drivers/net/virtio/virtio_user_ethdev.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2016 Intel Corporation
   3  */
   4
   5 #include <stdint.h>
   6 #include <sys/types.h>
   7 #include <unistd.h>
   8 #include <fcntl.h>
   9 #include <linux/major.h>
  10 #include <sys/stat.h>
  11 #include <sys/sysmacros.h>
  12 #include <sys/socket.h>
  13
  14 #include <rte_malloc.h>
  15 #include <rte_kvargs.h>
  16 #include <rte_ethdev_vdev.h>
  17 #include <rte_bus_vdev.h>
  18 #include <rte_alarm.h>
  19 #include <rte_cycles.h>
  20
  21 #include "virtio_ethdev.h"
  22 #include "virtio_logs.h"
  23 #include "virtio_pci.h"
  24 #include "virtqueue.h"
  25 #include "virtio_rxtx.h"
  26 #include "virtio_user/virtio_user_dev.h"
  27 #include "virtio_user/vhost.h"
  28
  29 #define virtio_user_get_dev(hw) \
  30         ((struct virtio_user_dev *)(hw)->virtio_user_dev)
  31
  32 static void
  33 virtio_user_reset_queues_packed(struct rte_eth_dev *dev)
  34 {
  35         struct virtio_hw *hw = dev->data->dev_private;
  36         struct virtnet_rx *rxvq;
  37         struct virtnet_tx *txvq;
  38         uint16_t i;
  39
  40         /* Add lock to avoid queue contention. */
  41         rte_spinlock_lock(&hw->state_lock);
  42         hw->started = 0;
  43
  44         /*
  45          * Waitting for datapath to complete before resetting queues.
  46          * 1 ms should be enough for the ongoing Tx/Rx function to finish.
  47          */
  48         rte_delay_ms(1);
  49
  50         /* Vring reset for each Tx queue and Rx queue. */
  51         for (i = 0; i < dev->data->nb_rx_queues; i++) {
  52                 rxvq = dev->data->rx_queues[i];
  53                 virtqueue_rxvq_reset_packed(rxvq->vq);
  54                 virtio_dev_rx_queue_setup_finish(dev, i);
  55         }
  56
  57         for (i = 0; i < dev->data->nb_tx_queues; i++) {
  58                 txvq = dev->data->tx_queues[i];
  59                 virtqueue_txvq_reset_packed(txvq->vq);
  60         }
  61
  62         hw->started = 1;
  63         rte_spinlock_unlock(&hw->state_lock);
  64 }
  65
  66
  67 static int
  68 virtio_user_server_reconnect(struct virtio_user_dev *dev)
  69 {
  70         int ret;
  71         int connectfd;
  72         struct rte_eth_dev *eth_dev = &rte_eth_devices[dev->port_id];
  73         struct virtio_hw *hw = eth_dev->data->dev_private;
  74         uint64_t protocol_features;
  75
  76         connectfd = accept(dev->listenfd, NULL, NULL);
  77         if (connectfd < 0)
  78                 return -1;
  79
  80         dev->vhostfd = connectfd;
  81         if (dev->ops->send_request(dev, VHOST_USER_GET_FEATURES,
  82                                    &dev->device_features) < 0) {
  83                 PMD_INIT_LOG(ERR, "get_features failed: %s",
  84                              strerror(errno));
  85                 return -1;
  86         }
  87
  88         if (dev->device_features &
  89                         (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) {
  90                 if (dev->ops->send_request(dev,
  91                                         VHOST_USER_GET_PROTOCOL_FEATURES,
  92                                         &protocol_features))
  93                         return -1;
  94
  95                 /* Offer VHOST_USER_PROTOCOL_F_STATUS */
  96                 dev->protocol_features |=
  97                         (1ULL << VHOST_USER_PROTOCOL_F_STATUS);
  98                 dev->protocol_features &= protocol_features;
  99
 100                 if (dev->ops->send_request(dev,
 101                                         VHOST_USER_SET_PROTOCOL_FEATURES,
 102                                         &dev->protocol_features))
 103                         return -1;
 104
 105                 if (!(dev->protocol_features &
 106                                 (1ULL << VHOST_USER_PROTOCOL_F_MQ)))
 107                         dev->unsupported_features |= (1ull << VIRTIO_NET_F_MQ);
 108         }
 109
 110         dev->device_features |= dev->frontend_features;
 111
 112         /* umask vhost-user unsupported features */
 113         dev->device_features &= ~(dev->unsupported_features);
 114
 115         dev->features &= dev->device_features;
 116
 117         /* For packed ring, resetting queues is required in reconnection. */
 118         if (vtpci_packed_queue(hw) &&
 119            (vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_DRIVER_OK)) {
 120                 PMD_INIT_LOG(NOTICE, "Packets on the fly will be dropped"
 121                                 " when packed ring reconnecting.");
 122                 virtio_user_reset_queues_packed(eth_dev);
 123         }
 124
 125         ret = virtio_user_start_device(dev);
 126         if (ret < 0)
 127                 return -1;
 128
 129         if (dev->queue_pairs > 1) {
 130                 ret = virtio_user_handle_mq(dev, dev->queue_pairs);
 131                 if (ret != 0) {
 132                         PMD_INIT_LOG(ERR, "Fails to enable multi-queue pairs!");
 133                         return -1;
 134                 }
 135         }
 136         if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
 137                 if (rte_intr_disable(eth_dev->intr_handle) < 0) {
 138                         PMD_DRV_LOG(ERR, "interrupt disable failed");
 139                         return -1;
 140                 }
 141                 rte_intr_callback_unregister(eth_dev->intr_handle,
 142                                              virtio_interrupt_handler,
 143                                              eth_dev);
 144                 eth_dev->intr_handle->fd = connectfd;
 145                 rte_intr_callback_register(eth_dev->intr_handle,
 146                                            virtio_interrupt_handler, eth_dev);
 147
 148                 if (rte_intr_enable(eth_dev->intr_handle) < 0) {
 149                         PMD_DRV_LOG(ERR, "interrupt enable failed");
 150                         return -1;
 151                 }
 152         }
 153         PMD_INIT_LOG(NOTICE, "server mode virtio-user reconnection succeeds!");
 154         return 0;
 155 }
 156
 157 static void
 158 virtio_user_delayed_handler(void *param)
 159 {
 160         struct virtio_hw *hw = (struct virtio_hw *)param;
 161         struct rte_eth_dev *eth_dev = &rte_eth_devices[hw->port_id];
 162         struct virtio_user_dev *dev = virtio_user_get_dev(hw);
 163
 164         if (rte_intr_disable(eth_dev->intr_handle) < 0) {
 165                 PMD_DRV_LOG(ERR, "interrupt disable failed");
 166                 return;
 167         }
 168         rte_intr_callback_unregister(eth_dev->intr_handle,
 169                                      virtio_interrupt_handler, eth_dev);
 170         if (dev->is_server) {
 171                 if (dev->vhostfd >= 0) {
 172                         close(dev->vhostfd);
 173                         dev->vhostfd = -1;
 174                         /* Until the featuers are negotiated again, don't assume
 175                          * the backend supports VHOST_USER_PROTOCOL_F_STATUS
 176                          */
 177                         dev->protocol_features &=
 178                                 ~(1ULL << VHOST_USER_PROTOCOL_F_STATUS);
 179                 }
 180                 eth_dev->intr_handle->fd = dev->listenfd;
 181                 rte_intr_callback_register(eth_dev->intr_handle,
 182                                            virtio_interrupt_handler, eth_dev);
 183                 if (rte_intr_enable(eth_dev->intr_handle) < 0) {
 184                         PMD_DRV_LOG(ERR, "interrupt enable failed");
 185                         return;
 186                 }
 187         }
 188 }
 189
 190 static void
 191 virtio_user_read_dev_config(struct virtio_hw *hw, size_t offset,
 192                      void *dst, int length)
 193 {
 194         int i;
 195         struct virtio_user_dev *dev = virtio_user_get_dev(hw);
 196
 197         if (offset == offsetof(struct virtio_net_config, mac) &&
 198             length == RTE_ETHER_ADDR_LEN) {
 199                 for (i = 0; i < RTE_ETHER_ADDR_LEN; ++i)
 200                         ((uint8_t *)dst)[i] = dev->mac_addr[i];
 201                 return;
 202         }
 203
 204         if (offset == offsetof(struct virtio_net_config, status)) {
 205                 char buf[128];
 206
 207                 if (dev->vhostfd >= 0) {
 208                         int r;
 209                         int flags;
 210
 211                         flags = fcntl(dev->vhostfd, F_GETFL);
 212                         if (fcntl(dev->vhostfd, F_SETFL,
 213                                         flags | O_NONBLOCK) == -1) {
 214                                 PMD_DRV_LOG(ERR, "error setting O_NONBLOCK flag");
 215                                 return;
 216                         }
 217                         r = recv(dev->vhostfd, buf, 128, MSG_PEEK);
 218                         if (r == 0 || (r < 0 && errno != EAGAIN)) {
 219                                 dev->net_status &= (~VIRTIO_NET_S_LINK_UP);
 220                                 PMD_DRV_LOG(ERR, "virtio-user port %u is down",
 221                                             hw->port_id);
 222
 223                                 /* This function could be called in the process
 224                                  * of interrupt handling, callback cannot be
 225                                  * unregistered here, set an alarm to do it.
 226                                  */
 227                                 rte_eal_alarm_set(1,
 228                                                   virtio_user_delayed_handler,
 229                                                   (void *)hw);
 230                         } else {
 231                                 dev->net_status |= VIRTIO_NET_S_LINK_UP;
 232                         }
 233                         if (fcntl(dev->vhostfd, F_SETFL,
 234                                         flags & ~O_NONBLOCK) == -1) {
 235                                 PMD_DRV_LOG(ERR, "error clearing O_NONBLOCK flag");
 236                                 return;
 237                         }
 238                 } else if (dev->is_server) {
 239                         dev->net_status &= (~VIRTIO_NET_S_LINK_UP);
 240                         if (virtio_user_server_reconnect(dev) >= 0)
 241                                 dev->net_status |= VIRTIO_NET_S_LINK_UP;
 242                 }
 243
 244                 *(uint16_t *)dst = dev->net_status;
 245         }
 246
 247         if (offset == offsetof(struct virtio_net_config, max_virtqueue_pairs))
 248                 *(uint16_t *)dst = dev->max_queue_pairs;
 249 }
 250
 251 static void
 252 virtio_user_write_dev_config(struct virtio_hw *hw, size_t offset,
 253                       const void *src, int length)
 254 {
 255         int i;
 256         struct virtio_user_dev *dev = virtio_user_get_dev(hw);
 257
 258         if ((offset == offsetof(struct virtio_net_config, mac)) &&
 259             (length == RTE_ETHER_ADDR_LEN))
 260                 for (i = 0; i < RTE_ETHER_ADDR_LEN; ++i)
 261                         dev->mac_addr[i] = ((const uint8_t *)src)[i];
 262         else
 263                 PMD_DRV_LOG(ERR, "not supported offset=%zu, len=%d",
 264                             offset, length);
 265 }
 266
 267 static void
 268 virtio_user_reset(struct virtio_hw *hw)
 269 {
 270         struct virtio_user_dev *dev = virtio_user_get_dev(hw);
 271
 272         if (dev->status & VIRTIO_CONFIG_STATUS_DRIVER_OK)
 273                 virtio_user_stop_device(dev);
 274 }
 275
 276 static void
 277 virtio_user_set_status(struct virtio_hw *hw, uint8_t status)
 278 {
 279         struct virtio_user_dev *dev = virtio_user_get_dev(hw);
 280         uint8_t old_status = dev->status;
 281
 282         if (status & VIRTIO_CONFIG_STATUS_FEATURES_OK &&
 283                         ~old_status & VIRTIO_CONFIG_STATUS_FEATURES_OK)
 284                 virtio_user_dev_set_features(dev);
 285         if (status & VIRTIO_CONFIG_STATUS_DRIVER_OK)
 286                 virtio_user_start_device(dev);
 287         else if (status == VIRTIO_CONFIG_STATUS_RESET)
 288                 virtio_user_reset(hw);
 289
 290         virtio_user_dev_set_status(dev, status);
 291 }
 292
 293 static uint8_t
 294 virtio_user_get_status(struct virtio_hw *hw)
 295 {
 296         struct virtio_user_dev *dev = virtio_user_get_dev(hw);
 297
 298         virtio_user_dev_update_status(dev);
 299
 300         return dev->status;
 301 }
 302
 303 static uint64_t
 304 virtio_user_get_features(struct virtio_hw *hw)
 305 {
 306         struct virtio_user_dev *dev = virtio_user_get_dev(hw);
 307
 308         /* unmask feature bits defined in vhost user protocol */
 309         return dev->device_features & VIRTIO_PMD_SUPPORTED_GUEST_FEATURES;
 310 }
 311
 312 static void
 313 virtio_user_set_features(struct virtio_hw *hw, uint64_t features)
 314 {
 315         struct virtio_user_dev *dev = virtio_user_get_dev(hw);
 316
 317         dev->features = features & dev->device_features;
 318 }
 319
 320 static uint8_t
 321 virtio_user_get_isr(struct virtio_hw *hw __rte_unused)
 322 {
 323         /* rxq interrupts and config interrupt are separated in virtio-user,
 324          * here we only report config change.
 325          */
 326         return VIRTIO_PCI_ISR_CONFIG;
 327 }
 328
 329 static uint16_t
 330 virtio_user_set_config_irq(struct virtio_hw *hw __rte_unused,
 331                     uint16_t vec __rte_unused)
 332 {
 333         return 0;
 334 }
 335
 336 static uint16_t
 337 virtio_user_set_queue_irq(struct virtio_hw *hw __rte_unused,
 338                           struct virtqueue *vq __rte_unused,
 339                           uint16_t vec)
 340 {
 341         /* pretend we have done that */
 342         return vec;
 343 }
 344
 345 /* This function is to get the queue size, aka, number of descs, of a specified
 346  * queue. Different with the VHOST_USER_GET_QUEUE_NUM, which is used to get the
 347  * max supported queues.
 348  */
 349 static uint16_t
 350 virtio_user_get_queue_num(struct virtio_hw *hw, uint16_t queue_id __rte_unused)
 351 {
 352         struct virtio_user_dev *dev = virtio_user_get_dev(hw);
 353
 354         /* Currently, each queue has same queue size */
 355         return dev->queue_size;
 356 }
 357
 358 static void
 359 virtio_user_setup_queue_packed(struct virtqueue *vq,
 360                                struct virtio_user_dev *dev)
 361 {
 362         uint16_t queue_idx = vq->vq_queue_index;
 363         struct vring_packed *vring;
 364         uint64_t desc_addr;
 365         uint64_t avail_addr;
 366         uint64_t used_addr;
 367         uint16_t i;
 368
 369         vring  = &dev->packed_vrings[queue_idx];
 370         desc_addr = (uintptr_t)vq->vq_ring_virt_mem;
 371         avail_addr = desc_addr + vq->vq_nentries *
 372                 sizeof(struct vring_packed_desc);
 373         used_addr = RTE_ALIGN_CEIL(avail_addr +
 374                            sizeof(struct vring_packed_desc_event),
 375                            VIRTIO_PCI_VRING_ALIGN);
 376         vring->num = vq->vq_nentries;
 377         vring->desc = (void *)(uintptr_t)desc_addr;
 378         vring->driver = (void *)(uintptr_t)avail_addr;
 379         vring->device = (void *)(uintptr_t)used_addr;
 380         dev->packed_queues[queue_idx].avail_wrap_counter = true;
 381         dev->packed_queues[queue_idx].used_wrap_counter = true;
 382
 383         for (i = 0; i < vring->num; i++)
 384                 vring->desc[i].flags = 0;
 385 }
 386
 387 static void
 388 virtio_user_setup_queue_split(struct virtqueue *vq, struct virtio_user_dev *dev)
 389 {
 390         uint16_t queue_idx = vq->vq_queue_index;
 391         uint64_t desc_addr, avail_addr, used_addr;
 392
 393         desc_addr = (uintptr_t)vq->vq_ring_virt_mem;
 394         avail_addr = desc_addr + vq->vq_nentries * sizeof(struct vring_desc);
 395         used_addr = RTE_ALIGN_CEIL(avail_addr + offsetof(struct vring_avail,
 396                                                          ring[vq->vq_nentries]),
 397                                    VIRTIO_PCI_VRING_ALIGN);
 398
 399         dev->vrings[queue_idx].num = vq->vq_nentries;
 400         dev->vrings[queue_idx].desc = (void *)(uintptr_t)desc_addr;
 401         dev->vrings[queue_idx].avail = (void *)(uintptr_t)avail_addr;
 402         dev->vrings[queue_idx].used = (void *)(uintptr_t)used_addr;
 403 }
 404
 405 static int
 406 virtio_user_setup_queue(struct virtio_hw *hw, struct virtqueue *vq)
 407 {
 408         struct virtio_user_dev *dev = virtio_user_get_dev(hw);
 409
 410         if (vtpci_packed_queue(hw))
 411                 virtio_user_setup_queue_packed(vq, dev);
 412         else
 413                 virtio_user_setup_queue_split(vq, dev);
 414
 415         return 0;
 416 }
 417
 418 static void
 419 virtio_user_del_queue(struct virtio_hw *hw, struct virtqueue *vq)
 420 {
 421         /* For legacy devices, write 0 to VIRTIO_PCI_QUEUE_PFN port, QEMU
 422          * correspondingly stops the ioeventfds, and reset the status of
 423          * the device.
 424          * For modern devices, set queue desc, avail, used in PCI bar to 0,
 425          * not see any more behavior in QEMU.
 426          *
 427          * Here we just care about what information to deliver to vhost-user
 428          * or vhost-kernel. So we just close ioeventfd for now.
 429          */
 430         struct virtio_user_dev *dev = virtio_user_get_dev(hw);
 431
 432         close(dev->callfds[vq->vq_queue_index]);
 433         close(dev->kickfds[vq->vq_queue_index]);
 434 }
 435
 436 static void
 437 virtio_user_notify_queue(struct virtio_hw *hw, struct virtqueue *vq)
 438 {
 439         uint64_t buf = 1;
 440         struct virtio_user_dev *dev = virtio_user_get_dev(hw);
 441
 442         if (hw->cvq && (hw->cvq->vq == vq)) {
 443                 if (vtpci_packed_queue(vq->hw))
 444                         virtio_user_handle_cq_packed(dev, vq->vq_queue_index);
 445                 else
 446                         virtio_user_handle_cq(dev, vq->vq_queue_index);
 447                 return;
 448         }
 449
 450         if (write(dev->kickfds[vq->vq_queue_index], &buf, sizeof(buf)) < 0)
 451                 PMD_DRV_LOG(ERR, "failed to kick backend: %s",
 452                             strerror(errno));
 453 }
 454
 455 const struct virtio_pci_ops virtio_user_ops = {
 456         .read_dev_cfg   = virtio_user_read_dev_config,
 457         .write_dev_cfg  = virtio_user_write_dev_config,
 458         .get_status     = virtio_user_get_status,
 459         .set_status     = virtio_user_set_status,
 460         .get_features   = virtio_user_get_features,
 461         .set_features   = virtio_user_set_features,
 462         .get_isr        = virtio_user_get_isr,
 463         .set_config_irq = virtio_user_set_config_irq,
 464         .set_queue_irq  = virtio_user_set_queue_irq,
 465         .get_queue_num  = virtio_user_get_queue_num,
 466         .setup_queue    = virtio_user_setup_queue,
 467         .del_queue      = virtio_user_del_queue,
 468         .notify_queue   = virtio_user_notify_queue,
 469 };
 470
 471 static const char *valid_args[] = {
 472 #define VIRTIO_USER_ARG_QUEUES_NUM     "queues"
 473         VIRTIO_USER_ARG_QUEUES_NUM,
 474 #define VIRTIO_USER_ARG_CQ_NUM         "cq"
 475         VIRTIO_USER_ARG_CQ_NUM,
 476 #define VIRTIO_USER_ARG_MAC            "mac"
 477         VIRTIO_USER_ARG_MAC,
 478 #define VIRTIO_USER_ARG_PATH           "path"
 479         VIRTIO_USER_ARG_PATH,
 480 #define VIRTIO_USER_ARG_QUEUE_SIZE     "queue_size"
 481         VIRTIO_USER_ARG_QUEUE_SIZE,
 482 #define VIRTIO_USER_ARG_INTERFACE_NAME "iface"
 483         VIRTIO_USER_ARG_INTERFACE_NAME,
 484 #define VIRTIO_USER_ARG_SERVER_MODE    "server"
 485         VIRTIO_USER_ARG_SERVER_MODE,
 486 #define VIRTIO_USER_ARG_MRG_RXBUF      "mrg_rxbuf"
 487         VIRTIO_USER_ARG_MRG_RXBUF,
 488 #define VIRTIO_USER_ARG_IN_ORDER       "in_order"
 489         VIRTIO_USER_ARG_IN_ORDER,
 490 #define VIRTIO_USER_ARG_PACKED_VQ      "packed_vq"
 491         VIRTIO_USER_ARG_PACKED_VQ,
 492 #define VIRTIO_USER_ARG_SPEED          "speed"
 493         VIRTIO_USER_ARG_SPEED,
 494 #define VIRTIO_USER_ARG_VECTORIZED     "vectorized"
 495         VIRTIO_USER_ARG_VECTORIZED,
 496         NULL
 497 };
 498
 499 #define VIRTIO_USER_DEF_CQ_EN   0
 500 #define VIRTIO_USER_DEF_Q_NUM   1
 501 #define VIRTIO_USER_DEF_Q_SZ    256
 502 #define VIRTIO_USER_DEF_SERVER_MODE     0
 503
 504 static int
 505 get_string_arg(const char *key __rte_unused,
 506                const char *value, void *extra_args)
 507 {
 508         if (!value || !extra_args)
 509                 return -EINVAL;
 510
 511         *(char **)extra_args = strdup(value);
 512
 513         if (!*(char **)extra_args)
 514                 return -ENOMEM;
 515
 516         return 0;
 517 }
 518
 519 static int
 520 get_integer_arg(const char *key __rte_unused,
 521                 const char *value, void *extra_args)
 522 {
 523         uint64_t integer = 0;
 524         if (!value || !extra_args)
 525                 return -EINVAL;
 526         errno = 0;
 527         integer = strtoull(value, NULL, 0);
 528         /* extra_args keeps default value, it should be replaced
 529          * only in case of successful parsing of the 'value' arg
 530          */
 531         if (errno == 0)
 532                 *(uint64_t *)extra_args = integer;
 533         return -errno;
 534 }
 535
 536 static uint32_t
 537 vdpa_dynamic_major_num(void)
 538 {
 539         FILE *fp;
 540         char *line = NULL;
 541         size_t size;
 542         char name[11];
 543         bool found = false;
 544         uint32_t num;
 545
 546         fp = fopen("/proc/devices", "r");
 547         if (fp == NULL) {
 548                 PMD_INIT_LOG(ERR, "Cannot open /proc/devices: %s",
 549                              strerror(errno));
 550                 return UNNAMED_MAJOR;
 551         }
 552
 553         while (getline(&line, &size, fp) > 0) {
 554                 char *stripped = line + strspn(line, " ");
 555                 if ((sscanf(stripped, "%u %10s", &num, name) == 2) &&
 556                     (strncmp(name, "vhost-vdpa", 10) == 0)) {
 557                         found = true;
 558                         break;
 559                 }
 560         }
 561         fclose(fp);
 562         return found ? num : UNNAMED_MAJOR;
 563 }
 564
 565 static enum virtio_user_backend_type
 566 virtio_user_backend_type(const char *path)
 567 {
 568         struct stat sb;
 569
 570         if (stat(path, &sb) == -1) {
 571                 if (errno == ENOENT)
 572                         return VIRTIO_USER_BACKEND_VHOST_USER;
 573
 574                 PMD_INIT_LOG(ERR, "Stat fails: %s (%s)\n", path,
 575                              strerror(errno));
 576                 return VIRTIO_USER_BACKEND_UNKNOWN;
 577         }
 578
 579         if (S_ISSOCK(sb.st_mode)) {
 580                 return VIRTIO_USER_BACKEND_VHOST_USER;
 581         } else if (S_ISCHR(sb.st_mode)) {
 582                 if (major(sb.st_rdev) == MISC_MAJOR)
 583                         return VIRTIO_USER_BACKEND_VHOST_KERNEL;
 584                 if (major(sb.st_rdev) == vdpa_dynamic_major_num())
 585                         return VIRTIO_USER_BACKEND_VHOST_VDPA;
 586         }
 587         return VIRTIO_USER_BACKEND_UNKNOWN;
 588 }
 589
 590 static struct rte_eth_dev *
 591 virtio_user_eth_dev_alloc(struct rte_vdev_device *vdev)
 592 {
 593         struct rte_eth_dev *eth_dev;
 594         struct rte_eth_dev_data *data;
 595         struct virtio_hw *hw;
 596         struct virtio_user_dev *dev;
 597
 598         eth_dev = rte_eth_vdev_allocate(vdev, sizeof(*hw));
 599         if (!eth_dev) {
 600                 PMD_INIT_LOG(ERR, "cannot alloc rte_eth_dev");
 601                 return NULL;
 602         }
 603
 604         data = eth_dev->data;
 605         hw = eth_dev->data->dev_private;
 606
 607         dev = rte_zmalloc(NULL, sizeof(*dev), 0);
 608         if (!dev) {
 609                 PMD_INIT_LOG(ERR, "malloc virtio_user_dev failed");
 610                 rte_eth_dev_release_port(eth_dev);
 611                 return NULL;
 612         }
 613
 614         hw->port_id = data->port_id;
 615         dev->port_id = data->port_id;
 616         virtio_hw_internal[hw->port_id].vtpci_ops = &virtio_user_ops;
 617         /*
 618          * MSIX is required to enable LSC (see virtio_init_device).
 619          * Here just pretend that we support msix.
 620          */
 621         hw->use_msix = 1;
 622         hw->modern   = 0;
 623         hw->use_vec_rx = 0;
 624         hw->use_vec_tx = 0;
 625         hw->use_inorder_rx = 0;
 626         hw->use_inorder_tx = 0;
 627         hw->virtio_user_dev = dev;
 628         return eth_dev;
 629 }
 630
 631 static void
 632 virtio_user_eth_dev_free(struct rte_eth_dev *eth_dev)
 633 {
 634         struct rte_eth_dev_data *data = eth_dev->data;
 635         struct virtio_hw *hw = data->dev_private;
 636
 637         rte_free(hw->virtio_user_dev);
 638         rte_eth_dev_release_port(eth_dev);
 639 }
 640
 641 /* Dev initialization routine. Invoked once for each virtio vdev at
 642  * EAL init time, see rte_bus_probe().
 643  * Returns 0 on success.
 644  */
 645 static int
 646 virtio_user_pmd_probe(struct rte_vdev_device *dev)
 647 {
 648         struct rte_kvargs *kvlist = NULL;
 649         struct rte_eth_dev *eth_dev;
 650         struct virtio_hw *hw;
 651         enum virtio_user_backend_type backend_type = VIRTIO_USER_BACKEND_UNKNOWN;
 652         uint64_t queues = VIRTIO_USER_DEF_Q_NUM;
 653         uint64_t cq = VIRTIO_USER_DEF_CQ_EN;
 654         uint64_t queue_size = VIRTIO_USER_DEF_Q_SZ;
 655         uint64_t server_mode = VIRTIO_USER_DEF_SERVER_MODE;
 656         uint64_t mrg_rxbuf = 1;
 657         uint64_t in_order = 1;
 658         uint64_t packed_vq = 0;
 659         uint64_t vectorized = 0;
 660         char *path = NULL;
 661         char *ifname = NULL;
 662         char *mac_addr = NULL;
 663         int ret = -1;
 664
 665         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
 666                 const char *name = rte_vdev_device_name(dev);
 667                 eth_dev = rte_eth_dev_attach_secondary(name);
 668                 if (!eth_dev) {
 669                         PMD_INIT_LOG(ERR, "Failed to probe %s", name);
 670                         return -1;
 671                 }
 672
 673                 if (eth_virtio_dev_init(eth_dev) < 0) {
 674                         PMD_INIT_LOG(ERR, "eth_virtio_dev_init fails");
 675                         rte_eth_dev_release_port(eth_dev);
 676                         return -1;
 677                 }
 678
 679                 eth_dev->dev_ops = &virtio_user_secondary_eth_dev_ops;
 680                 eth_dev->device = &dev->device;
 681                 rte_eth_dev_probing_finish(eth_dev);
 682                 return 0;
 683         }
 684
 685         kvlist = rte_kvargs_parse(rte_vdev_device_args(dev), valid_args);
 686         if (!kvlist) {
 687                 PMD_INIT_LOG(ERR, "error when parsing param");
 688                 goto end;
 689         }
 690
 691         if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_PATH) == 1) {
 692                 if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_PATH,
 693                                        &get_string_arg, &path) < 0) {
 694                         PMD_INIT_LOG(ERR, "error to parse %s",
 695                                      VIRTIO_USER_ARG_PATH);
 696                         goto end;
 697                 }
 698         } else {
 699                 PMD_INIT_LOG(ERR, "arg %s is mandatory for virtio_user",
 700                              VIRTIO_USER_ARG_PATH);
 701                 goto end;
 702         }
 703
 704         backend_type = virtio_user_backend_type(path);
 705         if (backend_type == VIRTIO_USER_BACKEND_UNKNOWN) {
 706                 PMD_INIT_LOG(ERR,
 707                              "unable to determine backend type for path %s",
 708                         path);
 709                 goto end;
 710         }
 711         PMD_INIT_LOG(INFO, "Backend type detected: %s",
 712                      virtio_user_backend_strings[backend_type]);
 713
 714         if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_INTERFACE_NAME) == 1) {
 715                 if (backend_type != VIRTIO_USER_BACKEND_VHOST_KERNEL) {
 716                         PMD_INIT_LOG(ERR,
 717                                 "arg %s applies only to vhost-kernel backend",
 718                                 VIRTIO_USER_ARG_INTERFACE_NAME);
 719                         goto end;
 720                 }
 721
 722                 if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_INTERFACE_NAME,
 723                                        &get_string_arg, &ifname) < 0) {
 724                         PMD_INIT_LOG(ERR, "error to parse %s",
 725                                      VIRTIO_USER_ARG_INTERFACE_NAME);
 726                         goto end;
 727                 }
 728         }
 729
 730         if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_MAC) == 1) {
 731                 if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_MAC,
 732                                        &get_string_arg, &mac_addr) < 0) {
 733                         PMD_INIT_LOG(ERR, "error to parse %s",
 734                                      VIRTIO_USER_ARG_MAC);
 735                         goto end;
 736                 }
 737         }
 738
 739         if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_QUEUE_SIZE) == 1) {
 740                 if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_QUEUE_SIZE,
 741                                        &get_integer_arg, &queue_size) < 0) {
 742                         PMD_INIT_LOG(ERR, "error to parse %s",
 743                                      VIRTIO_USER_ARG_QUEUE_SIZE);
 744                         goto end;
 745                 }
 746         }
 747
 748         if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_QUEUES_NUM) == 1) {
 749                 if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_QUEUES_NUM,
 750                                        &get_integer_arg, &queues) < 0) {
 751                         PMD_INIT_LOG(ERR, "error to parse %s",
 752                                      VIRTIO_USER_ARG_QUEUES_NUM);
 753                         goto end;
 754                 }
 755         }
 756
 757         if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_SERVER_MODE) == 1) {
 758                 if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_SERVER_MODE,
 759                                        &get_integer_arg, &server_mode) < 0) {
 760                         PMD_INIT_LOG(ERR, "error to parse %s",
 761                                      VIRTIO_USER_ARG_SERVER_MODE);
 762                         goto end;
 763                 }
 764         }
 765
 766         if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_CQ_NUM) == 1) {
 767                 if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_CQ_NUM,
 768                                        &get_integer_arg, &cq) < 0) {
 769                         PMD_INIT_LOG(ERR, "error to parse %s",
 770                                      VIRTIO_USER_ARG_CQ_NUM);
 771                         goto end;
 772                 }
 773         } else if (queues > 1) {
 774                 cq = 1;
 775         }
 776
 777         if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_PACKED_VQ) == 1) {
 778                 if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_PACKED_VQ,
 779                                        &get_integer_arg, &packed_vq) < 0) {
 780                         PMD_INIT_LOG(ERR, "error to parse %s",
 781                                      VIRTIO_USER_ARG_PACKED_VQ);
 782                         goto end;
 783                 }
 784         }
 785
 786         if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_VECTORIZED) == 1) {
 787                 if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_VECTORIZED,
 788                                        &get_integer_arg, &vectorized) < 0) {
 789                         PMD_INIT_LOG(ERR, "error to parse %s",
 790                                      VIRTIO_USER_ARG_VECTORIZED);
 791                         goto end;
 792                 }
 793         }
 794
 795         if (queues > 1 && cq == 0) {
 796                 PMD_INIT_LOG(ERR, "multi-q requires ctrl-q");
 797                 goto end;
 798         }
 799
 800         if (queues > VIRTIO_MAX_VIRTQUEUE_PAIRS) {
 801                 PMD_INIT_LOG(ERR, "arg %s %" PRIu64 " exceeds the limit %u",
 802                         VIRTIO_USER_ARG_QUEUES_NUM, queues,
 803                         VIRTIO_MAX_VIRTQUEUE_PAIRS);
 804                 goto end;
 805         }
 806
 807         if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_MRG_RXBUF) == 1) {
 808                 if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_MRG_RXBUF,
 809                                        &get_integer_arg, &mrg_rxbuf) < 0) {
 810                         PMD_INIT_LOG(ERR, "error to parse %s",
 811                                      VIRTIO_USER_ARG_MRG_RXBUF);
 812                         goto end;
 813                 }
 814         }
 815
 816         if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_IN_ORDER) == 1) {
 817                 if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_IN_ORDER,
 818                                        &get_integer_arg, &in_order) < 0) {
 819                         PMD_INIT_LOG(ERR, "error to parse %s",
 820                                      VIRTIO_USER_ARG_IN_ORDER);
 821                         goto end;
 822                 }
 823         }
 824
 825         eth_dev = virtio_user_eth_dev_alloc(dev);
 826         if (!eth_dev) {
 827                 PMD_INIT_LOG(ERR, "virtio_user fails to alloc device");
 828                 goto end;
 829         }
 830
 831         hw = eth_dev->data->dev_private;
 832         if (virtio_user_dev_init(hw->virtio_user_dev, path, queues, cq,
 833                          queue_size, mac_addr, &ifname, server_mode,
 834                          mrg_rxbuf, in_order, packed_vq, backend_type) < 0) {
 835                 PMD_INIT_LOG(ERR, "virtio_user_dev_init fails");
 836                 virtio_user_eth_dev_free(eth_dev);
 837                 goto end;
 838         }
 839
 840         /* previously called by pci probing for physical dev */
 841         if (eth_virtio_dev_init(eth_dev) < 0) {
 842                 PMD_INIT_LOG(ERR, "eth_virtio_dev_init fails");
 843                 virtio_user_eth_dev_free(eth_dev);
 844                 goto end;
 845         }
 846
 847         if (vectorized) {
 848                 if (packed_vq) {
 849 #if defined(CC_AVX512_SUPPORT)
 850                         hw->use_vec_rx = 1;
 851                         hw->use_vec_tx = 1;
 852 #else
 853                         PMD_INIT_LOG(INFO,
 854                                 "building environment do not support packed ring vectorized");
 855 #endif
 856                 } else {
 857                         hw->use_vec_rx = 1;
 858                 }
 859         }
 860
 861         rte_eth_dev_probing_finish(eth_dev);
 862         ret = 0;
 863
 864 end:
 865         if (kvlist)
 866                 rte_kvargs_free(kvlist);
 867         if (path)
 868                 free(path);
 869         if (mac_addr)
 870                 free(mac_addr);
 871         if (ifname)
 872                 free(ifname);
 873         return ret;
 874 }
 875
 876 static int
 877 virtio_user_pmd_remove(struct rte_vdev_device *vdev)
 878 {
 879         const char *name;
 880         struct rte_eth_dev *eth_dev;
 881
 882         if (!vdev)
 883                 return -EINVAL;
 884
 885         name = rte_vdev_device_name(vdev);
 886         PMD_DRV_LOG(INFO, "Un-Initializing %s", name);
 887         eth_dev = rte_eth_dev_allocated(name);
 888         /* Port has already been released by close. */
 889         if (!eth_dev)
 890                 return 0;
 891
 892         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 893                 return rte_eth_dev_release_port(eth_dev);
 894
 895         /* make sure the device is stopped, queues freed */
 896         return rte_eth_dev_close(eth_dev->data->port_id);
 897 }
 898
 899 static int virtio_user_pmd_dma_map(struct rte_vdev_device *vdev, void *addr,
 900                 uint64_t iova, size_t len)
 901 {
 902         const char *name;
 903         struct rte_eth_dev *eth_dev;
 904         struct virtio_user_dev *dev;
 905         struct virtio_hw *hw;
 906
 907         if (!vdev)
 908                 return -EINVAL;
 909
 910         name = rte_vdev_device_name(vdev);
 911         eth_dev = rte_eth_dev_allocated(name);
 912         /* Port has already been released by close. */
 913         if (!eth_dev)
 914                 return 0;
 915
 916         hw = (struct virtio_hw *)eth_dev->data->dev_private;
 917         dev = hw->virtio_user_dev;
 918
 919         if (dev->ops->dma_map)
 920                 return dev->ops->dma_map(dev, addr, iova, len);
 921
 922         return 0;
 923 }
 924
 925 static int virtio_user_pmd_dma_unmap(struct rte_vdev_device *vdev, void *addr,
 926                 uint64_t iova, size_t len)
 927 {
 928         const char *name;
 929         struct rte_eth_dev *eth_dev;
 930         struct virtio_user_dev *dev;
 931         struct virtio_hw *hw;
 932
 933         if (!vdev)
 934                 return -EINVAL;
 935
 936         name = rte_vdev_device_name(vdev);
 937         eth_dev = rte_eth_dev_allocated(name);
 938         /* Port has already been released by close. */
 939         if (!eth_dev)
 940                 return 0;
 941
 942         hw = (struct virtio_hw *)eth_dev->data->dev_private;
 943         dev = hw->virtio_user_dev;
 944
 945         if (dev->ops->dma_unmap)
 946                 return dev->ops->dma_unmap(dev, addr, iova, len);
 947
 948         return 0;
 949 }
 950
 951 static struct rte_vdev_driver virtio_user_driver = {
 952         .probe = virtio_user_pmd_probe,
 953         .remove = virtio_user_pmd_remove,
 954         .dma_map = virtio_user_pmd_dma_map,
 955         .dma_unmap = virtio_user_pmd_dma_unmap,
 956 };
 957
 958 RTE_PMD_REGISTER_VDEV(net_virtio_user, virtio_user_driver);
 959 RTE_PMD_REGISTER_ALIAS(net_virtio_user, virtio_user);
 960 RTE_PMD_REGISTER_PARAM_STRING(net_virtio_user,
 961         "path=<path> "
 962         "mac=<mac addr> "
 963         "cq=<int> "
 964         "queue_size=<int> "
 965         "queues=<int> "
 966         "iface=<string> "
 967         "server=<0|1> "
 968         "mrg_rxbuf=<0|1> "
 969         "in_order=<0|1> "
 970         "packed_vq=<0|1> "
 971         "speed=<int> "
 972         "vectorized=<0|1>");