr_handle){
 275                 .fd = -1,
 276                 .type = RTE_INTR_HANDLE_EXT,
 277         };
 278         rte_eth_dev_probing_finish(dev);
 279         return 0;
 280 cancel_alarm:
 281         failsafe_hotplug_alarm_cancel(dev);
 282 unregister_new_callback:
 283         rte_eth_dev_callback_unregister(RTE_ETH_ALL, RTE_ETH_EVENT_NEW,
 284                                         failsafe_eth_new_event_callback, dev);
 285 free_args:
 286         failsafe_args_free(dev);
 287 free_subs:
 288         fs_sub_device_free(dev);
 289 free_dev:
 290         /* mac_addrs must not be freed alone because part of dev_private */
 291         dev->data->mac_addrs = NULL;
 292         rte_eth_dev_release_port(dev);
 293         return -1;
 294 }
 295
 296 static int
 297 fs_rte_eth_free(const char *name)
 298 {
 299         struct rte_eth_dev *dev;
 300         int ret;
 301
 302         dev = rte_eth_dev_allocated(name);
 303         if (dev == NULL)
 304                 return -ENODEV;
 305         rte_eth_dev_callback_unregister(RTE_ETH_ALL, RTE_ETH_EVENT_NEW,
 306                                         failsafe_eth_new_event_callback, dev);
 307         ret = failsafe_eal_uninit(dev);
 308         if (ret)
 309                 ERROR("Error while uninitializing sub-EAL");
 310         failsafe_args_free(dev);
 311         fs_sub_device_free(dev);
 312         ret = pthread_mutex_destroy(&PRIV(dev)->hotplug_mutex);
 313         if (ret)
 314                 ERROR("Error while destroying hotplug mutex");
 315         rte_free(PRIV(dev)->mcast_addrs);
 316         /* mac_addrs must not be freed alone because part of dev_private */
 317         dev->data->mac_addrs = NULL;
 318         rte_eth_dev_release_port(dev);
 319         return ret;
 320 }
 321
 322 static bool
 323 devargs_already_listed(struct rte_devargs *devargs)
 324 {
 325         struct rte_devargs *list_da;
 326
 327         RTE_EAL_DEVARGS_FOREACH(devargs->bus->name, list_da) {
 328                 if (strcmp(list_da->name, devargs->name) == 0)
 329                         /* devargs already in the list */
 330                         return true;
 331         }
 332         return false;
 333 }
 334
 335 static int
 336 rte_pmd_failsafe_probe(struct rte_vdev_device *vdev)
 337 {
 338         const char *name;
 339         struct rte_eth_dev *eth_dev;
 340         struct sub_device  *sdev;
 341         struct rte_devargs devargs;
 342         uint8_t i;
 343         int ret;
 344
 345         name = rte_vdev_device_name(vdev);
 346         INFO("Initializing " FAILSAFE_DRIVER_NAME " for %s",
 347                         name);
 348
 349         if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
 350             strlen(rte_vdev_device_args(vdev)) == 0) {
 351                 eth_dev = rte_eth_dev_attach_secondary(name);
 352                 if (!eth_dev) {
 353                         ERROR("Failed to probe %s", name);
 354                         return -1;
 355                 }
 356                 eth_dev->dev_ops = &failsafe_ops;
 357                 eth_dev->device = &vdev->device;
 358                 eth_dev->rx_pkt_burst = (eth_rx_burst_t)&failsafe_rx_burst;
 359                 eth_dev->tx_pkt_burst = (eth_tx_burst_t)&failsafe_tx_burst;
 360                 /*
 361                  * Failsafe will attempt to probe all of its sub-devices.
 362                  * Any failure in sub-devices is not a fatal error.
 363                  * A sub-device can be plugged later.
 364                  */
 365                 FOREACH_SUBDEV(sdev, i, eth_dev) {
 366                         /* skip empty devargs */
 367                         if (sdev->devargs.name[0] == '\0')
 368                                 continue;
 369
 370                         /* rebuild devargs to be able to get the bus name. */
 371                         ret = rte_devargs_parse(&devargs,
 372                                                 sdev->devargs.name);
 373                         if (ret != 0) {
 374                                 ERROR("Failed to parse devargs %s",
 375                                         devargs.name);
 376                                 continue;
 377                         }
 378                         if (!devargs_already_listed(&devargs)) {
 379                                 ret = rte_dev_probe(devargs.name);
 380                                 if (ret < 0) {
 381                                         ERROR("Failed to probe devargs %s",
 382                                               devargs.name);
 383                                         continue;
 384                                 }
 385                         }
 386                 }
 387                 rte_eth_dev_probing_finish(eth_dev);
 388                 return 0;
 389         }
 390
 391         return fs_eth_dev_create(vdev);
 392 }
 393
 394 static int
 395 rte_pmd_failsafe_remove(struct rte_vdev_device *vdev)
 396 {
 397         const char *name;
 398
 399         name = rte_vdev_device_name(vdev);
 400         INFO("Uninitializing " FAILSAFE_DRIVER_NAME " for %s", name);
 401         return fs_rte_eth_free(name);
 402 }
 403
 404 static struct rte_vdev_driver failsafe_drv = {
 405         .probe = rte_pmd_failsafe_probe,
 406         .remove = rte_pmd_failsafe_remove,
 407 };
 408
 409 RTE_PMD_REGISTER_VDEV(net_failsafe, failsafe_drv);
 410 RTE_PMD_REGISTER_PARAM_STRING(net_failsafe, PMD_FAILSAFE_PARAM_STRING);
 411 RTE_LOG_REGISTER(failsafe_logtype, pmd.net.failsafe, NOTICE)