ceph/src/seastar/dpdk/drivers/net/kni/rte_eth_kni.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2017 Intel Corporation
   3  */
   4
   5 #include <fcntl.h>
   6 #include <pthread.h>
   7 #include <unistd.h>
   8
   9 #include <rte_string_fns.h>
  10 #include <rte_ethdev_driver.h>
  11 #include <rte_ethdev_vdev.h>
  12 #include <rte_kni.h>
  13 #include <rte_kvargs.h>
  14 #include <rte_malloc.h>
  15 #include <rte_bus_vdev.h>
  16
  17 /* Only single queue supported */
  18 #define KNI_MAX_QUEUE_PER_PORT 1
  19
  20 #define MAX_KNI_PORTS 8
  21
  22 #define KNI_ETHER_MTU(mbuf_size)       \
  23         ((mbuf_size) - ETHER_HDR_LEN) /**< Ethernet MTU. */
  24
  25 #define ETH_KNI_NO_REQUEST_THREAD_ARG   "no_request_thread"
  26 static const char * const valid_arguments[] = {
  27         ETH_KNI_NO_REQUEST_THREAD_ARG,
  28         NULL
  29 };
  30
  31 struct eth_kni_args {
  32         int no_request_thread;
  33 };
  34
  35 struct pmd_queue_stats {
  36         uint64_t pkts;
  37         uint64_t bytes;
  38         uint64_t err_pkts;
  39 };
  40
  41 struct pmd_queue {
  42         struct pmd_internals *internals;
  43         struct rte_mempool *mb_pool;
  44
  45         struct pmd_queue_stats rx;
  46         struct pmd_queue_stats tx;
  47 };
  48
  49 struct pmd_internals {
  50         struct rte_kni *kni;
  51         int is_kni_started;
  52
  53         pthread_t thread;
  54         int stop_thread;
  55         int no_request_thread;
  56
  57         struct ether_addr eth_addr;
  58
  59         struct pmd_queue rx_queues[KNI_MAX_QUEUE_PER_PORT];
  60         struct pmd_queue tx_queues[KNI_MAX_QUEUE_PER_PORT];
  61 };
  62
  63 static const struct rte_eth_link pmd_link = {
  64                 .link_speed = ETH_SPEED_NUM_10G,
  65                 .link_duplex = ETH_LINK_FULL_DUPLEX,
  66                 .link_status = ETH_LINK_DOWN,
  67                 .link_autoneg = ETH_LINK_FIXED,
  68 };
  69 static int is_kni_initialized;
  70
  71 static int eth_kni_logtype;
  72
  73 #define PMD_LOG(level, fmt, args...) \
  74         rte_log(RTE_LOG_ ## level, eth_kni_logtype, \
  75                 "%s(): " fmt "\n", __func__, ##args)
  76 static uint16_t
  77 eth_kni_rx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
  78 {
  79         struct pmd_queue *kni_q = q;
  80         struct rte_kni *kni = kni_q->internals->kni;
  81         uint16_t nb_pkts;
  82
  83         nb_pkts = rte_kni_rx_burst(kni, bufs, nb_bufs);
  84
  85         kni_q->rx.pkts += nb_pkts;
  86         kni_q->rx.err_pkts += nb_bufs - nb_pkts;
  87
  88         return nb_pkts;
  89 }
  90
  91 static uint16_t
  92 eth_kni_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
  93 {
  94         struct pmd_queue *kni_q = q;
  95         struct rte_kni *kni = kni_q->internals->kni;
  96         uint16_t nb_pkts;
  97
  98         nb_pkts =  rte_kni_tx_burst(kni, bufs, nb_bufs);
  99
 100         kni_q->tx.pkts += nb_pkts;
 101         kni_q->tx.err_pkts += nb_bufs - nb_pkts;
 102
 103         return nb_pkts;
 104 }
 105
 106 static void *
 107 kni_handle_request(void *param)
 108 {
 109         struct pmd_internals *internals = param;
 110 #define MS 1000
 111
 112         while (!internals->stop_thread) {
 113                 rte_kni_handle_request(internals->kni);
 114                 usleep(500 * MS);
 115         }
 116
 117         return param;
 118 }
 119
 120 static int
 121 eth_kni_start(struct rte_eth_dev *dev)
 122 {
 123         struct pmd_internals *internals = dev->data->dev_private;
 124         uint16_t port_id = dev->data->port_id;
 125         struct rte_mempool *mb_pool;
 126         struct rte_kni_conf conf;
 127         const char *name = dev->device->name + 4; /* remove net_ */
 128
 129         mb_pool = internals->rx_queues[0].mb_pool;
 130         strlcpy(conf.name, name, RTE_KNI_NAMESIZE);
 131         conf.force_bind = 0;
 132         conf.group_id = port_id;
 133         conf.mbuf_size =
 134                 rte_pktmbuf_data_room_size(mb_pool) - RTE_PKTMBUF_HEADROOM;
 135         conf.mtu = KNI_ETHER_MTU(conf.mbuf_size);
 136
 137         internals->kni = rte_kni_alloc(mb_pool, &conf, NULL);
 138         if (internals->kni == NULL) {
 139                 PMD_LOG(ERR,
 140                         "Fail to create kni interface for port: %d",
 141                         port_id);
 142                 return -1;
 143         }
 144
 145         return 0;
 146 }
 147
 148 static int
 149 eth_kni_dev_start(struct rte_eth_dev *dev)
 150 {
 151         struct pmd_internals *internals = dev->data->dev_private;
 152         int ret;
 153
 154         if (internals->is_kni_started == 0) {
 155                 ret = eth_kni_start(dev);
 156                 if (ret)
 157                         return -1;
 158                 internals->is_kni_started = 1;
 159         }
 160
 161         if (internals->no_request_thread == 0) {
 162                 ret = rte_ctrl_thread_create(&internals->thread,
 163                         "kni_handle_req", NULL,
 164                         kni_handle_request, internals);
 165                 if (ret) {
 166                         PMD_LOG(ERR,
 167                                 "Fail to create kni request thread");
 168                         return -1;
 169                 }
 170         }
 171
 172         dev->data->dev_link.link_status = 1;
 173
 174         return 0;
 175 }
 176
 177 static void
 178 eth_kni_dev_stop(struct rte_eth_dev *dev)
 179 {
 180         struct pmd_internals *internals = dev->data->dev_private;
 181         int ret;
 182
 183         if (internals->no_request_thread == 0) {
 184                 internals->stop_thread = 1;
 185
 186                 ret = pthread_cancel(internals->thread);
 187                 if (ret)
 188                         PMD_LOG(ERR, "Can't cancel the thread");
 189
 190                 ret = pthread_join(internals->thread, NULL);
 191                 if (ret)
 192                         PMD_LOG(ERR, "Can't join the thread");
 193
 194                 internals->stop_thread = 0;
 195         }
 196
 197         dev->data->dev_link.link_status = 0;
 198 }
 199
 200 static int
 201 eth_kni_dev_configure(struct rte_eth_dev *dev __rte_unused)
 202 {
 203         return 0;
 204 }
 205
 206 static void
 207 eth_kni_dev_info(struct rte_eth_dev *dev __rte_unused,
 208                 struct rte_eth_dev_info *dev_info)
 209 {
 210         dev_info->max_mac_addrs = 1;
 211         dev_info->max_rx_pktlen = UINT32_MAX;
 212         dev_info->max_rx_queues = KNI_MAX_QUEUE_PER_PORT;
 213         dev_info->max_tx_queues = KNI_MAX_QUEUE_PER_PORT;
 214         dev_info->min_rx_bufsize = 0;
 215 }
 216
 217 static int
 218 eth_kni_rx_queue_setup(struct rte_eth_dev *dev,
 219                 uint16_t rx_queue_id,
 220                 uint16_t nb_rx_desc __rte_unused,
 221                 unsigned int socket_id __rte_unused,
 222                 const struct rte_eth_rxconf *rx_conf __rte_unused,
 223                 struct rte_mempool *mb_pool)
 224 {
 225         struct pmd_internals *internals = dev->data->dev_private;
 226         struct pmd_queue *q;
 227
 228         q = &internals->rx_queues[rx_queue_id];
 229         q->internals = internals;
 230         q->mb_pool = mb_pool;
 231
 232         dev->data->rx_queues[rx_queue_id] = q;
 233
 234         return 0;
 235 }
 236
 237 static int
 238 eth_kni_tx_queue_setup(struct rte_eth_dev *dev,
 239                 uint16_t tx_queue_id,
 240                 uint16_t nb_tx_desc __rte_unused,
 241                 unsigned int socket_id __rte_unused,
 242                 const struct rte_eth_txconf *tx_conf __rte_unused)
 243 {
 244         struct pmd_internals *internals = dev->data->dev_private;
 245         struct pmd_queue *q;
 246
 247         q = &internals->tx_queues[tx_queue_id];
 248         q->internals = internals;
 249
 250         dev->data->tx_queues[tx_queue_id] = q;
 251
 252         return 0;
 253 }
 254
 255 static void
 256 eth_kni_queue_release(void *q __rte_unused)
 257 {
 258 }
 259
 260 static int
 261 eth_kni_link_update(struct rte_eth_dev *dev __rte_unused,
 262                 int wait_to_complete __rte_unused)
 263 {
 264         return 0;
 265 }
 266
 267 static int
 268 eth_kni_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 269 {
 270         unsigned long rx_packets_total = 0, rx_bytes_total = 0;
 271         unsigned long tx_packets_total = 0, tx_bytes_total = 0;
 272         struct rte_eth_dev_data *data = dev->data;
 273         unsigned long tx_packets_err_total = 0;
 274         unsigned int i, num_stats;
 275         struct pmd_queue *q;
 276
 277         num_stats = RTE_MIN((unsigned int)RTE_ETHDEV_QUEUE_STAT_CNTRS,
 278                         data->nb_rx_queues);
 279         for (i = 0; i < num_stats; i++) {
 280                 q = data->rx_queues[i];
 281                 stats->q_ipackets[i] = q->rx.pkts;
 282                 stats->q_ibytes[i] = q->rx.bytes;
 283                 rx_packets_total += stats->q_ipackets[i];
 284                 rx_bytes_total += stats->q_ibytes[i];
 285         }
 286
 287         num_stats = RTE_MIN((unsigned int)RTE_ETHDEV_QUEUE_STAT_CNTRS,
 288                         data->nb_tx_queues);
 289         for (i = 0; i < num_stats; i++) {
 290                 q = data->tx_queues[i];
 291                 stats->q_opackets[i] = q->tx.pkts;
 292                 stats->q_obytes[i] = q->tx.bytes;
 293                 stats->q_errors[i] = q->tx.err_pkts;
 294                 tx_packets_total += stats->q_opackets[i];
 295                 tx_bytes_total += stats->q_obytes[i];
 296                 tx_packets_err_total += stats->q_errors[i];
 297         }
 298
 299         stats->ipackets = rx_packets_total;
 300         stats->ibytes = rx_bytes_total;
 301         stats->opackets = tx_packets_total;
 302         stats->obytes = tx_bytes_total;
 303         stats->oerrors = tx_packets_err_total;
 304
 305         return 0;
 306 }
 307
 308 static void
 309 eth_kni_stats_reset(struct rte_eth_dev *dev)
 310 {
 311         struct rte_eth_dev_data *data = dev->data;
 312         struct pmd_queue *q;
 313         unsigned int i;
 314
 315         for (i = 0; i < data->nb_rx_queues; i++) {
 316                 q = data->rx_queues[i];
 317                 q->rx.pkts = 0;
 318                 q->rx.bytes = 0;
 319         }
 320         for (i = 0; i < data->nb_tx_queues; i++) {
 321                 q = data->tx_queues[i];
 322                 q->tx.pkts = 0;
 323                 q->tx.bytes = 0;
 324                 q->tx.err_pkts = 0;
 325         }
 326 }
 327
 328 static const struct eth_dev_ops eth_kni_ops = {
 329         .dev_start = eth_kni_dev_start,
 330         .dev_stop = eth_kni_dev_stop,
 331         .dev_configure = eth_kni_dev_configure,
 332         .dev_infos_get = eth_kni_dev_info,
 333         .rx_queue_setup = eth_kni_rx_queue_setup,
 334         .tx_queue_setup = eth_kni_tx_queue_setup,
 335         .rx_queue_release = eth_kni_queue_release,
 336         .tx_queue_release = eth_kni_queue_release,
 337         .link_update = eth_kni_link_update,
 338         .stats_get = eth_kni_stats_get,
 339         .stats_reset = eth_kni_stats_reset,
 340 };
 341
 342 static struct rte_eth_dev *
 343 eth_kni_create(struct rte_vdev_device *vdev,
 344                 struct eth_kni_args *args,
 345                 unsigned int numa_node)
 346 {
 347         struct pmd_internals *internals;
 348         struct rte_eth_dev_data *data;
 349         struct rte_eth_dev *eth_dev;
 350
 351         PMD_LOG(INFO, "Creating kni ethdev on numa socket %u",
 352                         numa_node);
 353
 354         /* reserve an ethdev entry */
 355         eth_dev = rte_eth_vdev_allocate(vdev, sizeof(*internals));
 356         if (!eth_dev)
 357                 return NULL;
 358
 359         internals = eth_dev->data->dev_private;
 360         data = eth_dev->data;
 361         data->nb_rx_queues = 1;
 362         data->nb_tx_queues = 1;
 363         data->dev_link = pmd_link;
 364         data->mac_addrs = &internals->eth_addr;
 365
 366         eth_random_addr(internals->eth_addr.addr_bytes);
 367
 368         eth_dev->dev_ops = &eth_kni_ops;
 369
 370         internals->no_request_thread = args->no_request_thread;
 371
 372         return eth_dev;
 373 }
 374
 375 static int
 376 kni_init(void)
 377 {
 378         if (is_kni_initialized == 0)
 379                 rte_kni_init(MAX_KNI_PORTS);
 380
 381         is_kni_initialized++;
 382
 383         return 0;
 384 }
 385
 386 static int
 387 eth_kni_kvargs_process(struct eth_kni_args *args, const char *params)
 388 {
 389         struct rte_kvargs *kvlist;
 390
 391         kvlist = rte_kvargs_parse(params, valid_arguments);
 392         if (kvlist == NULL)
 393                 return -1;
 394
 395         memset(args, 0, sizeof(struct eth_kni_args));
 396
 397         if (rte_kvargs_count(kvlist, ETH_KNI_NO_REQUEST_THREAD_ARG) == 1)
 398                 args->no_request_thread = 1;
 399
 400         rte_kvargs_free(kvlist);
 401
 402         return 0;
 403 }
 404
 405 static int
 406 eth_kni_probe(struct rte_vdev_device *vdev)
 407 {
 408         struct rte_eth_dev *eth_dev;
 409         struct eth_kni_args args;
 410         const char *name;
 411         const char *params;
 412         int ret;
 413
 414         name = rte_vdev_device_name(vdev);
 415         params = rte_vdev_device_args(vdev);
 416         PMD_LOG(INFO, "Initializing eth_kni for %s", name);
 417
 418         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
 419                 eth_dev = rte_eth_dev_attach_secondary(name);
 420                 if (!eth_dev) {
 421                         PMD_LOG(ERR, "Failed to probe %s", name);
 422                         return -1;
 423                 }
 424                 /* TODO: request info from primary to set up Rx and Tx */
 425                 eth_dev->dev_ops = &eth_kni_ops;
 426                 eth_dev->device = &vdev->device;
 427                 rte_eth_dev_probing_finish(eth_dev);
 428                 return 0;
 429         }
 430
 431         ret = eth_kni_kvargs_process(&args, params);
 432         if (ret < 0)
 433                 return ret;
 434
 435         ret = kni_init();
 436         if (ret < 0)
 437                 return ret;
 438
 439         eth_dev = eth_kni_create(vdev, &args, rte_socket_id());
 440         if (eth_dev == NULL)
 441                 goto kni_uninit;
 442
 443         eth_dev->rx_pkt_burst = eth_kni_rx;
 444         eth_dev->tx_pkt_burst = eth_kni_tx;
 445
 446         rte_eth_dev_probing_finish(eth_dev);
 447         return 0;
 448
 449 kni_uninit:
 450         is_kni_initialized--;
 451         if (is_kni_initialized == 0)
 452                 rte_kni_close();
 453         return -1;
 454 }
 455
 456 static int
 457 eth_kni_remove(struct rte_vdev_device *vdev)
 458 {
 459         struct rte_eth_dev *eth_dev;
 460         struct pmd_internals *internals;
 461         const char *name;
 462         int ret;
 463
 464         name = rte_vdev_device_name(vdev);
 465         PMD_LOG(INFO, "Un-Initializing eth_kni for %s", name);
 466
 467         /* find the ethdev entry */
 468         eth_dev = rte_eth_dev_allocated(name);
 469         if (eth_dev == NULL)
 470                 return -1;
 471
 472         /* mac_addrs must not be freed alone because part of dev_private */
 473         eth_dev->data->mac_addrs = NULL;
 474
 475         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 476                 return rte_eth_dev_release_port(eth_dev);
 477
 478         eth_kni_dev_stop(eth_dev);
 479
 480         internals = eth_dev->data->dev_private;
 481         ret = rte_kni_release(internals->kni);
 482         if (ret)
 483                 PMD_LOG(WARNING, "Not able to release kni for %s", name);
 484
 485         rte_eth_dev_release_port(eth_dev);
 486
 487         is_kni_initialized--;
 488         if (is_kni_initialized == 0)
 489                 rte_kni_close();
 490
 491         return 0;
 492 }
 493
 494 static struct rte_vdev_driver eth_kni_drv = {
 495         .probe = eth_kni_probe,
 496         .remove = eth_kni_remove,
 497 };
 498
 499 RTE_PMD_REGISTER_VDEV(net_kni, eth_kni_drv);
 500 RTE_PMD_REGISTER_PARAM_STRING(net_kni, ETH_KNI_NO_REQUEST_THREAD_ARG "=<int>");
 501
 502 RTE_INIT(eth_kni_init_log)
 503 {
 504         eth_kni_logtype = rte_log_register("pmd.net.kni");
 505         if (eth_kni_logtype >= 0)
 506                 rte_log_set_level(eth_kni_logtype, RTE_LOG_NOTICE);
 507 }