drivers/net/virtio/virtio_ethdev.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2016 Intel Corporation
   3  */
   4
   5 #include <stdint.h>
   6 #include <string.h>
   7 #include <stdio.h>
   8 #include <errno.h>
   9 #include <unistd.h>
  10
  11 #include <ethdev_driver.h>
  12 #include <ethdev_pci.h>
  13 #include <rte_memcpy.h>
  14 #include <rte_string_fns.h>
  15 #include <rte_memzone.h>
  16 #include <rte_malloc.h>
  17 #include <rte_branch_prediction.h>
  18 #include <rte_pci.h>
  19 #include <rte_bus_pci.h>
  20 #include <rte_ether.h>
  21 #include <rte_ip.h>
  22 #include <rte_arp.h>
  23 #include <rte_common.h>
  24 #include <rte_errno.h>
  25 #include <rte_cpuflags.h>
  26 #include <rte_vect.h>
  27
  28 #include <rte_memory.h>
  29 #include <rte_eal.h>
  30 #include <rte_dev.h>
  31 #include <rte_cycles.h>
  32 #include <rte_kvargs.h>
  33
  34 #include "virtio_ethdev.h"
  35 #include "virtio_pci.h"
  36 #include "virtio_logs.h"
  37 #include "virtqueue.h"
  38 #include "virtio_rxtx.h"
  39 #include "virtio_user/virtio_user_dev.h"
  40
  41 static int  virtio_dev_configure(struct rte_eth_dev *dev);
  42 static int  virtio_dev_start(struct rte_eth_dev *dev);
  43 static int virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
  44 static int virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
  45 static int virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
  46 static int virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
  47 static uint32_t virtio_dev_speed_capa_get(uint32_t speed);
  48 static int virtio_dev_devargs_parse(struct rte_devargs *devargs,
  49         uint32_t *speed,
  50         int *vectorized);
  51 static int virtio_dev_info_get(struct rte_eth_dev *dev,
  52                                 struct rte_eth_dev_info *dev_info);
  53 static int virtio_dev_link_update(struct rte_eth_dev *dev,
  54         int wait_to_complete);
  55 static int virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask);
  56
  57 static void virtio_set_hwaddr(struct virtio_hw *hw);
  58 static void virtio_get_hwaddr(struct virtio_hw *hw);
  59
  60 static int virtio_dev_stats_get(struct rte_eth_dev *dev,
  61                                  struct rte_eth_stats *stats);
  62 static int virtio_dev_xstats_get(struct rte_eth_dev *dev,
  63                                  struct rte_eth_xstat *xstats, unsigned n);
  64 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
  65                                        struct rte_eth_xstat_name *xstats_names,
  66                                        unsigned limit);
  67 static int virtio_dev_stats_reset(struct rte_eth_dev *dev);
  68 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
  69 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
  70                                 uint16_t vlan_id, int on);
  71 static int virtio_mac_addr_add(struct rte_eth_dev *dev,
  72                                 struct rte_ether_addr *mac_addr,
  73                                 uint32_t index, uint32_t vmdq);
  74 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
  75 static int virtio_mac_addr_set(struct rte_eth_dev *dev,
  76                                 struct rte_ether_addr *mac_addr);
  77
  78 static int virtio_intr_disable(struct rte_eth_dev *dev);
  79
  80 static int virtio_dev_queue_stats_mapping_set(
  81         struct rte_eth_dev *eth_dev,
  82         uint16_t queue_id,
  83         uint8_t stat_idx,
  84         uint8_t is_rx);
  85
  86 static void virtio_notify_peers(struct rte_eth_dev *dev);
  87 static void virtio_ack_link_announce(struct rte_eth_dev *dev);
  88
  89 struct rte_virtio_xstats_name_off {
  90         char name[RTE_ETH_XSTATS_NAME_SIZE];
  91         unsigned offset;
  92 };
  93
  94 /* [rt]x_qX_ is prepended to the name string here */
  95 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
  96         {"good_packets",           offsetof(struct virtnet_rx, stats.packets)},
  97         {"good_bytes",             offsetof(struct virtnet_rx, stats.bytes)},
  98         {"errors",                 offsetof(struct virtnet_rx, stats.errors)},
  99         {"multicast_packets",      offsetof(struct virtnet_rx, stats.multicast)},
 100         {"broadcast_packets",      offsetof(struct virtnet_rx, stats.broadcast)},
 101         {"undersize_packets",      offsetof(struct virtnet_rx, stats.size_bins[0])},
 102         {"size_64_packets",        offsetof(struct virtnet_rx, stats.size_bins[1])},
 103         {"size_65_127_packets",    offsetof(struct virtnet_rx, stats.size_bins[2])},
 104         {"size_128_255_packets",   offsetof(struct virtnet_rx, stats.size_bins[3])},
 105         {"size_256_511_packets",   offsetof(struct virtnet_rx, stats.size_bins[4])},
 106         {"size_512_1023_packets",  offsetof(struct virtnet_rx, stats.size_bins[5])},
 107         {"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
 108         {"size_1519_max_packets",  offsetof(struct virtnet_rx, stats.size_bins[7])},
 109 };
 110
 111 /* [rt]x_qX_ is prepended to the name string here */
 112 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
 113         {"good_packets",           offsetof(struct virtnet_tx, stats.packets)},
 114         {"good_bytes",             offsetof(struct virtnet_tx, stats.bytes)},
 115         {"multicast_packets",      offsetof(struct virtnet_tx, stats.multicast)},
 116         {"broadcast_packets",      offsetof(struct virtnet_tx, stats.broadcast)},
 117         {"undersize_packets",      offsetof(struct virtnet_tx, stats.size_bins[0])},
 118         {"size_64_packets",        offsetof(struct virtnet_tx, stats.size_bins[1])},
 119         {"size_65_127_packets",    offsetof(struct virtnet_tx, stats.size_bins[2])},
 120         {"size_128_255_packets",   offsetof(struct virtnet_tx, stats.size_bins[3])},
 121         {"size_256_511_packets",   offsetof(struct virtnet_tx, stats.size_bins[4])},
 122         {"size_512_1023_packets",  offsetof(struct virtnet_tx, stats.size_bins[5])},
 123         {"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
 124         {"size_1519_max_packets",  offsetof(struct virtnet_tx, stats.size_bins[7])},
 125 };
 126
 127 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
 128                             sizeof(rte_virtio_rxq_stat_strings[0]))
 129 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
 130                             sizeof(rte_virtio_txq_stat_strings[0]))
 131
 132 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
 133
 134 static struct virtio_pmd_ctrl *
 135 virtio_send_command_packed(struct virtnet_ctl *cvq,
 136                            struct virtio_pmd_ctrl *ctrl,
 137                            int *dlen, int pkt_num)
 138 {
 139         struct virtqueue *vq = cvq->vq;
 140         int head;
 141         struct vring_packed_desc *desc = vq->vq_packed.ring.desc;
 142         struct virtio_pmd_ctrl *result;
 143         uint16_t flags;
 144         int sum = 0;
 145         int nb_descs = 0;
 146         int k;
 147
 148         /*
 149          * Format is enforced in qemu code:
 150          * One TX packet for header;
 151          * At least one TX packet per argument;
 152          * One RX packet for ACK.
 153          */
 154         head = vq->vq_avail_idx;
 155         flags = vq->vq_packed.cached_flags;
 156         desc[head].addr = cvq->virtio_net_hdr_mem;
 157         desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
 158         vq->vq_free_cnt--;
 159         nb_descs++;
 160         if (++vq->vq_avail_idx >= vq->vq_nentries) {
 161                 vq->vq_avail_idx -= vq->vq_nentries;
 162                 vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
 163         }
 164
 165         for (k = 0; k < pkt_num; k++) {
 166                 desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
 167                         + sizeof(struct virtio_net_ctrl_hdr)
 168                         + sizeof(ctrl->status) + sizeof(uint8_t) * sum;
 169                 desc[vq->vq_avail_idx].len = dlen[k];
 170                 desc[vq->vq_avail_idx].flags = VRING_DESC_F_NEXT |
 171                         vq->vq_packed.cached_flags;
 172                 sum += dlen[k];
 173                 vq->vq_free_cnt--;
 174                 nb_descs++;
 175                 if (++vq->vq_avail_idx >= vq->vq_nentries) {
 176                         vq->vq_avail_idx -= vq->vq_nentries;
 177                         vq->vq_packed.cached_flags ^=
 178                                 VRING_PACKED_DESC_F_AVAIL_USED;
 179                 }
 180         }
 181
 182         desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
 183                 + sizeof(struct virtio_net_ctrl_hdr);
 184         desc[vq->vq_avail_idx].len = sizeof(ctrl->status);
 185         desc[vq->vq_avail_idx].flags = VRING_DESC_F_WRITE |
 186                 vq->vq_packed.cached_flags;
 187         vq->vq_free_cnt--;
 188         nb_descs++;
 189         if (++vq->vq_avail_idx >= vq->vq_nentries) {
 190                 vq->vq_avail_idx -= vq->vq_nentries;
 191                 vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
 192         }
 193
 194         virtqueue_store_flags_packed(&desc[head], VRING_DESC_F_NEXT | flags,
 195                         vq->hw->weak_barriers);
 196
 197         virtio_wmb(vq->hw->weak_barriers);
 198         virtqueue_notify(vq);
 199
 200         /* wait for used desc in virtqueue
 201          * desc_is_used has a load-acquire or rte_io_rmb inside
 202          */
 203         while (!desc_is_used(&desc[head], vq))
 204                 usleep(100);
 205
 206         /* now get used descriptors */
 207         vq->vq_free_cnt += nb_descs;
 208         vq->vq_used_cons_idx += nb_descs;
 209         if (vq->vq_used_cons_idx >= vq->vq_nentries) {
 210                 vq->vq_used_cons_idx -= vq->vq_nentries;
 211                 vq->vq_packed.used_wrap_counter ^= 1;
 212         }
 213
 214         PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\n"
 215                         "vq->vq_avail_idx=%d\n"
 216                         "vq->vq_used_cons_idx=%d\n"
 217                         "vq->vq_packed.cached_flags=0x%x\n"
 218                         "vq->vq_packed.used_wrap_counter=%d\n",
 219                         vq->vq_free_cnt,
 220                         vq->vq_avail_idx,
 221                         vq->vq_used_cons_idx,
 222                         vq->vq_packed.cached_flags,
 223                         vq->vq_packed.used_wrap_counter);
 224
 225         result = cvq->virtio_net_hdr_mz->addr;
 226         return result;
 227 }
 228
 229 static struct virtio_pmd_ctrl *
 230 virtio_send_command_split(struct virtnet_ctl *cvq,
 231                           struct virtio_pmd_ctrl *ctrl,
 232                           int *dlen, int pkt_num)
 233 {
 234         struct virtio_pmd_ctrl *result;
 235         struct virtqueue *vq = cvq->vq;
 236         uint32_t head, i;
 237         int k, sum = 0;
 238
 239         head = vq->vq_desc_head_idx;
 240
 241         /*
 242          * Format is enforced in qemu code:
 243          * One TX packet for header;
 244          * At least one TX packet per argument;
 245          * One RX packet for ACK.
 246          */
 247         vq->vq_split.ring.desc[head].flags = VRING_DESC_F_NEXT;
 248         vq->vq_split.ring.desc[head].addr = cvq->virtio_net_hdr_mem;
 249         vq->vq_split.ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
 250         vq->vq_free_cnt--;
 251         i = vq->vq_split.ring.desc[head].next;
 252
 253         for (k = 0; k < pkt_num; k++) {
 254                 vq->vq_split.ring.desc[i].flags = VRING_DESC_F_NEXT;
 255                 vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
 256                         + sizeof(struct virtio_net_ctrl_hdr)
 257                         + sizeof(ctrl->status) + sizeof(uint8_t)*sum;
 258                 vq->vq_split.ring.desc[i].len = dlen[k];
 259                 sum += dlen[k];
 260                 vq->vq_free_cnt--;
 261                 i = vq->vq_split.ring.desc[i].next;
 262         }
 263
 264         vq->vq_split.ring.desc[i].flags = VRING_DESC_F_WRITE;
 265         vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
 266                         + sizeof(struct virtio_net_ctrl_hdr);
 267         vq->vq_split.ring.desc[i].len = sizeof(ctrl->status);
 268         vq->vq_free_cnt--;
 269
 270         vq->vq_desc_head_idx = vq->vq_split.ring.desc[i].next;
 271
 272         vq_update_avail_ring(vq, head);
 273         vq_update_avail_idx(vq);
 274
 275         PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
 276
 277         virtqueue_notify(vq);
 278
 279         while (virtqueue_nused(vq) == 0)
 280                 usleep(100);
 281
 282         while (virtqueue_nused(vq)) {
 283                 uint32_t idx, desc_idx, used_idx;
 284                 struct vring_used_elem *uep;
 285
 286                 used_idx = (uint32_t)(vq->vq_used_cons_idx
 287                                 & (vq->vq_nentries - 1));
 288                 uep = &vq->vq_split.ring.used->ring[used_idx];
 289                 idx = (uint32_t) uep->id;
 290                 desc_idx = idx;
 291
 292                 while (vq->vq_split.ring.desc[desc_idx].flags &
 293                                 VRING_DESC_F_NEXT) {
 294                         desc_idx = vq->vq_split.ring.desc[desc_idx].next;
 295                         vq->vq_free_cnt++;
 296                 }
 297
 298                 vq->vq_split.ring.desc[desc_idx].next = vq->vq_desc_head_idx;
 299                 vq->vq_desc_head_idx = idx;
 300
 301                 vq->vq_used_cons_idx++;
 302                 vq->vq_free_cnt++;
 303         }
 304
 305         PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
 306                         vq->vq_free_cnt, vq->vq_desc_head_idx);
 307
 308         result = cvq->virtio_net_hdr_mz->addr;
 309         return result;
 310 }
 311
 312 static int
 313 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
 314                     int *dlen, int pkt_num)
 315 {
 316         virtio_net_ctrl_ack status = ~0;
 317         struct virtio_pmd_ctrl *result;
 318         struct virtqueue *vq;
 319
 320         ctrl->status = status;
 321
 322         if (!cvq || !cvq->vq) {
 323                 PMD_INIT_LOG(ERR, "Control queue is not supported.");
 324                 return -1;
 325         }
 326
 327         rte_spinlock_lock(&cvq->lock);
 328         vq = cvq->vq;
 329
 330         PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
 331                 "vq->hw->cvq = %p vq = %p",
 332                 vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
 333
 334         if (vq->vq_free_cnt < pkt_num + 2 || pkt_num < 1) {
 335                 rte_spinlock_unlock(&cvq->lock);
 336                 return -1;
 337         }
 338
 339         memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
 340                 sizeof(struct virtio_pmd_ctrl));
 341
 342         if (vtpci_packed_queue(vq->hw))
 343                 result = virtio_send_command_packed(cvq, ctrl, dlen, pkt_num);
 344         else
 345                 result = virtio_send_command_split(cvq, ctrl, dlen, pkt_num);
 346
 347         rte_spinlock_unlock(&cvq->lock);
 348         return result->status;
 349 }
 350
 351 static int
 352 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
 353 {
 354         struct virtio_hw *hw = dev->data->dev_private;
 355         struct virtio_pmd_ctrl ctrl;
 356         int dlen[1];
 357         int ret;
 358
 359         ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
 360         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
 361         memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
 362
 363         dlen[0] = sizeof(uint16_t);
 364
 365         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 366         if (ret) {
 367                 PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
 368                           "failed, this is too late now...");
 369                 return -EINVAL;
 370         }
 371
 372         return 0;
 373 }
 374
 375 static void
 376 virtio_dev_queue_release(void *queue __rte_unused)
 377 {
 378         /* do nothing */
 379 }
 380
 381 static uint16_t
 382 virtio_get_nr_vq(struct virtio_hw *hw)
 383 {
 384         uint16_t nr_vq = hw->max_queue_pairs * 2;
 385
 386         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
 387                 nr_vq += 1;
 388
 389         return nr_vq;
 390 }
 391
 392 static void
 393 virtio_init_vring(struct virtqueue *vq)
 394 {
 395         int size = vq->vq_nentries;
 396         uint8_t *ring_mem = vq->vq_ring_virt_mem;
 397
 398         PMD_INIT_FUNC_TRACE();
 399
 400         memset(ring_mem, 0, vq->vq_ring_size);
 401
 402         vq->vq_used_cons_idx = 0;
 403         vq->vq_desc_head_idx = 0;
 404         vq->vq_avail_idx = 0;
 405         vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
 406         vq->vq_free_cnt = vq->vq_nentries;
 407         memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
 408         if (vtpci_packed_queue(vq->hw)) {
 409                 vring_init_packed(&vq->vq_packed.ring, ring_mem,
 410                                   VIRTIO_PCI_VRING_ALIGN, size);
 411                 vring_desc_init_packed(vq, size);
 412         } else {
 413                 struct vring *vr = &vq->vq_split.ring;
 414
 415                 vring_init_split(vr, ring_mem, VIRTIO_PCI_VRING_ALIGN, size);
 416                 vring_desc_init_split(vr->desc, size);
 417         }
 418         /*
 419          * Disable device(host) interrupting guest
 420          */
 421         virtqueue_disable_intr(vq);
 422 }
 423
 424 static int
 425 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
 426 {
 427         char vq_name[VIRTQUEUE_MAX_NAME_SZ];
 428         char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
 429         const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
 430         unsigned int vq_size, size;
 431         struct virtio_hw *hw = dev->data->dev_private;
 432         struct virtnet_rx *rxvq = NULL;
 433         struct virtnet_tx *txvq = NULL;
 434         struct virtnet_ctl *cvq = NULL;
 435         struct virtqueue *vq;
 436         size_t sz_hdr_mz = 0;
 437         void *sw_ring = NULL;
 438         int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
 439         int ret;
 440         int numa_node = dev->device->numa_node;
 441
 442         PMD_INIT_LOG(INFO, "setting up queue: %u on NUMA node %d",
 443                         vtpci_queue_idx, numa_node);
 444
 445         /*
 446          * Read the virtqueue size from the Queue Size field
 447          * Always power of 2 and if 0 virtqueue does not exist
 448          */
 449         vq_size = VTPCI_OPS(hw)->get_queue_num(hw, vtpci_queue_idx);
 450         PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
 451         if (vq_size == 0) {
 452                 PMD_INIT_LOG(ERR, "virtqueue does not exist");
 453                 return -EINVAL;
 454         }
 455
 456         if (!vtpci_packed_queue(hw) && !rte_is_power_of_2(vq_size)) {
 457                 PMD_INIT_LOG(ERR, "split virtqueue size is not power of 2");
 458                 return -EINVAL;
 459         }
 460
 461         snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
 462                  dev->data->port_id, vtpci_queue_idx);
 463
 464         size = RTE_ALIGN_CEIL(sizeof(*vq) +
 465                                 vq_size * sizeof(struct vq_desc_extra),
 466                                 RTE_CACHE_LINE_SIZE);
 467         if (queue_type == VTNET_TQ) {
 468                 /*
 469                  * For each xmit packet, allocate a virtio_net_hdr
 470                  * and indirect ring elements
 471                  */
 472                 sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
 473         } else if (queue_type == VTNET_CQ) {
 474                 /* Allocate a page for control vq command, data and status */
 475                 sz_hdr_mz = PAGE_SIZE;
 476         }
 477
 478         vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
 479                                 numa_node);
 480         if (vq == NULL) {
 481                 PMD_INIT_LOG(ERR, "can not allocate vq");
 482                 return -ENOMEM;
 483         }
 484         hw->vqs[vtpci_queue_idx] = vq;
 485
 486         vq->hw = hw;
 487         vq->vq_queue_index = vtpci_queue_idx;
 488         vq->vq_nentries = vq_size;
 489         if (vtpci_packed_queue(hw)) {
 490                 vq->vq_packed.used_wrap_counter = 1;
 491                 vq->vq_packed.cached_flags = VRING_PACKED_DESC_F_AVAIL;
 492                 vq->vq_packed.event_flags_shadow = 0;
 493                 if (queue_type == VTNET_RQ)
 494                         vq->vq_packed.cached_flags |= VRING_DESC_F_WRITE;
 495         }
 496
 497         /*
 498          * Reserve a memzone for vring elements
 499          */
 500         size = vring_size(hw, vq_size, VIRTIO_PCI_VRING_ALIGN);
 501         vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
 502         PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
 503                      size, vq->vq_ring_size);
 504
 505         mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
 506                         numa_node, RTE_MEMZONE_IOVA_CONTIG,
 507                         VIRTIO_PCI_VRING_ALIGN);
 508         if (mz == NULL) {
 509                 if (rte_errno == EEXIST)
 510                         mz = rte_memzone_lookup(vq_name);
 511                 if (mz == NULL) {
 512                         ret = -ENOMEM;
 513                         goto fail_q_alloc;
 514                 }
 515         }
 516
 517         memset(mz->addr, 0, mz->len);
 518
 519         vq->vq_ring_mem = mz->iova;
 520         vq->vq_ring_virt_mem = mz->addr;
 521         PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%" PRIx64,
 522                      (uint64_t)mz->iova);
 523         PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
 524                      (uint64_t)(uintptr_t)mz->addr);
 525
 526         virtio_init_vring(vq);
 527
 528         if (sz_hdr_mz) {
 529                 snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
 530                          dev->data->port_id, vtpci_queue_idx);
 531                 hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
 532                                 numa_node, RTE_MEMZONE_IOVA_CONTIG,
 533                                 RTE_CACHE_LINE_SIZE);
 534                 if (hdr_mz == NULL) {
 535                         if (rte_errno == EEXIST)
 536                                 hdr_mz = rte_memzone_lookup(vq_hdr_name);
 537                         if (hdr_mz == NULL) {
 538                                 ret = -ENOMEM;
 539                                 goto fail_q_alloc;
 540                         }
 541                 }
 542         }
 543
 544         if (queue_type == VTNET_RQ) {
 545                 size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
 546                                sizeof(vq->sw_ring[0]);
 547
 548                 sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
 549                                 RTE_CACHE_LINE_SIZE, numa_node);
 550                 if (!sw_ring) {
 551                         PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
 552                         ret = -ENOMEM;
 553                         goto fail_q_alloc;
 554                 }
 555
 556                 vq->sw_ring = sw_ring;
 557                 rxvq = &vq->rxq;
 558                 rxvq->vq = vq;
 559                 rxvq->port_id = dev->data->port_id;
 560                 rxvq->mz = mz;
 561         } else if (queue_type == VTNET_TQ) {
 562                 txvq = &vq->txq;
 563                 txvq->vq = vq;
 564                 txvq->port_id = dev->data->port_id;
 565                 txvq->mz = mz;
 566                 txvq->virtio_net_hdr_mz = hdr_mz;
 567                 txvq->virtio_net_hdr_mem = hdr_mz->iova;
 568         } else if (queue_type == VTNET_CQ) {
 569                 cvq = &vq->cq;
 570                 cvq->vq = vq;
 571                 cvq->mz = mz;
 572                 cvq->virtio_net_hdr_mz = hdr_mz;
 573                 cvq->virtio_net_hdr_mem = hdr_mz->iova;
 574                 memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
 575
 576                 hw->cvq = cvq;
 577         }
 578
 579         if (queue_type == VTNET_TQ) {
 580                 struct virtio_tx_region *txr;
 581                 unsigned int i;
 582
 583                 txr = hdr_mz->addr;
 584                 memset(txr, 0, vq_size * sizeof(*txr));
 585                 for (i = 0; i < vq_size; i++) {
 586                         /* first indirect descriptor is always the tx header */
 587                         if (!vtpci_packed_queue(hw)) {
 588                                 struct vring_desc *start_dp = txr[i].tx_indir;
 589                                 vring_desc_init_split(start_dp,
 590                                                       RTE_DIM(txr[i].tx_indir));
 591                                 start_dp->addr = txvq->virtio_net_hdr_mem
 592                                         + i * sizeof(*txr)
 593                                         + offsetof(struct virtio_tx_region,
 594                                                    tx_hdr);
 595                                 start_dp->len = hw->vtnet_hdr_size;
 596                                 start_dp->flags = VRING_DESC_F_NEXT;
 597                         } else {
 598                                 struct vring_packed_desc *start_dp =
 599                                         txr[i].tx_packed_indir;
 600                                 vring_desc_init_indirect_packed(start_dp,
 601                                       RTE_DIM(txr[i].tx_packed_indir));
 602                                 start_dp->addr = txvq->virtio_net_hdr_mem
 603                                         + i * sizeof(*txr)
 604                                         + offsetof(struct virtio_tx_region,
 605                                                    tx_hdr);
 606                                 start_dp->len = hw->vtnet_hdr_size;
 607                         }
 608                 }
 609         }
 610
 611         if (VTPCI_OPS(hw)->setup_queue(hw, vq) < 0) {
 612                 PMD_INIT_LOG(ERR, "setup_queue failed");
 613                 return -EINVAL;
 614         }
 615
 616         return 0;
 617
 618 fail_q_alloc:
 619         rte_free(sw_ring);
 620         rte_memzone_free(hdr_mz);
 621         rte_memzone_free(mz);
 622         rte_free(vq);
 623
 624         return ret;
 625 }
 626
 627 static void
 628 virtio_free_queues(struct virtio_hw *hw)
 629 {
 630         uint16_t nr_vq = virtio_get_nr_vq(hw);
 631         struct virtqueue *vq;
 632         int queue_type;
 633         uint16_t i;
 634
 635         if (hw->vqs == NULL)
 636                 return;
 637
 638         for (i = 0; i < nr_vq; i++) {
 639                 vq = hw->vqs[i];
 640                 if (!vq)
 641                         continue;
 642
 643                 queue_type = virtio_get_queue_type(hw, i);
 644                 if (queue_type == VTNET_RQ) {
 645                         rte_free(vq->sw_ring);
 646                         rte_memzone_free(vq->rxq.mz);
 647                 } else if (queue_type == VTNET_TQ) {
 648                         rte_memzone_free(vq->txq.mz);
 649                         rte_memzone_free(vq->txq.virtio_net_hdr_mz);
 650                 } else {
 651                         rte_memzone_free(vq->cq.mz);
 652                         rte_memzone_free(vq->cq.virtio_net_hdr_mz);
 653                 }
 654
 655                 rte_free(vq);
 656                 hw->vqs[i] = NULL;
 657         }
 658
 659         rte_free(hw->vqs);
 660         hw->vqs = NULL;
 661 }
 662
 663 static int
 664 virtio_alloc_queues(struct rte_eth_dev *dev)
 665 {
 666         struct virtio_hw *hw = dev->data->dev_private;
 667         uint16_t nr_vq = virtio_get_nr_vq(hw);
 668         uint16_t i;
 669         int ret;
 670
 671         hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
 672         if (!hw->vqs) {
 673                 PMD_INIT_LOG(ERR, "failed to allocate vqs");
 674                 return -ENOMEM;
 675         }
 676
 677         for (i = 0; i < nr_vq; i++) {
 678                 ret = virtio_init_queue(dev, i);
 679                 if (ret < 0) {
 680                         virtio_free_queues(hw);
 681                         return ret;
 682                 }
 683         }
 684
 685         return 0;
 686 }
 687
 688 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
 689
 690 int
 691 virtio_dev_close(struct rte_eth_dev *dev)
 692 {
 693         struct virtio_hw *hw = dev->data->dev_private;
 694         struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
 695
 696         PMD_INIT_LOG(DEBUG, "virtio_dev_close");
 697         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 698                 return 0;
 699
 700         if (!hw->opened)
 701                 return 0;
 702         hw->opened = false;
 703
 704         /* reset the NIC */
 705         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
 706                 VTPCI_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
 707         if (intr_conf->rxq)
 708                 virtio_queues_unbind_intr(dev);
 709
 710         if (intr_conf->lsc || intr_conf->rxq) {
 711                 virtio_intr_disable(dev);
 712                 rte_intr_efd_disable(dev->intr_handle);
 713                 rte_free(dev->intr_handle->intr_vec);
 714                 dev->intr_handle->intr_vec = NULL;
 715         }
 716
 717         vtpci_reset(hw);
 718         virtio_dev_free_mbufs(dev);
 719         virtio_free_queues(hw);
 720
 721         return VTPCI_OPS(hw)->dev_close(hw);
 722 }
 723
 724 static int
 725 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
 726 {
 727         struct virtio_hw *hw = dev->data->dev_private;
 728         struct virtio_pmd_ctrl ctrl;
 729         int dlen[1];
 730         int ret;
 731
 732         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 733                 PMD_INIT_LOG(INFO, "host does not support rx control");
 734                 return -ENOTSUP;
 735         }
 736
 737         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 738         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 739         ctrl.data[0] = 1;
 740         dlen[0] = 1;
 741
 742         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 743         if (ret) {
 744                 PMD_INIT_LOG(ERR, "Failed to enable promisc");
 745                 return -EAGAIN;
 746         }
 747
 748         return 0;
 749 }
 750
 751 static int
 752 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
 753 {
 754         struct virtio_hw *hw = dev->data->dev_private;
 755         struct virtio_pmd_ctrl ctrl;
 756         int dlen[1];
 757         int ret;
 758
 759         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 760                 PMD_INIT_LOG(INFO, "host does not support rx control");
 761                 return -ENOTSUP;
 762         }
 763
 764         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 765         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 766         ctrl.data[0] = 0;
 767         dlen[0] = 1;
 768
 769         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 770         if (ret) {
 771                 PMD_INIT_LOG(ERR, "Failed to disable promisc");
 772                 return -EAGAIN;
 773         }
 774
 775         return 0;
 776 }
 777
 778 static int
 779 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
 780 {
 781         struct virtio_hw *hw = dev->data->dev_private;
 782         struct virtio_pmd_ctrl ctrl;
 783         int dlen[1];
 784         int ret;
 785
 786         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 787                 PMD_INIT_LOG(INFO, "host does not support rx control");
 788                 return -ENOTSUP;
 789         }
 790
 791         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 792         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 793         ctrl.data[0] = 1;
 794         dlen[0] = 1;
 795
 796         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 797         if (ret) {
 798                 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
 799                 return -EAGAIN;
 800         }
 801
 802         return 0;
 803 }
 804
 805 static int
 806 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
 807 {
 808         struct virtio_hw *hw = dev->data->dev_private;
 809         struct virtio_pmd_ctrl ctrl;
 810         int dlen[1];
 811         int ret;
 812
 813         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 814                 PMD_INIT_LOG(INFO, "host does not support rx control");
 815                 return -ENOTSUP;
 816         }
 817
 818         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 819         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 820         ctrl.data[0] = 0;
 821         dlen[0] = 1;
 822
 823         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 824         if (ret) {
 825                 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
 826                 return -EAGAIN;
 827         }
 828
 829         return 0;
 830 }
 831
 832 #define VLAN_TAG_LEN           4    /* 802.3ac tag (not DMA'd) */
 833 static int
 834 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 835 {
 836         struct virtio_hw *hw = dev->data->dev_private;
 837         uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
 838                                  hw->vtnet_hdr_size;
 839         uint32_t frame_size = mtu + ether_hdr_len;
 840         uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
 841
 842         max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
 843
 844         if (mtu < RTE_ETHER_MIN_MTU || frame_size > max_frame_size) {
 845                 PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
 846                         RTE_ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
 847                 return -EINVAL;
 848         }
 849         return 0;
 850 }
 851
 852 static int
 853 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
 854 {
 855         struct virtio_hw *hw = dev->data->dev_private;
 856         struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
 857         struct virtqueue *vq = rxvq->vq;
 858
 859         virtqueue_enable_intr(vq);
 860         virtio_mb(hw->weak_barriers);
 861         return 0;
 862 }
 863
 864 static int
 865 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
 866 {
 867         struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
 868         struct virtqueue *vq = rxvq->vq;
 869
 870         virtqueue_disable_intr(vq);
 871         return 0;
 872 }
 873
 874 /*
 875  * dev_ops for virtio, bare necessities for basic operation
 876  */
 877 static const struct eth_dev_ops virtio_eth_dev_ops = {
 878         .dev_configure           = virtio_dev_configure,
 879         .dev_start               = virtio_dev_start,
 880         .dev_stop                = virtio_dev_stop,
 881         .dev_close               = virtio_dev_close,
 882         .promiscuous_enable      = virtio_dev_promiscuous_enable,
 883         .promiscuous_disable     = virtio_dev_promiscuous_disable,
 884         .allmulticast_enable     = virtio_dev_allmulticast_enable,
 885         .allmulticast_disable    = virtio_dev_allmulticast_disable,
 886         .mtu_set                 = virtio_mtu_set,
 887         .dev_infos_get           = virtio_dev_info_get,
 888         .stats_get               = virtio_dev_stats_get,
 889         .xstats_get              = virtio_dev_xstats_get,
 890         .xstats_get_names        = virtio_dev_xstats_get_names,
 891         .stats_reset             = virtio_dev_stats_reset,
 892         .xstats_reset            = virtio_dev_stats_reset,
 893         .link_update             = virtio_dev_link_update,
 894         .vlan_offload_set        = virtio_dev_vlan_offload_set,
 895         .rx_queue_setup          = virtio_dev_rx_queue_setup,
 896         .rx_queue_intr_enable    = virtio_dev_rx_queue_intr_enable,
 897         .rx_queue_intr_disable   = virtio_dev_rx_queue_intr_disable,
 898         .rx_queue_release        = virtio_dev_queue_release,
 899         .tx_queue_setup          = virtio_dev_tx_queue_setup,
 900         .tx_queue_release        = virtio_dev_queue_release,
 901         /* collect stats per queue */
 902         .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
 903         .vlan_filter_set         = virtio_vlan_filter_set,
 904         .mac_addr_add            = virtio_mac_addr_add,
 905         .mac_addr_remove         = virtio_mac_addr_remove,
 906         .mac_addr_set            = virtio_mac_addr_set,
 907 };
 908
 909 /*
 910  * dev_ops for virtio-user in secondary processes, as we just have
 911  * some limited supports currently.
 912  */
 913 const struct eth_dev_ops virtio_user_secondary_eth_dev_ops = {
 914         .dev_infos_get           = virtio_dev_info_get,
 915         .stats_get               = virtio_dev_stats_get,
 916         .xstats_get              = virtio_dev_xstats_get,
 917         .xstats_get_names        = virtio_dev_xstats_get_names,
 918         .stats_reset             = virtio_dev_stats_reset,
 919         .xstats_reset            = virtio_dev_stats_reset,
 920         /* collect stats per queue */
 921         .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
 922 };
 923
 924 static void
 925 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 926 {
 927         unsigned i;
 928
 929         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 930                 const struct virtnet_tx *txvq = dev->data->tx_queues[i];
 931                 if (txvq == NULL)
 932                         continue;
 933
 934                 stats->opackets += txvq->stats.packets;
 935                 stats->obytes += txvq->stats.bytes;
 936
 937                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 938                         stats->q_opackets[i] = txvq->stats.packets;
 939                         stats->q_obytes[i] = txvq->stats.bytes;
 940                 }
 941         }
 942
 943         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 944                 const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
 945                 if (rxvq == NULL)
 946                         continue;
 947
 948                 stats->ipackets += rxvq->stats.packets;
 949                 stats->ibytes += rxvq->stats.bytes;
 950                 stats->ierrors += rxvq->stats.errors;
 951
 952                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 953                         stats->q_ipackets[i] = rxvq->stats.packets;
 954                         stats->q_ibytes[i] = rxvq->stats.bytes;
 955                 }
 956         }
 957
 958         stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
 959 }
 960
 961 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
 962                                        struct rte_eth_xstat_name *xstats_names,
 963                                        __rte_unused unsigned limit)
 964 {
 965         unsigned i;
 966         unsigned count = 0;
 967         unsigned t;
 968
 969         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
 970                 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
 971
 972         if (xstats_names != NULL) {
 973                 /* Note: limit checked in rte_eth_xstats_names() */
 974
 975                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
 976                         struct virtnet_rx *rxvq = dev->data->rx_queues[i];
 977                         if (rxvq == NULL)
 978                                 continue;
 979                         for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
 980                                 snprintf(xstats_names[count].name,
 981                                         sizeof(xstats_names[count].name),
 982                                         "rx_q%u_%s", i,
 983                                         rte_virtio_rxq_stat_strings[t].name);
 984                                 count++;
 985                         }
 986                 }
 987
 988                 for (i = 0; i < dev->data->nb_tx_queues; i++) {
 989                         struct virtnet_tx *txvq = dev->data->tx_queues[i];
 990                         if (txvq == NULL)
 991                                 continue;
 992                         for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
 993                                 snprintf(xstats_names[count].name,
 994                                         sizeof(xstats_names[count].name),
 995                                         "tx_q%u_%s", i,
 996                                         rte_virtio_txq_stat_strings[t].name);
 997                                 count++;
 998                         }
 999                 }
1000                 return count;
1001         }
1002         return nstats;
1003 }
1004
1005 static int
1006 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1007                       unsigned n)
1008 {
1009         unsigned i;
1010         unsigned count = 0;
1011
1012         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1013                 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1014
1015         if (n < nstats)
1016                 return nstats;
1017
1018         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1019                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1020
1021                 if (rxvq == NULL)
1022                         continue;
1023
1024                 unsigned t;
1025
1026                 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1027                         xstats[count].value = *(uint64_t *)(((char *)rxvq) +
1028                                 rte_virtio_rxq_stat_strings[t].offset);
1029                         xstats[count].id = count;
1030                         count++;
1031                 }
1032         }
1033
1034         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1035                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1036
1037                 if (txvq == NULL)
1038                         continue;
1039
1040                 unsigned t;
1041
1042                 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1043                         xstats[count].value = *(uint64_t *)(((char *)txvq) +
1044                                 rte_virtio_txq_stat_strings[t].offset);
1045                         xstats[count].id = count;
1046                         count++;
1047                 }
1048         }
1049
1050         return count;
1051 }
1052
1053 static int
1054 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1055 {
1056         virtio_update_stats(dev, stats);
1057
1058         return 0;
1059 }
1060
1061 static int
1062 virtio_dev_stats_reset(struct rte_eth_dev *dev)
1063 {
1064         unsigned int i;
1065
1066         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1067                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1068                 if (txvq == NULL)
1069                         continue;
1070
1071                 txvq->stats.packets = 0;
1072                 txvq->stats.bytes = 0;
1073                 txvq->stats.multicast = 0;
1074                 txvq->stats.broadcast = 0;
1075                 memset(txvq->stats.size_bins, 0,
1076                        sizeof(txvq->stats.size_bins[0]) * 8);
1077         }
1078
1079         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1080                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1081                 if (rxvq == NULL)
1082                         continue;
1083
1084                 rxvq->stats.packets = 0;
1085                 rxvq->stats.bytes = 0;
1086                 rxvq->stats.errors = 0;
1087                 rxvq->stats.multicast = 0;
1088                 rxvq->stats.broadcast = 0;
1089                 memset(rxvq->stats.size_bins, 0,
1090                        sizeof(rxvq->stats.size_bins[0]) * 8);
1091         }
1092
1093         return 0;
1094 }
1095
1096 static void
1097 virtio_set_hwaddr(struct virtio_hw *hw)
1098 {
1099         vtpci_write_dev_config(hw,
1100                         offsetof(struct virtio_net_config, mac),
1101                         &hw->mac_addr, RTE_ETHER_ADDR_LEN);
1102 }
1103
1104 static void
1105 virtio_get_hwaddr(struct virtio_hw *hw)
1106 {
1107         if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1108                 vtpci_read_dev_config(hw,
1109                         offsetof(struct virtio_net_config, mac),
1110                         &hw->mac_addr, RTE_ETHER_ADDR_LEN);
1111         } else {
1112                 rte_eth_random_addr(&hw->mac_addr[0]);
1113                 virtio_set_hwaddr(hw);
1114         }
1115 }
1116
1117 static int
1118 virtio_mac_table_set(struct virtio_hw *hw,
1119                      const struct virtio_net_ctrl_mac *uc,
1120                      const struct virtio_net_ctrl_mac *mc)
1121 {
1122         struct virtio_pmd_ctrl ctrl;
1123         int err, len[2];
1124
1125         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1126                 PMD_DRV_LOG(INFO, "host does not support mac table");
1127                 return -1;
1128         }
1129
1130         ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1131         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1132
1133         len[0] = uc->entries * RTE_ETHER_ADDR_LEN + sizeof(uc->entries);
1134         memcpy(ctrl.data, uc, len[0]);
1135
1136         len[1] = mc->entries * RTE_ETHER_ADDR_LEN + sizeof(mc->entries);
1137         memcpy(ctrl.data + len[0], mc, len[1]);
1138
1139         err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1140         if (err != 0)
1141                 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1142         return err;
1143 }
1144
1145 static int
1146 virtio_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1147                     uint32_t index, uint32_t vmdq __rte_unused)
1148 {
1149         struct virtio_hw *hw = dev->data->dev_private;
1150         const struct rte_ether_addr *addrs = dev->data->mac_addrs;
1151         unsigned int i;
1152         struct virtio_net_ctrl_mac *uc, *mc;
1153
1154         if (index >= VIRTIO_MAX_MAC_ADDRS) {
1155                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1156                 return -EINVAL;
1157         }
1158
1159         uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1160                 sizeof(uc->entries));
1161         uc->entries = 0;
1162         mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1163                 sizeof(mc->entries));
1164         mc->entries = 0;
1165
1166         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1167                 const struct rte_ether_addr *addr
1168                         = (i == index) ? mac_addr : addrs + i;
1169                 struct virtio_net_ctrl_mac *tbl
1170                         = rte_is_multicast_ether_addr(addr) ? mc : uc;
1171
1172                 memcpy(&tbl->macs[tbl->entries++], addr, RTE_ETHER_ADDR_LEN);
1173         }
1174
1175         return virtio_mac_table_set(hw, uc, mc);
1176 }
1177
1178 static void
1179 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1180 {
1181         struct virtio_hw *hw = dev->data->dev_private;
1182         struct rte_ether_addr *addrs = dev->data->mac_addrs;
1183         struct virtio_net_ctrl_mac *uc, *mc;
1184         unsigned int i;
1185
1186         if (index >= VIRTIO_MAX_MAC_ADDRS) {
1187                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1188                 return;
1189         }
1190
1191         uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1192                 sizeof(uc->entries));
1193         uc->entries = 0;
1194         mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1195                 sizeof(mc->entries));
1196         mc->entries = 0;
1197
1198         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1199                 struct virtio_net_ctrl_mac *tbl;
1200
1201                 if (i == index || rte_is_zero_ether_addr(addrs + i))
1202                         continue;
1203
1204                 tbl = rte_is_multicast_ether_addr(addrs + i) ? mc : uc;
1205                 memcpy(&tbl->macs[tbl->entries++], addrs + i,
1206                         RTE_ETHER_ADDR_LEN);
1207         }
1208
1209         virtio_mac_table_set(hw, uc, mc);
1210 }
1211
1212 static int
1213 virtio_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1214 {
1215         struct virtio_hw *hw = dev->data->dev_private;
1216
1217         memcpy(hw->mac_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1218
1219         /* Use atomic update if available */
1220         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1221                 struct virtio_pmd_ctrl ctrl;
1222                 int len = RTE_ETHER_ADDR_LEN;
1223
1224                 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1225                 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1226
1227                 memcpy(ctrl.data, mac_addr, RTE_ETHER_ADDR_LEN);
1228                 return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1229         }
1230
1231         if (!vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1232                 return -ENOTSUP;
1233
1234         virtio_set_hwaddr(hw);
1235         return 0;
1236 }
1237
1238 static int
1239 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1240 {
1241         struct virtio_hw *hw = dev->data->dev_private;
1242         struct virtio_pmd_ctrl ctrl;
1243         int len;
1244
1245         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1246                 return -ENOTSUP;
1247
1248         ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1249         ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1250         memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1251         len = sizeof(vlan_id);
1252
1253         return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1254 }
1255
1256 static int
1257 virtio_intr_unmask(struct rte_eth_dev *dev)
1258 {
1259         struct virtio_hw *hw = dev->data->dev_private;
1260
1261         if (rte_intr_ack(dev->intr_handle) < 0)
1262                 return -1;
1263
1264         if (VTPCI_OPS(hw)->intr_detect)
1265                 VTPCI_OPS(hw)->intr_detect(hw);
1266
1267         return 0;
1268 }
1269
1270 static int
1271 virtio_intr_enable(struct rte_eth_dev *dev)
1272 {
1273         struct virtio_hw *hw = dev->data->dev_private;
1274
1275         if (rte_intr_enable(dev->intr_handle) < 0)
1276                 return -1;
1277
1278         if (VTPCI_OPS(hw)->intr_detect)
1279                 VTPCI_OPS(hw)->intr_detect(hw);
1280
1281         return 0;
1282 }
1283
1284 static int
1285 virtio_intr_disable(struct rte_eth_dev *dev)
1286 {
1287         struct virtio_hw *hw = dev->data->dev_private;
1288
1289         if (rte_intr_disable(dev->intr_handle) < 0)
1290                 return -1;
1291
1292         if (VTPCI_OPS(hw)->intr_detect)
1293                 VTPCI_OPS(hw)->intr_detect(hw);
1294
1295         return 0;
1296 }
1297
1298 static int
1299 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1300 {
1301         uint64_t host_features;
1302
1303         /* Prepare guest_features: feature that driver wants to support */
1304         PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1305                 req_features);
1306
1307         /* Read device(host) feature bits */
1308         host_features = VTPCI_OPS(hw)->get_features(hw);
1309         PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1310                 host_features);
1311
1312         /* If supported, ensure MTU value is valid before acknowledging it. */
1313         if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1314                 struct virtio_net_config config;
1315
1316                 vtpci_read_dev_config(hw,
1317                         offsetof(struct virtio_net_config, mtu),
1318                         &config.mtu, sizeof(config.mtu));
1319
1320                 if (config.mtu < RTE_ETHER_MIN_MTU)
1321                         req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1322         }
1323
1324         /*
1325          * Negotiate features: Subset of device feature bits are written back
1326          * guest feature bits.
1327          */
1328         hw->guest_features = req_features;
1329         hw->guest_features = vtpci_negotiate_features(hw, host_features);
1330         PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1331                 hw->guest_features);
1332
1333         if (VTPCI_OPS(hw)->features_ok(hw) < 0)
1334                 return -1;
1335
1336         if (vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1337                 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1338
1339                 if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1340                         PMD_INIT_LOG(ERR, "Failed to set FEATURES_OK status!");
1341                         return -1;
1342                 }
1343         }
1344
1345         hw->req_guest_features = req_features;
1346
1347         return 0;
1348 }
1349
1350 int
1351 virtio_dev_pause(struct rte_eth_dev *dev)
1352 {
1353         struct virtio_hw *hw = dev->data->dev_private;
1354
1355         rte_spinlock_lock(&hw->state_lock);
1356
1357         if (hw->started == 0) {
1358                 /* Device is just stopped. */
1359                 rte_spinlock_unlock(&hw->state_lock);
1360                 return -1;
1361         }
1362         hw->started = 0;
1363         /*
1364          * Prevent the worker threads from touching queues to avoid contention,
1365          * 1 ms should be enough for the ongoing Tx function to finish.
1366          */
1367         rte_delay_ms(1);
1368         return 0;
1369 }
1370
1371 /*
1372  * Recover hw state to let the worker threads continue.
1373  */
1374 void
1375 virtio_dev_resume(struct rte_eth_dev *dev)
1376 {
1377         struct virtio_hw *hw = dev->data->dev_private;
1378
1379         hw->started = 1;
1380         rte_spinlock_unlock(&hw->state_lock);
1381 }
1382
1383 /*
1384  * Should be called only after device is paused.
1385  */
1386 int
1387 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1388                 int nb_pkts)
1389 {
1390         struct virtio_hw *hw = dev->data->dev_private;
1391         struct virtnet_tx *txvq = dev->data->tx_queues[0];
1392         int ret;
1393
1394         hw->inject_pkts = tx_pkts;
1395         ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1396         hw->inject_pkts = NULL;
1397
1398         return ret;
1399 }
1400
1401 static void
1402 virtio_notify_peers(struct rte_eth_dev *dev)
1403 {
1404         struct virtio_hw *hw = dev->data->dev_private;
1405         struct virtnet_rx *rxvq;
1406         struct rte_mbuf *rarp_mbuf;
1407
1408         if (!dev->data->rx_queues)
1409                 return;
1410
1411         rxvq = dev->data->rx_queues[0];
1412         if (!rxvq)
1413                 return;
1414
1415         rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1416                         (struct rte_ether_addr *)hw->mac_addr);
1417         if (rarp_mbuf == NULL) {
1418                 PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1419                 return;
1420         }
1421
1422         /* If virtio port just stopped, no need to send RARP */
1423         if (virtio_dev_pause(dev) < 0) {
1424                 rte_pktmbuf_free(rarp_mbuf);
1425                 return;
1426         }
1427
1428         virtio_inject_pkts(dev, &rarp_mbuf, 1);
1429         virtio_dev_resume(dev);
1430 }
1431
1432 static void
1433 virtio_ack_link_announce(struct rte_eth_dev *dev)
1434 {
1435         struct virtio_hw *hw = dev->data->dev_private;
1436         struct virtio_pmd_ctrl ctrl;
1437
1438         ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1439         ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1440
1441         virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1442 }
1443
1444 /*
1445  * Process virtio config changed interrupt. Call the callback
1446  * if link state changed, generate gratuitous RARP packet if
1447  * the status indicates an ANNOUNCE.
1448  */
1449 void
1450 virtio_interrupt_handler(void *param)
1451 {
1452         struct rte_eth_dev *dev = param;
1453         struct virtio_hw *hw = dev->data->dev_private;
1454         uint8_t isr;
1455         uint16_t status;
1456
1457         /* Read interrupt status which clears interrupt */
1458         isr = vtpci_isr(hw);
1459         PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1460
1461         if (virtio_intr_unmask(dev) < 0)
1462                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1463
1464         if (isr & VIRTIO_PCI_ISR_CONFIG) {
1465                 if (virtio_dev_link_update(dev, 0) == 0)
1466                         rte_eth_dev_callback_process(dev,
1467                                                      RTE_ETH_EVENT_INTR_LSC,
1468                                                      NULL);
1469
1470                 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1471                         vtpci_read_dev_config(hw,
1472                                 offsetof(struct virtio_net_config, status),
1473                                 &status, sizeof(status));
1474                         if (status & VIRTIO_NET_S_ANNOUNCE) {
1475                                 virtio_notify_peers(dev);
1476                                 if (hw->cvq)
1477                                         virtio_ack_link_announce(dev);
1478                         }
1479                 }
1480         }
1481 }
1482
1483 /* set rx and tx handlers according to what is supported */
1484 static void
1485 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1486 {
1487         struct virtio_hw *hw = eth_dev->data->dev_private;
1488
1489         eth_dev->tx_pkt_prepare = virtio_xmit_pkts_prepare;
1490         if (vtpci_packed_queue(hw)) {
1491                 PMD_INIT_LOG(INFO,
1492                         "virtio: using packed ring %s Tx path on port %u",
1493                         hw->use_vec_tx ? "vectorized" : "standard",
1494                         eth_dev->data->port_id);
1495                 if (hw->use_vec_tx)
1496                         eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed_vec;
1497                 else
1498                         eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed;
1499         } else {
1500                 if (hw->use_inorder_tx) {
1501                         PMD_INIT_LOG(INFO, "virtio: using inorder Tx path on port %u",
1502                                 eth_dev->data->port_id);
1503                         eth_dev->tx_pkt_burst = virtio_xmit_pkts_inorder;
1504                 } else {
1505                         PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1506                                 eth_dev->data->port_id);
1507                         eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1508                 }
1509         }
1510
1511         if (vtpci_packed_queue(hw)) {
1512                 if (hw->use_vec_rx) {
1513                         PMD_INIT_LOG(INFO,
1514                                 "virtio: using packed ring vectorized Rx path on port %u",
1515                                 eth_dev->data->port_id);
1516                         eth_dev->rx_pkt_burst =
1517                                 &virtio_recv_pkts_packed_vec;
1518                 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1519                         PMD_INIT_LOG(INFO,
1520                                 "virtio: using packed ring mergeable buffer Rx path on port %u",
1521                                 eth_dev->data->port_id);
1522                         eth_dev->rx_pkt_burst =
1523                                 &virtio_recv_mergeable_pkts_packed;
1524                 } else {
1525                         PMD_INIT_LOG(INFO,
1526                                 "virtio: using packed ring standard Rx path on port %u",
1527                                 eth_dev->data->port_id);
1528                         eth_dev->rx_pkt_burst = &virtio_recv_pkts_packed;
1529                 }
1530         } else {
1531                 if (hw->use_vec_rx) {
1532                         PMD_INIT_LOG(INFO, "virtio: using vectorized Rx path on port %u",
1533                                 eth_dev->data->port_id);
1534                         eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1535                 } else if (hw->use_inorder_rx) {
1536                         PMD_INIT_LOG(INFO,
1537                                 "virtio: using inorder Rx path on port %u",
1538                                 eth_dev->data->port_id);
1539                         eth_dev->rx_pkt_burst = &virtio_recv_pkts_inorder;
1540                 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1541                         PMD_INIT_LOG(INFO,
1542                                 "virtio: using mergeable buffer Rx path on port %u",
1543                                 eth_dev->data->port_id);
1544                         eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1545                 } else {
1546                         PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1547                                 eth_dev->data->port_id);
1548                         eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1549                 }
1550         }
1551
1552 }
1553
1554 /* Only support 1:1 queue/interrupt mapping so far.
1555  * TODO: support n:1 queue/interrupt mapping when there are limited number of
1556  * interrupt vectors (<N+1).
1557  */
1558 static int
1559 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1560 {
1561         uint32_t i;
1562         struct virtio_hw *hw = dev->data->dev_private;
1563
1564         PMD_INIT_LOG(INFO, "queue/interrupt binding");
1565         for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1566                 dev->intr_handle->intr_vec[i] = i + 1;
1567                 if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1568                                                  VIRTIO_MSI_NO_VECTOR) {
1569                         PMD_DRV_LOG(ERR, "failed to set queue vector");
1570                         return -EBUSY;
1571                 }
1572         }
1573
1574         return 0;
1575 }
1576
1577 static void
1578 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1579 {
1580         uint32_t i;
1581         struct virtio_hw *hw = dev->data->dev_private;
1582
1583         PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1584         for (i = 0; i < dev->data->nb_rx_queues; ++i)
1585                 VTPCI_OPS(hw)->set_queue_irq(hw,
1586                                              hw->vqs[i * VTNET_CQ],
1587                                              VIRTIO_MSI_NO_VECTOR);
1588 }
1589
1590 static int
1591 virtio_configure_intr(struct rte_eth_dev *dev)
1592 {
1593         struct virtio_hw *hw = dev->data->dev_private;
1594
1595         if (!rte_intr_cap_multiple(dev->intr_handle)) {
1596                 PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1597                 return -ENOTSUP;
1598         }
1599
1600         if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1601                 PMD_INIT_LOG(ERR, "Fail to create eventfd");
1602                 return -1;
1603         }
1604
1605         if (!dev->intr_handle->intr_vec) {
1606                 dev->intr_handle->intr_vec =
1607                         rte_zmalloc("intr_vec",
1608                                     hw->max_queue_pairs * sizeof(int), 0);
1609                 if (!dev->intr_handle->intr_vec) {
1610                         PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1611                                      hw->max_queue_pairs);
1612                         return -ENOMEM;
1613                 }
1614         }
1615
1616         /* Re-register callback to update max_intr */
1617         rte_intr_callback_unregister(dev->intr_handle,
1618                                      virtio_interrupt_handler,
1619                                      dev);
1620         rte_intr_callback_register(dev->intr_handle,
1621                                    virtio_interrupt_handler,
1622                                    dev);
1623
1624         /* DO NOT try to remove this! This function will enable msix, or QEMU
1625          * will encounter SIGSEGV when DRIVER_OK is sent.
1626          * And for legacy devices, this should be done before queue/vec binding
1627          * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1628          * (22) will be ignored.
1629          */
1630         if (virtio_intr_enable(dev) < 0) {
1631                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1632                 return -1;
1633         }
1634
1635         if (virtio_queues_bind_intr(dev) < 0) {
1636                 PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1637                 return -1;
1638         }
1639
1640         return 0;
1641 }
1642 #define DUPLEX_UNKNOWN   0xff
1643 /* reset device and renegotiate features if needed */
1644 static int
1645 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1646 {
1647         struct virtio_hw *hw = eth_dev->data->dev_private;
1648         struct virtio_net_config *config;
1649         struct virtio_net_config local_config;
1650         int ret;
1651
1652         /* Reset the device although not necessary at startup */
1653         vtpci_reset(hw);
1654
1655         if (hw->vqs) {
1656                 virtio_dev_free_mbufs(eth_dev);
1657                 virtio_free_queues(hw);
1658         }
1659
1660         /* Tell the host we've noticed this device. */
1661         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1662
1663         /* Tell the host we've known how to drive the device. */
1664         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1665         if (virtio_negotiate_features(hw, req_features) < 0)
1666                 return -1;
1667
1668         hw->weak_barriers = !vtpci_with_feature(hw, VIRTIO_F_ORDER_PLATFORM);
1669
1670         /* If host does not support both status and MSI-X then disable LSC */
1671         if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) &&
1672             hw->use_msix != VIRTIO_MSIX_NONE)
1673                 eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1674         else
1675                 eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1676
1677         eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
1678
1679         /* Setting up rx_header size for the device */
1680         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1681             vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
1682             vtpci_with_feature(hw, VIRTIO_F_RING_PACKED))
1683                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1684         else
1685                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1686
1687         /* Copy the permanent MAC address to: virtio_hw */
1688         virtio_get_hwaddr(hw);
1689         rte_ether_addr_copy((struct rte_ether_addr *)hw->mac_addr,
1690                         &eth_dev->data->mac_addrs[0]);
1691         PMD_INIT_LOG(DEBUG,
1692                      "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1693                      hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1694                      hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1695
1696         if (hw->speed == ETH_SPEED_NUM_UNKNOWN) {
1697                 if (vtpci_with_feature(hw, VIRTIO_NET_F_SPEED_DUPLEX)) {
1698                         config = &local_config;
1699                         vtpci_read_dev_config(hw,
1700                                 offsetof(struct virtio_net_config, speed),
1701                                 &config->speed, sizeof(config->speed));
1702                         vtpci_read_dev_config(hw,
1703                                 offsetof(struct virtio_net_config, duplex),
1704                                 &config->duplex, sizeof(config->duplex));
1705                         hw->speed = config->speed;
1706                         hw->duplex = config->duplex;
1707                 }
1708         }
1709         if (hw->duplex == DUPLEX_UNKNOWN)
1710                 hw->duplex = ETH_LINK_FULL_DUPLEX;
1711         PMD_INIT_LOG(DEBUG, "link speed = %d, duplex = %d",
1712                 hw->speed, hw->duplex);
1713         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1714                 config = &local_config;
1715
1716                 vtpci_read_dev_config(hw,
1717                         offsetof(struct virtio_net_config, mac),
1718                         &config->mac, sizeof(config->mac));
1719
1720                 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1721                         vtpci_read_dev_config(hw,
1722                                 offsetof(struct virtio_net_config, status),
1723                                 &config->status, sizeof(config->status));
1724                 } else {
1725                         PMD_INIT_LOG(DEBUG,
1726                                      "VIRTIO_NET_F_STATUS is not supported");
1727                         config->status = 0;
1728                 }
1729
1730                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1731                         vtpci_read_dev_config(hw,
1732                                 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1733                                 &config->max_virtqueue_pairs,
1734                                 sizeof(config->max_virtqueue_pairs));
1735                 } else {
1736                         PMD_INIT_LOG(DEBUG,
1737                                      "VIRTIO_NET_F_MQ is not supported");
1738                         config->max_virtqueue_pairs = 1;
1739                 }
1740
1741                 hw->max_queue_pairs = config->max_virtqueue_pairs;
1742
1743                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1744                         vtpci_read_dev_config(hw,
1745                                 offsetof(struct virtio_net_config, mtu),
1746                                 &config->mtu,
1747                                 sizeof(config->mtu));
1748
1749                         /*
1750                          * MTU value has already been checked at negotiation
1751                          * time, but check again in case it has changed since
1752                          * then, which should not happen.
1753                          */
1754                         if (config->mtu < RTE_ETHER_MIN_MTU) {
1755                                 PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1756                                                 config->mtu);
1757                                 return -1;
1758                         }
1759
1760                         hw->max_mtu = config->mtu;
1761                         /* Set initial MTU to maximum one supported by vhost */
1762                         eth_dev->data->mtu = config->mtu;
1763
1764                 } else {
1765                         hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1766                                 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1767                 }
1768
1769                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1770                                 config->max_virtqueue_pairs);
1771                 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1772                 PMD_INIT_LOG(DEBUG,
1773                                 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1774                                 config->mac[0], config->mac[1],
1775                                 config->mac[2], config->mac[3],
1776                                 config->mac[4], config->mac[5]);
1777         } else {
1778                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1779                 hw->max_queue_pairs = 1;
1780                 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1781                         VLAN_TAG_LEN - hw->vtnet_hdr_size;
1782         }
1783
1784         ret = virtio_alloc_queues(eth_dev);
1785         if (ret < 0)
1786                 return ret;
1787
1788         if (eth_dev->data->dev_conf.intr_conf.rxq) {
1789                 if (virtio_configure_intr(eth_dev) < 0) {
1790                         PMD_INIT_LOG(ERR, "failed to configure interrupt");
1791                         virtio_free_queues(hw);
1792                         return -1;
1793                 }
1794         }
1795
1796         vtpci_reinit_complete(hw);
1797
1798         return 0;
1799 }
1800
1801
1802 static void
1803 virtio_set_vtpci_ops(struct virtio_hw *hw)
1804 {
1805 #ifdef RTE_VIRTIO_USER
1806         if (hw->bus_type == VIRTIO_BUS_USER)
1807                 VTPCI_OPS(hw) = &virtio_user_ops;
1808         else
1809 #endif
1810         if (hw->bus_type == VIRTIO_BUS_PCI_MODERN)
1811                 VTPCI_OPS(hw) = &modern_ops;
1812         else if (hw->bus_type == VIRTIO_BUS_PCI_LEGACY)
1813                 VTPCI_OPS(hw) = &legacy_ops;
1814
1815         return;
1816 }
1817
1818 /*
1819  * This function is based on probe() function in virtio_pci.c
1820  * It returns 0 on success.
1821  */
1822 int
1823 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1824 {
1825         struct virtio_hw *hw = eth_dev->data->dev_private;
1826         uint32_t speed = ETH_SPEED_NUM_UNKNOWN;
1827         int vectorized = 0;
1828         int ret;
1829
1830         if (sizeof(struct virtio_net_hdr_mrg_rxbuf) > RTE_PKTMBUF_HEADROOM) {
1831                 PMD_INIT_LOG(ERR,
1832                         "Not sufficient headroom required = %d, avail = %d",
1833                         (int)sizeof(struct virtio_net_hdr_mrg_rxbuf),
1834                         RTE_PKTMBUF_HEADROOM);
1835
1836                 return -1;
1837         }
1838
1839         eth_dev->dev_ops = &virtio_eth_dev_ops;
1840         eth_dev->rx_descriptor_done = virtio_dev_rx_queue_done;
1841
1842         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1843                 virtio_set_vtpci_ops(hw);
1844                 set_rxtx_funcs(eth_dev);
1845                 return 0;
1846         }
1847
1848         ret = virtio_dev_devargs_parse(eth_dev->device->devargs, &speed, &vectorized);
1849         if (ret < 0)
1850                 return ret;
1851         hw->speed = speed;
1852
1853         /* Allocate memory for storing MAC addresses */
1854         eth_dev->data->mac_addrs = rte_zmalloc("virtio",
1855                                 VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN, 0);
1856         if (eth_dev->data->mac_addrs == NULL) {
1857                 PMD_INIT_LOG(ERR,
1858                         "Failed to allocate %d bytes needed to store MAC addresses",
1859                         VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN);
1860                 return -ENOMEM;
1861         }
1862
1863         hw->port_id = eth_dev->data->port_id;
1864         rte_spinlock_init(&hw->state_lock);
1865
1866         /* reset device and negotiate default features */
1867         ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1868         if (ret < 0)
1869                 goto err_virtio_init;
1870
1871         if (vectorized) {
1872                 if (!vtpci_packed_queue(hw)) {
1873                         hw->use_vec_rx = 1;
1874                 } else {
1875 #if defined(CC_AVX512_SUPPORT) || defined(RTE_ARCH_ARM)
1876                         hw->use_vec_rx = 1;
1877                         hw->use_vec_tx = 1;
1878 #else
1879                         PMD_DRV_LOG(INFO,
1880                                 "building environment do not support packed ring vectorized");
1881 #endif
1882                 }
1883         }
1884
1885         hw->opened = true;
1886
1887         return 0;
1888
1889 err_virtio_init:
1890         rte_free(eth_dev->data->mac_addrs);
1891         eth_dev->data->mac_addrs = NULL;
1892         return ret;
1893 }
1894
1895 static uint32_t
1896 virtio_dev_speed_capa_get(uint32_t speed)
1897 {
1898         switch (speed) {
1899         case ETH_SPEED_NUM_10G:
1900                 return ETH_LINK_SPEED_10G;
1901         case ETH_SPEED_NUM_20G:
1902                 return ETH_LINK_SPEED_20G;
1903         case ETH_SPEED_NUM_25G:
1904                 return ETH_LINK_SPEED_25G;
1905         case ETH_SPEED_NUM_40G:
1906                 return ETH_LINK_SPEED_40G;
1907         case ETH_SPEED_NUM_50G:
1908                 return ETH_LINK_SPEED_50G;
1909         case ETH_SPEED_NUM_56G:
1910                 return ETH_LINK_SPEED_56G;
1911         case ETH_SPEED_NUM_100G:
1912                 return ETH_LINK_SPEED_100G;
1913         case ETH_SPEED_NUM_200G:
1914                 return ETH_LINK_SPEED_200G;
1915         default:
1916                 return 0;
1917         }
1918 }
1919
1920 static int vectorized_check_handler(__rte_unused const char *key,
1921                 const char *value, void *ret_val)
1922 {
1923         if (strcmp(value, "1") == 0)
1924                 *(int *)ret_val = 1;
1925         else
1926                 *(int *)ret_val = 0;
1927
1928         return 0;
1929 }
1930
1931 #define VIRTIO_ARG_SPEED      "speed"
1932 #define VIRTIO_ARG_VECTORIZED "vectorized"
1933
1934 static int
1935 link_speed_handler(const char *key __rte_unused,
1936                 const char *value, void *ret_val)
1937 {
1938         uint32_t val;
1939         if (!value || !ret_val)
1940                 return -EINVAL;
1941         val = strtoul(value, NULL, 0);
1942         /* validate input */
1943         if (virtio_dev_speed_capa_get(val) == 0)
1944                 return -EINVAL;
1945         *(uint32_t *)ret_val = val;
1946
1947         return 0;
1948 }
1949
1950
1951 static int
1952 virtio_dev_devargs_parse(struct rte_devargs *devargs, uint32_t *speed, int *vectorized)
1953 {
1954         struct rte_kvargs *kvlist;
1955         int ret = 0;
1956
1957         if (devargs == NULL)
1958                 return 0;
1959
1960         kvlist = rte_kvargs_parse(devargs->args, NULL);
1961         if (kvlist == NULL) {
1962                 PMD_INIT_LOG(ERR, "error when parsing param");
1963                 return 0;
1964         }
1965
1966         if (speed && rte_kvargs_count(kvlist, VIRTIO_ARG_SPEED) == 1) {
1967                 ret = rte_kvargs_process(kvlist,
1968                                         VIRTIO_ARG_SPEED,
1969                                         link_speed_handler, speed);
1970                 if (ret < 0) {
1971                         PMD_INIT_LOG(ERR, "Failed to parse %s",
1972                                         VIRTIO_ARG_SPEED);
1973                         goto exit;
1974                 }
1975         }
1976
1977         if (vectorized &&
1978                 rte_kvargs_count(kvlist, VIRTIO_ARG_VECTORIZED) == 1) {
1979                 ret = rte_kvargs_process(kvlist,
1980                                 VIRTIO_ARG_VECTORIZED,
1981                                 vectorized_check_handler, vectorized);
1982                 if (ret < 0) {
1983                         PMD_INIT_LOG(ERR, "Failed to parse %s",
1984                                         VIRTIO_ARG_VECTORIZED);
1985                         goto exit;
1986                 }
1987         }
1988
1989 exit:
1990         rte_kvargs_free(kvlist);
1991         return ret;
1992 }
1993
1994 static bool
1995 rx_offload_enabled(struct virtio_hw *hw)
1996 {
1997         return vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM) ||
1998                 vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
1999                 vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6);
2000 }
2001
2002 static bool
2003 tx_offload_enabled(struct virtio_hw *hw)
2004 {
2005         return vtpci_with_feature(hw, VIRTIO_NET_F_CSUM) ||
2006                 vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO4) ||
2007                 vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO6);
2008 }
2009
2010 /*
2011  * Configure virtio device
2012  * It returns 0 on success.
2013  */
2014 static int
2015 virtio_dev_configure(struct rte_eth_dev *dev)
2016 {
2017         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2018         const struct rte_eth_txmode *txmode = &dev->data->dev_conf.txmode;
2019         struct virtio_hw *hw = dev->data->dev_private;
2020         uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
2021                 hw->vtnet_hdr_size;
2022         uint64_t rx_offloads = rxmode->offloads;
2023         uint64_t tx_offloads = txmode->offloads;
2024         uint64_t req_features;
2025         int ret;
2026
2027         PMD_INIT_LOG(DEBUG, "configure");
2028         req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
2029
2030         if (rxmode->mq_mode != ETH_MQ_RX_NONE) {
2031                 PMD_DRV_LOG(ERR,
2032                         "Unsupported Rx multi queue mode %d",
2033                         rxmode->mq_mode);
2034                 return -EINVAL;
2035         }
2036
2037         if (txmode->mq_mode != ETH_MQ_TX_NONE) {
2038                 PMD_DRV_LOG(ERR,
2039                         "Unsupported Tx multi queue mode %d",
2040                         txmode->mq_mode);
2041                 return -EINVAL;
2042         }
2043
2044         if (dev->data->dev_conf.intr_conf.rxq) {
2045                 ret = virtio_init_device(dev, hw->req_guest_features);
2046                 if (ret < 0)
2047                         return ret;
2048         }
2049
2050         if (rxmode->max_rx_pkt_len > hw->max_mtu + ether_hdr_len)
2051                 req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
2052
2053         if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2054                            DEV_RX_OFFLOAD_TCP_CKSUM))
2055                 req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
2056
2057         if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
2058                 req_features |=
2059                         (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2060                         (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2061
2062         if (tx_offloads & (DEV_TX_OFFLOAD_UDP_CKSUM |
2063                            DEV_TX_OFFLOAD_TCP_CKSUM))
2064                 req_features |= (1ULL << VIRTIO_NET_F_CSUM);
2065
2066         if (tx_offloads & DEV_TX_OFFLOAD_TCP_TSO)
2067                 req_features |=
2068                         (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2069                         (1ULL << VIRTIO_NET_F_HOST_TSO6);
2070
2071         /* if request features changed, reinit the device */
2072         if (req_features != hw->req_guest_features) {
2073                 ret = virtio_init_device(dev, req_features);
2074                 if (ret < 0)
2075                         return ret;
2076         }
2077
2078         if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2079                             DEV_RX_OFFLOAD_TCP_CKSUM)) &&
2080                 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
2081                 PMD_DRV_LOG(ERR,
2082                         "rx checksum not available on this host");
2083                 return -ENOTSUP;
2084         }
2085
2086         if ((rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) &&
2087                 (!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2088                  !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
2089                 PMD_DRV_LOG(ERR,
2090                         "Large Receive Offload not available on this host");
2091                 return -ENOTSUP;
2092         }
2093
2094         /* start control queue */
2095         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
2096                 virtio_dev_cq_start(dev);
2097
2098         if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2099                 hw->vlan_strip = 1;
2100
2101         if ((rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
2102             && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2103                 PMD_DRV_LOG(ERR,
2104                             "vlan filtering not available on this host");
2105                 return -ENOTSUP;
2106         }
2107
2108         hw->has_tx_offload = tx_offload_enabled(hw);
2109         hw->has_rx_offload = rx_offload_enabled(hw);
2110
2111         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2112                 /* Enable vector (0) for Link State Intrerrupt */
2113                 if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
2114                                 VIRTIO_MSI_NO_VECTOR) {
2115                         PMD_DRV_LOG(ERR, "failed to set config vector");
2116                         return -EBUSY;
2117                 }
2118
2119         if (vtpci_packed_queue(hw)) {
2120 #if defined(RTE_ARCH_X86_64) && defined(CC_AVX512_SUPPORT)
2121                 if ((hw->use_vec_rx || hw->use_vec_tx) &&
2122                     (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F) ||
2123                      !vtpci_with_feature(hw, VIRTIO_F_IN_ORDER) ||
2124                      !vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
2125                      rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_512)) {
2126                         PMD_DRV_LOG(INFO,
2127                                 "disabled packed ring vectorized path for requirements not met");
2128                         hw->use_vec_rx = 0;
2129                         hw->use_vec_tx = 0;
2130                 }
2131 #elif defined(RTE_ARCH_ARM)
2132                 if ((hw->use_vec_rx || hw->use_vec_tx) &&
2133                     (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON) ||
2134                      !vtpci_with_feature(hw, VIRTIO_F_IN_ORDER) ||
2135                      !vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
2136                      rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_128)) {
2137                         PMD_DRV_LOG(INFO,
2138                                 "disabled packed ring vectorized path for requirements not met");
2139                         hw->use_vec_rx = 0;
2140                         hw->use_vec_tx = 0;
2141                 }
2142 #else
2143                 hw->use_vec_rx = 0;
2144                 hw->use_vec_tx = 0;
2145 #endif
2146
2147                 if (hw->use_vec_rx) {
2148                         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2149                                 PMD_DRV_LOG(INFO,
2150                                         "disabled packed ring vectorized rx for mrg_rxbuf enabled");
2151                                 hw->use_vec_rx = 0;
2152                         }
2153
2154                         if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) {
2155                                 PMD_DRV_LOG(INFO,
2156                                         "disabled packed ring vectorized rx for TCP_LRO enabled");
2157                                 hw->use_vec_rx = 0;
2158                         }
2159                 }
2160         } else {
2161                 if (vtpci_with_feature(hw, VIRTIO_F_IN_ORDER)) {
2162                         hw->use_inorder_tx = 1;
2163                         hw->use_inorder_rx = 1;
2164                         hw->use_vec_rx = 0;
2165                 }
2166
2167                 if (hw->use_vec_rx) {
2168 #if defined RTE_ARCH_ARM
2169                         if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
2170                                 PMD_DRV_LOG(INFO,
2171                                         "disabled split ring vectorized path for requirement not met");
2172                                 hw->use_vec_rx = 0;
2173                         }
2174 #endif
2175                         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2176                                 PMD_DRV_LOG(INFO,
2177                                         "disabled split ring vectorized rx for mrg_rxbuf enabled");
2178                                 hw->use_vec_rx = 0;
2179                         }
2180
2181                         if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2182                                            DEV_RX_OFFLOAD_TCP_CKSUM |
2183                                            DEV_RX_OFFLOAD_TCP_LRO |
2184                                            DEV_RX_OFFLOAD_VLAN_STRIP)) {
2185                                 PMD_DRV_LOG(INFO,
2186                                         "disabled split ring vectorized rx for offloading enabled");
2187                                 hw->use_vec_rx = 0;
2188                         }
2189
2190                         if (rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_128) {
2191                                 PMD_DRV_LOG(INFO,
2192                                         "disabled split ring vectorized rx, max SIMD bitwidth too low");
2193                                 hw->use_vec_rx = 0;
2194                         }
2195                 }
2196         }
2197
2198         return 0;
2199 }
2200
2201
2202 static int
2203 virtio_dev_start(struct rte_eth_dev *dev)
2204 {
2205         uint16_t nb_queues, i;
2206         struct virtnet_rx *rxvq;
2207         struct virtnet_tx *txvq __rte_unused;
2208         struct virtio_hw *hw = dev->data->dev_private;
2209         int ret;
2210
2211         /* Finish the initialization of the queues */
2212         for (i = 0; i < dev->data->nb_rx_queues; i++) {
2213                 ret = virtio_dev_rx_queue_setup_finish(dev, i);
2214                 if (ret < 0)
2215                         return ret;
2216         }
2217         for (i = 0; i < dev->data->nb_tx_queues; i++) {
2218                 ret = virtio_dev_tx_queue_setup_finish(dev, i);
2219                 if (ret < 0)
2220                         return ret;
2221         }
2222
2223         /* check if lsc interrupt feature is enabled */
2224         if (dev->data->dev_conf.intr_conf.lsc) {
2225                 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
2226                         PMD_DRV_LOG(ERR, "link status not supported by host");
2227                         return -ENOTSUP;
2228                 }
2229         }
2230
2231         /* Enable uio/vfio intr/eventfd mapping: althrough we already did that
2232          * in device configure, but it could be unmapped  when device is
2233          * stopped.
2234          */
2235         if (dev->data->dev_conf.intr_conf.lsc ||
2236             dev->data->dev_conf.intr_conf.rxq) {
2237                 virtio_intr_disable(dev);
2238
2239                 /* Setup interrupt callback  */
2240                 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2241                         rte_intr_callback_register(dev->intr_handle,
2242                                                    virtio_interrupt_handler,
2243                                                    dev);
2244
2245                 if (virtio_intr_enable(dev) < 0) {
2246                         PMD_DRV_LOG(ERR, "interrupt enable failed");
2247                         return -EIO;
2248                 }
2249         }
2250
2251         /*Notify the backend
2252          *Otherwise the tap backend might already stop its queue due to fullness.
2253          *vhost backend will have no chance to be waked up
2254          */
2255         nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
2256         if (hw->max_queue_pairs > 1) {
2257                 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
2258                         return -EINVAL;
2259         }
2260
2261         PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
2262
2263         for (i = 0; i < dev->data->nb_rx_queues; i++) {
2264                 rxvq = dev->data->rx_queues[i];
2265                 /* Flush the old packets */
2266                 virtqueue_rxvq_flush(rxvq->vq);
2267                 virtqueue_notify(rxvq->vq);
2268         }
2269
2270         for (i = 0; i < dev->data->nb_tx_queues; i++) {
2271                 txvq = dev->data->tx_queues[i];
2272                 virtqueue_notify(txvq->vq);
2273         }
2274
2275         PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
2276
2277         for (i = 0; i < dev->data->nb_rx_queues; i++) {
2278                 rxvq = dev->data->rx_queues[i];
2279                 VIRTQUEUE_DUMP(rxvq->vq);
2280         }
2281
2282         for (i = 0; i < dev->data->nb_tx_queues; i++) {
2283                 txvq = dev->data->tx_queues[i];
2284                 VIRTQUEUE_DUMP(txvq->vq);
2285         }
2286
2287         set_rxtx_funcs(dev);
2288         hw->started = true;
2289
2290         /* Initialize Link state */
2291         virtio_dev_link_update(dev, 0);
2292
2293         return 0;
2294 }
2295
2296 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
2297 {
2298         struct virtio_hw *hw = dev->data->dev_private;
2299         uint16_t nr_vq = virtio_get_nr_vq(hw);
2300         const char *type __rte_unused;
2301         unsigned int i, mbuf_num = 0;
2302         struct virtqueue *vq;
2303         struct rte_mbuf *buf;
2304         int queue_type;
2305
2306         if (hw->vqs == NULL)
2307                 return;
2308
2309         for (i = 0; i < nr_vq; i++) {
2310                 vq = hw->vqs[i];
2311                 if (!vq)
2312                         continue;
2313
2314                 queue_type = virtio_get_queue_type(hw, i);
2315                 if (queue_type == VTNET_RQ)
2316                         type = "rxq";
2317                 else if (queue_type == VTNET_TQ)
2318                         type = "txq";
2319                 else
2320                         continue;
2321
2322                 PMD_INIT_LOG(DEBUG,
2323                         "Before freeing %s[%d] used and unused buf",
2324                         type, i);
2325                 VIRTQUEUE_DUMP(vq);
2326
2327                 while ((buf = virtqueue_detach_unused(vq)) != NULL) {
2328                         rte_pktmbuf_free(buf);
2329                         mbuf_num++;
2330                 }
2331
2332                 PMD_INIT_LOG(DEBUG,
2333                         "After freeing %s[%d] used and unused buf",
2334                         type, i);
2335                 VIRTQUEUE_DUMP(vq);
2336         }
2337
2338         PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
2339 }
2340
2341 /*
2342  * Stop device: disable interrupt and mark link down
2343  */
2344 int
2345 virtio_dev_stop(struct rte_eth_dev *dev)
2346 {
2347         struct virtio_hw *hw = dev->data->dev_private;
2348         struct rte_eth_link link;
2349         struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
2350
2351         PMD_INIT_LOG(DEBUG, "stop");
2352         dev->data->dev_started = 0;
2353
2354         rte_spinlock_lock(&hw->state_lock);
2355         if (!hw->started)
2356                 goto out_unlock;
2357         hw->started = false;
2358
2359         if (intr_conf->lsc || intr_conf->rxq) {
2360                 virtio_intr_disable(dev);
2361
2362                 /* Reset interrupt callback  */
2363                 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
2364                         rte_intr_callback_unregister(dev->intr_handle,
2365                                                      virtio_interrupt_handler,
2366                                                      dev);
2367                 }
2368         }
2369
2370         memset(&link, 0, sizeof(link));
2371         rte_eth_linkstatus_set(dev, &link);
2372 out_unlock:
2373         rte_spinlock_unlock(&hw->state_lock);
2374
2375         return 0;
2376 }
2377
2378 static int
2379 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2380 {
2381         struct rte_eth_link link;
2382         uint16_t status;
2383         struct virtio_hw *hw = dev->data->dev_private;
2384
2385         memset(&link, 0, sizeof(link));
2386         link.link_duplex = hw->duplex;
2387         link.link_speed  = hw->speed;
2388         link.link_autoneg = ETH_LINK_AUTONEG;
2389
2390         if (!hw->started) {
2391                 link.link_status = ETH_LINK_DOWN;
2392                 link.link_speed = ETH_SPEED_NUM_NONE;
2393         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2394                 PMD_INIT_LOG(DEBUG, "Get link status from hw");
2395                 vtpci_read_dev_config(hw,
2396                                 offsetof(struct virtio_net_config, status),
2397                                 &status, sizeof(status));
2398                 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2399                         link.link_status = ETH_LINK_DOWN;
2400                         link.link_speed = ETH_SPEED_NUM_NONE;
2401                         PMD_INIT_LOG(DEBUG, "Port %d is down",
2402                                      dev->data->port_id);
2403                 } else {
2404                         link.link_status = ETH_LINK_UP;
2405                         PMD_INIT_LOG(DEBUG, "Port %d is up",
2406                                      dev->data->port_id);
2407                 }
2408         } else {
2409                 link.link_status = ETH_LINK_UP;
2410         }
2411
2412         return rte_eth_linkstatus_set(dev, &link);
2413 }
2414
2415 static int
2416 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2417 {
2418         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2419         struct virtio_hw *hw = dev->data->dev_private;
2420         uint64_t offloads = rxmode->offloads;
2421
2422         if (mask & ETH_VLAN_FILTER_MASK) {
2423                 if ((offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2424                                 !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2425
2426                         PMD_DRV_LOG(NOTICE,
2427                                 "vlan filtering not available on this host");
2428
2429                         return -ENOTSUP;
2430                 }
2431         }
2432
2433         if (mask & ETH_VLAN_STRIP_MASK)
2434                 hw->vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
2435
2436         return 0;
2437 }
2438
2439 static int
2440 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2441 {
2442         uint64_t tso_mask, host_features;
2443         struct virtio_hw *hw = dev->data->dev_private;
2444         dev_info->speed_capa = virtio_dev_speed_capa_get(hw->speed);
2445
2446         dev_info->max_rx_queues =
2447                 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2448         dev_info->max_tx_queues =
2449                 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2450         dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2451         dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2452         dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2453
2454         host_features = VTPCI_OPS(hw)->get_features(hw);
2455         dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
2456         dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2457         if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2458                 dev_info->rx_offload_capa |=
2459                         DEV_RX_OFFLOAD_TCP_CKSUM |
2460                         DEV_RX_OFFLOAD_UDP_CKSUM;
2461         }
2462         if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN))
2463                 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_FILTER;
2464         tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2465                 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2466         if ((host_features & tso_mask) == tso_mask)
2467                 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2468
2469         dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
2470                                     DEV_TX_OFFLOAD_VLAN_INSERT;
2471         if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2472                 dev_info->tx_offload_capa |=
2473                         DEV_TX_OFFLOAD_UDP_CKSUM |
2474                         DEV_TX_OFFLOAD_TCP_CKSUM;
2475         }
2476         tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2477                 (1ULL << VIRTIO_NET_F_HOST_TSO6);
2478         if ((host_features & tso_mask) == tso_mask)
2479                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2480
2481         return 0;
2482 }
2483
2484 /*
2485  * It enables testpmd to collect per queue stats.
2486  */
2487 static int
2488 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2489 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2490 __rte_unused uint8_t is_rx)
2491 {
2492         return 0;
2493 }
2494
2495 RTE_LOG_REGISTER(virtio_logtype_init, pmd.net.virtio.init, NOTICE);
2496 RTE_LOG_REGISTER(virtio_logtype_driver, pmd.net.virtio.driver, NOTICE);