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