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 <rte_ethdev_driver.h>
  12 #include <rte_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
  27 #include <rte_memory.h>
  28 #include <rte_eal.h>
  29 #include <rte_dev.h>
  30 #include <rte_cycles.h>
  31 #include <rte_kvargs.h>
  32
  33 #include "virtio_ethdev.h"
  34 #include "virtio_pci.h"
  35 #include "virtio_logs.h"
  36 #include "virtqueue.h"
  37 #include "virtio_rxtx.h"
  38 #include "virtio_user/virtio_user_dev.h"
  39
  40 static int eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev);
  41 static int  virtio_dev_configure(struct rte_eth_dev *dev);
  42 static int  virtio_dev_start(struct rte_eth_dev *dev);
  43 static void virtio_dev_stop(struct rte_eth_dev *dev);
  44 static void virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
  45 static void virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
  46 static void virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
  47 static void virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
  48 static void virtio_dev_info_get(struct rte_eth_dev *dev,
  49                                 struct rte_eth_dev_info *dev_info);
  50 static int virtio_dev_link_update(struct rte_eth_dev *dev,
  51         int wait_to_complete);
  52 static int virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask);
  53
  54 static void virtio_set_hwaddr(struct virtio_hw *hw);
  55 static void virtio_get_hwaddr(struct virtio_hw *hw);
  56
  57 static int virtio_dev_stats_get(struct rte_eth_dev *dev,
  58                                  struct rte_eth_stats *stats);
  59 static int virtio_dev_xstats_get(struct rte_eth_dev *dev,
  60                                  struct rte_eth_xstat *xstats, unsigned n);
  61 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
  62                                        struct rte_eth_xstat_name *xstats_names,
  63                                        unsigned limit);
  64 static void virtio_dev_stats_reset(struct rte_eth_dev *dev);
  65 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
  66 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
  67                                 uint16_t vlan_id, int on);
  68 static int virtio_mac_addr_add(struct rte_eth_dev *dev,
  69                                 struct rte_ether_addr *mac_addr,
  70                                 uint32_t index, uint32_t vmdq);
  71 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
  72 static int virtio_mac_addr_set(struct rte_eth_dev *dev,
  73                                 struct rte_ether_addr *mac_addr);
  74
  75 static int virtio_intr_disable(struct rte_eth_dev *dev);
  76
  77 static int virtio_dev_queue_stats_mapping_set(
  78         struct rte_eth_dev *eth_dev,
  79         uint16_t queue_id,
  80         uint8_t stat_idx,
  81         uint8_t is_rx);
  82
  83 int virtio_logtype_init;
  84 int virtio_logtype_driver;
  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 /*
  90  * The set of PCI devices this driver supports
  91  */
  92 static const struct rte_pci_id pci_id_virtio_map[] = {
  93         { RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_LEGACY_DEVICEID_NET) },
  94         { RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_MODERN_DEVICEID_NET) },
  95         { .vendor_id = 0, /* sentinel */ },
  96 };
  97
  98 struct rte_virtio_xstats_name_off {
  99         char name[RTE_ETH_XSTATS_NAME_SIZE];
 100         unsigned offset;
 101 };
 102
 103 /* [rt]x_qX_ is prepended to the name string here */
 104 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
 105         {"good_packets",           offsetof(struct virtnet_rx, stats.packets)},
 106         {"good_bytes",             offsetof(struct virtnet_rx, stats.bytes)},
 107         {"errors",                 offsetof(struct virtnet_rx, stats.errors)},
 108         {"multicast_packets",      offsetof(struct virtnet_rx, stats.multicast)},
 109         {"broadcast_packets",      offsetof(struct virtnet_rx, stats.broadcast)},
 110         {"undersize_packets",      offsetof(struct virtnet_rx, stats.size_bins[0])},
 111         {"size_64_packets",        offsetof(struct virtnet_rx, stats.size_bins[1])},
 112         {"size_65_127_packets",    offsetof(struct virtnet_rx, stats.size_bins[2])},
 113         {"size_128_255_packets",   offsetof(struct virtnet_rx, stats.size_bins[3])},
 114         {"size_256_511_packets",   offsetof(struct virtnet_rx, stats.size_bins[4])},
 115         {"size_512_1023_packets",  offsetof(struct virtnet_rx, stats.size_bins[5])},
 116         {"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
 117         {"size_1519_max_packets",  offsetof(struct virtnet_rx, stats.size_bins[7])},
 118 };
 119
 120 /* [rt]x_qX_ is prepended to the name string here */
 121 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
 122         {"good_packets",           offsetof(struct virtnet_tx, stats.packets)},
 123         {"good_bytes",             offsetof(struct virtnet_tx, stats.bytes)},
 124         {"errors",                 offsetof(struct virtnet_tx, stats.errors)},
 125         {"multicast_packets",      offsetof(struct virtnet_tx, stats.multicast)},
 126         {"broadcast_packets",      offsetof(struct virtnet_tx, stats.broadcast)},
 127         {"undersize_packets",      offsetof(struct virtnet_tx, stats.size_bins[0])},
 128         {"size_64_packets",        offsetof(struct virtnet_tx, stats.size_bins[1])},
 129         {"size_65_127_packets",    offsetof(struct virtnet_tx, stats.size_bins[2])},
 130         {"size_128_255_packets",   offsetof(struct virtnet_tx, stats.size_bins[3])},
 131         {"size_256_511_packets",   offsetof(struct virtnet_tx, stats.size_bins[4])},
 132         {"size_512_1023_packets",  offsetof(struct virtnet_tx, stats.size_bins[5])},
 133         {"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
 134         {"size_1519_max_packets",  offsetof(struct virtnet_tx, stats.size_bins[7])},
 135 };
 136
 137 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
 138                             sizeof(rte_virtio_rxq_stat_strings[0]))
 139 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
 140                             sizeof(rte_virtio_txq_stat_strings[0]))
 141
 142 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
 143
 144 static struct virtio_pmd_ctrl *
 145 virtio_send_command_packed(struct virtnet_ctl *cvq,
 146                            struct virtio_pmd_ctrl *ctrl,
 147                            int *dlen, int pkt_num)
 148 {
 149         struct virtqueue *vq = cvq->vq;
 150         int head;
 151         struct vring_packed_desc *desc = vq->vq_packed.ring.desc;
 152         struct virtio_pmd_ctrl *result;
 153         uint16_t flags;
 154         int sum = 0;
 155         int nb_descs = 0;
 156         int k;
 157
 158         /*
 159          * Format is enforced in qemu code:
 160          * One TX packet for header;
 161          * At least one TX packet per argument;
 162          * One RX packet for ACK.
 163          */
 164         head = vq->vq_avail_idx;
 165         flags = vq->vq_packed.cached_flags;
 166         desc[head].addr = cvq->virtio_net_hdr_mem;
 167         desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
 168         vq->vq_free_cnt--;
 169         nb_descs++;
 170         if (++vq->vq_avail_idx >= vq->vq_nentries) {
 171                 vq->vq_avail_idx -= vq->vq_nentries;
 172                 vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
 173         }
 174
 175         for (k = 0; k < pkt_num; k++) {
 176                 desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
 177                         + sizeof(struct virtio_net_ctrl_hdr)
 178                         + sizeof(ctrl->status) + sizeof(uint8_t) * sum;
 179                 desc[vq->vq_avail_idx].len = dlen[k];
 180                 desc[vq->vq_avail_idx].flags = VRING_DESC_F_NEXT |
 181                         vq->vq_packed.cached_flags;
 182                 sum += dlen[k];
 183                 vq->vq_free_cnt--;
 184                 nb_descs++;
 185                 if (++vq->vq_avail_idx >= vq->vq_nentries) {
 186                         vq->vq_avail_idx -= vq->vq_nentries;
 187                         vq->vq_packed.cached_flags ^=
 188                                 VRING_PACKED_DESC_F_AVAIL_USED;
 189                 }
 190         }
 191
 192         desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
 193                 + sizeof(struct virtio_net_ctrl_hdr);
 194         desc[vq->vq_avail_idx].len = sizeof(ctrl->status);
 195         desc[vq->vq_avail_idx].flags = VRING_DESC_F_WRITE |
 196                 vq->vq_packed.cached_flags;
 197         vq->vq_free_cnt--;
 198         nb_descs++;
 199         if (++vq->vq_avail_idx >= vq->vq_nentries) {
 200                 vq->vq_avail_idx -= vq->vq_nentries;
 201                 vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
 202         }
 203
 204         virtio_wmb(vq->hw->weak_barriers);
 205         desc[head].flags = VRING_DESC_F_NEXT | flags;
 206
 207         virtio_wmb(vq->hw->weak_barriers);
 208         virtqueue_notify(vq);
 209
 210         /* wait for used descriptors in virtqueue */
 211         while (!desc_is_used(&desc[head], vq))
 212                 usleep(100);
 213
 214         virtio_rmb(vq->hw->weak_barriers);
 215
 216         /* now get used descriptors */
 217         vq->vq_free_cnt += nb_descs;
 218         vq->vq_used_cons_idx += nb_descs;
 219         if (vq->vq_used_cons_idx >= vq->vq_nentries) {
 220                 vq->vq_used_cons_idx -= vq->vq_nentries;
 221                 vq->vq_packed.used_wrap_counter ^= 1;
 222         }
 223
 224         PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\n"
 225                         "vq->vq_avail_idx=%d\n"
 226                         "vq->vq_used_cons_idx=%d\n"
 227                         "vq->vq_packed.cached_flags=0x%x\n"
 228                         "vq->vq_packed.used_wrap_counter=%d\n",
 229                         vq->vq_free_cnt,
 230                         vq->vq_avail_idx,
 231                         vq->vq_used_cons_idx,
 232                         vq->vq_packed.cached_flags,
 233                         vq->vq_packed.used_wrap_counter);
 234
 235         result = cvq->virtio_net_hdr_mz->addr;
 236         return result;
 237 }
 238
 239 static struct virtio_pmd_ctrl *
 240 virtio_send_command_split(struct virtnet_ctl *cvq,
 241                           struct virtio_pmd_ctrl *ctrl,
 242                           int *dlen, int pkt_num)
 243 {
 244         struct virtio_pmd_ctrl *result;
 245         struct virtqueue *vq = cvq->vq;
 246         uint32_t head, i;
 247         int k, sum = 0;
 248
 249         head = vq->vq_desc_head_idx;
 250
 251         /*
 252          * Format is enforced in qemu code:
 253          * One TX packet for header;
 254          * At least one TX packet per argument;
 255          * One RX packet for ACK.
 256          */
 257         vq->vq_split.ring.desc[head].flags = VRING_DESC_F_NEXT;
 258         vq->vq_split.ring.desc[head].addr = cvq->virtio_net_hdr_mem;
 259         vq->vq_split.ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
 260         vq->vq_free_cnt--;
 261         i = vq->vq_split.ring.desc[head].next;
 262
 263         for (k = 0; k < pkt_num; k++) {
 264                 vq->vq_split.ring.desc[i].flags = VRING_DESC_F_NEXT;
 265                 vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
 266                         + sizeof(struct virtio_net_ctrl_hdr)
 267                         + sizeof(ctrl->status) + sizeof(uint8_t)*sum;
 268                 vq->vq_split.ring.desc[i].len = dlen[k];
 269                 sum += dlen[k];
 270                 vq->vq_free_cnt--;
 271                 i = vq->vq_split.ring.desc[i].next;
 272         }
 273
 274         vq->vq_split.ring.desc[i].flags = VRING_DESC_F_WRITE;
 275         vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
 276                         + sizeof(struct virtio_net_ctrl_hdr);
 277         vq->vq_split.ring.desc[i].len = sizeof(ctrl->status);
 278         vq->vq_free_cnt--;
 279
 280         vq->vq_desc_head_idx = vq->vq_split.ring.desc[i].next;
 281
 282         vq_update_avail_ring(vq, head);
 283         vq_update_avail_idx(vq);
 284
 285         PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
 286
 287         virtqueue_notify(vq);
 288
 289         rte_rmb();
 290         while (VIRTQUEUE_NUSED(vq) == 0) {
 291                 rte_rmb();
 292                 usleep(100);
 293         }
 294
 295         while (VIRTQUEUE_NUSED(vq)) {
 296                 uint32_t idx, desc_idx, used_idx;
 297                 struct vring_used_elem *uep;
 298
 299                 used_idx = (uint32_t)(vq->vq_used_cons_idx
 300                                 & (vq->vq_nentries - 1));
 301                 uep = &vq->vq_split.ring.used->ring[used_idx];
 302                 idx = (uint32_t) uep->id;
 303                 desc_idx = idx;
 304
 305                 while (vq->vq_split.ring.desc[desc_idx].flags &
 306                                 VRING_DESC_F_NEXT) {
 307                         desc_idx = vq->vq_split.ring.desc[desc_idx].next;
 308                         vq->vq_free_cnt++;
 309                 }
 310
 311                 vq->vq_split.ring.desc[desc_idx].next = vq->vq_desc_head_idx;
 312                 vq->vq_desc_head_idx = idx;
 313
 314                 vq->vq_used_cons_idx++;
 315                 vq->vq_free_cnt++;
 316         }
 317
 318         PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
 319                         vq->vq_free_cnt, vq->vq_desc_head_idx);
 320
 321         result = cvq->virtio_net_hdr_mz->addr;
 322         return result;
 323 }
 324
 325 static int
 326 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
 327                     int *dlen, int pkt_num)
 328 {
 329         virtio_net_ctrl_ack status = ~0;
 330         struct virtio_pmd_ctrl *result;
 331         struct virtqueue *vq;
 332
 333         ctrl->status = status;
 334
 335         if (!cvq || !cvq->vq) {
 336                 PMD_INIT_LOG(ERR, "Control queue is not supported.");
 337                 return -1;
 338         }
 339
 340         rte_spinlock_lock(&cvq->lock);
 341         vq = cvq->vq;
 342
 343         PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
 344                 "vq->hw->cvq = %p vq = %p",
 345                 vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
 346
 347         if (vq->vq_free_cnt < pkt_num + 2 || pkt_num < 1) {
 348                 rte_spinlock_unlock(&cvq->lock);
 349                 return -1;
 350         }
 351
 352         memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
 353                 sizeof(struct virtio_pmd_ctrl));
 354
 355         if (vtpci_packed_queue(vq->hw))
 356                 result = virtio_send_command_packed(cvq, ctrl, dlen, pkt_num);
 357         else
 358                 result = virtio_send_command_split(cvq, ctrl, dlen, pkt_num);
 359
 360         rte_spinlock_unlock(&cvq->lock);
 361         return result->status;
 362 }
 363
 364 static int
 365 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
 366 {
 367         struct virtio_hw *hw = dev->data->dev_private;
 368         struct virtio_pmd_ctrl ctrl;
 369         int dlen[1];
 370         int ret;
 371
 372         ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
 373         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
 374         memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
 375
 376         dlen[0] = sizeof(uint16_t);
 377
 378         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 379         if (ret) {
 380                 PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
 381                           "failed, this is too late now...");
 382                 return -EINVAL;
 383         }
 384
 385         return 0;
 386 }
 387
 388 static void
 389 virtio_dev_queue_release(void *queue __rte_unused)
 390 {
 391         /* do nothing */
 392 }
 393
 394 static uint16_t
 395 virtio_get_nr_vq(struct virtio_hw *hw)
 396 {
 397         uint16_t nr_vq = hw->max_queue_pairs * 2;
 398
 399         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
 400                 nr_vq += 1;
 401
 402         return nr_vq;
 403 }
 404
 405 static void
 406 virtio_init_vring(struct virtqueue *vq)
 407 {
 408         int size = vq->vq_nentries;
 409         uint8_t *ring_mem = vq->vq_ring_virt_mem;
 410
 411         PMD_INIT_FUNC_TRACE();
 412
 413         memset(ring_mem, 0, vq->vq_ring_size);
 414
 415         vq->vq_used_cons_idx = 0;
 416         vq->vq_desc_head_idx = 0;
 417         vq->vq_avail_idx = 0;
 418         vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
 419         vq->vq_free_cnt = vq->vq_nentries;
 420         memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
 421         if (vtpci_packed_queue(vq->hw)) {
 422                 vring_init_packed(&vq->vq_packed.ring, ring_mem,
 423                                   VIRTIO_PCI_VRING_ALIGN, size);
 424                 vring_desc_init_packed(vq, size);
 425         } else {
 426                 struct vring *vr = &vq->vq_split.ring;
 427
 428                 vring_init_split(vr, ring_mem, VIRTIO_PCI_VRING_ALIGN, size);
 429                 vring_desc_init_split(vr->desc, size);
 430         }
 431         /*
 432          * Disable device(host) interrupting guest
 433          */
 434         virtqueue_disable_intr(vq);
 435 }
 436
 437 static int
 438 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
 439 {
 440         char vq_name[VIRTQUEUE_MAX_NAME_SZ];
 441         char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
 442         const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
 443         unsigned int vq_size, size;
 444         struct virtio_hw *hw = dev->data->dev_private;
 445         struct virtnet_rx *rxvq = NULL;
 446         struct virtnet_tx *txvq = NULL;
 447         struct virtnet_ctl *cvq = NULL;
 448         struct virtqueue *vq;
 449         size_t sz_hdr_mz = 0;
 450         void *sw_ring = NULL;
 451         int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
 452         int ret;
 453         int numa_node = dev->device->numa_node;
 454
 455         PMD_INIT_LOG(INFO, "setting up queue: %u on NUMA node %d",
 456                         vtpci_queue_idx, numa_node);
 457
 458         /*
 459          * Read the virtqueue size from the Queue Size field
 460          * Always power of 2 and if 0 virtqueue does not exist
 461          */
 462         vq_size = VTPCI_OPS(hw)->get_queue_num(hw, vtpci_queue_idx);
 463         PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
 464         if (vq_size == 0) {
 465                 PMD_INIT_LOG(ERR, "virtqueue does not exist");
 466                 return -EINVAL;
 467         }
 468
 469         if (!rte_is_power_of_2(vq_size)) {
 470                 PMD_INIT_LOG(ERR, "virtqueue size is not powerof 2");
 471                 return -EINVAL;
 472         }
 473
 474         snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
 475                  dev->data->port_id, vtpci_queue_idx);
 476
 477         size = RTE_ALIGN_CEIL(sizeof(*vq) +
 478                                 vq_size * sizeof(struct vq_desc_extra),
 479                                 RTE_CACHE_LINE_SIZE);
 480         if (queue_type == VTNET_TQ) {
 481                 /*
 482                  * For each xmit packet, allocate a virtio_net_hdr
 483                  * and indirect ring elements
 484                  */
 485                 sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
 486         } else if (queue_type == VTNET_CQ) {
 487                 /* Allocate a page for control vq command, data and status */
 488                 sz_hdr_mz = PAGE_SIZE;
 489         }
 490
 491         vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
 492                                 numa_node);
 493         if (vq == NULL) {
 494                 PMD_INIT_LOG(ERR, "can not allocate vq");
 495                 return -ENOMEM;
 496         }
 497         hw->vqs[vtpci_queue_idx] = vq;
 498
 499         vq->hw = hw;
 500         vq->vq_queue_index = vtpci_queue_idx;
 501         vq->vq_nentries = vq_size;
 502         if (vtpci_packed_queue(hw)) {
 503                 vq->vq_packed.used_wrap_counter = 1;
 504                 vq->vq_packed.cached_flags = VRING_PACKED_DESC_F_AVAIL;
 505                 vq->vq_packed.event_flags_shadow = 0;
 506                 if (queue_type == VTNET_RQ)
 507                         vq->vq_packed.cached_flags |= VRING_DESC_F_WRITE;
 508         }
 509
 510         /*
 511          * Reserve a memzone for vring elements
 512          */
 513         size = vring_size(hw, vq_size, VIRTIO_PCI_VRING_ALIGN);
 514         vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
 515         PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
 516                      size, vq->vq_ring_size);
 517
 518         mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
 519                         numa_node, RTE_MEMZONE_IOVA_CONTIG,
 520                         VIRTIO_PCI_VRING_ALIGN);
 521         if (mz == NULL) {
 522                 if (rte_errno == EEXIST)
 523                         mz = rte_memzone_lookup(vq_name);
 524                 if (mz == NULL) {
 525                         ret = -ENOMEM;
 526                         goto fail_q_alloc;
 527                 }
 528         }
 529
 530         memset(mz->addr, 0, mz->len);
 531
 532         vq->vq_ring_mem = mz->iova;
 533         vq->vq_ring_virt_mem = mz->addr;
 534         PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%" PRIx64,
 535                      (uint64_t)mz->iova);
 536         PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
 537                      (uint64_t)(uintptr_t)mz->addr);
 538
 539         virtio_init_vring(vq);
 540
 541         if (sz_hdr_mz) {
 542                 snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
 543                          dev->data->port_id, vtpci_queue_idx);
 544                 hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
 545                                 numa_node, RTE_MEMZONE_IOVA_CONTIG,
 546                                 RTE_CACHE_LINE_SIZE);
 547                 if (hdr_mz == NULL) {
 548                         if (rte_errno == EEXIST)
 549                                 hdr_mz = rte_memzone_lookup(vq_hdr_name);
 550                         if (hdr_mz == NULL) {
 551                                 ret = -ENOMEM;
 552                                 goto fail_q_alloc;
 553                         }
 554                 }
 555         }
 556
 557         if (queue_type == VTNET_RQ) {
 558                 size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
 559                                sizeof(vq->sw_ring[0]);
 560
 561                 sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
 562                                 RTE_CACHE_LINE_SIZE, numa_node);
 563                 if (!sw_ring) {
 564                         PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
 565                         ret = -ENOMEM;
 566                         goto fail_q_alloc;
 567                 }
 568
 569                 vq->sw_ring = sw_ring;
 570                 rxvq = &vq->rxq;
 571                 rxvq->vq = vq;
 572                 rxvq->port_id = dev->data->port_id;
 573                 rxvq->mz = mz;
 574         } else if (queue_type == VTNET_TQ) {
 575                 txvq = &vq->txq;
 576                 txvq->vq = vq;
 577                 txvq->port_id = dev->data->port_id;
 578                 txvq->mz = mz;
 579                 txvq->virtio_net_hdr_mz = hdr_mz;
 580                 txvq->virtio_net_hdr_mem = hdr_mz->iova;
 581         } else if (queue_type == VTNET_CQ) {
 582                 cvq = &vq->cq;
 583                 cvq->vq = vq;
 584                 cvq->mz = mz;
 585                 cvq->virtio_net_hdr_mz = hdr_mz;
 586                 cvq->virtio_net_hdr_mem = hdr_mz->iova;
 587                 memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
 588
 589                 hw->cvq = cvq;
 590         }
 591
 592         /* For virtio_user case (that is when hw->dev is NULL), we use
 593          * virtual address. And we need properly set _offset_, please see
 594          * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
 595          */
 596         if (!hw->virtio_user_dev)
 597                 vq->offset = offsetof(struct rte_mbuf, buf_iova);
 598         else {
 599                 vq->vq_ring_mem = (uintptr_t)mz->addr;
 600                 vq->offset = offsetof(struct rte_mbuf, buf_addr);
 601                 if (queue_type == VTNET_TQ)
 602                         txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
 603                 else if (queue_type == VTNET_CQ)
 604                         cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
 605         }
 606
 607         if (queue_type == VTNET_TQ) {
 608                 struct virtio_tx_region *txr;
 609                 unsigned int i;
 610
 611                 txr = hdr_mz->addr;
 612                 memset(txr, 0, vq_size * sizeof(*txr));
 613                 for (i = 0; i < vq_size; i++) {
 614                         struct vring_desc *start_dp = txr[i].tx_indir;
 615
 616                         /* first indirect descriptor is always the tx header */
 617                         if (!vtpci_packed_queue(hw)) {
 618                                 vring_desc_init_split(start_dp,
 619                                                       RTE_DIM(txr[i].tx_indir));
 620                                 start_dp->addr = txvq->virtio_net_hdr_mem
 621                                         + i * sizeof(*txr)
 622                                         + offsetof(struct virtio_tx_region,
 623                                                    tx_hdr);
 624                                 start_dp->len = hw->vtnet_hdr_size;
 625                                 start_dp->flags = VRING_DESC_F_NEXT;
 626                         }
 627                 }
 628         }
 629
 630         if (VTPCI_OPS(hw)->setup_queue(hw, vq) < 0) {
 631                 PMD_INIT_LOG(ERR, "setup_queue failed");
 632                 return -EINVAL;
 633         }
 634
 635         return 0;
 636
 637 fail_q_alloc:
 638         rte_free(sw_ring);
 639         rte_memzone_free(hdr_mz);
 640         rte_memzone_free(mz);
 641         rte_free(vq);
 642
 643         return ret;
 644 }
 645
 646 static void
 647 virtio_free_queues(struct virtio_hw *hw)
 648 {
 649         uint16_t nr_vq = virtio_get_nr_vq(hw);
 650         struct virtqueue *vq;
 651         int queue_type;
 652         uint16_t i;
 653
 654         if (hw->vqs == NULL)
 655                 return;
 656
 657         for (i = 0; i < nr_vq; i++) {
 658                 vq = hw->vqs[i];
 659                 if (!vq)
 660                         continue;
 661
 662                 queue_type = virtio_get_queue_type(hw, i);
 663                 if (queue_type == VTNET_RQ) {
 664                         rte_free(vq->sw_ring);
 665                         rte_memzone_free(vq->rxq.mz);
 666                 } else if (queue_type == VTNET_TQ) {
 667                         rte_memzone_free(vq->txq.mz);
 668                         rte_memzone_free(vq->txq.virtio_net_hdr_mz);
 669                 } else {
 670                         rte_memzone_free(vq->cq.mz);
 671                         rte_memzone_free(vq->cq.virtio_net_hdr_mz);
 672                 }
 673
 674                 rte_free(vq);
 675                 hw->vqs[i] = NULL;
 676         }
 677
 678         rte_free(hw->vqs);
 679         hw->vqs = NULL;
 680 }
 681
 682 static int
 683 virtio_alloc_queues(struct rte_eth_dev *dev)
 684 {
 685         struct virtio_hw *hw = dev->data->dev_private;
 686         uint16_t nr_vq = virtio_get_nr_vq(hw);
 687         uint16_t i;
 688         int ret;
 689
 690         hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
 691         if (!hw->vqs) {
 692                 PMD_INIT_LOG(ERR, "failed to allocate vqs");
 693                 return -ENOMEM;
 694         }
 695
 696         for (i = 0; i < nr_vq; i++) {
 697                 ret = virtio_init_queue(dev, i);
 698                 if (ret < 0) {
 699                         virtio_free_queues(hw);
 700                         return ret;
 701                 }
 702         }
 703
 704         return 0;
 705 }
 706
 707 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
 708
 709 static void
 710 virtio_dev_close(struct rte_eth_dev *dev)
 711 {
 712         struct virtio_hw *hw = dev->data->dev_private;
 713         struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
 714
 715         PMD_INIT_LOG(DEBUG, "virtio_dev_close");
 716
 717         if (!hw->opened)
 718                 return;
 719         hw->opened = false;
 720
 721         /* reset the NIC */
 722         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
 723                 VTPCI_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
 724         if (intr_conf->rxq)
 725                 virtio_queues_unbind_intr(dev);
 726
 727         if (intr_conf->lsc || intr_conf->rxq) {
 728                 virtio_intr_disable(dev);
 729                 rte_intr_efd_disable(dev->intr_handle);
 730                 rte_free(dev->intr_handle->intr_vec);
 731                 dev->intr_handle->intr_vec = NULL;
 732         }
 733
 734         vtpci_reset(hw);
 735         virtio_dev_free_mbufs(dev);
 736         virtio_free_queues(hw);
 737
 738 #ifdef RTE_VIRTIO_USER
 739         if (hw->virtio_user_dev)
 740                 virtio_user_dev_uninit(hw->virtio_user_dev);
 741         else
 742 #endif
 743         if (dev->device) {
 744                 rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(dev));
 745                 if (!hw->modern)
 746                         rte_pci_ioport_unmap(VTPCI_IO(hw));
 747         }
 748 }
 749
 750 static void
 751 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
 752 {
 753         struct virtio_hw *hw = dev->data->dev_private;
 754         struct virtio_pmd_ctrl ctrl;
 755         int dlen[1];
 756         int ret;
 757
 758         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 759                 PMD_INIT_LOG(INFO, "host does not support rx control");
 760                 return;
 761         }
 762
 763         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 764         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 765         ctrl.data[0] = 1;
 766         dlen[0] = 1;
 767
 768         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 769         if (ret)
 770                 PMD_INIT_LOG(ERR, "Failed to enable promisc");
 771 }
 772
 773 static void
 774 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
 775 {
 776         struct virtio_hw *hw = dev->data->dev_private;
 777         struct virtio_pmd_ctrl ctrl;
 778         int dlen[1];
 779         int ret;
 780
 781         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 782                 PMD_INIT_LOG(INFO, "host does not support rx control");
 783                 return;
 784         }
 785
 786         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 787         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 788         ctrl.data[0] = 0;
 789         dlen[0] = 1;
 790
 791         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 792         if (ret)
 793                 PMD_INIT_LOG(ERR, "Failed to disable promisc");
 794 }
 795
 796 static void
 797 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
 798 {
 799         struct virtio_hw *hw = dev->data->dev_private;
 800         struct virtio_pmd_ctrl ctrl;
 801         int dlen[1];
 802         int ret;
 803
 804         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 805                 PMD_INIT_LOG(INFO, "host does not support rx control");
 806                 return;
 807         }
 808
 809         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 810         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 811         ctrl.data[0] = 1;
 812         dlen[0] = 1;
 813
 814         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 815         if (ret)
 816                 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
 817 }
 818
 819 static void
 820 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
 821 {
 822         struct virtio_hw *hw = dev->data->dev_private;
 823         struct virtio_pmd_ctrl ctrl;
 824         int dlen[1];
 825         int ret;
 826
 827         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 828                 PMD_INIT_LOG(INFO, "host does not support rx control");
 829                 return;
 830         }
 831
 832         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 833         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 834         ctrl.data[0] = 0;
 835         dlen[0] = 1;
 836
 837         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 838         if (ret)
 839                 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
 840 }
 841
 842 #define VLAN_TAG_LEN           4    /* 802.3ac tag (not DMA'd) */
 843 static int
 844 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 845 {
 846         struct virtio_hw *hw = dev->data->dev_private;
 847         uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
 848                                  hw->vtnet_hdr_size;
 849         uint32_t frame_size = mtu + ether_hdr_len;
 850         uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
 851
 852         max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
 853
 854         if (mtu < RTE_ETHER_MIN_MTU || frame_size > max_frame_size) {
 855                 PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
 856                         RTE_ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
 857                 return -EINVAL;
 858         }
 859         return 0;
 860 }
 861
 862 static int
 863 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
 864 {
 865         struct virtio_hw *hw = dev->data->dev_private;
 866         struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
 867         struct virtqueue *vq = rxvq->vq;
 868
 869         virtqueue_enable_intr(vq);
 870         virtio_mb(hw->weak_barriers);
 871         return 0;
 872 }
 873
 874 static int
 875 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
 876 {
 877         struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
 878         struct virtqueue *vq = rxvq->vq;
 879
 880         virtqueue_disable_intr(vq);
 881         return 0;
 882 }
 883
 884 /*
 885  * dev_ops for virtio, bare necessities for basic operation
 886  */
 887 static const struct eth_dev_ops virtio_eth_dev_ops = {
 888         .dev_configure           = virtio_dev_configure,
 889         .dev_start               = virtio_dev_start,
 890         .dev_stop                = virtio_dev_stop,
 891         .dev_close               = virtio_dev_close,
 892         .promiscuous_enable      = virtio_dev_promiscuous_enable,
 893         .promiscuous_disable     = virtio_dev_promiscuous_disable,
 894         .allmulticast_enable     = virtio_dev_allmulticast_enable,
 895         .allmulticast_disable    = virtio_dev_allmulticast_disable,
 896         .mtu_set                 = virtio_mtu_set,
 897         .dev_infos_get           = virtio_dev_info_get,
 898         .stats_get               = virtio_dev_stats_get,
 899         .xstats_get              = virtio_dev_xstats_get,
 900         .xstats_get_names        = virtio_dev_xstats_get_names,
 901         .stats_reset             = virtio_dev_stats_reset,
 902         .xstats_reset            = virtio_dev_stats_reset,
 903         .link_update             = virtio_dev_link_update,
 904         .vlan_offload_set        = virtio_dev_vlan_offload_set,
 905         .rx_queue_setup          = virtio_dev_rx_queue_setup,
 906         .rx_queue_intr_enable    = virtio_dev_rx_queue_intr_enable,
 907         .rx_queue_intr_disable   = virtio_dev_rx_queue_intr_disable,
 908         .rx_queue_release        = virtio_dev_queue_release,
 909         .rx_descriptor_done      = virtio_dev_rx_queue_done,
 910         .tx_queue_setup          = virtio_dev_tx_queue_setup,
 911         .tx_queue_release        = virtio_dev_queue_release,
 912         /* collect stats per queue */
 913         .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
 914         .vlan_filter_set         = virtio_vlan_filter_set,
 915         .mac_addr_add            = virtio_mac_addr_add,
 916         .mac_addr_remove         = virtio_mac_addr_remove,
 917         .mac_addr_set            = virtio_mac_addr_set,
 918 };
 919
 920 /*
 921  * dev_ops for virtio-user in secondary processes, as we just have
 922  * some limited supports currently.
 923  */
 924 const struct eth_dev_ops virtio_user_secondary_eth_dev_ops = {
 925         .dev_infos_get           = virtio_dev_info_get,
 926         .stats_get               = virtio_dev_stats_get,
 927         .xstats_get              = virtio_dev_xstats_get,
 928         .xstats_get_names        = virtio_dev_xstats_get_names,
 929         .stats_reset             = virtio_dev_stats_reset,
 930         .xstats_reset            = virtio_dev_stats_reset,
 931         /* collect stats per queue */
 932         .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
 933 };
 934
 935 static void
 936 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 937 {
 938         unsigned i;
 939
 940         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 941                 const struct virtnet_tx *txvq = dev->data->tx_queues[i];
 942                 if (txvq == NULL)
 943                         continue;
 944
 945                 stats->opackets += txvq->stats.packets;
 946                 stats->obytes += txvq->stats.bytes;
 947                 stats->oerrors += txvq->stats.errors;
 948
 949                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 950                         stats->q_opackets[i] = txvq->stats.packets;
 951                         stats->q_obytes[i] = txvq->stats.bytes;
 952                 }
 953         }
 954
 955         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 956                 const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
 957                 if (rxvq == NULL)
 958                         continue;
 959
 960                 stats->ipackets += rxvq->stats.packets;
 961                 stats->ibytes += rxvq->stats.bytes;
 962                 stats->ierrors += rxvq->stats.errors;
 963
 964                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 965                         stats->q_ipackets[i] = rxvq->stats.packets;
 966                         stats->q_ibytes[i] = rxvq->stats.bytes;
 967                 }
 968         }
 969
 970         stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
 971 }
 972
 973 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
 974                                        struct rte_eth_xstat_name *xstats_names,
 975                                        __rte_unused unsigned limit)
 976 {
 977         unsigned i;
 978         unsigned count = 0;
 979         unsigned t;
 980
 981         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
 982                 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
 983
 984         if (xstats_names != NULL) {
 985                 /* Note: limit checked in rte_eth_xstats_names() */
 986
 987                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
 988                         struct virtnet_rx *rxvq = dev->data->rx_queues[i];
 989                         if (rxvq == NULL)
 990                                 continue;
 991                         for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
 992                                 snprintf(xstats_names[count].name,
 993                                         sizeof(xstats_names[count].name),
 994                                         "rx_q%u_%s", i,
 995                                         rte_virtio_rxq_stat_strings[t].name);
 996                                 count++;
 997                         }
 998                 }
 999
1000                 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1001                         struct virtnet_tx *txvq = dev->data->tx_queues[i];
1002                         if (txvq == NULL)
1003                                 continue;
1004                         for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1005                                 snprintf(xstats_names[count].name,
1006                                         sizeof(xstats_names[count].name),
1007                                         "tx_q%u_%s", i,
1008                                         rte_virtio_txq_stat_strings[t].name);
1009                                 count++;
1010                         }
1011                 }
1012                 return count;
1013         }
1014         return nstats;
1015 }
1016
1017 static int
1018 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1019                       unsigned n)
1020 {
1021         unsigned i;
1022         unsigned count = 0;
1023
1024         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1025                 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1026
1027         if (n < nstats)
1028                 return nstats;
1029
1030         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1031                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1032
1033                 if (rxvq == NULL)
1034                         continue;
1035
1036                 unsigned t;
1037
1038                 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1039                         xstats[count].value = *(uint64_t *)(((char *)rxvq) +
1040                                 rte_virtio_rxq_stat_strings[t].offset);
1041                         xstats[count].id = count;
1042                         count++;
1043                 }
1044         }
1045
1046         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1047                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1048
1049                 if (txvq == NULL)
1050                         continue;
1051
1052                 unsigned t;
1053
1054                 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1055                         xstats[count].value = *(uint64_t *)(((char *)txvq) +
1056                                 rte_virtio_txq_stat_strings[t].offset);
1057                         xstats[count].id = count;
1058                         count++;
1059                 }
1060         }
1061
1062         return count;
1063 }
1064
1065 static int
1066 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1067 {
1068         virtio_update_stats(dev, stats);
1069
1070         return 0;
1071 }
1072
1073 static void
1074 virtio_dev_stats_reset(struct rte_eth_dev *dev)
1075 {
1076         unsigned int i;
1077
1078         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1079                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1080                 if (txvq == NULL)
1081                         continue;
1082
1083                 txvq->stats.packets = 0;
1084                 txvq->stats.bytes = 0;
1085                 txvq->stats.errors = 0;
1086                 txvq->stats.multicast = 0;
1087                 txvq->stats.broadcast = 0;
1088                 memset(txvq->stats.size_bins, 0,
1089                        sizeof(txvq->stats.size_bins[0]) * 8);
1090         }
1091
1092         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1093                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1094                 if (rxvq == NULL)
1095                         continue;
1096
1097                 rxvq->stats.packets = 0;
1098                 rxvq->stats.bytes = 0;
1099                 rxvq->stats.errors = 0;
1100                 rxvq->stats.multicast = 0;
1101                 rxvq->stats.broadcast = 0;
1102                 memset(rxvq->stats.size_bins, 0,
1103                        sizeof(rxvq->stats.size_bins[0]) * 8);
1104         }
1105 }
1106
1107 static void
1108 virtio_set_hwaddr(struct virtio_hw *hw)
1109 {
1110         vtpci_write_dev_config(hw,
1111                         offsetof(struct virtio_net_config, mac),
1112                         &hw->mac_addr, RTE_ETHER_ADDR_LEN);
1113 }
1114
1115 static void
1116 virtio_get_hwaddr(struct virtio_hw *hw)
1117 {
1118         if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1119                 vtpci_read_dev_config(hw,
1120                         offsetof(struct virtio_net_config, mac),
1121                         &hw->mac_addr, RTE_ETHER_ADDR_LEN);
1122         } else {
1123                 rte_eth_random_addr(&hw->mac_addr[0]);
1124                 virtio_set_hwaddr(hw);
1125         }
1126 }
1127
1128 static int
1129 virtio_mac_table_set(struct virtio_hw *hw,
1130                      const struct virtio_net_ctrl_mac *uc,
1131                      const struct virtio_net_ctrl_mac *mc)
1132 {
1133         struct virtio_pmd_ctrl ctrl;
1134         int err, len[2];
1135
1136         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1137                 PMD_DRV_LOG(INFO, "host does not support mac table");
1138                 return -1;
1139         }
1140
1141         ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1142         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1143
1144         len[0] = uc->entries * RTE_ETHER_ADDR_LEN + sizeof(uc->entries);
1145         memcpy(ctrl.data, uc, len[0]);
1146
1147         len[1] = mc->entries * RTE_ETHER_ADDR_LEN + sizeof(mc->entries);
1148         memcpy(ctrl.data + len[0], mc, len[1]);
1149
1150         err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1151         if (err != 0)
1152                 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1153         return err;
1154 }
1155
1156 static int
1157 virtio_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1158                     uint32_t index, uint32_t vmdq __rte_unused)
1159 {
1160         struct virtio_hw *hw = dev->data->dev_private;
1161         const struct rte_ether_addr *addrs = dev->data->mac_addrs;
1162         unsigned int i;
1163         struct virtio_net_ctrl_mac *uc, *mc;
1164
1165         if (index >= VIRTIO_MAX_MAC_ADDRS) {
1166                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1167                 return -EINVAL;
1168         }
1169
1170         uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1171                 sizeof(uc->entries));
1172         uc->entries = 0;
1173         mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1174                 sizeof(mc->entries));
1175         mc->entries = 0;
1176
1177         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1178                 const struct rte_ether_addr *addr
1179                         = (i == index) ? mac_addr : addrs + i;
1180                 struct virtio_net_ctrl_mac *tbl
1181                         = rte_is_multicast_ether_addr(addr) ? mc : uc;
1182
1183                 memcpy(&tbl->macs[tbl->entries++], addr, RTE_ETHER_ADDR_LEN);
1184         }
1185
1186         return virtio_mac_table_set(hw, uc, mc);
1187 }
1188
1189 static void
1190 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1191 {
1192         struct virtio_hw *hw = dev->data->dev_private;
1193         struct rte_ether_addr *addrs = dev->data->mac_addrs;
1194         struct virtio_net_ctrl_mac *uc, *mc;
1195         unsigned int i;
1196
1197         if (index >= VIRTIO_MAX_MAC_ADDRS) {
1198                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1199                 return;
1200         }
1201
1202         uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1203                 sizeof(uc->entries));
1204         uc->entries = 0;
1205         mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1206                 sizeof(mc->entries));
1207         mc->entries = 0;
1208
1209         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1210                 struct virtio_net_ctrl_mac *tbl;
1211
1212                 if (i == index || rte_is_zero_ether_addr(addrs + i))
1213                         continue;
1214
1215                 tbl = rte_is_multicast_ether_addr(addrs + i) ? mc : uc;
1216                 memcpy(&tbl->macs[tbl->entries++], addrs + i,
1217                         RTE_ETHER_ADDR_LEN);
1218         }
1219
1220         virtio_mac_table_set(hw, uc, mc);
1221 }
1222
1223 static int
1224 virtio_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1225 {
1226         struct virtio_hw *hw = dev->data->dev_private;
1227
1228         memcpy(hw->mac_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1229
1230         /* Use atomic update if available */
1231         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1232                 struct virtio_pmd_ctrl ctrl;
1233                 int len = RTE_ETHER_ADDR_LEN;
1234
1235                 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1236                 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1237
1238                 memcpy(ctrl.data, mac_addr, RTE_ETHER_ADDR_LEN);
1239                 return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1240         }
1241
1242         if (!vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1243                 return -ENOTSUP;
1244
1245         virtio_set_hwaddr(hw);
1246         return 0;
1247 }
1248
1249 static int
1250 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1251 {
1252         struct virtio_hw *hw = dev->data->dev_private;
1253         struct virtio_pmd_ctrl ctrl;
1254         int len;
1255
1256         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1257                 return -ENOTSUP;
1258
1259         ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1260         ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1261         memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1262         len = sizeof(vlan_id);
1263
1264         return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1265 }
1266
1267 static int
1268 virtio_intr_enable(struct rte_eth_dev *dev)
1269 {
1270         struct virtio_hw *hw = dev->data->dev_private;
1271
1272         if (rte_intr_enable(dev->intr_handle) < 0)
1273                 return -1;
1274
1275         if (!hw->virtio_user_dev)
1276                 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1277
1278         return 0;
1279 }
1280
1281 static int
1282 virtio_intr_disable(struct rte_eth_dev *dev)
1283 {
1284         struct virtio_hw *hw = dev->data->dev_private;
1285
1286         if (rte_intr_disable(dev->intr_handle) < 0)
1287                 return -1;
1288
1289         if (!hw->virtio_user_dev)
1290                 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1291
1292         return 0;
1293 }
1294
1295 static int
1296 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1297 {
1298         uint64_t host_features;
1299
1300         /* Prepare guest_features: feature that driver wants to support */
1301         PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1302                 req_features);
1303
1304         /* Read device(host) feature bits */
1305         host_features = VTPCI_OPS(hw)->get_features(hw);
1306         PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1307                 host_features);
1308
1309         /* If supported, ensure MTU value is valid before acknowledging it. */
1310         if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1311                 struct virtio_net_config config;
1312
1313                 vtpci_read_dev_config(hw,
1314                         offsetof(struct virtio_net_config, mtu),
1315                         &config.mtu, sizeof(config.mtu));
1316
1317                 if (config.mtu < RTE_ETHER_MIN_MTU)
1318                         req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1319         }
1320
1321         /*
1322          * Negotiate features: Subset of device feature bits are written back
1323          * guest feature bits.
1324          */
1325         hw->guest_features = req_features;
1326         hw->guest_features = vtpci_negotiate_features(hw, host_features);
1327         PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1328                 hw->guest_features);
1329
1330         if (hw->modern) {
1331                 if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1332                         PMD_INIT_LOG(ERR,
1333                                 "VIRTIO_F_VERSION_1 features is not enabled.");
1334                         return -1;
1335                 }
1336                 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1337                 if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1338                         PMD_INIT_LOG(ERR,
1339                                 "failed to set FEATURES_OK status!");
1340                         return -1;
1341                 }
1342         }
1343
1344         hw->req_guest_features = req_features;
1345
1346         return 0;
1347 }
1348
1349 int
1350 virtio_dev_pause(struct rte_eth_dev *dev)
1351 {
1352         struct virtio_hw *hw = dev->data->dev_private;
1353
1354         rte_spinlock_lock(&hw->state_lock);
1355
1356         if (hw->started == 0) {
1357                 /* Device is just stopped. */
1358                 rte_spinlock_unlock(&hw->state_lock);
1359                 return -1;
1360         }
1361         hw->started = 0;
1362         /*
1363          * Prevent the worker threads from touching queues to avoid contention,
1364          * 1 ms should be enough for the ongoing Tx function to finish.
1365          */
1366         rte_delay_ms(1);
1367         return 0;
1368 }
1369
1370 /*
1371  * Recover hw state to let the worker threads continue.
1372  */
1373 void
1374 virtio_dev_resume(struct rte_eth_dev *dev)
1375 {
1376         struct virtio_hw *hw = dev->data->dev_private;
1377
1378         hw->started = 1;
1379         rte_spinlock_unlock(&hw->state_lock);
1380 }
1381
1382 /*
1383  * Should be called only after device is paused.
1384  */
1385 int
1386 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1387                 int nb_pkts)
1388 {
1389         struct virtio_hw *hw = dev->data->dev_private;
1390         struct virtnet_tx *txvq = dev->data->tx_queues[0];
1391         int ret;
1392
1393         hw->inject_pkts = tx_pkts;
1394         ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1395         hw->inject_pkts = NULL;
1396
1397         return ret;
1398 }
1399
1400 static void
1401 virtio_notify_peers(struct rte_eth_dev *dev)
1402 {
1403         struct virtio_hw *hw = dev->data->dev_private;
1404         struct virtnet_rx *rxvq;
1405         struct rte_mbuf *rarp_mbuf;
1406
1407         if (!dev->data->rx_queues)
1408                 return;
1409
1410         rxvq = dev->data->rx_queues[0];
1411         if (!rxvq)
1412                 return;
1413
1414         rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1415                         (struct rte_ether_addr *)hw->mac_addr);
1416         if (rarp_mbuf == NULL) {
1417                 PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1418                 return;
1419         }
1420
1421         /* If virtio port just stopped, no need to send RARP */
1422         if (virtio_dev_pause(dev) < 0) {
1423                 rte_pktmbuf_free(rarp_mbuf);
1424                 return;
1425         }
1426
1427         virtio_inject_pkts(dev, &rarp_mbuf, 1);
1428         virtio_dev_resume(dev);
1429 }
1430
1431 static void
1432 virtio_ack_link_announce(struct rte_eth_dev *dev)
1433 {
1434         struct virtio_hw *hw = dev->data->dev_private;
1435         struct virtio_pmd_ctrl ctrl;
1436
1437         ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1438         ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1439
1440         virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1441 }
1442
1443 /*
1444  * Process virtio config changed interrupt. Call the callback
1445  * if link state changed, generate gratuitous RARP packet if
1446  * the status indicates an ANNOUNCE.
1447  */
1448 void
1449 virtio_interrupt_handler(void *param)
1450 {
1451         struct rte_eth_dev *dev = param;
1452         struct virtio_hw *hw = dev->data->dev_private;
1453         uint8_t isr;
1454         uint16_t status;
1455
1456         /* Read interrupt status which clears interrupt */
1457         isr = vtpci_isr(hw);
1458         PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1459
1460         if (virtio_intr_enable(dev) < 0)
1461                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1462
1463         if (isr & VIRTIO_PCI_ISR_CONFIG) {
1464                 if (virtio_dev_link_update(dev, 0) == 0)
1465                         _rte_eth_dev_callback_process(dev,
1466                                                       RTE_ETH_EVENT_INTR_LSC,
1467                                                       NULL);
1468
1469                 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1470                         vtpci_read_dev_config(hw,
1471                                 offsetof(struct virtio_net_config, status),
1472                                 &status, sizeof(status));
1473                         if (status & VIRTIO_NET_S_ANNOUNCE) {
1474                                 virtio_notify_peers(dev);
1475                                 if (hw->cvq)
1476                                         virtio_ack_link_announce(dev);
1477                         }
1478                 }
1479         }
1480 }
1481
1482 /* set rx and tx handlers according to what is supported */
1483 static void
1484 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1485 {
1486         struct virtio_hw *hw = eth_dev->data->dev_private;
1487
1488         if (vtpci_packed_queue(hw)) {
1489                 PMD_INIT_LOG(INFO,
1490                         "virtio: using packed ring %s Tx path on port %u",
1491                         hw->use_inorder_tx ? "inorder" : "standard",
1492                         eth_dev->data->port_id);
1493                 eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed;
1494         } else {
1495                 if (hw->use_inorder_tx) {
1496                         PMD_INIT_LOG(INFO, "virtio: using inorder Tx path on port %u",
1497                                 eth_dev->data->port_id);
1498                         eth_dev->tx_pkt_burst = virtio_xmit_pkts_inorder;
1499                 } else {
1500                         PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1501                                 eth_dev->data->port_id);
1502                         eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1503                 }
1504         }
1505
1506         if (vtpci_packed_queue(hw)) {
1507                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1508                         PMD_INIT_LOG(INFO,
1509                                 "virtio: using packed ring mergeable buffer Rx path on port %u",
1510                                 eth_dev->data->port_id);
1511                         eth_dev->rx_pkt_burst =
1512                                 &virtio_recv_mergeable_pkts_packed;
1513                 } else {
1514                         PMD_INIT_LOG(INFO,
1515                                 "virtio: using packed ring standard Rx path on port %u",
1516                                 eth_dev->data->port_id);
1517                         eth_dev->rx_pkt_burst = &virtio_recv_pkts_packed;
1518                 }
1519         } else {
1520                 if (hw->use_simple_rx) {
1521                         PMD_INIT_LOG(INFO, "virtio: using simple Rx path on port %u",
1522                                 eth_dev->data->port_id);
1523                         eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1524                 } else if (hw->use_inorder_rx) {
1525                         PMD_INIT_LOG(INFO,
1526                                 "virtio: using inorder Rx path on port %u",
1527                                 eth_dev->data->port_id);
1528                         eth_dev->rx_pkt_burst = &virtio_recv_pkts_inorder;
1529                 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1530                         PMD_INIT_LOG(INFO,
1531                                 "virtio: using mergeable buffer Rx path on port %u",
1532                                 eth_dev->data->port_id);
1533                         eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1534                 } else {
1535                         PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1536                                 eth_dev->data->port_id);
1537                         eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1538                 }
1539         }
1540
1541 }
1542
1543 /* Only support 1:1 queue/interrupt mapping so far.
1544  * TODO: support n:1 queue/interrupt mapping when there are limited number of
1545  * interrupt vectors (<N+1).
1546  */
1547 static int
1548 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1549 {
1550         uint32_t i;
1551         struct virtio_hw *hw = dev->data->dev_private;
1552
1553         PMD_INIT_LOG(INFO, "queue/interrupt binding");
1554         for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1555                 dev->intr_handle->intr_vec[i] = i + 1;
1556                 if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1557                                                  VIRTIO_MSI_NO_VECTOR) {
1558                         PMD_DRV_LOG(ERR, "failed to set queue vector");
1559                         return -EBUSY;
1560                 }
1561         }
1562
1563         return 0;
1564 }
1565
1566 static void
1567 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1568 {
1569         uint32_t i;
1570         struct virtio_hw *hw = dev->data->dev_private;
1571
1572         PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1573         for (i = 0; i < dev->data->nb_rx_queues; ++i)
1574                 VTPCI_OPS(hw)->set_queue_irq(hw,
1575                                              hw->vqs[i * VTNET_CQ],
1576                                              VIRTIO_MSI_NO_VECTOR);
1577 }
1578
1579 static int
1580 virtio_configure_intr(struct rte_eth_dev *dev)
1581 {
1582         struct virtio_hw *hw = dev->data->dev_private;
1583
1584         if (!rte_intr_cap_multiple(dev->intr_handle)) {
1585                 PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1586                 return -ENOTSUP;
1587         }
1588
1589         if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1590                 PMD_INIT_LOG(ERR, "Fail to create eventfd");
1591                 return -1;
1592         }
1593
1594         if (!dev->intr_handle->intr_vec) {
1595                 dev->intr_handle->intr_vec =
1596                         rte_zmalloc("intr_vec",
1597                                     hw->max_queue_pairs * sizeof(int), 0);
1598                 if (!dev->intr_handle->intr_vec) {
1599                         PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1600                                      hw->max_queue_pairs);
1601                         return -ENOMEM;
1602                 }
1603         }
1604
1605         /* Re-register callback to update max_intr */
1606         rte_intr_callback_unregister(dev->intr_handle,
1607                                      virtio_interrupt_handler,
1608                                      dev);
1609         rte_intr_callback_register(dev->intr_handle,
1610                                    virtio_interrupt_handler,
1611                                    dev);
1612
1613         /* DO NOT try to remove this! This function will enable msix, or QEMU
1614          * will encounter SIGSEGV when DRIVER_OK is sent.
1615          * And for legacy devices, this should be done before queue/vec binding
1616          * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1617          * (22) will be ignored.
1618          */
1619         if (virtio_intr_enable(dev) < 0) {
1620                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1621                 return -1;
1622         }
1623
1624         if (virtio_queues_bind_intr(dev) < 0) {
1625                 PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1626                 return -1;
1627         }
1628
1629         return 0;
1630 }
1631
1632 /* reset device and renegotiate features if needed */
1633 static int
1634 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1635 {
1636         struct virtio_hw *hw = eth_dev->data->dev_private;
1637         struct virtio_net_config *config;
1638         struct virtio_net_config local_config;
1639         struct rte_pci_device *pci_dev = NULL;
1640         int ret;
1641
1642         /* Reset the device although not necessary at startup */
1643         vtpci_reset(hw);
1644
1645         if (hw->vqs) {
1646                 virtio_dev_free_mbufs(eth_dev);
1647                 virtio_free_queues(hw);
1648         }
1649
1650         /* Tell the host we've noticed this device. */
1651         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1652
1653         /* Tell the host we've known how to drive the device. */
1654         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1655         if (virtio_negotiate_features(hw, req_features) < 0)
1656                 return -1;
1657
1658         hw->weak_barriers = !vtpci_with_feature(hw, VIRTIO_F_ORDER_PLATFORM);
1659
1660         if (!hw->virtio_user_dev)
1661                 pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1662
1663         /* If host does not support both status and MSI-X then disable LSC */
1664         if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) &&
1665             hw->use_msix != VIRTIO_MSIX_NONE)
1666                 eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1667         else
1668                 eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1669
1670         /* Setting up rx_header size for the device */
1671         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1672             vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
1673             vtpci_with_feature(hw, VIRTIO_F_RING_PACKED))
1674                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1675         else
1676                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1677
1678         /* Copy the permanent MAC address to: virtio_hw */
1679         virtio_get_hwaddr(hw);
1680         rte_ether_addr_copy((struct rte_ether_addr *)hw->mac_addr,
1681                         &eth_dev->data->mac_addrs[0]);
1682         PMD_INIT_LOG(DEBUG,
1683                      "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1684                      hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1685                      hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1686
1687         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1688                 config = &local_config;
1689
1690                 vtpci_read_dev_config(hw,
1691                         offsetof(struct virtio_net_config, mac),
1692                         &config->mac, sizeof(config->mac));
1693
1694                 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1695                         vtpci_read_dev_config(hw,
1696                                 offsetof(struct virtio_net_config, status),
1697                                 &config->status, sizeof(config->status));
1698                 } else {
1699                         PMD_INIT_LOG(DEBUG,
1700                                      "VIRTIO_NET_F_STATUS is not supported");
1701                         config->status = 0;
1702                 }
1703
1704                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1705                         vtpci_read_dev_config(hw,
1706                                 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1707                                 &config->max_virtqueue_pairs,
1708                                 sizeof(config->max_virtqueue_pairs));
1709                 } else {
1710                         PMD_INIT_LOG(DEBUG,
1711                                      "VIRTIO_NET_F_MQ is not supported");
1712                         config->max_virtqueue_pairs = 1;
1713                 }
1714
1715                 hw->max_queue_pairs = config->max_virtqueue_pairs;
1716
1717                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1718                         vtpci_read_dev_config(hw,
1719                                 offsetof(struct virtio_net_config, mtu),
1720                                 &config->mtu,
1721                                 sizeof(config->mtu));
1722
1723                         /*
1724                          * MTU value has already been checked at negotiation
1725                          * time, but check again in case it has changed since
1726                          * then, which should not happen.
1727                          */
1728                         if (config->mtu < RTE_ETHER_MIN_MTU) {
1729                                 PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1730                                                 config->mtu);
1731                                 return -1;
1732                         }
1733
1734                         hw->max_mtu = config->mtu;
1735                         /* Set initial MTU to maximum one supported by vhost */
1736                         eth_dev->data->mtu = config->mtu;
1737
1738                 } else {
1739                         hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1740                                 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1741                 }
1742
1743                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1744                                 config->max_virtqueue_pairs);
1745                 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1746                 PMD_INIT_LOG(DEBUG,
1747                                 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1748                                 config->mac[0], config->mac[1],
1749                                 config->mac[2], config->mac[3],
1750                                 config->mac[4], config->mac[5]);
1751         } else {
1752                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1753                 hw->max_queue_pairs = 1;
1754                 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1755                         VLAN_TAG_LEN - hw->vtnet_hdr_size;
1756         }
1757
1758         ret = virtio_alloc_queues(eth_dev);
1759         if (ret < 0)
1760                 return ret;
1761
1762         if (eth_dev->data->dev_conf.intr_conf.rxq) {
1763                 if (virtio_configure_intr(eth_dev) < 0) {
1764                         PMD_INIT_LOG(ERR, "failed to configure interrupt");
1765                         virtio_free_queues(hw);
1766                         return -1;
1767                 }
1768         }
1769
1770         vtpci_reinit_complete(hw);
1771
1772         if (pci_dev)
1773                 PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1774                         eth_dev->data->port_id, pci_dev->id.vendor_id,
1775                         pci_dev->id.device_id);
1776
1777         return 0;
1778 }
1779
1780 /*
1781  * Remap the PCI device again (IO port map for legacy device and
1782  * memory map for modern device), so that the secondary process
1783  * could have the PCI initiated correctly.
1784  */
1785 static int
1786 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1787 {
1788         if (hw->modern) {
1789                 /*
1790                  * We don't have to re-parse the PCI config space, since
1791                  * rte_pci_map_device() makes sure the mapped address
1792                  * in secondary process would equal to the one mapped in
1793                  * the primary process: error will be returned if that
1794                  * requirement is not met.
1795                  *
1796                  * That said, we could simply reuse all cap pointers
1797                  * (such as dev_cfg, common_cfg, etc.) parsed from the
1798                  * primary process, which is stored in shared memory.
1799                  */
1800                 if (rte_pci_map_device(pci_dev)) {
1801                         PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1802                         return -1;
1803                 }
1804         } else {
1805                 if (rte_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1806                         return -1;
1807         }
1808
1809         return 0;
1810 }
1811
1812 static void
1813 virtio_set_vtpci_ops(struct virtio_hw *hw)
1814 {
1815 #ifdef RTE_VIRTIO_USER
1816         if (hw->virtio_user_dev)
1817                 VTPCI_OPS(hw) = &virtio_user_ops;
1818         else
1819 #endif
1820         if (hw->modern)
1821                 VTPCI_OPS(hw) = &modern_ops;
1822         else
1823                 VTPCI_OPS(hw) = &legacy_ops;
1824 }
1825
1826 /*
1827  * This function is based on probe() function in virtio_pci.c
1828  * It returns 0 on success.
1829  */
1830 int
1831 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1832 {
1833         struct virtio_hw *hw = eth_dev->data->dev_private;
1834         int ret;
1835
1836         RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr_mrg_rxbuf));
1837
1838         eth_dev->dev_ops = &virtio_eth_dev_ops;
1839
1840         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1841                 if (!hw->virtio_user_dev) {
1842                         ret = virtio_remap_pci(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1843                         if (ret)
1844                                 return ret;
1845                 }
1846
1847                 virtio_set_vtpci_ops(hw);
1848                 set_rxtx_funcs(eth_dev);
1849
1850                 return 0;
1851         }
1852
1853         /*
1854          * Pass the information to the rte_eth_dev_close() that it should also
1855          * release the private port resources.
1856          */
1857         eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
1858
1859         /* Allocate memory for storing MAC addresses */
1860         eth_dev->data->mac_addrs = rte_zmalloc("virtio",
1861                                 VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN, 0);
1862         if (eth_dev->data->mac_addrs == NULL) {
1863                 PMD_INIT_LOG(ERR,
1864                         "Failed to allocate %d bytes needed to store MAC addresses",
1865                         VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN);
1866                 return -ENOMEM;
1867         }
1868
1869         hw->port_id = eth_dev->data->port_id;
1870         /* For virtio_user case the hw->virtio_user_dev is populated by
1871          * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1872          */
1873         if (!hw->virtio_user_dev) {
1874                 ret = vtpci_init(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1875                 if (ret)
1876                         goto err_vtpci_init;
1877         }
1878
1879         /* reset device and negotiate default features */
1880         ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1881         if (ret < 0)
1882                 goto err_virtio_init;
1883
1884         hw->opened = true;
1885
1886         return 0;
1887
1888 err_virtio_init:
1889         if (!hw->virtio_user_dev) {
1890                 rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev));
1891                 if (!hw->modern)
1892                         rte_pci_ioport_unmap(VTPCI_IO(hw));
1893         }
1894 err_vtpci_init:
1895         rte_free(eth_dev->data->mac_addrs);
1896         eth_dev->data->mac_addrs = NULL;
1897         return ret;
1898 }
1899
1900 static int
1901 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1902 {
1903         PMD_INIT_FUNC_TRACE();
1904
1905         if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1906                 return 0;
1907
1908         virtio_dev_stop(eth_dev);
1909         virtio_dev_close(eth_dev);
1910
1911         eth_dev->dev_ops = NULL;
1912         eth_dev->tx_pkt_burst = NULL;
1913         eth_dev->rx_pkt_burst = NULL;
1914
1915         PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1916
1917         return 0;
1918 }
1919
1920 static int vdpa_check_handler(__rte_unused const char *key,
1921                 const char *value, __rte_unused void *opaque)
1922 {
1923         if (strcmp(value, "1"))
1924                 return -1;
1925
1926         return 0;
1927 }
1928
1929 static int
1930 vdpa_mode_selected(struct rte_devargs *devargs)
1931 {
1932         struct rte_kvargs *kvlist;
1933         const char *key = "vdpa";
1934         int ret = 0;
1935
1936         if (devargs == NULL)
1937                 return 0;
1938
1939         kvlist = rte_kvargs_parse(devargs->args, NULL);
1940         if (kvlist == NULL)
1941                 return 0;
1942
1943         if (!rte_kvargs_count(kvlist, key))
1944                 goto exit;
1945
1946         /* vdpa mode selected when there's a key-value pair: vdpa=1 */
1947         if (rte_kvargs_process(kvlist, key,
1948                                 vdpa_check_handler, NULL) < 0) {
1949                 goto exit;
1950         }
1951         ret = 1;
1952
1953 exit:
1954         rte_kvargs_free(kvlist);
1955         return ret;
1956 }
1957
1958 static int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1959         struct rte_pci_device *pci_dev)
1960 {
1961         if (rte_eal_iopl_init() != 0) {
1962                 PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1963                 return 1;
1964         }
1965
1966         /* virtio pmd skips probe if device needs to work in vdpa mode */
1967         if (vdpa_mode_selected(pci_dev->device.devargs))
1968                 return 1;
1969
1970         return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw),
1971                 eth_virtio_dev_init);
1972 }
1973
1974 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev)
1975 {
1976         int ret;
1977
1978         ret = rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
1979         /* Port has already been released by close. */
1980         if (ret == -ENODEV)
1981                 ret = 0;
1982         return ret;
1983 }
1984
1985 static struct rte_pci_driver rte_virtio_pmd = {
1986         .driver = {
1987                 .name = "net_virtio",
1988         },
1989         .id_table = pci_id_virtio_map,
1990         .drv_flags = 0,
1991         .probe = eth_virtio_pci_probe,
1992         .remove = eth_virtio_pci_remove,
1993 };
1994
1995 RTE_INIT(rte_virtio_pmd_init)
1996 {
1997         rte_eal_iopl_init();
1998         rte_pci_register(&rte_virtio_pmd);
1999 }
2000
2001 static bool
2002 rx_offload_enabled(struct virtio_hw *hw)
2003 {
2004         return vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM) ||
2005                 vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2006                 vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6);
2007 }
2008
2009 static bool
2010 tx_offload_enabled(struct virtio_hw *hw)
2011 {
2012         return vtpci_with_feature(hw, VIRTIO_NET_F_CSUM) ||
2013                 vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO4) ||
2014                 vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO6);
2015 }
2016
2017 /*
2018  * Configure virtio device
2019  * It returns 0 on success.
2020  */
2021 static int
2022 virtio_dev_configure(struct rte_eth_dev *dev)
2023 {
2024         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2025         const struct rte_eth_txmode *txmode = &dev->data->dev_conf.txmode;
2026         struct virtio_hw *hw = dev->data->dev_private;
2027         uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
2028                 hw->vtnet_hdr_size;
2029         uint64_t rx_offloads = rxmode->offloads;
2030         uint64_t tx_offloads = txmode->offloads;
2031         uint64_t req_features;
2032         int ret;
2033
2034         PMD_INIT_LOG(DEBUG, "configure");
2035         req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
2036
2037         if (dev->data->dev_conf.intr_conf.rxq) {
2038                 ret = virtio_init_device(dev, hw->req_guest_features);
2039                 if (ret < 0)
2040                         return ret;
2041         }
2042
2043         if (rxmode->max_rx_pkt_len > hw->max_mtu + ether_hdr_len)
2044                 req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
2045
2046         if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2047                            DEV_RX_OFFLOAD_TCP_CKSUM))
2048                 req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
2049
2050         if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
2051                 req_features |=
2052                         (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2053                         (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2054
2055         if (tx_offloads & (DEV_TX_OFFLOAD_UDP_CKSUM |
2056                            DEV_TX_OFFLOAD_TCP_CKSUM))
2057                 req_features |= (1ULL << VIRTIO_NET_F_CSUM);
2058
2059         if (tx_offloads & DEV_TX_OFFLOAD_TCP_TSO)
2060                 req_features |=
2061                         (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2062                         (1ULL << VIRTIO_NET_F_HOST_TSO6);
2063
2064         /* if request features changed, reinit the device */
2065         if (req_features != hw->req_guest_features) {
2066                 ret = virtio_init_device(dev, req_features);
2067                 if (ret < 0)
2068                         return ret;
2069         }
2070
2071         if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2072                             DEV_RX_OFFLOAD_TCP_CKSUM)) &&
2073                 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
2074                 PMD_DRV_LOG(ERR,
2075                         "rx checksum not available on this host");
2076                 return -ENOTSUP;
2077         }
2078
2079         if ((rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) &&
2080                 (!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2081                  !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
2082                 PMD_DRV_LOG(ERR,
2083                         "Large Receive Offload not available on this host");
2084                 return -ENOTSUP;
2085         }
2086
2087         /* start control queue */
2088         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
2089                 virtio_dev_cq_start(dev);
2090
2091         if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2092                 hw->vlan_strip = 1;
2093
2094         if ((rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
2095             && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2096                 PMD_DRV_LOG(ERR,
2097                             "vlan filtering not available on this host");
2098                 return -ENOTSUP;
2099         }
2100
2101         hw->has_tx_offload = tx_offload_enabled(hw);
2102         hw->has_rx_offload = rx_offload_enabled(hw);
2103
2104         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2105                 /* Enable vector (0) for Link State Intrerrupt */
2106                 if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
2107                                 VIRTIO_MSI_NO_VECTOR) {
2108                         PMD_DRV_LOG(ERR, "failed to set config vector");
2109                         return -EBUSY;
2110                 }
2111
2112         rte_spinlock_init(&hw->state_lock);
2113
2114         hw->use_simple_rx = 1;
2115
2116         if (vtpci_with_feature(hw, VIRTIO_F_IN_ORDER)) {
2117                 hw->use_inorder_tx = 1;
2118                 hw->use_inorder_rx = 1;
2119                 hw->use_simple_rx = 0;
2120         }
2121
2122         if (vtpci_packed_queue(hw)) {
2123                 hw->use_simple_rx = 0;
2124                 hw->use_inorder_rx = 0;
2125         }
2126
2127 #if defined RTE_ARCH_ARM64 || defined RTE_ARCH_ARM
2128         if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
2129                 hw->use_simple_rx = 0;
2130         }
2131 #endif
2132         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2133                  hw->use_simple_rx = 0;
2134         }
2135
2136         if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2137                            DEV_RX_OFFLOAD_TCP_CKSUM |
2138                            DEV_RX_OFFLOAD_TCP_LRO |
2139                            DEV_RX_OFFLOAD_VLAN_STRIP))
2140                 hw->use_simple_rx = 0;
2141
2142         return 0;
2143 }
2144
2145
2146 static int
2147 virtio_dev_start(struct rte_eth_dev *dev)
2148 {
2149         uint16_t nb_queues, i;
2150         struct virtnet_rx *rxvq;
2151         struct virtnet_tx *txvq __rte_unused;
2152         struct virtio_hw *hw = dev->data->dev_private;
2153         int ret;
2154
2155         /* Finish the initialization of the queues */
2156         for (i = 0; i < dev->data->nb_rx_queues; i++) {
2157                 ret = virtio_dev_rx_queue_setup_finish(dev, i);
2158                 if (ret < 0)
2159                         return ret;
2160         }
2161         for (i = 0; i < dev->data->nb_tx_queues; i++) {
2162                 ret = virtio_dev_tx_queue_setup_finish(dev, i);
2163                 if (ret < 0)
2164                         return ret;
2165         }
2166
2167         /* check if lsc interrupt feature is enabled */
2168         if (dev->data->dev_conf.intr_conf.lsc) {
2169                 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
2170                         PMD_DRV_LOG(ERR, "link status not supported by host");
2171                         return -ENOTSUP;
2172                 }
2173         }
2174
2175         /* Enable uio/vfio intr/eventfd mapping: althrough we already did that
2176          * in device configure, but it could be unmapped  when device is
2177          * stopped.
2178          */
2179         if (dev->data->dev_conf.intr_conf.lsc ||
2180             dev->data->dev_conf.intr_conf.rxq) {
2181                 virtio_intr_disable(dev);
2182
2183                 /* Setup interrupt callback  */
2184                 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2185                         rte_intr_callback_register(dev->intr_handle,
2186                                                    virtio_interrupt_handler,
2187                                                    dev);
2188
2189                 if (virtio_intr_enable(dev) < 0) {
2190                         PMD_DRV_LOG(ERR, "interrupt enable failed");
2191                         return -EIO;
2192                 }
2193         }
2194
2195         /*Notify the backend
2196          *Otherwise the tap backend might already stop its queue due to fullness.
2197          *vhost backend will have no chance to be waked up
2198          */
2199         nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
2200         if (hw->max_queue_pairs > 1) {
2201                 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
2202                         return -EINVAL;
2203         }
2204
2205         PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
2206
2207         for (i = 0; i < dev->data->nb_rx_queues; i++) {
2208                 rxvq = dev->data->rx_queues[i];
2209                 /* Flush the old packets */
2210                 virtqueue_rxvq_flush(rxvq->vq);
2211                 virtqueue_notify(rxvq->vq);
2212         }
2213
2214         for (i = 0; i < dev->data->nb_tx_queues; i++) {
2215                 txvq = dev->data->tx_queues[i];
2216                 virtqueue_notify(txvq->vq);
2217         }
2218
2219         PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
2220
2221         for (i = 0; i < dev->data->nb_rx_queues; i++) {
2222                 rxvq = dev->data->rx_queues[i];
2223                 VIRTQUEUE_DUMP(rxvq->vq);
2224         }
2225
2226         for (i = 0; i < dev->data->nb_tx_queues; i++) {
2227                 txvq = dev->data->tx_queues[i];
2228                 VIRTQUEUE_DUMP(txvq->vq);
2229         }
2230
2231         set_rxtx_funcs(dev);
2232         hw->started = true;
2233
2234         /* Initialize Link state */
2235         virtio_dev_link_update(dev, 0);
2236
2237         return 0;
2238 }
2239
2240 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
2241 {
2242         struct virtio_hw *hw = dev->data->dev_private;
2243         uint16_t nr_vq = virtio_get_nr_vq(hw);
2244         const char *type __rte_unused;
2245         unsigned int i, mbuf_num = 0;
2246         struct virtqueue *vq;
2247         struct rte_mbuf *buf;
2248         int queue_type;
2249
2250         if (hw->vqs == NULL)
2251                 return;
2252
2253         for (i = 0; i < nr_vq; i++) {
2254                 vq = hw->vqs[i];
2255                 if (!vq)
2256                         continue;
2257
2258                 queue_type = virtio_get_queue_type(hw, i);
2259                 if (queue_type == VTNET_RQ)
2260                         type = "rxq";
2261                 else if (queue_type == VTNET_TQ)
2262                         type = "txq";
2263                 else
2264                         continue;
2265
2266                 PMD_INIT_LOG(DEBUG,
2267                         "Before freeing %s[%d] used and unused buf",
2268                         type, i);
2269                 VIRTQUEUE_DUMP(vq);
2270
2271                 while ((buf = virtqueue_detach_unused(vq)) != NULL) {
2272                         rte_pktmbuf_free(buf);
2273                         mbuf_num++;
2274                 }
2275
2276                 PMD_INIT_LOG(DEBUG,
2277                         "After freeing %s[%d] used and unused buf",
2278                         type, i);
2279                 VIRTQUEUE_DUMP(vq);
2280         }
2281
2282         PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
2283 }
2284
2285 /*
2286  * Stop device: disable interrupt and mark link down
2287  */
2288 static void
2289 virtio_dev_stop(struct rte_eth_dev *dev)
2290 {
2291         struct virtio_hw *hw = dev->data->dev_private;
2292         struct rte_eth_link link;
2293         struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
2294
2295         PMD_INIT_LOG(DEBUG, "stop");
2296
2297         rte_spinlock_lock(&hw->state_lock);
2298         if (!hw->started)
2299                 goto out_unlock;
2300         hw->started = false;
2301
2302         if (intr_conf->lsc || intr_conf->rxq) {
2303                 virtio_intr_disable(dev);
2304
2305                 /* Reset interrupt callback  */
2306                 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
2307                         rte_intr_callback_unregister(dev->intr_handle,
2308                                                      virtio_interrupt_handler,
2309                                                      dev);
2310                 }
2311         }
2312
2313         memset(&link, 0, sizeof(link));
2314         rte_eth_linkstatus_set(dev, &link);
2315 out_unlock:
2316         rte_spinlock_unlock(&hw->state_lock);
2317 }
2318
2319 static int
2320 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2321 {
2322         struct rte_eth_link link;
2323         uint16_t status;
2324         struct virtio_hw *hw = dev->data->dev_private;
2325
2326         memset(&link, 0, sizeof(link));
2327         link.link_duplex = ETH_LINK_FULL_DUPLEX;
2328         link.link_speed  = ETH_SPEED_NUM_10G;
2329         link.link_autoneg = ETH_LINK_FIXED;
2330
2331         if (!hw->started) {
2332                 link.link_status = ETH_LINK_DOWN;
2333         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2334                 PMD_INIT_LOG(DEBUG, "Get link status from hw");
2335                 vtpci_read_dev_config(hw,
2336                                 offsetof(struct virtio_net_config, status),
2337                                 &status, sizeof(status));
2338                 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2339                         link.link_status = ETH_LINK_DOWN;
2340                         PMD_INIT_LOG(DEBUG, "Port %d is down",
2341                                      dev->data->port_id);
2342                 } else {
2343                         link.link_status = ETH_LINK_UP;
2344                         PMD_INIT_LOG(DEBUG, "Port %d is up",
2345                                      dev->data->port_id);
2346                 }
2347         } else {
2348                 link.link_status = ETH_LINK_UP;
2349         }
2350
2351         return rte_eth_linkstatus_set(dev, &link);
2352 }
2353
2354 static int
2355 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2356 {
2357         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2358         struct virtio_hw *hw = dev->data->dev_private;
2359         uint64_t offloads = rxmode->offloads;
2360
2361         if (mask & ETH_VLAN_FILTER_MASK) {
2362                 if ((offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2363                                 !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2364
2365                         PMD_DRV_LOG(NOTICE,
2366                                 "vlan filtering not available on this host");
2367
2368                         return -ENOTSUP;
2369                 }
2370         }
2371
2372         if (mask & ETH_VLAN_STRIP_MASK)
2373                 hw->vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
2374
2375         return 0;
2376 }
2377
2378 static void
2379 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2380 {
2381         uint64_t tso_mask, host_features;
2382         struct virtio_hw *hw = dev->data->dev_private;
2383
2384         dev_info->speed_capa = ETH_LINK_SPEED_10G; /* fake value */
2385
2386         dev_info->max_rx_queues =
2387                 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2388         dev_info->max_tx_queues =
2389                 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2390         dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2391         dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2392         dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2393
2394         host_features = VTPCI_OPS(hw)->get_features(hw);
2395         dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
2396         dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2397         if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2398                 dev_info->rx_offload_capa |=
2399                         DEV_RX_OFFLOAD_TCP_CKSUM |
2400                         DEV_RX_OFFLOAD_UDP_CKSUM;
2401         }
2402         if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN))
2403                 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_FILTER;
2404         tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2405                 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2406         if ((host_features & tso_mask) == tso_mask)
2407                 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2408
2409         dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
2410                                     DEV_TX_OFFLOAD_VLAN_INSERT;
2411         if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2412                 dev_info->tx_offload_capa |=
2413                         DEV_TX_OFFLOAD_UDP_CKSUM |
2414                         DEV_TX_OFFLOAD_TCP_CKSUM;
2415         }
2416         tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2417                 (1ULL << VIRTIO_NET_F_HOST_TSO6);
2418         if ((host_features & tso_mask) == tso_mask)
2419                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2420 }
2421
2422 /*
2423  * It enables testpmd to collect per queue stats.
2424  */
2425 static int
2426 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2427 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2428 __rte_unused uint8_t is_rx)
2429 {
2430         return 0;
2431 }
2432
2433 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
2434 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
2435 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio-pci");
2436
2437 RTE_INIT(virtio_init_log)
2438 {
2439         virtio_logtype_init = rte_log_register("pmd.net.virtio.init");
2440         if (virtio_logtype_init >= 0)
2441                 rte_log_set_level(virtio_logtype_init, RTE_LOG_NOTICE);
2442         virtio_logtype_driver = rte_log_register("pmd.net.virtio.driver");
2443         if (virtio_logtype_driver >= 0)
2444                 rte_log_set_level(virtio_logtype_driver, RTE_LOG_NOTICE);
2445 }