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_disable(struct rte_eth_dev *dev);
  75
  76 static int virtio_dev_queue_stats_mapping_set(
  77         struct rte_eth_dev *eth_dev,
  78         uint16_t queue_id,
  79         uint8_t stat_idx,
  80         uint8_t is_rx);
  81
  82 int virtio_logtype_init;
  83 int virtio_logtype_driver;
  84
  85 static void virtio_notify_peers(struct rte_eth_dev *dev);
  86 static void virtio_ack_link_announce(struct rte_eth_dev *dev);
  87
  88 /*
  89  * The set of PCI devices this driver supports
  90  */
  91 static const struct rte_pci_id pci_id_virtio_map[] = {
  92         { RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_LEGACY_DEVICEID_NET) },
  93         { RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_MODERN_DEVICEID_NET) },
  94         { .vendor_id = 0, /* sentinel */ },
  95 };
  96
  97 struct rte_virtio_xstats_name_off {
  98         char name[RTE_ETH_XSTATS_NAME_SIZE];
  99         unsigned offset;
 100 };
 101
 102 /* [rt]x_qX_ is prepended to the name string here */
 103 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
 104         {"good_packets",           offsetof(struct virtnet_rx, stats.packets)},
 105         {"good_bytes",             offsetof(struct virtnet_rx, stats.bytes)},
 106         {"errors",                 offsetof(struct virtnet_rx, stats.errors)},
 107         {"multicast_packets",      offsetof(struct virtnet_rx, stats.multicast)},
 108         {"broadcast_packets",      offsetof(struct virtnet_rx, stats.broadcast)},
 109         {"undersize_packets",      offsetof(struct virtnet_rx, stats.size_bins[0])},
 110         {"size_64_packets",        offsetof(struct virtnet_rx, stats.size_bins[1])},
 111         {"size_65_127_packets",    offsetof(struct virtnet_rx, stats.size_bins[2])},
 112         {"size_128_255_packets",   offsetof(struct virtnet_rx, stats.size_bins[3])},
 113         {"size_256_511_packets",   offsetof(struct virtnet_rx, stats.size_bins[4])},
 114         {"size_512_1023_packets",  offsetof(struct virtnet_rx, stats.size_bins[5])},
 115         {"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
 116         {"size_1519_max_packets",  offsetof(struct virtnet_rx, stats.size_bins[7])},
 117 };
 118
 119 /* [rt]x_qX_ is prepended to the name string here */
 120 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
 121         {"good_packets",           offsetof(struct virtnet_tx, stats.packets)},
 122         {"good_bytes",             offsetof(struct virtnet_tx, stats.bytes)},
 123         {"errors",                 offsetof(struct virtnet_tx, stats.errors)},
 124         {"multicast_packets",      offsetof(struct virtnet_tx, stats.multicast)},
 125         {"broadcast_packets",      offsetof(struct virtnet_tx, stats.broadcast)},
 126         {"undersize_packets",      offsetof(struct virtnet_tx, stats.size_bins[0])},
 127         {"size_64_packets",        offsetof(struct virtnet_tx, stats.size_bins[1])},
 128         {"size_65_127_packets",    offsetof(struct virtnet_tx, stats.size_bins[2])},
 129         {"size_128_255_packets",   offsetof(struct virtnet_tx, stats.size_bins[3])},
 130         {"size_256_511_packets",   offsetof(struct virtnet_tx, stats.size_bins[4])},
 131         {"size_512_1023_packets",  offsetof(struct virtnet_tx, stats.size_bins[5])},
 132         {"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
 133         {"size_1519_max_packets",  offsetof(struct virtnet_tx, stats.size_bins[7])},
 134 };
 135
 136 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
 137                             sizeof(rte_virtio_rxq_stat_strings[0]))
 138 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
 139                             sizeof(rte_virtio_txq_stat_strings[0]))
 140
 141 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
 142
 143 static struct virtio_pmd_ctrl *
 144 virtio_send_command_packed(struct virtnet_ctl *cvq,
 145                            struct virtio_pmd_ctrl *ctrl,
 146                            int *dlen, int pkt_num)
 147 {
 148         struct virtqueue *vq = cvq->vq;
 149         int head;
 150         struct vring_packed_desc *desc = vq->vq_packed.ring.desc;
 151         struct virtio_pmd_ctrl *result;
 152         uint16_t flags;
 153         int sum = 0;
 154         int nb_descs = 0;
 155         int k;
 156
 157         /*
 158          * Format is enforced in qemu code:
 159          * One TX packet for header;
 160          * At least one TX packet per argument;
 161          * One RX packet for ACK.
 162          */
 163         head = vq->vq_avail_idx;
 164         flags = vq->vq_packed.cached_flags;
 165         desc[head].addr = cvq->virtio_net_hdr_mem;
 166         desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
 167         vq->vq_free_cnt--;
 168         nb_descs++;
 169         if (++vq->vq_avail_idx >= vq->vq_nentries) {
 170                 vq->vq_avail_idx -= vq->vq_nentries;
 171                 vq->vq_packed.cached_flags ^=
 172                         VRING_DESC_F_AVAIL(1) | VRING_DESC_F_USED(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                         vq->vq_packed.cached_flags;
 182                 sum += dlen[k];
 183                 vq->vq_free_cnt--;
 184                 nb_descs++;
 185                 if (++vq->vq_avail_idx >= vq->vq_nentries) {
 186                         vq->vq_avail_idx -= vq->vq_nentries;
 187                         vq->vq_packed.cached_flags ^=
 188                                 VRING_DESC_F_AVAIL(1) | VRING_DESC_F_USED(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                 vq->vq_packed.cached_flags;
 197         vq->vq_free_cnt--;
 198         nb_descs++;
 199         if (++vq->vq_avail_idx >= vq->vq_nentries) {
 200                 vq->vq_avail_idx -= vq->vq_nentries;
 201                 vq->vq_packed.cached_flags ^=
 202                         VRING_DESC_F_AVAIL(1) | VRING_DESC_F_USED(1);
 203         }
 204
 205         virtio_wmb(vq->hw->weak_barriers);
 206         desc[head].flags = VRING_DESC_F_NEXT | flags;
 207
 208         virtio_wmb(vq->hw->weak_barriers);
 209         virtqueue_notify(vq);
 210
 211         /* wait for used descriptors in virtqueue */
 212         while (!desc_is_used(&desc[head], vq))
 213                 usleep(100);
 214
 215         virtio_rmb(vq->hw->weak_barriers);
 216
 217         /* now get used descriptors */
 218         vq->vq_free_cnt += nb_descs;
 219         vq->vq_used_cons_idx += nb_descs;
 220         if (vq->vq_used_cons_idx >= vq->vq_nentries) {
 221                 vq->vq_used_cons_idx -= vq->vq_nentries;
 222                 vq->vq_packed.used_wrap_counter ^= 1;
 223         }
 224
 225         PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\n"
 226                         "vq->vq_avail_idx=%d\n"
 227                         "vq->vq_used_cons_idx=%d\n"
 228                         "vq->vq_packed.cached_flags=0x%x\n"
 229                         "vq->vq_packed.used_wrap_counter=%d\n",
 230                         vq->vq_free_cnt,
 231                         vq->vq_avail_idx,
 232                         vq->vq_used_cons_idx,
 233                         vq->vq_packed.cached_flags,
 234                         vq->vq_packed.used_wrap_counter);
 235
 236         result = cvq->virtio_net_hdr_mz->addr;
 237         return result;
 238 }
 239
 240 static int
 241 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
 242                 int *dlen, int pkt_num)
 243 {
 244         uint32_t head, i;
 245         int k, sum = 0;
 246         virtio_net_ctrl_ack status = ~0;
 247         struct virtio_pmd_ctrl *result;
 248         struct virtqueue *vq;
 249
 250         ctrl->status = status;
 251
 252         if (!cvq || !cvq->vq) {
 253                 PMD_INIT_LOG(ERR, "Control queue is not supported.");
 254                 return -1;
 255         }
 256
 257         rte_spinlock_lock(&cvq->lock);
 258         vq = cvq->vq;
 259         head = vq->vq_desc_head_idx;
 260
 261         PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
 262                 "vq->hw->cvq = %p vq = %p",
 263                 vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
 264
 265         if (vq->vq_free_cnt < pkt_num + 2 || pkt_num < 1) {
 266                 rte_spinlock_unlock(&cvq->lock);
 267                 return -1;
 268         }
 269
 270         memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
 271                 sizeof(struct virtio_pmd_ctrl));
 272
 273         if (vtpci_packed_queue(vq->hw)) {
 274                 result = virtio_send_command_packed(cvq, ctrl, dlen, pkt_num);
 275                 goto out_unlock;
 276         }
 277
 278         /*
 279          * Format is enforced in qemu code:
 280          * One TX packet for header;
 281          * At least one TX packet per argument;
 282          * One RX packet for ACK.
 283          */
 284         vq->vq_split.ring.desc[head].flags = VRING_DESC_F_NEXT;
 285         vq->vq_split.ring.desc[head].addr = cvq->virtio_net_hdr_mem;
 286         vq->vq_split.ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
 287         vq->vq_free_cnt--;
 288         i = vq->vq_split.ring.desc[head].next;
 289
 290         for (k = 0; k < pkt_num; k++) {
 291                 vq->vq_split.ring.desc[i].flags = VRING_DESC_F_NEXT;
 292                 vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
 293                         + sizeof(struct virtio_net_ctrl_hdr)
 294                         + sizeof(ctrl->status) + sizeof(uint8_t)*sum;
 295                 vq->vq_split.ring.desc[i].len = dlen[k];
 296                 sum += dlen[k];
 297                 vq->vq_free_cnt--;
 298                 i = vq->vq_split.ring.desc[i].next;
 299         }
 300
 301         vq->vq_split.ring.desc[i].flags = VRING_DESC_F_WRITE;
 302         vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
 303                         + sizeof(struct virtio_net_ctrl_hdr);
 304         vq->vq_split.ring.desc[i].len = sizeof(ctrl->status);
 305         vq->vq_free_cnt--;
 306
 307         vq->vq_desc_head_idx = vq->vq_split.ring.desc[i].next;
 308
 309         vq_update_avail_ring(vq, head);
 310         vq_update_avail_idx(vq);
 311
 312         PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
 313
 314         virtqueue_notify(vq);
 315
 316         rte_rmb();
 317         while (VIRTQUEUE_NUSED(vq) == 0) {
 318                 rte_rmb();
 319                 usleep(100);
 320         }
 321
 322         while (VIRTQUEUE_NUSED(vq)) {
 323                 uint32_t idx, desc_idx, used_idx;
 324                 struct vring_used_elem *uep;
 325
 326                 used_idx = (uint32_t)(vq->vq_used_cons_idx
 327                                 & (vq->vq_nentries - 1));
 328                 uep = &vq->vq_split.ring.used->ring[used_idx];
 329                 idx = (uint32_t) uep->id;
 330                 desc_idx = idx;
 331
 332                 while (vq->vq_split.ring.desc[desc_idx].flags &
 333                                 VRING_DESC_F_NEXT) {
 334                         desc_idx = vq->vq_split.ring.desc[desc_idx].next;
 335                         vq->vq_free_cnt++;
 336                 }
 337
 338                 vq->vq_split.ring.desc[desc_idx].next = vq->vq_desc_head_idx;
 339                 vq->vq_desc_head_idx = idx;
 340
 341                 vq->vq_used_cons_idx++;
 342                 vq->vq_free_cnt++;
 343         }
 344
 345         PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
 346                         vq->vq_free_cnt, vq->vq_desc_head_idx);
 347
 348         result = cvq->virtio_net_hdr_mz->addr;
 349
 350 out_unlock:
 351         rte_spinlock_unlock(&cvq->lock);
 352         return result->status;
 353 }
 354
 355 static int
 356 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
 357 {
 358         struct virtio_hw *hw = dev->data->dev_private;
 359         struct virtio_pmd_ctrl ctrl;
 360         int dlen[1];
 361         int ret;
 362
 363         ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
 364         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
 365         memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
 366
 367         dlen[0] = sizeof(uint16_t);
 368
 369         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 370         if (ret) {
 371                 PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
 372                           "failed, this is too late now...");
 373                 return -EINVAL;
 374         }
 375
 376         return 0;
 377 }
 378
 379 static void
 380 virtio_dev_queue_release(void *queue __rte_unused)
 381 {
 382         /* do nothing */
 383 }
 384
 385 static uint16_t
 386 virtio_get_nr_vq(struct virtio_hw *hw)
 387 {
 388         uint16_t nr_vq = hw->max_queue_pairs * 2;
 389
 390         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
 391                 nr_vq += 1;
 392
 393         return nr_vq;
 394 }
 395
 396 static void
 397 virtio_init_vring(struct virtqueue *vq)
 398 {
 399         int size = vq->vq_nentries;
 400         uint8_t *ring_mem = vq->vq_ring_virt_mem;
 401
 402         PMD_INIT_FUNC_TRACE();
 403
 404         memset(ring_mem, 0, vq->vq_ring_size);
 405
 406         vq->vq_used_cons_idx = 0;
 407         vq->vq_desc_head_idx = 0;
 408         vq->vq_avail_idx = 0;
 409         vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
 410         vq->vq_free_cnt = vq->vq_nentries;
 411         memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
 412         if (vtpci_packed_queue(vq->hw)) {
 413                 vring_init_packed(&vq->vq_packed.ring, ring_mem,
 414                                   VIRTIO_PCI_VRING_ALIGN, size);
 415                 vring_desc_init_packed(vq, size);
 416         } else {
 417                 struct vring *vr = &vq->vq_split.ring;
 418
 419                 vring_init_split(vr, ring_mem, VIRTIO_PCI_VRING_ALIGN, size);
 420                 vring_desc_init_split(vr->desc, size);
 421         }
 422         /*
 423          * Disable device(host) interrupting guest
 424          */
 425         virtqueue_disable_intr(vq);
 426 }
 427
 428 static int
 429 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
 430 {
 431         char vq_name[VIRTQUEUE_MAX_NAME_SZ];
 432         char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
 433         const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
 434         unsigned int vq_size, size;
 435         struct virtio_hw *hw = dev->data->dev_private;
 436         struct virtnet_rx *rxvq = NULL;
 437         struct virtnet_tx *txvq = NULL;
 438         struct virtnet_ctl *cvq = NULL;
 439         struct virtqueue *vq;
 440         size_t sz_hdr_mz = 0;
 441         void *sw_ring = NULL;
 442         int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
 443         int ret;
 444         int numa_node = dev->device->numa_node;
 445
 446         PMD_INIT_LOG(INFO, "setting up queue: %u on NUMA node %d",
 447                         vtpci_queue_idx, numa_node);
 448
 449         /*
 450          * Read the virtqueue size from the Queue Size field
 451          * Always power of 2 and if 0 virtqueue does not exist
 452          */
 453         vq_size = VTPCI_OPS(hw)->get_queue_num(hw, vtpci_queue_idx);
 454         PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
 455         if (vq_size == 0) {
 456                 PMD_INIT_LOG(ERR, "virtqueue does not exist");
 457                 return -EINVAL;
 458         }
 459
 460         if (!rte_is_power_of_2(vq_size)) {
 461                 PMD_INIT_LOG(ERR, "virtqueue size is not powerof 2");
 462                 return -EINVAL;
 463         }
 464
 465         snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
 466                  dev->data->port_id, vtpci_queue_idx);
 467
 468         size = RTE_ALIGN_CEIL(sizeof(*vq) +
 469                                 vq_size * sizeof(struct vq_desc_extra),
 470                                 RTE_CACHE_LINE_SIZE);
 471         if (queue_type == VTNET_TQ) {
 472                 /*
 473                  * For each xmit packet, allocate a virtio_net_hdr
 474                  * and indirect ring elements
 475                  */
 476                 sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
 477         } else if (queue_type == VTNET_CQ) {
 478                 /* Allocate a page for control vq command, data and status */
 479                 sz_hdr_mz = PAGE_SIZE;
 480         }
 481
 482         vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
 483                                 numa_node);
 484         if (vq == NULL) {
 485                 PMD_INIT_LOG(ERR, "can not allocate vq");
 486                 return -ENOMEM;
 487         }
 488         hw->vqs[vtpci_queue_idx] = vq;
 489
 490         vq->hw = hw;
 491         vq->vq_queue_index = vtpci_queue_idx;
 492         vq->vq_nentries = vq_size;
 493         if (vtpci_packed_queue(hw)) {
 494                 vq->vq_packed.used_wrap_counter = 1;
 495                 vq->vq_packed.cached_flags = VRING_DESC_F_AVAIL(1);
 496                 vq->vq_packed.event_flags_shadow = 0;
 497                 if (queue_type == VTNET_RQ)
 498                         vq->vq_packed.cached_flags |= VRING_DESC_F_WRITE;
 499         }
 500
 501         /*
 502          * Reserve a memzone for vring elements
 503          */
 504         size = vring_size(hw, vq_size, VIRTIO_PCI_VRING_ALIGN);
 505         vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
 506         PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
 507                      size, vq->vq_ring_size);
 508
 509         mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
 510                         numa_node, RTE_MEMZONE_IOVA_CONTIG,
 511                         VIRTIO_PCI_VRING_ALIGN);
 512         if (mz == NULL) {
 513                 if (rte_errno == EEXIST)
 514                         mz = rte_memzone_lookup(vq_name);
 515                 if (mz == NULL) {
 516                         ret = -ENOMEM;
 517                         goto fail_q_alloc;
 518                 }
 519         }
 520
 521         memset(mz->addr, 0, mz->len);
 522
 523         vq->vq_ring_mem = mz->iova;
 524         vq->vq_ring_virt_mem = mz->addr;
 525         PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%" PRIx64,
 526                      (uint64_t)mz->iova);
 527         PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
 528                      (uint64_t)(uintptr_t)mz->addr);
 529
 530         virtio_init_vring(vq);
 531
 532         if (sz_hdr_mz) {
 533                 snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
 534                          dev->data->port_id, vtpci_queue_idx);
 535                 hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
 536                                 numa_node, RTE_MEMZONE_IOVA_CONTIG,
 537                                 RTE_CACHE_LINE_SIZE);
 538                 if (hdr_mz == NULL) {
 539                         if (rte_errno == EEXIST)
 540                                 hdr_mz = rte_memzone_lookup(vq_hdr_name);
 541                         if (hdr_mz == NULL) {
 542                                 ret = -ENOMEM;
 543                                 goto fail_q_alloc;
 544                         }
 545                 }
 546         }
 547
 548         if (queue_type == VTNET_RQ) {
 549                 size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
 550                                sizeof(vq->sw_ring[0]);
 551
 552                 sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
 553                                 RTE_CACHE_LINE_SIZE, numa_node);
 554                 if (!sw_ring) {
 555                         PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
 556                         ret = -ENOMEM;
 557                         goto fail_q_alloc;
 558                 }
 559
 560                 vq->sw_ring = sw_ring;
 561                 rxvq = &vq->rxq;
 562                 rxvq->vq = vq;
 563                 rxvq->port_id = dev->data->port_id;
 564                 rxvq->mz = mz;
 565         } else if (queue_type == VTNET_TQ) {
 566                 txvq = &vq->txq;
 567                 txvq->vq = vq;
 568                 txvq->port_id = dev->data->port_id;
 569                 txvq->mz = mz;
 570                 txvq->virtio_net_hdr_mz = hdr_mz;
 571                 txvq->virtio_net_hdr_mem = hdr_mz->iova;
 572         } else if (queue_type == VTNET_CQ) {
 573                 cvq = &vq->cq;
 574                 cvq->vq = vq;
 575                 cvq->mz = mz;
 576                 cvq->virtio_net_hdr_mz = hdr_mz;
 577                 cvq->virtio_net_hdr_mem = hdr_mz->iova;
 578                 memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
 579
 580                 hw->cvq = cvq;
 581         }
 582
 583         /* For virtio_user case (that is when hw->dev is NULL), we use
 584          * virtual address. And we need properly set _offset_, please see
 585          * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
 586          */
 587         if (!hw->virtio_user_dev)
 588                 vq->offset = offsetof(struct rte_mbuf, buf_iova);
 589         else {
 590                 vq->vq_ring_mem = (uintptr_t)mz->addr;
 591                 vq->offset = offsetof(struct rte_mbuf, buf_addr);
 592                 if (queue_type == VTNET_TQ)
 593                         txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
 594                 else if (queue_type == VTNET_CQ)
 595                         cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
 596         }
 597
 598         if (queue_type == VTNET_TQ) {
 599                 struct virtio_tx_region *txr;
 600                 unsigned int i;
 601
 602                 txr = hdr_mz->addr;
 603                 memset(txr, 0, vq_size * sizeof(*txr));
 604                 for (i = 0; i < vq_size; i++) {
 605                         struct vring_desc *start_dp = txr[i].tx_indir;
 606                         struct vring_packed_desc *start_dp_packed =
 607                                 txr[i].tx_indir_pq;
 608
 609                         /* first indirect descriptor is always the tx header */
 610                         if (vtpci_packed_queue(hw)) {
 611                                 start_dp_packed->addr = txvq->virtio_net_hdr_mem
 612                                         + i * sizeof(*txr)
 613                                         + offsetof(struct virtio_tx_region,
 614                                                    tx_hdr);
 615                                 start_dp_packed->len = hw->vtnet_hdr_size;
 616                         } else {
 617                                 vring_desc_init_split(start_dp,
 618                                                       RTE_DIM(txr[i].tx_indir));
 619                                 start_dp->addr = txvq->virtio_net_hdr_mem
 620                                         + i * sizeof(*txr)
 621                                         + offsetof(struct virtio_tx_region,
 622                                                    tx_hdr);
 623                                 start_dp->len = hw->vtnet_hdr_size;
 624                                 start_dp->flags = VRING_DESC_F_NEXT;
 625                         }
 626                 }
 627         }
 628
 629         if (VTPCI_OPS(hw)->setup_queue(hw, vq) < 0) {
 630                 PMD_INIT_LOG(ERR, "setup_queue failed");
 631                 return -EINVAL;
 632         }
 633
 634         return 0;
 635
 636 fail_q_alloc:
 637         rte_free(sw_ring);
 638         rte_memzone_free(hdr_mz);
 639         rte_memzone_free(mz);
 640         rte_free(vq);
 641
 642         return ret;
 643 }
 644
 645 static void
 646 virtio_free_queues(struct virtio_hw *hw)
 647 {
 648         uint16_t nr_vq = virtio_get_nr_vq(hw);
 649         struct virtqueue *vq;
 650         int queue_type;
 651         uint16_t i;
 652
 653         if (hw->vqs == NULL)
 654                 return;
 655
 656         for (i = 0; i < nr_vq; i++) {
 657                 vq = hw->vqs[i];
 658                 if (!vq)
 659                         continue;
 660
 661                 queue_type = virtio_get_queue_type(hw, i);
 662                 if (queue_type == VTNET_RQ) {
 663                         rte_free(vq->sw_ring);
 664                         rte_memzone_free(vq->rxq.mz);
 665                 } else if (queue_type == VTNET_TQ) {
 666                         rte_memzone_free(vq->txq.mz);
 667                         rte_memzone_free(vq->txq.virtio_net_hdr_mz);
 668                 } else {
 669                         rte_memzone_free(vq->cq.mz);
 670                         rte_memzone_free(vq->cq.virtio_net_hdr_mz);
 671                 }
 672
 673                 rte_free(vq);
 674                 hw->vqs[i] = NULL;
 675         }
 676
 677         rte_free(hw->vqs);
 678         hw->vqs = NULL;
 679 }
 680
 681 static int
 682 virtio_alloc_queues(struct rte_eth_dev *dev)
 683 {
 684         struct virtio_hw *hw = dev->data->dev_private;
 685         uint16_t nr_vq = virtio_get_nr_vq(hw);
 686         uint16_t i;
 687         int ret;
 688
 689         hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
 690         if (!hw->vqs) {
 691                 PMD_INIT_LOG(ERR, "failed to allocate vqs");
 692                 return -ENOMEM;
 693         }
 694
 695         for (i = 0; i < nr_vq; i++) {
 696                 ret = virtio_init_queue(dev, i);
 697                 if (ret < 0) {
 698                         virtio_free_queues(hw);
 699                         return ret;
 700                 }
 701         }
 702
 703         return 0;
 704 }
 705
 706 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
 707
 708 static void
 709 virtio_dev_close(struct rte_eth_dev *dev)
 710 {
 711         struct virtio_hw *hw = dev->data->dev_private;
 712         struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
 713
 714         PMD_INIT_LOG(DEBUG, "virtio_dev_close");
 715
 716         if (!hw->opened)
 717                 return;
 718         hw->opened = false;
 719
 720         /* reset the NIC */
 721         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
 722                 VTPCI_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
 723         if (intr_conf->rxq)
 724                 virtio_queues_unbind_intr(dev);
 725
 726         if (intr_conf->lsc || intr_conf->rxq) {
 727                 virtio_intr_disable(dev);
 728                 rte_intr_efd_disable(dev->intr_handle);
 729                 rte_free(dev->intr_handle->intr_vec);
 730                 dev->intr_handle->intr_vec = NULL;
 731         }
 732
 733         vtpci_reset(hw);
 734         virtio_dev_free_mbufs(dev);
 735         virtio_free_queues(hw);
 736 }
 737
 738 static void
 739 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
 740 {
 741         struct virtio_hw *hw = dev->data->dev_private;
 742         struct virtio_pmd_ctrl ctrl;
 743         int dlen[1];
 744         int ret;
 745
 746         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 747                 PMD_INIT_LOG(INFO, "host does not support rx control");
 748                 return;
 749         }
 750
 751         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 752         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 753         ctrl.data[0] = 1;
 754         dlen[0] = 1;
 755
 756         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 757         if (ret)
 758                 PMD_INIT_LOG(ERR, "Failed to enable promisc");
 759 }
 760
 761 static void
 762 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
 763 {
 764         struct virtio_hw *hw = dev->data->dev_private;
 765         struct virtio_pmd_ctrl ctrl;
 766         int dlen[1];
 767         int ret;
 768
 769         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 770                 PMD_INIT_LOG(INFO, "host does not support rx control");
 771                 return;
 772         }
 773
 774         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 775         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 776         ctrl.data[0] = 0;
 777         dlen[0] = 1;
 778
 779         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 780         if (ret)
 781                 PMD_INIT_LOG(ERR, "Failed to disable promisc");
 782 }
 783
 784 static void
 785 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
 786 {
 787         struct virtio_hw *hw = dev->data->dev_private;
 788         struct virtio_pmd_ctrl ctrl;
 789         int dlen[1];
 790         int ret;
 791
 792         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 793                 PMD_INIT_LOG(INFO, "host does not support rx control");
 794                 return;
 795         }
 796
 797         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 798         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 799         ctrl.data[0] = 1;
 800         dlen[0] = 1;
 801
 802         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 803         if (ret)
 804                 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
 805 }
 806
 807 static void
 808 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
 809 {
 810         struct virtio_hw *hw = dev->data->dev_private;
 811         struct virtio_pmd_ctrl ctrl;
 812         int dlen[1];
 813         int ret;
 814
 815         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 816                 PMD_INIT_LOG(INFO, "host does not support rx control");
 817                 return;
 818         }
 819
 820         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 821         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 822         ctrl.data[0] = 0;
 823         dlen[0] = 1;
 824
 825         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 826         if (ret)
 827                 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
 828 }
 829
 830 #define VLAN_TAG_LEN           4    /* 802.3ac tag (not DMA'd) */
 831 static int
 832 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 833 {
 834         struct virtio_hw *hw = dev->data->dev_private;
 835         uint32_t ether_hdr_len = ETHER_HDR_LEN + VLAN_TAG_LEN +
 836                                  hw->vtnet_hdr_size;
 837         uint32_t frame_size = mtu + ether_hdr_len;
 838         uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
 839
 840         max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
 841
 842         if (mtu < ETHER_MIN_MTU || frame_size > max_frame_size) {
 843                 PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
 844                         ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
 845                 return -EINVAL;
 846         }
 847         return 0;
 848 }
 849
 850 static int
 851 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
 852 {
 853         struct virtio_hw *hw = dev->data->dev_private;
 854         struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
 855         struct virtqueue *vq = rxvq->vq;
 856
 857         virtqueue_enable_intr(vq);
 858         virtio_mb(hw->weak_barriers);
 859         return 0;
 860 }
 861
 862 static int
 863 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
 864 {
 865         struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
 866         struct virtqueue *vq = rxvq->vq;
 867
 868         virtqueue_disable_intr(vq);
 869         return 0;
 870 }
 871
 872 /*
 873  * dev_ops for virtio, bare necessities for basic operation
 874  */
 875 static const struct eth_dev_ops virtio_eth_dev_ops = {
 876         .dev_configure           = virtio_dev_configure,
 877         .dev_start               = virtio_dev_start,
 878         .dev_stop                = virtio_dev_stop,
 879         .dev_close               = virtio_dev_close,
 880         .promiscuous_enable      = virtio_dev_promiscuous_enable,
 881         .promiscuous_disable     = virtio_dev_promiscuous_disable,
 882         .allmulticast_enable     = virtio_dev_allmulticast_enable,
 883         .allmulticast_disable    = virtio_dev_allmulticast_disable,
 884         .mtu_set                 = virtio_mtu_set,
 885         .dev_infos_get           = virtio_dev_info_get,
 886         .stats_get               = virtio_dev_stats_get,
 887         .xstats_get              = virtio_dev_xstats_get,
 888         .xstats_get_names        = virtio_dev_xstats_get_names,
 889         .stats_reset             = virtio_dev_stats_reset,
 890         .xstats_reset            = virtio_dev_stats_reset,
 891         .link_update             = virtio_dev_link_update,
 892         .vlan_offload_set        = virtio_dev_vlan_offload_set,
 893         .rx_queue_setup          = virtio_dev_rx_queue_setup,
 894         .rx_queue_intr_enable    = virtio_dev_rx_queue_intr_enable,
 895         .rx_queue_intr_disable   = virtio_dev_rx_queue_intr_disable,
 896         .rx_queue_release        = virtio_dev_queue_release,
 897         .rx_descriptor_done      = virtio_dev_rx_queue_done,
 898         .tx_queue_setup          = virtio_dev_tx_queue_setup,
 899         .tx_queue_release        = virtio_dev_queue_release,
 900         /* collect stats per queue */
 901         .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
 902         .vlan_filter_set         = virtio_vlan_filter_set,
 903         .mac_addr_add            = virtio_mac_addr_add,
 904         .mac_addr_remove         = virtio_mac_addr_remove,
 905         .mac_addr_set            = virtio_mac_addr_set,
 906 };
 907
 908 static void
 909 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 910 {
 911         unsigned i;
 912
 913         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 914                 const struct virtnet_tx *txvq = dev->data->tx_queues[i];
 915                 if (txvq == NULL)
 916                         continue;
 917
 918                 stats->opackets += txvq->stats.packets;
 919                 stats->obytes += txvq->stats.bytes;
 920                 stats->oerrors += txvq->stats.errors;
 921
 922                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 923                         stats->q_opackets[i] = txvq->stats.packets;
 924                         stats->q_obytes[i] = txvq->stats.bytes;
 925                 }
 926         }
 927
 928         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 929                 const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
 930                 if (rxvq == NULL)
 931                         continue;
 932
 933                 stats->ipackets += rxvq->stats.packets;
 934                 stats->ibytes += rxvq->stats.bytes;
 935                 stats->ierrors += rxvq->stats.errors;
 936
 937                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 938                         stats->q_ipackets[i] = rxvq->stats.packets;
 939                         stats->q_ibytes[i] = rxvq->stats.bytes;
 940                 }
 941         }
 942
 943         stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
 944 }
 945
 946 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
 947                                        struct rte_eth_xstat_name *xstats_names,
 948                                        __rte_unused unsigned limit)
 949 {
 950         unsigned i;
 951         unsigned count = 0;
 952         unsigned t;
 953
 954         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
 955                 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
 956
 957         if (xstats_names != NULL) {
 958                 /* Note: limit checked in rte_eth_xstats_names() */
 959
 960                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
 961                         struct virtnet_rx *rxvq = dev->data->rx_queues[i];
 962                         if (rxvq == NULL)
 963                                 continue;
 964                         for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
 965                                 snprintf(xstats_names[count].name,
 966                                         sizeof(xstats_names[count].name),
 967                                         "rx_q%u_%s", i,
 968                                         rte_virtio_rxq_stat_strings[t].name);
 969                                 count++;
 970                         }
 971                 }
 972
 973                 for (i = 0; i < dev->data->nb_tx_queues; i++) {
 974                         struct virtnet_tx *txvq = dev->data->tx_queues[i];
 975                         if (txvq == NULL)
 976                                 continue;
 977                         for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
 978                                 snprintf(xstats_names[count].name,
 979                                         sizeof(xstats_names[count].name),
 980                                         "tx_q%u_%s", i,
 981                                         rte_virtio_txq_stat_strings[t].name);
 982                                 count++;
 983                         }
 984                 }
 985                 return count;
 986         }
 987         return nstats;
 988 }
 989
 990 static int
 991 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 992                       unsigned n)
 993 {
 994         unsigned i;
 995         unsigned count = 0;
 996
 997         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
 998                 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
 999
1000         if (n < nstats)
1001                 return nstats;
1002
1003         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1004                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1005
1006                 if (rxvq == NULL)
1007                         continue;
1008
1009                 unsigned t;
1010
1011                 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1012                         xstats[count].value = *(uint64_t *)(((char *)rxvq) +
1013                                 rte_virtio_rxq_stat_strings[t].offset);
1014                         xstats[count].id = count;
1015                         count++;
1016                 }
1017         }
1018
1019         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1020                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1021
1022                 if (txvq == NULL)
1023                         continue;
1024
1025                 unsigned t;
1026
1027                 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1028                         xstats[count].value = *(uint64_t *)(((char *)txvq) +
1029                                 rte_virtio_txq_stat_strings[t].offset);
1030                         xstats[count].id = count;
1031                         count++;
1032                 }
1033         }
1034
1035         return count;
1036 }
1037
1038 static int
1039 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1040 {
1041         virtio_update_stats(dev, stats);
1042
1043         return 0;
1044 }
1045
1046 static void
1047 virtio_dev_stats_reset(struct rte_eth_dev *dev)
1048 {
1049         unsigned int i;
1050
1051         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1052                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1053                 if (txvq == NULL)
1054                         continue;
1055
1056                 txvq->stats.packets = 0;
1057                 txvq->stats.bytes = 0;
1058                 txvq->stats.errors = 0;
1059                 txvq->stats.multicast = 0;
1060                 txvq->stats.broadcast = 0;
1061                 memset(txvq->stats.size_bins, 0,
1062                        sizeof(txvq->stats.size_bins[0]) * 8);
1063         }
1064
1065         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1066                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1067                 if (rxvq == NULL)
1068                         continue;
1069
1070                 rxvq->stats.packets = 0;
1071                 rxvq->stats.bytes = 0;
1072                 rxvq->stats.errors = 0;
1073                 rxvq->stats.multicast = 0;
1074                 rxvq->stats.broadcast = 0;
1075                 memset(rxvq->stats.size_bins, 0,
1076                        sizeof(rxvq->stats.size_bins[0]) * 8);
1077         }
1078 }
1079
1080 static void
1081 virtio_set_hwaddr(struct virtio_hw *hw)
1082 {
1083         vtpci_write_dev_config(hw,
1084                         offsetof(struct virtio_net_config, mac),
1085                         &hw->mac_addr, ETHER_ADDR_LEN);
1086 }
1087
1088 static void
1089 virtio_get_hwaddr(struct virtio_hw *hw)
1090 {
1091         if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1092                 vtpci_read_dev_config(hw,
1093                         offsetof(struct virtio_net_config, mac),
1094                         &hw->mac_addr, ETHER_ADDR_LEN);
1095         } else {
1096                 eth_random_addr(&hw->mac_addr[0]);
1097                 virtio_set_hwaddr(hw);
1098         }
1099 }
1100
1101 static int
1102 virtio_mac_table_set(struct virtio_hw *hw,
1103                      const struct virtio_net_ctrl_mac *uc,
1104                      const struct virtio_net_ctrl_mac *mc)
1105 {
1106         struct virtio_pmd_ctrl ctrl;
1107         int err, len[2];
1108
1109         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1110                 PMD_DRV_LOG(INFO, "host does not support mac table");
1111                 return -1;
1112         }
1113
1114         ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1115         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1116
1117         len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
1118         memcpy(ctrl.data, uc, len[0]);
1119
1120         len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
1121         memcpy(ctrl.data + len[0], mc, len[1]);
1122
1123         err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1124         if (err != 0)
1125                 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1126         return err;
1127 }
1128
1129 static int
1130 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
1131                     uint32_t index, uint32_t vmdq __rte_unused)
1132 {
1133         struct virtio_hw *hw = dev->data->dev_private;
1134         const struct ether_addr *addrs = dev->data->mac_addrs;
1135         unsigned int i;
1136         struct virtio_net_ctrl_mac *uc, *mc;
1137
1138         if (index >= VIRTIO_MAX_MAC_ADDRS) {
1139                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1140                 return -EINVAL;
1141         }
1142
1143         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1144         uc->entries = 0;
1145         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1146         mc->entries = 0;
1147
1148         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1149                 const struct ether_addr *addr
1150                         = (i == index) ? mac_addr : addrs + i;
1151                 struct virtio_net_ctrl_mac *tbl
1152                         = is_multicast_ether_addr(addr) ? mc : uc;
1153
1154                 memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
1155         }
1156
1157         return virtio_mac_table_set(hw, uc, mc);
1158 }
1159
1160 static void
1161 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1162 {
1163         struct virtio_hw *hw = dev->data->dev_private;
1164         struct ether_addr *addrs = dev->data->mac_addrs;
1165         struct virtio_net_ctrl_mac *uc, *mc;
1166         unsigned int i;
1167
1168         if (index >= VIRTIO_MAX_MAC_ADDRS) {
1169                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1170                 return;
1171         }
1172
1173         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1174         uc->entries = 0;
1175         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1176         mc->entries = 0;
1177
1178         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1179                 struct virtio_net_ctrl_mac *tbl;
1180
1181                 if (i == index || is_zero_ether_addr(addrs + i))
1182                         continue;
1183
1184                 tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
1185                 memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
1186         }
1187
1188         virtio_mac_table_set(hw, uc, mc);
1189 }
1190
1191 static int
1192 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1193 {
1194         struct virtio_hw *hw = dev->data->dev_private;
1195
1196         memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
1197
1198         /* Use atomic update if available */
1199         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1200                 struct virtio_pmd_ctrl ctrl;
1201                 int len = ETHER_ADDR_LEN;
1202
1203                 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1204                 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1205
1206                 memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
1207                 return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1208         }
1209
1210         if (!vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1211                 return -ENOTSUP;
1212
1213         virtio_set_hwaddr(hw);
1214         return 0;
1215 }
1216
1217 static int
1218 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1219 {
1220         struct virtio_hw *hw = dev->data->dev_private;
1221         struct virtio_pmd_ctrl ctrl;
1222         int len;
1223
1224         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1225                 return -ENOTSUP;
1226
1227         ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1228         ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1229         memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1230         len = sizeof(vlan_id);
1231
1232         return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1233 }
1234
1235 static int
1236 virtio_intr_enable(struct rte_eth_dev *dev)
1237 {
1238         struct virtio_hw *hw = dev->data->dev_private;
1239
1240         if (rte_intr_enable(dev->intr_handle) < 0)
1241                 return -1;
1242
1243         if (!hw->virtio_user_dev)
1244                 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1245
1246         return 0;
1247 }
1248
1249 static int
1250 virtio_intr_disable(struct rte_eth_dev *dev)
1251 {
1252         struct virtio_hw *hw = dev->data->dev_private;
1253
1254         if (rte_intr_disable(dev->intr_handle) < 0)
1255                 return -1;
1256
1257         if (!hw->virtio_user_dev)
1258                 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1259
1260         return 0;
1261 }
1262
1263 static int
1264 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1265 {
1266         uint64_t host_features;
1267
1268         /* Prepare guest_features: feature that driver wants to support */
1269         PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1270                 req_features);
1271
1272         /* Read device(host) feature bits */
1273         host_features = VTPCI_OPS(hw)->get_features(hw);
1274         PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1275                 host_features);
1276
1277         /* If supported, ensure MTU value is valid before acknowledging it. */
1278         if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1279                 struct virtio_net_config config;
1280
1281                 vtpci_read_dev_config(hw,
1282                         offsetof(struct virtio_net_config, mtu),
1283                         &config.mtu, sizeof(config.mtu));
1284
1285                 if (config.mtu < ETHER_MIN_MTU)
1286                         req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1287         }
1288
1289         /*
1290          * Negotiate features: Subset of device feature bits are written back
1291          * guest feature bits.
1292          */
1293         hw->guest_features = req_features;
1294         hw->guest_features = vtpci_negotiate_features(hw, host_features);
1295         PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1296                 hw->guest_features);
1297
1298         if (hw->modern) {
1299                 if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1300                         PMD_INIT_LOG(ERR,
1301                                 "VIRTIO_F_VERSION_1 features is not enabled.");
1302                         return -1;
1303                 }
1304                 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1305                 if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1306                         PMD_INIT_LOG(ERR,
1307                                 "failed to set FEATURES_OK status!");
1308                         return -1;
1309                 }
1310         }
1311
1312         hw->req_guest_features = req_features;
1313
1314         return 0;
1315 }
1316
1317 int
1318 virtio_dev_pause(struct rte_eth_dev *dev)
1319 {
1320         struct virtio_hw *hw = dev->data->dev_private;
1321
1322         rte_spinlock_lock(&hw->state_lock);
1323
1324         if (hw->started == 0) {
1325                 /* Device is just stopped. */
1326                 rte_spinlock_unlock(&hw->state_lock);
1327                 return -1;
1328         }
1329         hw->started = 0;
1330         /*
1331          * Prevent the worker threads from touching queues to avoid contention,
1332          * 1 ms should be enough for the ongoing Tx function to finish.
1333          */
1334         rte_delay_ms(1);
1335         return 0;
1336 }
1337
1338 /*
1339  * Recover hw state to let the worker threads continue.
1340  */
1341 void
1342 virtio_dev_resume(struct rte_eth_dev *dev)
1343 {
1344         struct virtio_hw *hw = dev->data->dev_private;
1345
1346         hw->started = 1;
1347         rte_spinlock_unlock(&hw->state_lock);
1348 }
1349
1350 /*
1351  * Should be called only after device is paused.
1352  */
1353 int
1354 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1355                 int nb_pkts)
1356 {
1357         struct virtio_hw *hw = dev->data->dev_private;
1358         struct virtnet_tx *txvq = dev->data->tx_queues[0];
1359         int ret;
1360
1361         hw->inject_pkts = tx_pkts;
1362         ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1363         hw->inject_pkts = NULL;
1364
1365         return ret;
1366 }
1367
1368 static void
1369 virtio_notify_peers(struct rte_eth_dev *dev)
1370 {
1371         struct virtio_hw *hw = dev->data->dev_private;
1372         struct virtnet_rx *rxvq;
1373         struct rte_mbuf *rarp_mbuf;
1374
1375         if (!dev->data->rx_queues)
1376                 return;
1377
1378         rxvq = dev->data->rx_queues[0];
1379         if (!rxvq)
1380                 return;
1381
1382         rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1383                         (struct ether_addr *)hw->mac_addr);
1384         if (rarp_mbuf == NULL) {
1385                 PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1386                 return;
1387         }
1388
1389         /* If virtio port just stopped, no need to send RARP */
1390         if (virtio_dev_pause(dev) < 0) {
1391                 rte_pktmbuf_free(rarp_mbuf);
1392                 return;
1393         }
1394
1395         virtio_inject_pkts(dev, &rarp_mbuf, 1);
1396         virtio_dev_resume(dev);
1397 }
1398
1399 static void
1400 virtio_ack_link_announce(struct rte_eth_dev *dev)
1401 {
1402         struct virtio_hw *hw = dev->data->dev_private;
1403         struct virtio_pmd_ctrl ctrl;
1404
1405         ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1406         ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1407
1408         virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1409 }
1410
1411 /*
1412  * Process virtio config changed interrupt. Call the callback
1413  * if link state changed, generate gratuitous RARP packet if
1414  * the status indicates an ANNOUNCE.
1415  */
1416 void
1417 virtio_interrupt_handler(void *param)
1418 {
1419         struct rte_eth_dev *dev = param;
1420         struct virtio_hw *hw = dev->data->dev_private;
1421         uint8_t isr;
1422         uint16_t status;
1423
1424         /* Read interrupt status which clears interrupt */
1425         isr = vtpci_isr(hw);
1426         PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1427
1428         if (virtio_intr_enable(dev) < 0)
1429                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1430
1431         if (isr & VIRTIO_PCI_ISR_CONFIG) {
1432                 if (virtio_dev_link_update(dev, 0) == 0)
1433                         _rte_eth_dev_callback_process(dev,
1434                                                       RTE_ETH_EVENT_INTR_LSC,
1435                                                       NULL);
1436
1437                 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1438                         vtpci_read_dev_config(hw,
1439                                 offsetof(struct virtio_net_config, status),
1440                                 &status, sizeof(status));
1441                         if (status & VIRTIO_NET_S_ANNOUNCE) {
1442                                 virtio_notify_peers(dev);
1443                                 if (hw->cvq)
1444                                         virtio_ack_link_announce(dev);
1445                         }
1446                 }
1447         }
1448 }
1449
1450 /* set rx and tx handlers according to what is supported */
1451 static void
1452 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1453 {
1454         struct virtio_hw *hw = eth_dev->data->dev_private;
1455
1456         if (vtpci_packed_queue(hw)) {
1457                 PMD_INIT_LOG(INFO,
1458                         "virtio: using packed ring %s Tx path on port %u",
1459                         hw->use_inorder_tx ? "inorder" : "standard",
1460                         eth_dev->data->port_id);
1461                 eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed;
1462         } else {
1463                 if (hw->use_inorder_tx) {
1464                         PMD_INIT_LOG(INFO, "virtio: using inorder Tx path on port %u",
1465                                 eth_dev->data->port_id);
1466                         eth_dev->tx_pkt_burst = virtio_xmit_pkts_inorder;
1467                 } else {
1468                         PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1469                                 eth_dev->data->port_id);
1470                         eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1471                 }
1472         }
1473
1474         if (vtpci_packed_queue(hw)) {
1475                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1476                         PMD_INIT_LOG(INFO,
1477                                 "virtio: using packed ring mergeable buffer Rx path on port %u",
1478                                 eth_dev->data->port_id);
1479                         eth_dev->rx_pkt_burst =
1480                                 &virtio_recv_mergeable_pkts_packed;
1481                 } else {
1482                         PMD_INIT_LOG(INFO,
1483                                 "virtio: using packed ring standard Rx path on port %u",
1484                                 eth_dev->data->port_id);
1485                         eth_dev->rx_pkt_burst = &virtio_recv_pkts_packed;
1486                 }
1487         } else {
1488                 if (hw->use_simple_rx) {
1489                         PMD_INIT_LOG(INFO, "virtio: using simple Rx path on port %u",
1490                                 eth_dev->data->port_id);
1491                         eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1492                 } else if (hw->use_inorder_rx) {
1493                         PMD_INIT_LOG(INFO,
1494                                 "virtio: using inorder Rx path on port %u",
1495                                 eth_dev->data->port_id);
1496                         eth_dev->rx_pkt_burst = &virtio_recv_pkts_inorder;
1497                 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1498                         PMD_INIT_LOG(INFO,
1499                                 "virtio: using mergeable buffer Rx path on port %u",
1500                                 eth_dev->data->port_id);
1501                         eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1502                 } else {
1503                         PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1504                                 eth_dev->data->port_id);
1505                         eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1506                 }
1507         }
1508
1509 }
1510
1511 /* Only support 1:1 queue/interrupt mapping so far.
1512  * TODO: support n:1 queue/interrupt mapping when there are limited number of
1513  * interrupt vectors (<N+1).
1514  */
1515 static int
1516 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1517 {
1518         uint32_t i;
1519         struct virtio_hw *hw = dev->data->dev_private;
1520
1521         PMD_INIT_LOG(INFO, "queue/interrupt binding");
1522         for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1523                 dev->intr_handle->intr_vec[i] = i + 1;
1524                 if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1525                                                  VIRTIO_MSI_NO_VECTOR) {
1526                         PMD_DRV_LOG(ERR, "failed to set queue vector");
1527                         return -EBUSY;
1528                 }
1529         }
1530
1531         return 0;
1532 }
1533
1534 static void
1535 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1536 {
1537         uint32_t i;
1538         struct virtio_hw *hw = dev->data->dev_private;
1539
1540         PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1541         for (i = 0; i < dev->data->nb_rx_queues; ++i)
1542                 VTPCI_OPS(hw)->set_queue_irq(hw,
1543                                              hw->vqs[i * VTNET_CQ],
1544                                              VIRTIO_MSI_NO_VECTOR);
1545 }
1546
1547 static int
1548 virtio_configure_intr(struct rte_eth_dev *dev)
1549 {
1550         struct virtio_hw *hw = dev->data->dev_private;
1551
1552         if (!rte_intr_cap_multiple(dev->intr_handle)) {
1553                 PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1554                 return -ENOTSUP;
1555         }
1556
1557         if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1558                 PMD_INIT_LOG(ERR, "Fail to create eventfd");
1559                 return -1;
1560         }
1561
1562         if (!dev->intr_handle->intr_vec) {
1563                 dev->intr_handle->intr_vec =
1564                         rte_zmalloc("intr_vec",
1565                                     hw->max_queue_pairs * sizeof(int), 0);
1566                 if (!dev->intr_handle->intr_vec) {
1567                         PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1568                                      hw->max_queue_pairs);
1569                         return -ENOMEM;
1570                 }
1571         }
1572
1573         /* Re-register callback to update max_intr */
1574         rte_intr_callback_unregister(dev->intr_handle,
1575                                      virtio_interrupt_handler,
1576                                      dev);
1577         rte_intr_callback_register(dev->intr_handle,
1578                                    virtio_interrupt_handler,
1579                                    dev);
1580
1581         /* DO NOT try to remove this! This function will enable msix, or QEMU
1582          * will encounter SIGSEGV when DRIVER_OK is sent.
1583          * And for legacy devices, this should be done before queue/vec binding
1584          * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1585          * (22) will be ignored.
1586          */
1587         if (virtio_intr_enable(dev) < 0) {
1588                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1589                 return -1;
1590         }
1591
1592         if (virtio_queues_bind_intr(dev) < 0) {
1593                 PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1594                 return -1;
1595         }
1596
1597         return 0;
1598 }
1599
1600 /* reset device and renegotiate features if needed */
1601 static int
1602 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1603 {
1604         struct virtio_hw *hw = eth_dev->data->dev_private;
1605         struct virtio_net_config *config;
1606         struct virtio_net_config local_config;
1607         struct rte_pci_device *pci_dev = NULL;
1608         int ret;
1609
1610         /* Reset the device although not necessary at startup */
1611         vtpci_reset(hw);
1612
1613         if (hw->vqs) {
1614                 virtio_dev_free_mbufs(eth_dev);
1615                 virtio_free_queues(hw);
1616         }
1617
1618         /* Tell the host we've noticed this device. */
1619         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1620
1621         /* Tell the host we've known how to drive the device. */
1622         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1623         if (virtio_negotiate_features(hw, req_features) < 0)
1624                 return -1;
1625
1626         hw->weak_barriers = !vtpci_with_feature(hw, VIRTIO_F_ORDER_PLATFORM);
1627
1628         if (!hw->virtio_user_dev) {
1629                 pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1630                 rte_eth_copy_pci_info(eth_dev, pci_dev);
1631         }
1632
1633         /* If host does not support both status and MSI-X then disable LSC */
1634         if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) &&
1635             hw->use_msix != VIRTIO_MSIX_NONE)
1636                 eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1637         else
1638                 eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1639
1640         /* Setting up rx_header size for the device */
1641         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1642             vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
1643             vtpci_with_feature(hw, VIRTIO_F_RING_PACKED))
1644                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1645         else
1646                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1647
1648         /* Copy the permanent MAC address to: virtio_hw */
1649         virtio_get_hwaddr(hw);
1650         ether_addr_copy((struct ether_addr *) hw->mac_addr,
1651                         &eth_dev->data->mac_addrs[0]);
1652         PMD_INIT_LOG(DEBUG,
1653                      "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1654                      hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1655                      hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1656
1657         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1658                 config = &local_config;
1659
1660                 vtpci_read_dev_config(hw,
1661                         offsetof(struct virtio_net_config, mac),
1662                         &config->mac, sizeof(config->mac));
1663
1664                 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1665                         vtpci_read_dev_config(hw,
1666                                 offsetof(struct virtio_net_config, status),
1667                                 &config->status, sizeof(config->status));
1668                 } else {
1669                         PMD_INIT_LOG(DEBUG,
1670                                      "VIRTIO_NET_F_STATUS is not supported");
1671                         config->status = 0;
1672                 }
1673
1674                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1675                         vtpci_read_dev_config(hw,
1676                                 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1677                                 &config->max_virtqueue_pairs,
1678                                 sizeof(config->max_virtqueue_pairs));
1679                 } else {
1680                         PMD_INIT_LOG(DEBUG,
1681                                      "VIRTIO_NET_F_MQ is not supported");
1682                         config->max_virtqueue_pairs = 1;
1683                 }
1684
1685                 hw->max_queue_pairs = config->max_virtqueue_pairs;
1686
1687                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1688                         vtpci_read_dev_config(hw,
1689                                 offsetof(struct virtio_net_config, mtu),
1690                                 &config->mtu,
1691                                 sizeof(config->mtu));
1692
1693                         /*
1694                          * MTU value has already been checked at negotiation
1695                          * time, but check again in case it has changed since
1696                          * then, which should not happen.
1697                          */
1698                         if (config->mtu < ETHER_MIN_MTU) {
1699                                 PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1700                                                 config->mtu);
1701                                 return -1;
1702                         }
1703
1704                         hw->max_mtu = config->mtu;
1705                         /* Set initial MTU to maximum one supported by vhost */
1706                         eth_dev->data->mtu = config->mtu;
1707
1708                 } else {
1709                         hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - ETHER_HDR_LEN -
1710                                 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1711                 }
1712
1713                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1714                                 config->max_virtqueue_pairs);
1715                 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1716                 PMD_INIT_LOG(DEBUG,
1717                                 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1718                                 config->mac[0], config->mac[1],
1719                                 config->mac[2], config->mac[3],
1720                                 config->mac[4], config->mac[5]);
1721         } else {
1722                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1723                 hw->max_queue_pairs = 1;
1724                 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - ETHER_HDR_LEN -
1725                         VLAN_TAG_LEN - hw->vtnet_hdr_size;
1726         }
1727
1728         ret = virtio_alloc_queues(eth_dev);
1729         if (ret < 0)
1730                 return ret;
1731
1732         if (eth_dev->data->dev_conf.intr_conf.rxq) {
1733                 if (virtio_configure_intr(eth_dev) < 0) {
1734                         PMD_INIT_LOG(ERR, "failed to configure interrupt");
1735                         return -1;
1736                 }
1737         }
1738
1739         vtpci_reinit_complete(hw);
1740
1741         if (pci_dev)
1742                 PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1743                         eth_dev->data->port_id, pci_dev->id.vendor_id,
1744                         pci_dev->id.device_id);
1745
1746         return 0;
1747 }
1748
1749 /*
1750  * Remap the PCI device again (IO port map for legacy device and
1751  * memory map for modern device), so that the secondary process
1752  * could have the PCI initiated correctly.
1753  */
1754 static int
1755 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1756 {
1757         if (hw->modern) {
1758                 /*
1759                  * We don't have to re-parse the PCI config space, since
1760                  * rte_pci_map_device() makes sure the mapped address
1761                  * in secondary process would equal to the one mapped in
1762                  * the primary process: error will be returned if that
1763                  * requirement is not met.
1764                  *
1765                  * That said, we could simply reuse all cap pointers
1766                  * (such as dev_cfg, common_cfg, etc.) parsed from the
1767                  * primary process, which is stored in shared memory.
1768                  */
1769                 if (rte_pci_map_device(pci_dev)) {
1770                         PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1771                         return -1;
1772                 }
1773         } else {
1774                 if (rte_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1775                         return -1;
1776         }
1777
1778         return 0;
1779 }
1780
1781 static void
1782 virtio_set_vtpci_ops(struct virtio_hw *hw)
1783 {
1784 #ifdef RTE_VIRTIO_USER
1785         if (hw->virtio_user_dev)
1786                 VTPCI_OPS(hw) = &virtio_user_ops;
1787         else
1788 #endif
1789         if (hw->modern)
1790                 VTPCI_OPS(hw) = &modern_ops;
1791         else
1792                 VTPCI_OPS(hw) = &legacy_ops;
1793 }
1794
1795 /*
1796  * This function is based on probe() function in virtio_pci.c
1797  * It returns 0 on success.
1798  */
1799 int
1800 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1801 {
1802         struct virtio_hw *hw = eth_dev->data->dev_private;
1803         int ret;
1804
1805         RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr_mrg_rxbuf));
1806
1807         eth_dev->dev_ops = &virtio_eth_dev_ops;
1808
1809         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1810                 if (!hw->virtio_user_dev) {
1811                         ret = virtio_remap_pci(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1812                         if (ret)
1813                                 return ret;
1814                 }
1815
1816                 virtio_set_vtpci_ops(hw);
1817                 set_rxtx_funcs(eth_dev);
1818
1819                 return 0;
1820         }
1821
1822         /* Allocate memory for storing MAC addresses */
1823         eth_dev->data->mac_addrs = rte_zmalloc("virtio", VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN, 0);
1824         if (eth_dev->data->mac_addrs == NULL) {
1825                 PMD_INIT_LOG(ERR,
1826                         "Failed to allocate %d bytes needed to store MAC addresses",
1827                         VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN);
1828                 return -ENOMEM;
1829         }
1830
1831         hw->port_id = eth_dev->data->port_id;
1832         /* For virtio_user case the hw->virtio_user_dev is populated by
1833          * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1834          */
1835         if (!hw->virtio_user_dev) {
1836                 ret = vtpci_init(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1837                 if (ret)
1838                         goto out;
1839         }
1840
1841         /* reset device and negotiate default features */
1842         ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1843         if (ret < 0)
1844                 goto out;
1845
1846         return 0;
1847
1848 out:
1849         rte_free(eth_dev->data->mac_addrs);
1850         return ret;
1851 }
1852
1853 static int
1854 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1855 {
1856         PMD_INIT_FUNC_TRACE();
1857
1858         if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1859                 return 0;
1860
1861         virtio_dev_stop(eth_dev);
1862         virtio_dev_close(eth_dev);
1863
1864         eth_dev->dev_ops = NULL;
1865         eth_dev->tx_pkt_burst = NULL;
1866         eth_dev->rx_pkt_burst = NULL;
1867
1868         if (eth_dev->device)
1869                 rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev));
1870
1871         PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1872
1873         return 0;
1874 }
1875
1876 static int vdpa_check_handler(__rte_unused const char *key,
1877                 const char *value, __rte_unused void *opaque)
1878 {
1879         if (strcmp(value, "1"))
1880                 return -1;
1881
1882         return 0;
1883 }
1884
1885 static int
1886 vdpa_mode_selected(struct rte_devargs *devargs)
1887 {
1888         struct rte_kvargs *kvlist;
1889         const char *key = "vdpa";
1890         int ret = 0;
1891
1892         if (devargs == NULL)
1893                 return 0;
1894
1895         kvlist = rte_kvargs_parse(devargs->args, NULL);
1896         if (kvlist == NULL)
1897                 return 0;
1898
1899         if (!rte_kvargs_count(kvlist, key))
1900                 goto exit;
1901
1902         /* vdpa mode selected when there's a key-value pair: vdpa=1 */
1903         if (rte_kvargs_process(kvlist, key,
1904                                 vdpa_check_handler, NULL) < 0) {
1905                 goto exit;
1906         }
1907         ret = 1;
1908
1909 exit:
1910         rte_kvargs_free(kvlist);
1911         return ret;
1912 }
1913
1914 static int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1915         struct rte_pci_device *pci_dev)
1916 {
1917         if (rte_eal_iopl_init() != 0) {
1918                 PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1919                 return 1;
1920         }
1921
1922         /* virtio pmd skips probe if device needs to work in vdpa mode */
1923         if (vdpa_mode_selected(pci_dev->device.devargs))
1924                 return 1;
1925
1926         return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw),
1927                 eth_virtio_dev_init);
1928 }
1929
1930 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev)
1931 {
1932         return rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
1933 }
1934
1935 static struct rte_pci_driver rte_virtio_pmd = {
1936         .driver = {
1937                 .name = "net_virtio",
1938         },
1939         .id_table = pci_id_virtio_map,
1940         .drv_flags = 0,
1941         .probe = eth_virtio_pci_probe,
1942         .remove = eth_virtio_pci_remove,
1943 };
1944
1945 RTE_INIT(rte_virtio_pmd_init)
1946 {
1947         rte_eal_iopl_init();
1948         rte_pci_register(&rte_virtio_pmd);
1949 }
1950
1951 static bool
1952 rx_offload_enabled(struct virtio_hw *hw)
1953 {
1954         return vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM) ||
1955                 vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
1956                 vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6);
1957 }
1958
1959 static bool
1960 tx_offload_enabled(struct virtio_hw *hw)
1961 {
1962         return vtpci_with_feature(hw, VIRTIO_NET_F_CSUM) ||
1963                 vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO4) ||
1964                 vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO6);
1965 }
1966
1967 /*
1968  * Configure virtio device
1969  * It returns 0 on success.
1970  */
1971 static int
1972 virtio_dev_configure(struct rte_eth_dev *dev)
1973 {
1974         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1975         const struct rte_eth_txmode *txmode = &dev->data->dev_conf.txmode;
1976         struct virtio_hw *hw = dev->data->dev_private;
1977         uint32_t ether_hdr_len = ETHER_HDR_LEN + VLAN_TAG_LEN +
1978                 hw->vtnet_hdr_size;
1979         uint64_t rx_offloads = rxmode->offloads;
1980         uint64_t tx_offloads = txmode->offloads;
1981         uint64_t req_features;
1982         int ret;
1983
1984         PMD_INIT_LOG(DEBUG, "configure");
1985         req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
1986
1987         if (dev->data->dev_conf.intr_conf.rxq) {
1988                 ret = virtio_init_device(dev, hw->req_guest_features);
1989                 if (ret < 0)
1990                         return ret;
1991         }
1992
1993         if (rxmode->max_rx_pkt_len > hw->max_mtu + ether_hdr_len)
1994                 req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1995
1996         if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
1997                            DEV_RX_OFFLOAD_TCP_CKSUM))
1998                 req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
1999
2000         if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
2001                 req_features |=
2002                         (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2003                         (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2004
2005         if (tx_offloads & (DEV_TX_OFFLOAD_UDP_CKSUM |
2006                            DEV_TX_OFFLOAD_TCP_CKSUM))
2007                 req_features |= (1ULL << VIRTIO_NET_F_CSUM);
2008
2009         if (tx_offloads & DEV_TX_OFFLOAD_TCP_TSO)
2010                 req_features |=
2011                         (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2012                         (1ULL << VIRTIO_NET_F_HOST_TSO6);
2013
2014         /* if request features changed, reinit the device */
2015         if (req_features != hw->req_guest_features) {
2016                 ret = virtio_init_device(dev, req_features);
2017                 if (ret < 0)
2018                         return ret;
2019         }
2020
2021         if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2022                             DEV_RX_OFFLOAD_TCP_CKSUM)) &&
2023                 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
2024                 PMD_DRV_LOG(ERR,
2025                         "rx checksum not available on this host");
2026                 return -ENOTSUP;
2027         }
2028
2029         if ((rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) &&
2030                 (!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2031                  !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
2032                 PMD_DRV_LOG(ERR,
2033                         "Large Receive Offload not available on this host");
2034                 return -ENOTSUP;
2035         }
2036
2037         /* start control queue */
2038         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
2039                 virtio_dev_cq_start(dev);
2040
2041         if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2042                 hw->vlan_strip = 1;
2043
2044         if ((rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
2045             && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2046                 PMD_DRV_LOG(ERR,
2047                             "vlan filtering not available on this host");
2048                 return -ENOTSUP;
2049         }
2050
2051         hw->has_tx_offload = tx_offload_enabled(hw);
2052         hw->has_rx_offload = rx_offload_enabled(hw);
2053
2054         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2055                 /* Enable vector (0) for Link State Intrerrupt */
2056                 if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
2057                                 VIRTIO_MSI_NO_VECTOR) {
2058                         PMD_DRV_LOG(ERR, "failed to set config vector");
2059                         return -EBUSY;
2060                 }
2061
2062         rte_spinlock_init(&hw->state_lock);
2063
2064         hw->use_simple_rx = 1;
2065
2066         if (vtpci_with_feature(hw, VIRTIO_F_IN_ORDER)) {
2067                 hw->use_inorder_tx = 1;
2068                 hw->use_inorder_rx = 1;
2069                 hw->use_simple_rx = 0;
2070         }
2071
2072         if (vtpci_packed_queue(hw)) {
2073                 hw->use_simple_rx = 0;
2074                 hw->use_inorder_rx = 0;
2075         }
2076
2077 #if defined RTE_ARCH_ARM64 || defined RTE_ARCH_ARM
2078         if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
2079                 hw->use_simple_rx = 0;
2080         }
2081 #endif
2082         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2083                  hw->use_simple_rx = 0;
2084         }
2085
2086         if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2087                            DEV_RX_OFFLOAD_TCP_CKSUM |
2088                            DEV_RX_OFFLOAD_TCP_LRO |
2089                            DEV_RX_OFFLOAD_VLAN_STRIP))
2090                 hw->use_simple_rx = 0;
2091
2092         hw->opened = true;
2093
2094         return 0;
2095 }
2096
2097
2098 static int
2099 virtio_dev_start(struct rte_eth_dev *dev)
2100 {
2101         uint16_t nb_queues, i;
2102         struct virtnet_rx *rxvq;
2103         struct virtnet_tx *txvq __rte_unused;
2104         struct virtio_hw *hw = dev->data->dev_private;
2105         int ret;
2106
2107         /* Finish the initialization of the queues */
2108         for (i = 0; i < dev->data->nb_rx_queues; i++) {
2109                 ret = virtio_dev_rx_queue_setup_finish(dev, i);
2110                 if (ret < 0)
2111                         return ret;
2112         }
2113         for (i = 0; i < dev->data->nb_tx_queues; i++) {
2114                 ret = virtio_dev_tx_queue_setup_finish(dev, i);
2115                 if (ret < 0)
2116                         return ret;
2117         }
2118
2119         /* check if lsc interrupt feature is enabled */
2120         if (dev->data->dev_conf.intr_conf.lsc) {
2121                 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
2122                         PMD_DRV_LOG(ERR, "link status not supported by host");
2123                         return -ENOTSUP;
2124                 }
2125         }
2126
2127         /* Enable uio/vfio intr/eventfd mapping: althrough we already did that
2128          * in device configure, but it could be unmapped  when device is
2129          * stopped.
2130          */
2131         if (dev->data->dev_conf.intr_conf.lsc ||
2132             dev->data->dev_conf.intr_conf.rxq) {
2133                 virtio_intr_disable(dev);
2134
2135                 /* Setup interrupt callback  */
2136                 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2137                         rte_intr_callback_register(dev->intr_handle,
2138                                                    virtio_interrupt_handler,
2139                                                    dev);
2140
2141                 if (virtio_intr_enable(dev) < 0) {
2142                         PMD_DRV_LOG(ERR, "interrupt enable failed");
2143                         return -EIO;
2144                 }
2145         }
2146
2147         /*Notify the backend
2148          *Otherwise the tap backend might already stop its queue due to fullness.
2149          *vhost backend will have no chance to be waked up
2150          */
2151         nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
2152         if (hw->max_queue_pairs > 1) {
2153                 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
2154                         return -EINVAL;
2155         }
2156
2157         PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
2158
2159         for (i = 0; i < dev->data->nb_rx_queues; i++) {
2160                 rxvq = dev->data->rx_queues[i];
2161                 /* Flush the old packets */
2162                 virtqueue_rxvq_flush(rxvq->vq);
2163                 virtqueue_notify(rxvq->vq);
2164         }
2165
2166         for (i = 0; i < dev->data->nb_tx_queues; i++) {
2167                 txvq = dev->data->tx_queues[i];
2168                 virtqueue_notify(txvq->vq);
2169         }
2170
2171         PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
2172
2173         for (i = 0; i < dev->data->nb_rx_queues; i++) {
2174                 rxvq = dev->data->rx_queues[i];
2175                 VIRTQUEUE_DUMP(rxvq->vq);
2176         }
2177
2178         for (i = 0; i < dev->data->nb_tx_queues; i++) {
2179                 txvq = dev->data->tx_queues[i];
2180                 VIRTQUEUE_DUMP(txvq->vq);
2181         }
2182
2183         set_rxtx_funcs(dev);
2184         hw->started = true;
2185
2186         /* Initialize Link state */
2187         virtio_dev_link_update(dev, 0);
2188
2189         return 0;
2190 }
2191
2192 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
2193 {
2194         struct virtio_hw *hw = dev->data->dev_private;
2195         uint16_t nr_vq = virtio_get_nr_vq(hw);
2196         const char *type __rte_unused;
2197         unsigned int i, mbuf_num = 0;
2198         struct virtqueue *vq;
2199         struct rte_mbuf *buf;
2200         int queue_type;
2201
2202         if (hw->vqs == NULL)
2203                 return;
2204
2205         for (i = 0; i < nr_vq; i++) {
2206                 vq = hw->vqs[i];
2207                 if (!vq)
2208                         continue;
2209
2210                 queue_type = virtio_get_queue_type(hw, i);
2211                 if (queue_type == VTNET_RQ)
2212                         type = "rxq";
2213                 else if (queue_type == VTNET_TQ)
2214                         type = "txq";
2215                 else
2216                         continue;
2217
2218                 PMD_INIT_LOG(DEBUG,
2219                         "Before freeing %s[%d] used and unused buf",
2220                         type, i);
2221                 VIRTQUEUE_DUMP(vq);
2222
2223                 while ((buf = virtqueue_detach_unused(vq)) != NULL) {
2224                         rte_pktmbuf_free(buf);
2225                         mbuf_num++;
2226                 }
2227
2228                 PMD_INIT_LOG(DEBUG,
2229                         "After freeing %s[%d] used and unused buf",
2230                         type, i);
2231                 VIRTQUEUE_DUMP(vq);
2232         }
2233
2234         PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
2235 }
2236
2237 /*
2238  * Stop device: disable interrupt and mark link down
2239  */
2240 static void
2241 virtio_dev_stop(struct rte_eth_dev *dev)
2242 {
2243         struct virtio_hw *hw = dev->data->dev_private;
2244         struct rte_eth_link link;
2245         struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
2246
2247         PMD_INIT_LOG(DEBUG, "stop");
2248
2249         rte_spinlock_lock(&hw->state_lock);
2250         if (!hw->started)
2251                 goto out_unlock;
2252         hw->started = false;
2253
2254         if (intr_conf->lsc || intr_conf->rxq) {
2255                 virtio_intr_disable(dev);
2256
2257                 /* Reset interrupt callback  */
2258                 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
2259                         rte_intr_callback_unregister(dev->intr_handle,
2260                                                      virtio_interrupt_handler,
2261                                                      dev);
2262                 }
2263         }
2264
2265         memset(&link, 0, sizeof(link));
2266         rte_eth_linkstatus_set(dev, &link);
2267 out_unlock:
2268         rte_spinlock_unlock(&hw->state_lock);
2269 }
2270
2271 static int
2272 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2273 {
2274         struct rte_eth_link link;
2275         uint16_t status;
2276         struct virtio_hw *hw = dev->data->dev_private;
2277
2278         memset(&link, 0, sizeof(link));
2279         link.link_duplex = ETH_LINK_FULL_DUPLEX;
2280         link.link_speed  = ETH_SPEED_NUM_10G;
2281         link.link_autoneg = ETH_LINK_FIXED;
2282
2283         if (!hw->started) {
2284                 link.link_status = ETH_LINK_DOWN;
2285         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2286                 PMD_INIT_LOG(DEBUG, "Get link status from hw");
2287                 vtpci_read_dev_config(hw,
2288                                 offsetof(struct virtio_net_config, status),
2289                                 &status, sizeof(status));
2290                 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2291                         link.link_status = ETH_LINK_DOWN;
2292                         PMD_INIT_LOG(DEBUG, "Port %d is down",
2293                                      dev->data->port_id);
2294                 } else {
2295                         link.link_status = ETH_LINK_UP;
2296                         PMD_INIT_LOG(DEBUG, "Port %d is up",
2297                                      dev->data->port_id);
2298                 }
2299         } else {
2300                 link.link_status = ETH_LINK_UP;
2301         }
2302
2303         return rte_eth_linkstatus_set(dev, &link);
2304 }
2305
2306 static int
2307 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2308 {
2309         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2310         struct virtio_hw *hw = dev->data->dev_private;
2311         uint64_t offloads = rxmode->offloads;
2312
2313         if (mask & ETH_VLAN_FILTER_MASK) {
2314                 if ((offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2315                                 !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2316
2317                         PMD_DRV_LOG(NOTICE,
2318                                 "vlan filtering not available on this host");
2319
2320                         return -ENOTSUP;
2321                 }
2322         }
2323
2324         if (mask & ETH_VLAN_STRIP_MASK)
2325                 hw->vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
2326
2327         return 0;
2328 }
2329
2330 static void
2331 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2332 {
2333         uint64_t tso_mask, host_features;
2334         struct virtio_hw *hw = dev->data->dev_private;
2335
2336         dev_info->speed_capa = ETH_LINK_SPEED_10G; /* fake value */
2337
2338         dev_info->max_rx_queues =
2339                 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2340         dev_info->max_tx_queues =
2341                 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2342         dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2343         dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2344         dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2345
2346         host_features = VTPCI_OPS(hw)->get_features(hw);
2347         dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
2348         dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2349         if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2350                 dev_info->rx_offload_capa |=
2351                         DEV_RX_OFFLOAD_TCP_CKSUM |
2352                         DEV_RX_OFFLOAD_UDP_CKSUM;
2353         }
2354         if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN))
2355                 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_FILTER;
2356         tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2357                 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2358         if ((host_features & tso_mask) == tso_mask)
2359                 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2360
2361         dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
2362                                     DEV_TX_OFFLOAD_VLAN_INSERT;
2363         if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2364                 dev_info->tx_offload_capa |=
2365                         DEV_TX_OFFLOAD_UDP_CKSUM |
2366                         DEV_TX_OFFLOAD_TCP_CKSUM;
2367         }
2368         tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2369                 (1ULL << VIRTIO_NET_F_HOST_TSO6);
2370         if ((host_features & tso_mask) == tso_mask)
2371                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2372 }
2373
2374 /*
2375  * It enables testpmd to collect per queue stats.
2376  */
2377 static int
2378 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2379 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2380 __rte_unused uint8_t is_rx)
2381 {
2382         return 0;
2383 }
2384
2385 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
2386 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
2387 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio-pci");
2388
2389 RTE_INIT(virtio_init_log)
2390 {
2391         virtio_logtype_init = rte_log_register("pmd.net.virtio.init");
2392         if (virtio_logtype_init >= 0)
2393                 rte_log_set_level(virtio_logtype_init, RTE_LOG_NOTICE);
2394         virtio_logtype_driver = rte_log_register("pmd.net.virtio.driver");
2395         if (virtio_logtype_driver >= 0)
2396                 rte_log_set_level(virtio_logtype_driver, RTE_LOG_NOTICE);
2397 }