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 (hw->bus_type == VIRTIO_BUS_PCI_MODERN && !vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1334                 PMD_INIT_LOG(ERR,
1335                         "VIRTIO_F_VERSION_1 features is not enabled.");
1336                 return -1;
1337         }
1338
1339         if (hw->bus_type == VIRTIO_BUS_PCI_MODERN || hw->bus_type == VIRTIO_BUS_USER) {
1340                 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1341                 if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1342                         PMD_INIT_LOG(ERR,
1343                                 "failed to set FEATURES_OK status!");
1344                         return -1;
1345                 }
1346         }
1347
1348         hw->req_guest_features = req_features;
1349
1350         return 0;
1351 }
1352
1353 int
1354 virtio_dev_pause(struct rte_eth_dev *dev)
1355 {
1356         struct virtio_hw *hw = dev->data->dev_private;
1357
1358         rte_spinlock_lock(&hw->state_lock);
1359
1360         if (hw->started == 0) {
1361                 /* Device is just stopped. */
1362                 rte_spinlock_unlock(&hw->state_lock);
1363                 return -1;
1364         }
1365         hw->started = 0;
1366         /*
1367          * Prevent the worker threads from touching queues to avoid contention,
1368          * 1 ms should be enough for the ongoing Tx function to finish.
1369          */
1370         rte_delay_ms(1);
1371         return 0;
1372 }
1373
1374 /*
1375  * Recover hw state to let the worker threads continue.
1376  */
1377 void
1378 virtio_dev_resume(struct rte_eth_dev *dev)
1379 {
1380         struct virtio_hw *hw = dev->data->dev_private;
1381
1382         hw->started = 1;
1383         rte_spinlock_unlock(&hw->state_lock);
1384 }
1385
1386 /*
1387  * Should be called only after device is paused.
1388  */
1389 int
1390 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1391                 int nb_pkts)
1392 {
1393         struct virtio_hw *hw = dev->data->dev_private;
1394         struct virtnet_tx *txvq = dev->data->tx_queues[0];
1395         int ret;
1396
1397         hw->inject_pkts = tx_pkts;
1398         ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1399         hw->inject_pkts = NULL;
1400
1401         return ret;
1402 }
1403
1404 static void
1405 virtio_notify_peers(struct rte_eth_dev *dev)
1406 {
1407         struct virtio_hw *hw = dev->data->dev_private;
1408         struct virtnet_rx *rxvq;
1409         struct rte_mbuf *rarp_mbuf;
1410
1411         if (!dev->data->rx_queues)
1412                 return;
1413
1414         rxvq = dev->data->rx_queues[0];
1415         if (!rxvq)
1416                 return;
1417
1418         rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1419                         (struct rte_ether_addr *)hw->mac_addr);
1420         if (rarp_mbuf == NULL) {
1421                 PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1422                 return;
1423         }
1424
1425         /* If virtio port just stopped, no need to send RARP */
1426         if (virtio_dev_pause(dev) < 0) {
1427                 rte_pktmbuf_free(rarp_mbuf);
1428                 return;
1429         }
1430
1431         virtio_inject_pkts(dev, &rarp_mbuf, 1);
1432         virtio_dev_resume(dev);
1433 }
1434
1435 static void
1436 virtio_ack_link_announce(struct rte_eth_dev *dev)
1437 {
1438         struct virtio_hw *hw = dev->data->dev_private;
1439         struct virtio_pmd_ctrl ctrl;
1440
1441         ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1442         ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1443
1444         virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1445 }
1446
1447 /*
1448  * Process virtio config changed interrupt. Call the callback
1449  * if link state changed, generate gratuitous RARP packet if
1450  * the status indicates an ANNOUNCE.
1451  */
1452 void
1453 virtio_interrupt_handler(void *param)
1454 {
1455         struct rte_eth_dev *dev = param;
1456         struct virtio_hw *hw = dev->data->dev_private;
1457         uint8_t isr;
1458         uint16_t status;
1459
1460         /* Read interrupt status which clears interrupt */
1461         isr = vtpci_isr(hw);
1462         PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1463
1464         if (virtio_intr_unmask(dev) < 0)
1465                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1466
1467         if (isr & VIRTIO_PCI_ISR_CONFIG) {
1468                 if (virtio_dev_link_update(dev, 0) == 0)
1469                         rte_eth_dev_callback_process(dev,
1470                                                      RTE_ETH_EVENT_INTR_LSC,
1471                                                      NULL);
1472
1473                 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1474                         vtpci_read_dev_config(hw,
1475                                 offsetof(struct virtio_net_config, status),
1476                                 &status, sizeof(status));
1477                         if (status & VIRTIO_NET_S_ANNOUNCE) {
1478                                 virtio_notify_peers(dev);
1479                                 if (hw->cvq)
1480                                         virtio_ack_link_announce(dev);
1481                         }
1482                 }
1483         }
1484 }
1485
1486 /* set rx and tx handlers according to what is supported */
1487 static void
1488 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1489 {
1490         struct virtio_hw *hw = eth_dev->data->dev_private;
1491
1492         eth_dev->tx_pkt_prepare = virtio_xmit_pkts_prepare;
1493         if (vtpci_packed_queue(hw)) {
1494                 PMD_INIT_LOG(INFO,
1495                         "virtio: using packed ring %s Tx path on port %u",
1496                         hw->use_vec_tx ? "vectorized" : "standard",
1497                         eth_dev->data->port_id);
1498                 if (hw->use_vec_tx)
1499                         eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed_vec;
1500                 else
1501                         eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed;
1502         } else {
1503                 if (hw->use_inorder_tx) {
1504                         PMD_INIT_LOG(INFO, "virtio: using inorder Tx path on port %u",
1505                                 eth_dev->data->port_id);
1506                         eth_dev->tx_pkt_burst = virtio_xmit_pkts_inorder;
1507                 } else {
1508                         PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1509                                 eth_dev->data->port_id);
1510                         eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1511                 }
1512         }
1513
1514         if (vtpci_packed_queue(hw)) {
1515                 if (hw->use_vec_rx) {
1516                         PMD_INIT_LOG(INFO,
1517                                 "virtio: using packed ring vectorized Rx path on port %u",
1518                                 eth_dev->data->port_id);
1519                         eth_dev->rx_pkt_burst =
1520                                 &virtio_recv_pkts_packed_vec;
1521                 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1522                         PMD_INIT_LOG(INFO,
1523                                 "virtio: using packed ring mergeable buffer Rx path on port %u",
1524                                 eth_dev->data->port_id);
1525                         eth_dev->rx_pkt_burst =
1526                                 &virtio_recv_mergeable_pkts_packed;
1527                 } else {
1528                         PMD_INIT_LOG(INFO,
1529                                 "virtio: using packed ring standard Rx path on port %u",
1530                                 eth_dev->data->port_id);
1531                         eth_dev->rx_pkt_burst = &virtio_recv_pkts_packed;
1532                 }
1533         } else {
1534                 if (hw->use_vec_rx) {
1535                         PMD_INIT_LOG(INFO, "virtio: using vectorized Rx path on port %u",
1536                                 eth_dev->data->port_id);
1537                         eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1538                 } else if (hw->use_inorder_rx) {
1539                         PMD_INIT_LOG(INFO,
1540                                 "virtio: using inorder Rx path on port %u",
1541                                 eth_dev->data->port_id);
1542                         eth_dev->rx_pkt_burst = &virtio_recv_pkts_inorder;
1543                 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1544                         PMD_INIT_LOG(INFO,
1545                                 "virtio: using mergeable buffer Rx path on port %u",
1546                                 eth_dev->data->port_id);
1547                         eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1548                 } else {
1549                         PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1550                                 eth_dev->data->port_id);
1551                         eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1552                 }
1553         }
1554
1555 }
1556
1557 /* Only support 1:1 queue/interrupt mapping so far.
1558  * TODO: support n:1 queue/interrupt mapping when there are limited number of
1559  * interrupt vectors (<N+1).
1560  */
1561 static int
1562 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1563 {
1564         uint32_t i;
1565         struct virtio_hw *hw = dev->data->dev_private;
1566
1567         PMD_INIT_LOG(INFO, "queue/interrupt binding");
1568         for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1569                 dev->intr_handle->intr_vec[i] = i + 1;
1570                 if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1571                                                  VIRTIO_MSI_NO_VECTOR) {
1572                         PMD_DRV_LOG(ERR, "failed to set queue vector");
1573                         return -EBUSY;
1574                 }
1575         }
1576
1577         return 0;
1578 }
1579
1580 static void
1581 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1582 {
1583         uint32_t i;
1584         struct virtio_hw *hw = dev->data->dev_private;
1585
1586         PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1587         for (i = 0; i < dev->data->nb_rx_queues; ++i)
1588                 VTPCI_OPS(hw)->set_queue_irq(hw,
1589                                              hw->vqs[i * VTNET_CQ],
1590                                              VIRTIO_MSI_NO_VECTOR);
1591 }
1592
1593 static int
1594 virtio_configure_intr(struct rte_eth_dev *dev)
1595 {
1596         struct virtio_hw *hw = dev->data->dev_private;
1597
1598         if (!rte_intr_cap_multiple(dev->intr_handle)) {
1599                 PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1600                 return -ENOTSUP;
1601         }
1602
1603         if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1604                 PMD_INIT_LOG(ERR, "Fail to create eventfd");
1605                 return -1;
1606         }
1607
1608         if (!dev->intr_handle->intr_vec) {
1609                 dev->intr_handle->intr_vec =
1610                         rte_zmalloc("intr_vec",
1611                                     hw->max_queue_pairs * sizeof(int), 0);
1612                 if (!dev->intr_handle->intr_vec) {
1613                         PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1614                                      hw->max_queue_pairs);
1615                         return -ENOMEM;
1616                 }
1617         }
1618
1619         /* Re-register callback to update max_intr */
1620         rte_intr_callback_unregister(dev->intr_handle,
1621                                      virtio_interrupt_handler,
1622                                      dev);
1623         rte_intr_callback_register(dev->intr_handle,
1624                                    virtio_interrupt_handler,
1625                                    dev);
1626
1627         /* DO NOT try to remove this! This function will enable msix, or QEMU
1628          * will encounter SIGSEGV when DRIVER_OK is sent.
1629          * And for legacy devices, this should be done before queue/vec binding
1630          * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1631          * (22) will be ignored.
1632          */
1633         if (virtio_intr_enable(dev) < 0) {
1634                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1635                 return -1;
1636         }
1637
1638         if (virtio_queues_bind_intr(dev) < 0) {
1639                 PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1640                 return -1;
1641         }
1642
1643         return 0;
1644 }
1645 #define DUPLEX_UNKNOWN   0xff
1646 /* reset device and renegotiate features if needed */
1647 static int
1648 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1649 {
1650         struct virtio_hw *hw = eth_dev->data->dev_private;
1651         struct virtio_net_config *config;
1652         struct virtio_net_config local_config;
1653         int ret;
1654
1655         /* Reset the device although not necessary at startup */
1656         vtpci_reset(hw);
1657
1658         if (hw->vqs) {
1659                 virtio_dev_free_mbufs(eth_dev);
1660                 virtio_free_queues(hw);
1661         }
1662
1663         /* Tell the host we've noticed this device. */
1664         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1665
1666         /* Tell the host we've known how to drive the device. */
1667         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1668         if (virtio_negotiate_features(hw, req_features) < 0)
1669                 return -1;
1670
1671         hw->weak_barriers = !vtpci_with_feature(hw, VIRTIO_F_ORDER_PLATFORM);
1672
1673         /* If host does not support both status and MSI-X then disable LSC */
1674         if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) &&
1675             hw->use_msix != VIRTIO_MSIX_NONE)
1676                 eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1677         else
1678                 eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1679
1680         eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
1681
1682         /* Setting up rx_header size for the device */
1683         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1684             vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
1685             vtpci_with_feature(hw, VIRTIO_F_RING_PACKED))
1686                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1687         else
1688                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1689
1690         /* Copy the permanent MAC address to: virtio_hw */
1691         virtio_get_hwaddr(hw);
1692         rte_ether_addr_copy((struct rte_ether_addr *)hw->mac_addr,
1693                         &eth_dev->data->mac_addrs[0]);
1694         PMD_INIT_LOG(DEBUG,
1695                      "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1696                      hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1697                      hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1698
1699         if (hw->speed == ETH_SPEED_NUM_UNKNOWN) {
1700                 if (vtpci_with_feature(hw, VIRTIO_NET_F_SPEED_DUPLEX)) {
1701                         config = &local_config;
1702                         vtpci_read_dev_config(hw,
1703                                 offsetof(struct virtio_net_config, speed),
1704                                 &config->speed, sizeof(config->speed));
1705                         vtpci_read_dev_config(hw,
1706                                 offsetof(struct virtio_net_config, duplex),
1707                                 &config->duplex, sizeof(config->duplex));
1708                         hw->speed = config->speed;
1709                         hw->duplex = config->duplex;
1710                 }
1711         }
1712         if (hw->duplex == DUPLEX_UNKNOWN)
1713                 hw->duplex = ETH_LINK_FULL_DUPLEX;
1714         PMD_INIT_LOG(DEBUG, "link speed = %d, duplex = %d",
1715                 hw->speed, hw->duplex);
1716         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1717                 config = &local_config;
1718
1719                 vtpci_read_dev_config(hw,
1720                         offsetof(struct virtio_net_config, mac),
1721                         &config->mac, sizeof(config->mac));
1722
1723                 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1724                         vtpci_read_dev_config(hw,
1725                                 offsetof(struct virtio_net_config, status),
1726                                 &config->status, sizeof(config->status));
1727                 } else {
1728                         PMD_INIT_LOG(DEBUG,
1729                                      "VIRTIO_NET_F_STATUS is not supported");
1730                         config->status = 0;
1731                 }
1732
1733                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1734                         vtpci_read_dev_config(hw,
1735                                 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1736                                 &config->max_virtqueue_pairs,
1737                                 sizeof(config->max_virtqueue_pairs));
1738                 } else {
1739                         PMD_INIT_LOG(DEBUG,
1740                                      "VIRTIO_NET_F_MQ is not supported");
1741                         config->max_virtqueue_pairs = 1;
1742                 }
1743
1744                 hw->max_queue_pairs = config->max_virtqueue_pairs;
1745
1746                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1747                         vtpci_read_dev_config(hw,
1748                                 offsetof(struct virtio_net_config, mtu),
1749                                 &config->mtu,
1750                                 sizeof(config->mtu));
1751
1752                         /*
1753                          * MTU value has already been checked at negotiation
1754                          * time, but check again in case it has changed since
1755                          * then, which should not happen.
1756                          */
1757                         if (config->mtu < RTE_ETHER_MIN_MTU) {
1758                                 PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1759                                                 config->mtu);
1760                                 return -1;
1761                         }
1762
1763                         hw->max_mtu = config->mtu;
1764                         /* Set initial MTU to maximum one supported by vhost */
1765                         eth_dev->data->mtu = config->mtu;
1766
1767                 } else {
1768                         hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1769                                 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1770                 }
1771
1772                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1773                                 config->max_virtqueue_pairs);
1774                 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1775                 PMD_INIT_LOG(DEBUG,
1776                                 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1777                                 config->mac[0], config->mac[1],
1778                                 config->mac[2], config->mac[3],
1779                                 config->mac[4], config->mac[5]);
1780         } else {
1781                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1782                 hw->max_queue_pairs = 1;
1783                 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1784                         VLAN_TAG_LEN - hw->vtnet_hdr_size;
1785         }
1786
1787         ret = virtio_alloc_queues(eth_dev);
1788         if (ret < 0)
1789                 return ret;
1790
1791         if (eth_dev->data->dev_conf.intr_conf.rxq) {
1792                 if (virtio_configure_intr(eth_dev) < 0) {
1793                         PMD_INIT_LOG(ERR, "failed to configure interrupt");
1794                         virtio_free_queues(hw);
1795                         return -1;
1796                 }
1797         }
1798
1799         vtpci_reinit_complete(hw);
1800
1801         return 0;
1802 }
1803
1804
1805 static void
1806 virtio_set_vtpci_ops(struct virtio_hw *hw)
1807 {
1808 #ifdef RTE_VIRTIO_USER
1809         if (hw->bus_type == VIRTIO_BUS_USER)
1810                 VTPCI_OPS(hw) = &virtio_user_ops;
1811         else
1812 #endif
1813         if (hw->bus_type == VIRTIO_BUS_PCI_MODERN)
1814                 VTPCI_OPS(hw) = &modern_ops;
1815         else if (hw->bus_type == VIRTIO_BUS_PCI_LEGACY)
1816                 VTPCI_OPS(hw) = &legacy_ops;
1817
1818         return;
1819 }
1820
1821 /*
1822  * This function is based on probe() function in virtio_pci.c
1823  * It returns 0 on success.
1824  */
1825 int
1826 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1827 {
1828         struct virtio_hw *hw = eth_dev->data->dev_private;
1829         uint32_t speed = ETH_SPEED_NUM_UNKNOWN;
1830         int vectorized = 0;
1831         int ret;
1832
1833         if (sizeof(struct virtio_net_hdr_mrg_rxbuf) > RTE_PKTMBUF_HEADROOM) {
1834                 PMD_INIT_LOG(ERR,
1835                         "Not sufficient headroom required = %d, avail = %d",
1836                         (int)sizeof(struct virtio_net_hdr_mrg_rxbuf),
1837                         RTE_PKTMBUF_HEADROOM);
1838
1839                 return -1;
1840         }
1841
1842         eth_dev->dev_ops = &virtio_eth_dev_ops;
1843         eth_dev->rx_descriptor_done = virtio_dev_rx_queue_done;
1844
1845         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1846                 virtio_set_vtpci_ops(hw);
1847                 set_rxtx_funcs(eth_dev);
1848                 return 0;
1849         }
1850
1851         ret = virtio_dev_devargs_parse(eth_dev->device->devargs, &speed, &vectorized);
1852         if (ret < 0)
1853                 return ret;
1854         hw->speed = speed;
1855
1856         /* Allocate memory for storing MAC addresses */
1857         eth_dev->data->mac_addrs = rte_zmalloc("virtio",
1858                                 VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN, 0);
1859         if (eth_dev->data->mac_addrs == NULL) {
1860                 PMD_INIT_LOG(ERR,
1861                         "Failed to allocate %d bytes needed to store MAC addresses",
1862                         VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN);
1863                 return -ENOMEM;
1864         }
1865
1866         hw->port_id = eth_dev->data->port_id;
1867         rte_spinlock_init(&hw->state_lock);
1868
1869         /* reset device and negotiate default features */
1870         ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1871         if (ret < 0)
1872                 goto err_virtio_init;
1873
1874         if (vectorized) {
1875                 if (!vtpci_packed_queue(hw)) {
1876                         hw->use_vec_rx = 1;
1877                 } else {
1878 #if defined(CC_AVX512_SUPPORT) || defined(RTE_ARCH_ARM)
1879                         hw->use_vec_rx = 1;
1880                         hw->use_vec_tx = 1;
1881 #else
1882                         PMD_DRV_LOG(INFO,
1883                                 "building environment do not support packed ring vectorized");
1884 #endif
1885                 }
1886         }
1887
1888         hw->opened = true;
1889
1890         return 0;
1891
1892 err_virtio_init:
1893         rte_free(eth_dev->data->mac_addrs);
1894         eth_dev->data->mac_addrs = NULL;
1895         return ret;
1896 }
1897
1898 static uint32_t
1899 virtio_dev_speed_capa_get(uint32_t speed)
1900 {
1901         switch (speed) {
1902         case ETH_SPEED_NUM_10G:
1903                 return ETH_LINK_SPEED_10G;
1904         case ETH_SPEED_NUM_20G:
1905                 return ETH_LINK_SPEED_20G;
1906         case ETH_SPEED_NUM_25G:
1907                 return ETH_LINK_SPEED_25G;
1908         case ETH_SPEED_NUM_40G:
1909                 return ETH_LINK_SPEED_40G;
1910         case ETH_SPEED_NUM_50G:
1911                 return ETH_LINK_SPEED_50G;
1912         case ETH_SPEED_NUM_56G:
1913                 return ETH_LINK_SPEED_56G;
1914         case ETH_SPEED_NUM_100G:
1915                 return ETH_LINK_SPEED_100G;
1916         case ETH_SPEED_NUM_200G:
1917                 return ETH_LINK_SPEED_200G;
1918         default:
1919                 return 0;
1920         }
1921 }
1922
1923 static int vectorized_check_handler(__rte_unused const char *key,
1924                 const char *value, void *ret_val)
1925 {
1926         if (strcmp(value, "1") == 0)
1927                 *(int *)ret_val = 1;
1928         else
1929                 *(int *)ret_val = 0;
1930
1931         return 0;
1932 }
1933
1934 #define VIRTIO_ARG_SPEED      "speed"
1935 #define VIRTIO_ARG_VECTORIZED "vectorized"
1936
1937 static int
1938 link_speed_handler(const char *key __rte_unused,
1939                 const char *value, void *ret_val)
1940 {
1941         uint32_t val;
1942         if (!value || !ret_val)
1943                 return -EINVAL;
1944         val = strtoul(value, NULL, 0);
1945         /* validate input */
1946         if (virtio_dev_speed_capa_get(val) == 0)
1947                 return -EINVAL;
1948         *(uint32_t *)ret_val = val;
1949
1950         return 0;
1951 }
1952
1953
1954 static int
1955 virtio_dev_devargs_parse(struct rte_devargs *devargs, uint32_t *speed, int *vectorized)
1956 {
1957         struct rte_kvargs *kvlist;
1958         int ret = 0;
1959
1960         if (devargs == NULL)
1961                 return 0;
1962
1963         kvlist = rte_kvargs_parse(devargs->args, NULL);
1964         if (kvlist == NULL) {
1965                 PMD_INIT_LOG(ERR, "error when parsing param");
1966                 return 0;
1967         }
1968
1969         if (speed && rte_kvargs_count(kvlist, VIRTIO_ARG_SPEED) == 1) {
1970                 ret = rte_kvargs_process(kvlist,
1971                                         VIRTIO_ARG_SPEED,
1972                                         link_speed_handler, speed);
1973                 if (ret < 0) {
1974                         PMD_INIT_LOG(ERR, "Failed to parse %s",
1975                                         VIRTIO_ARG_SPEED);
1976                         goto exit;
1977                 }
1978         }
1979
1980         if (vectorized &&
1981                 rte_kvargs_count(kvlist, VIRTIO_ARG_VECTORIZED) == 1) {
1982                 ret = rte_kvargs_process(kvlist,
1983                                 VIRTIO_ARG_VECTORIZED,
1984                                 vectorized_check_handler, vectorized);
1985                 if (ret < 0) {
1986                         PMD_INIT_LOG(ERR, "Failed to parse %s",
1987                                         VIRTIO_ARG_VECTORIZED);
1988                         goto exit;
1989                 }
1990         }
1991
1992 exit:
1993         rte_kvargs_free(kvlist);
1994         return ret;
1995 }
1996
1997 static bool
1998 rx_offload_enabled(struct virtio_hw *hw)
1999 {
2000         return vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM) ||
2001                 vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2002                 vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6);
2003 }
2004
2005 static bool
2006 tx_offload_enabled(struct virtio_hw *hw)
2007 {
2008         return vtpci_with_feature(hw, VIRTIO_NET_F_CSUM) ||
2009                 vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO4) ||
2010                 vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO6);
2011 }
2012
2013 /*
2014  * Configure virtio device
2015  * It returns 0 on success.
2016  */
2017 static int
2018 virtio_dev_configure(struct rte_eth_dev *dev)
2019 {
2020         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2021         const struct rte_eth_txmode *txmode = &dev->data->dev_conf.txmode;
2022         struct virtio_hw *hw = dev->data->dev_private;
2023         uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
2024                 hw->vtnet_hdr_size;
2025         uint64_t rx_offloads = rxmode->offloads;
2026         uint64_t tx_offloads = txmode->offloads;
2027         uint64_t req_features;
2028         int ret;
2029
2030         PMD_INIT_LOG(DEBUG, "configure");
2031         req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
2032
2033         if (rxmode->mq_mode != ETH_MQ_RX_NONE) {
2034                 PMD_DRV_LOG(ERR,
2035                         "Unsupported Rx multi queue mode %d",
2036                         rxmode->mq_mode);
2037                 return -EINVAL;
2038         }
2039
2040         if (txmode->mq_mode != ETH_MQ_TX_NONE) {
2041                 PMD_DRV_LOG(ERR,
2042                         "Unsupported Tx multi queue mode %d",
2043                         txmode->mq_mode);
2044                 return -EINVAL;
2045         }
2046
2047         if (dev->data->dev_conf.intr_conf.rxq) {
2048                 ret = virtio_init_device(dev, hw->req_guest_features);
2049                 if (ret < 0)
2050                         return ret;
2051         }
2052
2053         if (rxmode->max_rx_pkt_len > hw->max_mtu + ether_hdr_len)
2054                 req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
2055
2056         if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2057                            DEV_RX_OFFLOAD_TCP_CKSUM))
2058                 req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
2059
2060         if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
2061                 req_features |=
2062                         (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2063                         (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2064
2065         if (tx_offloads & (DEV_TX_OFFLOAD_UDP_CKSUM |
2066                            DEV_TX_OFFLOAD_TCP_CKSUM))
2067                 req_features |= (1ULL << VIRTIO_NET_F_CSUM);
2068
2069         if (tx_offloads & DEV_TX_OFFLOAD_TCP_TSO)
2070                 req_features |=
2071                         (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2072                         (1ULL << VIRTIO_NET_F_HOST_TSO6);
2073
2074         /* if request features changed, reinit the device */
2075         if (req_features != hw->req_guest_features) {
2076                 ret = virtio_init_device(dev, req_features);
2077                 if (ret < 0)
2078                         return ret;
2079         }
2080
2081         if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2082                             DEV_RX_OFFLOAD_TCP_CKSUM)) &&
2083                 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
2084                 PMD_DRV_LOG(ERR,
2085                         "rx checksum not available on this host");
2086                 return -ENOTSUP;
2087         }
2088
2089         if ((rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) &&
2090                 (!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2091                  !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
2092                 PMD_DRV_LOG(ERR,
2093                         "Large Receive Offload not available on this host");
2094                 return -ENOTSUP;
2095         }
2096
2097         /* start control queue */
2098         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
2099                 virtio_dev_cq_start(dev);
2100
2101         if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2102                 hw->vlan_strip = 1;
2103
2104         if ((rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
2105             && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2106                 PMD_DRV_LOG(ERR,
2107                             "vlan filtering not available on this host");
2108                 return -ENOTSUP;
2109         }
2110
2111         hw->has_tx_offload = tx_offload_enabled(hw);
2112         hw->has_rx_offload = rx_offload_enabled(hw);
2113
2114         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2115                 /* Enable vector (0) for Link State Intrerrupt */
2116                 if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
2117                                 VIRTIO_MSI_NO_VECTOR) {
2118                         PMD_DRV_LOG(ERR, "failed to set config vector");
2119                         return -EBUSY;
2120                 }
2121
2122         if (vtpci_packed_queue(hw)) {
2123 #if defined(RTE_ARCH_X86_64) && defined(CC_AVX512_SUPPORT)
2124                 if ((hw->use_vec_rx || hw->use_vec_tx) &&
2125                     (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F) ||
2126                      !vtpci_with_feature(hw, VIRTIO_F_IN_ORDER) ||
2127                      !vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
2128                      rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_512)) {
2129                         PMD_DRV_LOG(INFO,
2130                                 "disabled packed ring vectorized path for requirements not met");
2131                         hw->use_vec_rx = 0;
2132                         hw->use_vec_tx = 0;
2133                 }
2134 #elif defined(RTE_ARCH_ARM)
2135                 if ((hw->use_vec_rx || hw->use_vec_tx) &&
2136                     (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON) ||
2137                      !vtpci_with_feature(hw, VIRTIO_F_IN_ORDER) ||
2138                      !vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
2139                      rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_128)) {
2140                         PMD_DRV_LOG(INFO,
2141                                 "disabled packed ring vectorized path for requirements not met");
2142                         hw->use_vec_rx = 0;
2143                         hw->use_vec_tx = 0;
2144                 }
2145 #else
2146                 hw->use_vec_rx = 0;
2147                 hw->use_vec_tx = 0;
2148 #endif
2149
2150                 if (hw->use_vec_rx) {
2151                         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2152                                 PMD_DRV_LOG(INFO,
2153                                         "disabled packed ring vectorized rx for mrg_rxbuf enabled");
2154                                 hw->use_vec_rx = 0;
2155                         }
2156
2157                         if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) {
2158                                 PMD_DRV_LOG(INFO,
2159                                         "disabled packed ring vectorized rx for TCP_LRO enabled");
2160                                 hw->use_vec_rx = 0;
2161                         }
2162                 }
2163         } else {
2164                 if (vtpci_with_feature(hw, VIRTIO_F_IN_ORDER)) {
2165                         hw->use_inorder_tx = 1;
2166                         hw->use_inorder_rx = 1;
2167                         hw->use_vec_rx = 0;
2168                 }
2169
2170                 if (hw->use_vec_rx) {
2171 #if defined RTE_ARCH_ARM
2172                         if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
2173                                 PMD_DRV_LOG(INFO,
2174                                         "disabled split ring vectorized path for requirement not met");
2175                                 hw->use_vec_rx = 0;
2176                         }
2177 #endif
2178                         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2179                                 PMD_DRV_LOG(INFO,
2180                                         "disabled split ring vectorized rx for mrg_rxbuf enabled");
2181                                 hw->use_vec_rx = 0;
2182                         }
2183
2184                         if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2185                                            DEV_RX_OFFLOAD_TCP_CKSUM |
2186                                            DEV_RX_OFFLOAD_TCP_LRO |
2187                                            DEV_RX_OFFLOAD_VLAN_STRIP)) {
2188                                 PMD_DRV_LOG(INFO,
2189                                         "disabled split ring vectorized rx for offloading enabled");
2190                                 hw->use_vec_rx = 0;
2191                         }
2192
2193                         if (rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_128) {
2194                                 PMD_DRV_LOG(INFO,
2195                                         "disabled split ring vectorized rx, max SIMD bitwidth too low");
2196                                 hw->use_vec_rx = 0;
2197                         }
2198                 }
2199         }
2200
2201         return 0;
2202 }
2203
2204
2205 static int
2206 virtio_dev_start(struct rte_eth_dev *dev)
2207 {
2208         uint16_t nb_queues, i;
2209         struct virtnet_rx *rxvq;
2210         struct virtnet_tx *txvq __rte_unused;
2211         struct virtio_hw *hw = dev->data->dev_private;
2212         int ret;
2213
2214         /* Finish the initialization of the queues */
2215         for (i = 0; i < dev->data->nb_rx_queues; i++) {
2216                 ret = virtio_dev_rx_queue_setup_finish(dev, i);
2217                 if (ret < 0)
2218                         return ret;
2219         }
2220         for (i = 0; i < dev->data->nb_tx_queues; i++) {
2221                 ret = virtio_dev_tx_queue_setup_finish(dev, i);
2222                 if (ret < 0)
2223                         return ret;
2224         }
2225
2226         /* check if lsc interrupt feature is enabled */
2227         if (dev->data->dev_conf.intr_conf.lsc) {
2228                 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
2229                         PMD_DRV_LOG(ERR, "link status not supported by host");
2230                         return -ENOTSUP;
2231                 }
2232         }
2233
2234         /* Enable uio/vfio intr/eventfd mapping: althrough we already did that
2235          * in device configure, but it could be unmapped  when device is
2236          * stopped.
2237          */
2238         if (dev->data->dev_conf.intr_conf.lsc ||
2239             dev->data->dev_conf.intr_conf.rxq) {
2240                 virtio_intr_disable(dev);
2241
2242                 /* Setup interrupt callback  */
2243                 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2244                         rte_intr_callback_register(dev->intr_handle,
2245                                                    virtio_interrupt_handler,
2246                                                    dev);
2247
2248                 if (virtio_intr_enable(dev) < 0) {
2249                         PMD_DRV_LOG(ERR, "interrupt enable failed");
2250                         return -EIO;
2251                 }
2252         }
2253
2254         /*Notify the backend
2255          *Otherwise the tap backend might already stop its queue due to fullness.
2256          *vhost backend will have no chance to be waked up
2257          */
2258         nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
2259         if (hw->max_queue_pairs > 1) {
2260                 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
2261                         return -EINVAL;
2262         }
2263
2264         PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
2265
2266         for (i = 0; i < dev->data->nb_rx_queues; i++) {
2267                 rxvq = dev->data->rx_queues[i];
2268                 /* Flush the old packets */
2269                 virtqueue_rxvq_flush(rxvq->vq);
2270                 virtqueue_notify(rxvq->vq);
2271         }
2272
2273         for (i = 0; i < dev->data->nb_tx_queues; i++) {
2274                 txvq = dev->data->tx_queues[i];
2275                 virtqueue_notify(txvq->vq);
2276         }
2277
2278         PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
2279
2280         for (i = 0; i < dev->data->nb_rx_queues; i++) {
2281                 rxvq = dev->data->rx_queues[i];
2282                 VIRTQUEUE_DUMP(rxvq->vq);
2283         }
2284
2285         for (i = 0; i < dev->data->nb_tx_queues; i++) {
2286                 txvq = dev->data->tx_queues[i];
2287                 VIRTQUEUE_DUMP(txvq->vq);
2288         }
2289
2290         set_rxtx_funcs(dev);
2291         hw->started = true;
2292
2293         /* Initialize Link state */
2294         virtio_dev_link_update(dev, 0);
2295
2296         return 0;
2297 }
2298
2299 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
2300 {
2301         struct virtio_hw *hw = dev->data->dev_private;
2302         uint16_t nr_vq = virtio_get_nr_vq(hw);
2303         const char *type __rte_unused;
2304         unsigned int i, mbuf_num = 0;
2305         struct virtqueue *vq;
2306         struct rte_mbuf *buf;
2307         int queue_type;
2308
2309         if (hw->vqs == NULL)
2310                 return;
2311
2312         for (i = 0; i < nr_vq; i++) {
2313                 vq = hw->vqs[i];
2314                 if (!vq)
2315                         continue;
2316
2317                 queue_type = virtio_get_queue_type(hw, i);
2318                 if (queue_type == VTNET_RQ)
2319                         type = "rxq";
2320                 else if (queue_type == VTNET_TQ)
2321                         type = "txq";
2322                 else
2323                         continue;
2324
2325                 PMD_INIT_LOG(DEBUG,
2326                         "Before freeing %s[%d] used and unused buf",
2327                         type, i);
2328                 VIRTQUEUE_DUMP(vq);
2329
2330                 while ((buf = virtqueue_detach_unused(vq)) != NULL) {
2331                         rte_pktmbuf_free(buf);
2332                         mbuf_num++;
2333                 }
2334
2335                 PMD_INIT_LOG(DEBUG,
2336                         "After freeing %s[%d] used and unused buf",
2337                         type, i);
2338                 VIRTQUEUE_DUMP(vq);
2339         }
2340
2341         PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
2342 }
2343
2344 /*
2345  * Stop device: disable interrupt and mark link down
2346  */
2347 int
2348 virtio_dev_stop(struct rte_eth_dev *dev)
2349 {
2350         struct virtio_hw *hw = dev->data->dev_private;
2351         struct rte_eth_link link;
2352         struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
2353
2354         PMD_INIT_LOG(DEBUG, "stop");
2355         dev->data->dev_started = 0;
2356
2357         rte_spinlock_lock(&hw->state_lock);
2358         if (!hw->started)
2359                 goto out_unlock;
2360         hw->started = false;
2361
2362         if (intr_conf->lsc || intr_conf->rxq) {
2363                 virtio_intr_disable(dev);
2364
2365                 /* Reset interrupt callback  */
2366                 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
2367                         rte_intr_callback_unregister(dev->intr_handle,
2368                                                      virtio_interrupt_handler,
2369                                                      dev);
2370                 }
2371         }
2372
2373         memset(&link, 0, sizeof(link));
2374         rte_eth_linkstatus_set(dev, &link);
2375 out_unlock:
2376         rte_spinlock_unlock(&hw->state_lock);
2377
2378         return 0;
2379 }
2380
2381 static int
2382 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2383 {
2384         struct rte_eth_link link;
2385         uint16_t status;
2386         struct virtio_hw *hw = dev->data->dev_private;
2387
2388         memset(&link, 0, sizeof(link));
2389         link.link_duplex = hw->duplex;
2390         link.link_speed  = hw->speed;
2391         link.link_autoneg = ETH_LINK_AUTONEG;
2392
2393         if (!hw->started) {
2394                 link.link_status = ETH_LINK_DOWN;
2395                 link.link_speed = ETH_SPEED_NUM_NONE;
2396         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2397                 PMD_INIT_LOG(DEBUG, "Get link status from hw");
2398                 vtpci_read_dev_config(hw,
2399                                 offsetof(struct virtio_net_config, status),
2400                                 &status, sizeof(status));
2401                 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2402                         link.link_status = ETH_LINK_DOWN;
2403                         link.link_speed = ETH_SPEED_NUM_NONE;
2404                         PMD_INIT_LOG(DEBUG, "Port %d is down",
2405                                      dev->data->port_id);
2406                 } else {
2407                         link.link_status = ETH_LINK_UP;
2408                         PMD_INIT_LOG(DEBUG, "Port %d is up",
2409                                      dev->data->port_id);
2410                 }
2411         } else {
2412                 link.link_status = ETH_LINK_UP;
2413         }
2414
2415         return rte_eth_linkstatus_set(dev, &link);
2416 }
2417
2418 static int
2419 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2420 {
2421         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2422         struct virtio_hw *hw = dev->data->dev_private;
2423         uint64_t offloads = rxmode->offloads;
2424
2425         if (mask & ETH_VLAN_FILTER_MASK) {
2426                 if ((offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2427                                 !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2428
2429                         PMD_DRV_LOG(NOTICE,
2430                                 "vlan filtering not available on this host");
2431
2432                         return -ENOTSUP;
2433                 }
2434         }
2435
2436         if (mask & ETH_VLAN_STRIP_MASK)
2437                 hw->vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
2438
2439         return 0;
2440 }
2441
2442 static int
2443 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2444 {
2445         uint64_t tso_mask, host_features;
2446         struct virtio_hw *hw = dev->data->dev_private;
2447         dev_info->speed_capa = virtio_dev_speed_capa_get(hw->speed);
2448
2449         dev_info->max_rx_queues =
2450                 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2451         dev_info->max_tx_queues =
2452                 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2453         dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2454         dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2455         dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2456
2457         host_features = VTPCI_OPS(hw)->get_features(hw);
2458         dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
2459         dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2460         if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2461                 dev_info->rx_offload_capa |=
2462                         DEV_RX_OFFLOAD_TCP_CKSUM |
2463                         DEV_RX_OFFLOAD_UDP_CKSUM;
2464         }
2465         if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN))
2466                 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_FILTER;
2467         tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2468                 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2469         if ((host_features & tso_mask) == tso_mask)
2470                 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2471
2472         dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
2473                                     DEV_TX_OFFLOAD_VLAN_INSERT;
2474         if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2475                 dev_info->tx_offload_capa |=
2476                         DEV_TX_OFFLOAD_UDP_CKSUM |
2477                         DEV_TX_OFFLOAD_TCP_CKSUM;
2478         }
2479         tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2480                 (1ULL << VIRTIO_NET_F_HOST_TSO6);
2481         if ((host_features & tso_mask) == tso_mask)
2482                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2483
2484         return 0;
2485 }
2486
2487 /*
2488  * It enables testpmd to collect per queue stats.
2489  */
2490 static int
2491 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2492 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2493 __rte_unused uint8_t is_rx)
2494 {
2495         return 0;
2496 }
2497
2498 RTE_LOG_REGISTER(virtio_logtype_init, pmd.net.virtio.init, NOTICE);
2499 RTE_LOG_REGISTER(virtio_logtype_driver, pmd.net.virtio.driver, NOTICE);