drivers/net/virtio/virtio_ethdev.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2016 Intel Corporation
   3  */
   4
   5 #include <stdint.h>
   6 #include <string.h>
   7 #include <stdio.h>
   8 #include <errno.h>
   9 #include <unistd.h>
  10
  11 #include <rte_ethdev_driver.h>
  12 #include <rte_ethdev_pci.h>
  13 #include <rte_memcpy.h>
  14 #include <rte_string_fns.h>
  15 #include <rte_memzone.h>
  16 #include <rte_malloc.h>
  17 #include <rte_branch_prediction.h>
  18 #include <rte_pci.h>
  19 #include <rte_bus_pci.h>
  20 #include <rte_ether.h>
  21 #include <rte_ip.h>
  22 #include <rte_arp.h>
  23 #include <rte_common.h>
  24 #include <rte_errno.h>
  25 #include <rte_cpuflags.h>
  26
  27 #include <rte_memory.h>
  28 #include <rte_eal.h>
  29 #include <rte_dev.h>
  30 #include <rte_cycles.h>
  31 #include <rte_kvargs.h>
  32
  33 #include "virtio_ethdev.h"
  34 #include "virtio_pci.h"
  35 #include "virtio_logs.h"
  36 #include "virtqueue.h"
  37 #include "virtio_rxtx.h"
  38 #include "virtio_user/virtio_user_dev.h"
  39
  40 static int eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev);
  41 static int  virtio_dev_configure(struct rte_eth_dev *dev);
  42 static int  virtio_dev_start(struct rte_eth_dev *dev);
  43 static void virtio_dev_stop(struct rte_eth_dev *dev);
  44 static int virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
  45 static int virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
  46 static int virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
  47 static int virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
  48 static uint32_t virtio_dev_speed_capa_get(uint32_t speed);
  49 static int virtio_dev_devargs_parse(struct rte_devargs *devargs,
  50         int *vdpa,
  51         uint32_t *speed,
  52         int *vectorized);
  53 static int virtio_dev_info_get(struct rte_eth_dev *dev,
  54                                 struct rte_eth_dev_info *dev_info);
  55 static int virtio_dev_link_update(struct rte_eth_dev *dev,
  56         int wait_to_complete);
  57 static int virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask);
  58
  59 static void virtio_set_hwaddr(struct virtio_hw *hw);
  60 static void virtio_get_hwaddr(struct virtio_hw *hw);
  61
  62 static int virtio_dev_stats_get(struct rte_eth_dev *dev,
  63                                  struct rte_eth_stats *stats);
  64 static int virtio_dev_xstats_get(struct rte_eth_dev *dev,
  65                                  struct rte_eth_xstat *xstats, unsigned n);
  66 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
  67                                        struct rte_eth_xstat_name *xstats_names,
  68                                        unsigned limit);
  69 static int virtio_dev_stats_reset(struct rte_eth_dev *dev);
  70 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
  71 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
  72                                 uint16_t vlan_id, int on);
  73 static int virtio_mac_addr_add(struct rte_eth_dev *dev,
  74                                 struct rte_ether_addr *mac_addr,
  75                                 uint32_t index, uint32_t vmdq);
  76 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
  77 static int virtio_mac_addr_set(struct rte_eth_dev *dev,
  78                                 struct rte_ether_addr *mac_addr);
  79
  80 static int virtio_intr_disable(struct rte_eth_dev *dev);
  81
  82 static int virtio_dev_queue_stats_mapping_set(
  83         struct rte_eth_dev *eth_dev,
  84         uint16_t queue_id,
  85         uint8_t stat_idx,
  86         uint8_t is_rx);
  87
  88 static void virtio_notify_peers(struct rte_eth_dev *dev);
  89 static void virtio_ack_link_announce(struct rte_eth_dev *dev);
  90
  91 /*
  92  * The set of PCI devices this driver supports
  93  */
  94 static const struct rte_pci_id pci_id_virtio_map[] = {
  95         { RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_LEGACY_DEVICEID_NET) },
  96         { RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_MODERN_DEVICEID_NET) },
  97         { .vendor_id = 0, /* sentinel */ },
  98 };
  99
 100 struct rte_virtio_xstats_name_off {
 101         char name[RTE_ETH_XSTATS_NAME_SIZE];
 102         unsigned offset;
 103 };
 104
 105 /* [rt]x_qX_ is prepended to the name string here */
 106 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
 107         {"good_packets",           offsetof(struct virtnet_rx, stats.packets)},
 108         {"good_bytes",             offsetof(struct virtnet_rx, stats.bytes)},
 109         {"errors",                 offsetof(struct virtnet_rx, stats.errors)},
 110         {"multicast_packets",      offsetof(struct virtnet_rx, stats.multicast)},
 111         {"broadcast_packets",      offsetof(struct virtnet_rx, stats.broadcast)},
 112         {"undersize_packets",      offsetof(struct virtnet_rx, stats.size_bins[0])},
 113         {"size_64_packets",        offsetof(struct virtnet_rx, stats.size_bins[1])},
 114         {"size_65_127_packets",    offsetof(struct virtnet_rx, stats.size_bins[2])},
 115         {"size_128_255_packets",   offsetof(struct virtnet_rx, stats.size_bins[3])},
 116         {"size_256_511_packets",   offsetof(struct virtnet_rx, stats.size_bins[4])},
 117         {"size_512_1023_packets",  offsetof(struct virtnet_rx, stats.size_bins[5])},
 118         {"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
 119         {"size_1519_max_packets",  offsetof(struct virtnet_rx, stats.size_bins[7])},
 120 };
 121
 122 /* [rt]x_qX_ is prepended to the name string here */
 123 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
 124         {"good_packets",           offsetof(struct virtnet_tx, stats.packets)},
 125         {"good_bytes",             offsetof(struct virtnet_tx, stats.bytes)},
 126         {"multicast_packets",      offsetof(struct virtnet_tx, stats.multicast)},
 127         {"broadcast_packets",      offsetof(struct virtnet_tx, stats.broadcast)},
 128         {"undersize_packets",      offsetof(struct virtnet_tx, stats.size_bins[0])},
 129         {"size_64_packets",        offsetof(struct virtnet_tx, stats.size_bins[1])},
 130         {"size_65_127_packets",    offsetof(struct virtnet_tx, stats.size_bins[2])},
 131         {"size_128_255_packets",   offsetof(struct virtnet_tx, stats.size_bins[3])},
 132         {"size_256_511_packets",   offsetof(struct virtnet_tx, stats.size_bins[4])},
 133         {"size_512_1023_packets",  offsetof(struct virtnet_tx, stats.size_bins[5])},
 134         {"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
 135         {"size_1519_max_packets",  offsetof(struct virtnet_tx, stats.size_bins[7])},
 136 };
 137
 138 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
 139                             sizeof(rte_virtio_rxq_stat_strings[0]))
 140 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
 141                             sizeof(rte_virtio_txq_stat_strings[0]))
 142
 143 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
 144
 145 static struct virtio_pmd_ctrl *
 146 virtio_send_command_packed(struct virtnet_ctl *cvq,
 147                            struct virtio_pmd_ctrl *ctrl,
 148                            int *dlen, int pkt_num)
 149 {
 150         struct virtqueue *vq = cvq->vq;
 151         int head;
 152         struct vring_packed_desc *desc = vq->vq_packed.ring.desc;
 153         struct virtio_pmd_ctrl *result;
 154         uint16_t flags;
 155         int sum = 0;
 156         int nb_descs = 0;
 157         int k;
 158
 159         /*
 160          * Format is enforced in qemu code:
 161          * One TX packet for header;
 162          * At least one TX packet per argument;
 163          * One RX packet for ACK.
 164          */
 165         head = vq->vq_avail_idx;
 166         flags = vq->vq_packed.cached_flags;
 167         desc[head].addr = cvq->virtio_net_hdr_mem;
 168         desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
 169         vq->vq_free_cnt--;
 170         nb_descs++;
 171         if (++vq->vq_avail_idx >= vq->vq_nentries) {
 172                 vq->vq_avail_idx -= vq->vq_nentries;
 173                 vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
 174         }
 175
 176         for (k = 0; k < pkt_num; k++) {
 177                 desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
 178                         + sizeof(struct virtio_net_ctrl_hdr)
 179                         + sizeof(ctrl->status) + sizeof(uint8_t) * sum;
 180                 desc[vq->vq_avail_idx].len = dlen[k];
 181                 desc[vq->vq_avail_idx].flags = VRING_DESC_F_NEXT |
 182                         vq->vq_packed.cached_flags;
 183                 sum += dlen[k];
 184                 vq->vq_free_cnt--;
 185                 nb_descs++;
 186                 if (++vq->vq_avail_idx >= vq->vq_nentries) {
 187                         vq->vq_avail_idx -= vq->vq_nentries;
 188                         vq->vq_packed.cached_flags ^=
 189                                 VRING_PACKED_DESC_F_AVAIL_USED;
 190                 }
 191         }
 192
 193         desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
 194                 + sizeof(struct virtio_net_ctrl_hdr);
 195         desc[vq->vq_avail_idx].len = sizeof(ctrl->status);
 196         desc[vq->vq_avail_idx].flags = VRING_DESC_F_WRITE |
 197                 vq->vq_packed.cached_flags;
 198         vq->vq_free_cnt--;
 199         nb_descs++;
 200         if (++vq->vq_avail_idx >= vq->vq_nentries) {
 201                 vq->vq_avail_idx -= vq->vq_nentries;
 202                 vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
 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 struct virtio_pmd_ctrl *
 241 virtio_send_command_split(struct virtnet_ctl *cvq,
 242                           struct virtio_pmd_ctrl *ctrl,
 243                           int *dlen, int pkt_num)
 244 {
 245         struct virtio_pmd_ctrl *result;
 246         struct virtqueue *vq = cvq->vq;
 247         uint32_t head, i;
 248         int k, sum = 0;
 249
 250         head = vq->vq_desc_head_idx;
 251
 252         /*
 253          * Format is enforced in qemu code:
 254          * One TX packet for header;
 255          * At least one TX packet per argument;
 256          * One RX packet for ACK.
 257          */
 258         vq->vq_split.ring.desc[head].flags = VRING_DESC_F_NEXT;
 259         vq->vq_split.ring.desc[head].addr = cvq->virtio_net_hdr_mem;
 260         vq->vq_split.ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
 261         vq->vq_free_cnt--;
 262         i = vq->vq_split.ring.desc[head].next;
 263
 264         for (k = 0; k < pkt_num; k++) {
 265                 vq->vq_split.ring.desc[i].flags = VRING_DESC_F_NEXT;
 266                 vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
 267                         + sizeof(struct virtio_net_ctrl_hdr)
 268                         + sizeof(ctrl->status) + sizeof(uint8_t)*sum;
 269                 vq->vq_split.ring.desc[i].len = dlen[k];
 270                 sum += dlen[k];
 271                 vq->vq_free_cnt--;
 272                 i = vq->vq_split.ring.desc[i].next;
 273         }
 274
 275         vq->vq_split.ring.desc[i].flags = VRING_DESC_F_WRITE;
 276         vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
 277                         + sizeof(struct virtio_net_ctrl_hdr);
 278         vq->vq_split.ring.desc[i].len = sizeof(ctrl->status);
 279         vq->vq_free_cnt--;
 280
 281         vq->vq_desc_head_idx = vq->vq_split.ring.desc[i].next;
 282
 283         vq_update_avail_ring(vq, head);
 284         vq_update_avail_idx(vq);
 285
 286         PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
 287
 288         virtqueue_notify(vq);
 289
 290         while (virtqueue_nused(vq) == 0)
 291                 usleep(100);
 292
 293         while (virtqueue_nused(vq)) {
 294                 uint32_t idx, desc_idx, used_idx;
 295                 struct vring_used_elem *uep;
 296
 297                 used_idx = (uint32_t)(vq->vq_used_cons_idx
 298                                 & (vq->vq_nentries - 1));
 299                 uep = &vq->vq_split.ring.used->ring[used_idx];
 300                 idx = (uint32_t) uep->id;
 301                 desc_idx = idx;
 302
 303                 while (vq->vq_split.ring.desc[desc_idx].flags &
 304                                 VRING_DESC_F_NEXT) {
 305                         desc_idx = vq->vq_split.ring.desc[desc_idx].next;
 306                         vq->vq_free_cnt++;
 307                 }
 308
 309                 vq->vq_split.ring.desc[desc_idx].next = vq->vq_desc_head_idx;
 310                 vq->vq_desc_head_idx = idx;
 311
 312                 vq->vq_used_cons_idx++;
 313                 vq->vq_free_cnt++;
 314         }
 315
 316         PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
 317                         vq->vq_free_cnt, vq->vq_desc_head_idx);
 318
 319         result = cvq->virtio_net_hdr_mz->addr;
 320         return result;
 321 }
 322
 323 static int
 324 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
 325                     int *dlen, int pkt_num)
 326 {
 327         virtio_net_ctrl_ack status = ~0;
 328         struct virtio_pmd_ctrl *result;
 329         struct virtqueue *vq;
 330
 331         ctrl->status = status;
 332
 333         if (!cvq || !cvq->vq) {
 334                 PMD_INIT_LOG(ERR, "Control queue is not supported.");
 335                 return -1;
 336         }
 337
 338         rte_spinlock_lock(&cvq->lock);
 339         vq = cvq->vq;
 340
 341         PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
 342                 "vq->hw->cvq = %p vq = %p",
 343                 vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
 344
 345         if (vq->vq_free_cnt < pkt_num + 2 || pkt_num < 1) {
 346                 rte_spinlock_unlock(&cvq->lock);
 347                 return -1;
 348         }
 349
 350         memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
 351                 sizeof(struct virtio_pmd_ctrl));
 352
 353         if (vtpci_packed_queue(vq->hw))
 354                 result = virtio_send_command_packed(cvq, ctrl, dlen, pkt_num);
 355         else
 356                 result = virtio_send_command_split(cvq, ctrl, dlen, pkt_num);
 357
 358         rte_spinlock_unlock(&cvq->lock);
 359         return result->status;
 360 }
 361
 362 static int
 363 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
 364 {
 365         struct virtio_hw *hw = dev->data->dev_private;
 366         struct virtio_pmd_ctrl ctrl;
 367         int dlen[1];
 368         int ret;
 369
 370         ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
 371         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
 372         memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
 373
 374         dlen[0] = sizeof(uint16_t);
 375
 376         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 377         if (ret) {
 378                 PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
 379                           "failed, this is too late now...");
 380                 return -EINVAL;
 381         }
 382
 383         return 0;
 384 }
 385
 386 static void
 387 virtio_dev_queue_release(void *queue __rte_unused)
 388 {
 389         /* do nothing */
 390 }
 391
 392 static uint16_t
 393 virtio_get_nr_vq(struct virtio_hw *hw)
 394 {
 395         uint16_t nr_vq = hw->max_queue_pairs * 2;
 396
 397         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
 398                 nr_vq += 1;
 399
 400         return nr_vq;
 401 }
 402
 403 static void
 404 virtio_init_vring(struct virtqueue *vq)
 405 {
 406         int size = vq->vq_nentries;
 407         uint8_t *ring_mem = vq->vq_ring_virt_mem;
 408
 409         PMD_INIT_FUNC_TRACE();
 410
 411         memset(ring_mem, 0, vq->vq_ring_size);
 412
 413         vq->vq_used_cons_idx = 0;
 414         vq->vq_desc_head_idx = 0;
 415         vq->vq_avail_idx = 0;
 416         vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
 417         vq->vq_free_cnt = vq->vq_nentries;
 418         memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
 419         if (vtpci_packed_queue(vq->hw)) {
 420                 vring_init_packed(&vq->vq_packed.ring, ring_mem,
 421                                   VIRTIO_PCI_VRING_ALIGN, size);
 422                 vring_desc_init_packed(vq, size);
 423         } else {
 424                 struct vring *vr = &vq->vq_split.ring;
 425
 426                 vring_init_split(vr, ring_mem, VIRTIO_PCI_VRING_ALIGN, size);
 427                 vring_desc_init_split(vr->desc, size);
 428         }
 429         /*
 430          * Disable device(host) interrupting guest
 431          */
 432         virtqueue_disable_intr(vq);
 433 }
 434
 435 static int
 436 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
 437 {
 438         char vq_name[VIRTQUEUE_MAX_NAME_SZ];
 439         char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
 440         const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
 441         unsigned int vq_size, size;
 442         struct virtio_hw *hw = dev->data->dev_private;
 443         struct virtnet_rx *rxvq = NULL;
 444         struct virtnet_tx *txvq = NULL;
 445         struct virtnet_ctl *cvq = NULL;
 446         struct virtqueue *vq;
 447         size_t sz_hdr_mz = 0;
 448         void *sw_ring = NULL;
 449         int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
 450         int ret;
 451         int numa_node = dev->device->numa_node;
 452
 453         PMD_INIT_LOG(INFO, "setting up queue: %u on NUMA node %d",
 454                         vtpci_queue_idx, numa_node);
 455
 456         /*
 457          * Read the virtqueue size from the Queue Size field
 458          * Always power of 2 and if 0 virtqueue does not exist
 459          */
 460         vq_size = VTPCI_OPS(hw)->get_queue_num(hw, vtpci_queue_idx);
 461         PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
 462         if (vq_size == 0) {
 463                 PMD_INIT_LOG(ERR, "virtqueue does not exist");
 464                 return -EINVAL;
 465         }
 466
 467         if (!vtpci_packed_queue(hw) && !rte_is_power_of_2(vq_size)) {
 468                 PMD_INIT_LOG(ERR, "split virtqueue size is not power of 2");
 469                 return -EINVAL;
 470         }
 471
 472         snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
 473                  dev->data->port_id, vtpci_queue_idx);
 474
 475         size = RTE_ALIGN_CEIL(sizeof(*vq) +
 476                                 vq_size * sizeof(struct vq_desc_extra),
 477                                 RTE_CACHE_LINE_SIZE);
 478         if (queue_type == VTNET_TQ) {
 479                 /*
 480                  * For each xmit packet, allocate a virtio_net_hdr
 481                  * and indirect ring elements
 482                  */
 483                 sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
 484         } else if (queue_type == VTNET_CQ) {
 485                 /* Allocate a page for control vq command, data and status */
 486                 sz_hdr_mz = PAGE_SIZE;
 487         }
 488
 489         vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
 490                                 numa_node);
 491         if (vq == NULL) {
 492                 PMD_INIT_LOG(ERR, "can not allocate vq");
 493                 return -ENOMEM;
 494         }
 495         hw->vqs[vtpci_queue_idx] = vq;
 496
 497         vq->hw = hw;
 498         vq->vq_queue_index = vtpci_queue_idx;
 499         vq->vq_nentries = vq_size;
 500         if (vtpci_packed_queue(hw)) {
 501                 vq->vq_packed.used_wrap_counter = 1;
 502                 vq->vq_packed.cached_flags = VRING_PACKED_DESC_F_AVAIL;
 503                 vq->vq_packed.event_flags_shadow = 0;
 504                 if (queue_type == VTNET_RQ)
 505                         vq->vq_packed.cached_flags |= VRING_DESC_F_WRITE;
 506         }
 507
 508         /*
 509          * Reserve a memzone for vring elements
 510          */
 511         size = vring_size(hw, vq_size, VIRTIO_PCI_VRING_ALIGN);
 512         vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
 513         PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
 514                      size, vq->vq_ring_size);
 515
 516         mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
 517                         numa_node, RTE_MEMZONE_IOVA_CONTIG,
 518                         VIRTIO_PCI_VRING_ALIGN);
 519         if (mz == NULL) {
 520                 if (rte_errno == EEXIST)
 521                         mz = rte_memzone_lookup(vq_name);
 522                 if (mz == NULL) {
 523                         ret = -ENOMEM;
 524                         goto fail_q_alloc;
 525                 }
 526         }
 527
 528         memset(mz->addr, 0, mz->len);
 529
 530         vq->vq_ring_mem = mz->iova;
 531         vq->vq_ring_virt_mem = mz->addr;
 532         PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%" PRIx64,
 533                      (uint64_t)mz->iova);
 534         PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
 535                      (uint64_t)(uintptr_t)mz->addr);
 536
 537         virtio_init_vring(vq);
 538
 539         if (sz_hdr_mz) {
 540                 snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
 541                          dev->data->port_id, vtpci_queue_idx);
 542                 hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
 543                                 numa_node, RTE_MEMZONE_IOVA_CONTIG,
 544                                 RTE_CACHE_LINE_SIZE);
 545                 if (hdr_mz == NULL) {
 546                         if (rte_errno == EEXIST)
 547                                 hdr_mz = rte_memzone_lookup(vq_hdr_name);
 548                         if (hdr_mz == NULL) {
 549                                 ret = -ENOMEM;
 550                                 goto fail_q_alloc;
 551                         }
 552                 }
 553         }
 554
 555         if (queue_type == VTNET_RQ) {
 556                 size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
 557                                sizeof(vq->sw_ring[0]);
 558
 559                 sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
 560                                 RTE_CACHE_LINE_SIZE, numa_node);
 561                 if (!sw_ring) {
 562                         PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
 563                         ret = -ENOMEM;
 564                         goto fail_q_alloc;
 565                 }
 566
 567                 vq->sw_ring = sw_ring;
 568                 rxvq = &vq->rxq;
 569                 rxvq->vq = vq;
 570                 rxvq->port_id = dev->data->port_id;
 571                 rxvq->mz = mz;
 572         } else if (queue_type == VTNET_TQ) {
 573                 txvq = &vq->txq;
 574                 txvq->vq = vq;
 575                 txvq->port_id = dev->data->port_id;
 576                 txvq->mz = mz;
 577                 txvq->virtio_net_hdr_mz = hdr_mz;
 578                 txvq->virtio_net_hdr_mem = hdr_mz->iova;
 579         } else if (queue_type == VTNET_CQ) {
 580                 cvq = &vq->cq;
 581                 cvq->vq = vq;
 582                 cvq->mz = mz;
 583                 cvq->virtio_net_hdr_mz = hdr_mz;
 584                 cvq->virtio_net_hdr_mem = hdr_mz->iova;
 585                 memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
 586
 587                 hw->cvq = cvq;
 588         }
 589
 590         /* For virtio_user case (that is when hw->virtio_user_dev is not NULL),
 591          * we use virtual address. And we need properly set _offset_, please see
 592          * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
 593          */
 594         if (!hw->virtio_user_dev)
 595                 vq->offset = offsetof(struct rte_mbuf, buf_iova);
 596         else {
 597                 vq->vq_ring_mem = (uintptr_t)mz->addr;
 598                 vq->offset = offsetof(struct rte_mbuf, buf_addr);
 599                 if (queue_type == VTNET_TQ)
 600                         txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
 601                 else if (queue_type == VTNET_CQ)
 602                         cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
 603         }
 604
 605         if (queue_type == VTNET_TQ) {
 606                 struct virtio_tx_region *txr;
 607                 unsigned int i;
 608
 609                 txr = hdr_mz->addr;
 610                 memset(txr, 0, vq_size * sizeof(*txr));
 611                 for (i = 0; i < vq_size; i++) {
 612                         struct vring_desc *start_dp = txr[i].tx_indir;
 613
 614                         /* first indirect descriptor is always the tx header */
 615                         if (!vtpci_packed_queue(hw)) {
 616                                 vring_desc_init_split(start_dp,
 617                                                       RTE_DIM(txr[i].tx_indir));
 618                                 start_dp->addr = txvq->virtio_net_hdr_mem
 619                                         + i * sizeof(*txr)
 620                                         + offsetof(struct virtio_tx_region,
 621                                                    tx_hdr);
 622                                 start_dp->len = hw->vtnet_hdr_size;
 623                                 start_dp->flags = VRING_DESC_F_NEXT;
 624                         }
 625                 }
 626         }
 627
 628         if (VTPCI_OPS(hw)->setup_queue(hw, vq) < 0) {
 629                 PMD_INIT_LOG(ERR, "setup_queue failed");
 630                 return -EINVAL;
 631         }
 632
 633         return 0;
 634
 635 fail_q_alloc:
 636         rte_free(sw_ring);
 637         rte_memzone_free(hdr_mz);
 638         rte_memzone_free(mz);
 639         rte_free(vq);
 640
 641         return ret;
 642 }
 643
 644 static void
 645 virtio_free_queues(struct virtio_hw *hw)
 646 {
 647         uint16_t nr_vq = virtio_get_nr_vq(hw);
 648         struct virtqueue *vq;
 649         int queue_type;
 650         uint16_t i;
 651
 652         if (hw->vqs == NULL)
 653                 return;
 654
 655         for (i = 0; i < nr_vq; i++) {
 656                 vq = hw->vqs[i];
 657                 if (!vq)
 658                         continue;
 659
 660                 queue_type = virtio_get_queue_type(hw, i);
 661                 if (queue_type == VTNET_RQ) {
 662                         rte_free(vq->sw_ring);
 663                         rte_memzone_free(vq->rxq.mz);
 664                 } else if (queue_type == VTNET_TQ) {
 665                         rte_memzone_free(vq->txq.mz);
 666                         rte_memzone_free(vq->txq.virtio_net_hdr_mz);
 667                 } else {
 668                         rte_memzone_free(vq->cq.mz);
 669                         rte_memzone_free(vq->cq.virtio_net_hdr_mz);
 670                 }
 671
 672                 rte_free(vq);
 673                 hw->vqs[i] = NULL;
 674         }
 675
 676         rte_free(hw->vqs);
 677         hw->vqs = NULL;
 678 }
 679
 680 static int
 681 virtio_alloc_queues(struct rte_eth_dev *dev)
 682 {
 683         struct virtio_hw *hw = dev->data->dev_private;
 684         uint16_t nr_vq = virtio_get_nr_vq(hw);
 685         uint16_t i;
 686         int ret;
 687
 688         hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
 689         if (!hw->vqs) {
 690                 PMD_INIT_LOG(ERR, "failed to allocate vqs");
 691                 return -ENOMEM;
 692         }
 693
 694         for (i = 0; i < nr_vq; i++) {
 695                 ret = virtio_init_queue(dev, i);
 696                 if (ret < 0) {
 697                         virtio_free_queues(hw);
 698                         return ret;
 699                 }
 700         }
 701
 702         return 0;
 703 }
 704
 705 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
 706
 707 static void
 708 virtio_dev_close(struct rte_eth_dev *dev)
 709 {
 710         struct virtio_hw *hw = dev->data->dev_private;
 711         struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
 712
 713         PMD_INIT_LOG(DEBUG, "virtio_dev_close");
 714
 715         if (!hw->opened)
 716                 return;
 717         hw->opened = false;
 718
 719         /* reset the NIC */
 720         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
 721                 VTPCI_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
 722         if (intr_conf->rxq)
 723                 virtio_queues_unbind_intr(dev);
 724
 725         if (intr_conf->lsc || intr_conf->rxq) {
 726                 virtio_intr_disable(dev);
 727                 rte_intr_efd_disable(dev->intr_handle);
 728                 rte_free(dev->intr_handle->intr_vec);
 729                 dev->intr_handle->intr_vec = NULL;
 730         }
 731
 732         vtpci_reset(hw);
 733         virtio_dev_free_mbufs(dev);
 734         virtio_free_queues(hw);
 735
 736 #ifdef RTE_VIRTIO_USER
 737         if (hw->virtio_user_dev)
 738                 virtio_user_dev_uninit(hw->virtio_user_dev);
 739         else
 740 #endif
 741         if (dev->device) {
 742                 rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(dev));
 743                 if (!hw->modern)
 744                         rte_pci_ioport_unmap(VTPCI_IO(hw));
 745         }
 746 }
 747
 748 static int
 749 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
 750 {
 751         struct virtio_hw *hw = dev->data->dev_private;
 752         struct virtio_pmd_ctrl ctrl;
 753         int dlen[1];
 754         int ret;
 755
 756         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 757                 PMD_INIT_LOG(INFO, "host does not support rx control");
 758                 return -ENOTSUP;
 759         }
 760
 761         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 762         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 763         ctrl.data[0] = 1;
 764         dlen[0] = 1;
 765
 766         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 767         if (ret) {
 768                 PMD_INIT_LOG(ERR, "Failed to enable promisc");
 769                 return -EAGAIN;
 770         }
 771
 772         return 0;
 773 }
 774
 775 static int
 776 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
 777 {
 778         struct virtio_hw *hw = dev->data->dev_private;
 779         struct virtio_pmd_ctrl ctrl;
 780         int dlen[1];
 781         int ret;
 782
 783         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 784                 PMD_INIT_LOG(INFO, "host does not support rx control");
 785                 return -ENOTSUP;
 786         }
 787
 788         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 789         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 790         ctrl.data[0] = 0;
 791         dlen[0] = 1;
 792
 793         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 794         if (ret) {
 795                 PMD_INIT_LOG(ERR, "Failed to disable promisc");
 796                 return -EAGAIN;
 797         }
 798
 799         return 0;
 800 }
 801
 802 static int
 803 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
 804 {
 805         struct virtio_hw *hw = dev->data->dev_private;
 806         struct virtio_pmd_ctrl ctrl;
 807         int dlen[1];
 808         int ret;
 809
 810         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 811                 PMD_INIT_LOG(INFO, "host does not support rx control");
 812                 return -ENOTSUP;
 813         }
 814
 815         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 816         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 817         ctrl.data[0] = 1;
 818         dlen[0] = 1;
 819
 820         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 821         if (ret) {
 822                 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
 823                 return -EAGAIN;
 824         }
 825
 826         return 0;
 827 }
 828
 829 static int
 830 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
 831 {
 832         struct virtio_hw *hw = dev->data->dev_private;
 833         struct virtio_pmd_ctrl ctrl;
 834         int dlen[1];
 835         int ret;
 836
 837         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 838                 PMD_INIT_LOG(INFO, "host does not support rx control");
 839                 return -ENOTSUP;
 840         }
 841
 842         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 843         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 844         ctrl.data[0] = 0;
 845         dlen[0] = 1;
 846
 847         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 848         if (ret) {
 849                 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
 850                 return -EAGAIN;
 851         }
 852
 853         return 0;
 854 }
 855
 856 #define VLAN_TAG_LEN           4    /* 802.3ac tag (not DMA'd) */
 857 static int
 858 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 859 {
 860         struct virtio_hw *hw = dev->data->dev_private;
 861         uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
 862                                  hw->vtnet_hdr_size;
 863         uint32_t frame_size = mtu + ether_hdr_len;
 864         uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
 865
 866         max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
 867
 868         if (mtu < RTE_ETHER_MIN_MTU || frame_size > max_frame_size) {
 869                 PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
 870                         RTE_ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
 871                 return -EINVAL;
 872         }
 873         return 0;
 874 }
 875
 876 static int
 877 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
 878 {
 879         struct virtio_hw *hw = dev->data->dev_private;
 880         struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
 881         struct virtqueue *vq = rxvq->vq;
 882
 883         virtqueue_enable_intr(vq);
 884         virtio_mb(hw->weak_barriers);
 885         return 0;
 886 }
 887
 888 static int
 889 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
 890 {
 891         struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
 892         struct virtqueue *vq = rxvq->vq;
 893
 894         virtqueue_disable_intr(vq);
 895         return 0;
 896 }
 897
 898 /*
 899  * dev_ops for virtio, bare necessities for basic operation
 900  */
 901 static const struct eth_dev_ops virtio_eth_dev_ops = {
 902         .dev_configure           = virtio_dev_configure,
 903         .dev_start               = virtio_dev_start,
 904         .dev_stop                = virtio_dev_stop,
 905         .dev_close               = virtio_dev_close,
 906         .promiscuous_enable      = virtio_dev_promiscuous_enable,
 907         .promiscuous_disable     = virtio_dev_promiscuous_disable,
 908         .allmulticast_enable     = virtio_dev_allmulticast_enable,
 909         .allmulticast_disable    = virtio_dev_allmulticast_disable,
 910         .mtu_set                 = virtio_mtu_set,
 911         .dev_infos_get           = virtio_dev_info_get,
 912         .stats_get               = virtio_dev_stats_get,
 913         .xstats_get              = virtio_dev_xstats_get,
 914         .xstats_get_names        = virtio_dev_xstats_get_names,
 915         .stats_reset             = virtio_dev_stats_reset,
 916         .xstats_reset            = virtio_dev_stats_reset,
 917         .link_update             = virtio_dev_link_update,
 918         .vlan_offload_set        = virtio_dev_vlan_offload_set,
 919         .rx_queue_setup          = virtio_dev_rx_queue_setup,
 920         .rx_queue_intr_enable    = virtio_dev_rx_queue_intr_enable,
 921         .rx_queue_intr_disable   = virtio_dev_rx_queue_intr_disable,
 922         .rx_queue_release        = virtio_dev_queue_release,
 923         .tx_queue_setup          = virtio_dev_tx_queue_setup,
 924         .tx_queue_release        = virtio_dev_queue_release,
 925         /* collect stats per queue */
 926         .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
 927         .vlan_filter_set         = virtio_vlan_filter_set,
 928         .mac_addr_add            = virtio_mac_addr_add,
 929         .mac_addr_remove         = virtio_mac_addr_remove,
 930         .mac_addr_set            = virtio_mac_addr_set,
 931 };
 932
 933 /*
 934  * dev_ops for virtio-user in secondary processes, as we just have
 935  * some limited supports currently.
 936  */
 937 const struct eth_dev_ops virtio_user_secondary_eth_dev_ops = {
 938         .dev_infos_get           = virtio_dev_info_get,
 939         .stats_get               = virtio_dev_stats_get,
 940         .xstats_get              = virtio_dev_xstats_get,
 941         .xstats_get_names        = virtio_dev_xstats_get_names,
 942         .stats_reset             = virtio_dev_stats_reset,
 943         .xstats_reset            = virtio_dev_stats_reset,
 944         /* collect stats per queue */
 945         .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
 946 };
 947
 948 static void
 949 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 950 {
 951         unsigned i;
 952
 953         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 954                 const struct virtnet_tx *txvq = dev->data->tx_queues[i];
 955                 if (txvq == NULL)
 956                         continue;
 957
 958                 stats->opackets += txvq->stats.packets;
 959                 stats->obytes += txvq->stats.bytes;
 960
 961                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 962                         stats->q_opackets[i] = txvq->stats.packets;
 963                         stats->q_obytes[i] = txvq->stats.bytes;
 964                 }
 965         }
 966
 967         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 968                 const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
 969                 if (rxvq == NULL)
 970                         continue;
 971
 972                 stats->ipackets += rxvq->stats.packets;
 973                 stats->ibytes += rxvq->stats.bytes;
 974                 stats->ierrors += rxvq->stats.errors;
 975
 976                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 977                         stats->q_ipackets[i] = rxvq->stats.packets;
 978                         stats->q_ibytes[i] = rxvq->stats.bytes;
 979                 }
 980         }
 981
 982         stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
 983 }
 984
 985 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
 986                                        struct rte_eth_xstat_name *xstats_names,
 987                                        __rte_unused unsigned limit)
 988 {
 989         unsigned i;
 990         unsigned count = 0;
 991         unsigned t;
 992
 993         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
 994                 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
 995
 996         if (xstats_names != NULL) {
 997                 /* Note: limit checked in rte_eth_xstats_names() */
 998
 999                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1000                         struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1001                         if (rxvq == NULL)
1002                                 continue;
1003                         for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1004                                 snprintf(xstats_names[count].name,
1005                                         sizeof(xstats_names[count].name),
1006                                         "rx_q%u_%s", i,
1007                                         rte_virtio_rxq_stat_strings[t].name);
1008                                 count++;
1009                         }
1010                 }
1011
1012                 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1013                         struct virtnet_tx *txvq = dev->data->tx_queues[i];
1014                         if (txvq == NULL)
1015                                 continue;
1016                         for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1017                                 snprintf(xstats_names[count].name,
1018                                         sizeof(xstats_names[count].name),
1019                                         "tx_q%u_%s", i,
1020                                         rte_virtio_txq_stat_strings[t].name);
1021                                 count++;
1022                         }
1023                 }
1024                 return count;
1025         }
1026         return nstats;
1027 }
1028
1029 static int
1030 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1031                       unsigned n)
1032 {
1033         unsigned i;
1034         unsigned count = 0;
1035
1036         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1037                 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1038
1039         if (n < nstats)
1040                 return nstats;
1041
1042         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1043                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1044
1045                 if (rxvq == NULL)
1046                         continue;
1047
1048                 unsigned t;
1049
1050                 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1051                         xstats[count].value = *(uint64_t *)(((char *)rxvq) +
1052                                 rte_virtio_rxq_stat_strings[t].offset);
1053                         xstats[count].id = count;
1054                         count++;
1055                 }
1056         }
1057
1058         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1059                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1060
1061                 if (txvq == NULL)
1062                         continue;
1063
1064                 unsigned t;
1065
1066                 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1067                         xstats[count].value = *(uint64_t *)(((char *)txvq) +
1068                                 rte_virtio_txq_stat_strings[t].offset);
1069                         xstats[count].id = count;
1070                         count++;
1071                 }
1072         }
1073
1074         return count;
1075 }
1076
1077 static int
1078 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1079 {
1080         virtio_update_stats(dev, stats);
1081
1082         return 0;
1083 }
1084
1085 static int
1086 virtio_dev_stats_reset(struct rte_eth_dev *dev)
1087 {
1088         unsigned int i;
1089
1090         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1091                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1092                 if (txvq == NULL)
1093                         continue;
1094
1095                 txvq->stats.packets = 0;
1096                 txvq->stats.bytes = 0;
1097                 txvq->stats.multicast = 0;
1098                 txvq->stats.broadcast = 0;
1099                 memset(txvq->stats.size_bins, 0,
1100                        sizeof(txvq->stats.size_bins[0]) * 8);
1101         }
1102
1103         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1104                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1105                 if (rxvq == NULL)
1106                         continue;
1107
1108                 rxvq->stats.packets = 0;
1109                 rxvq->stats.bytes = 0;
1110                 rxvq->stats.errors = 0;
1111                 rxvq->stats.multicast = 0;
1112                 rxvq->stats.broadcast = 0;
1113                 memset(rxvq->stats.size_bins, 0,
1114                        sizeof(rxvq->stats.size_bins[0]) * 8);
1115         }
1116
1117         return 0;
1118 }
1119
1120 static void
1121 virtio_set_hwaddr(struct virtio_hw *hw)
1122 {
1123         vtpci_write_dev_config(hw,
1124                         offsetof(struct virtio_net_config, mac),
1125                         &hw->mac_addr, RTE_ETHER_ADDR_LEN);
1126 }
1127
1128 static void
1129 virtio_get_hwaddr(struct virtio_hw *hw)
1130 {
1131         if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1132                 vtpci_read_dev_config(hw,
1133                         offsetof(struct virtio_net_config, mac),
1134                         &hw->mac_addr, RTE_ETHER_ADDR_LEN);
1135         } else {
1136                 rte_eth_random_addr(&hw->mac_addr[0]);
1137                 virtio_set_hwaddr(hw);
1138         }
1139 }
1140
1141 static int
1142 virtio_mac_table_set(struct virtio_hw *hw,
1143                      const struct virtio_net_ctrl_mac *uc,
1144                      const struct virtio_net_ctrl_mac *mc)
1145 {
1146         struct virtio_pmd_ctrl ctrl;
1147         int err, len[2];
1148
1149         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1150                 PMD_DRV_LOG(INFO, "host does not support mac table");
1151                 return -1;
1152         }
1153
1154         ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1155         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1156
1157         len[0] = uc->entries * RTE_ETHER_ADDR_LEN + sizeof(uc->entries);
1158         memcpy(ctrl.data, uc, len[0]);
1159
1160         len[1] = mc->entries * RTE_ETHER_ADDR_LEN + sizeof(mc->entries);
1161         memcpy(ctrl.data + len[0], mc, len[1]);
1162
1163         err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1164         if (err != 0)
1165                 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1166         return err;
1167 }
1168
1169 static int
1170 virtio_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1171                     uint32_t index, uint32_t vmdq __rte_unused)
1172 {
1173         struct virtio_hw *hw = dev->data->dev_private;
1174         const struct rte_ether_addr *addrs = dev->data->mac_addrs;
1175         unsigned int i;
1176         struct virtio_net_ctrl_mac *uc, *mc;
1177
1178         if (index >= VIRTIO_MAX_MAC_ADDRS) {
1179                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1180                 return -EINVAL;
1181         }
1182
1183         uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1184                 sizeof(uc->entries));
1185         uc->entries = 0;
1186         mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1187                 sizeof(mc->entries));
1188         mc->entries = 0;
1189
1190         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1191                 const struct rte_ether_addr *addr
1192                         = (i == index) ? mac_addr : addrs + i;
1193                 struct virtio_net_ctrl_mac *tbl
1194                         = rte_is_multicast_ether_addr(addr) ? mc : uc;
1195
1196                 memcpy(&tbl->macs[tbl->entries++], addr, RTE_ETHER_ADDR_LEN);
1197         }
1198
1199         return virtio_mac_table_set(hw, uc, mc);
1200 }
1201
1202 static void
1203 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1204 {
1205         struct virtio_hw *hw = dev->data->dev_private;
1206         struct rte_ether_addr *addrs = dev->data->mac_addrs;
1207         struct virtio_net_ctrl_mac *uc, *mc;
1208         unsigned int i;
1209
1210         if (index >= VIRTIO_MAX_MAC_ADDRS) {
1211                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1212                 return;
1213         }
1214
1215         uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1216                 sizeof(uc->entries));
1217         uc->entries = 0;
1218         mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1219                 sizeof(mc->entries));
1220         mc->entries = 0;
1221
1222         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1223                 struct virtio_net_ctrl_mac *tbl;
1224
1225                 if (i == index || rte_is_zero_ether_addr(addrs + i))
1226                         continue;
1227
1228                 tbl = rte_is_multicast_ether_addr(addrs + i) ? mc : uc;
1229                 memcpy(&tbl->macs[tbl->entries++], addrs + i,
1230                         RTE_ETHER_ADDR_LEN);
1231         }
1232
1233         virtio_mac_table_set(hw, uc, mc);
1234 }
1235
1236 static int
1237 virtio_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1238 {
1239         struct virtio_hw *hw = dev->data->dev_private;
1240
1241         memcpy(hw->mac_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1242
1243         /* Use atomic update if available */
1244         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1245                 struct virtio_pmd_ctrl ctrl;
1246                 int len = RTE_ETHER_ADDR_LEN;
1247
1248                 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1249                 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1250
1251                 memcpy(ctrl.data, mac_addr, RTE_ETHER_ADDR_LEN);
1252                 return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1253         }
1254
1255         if (!vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1256                 return -ENOTSUP;
1257
1258         virtio_set_hwaddr(hw);
1259         return 0;
1260 }
1261
1262 static int
1263 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1264 {
1265         struct virtio_hw *hw = dev->data->dev_private;
1266         struct virtio_pmd_ctrl ctrl;
1267         int len;
1268
1269         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1270                 return -ENOTSUP;
1271
1272         ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1273         ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1274         memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1275         len = sizeof(vlan_id);
1276
1277         return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1278 }
1279
1280 static int
1281 virtio_intr_unmask(struct rte_eth_dev *dev)
1282 {
1283         struct virtio_hw *hw = dev->data->dev_private;
1284
1285         if (rte_intr_ack(dev->intr_handle) < 0)
1286                 return -1;
1287
1288         if (!hw->virtio_user_dev)
1289                 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1290
1291         return 0;
1292 }
1293
1294 static int
1295 virtio_intr_enable(struct rte_eth_dev *dev)
1296 {
1297         struct virtio_hw *hw = dev->data->dev_private;
1298
1299         if (rte_intr_enable(dev->intr_handle) < 0)
1300                 return -1;
1301
1302         if (!hw->virtio_user_dev)
1303                 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1304
1305         return 0;
1306 }
1307
1308 static int
1309 virtio_intr_disable(struct rte_eth_dev *dev)
1310 {
1311         struct virtio_hw *hw = dev->data->dev_private;
1312
1313         if (rte_intr_disable(dev->intr_handle) < 0)
1314                 return -1;
1315
1316         if (!hw->virtio_user_dev)
1317                 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1318
1319         return 0;
1320 }
1321
1322 static int
1323 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1324 {
1325         uint64_t host_features;
1326
1327         /* Prepare guest_features: feature that driver wants to support */
1328         PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1329                 req_features);
1330
1331         /* Read device(host) feature bits */
1332         host_features = VTPCI_OPS(hw)->get_features(hw);
1333         PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1334                 host_features);
1335
1336         /* If supported, ensure MTU value is valid before acknowledging it. */
1337         if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1338                 struct virtio_net_config config;
1339
1340                 vtpci_read_dev_config(hw,
1341                         offsetof(struct virtio_net_config, mtu),
1342                         &config.mtu, sizeof(config.mtu));
1343
1344                 if (config.mtu < RTE_ETHER_MIN_MTU)
1345                         req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1346         }
1347
1348         /*
1349          * Negotiate features: Subset of device feature bits are written back
1350          * guest feature bits.
1351          */
1352         hw->guest_features = req_features;
1353         hw->guest_features = vtpci_negotiate_features(hw, host_features);
1354         PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1355                 hw->guest_features);
1356
1357         if (hw->modern && !vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1358                 PMD_INIT_LOG(ERR,
1359                         "VIRTIO_F_VERSION_1 features is not enabled.");
1360                 return -1;
1361         }
1362
1363         if (hw->modern || hw->virtio_user_dev) {
1364                 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1365                 if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1366                         PMD_INIT_LOG(ERR,
1367                                 "failed to set FEATURES_OK status!");
1368                         return -1;
1369                 }
1370         }
1371
1372         hw->req_guest_features = req_features;
1373
1374         return 0;
1375 }
1376
1377 int
1378 virtio_dev_pause(struct rte_eth_dev *dev)
1379 {
1380         struct virtio_hw *hw = dev->data->dev_private;
1381
1382         rte_spinlock_lock(&hw->state_lock);
1383
1384         if (hw->started == 0) {
1385                 /* Device is just stopped. */
1386                 rte_spinlock_unlock(&hw->state_lock);
1387                 return -1;
1388         }
1389         hw->started = 0;
1390         /*
1391          * Prevent the worker threads from touching queues to avoid contention,
1392          * 1 ms should be enough for the ongoing Tx function to finish.
1393          */
1394         rte_delay_ms(1);
1395         return 0;
1396 }
1397
1398 /*
1399  * Recover hw state to let the worker threads continue.
1400  */
1401 void
1402 virtio_dev_resume(struct rte_eth_dev *dev)
1403 {
1404         struct virtio_hw *hw = dev->data->dev_private;
1405
1406         hw->started = 1;
1407         rte_spinlock_unlock(&hw->state_lock);
1408 }
1409
1410 /*
1411  * Should be called only after device is paused.
1412  */
1413 int
1414 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1415                 int nb_pkts)
1416 {
1417         struct virtio_hw *hw = dev->data->dev_private;
1418         struct virtnet_tx *txvq = dev->data->tx_queues[0];
1419         int ret;
1420
1421         hw->inject_pkts = tx_pkts;
1422         ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1423         hw->inject_pkts = NULL;
1424
1425         return ret;
1426 }
1427
1428 static void
1429 virtio_notify_peers(struct rte_eth_dev *dev)
1430 {
1431         struct virtio_hw *hw = dev->data->dev_private;
1432         struct virtnet_rx *rxvq;
1433         struct rte_mbuf *rarp_mbuf;
1434
1435         if (!dev->data->rx_queues)
1436                 return;
1437
1438         rxvq = dev->data->rx_queues[0];
1439         if (!rxvq)
1440                 return;
1441
1442         rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1443                         (struct rte_ether_addr *)hw->mac_addr);
1444         if (rarp_mbuf == NULL) {
1445                 PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1446                 return;
1447         }
1448
1449         /* If virtio port just stopped, no need to send RARP */
1450         if (virtio_dev_pause(dev) < 0) {
1451                 rte_pktmbuf_free(rarp_mbuf);
1452                 return;
1453         }
1454
1455         virtio_inject_pkts(dev, &rarp_mbuf, 1);
1456         virtio_dev_resume(dev);
1457 }
1458
1459 static void
1460 virtio_ack_link_announce(struct rte_eth_dev *dev)
1461 {
1462         struct virtio_hw *hw = dev->data->dev_private;
1463         struct virtio_pmd_ctrl ctrl;
1464
1465         ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1466         ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1467
1468         virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1469 }
1470
1471 /*
1472  * Process virtio config changed interrupt. Call the callback
1473  * if link state changed, generate gratuitous RARP packet if
1474  * the status indicates an ANNOUNCE.
1475  */
1476 void
1477 virtio_interrupt_handler(void *param)
1478 {
1479         struct rte_eth_dev *dev = param;
1480         struct virtio_hw *hw = dev->data->dev_private;
1481         uint8_t isr;
1482         uint16_t status;
1483
1484         /* Read interrupt status which clears interrupt */
1485         isr = vtpci_isr(hw);
1486         PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1487
1488         if (virtio_intr_unmask(dev) < 0)
1489                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1490
1491         if (isr & VIRTIO_PCI_ISR_CONFIG) {
1492                 if (virtio_dev_link_update(dev, 0) == 0)
1493                         rte_eth_dev_callback_process(dev,
1494                                                      RTE_ETH_EVENT_INTR_LSC,
1495                                                      NULL);
1496
1497                 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1498                         vtpci_read_dev_config(hw,
1499                                 offsetof(struct virtio_net_config, status),
1500                                 &status, sizeof(status));
1501                         if (status & VIRTIO_NET_S_ANNOUNCE) {
1502                                 virtio_notify_peers(dev);
1503                                 if (hw->cvq)
1504                                         virtio_ack_link_announce(dev);
1505                         }
1506                 }
1507         }
1508 }
1509
1510 /* set rx and tx handlers according to what is supported */
1511 static void
1512 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1513 {
1514         struct virtio_hw *hw = eth_dev->data->dev_private;
1515
1516         eth_dev->tx_pkt_prepare = virtio_xmit_pkts_prepare;
1517         if (vtpci_packed_queue(hw)) {
1518                 PMD_INIT_LOG(INFO,
1519                         "virtio: using packed ring %s Tx path on port %u",
1520                         hw->use_vec_tx ? "vectorized" : "standard",
1521                         eth_dev->data->port_id);
1522                 if (hw->use_vec_tx)
1523                         eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed_vec;
1524                 else
1525                         eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed;
1526         } else {
1527                 if (hw->use_inorder_tx) {
1528                         PMD_INIT_LOG(INFO, "virtio: using inorder Tx path on port %u",
1529                                 eth_dev->data->port_id);
1530                         eth_dev->tx_pkt_burst = virtio_xmit_pkts_inorder;
1531                 } else {
1532                         PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1533                                 eth_dev->data->port_id);
1534                         eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1535                 }
1536         }
1537
1538         if (vtpci_packed_queue(hw)) {
1539                 if (hw->use_vec_rx) {
1540                         PMD_INIT_LOG(INFO,
1541                                 "virtio: using packed ring vectorized Rx path on port %u",
1542                                 eth_dev->data->port_id);
1543                         eth_dev->rx_pkt_burst =
1544                                 &virtio_recv_pkts_packed_vec;
1545                 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1546                         PMD_INIT_LOG(INFO,
1547                                 "virtio: using packed ring mergeable buffer Rx path on port %u",
1548                                 eth_dev->data->port_id);
1549                         eth_dev->rx_pkt_burst =
1550                                 &virtio_recv_mergeable_pkts_packed;
1551                 } else {
1552                         PMD_INIT_LOG(INFO,
1553                                 "virtio: using packed ring standard Rx path on port %u",
1554                                 eth_dev->data->port_id);
1555                         eth_dev->rx_pkt_burst = &virtio_recv_pkts_packed;
1556                 }
1557         } else {
1558                 if (hw->use_vec_rx) {
1559                         PMD_INIT_LOG(INFO, "virtio: using vectorized Rx path on port %u",
1560                                 eth_dev->data->port_id);
1561                         eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1562                 } else if (hw->use_inorder_rx) {
1563                         PMD_INIT_LOG(INFO,
1564                                 "virtio: using inorder Rx path on port %u",
1565                                 eth_dev->data->port_id);
1566                         eth_dev->rx_pkt_burst = &virtio_recv_pkts_inorder;
1567                 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1568                         PMD_INIT_LOG(INFO,
1569                                 "virtio: using mergeable buffer Rx path on port %u",
1570                                 eth_dev->data->port_id);
1571                         eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1572                 } else {
1573                         PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1574                                 eth_dev->data->port_id);
1575                         eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1576                 }
1577         }
1578
1579 }
1580
1581 /* Only support 1:1 queue/interrupt mapping so far.
1582  * TODO: support n:1 queue/interrupt mapping when there are limited number of
1583  * interrupt vectors (<N+1).
1584  */
1585 static int
1586 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1587 {
1588         uint32_t i;
1589         struct virtio_hw *hw = dev->data->dev_private;
1590
1591         PMD_INIT_LOG(INFO, "queue/interrupt binding");
1592         for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1593                 dev->intr_handle->intr_vec[i] = i + 1;
1594                 if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1595                                                  VIRTIO_MSI_NO_VECTOR) {
1596                         PMD_DRV_LOG(ERR, "failed to set queue vector");
1597                         return -EBUSY;
1598                 }
1599         }
1600
1601         return 0;
1602 }
1603
1604 static void
1605 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1606 {
1607         uint32_t i;
1608         struct virtio_hw *hw = dev->data->dev_private;
1609
1610         PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1611         for (i = 0; i < dev->data->nb_rx_queues; ++i)
1612                 VTPCI_OPS(hw)->set_queue_irq(hw,
1613                                              hw->vqs[i * VTNET_CQ],
1614                                              VIRTIO_MSI_NO_VECTOR);
1615 }
1616
1617 static int
1618 virtio_configure_intr(struct rte_eth_dev *dev)
1619 {
1620         struct virtio_hw *hw = dev->data->dev_private;
1621
1622         if (!rte_intr_cap_multiple(dev->intr_handle)) {
1623                 PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1624                 return -ENOTSUP;
1625         }
1626
1627         if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1628                 PMD_INIT_LOG(ERR, "Fail to create eventfd");
1629                 return -1;
1630         }
1631
1632         if (!dev->intr_handle->intr_vec) {
1633                 dev->intr_handle->intr_vec =
1634                         rte_zmalloc("intr_vec",
1635                                     hw->max_queue_pairs * sizeof(int), 0);
1636                 if (!dev->intr_handle->intr_vec) {
1637                         PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1638                                      hw->max_queue_pairs);
1639                         return -ENOMEM;
1640                 }
1641         }
1642
1643         /* Re-register callback to update max_intr */
1644         rte_intr_callback_unregister(dev->intr_handle,
1645                                      virtio_interrupt_handler,
1646                                      dev);
1647         rte_intr_callback_register(dev->intr_handle,
1648                                    virtio_interrupt_handler,
1649                                    dev);
1650
1651         /* DO NOT try to remove this! This function will enable msix, or QEMU
1652          * will encounter SIGSEGV when DRIVER_OK is sent.
1653          * And for legacy devices, this should be done before queue/vec binding
1654          * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1655          * (22) will be ignored.
1656          */
1657         if (virtio_intr_enable(dev) < 0) {
1658                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1659                 return -1;
1660         }
1661
1662         if (virtio_queues_bind_intr(dev) < 0) {
1663                 PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1664                 return -1;
1665         }
1666
1667         return 0;
1668 }
1669 #define SPEED_UNKNOWN    0xffffffff
1670 #define DUPLEX_UNKNOWN   0xff
1671 /* reset device and renegotiate features if needed */
1672 static int
1673 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1674 {
1675         struct virtio_hw *hw = eth_dev->data->dev_private;
1676         struct virtio_net_config *config;
1677         struct virtio_net_config local_config;
1678         struct rte_pci_device *pci_dev = NULL;
1679         int ret;
1680
1681         /* Reset the device although not necessary at startup */
1682         vtpci_reset(hw);
1683
1684         if (hw->vqs) {
1685                 virtio_dev_free_mbufs(eth_dev);
1686                 virtio_free_queues(hw);
1687         }
1688
1689         /* Tell the host we've noticed this device. */
1690         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1691
1692         /* Tell the host we've known how to drive the device. */
1693         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1694         if (virtio_negotiate_features(hw, req_features) < 0)
1695                 return -1;
1696
1697         hw->weak_barriers = !vtpci_with_feature(hw, VIRTIO_F_ORDER_PLATFORM);
1698
1699         if (!hw->virtio_user_dev)
1700                 pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1701
1702         /* If host does not support both status and MSI-X then disable LSC */
1703         if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) &&
1704             hw->use_msix != VIRTIO_MSIX_NONE)
1705                 eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1706         else
1707                 eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1708
1709         /* Setting up rx_header size for the device */
1710         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1711             vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
1712             vtpci_with_feature(hw, VIRTIO_F_RING_PACKED))
1713                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1714         else
1715                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1716
1717         /* Copy the permanent MAC address to: virtio_hw */
1718         virtio_get_hwaddr(hw);
1719         rte_ether_addr_copy((struct rte_ether_addr *)hw->mac_addr,
1720                         &eth_dev->data->mac_addrs[0]);
1721         PMD_INIT_LOG(DEBUG,
1722                      "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1723                      hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1724                      hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1725
1726         if (hw->speed == SPEED_UNKNOWN) {
1727                 if (vtpci_with_feature(hw, VIRTIO_NET_F_SPEED_DUPLEX)) {
1728                         config = &local_config;
1729                         vtpci_read_dev_config(hw,
1730                                 offsetof(struct virtio_net_config, speed),
1731                                 &config->speed, sizeof(config->speed));
1732                         vtpci_read_dev_config(hw,
1733                                 offsetof(struct virtio_net_config, duplex),
1734                                 &config->duplex, sizeof(config->duplex));
1735                         hw->speed = config->speed;
1736                         hw->duplex = config->duplex;
1737                 }
1738         }
1739         if (hw->speed == SPEED_UNKNOWN)
1740                 hw->speed = ETH_SPEED_NUM_10G;
1741         if (hw->duplex == DUPLEX_UNKNOWN)
1742                 hw->duplex = ETH_LINK_FULL_DUPLEX;
1743         PMD_INIT_LOG(DEBUG, "link speed = %d, duplex = %d",
1744                 hw->speed, hw->duplex);
1745         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1746                 config = &local_config;
1747
1748                 vtpci_read_dev_config(hw,
1749                         offsetof(struct virtio_net_config, mac),
1750                         &config->mac, sizeof(config->mac));
1751
1752                 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1753                         vtpci_read_dev_config(hw,
1754                                 offsetof(struct virtio_net_config, status),
1755                                 &config->status, sizeof(config->status));
1756                 } else {
1757                         PMD_INIT_LOG(DEBUG,
1758                                      "VIRTIO_NET_F_STATUS is not supported");
1759                         config->status = 0;
1760                 }
1761
1762                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1763                         vtpci_read_dev_config(hw,
1764                                 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1765                                 &config->max_virtqueue_pairs,
1766                                 sizeof(config->max_virtqueue_pairs));
1767                 } else {
1768                         PMD_INIT_LOG(DEBUG,
1769                                      "VIRTIO_NET_F_MQ is not supported");
1770                         config->max_virtqueue_pairs = 1;
1771                 }
1772
1773                 hw->max_queue_pairs = config->max_virtqueue_pairs;
1774
1775                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1776                         vtpci_read_dev_config(hw,
1777                                 offsetof(struct virtio_net_config, mtu),
1778                                 &config->mtu,
1779                                 sizeof(config->mtu));
1780
1781                         /*
1782                          * MTU value has already been checked at negotiation
1783                          * time, but check again in case it has changed since
1784                          * then, which should not happen.
1785                          */
1786                         if (config->mtu < RTE_ETHER_MIN_MTU) {
1787                                 PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1788                                                 config->mtu);
1789                                 return -1;
1790                         }
1791
1792                         hw->max_mtu = config->mtu;
1793                         /* Set initial MTU to maximum one supported by vhost */
1794                         eth_dev->data->mtu = config->mtu;
1795
1796                 } else {
1797                         hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1798                                 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1799                 }
1800
1801                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1802                                 config->max_virtqueue_pairs);
1803                 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1804                 PMD_INIT_LOG(DEBUG,
1805                                 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1806                                 config->mac[0], config->mac[1],
1807                                 config->mac[2], config->mac[3],
1808                                 config->mac[4], config->mac[5]);
1809         } else {
1810                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1811                 hw->max_queue_pairs = 1;
1812                 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1813                         VLAN_TAG_LEN - hw->vtnet_hdr_size;
1814         }
1815
1816         ret = virtio_alloc_queues(eth_dev);
1817         if (ret < 0)
1818                 return ret;
1819
1820         if (eth_dev->data->dev_conf.intr_conf.rxq) {
1821                 if (virtio_configure_intr(eth_dev) < 0) {
1822                         PMD_INIT_LOG(ERR, "failed to configure interrupt");
1823                         virtio_free_queues(hw);
1824                         return -1;
1825                 }
1826         }
1827
1828         vtpci_reinit_complete(hw);
1829
1830         if (pci_dev)
1831                 PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1832                         eth_dev->data->port_id, pci_dev->id.vendor_id,
1833                         pci_dev->id.device_id);
1834
1835         return 0;
1836 }
1837
1838 /*
1839  * Remap the PCI device again (IO port map for legacy device and
1840  * memory map for modern device), so that the secondary process
1841  * could have the PCI initiated correctly.
1842  */
1843 static int
1844 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1845 {
1846         if (hw->modern) {
1847                 /*
1848                  * We don't have to re-parse the PCI config space, since
1849                  * rte_pci_map_device() makes sure the mapped address
1850                  * in secondary process would equal to the one mapped in
1851                  * the primary process: error will be returned if that
1852                  * requirement is not met.
1853                  *
1854                  * That said, we could simply reuse all cap pointers
1855                  * (such as dev_cfg, common_cfg, etc.) parsed from the
1856                  * primary process, which is stored in shared memory.
1857                  */
1858                 if (rte_pci_map_device(pci_dev)) {
1859                         PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1860                         return -1;
1861                 }
1862         } else {
1863                 if (rte_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1864                         return -1;
1865         }
1866
1867         return 0;
1868 }
1869
1870 static void
1871 virtio_set_vtpci_ops(struct virtio_hw *hw)
1872 {
1873 #ifdef RTE_VIRTIO_USER
1874         if (hw->virtio_user_dev)
1875                 VTPCI_OPS(hw) = &virtio_user_ops;
1876         else
1877 #endif
1878         if (hw->modern)
1879                 VTPCI_OPS(hw) = &modern_ops;
1880         else
1881                 VTPCI_OPS(hw) = &legacy_ops;
1882 }
1883
1884 /*
1885  * This function is based on probe() function in virtio_pci.c
1886  * It returns 0 on success.
1887  */
1888 int
1889 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1890 {
1891         struct virtio_hw *hw = eth_dev->data->dev_private;
1892         uint32_t speed = SPEED_UNKNOWN;
1893         int vectorized = 0;
1894         int ret;
1895
1896         if (sizeof(struct virtio_net_hdr_mrg_rxbuf) > RTE_PKTMBUF_HEADROOM) {
1897                 PMD_INIT_LOG(ERR,
1898                         "Not sufficient headroom required = %d, avail = %d",
1899                         (int)sizeof(struct virtio_net_hdr_mrg_rxbuf),
1900                         RTE_PKTMBUF_HEADROOM);
1901
1902                 return -1;
1903         }
1904
1905         eth_dev->dev_ops = &virtio_eth_dev_ops;
1906         eth_dev->rx_descriptor_done = virtio_dev_rx_queue_done;
1907
1908         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1909                 if (!hw->virtio_user_dev) {
1910                         ret = virtio_remap_pci(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1911                         if (ret)
1912                                 return ret;
1913                 }
1914
1915                 virtio_set_vtpci_ops(hw);
1916                 set_rxtx_funcs(eth_dev);
1917
1918                 return 0;
1919         }
1920         ret = virtio_dev_devargs_parse(eth_dev->device->devargs,
1921                  NULL, &speed, &vectorized);
1922         if (ret < 0)
1923                 return ret;
1924         hw->speed = speed;
1925         /*
1926          * Pass the information to the rte_eth_dev_close() that it should also
1927          * release the private port resources.
1928          */
1929         eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
1930
1931         /* Allocate memory for storing MAC addresses */
1932         eth_dev->data->mac_addrs = rte_zmalloc("virtio",
1933                                 VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN, 0);
1934         if (eth_dev->data->mac_addrs == NULL) {
1935                 PMD_INIT_LOG(ERR,
1936                         "Failed to allocate %d bytes needed to store MAC addresses",
1937                         VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN);
1938                 return -ENOMEM;
1939         }
1940
1941         hw->port_id = eth_dev->data->port_id;
1942         /* For virtio_user case the hw->virtio_user_dev is populated by
1943          * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1944          */
1945         if (!hw->virtio_user_dev) {
1946                 ret = vtpci_init(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1947                 if (ret)
1948                         goto err_vtpci_init;
1949         }
1950
1951         rte_spinlock_init(&hw->state_lock);
1952
1953         /* reset device and negotiate default features */
1954         ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1955         if (ret < 0)
1956                 goto err_virtio_init;
1957
1958         if (vectorized) {
1959                 if (!vtpci_packed_queue(hw)) {
1960                         hw->use_vec_rx = 1;
1961                 } else {
1962 #if !defined(CC_AVX512_SUPPORT)
1963                         PMD_DRV_LOG(INFO,
1964                                 "building environment do not support packed ring vectorized");
1965 #else
1966                         hw->use_vec_rx = 1;
1967                         hw->use_vec_tx = 1;
1968 #endif
1969                 }
1970         }
1971
1972         hw->opened = true;
1973
1974         return 0;
1975
1976 err_virtio_init:
1977         if (!hw->virtio_user_dev) {
1978                 rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev));
1979                 if (!hw->modern)
1980                         rte_pci_ioport_unmap(VTPCI_IO(hw));
1981         }
1982 err_vtpci_init:
1983         rte_free(eth_dev->data->mac_addrs);
1984         eth_dev->data->mac_addrs = NULL;
1985         return ret;
1986 }
1987
1988 static int
1989 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1990 {
1991         PMD_INIT_FUNC_TRACE();
1992
1993         if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1994                 return 0;
1995
1996         virtio_dev_stop(eth_dev);
1997         virtio_dev_close(eth_dev);
1998
1999         eth_dev->dev_ops = NULL;
2000         eth_dev->tx_pkt_burst = NULL;
2001         eth_dev->rx_pkt_burst = NULL;
2002
2003         PMD_INIT_LOG(DEBUG, "dev_uninit completed");
2004
2005         return 0;
2006 }
2007
2008
2009 static int vdpa_check_handler(__rte_unused const char *key,
2010                 const char *value, void *ret_val)
2011 {
2012         if (strcmp(value, "1") == 0)
2013                 *(int *)ret_val = 1;
2014         else
2015                 *(int *)ret_val = 0;
2016
2017         return 0;
2018 }
2019
2020
2021 static uint32_t
2022 virtio_dev_speed_capa_get(uint32_t speed)
2023 {
2024         switch (speed) {
2025         case ETH_SPEED_NUM_10G:
2026                 return ETH_LINK_SPEED_10G;
2027         case ETH_SPEED_NUM_20G:
2028                 return ETH_LINK_SPEED_20G;
2029         case ETH_SPEED_NUM_25G:
2030                 return ETH_LINK_SPEED_25G;
2031         case ETH_SPEED_NUM_40G:
2032                 return ETH_LINK_SPEED_40G;
2033         case ETH_SPEED_NUM_50G:
2034                 return ETH_LINK_SPEED_50G;
2035         case ETH_SPEED_NUM_56G:
2036                 return ETH_LINK_SPEED_56G;
2037         case ETH_SPEED_NUM_100G:
2038                 return ETH_LINK_SPEED_100G;
2039         default:
2040                 return 0;
2041         }
2042 }
2043
2044 static int vectorized_check_handler(__rte_unused const char *key,
2045                 const char *value, void *ret_val)
2046 {
2047         if (strcmp(value, "1") == 0)
2048                 *(int *)ret_val = 1;
2049         else
2050                 *(int *)ret_val = 0;
2051
2052         return 0;
2053 }
2054
2055 #define VIRTIO_ARG_SPEED      "speed"
2056 #define VIRTIO_ARG_VDPA       "vdpa"
2057 #define VIRTIO_ARG_VECTORIZED "vectorized"
2058
2059
2060 static int
2061 link_speed_handler(const char *key __rte_unused,
2062                 const char *value, void *ret_val)
2063 {
2064         uint32_t val;
2065         if (!value || !ret_val)
2066                 return -EINVAL;
2067         val = strtoul(value, NULL, 0);
2068         /* validate input */
2069         if (virtio_dev_speed_capa_get(val) == 0)
2070                 return -EINVAL;
2071         *(uint32_t *)ret_val = val;
2072
2073         return 0;
2074 }
2075
2076
2077 static int
2078 virtio_dev_devargs_parse(struct rte_devargs *devargs, int *vdpa,
2079         uint32_t *speed, int *vectorized)
2080 {
2081         struct rte_kvargs *kvlist;
2082         int ret = 0;
2083
2084         if (devargs == NULL)
2085                 return 0;
2086
2087         kvlist = rte_kvargs_parse(devargs->args, NULL);
2088         if (kvlist == NULL) {
2089                 PMD_INIT_LOG(ERR, "error when parsing param");
2090                 return 0;
2091         }
2092         if (vdpa && rte_kvargs_count(kvlist, VIRTIO_ARG_VDPA) == 1) {
2093                 /* vdpa mode selected when there's a key-value pair:
2094                  * vdpa=1
2095                  */
2096                 ret = rte_kvargs_process(kvlist, VIRTIO_ARG_VDPA,
2097                                 vdpa_check_handler, vdpa);
2098                 if (ret < 0) {
2099                         PMD_INIT_LOG(ERR, "Failed to parse %s",
2100                                 VIRTIO_ARG_VDPA);
2101                         goto exit;
2102                 }
2103         }
2104         if (speed && rte_kvargs_count(kvlist, VIRTIO_ARG_SPEED) == 1) {
2105                 ret = rte_kvargs_process(kvlist,
2106                                         VIRTIO_ARG_SPEED,
2107                                         link_speed_handler, speed);
2108                 if (ret < 0) {
2109                         PMD_INIT_LOG(ERR, "Failed to parse %s",
2110                                         VIRTIO_ARG_SPEED);
2111                         goto exit;
2112                 }
2113         }
2114
2115         if (vectorized &&
2116                 rte_kvargs_count(kvlist, VIRTIO_ARG_VECTORIZED) == 1) {
2117                 ret = rte_kvargs_process(kvlist,
2118                                 VIRTIO_ARG_VECTORIZED,
2119                                 vectorized_check_handler, vectorized);
2120                 if (ret < 0) {
2121                         PMD_INIT_LOG(ERR, "Failed to parse %s",
2122                                         VIRTIO_ARG_VECTORIZED);
2123                         goto exit;
2124                 }
2125         }
2126
2127 exit:
2128         rte_kvargs_free(kvlist);
2129         return ret;
2130 }
2131
2132 static int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2133         struct rte_pci_device *pci_dev)
2134 {
2135         int vdpa = 0;
2136         int ret = 0;
2137
2138         ret = virtio_dev_devargs_parse(pci_dev->device.devargs, &vdpa, NULL,
2139                 NULL);
2140         if (ret < 0) {
2141                 PMD_INIT_LOG(ERR, "devargs parsing is failed");
2142                 return ret;
2143         }
2144         /* virtio pmd skips probe if device needs to work in vdpa mode */
2145         if (vdpa == 1)
2146                 return 1;
2147
2148         return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw),
2149                 eth_virtio_dev_init);
2150 }
2151
2152 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev)
2153 {
2154         int ret;
2155
2156         ret = rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
2157         /* Port has already been released by close. */
2158         if (ret == -ENODEV)
2159                 ret = 0;
2160         return ret;
2161 }
2162
2163 static struct rte_pci_driver rte_virtio_pmd = {
2164         .driver = {
2165                 .name = "net_virtio",
2166         },
2167         .id_table = pci_id_virtio_map,
2168         .drv_flags = 0,
2169         .probe = eth_virtio_pci_probe,
2170         .remove = eth_virtio_pci_remove,
2171 };
2172
2173 RTE_INIT(rte_virtio_pmd_init)
2174 {
2175         rte_eal_iopl_init();
2176         rte_pci_register(&rte_virtio_pmd);
2177 }
2178
2179 static bool
2180 rx_offload_enabled(struct virtio_hw *hw)
2181 {
2182         return vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM) ||
2183                 vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2184                 vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6);
2185 }
2186
2187 static bool
2188 tx_offload_enabled(struct virtio_hw *hw)
2189 {
2190         return vtpci_with_feature(hw, VIRTIO_NET_F_CSUM) ||
2191                 vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO4) ||
2192                 vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO6);
2193 }
2194
2195 /*
2196  * Configure virtio device
2197  * It returns 0 on success.
2198  */
2199 static int
2200 virtio_dev_configure(struct rte_eth_dev *dev)
2201 {
2202         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2203         const struct rte_eth_txmode *txmode = &dev->data->dev_conf.txmode;
2204         struct virtio_hw *hw = dev->data->dev_private;
2205         uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
2206                 hw->vtnet_hdr_size;
2207         uint64_t rx_offloads = rxmode->offloads;
2208         uint64_t tx_offloads = txmode->offloads;
2209         uint64_t req_features;
2210         int ret;
2211
2212         PMD_INIT_LOG(DEBUG, "configure");
2213         req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
2214
2215         if (rxmode->mq_mode != ETH_MQ_RX_NONE) {
2216                 PMD_DRV_LOG(ERR,
2217                         "Unsupported Rx multi queue mode %d",
2218                         rxmode->mq_mode);
2219                 return -EINVAL;
2220         }
2221
2222         if (txmode->mq_mode != ETH_MQ_TX_NONE) {
2223                 PMD_DRV_LOG(ERR,
2224                         "Unsupported Tx multi queue mode %d",
2225                         txmode->mq_mode);
2226                 return -EINVAL;
2227         }
2228
2229         if (dev->data->dev_conf.intr_conf.rxq) {
2230                 ret = virtio_init_device(dev, hw->req_guest_features);
2231                 if (ret < 0)
2232                         return ret;
2233         }
2234
2235         if (rxmode->max_rx_pkt_len > hw->max_mtu + ether_hdr_len)
2236                 req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
2237
2238         if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2239                            DEV_RX_OFFLOAD_TCP_CKSUM))
2240                 req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
2241
2242         if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
2243                 req_features |=
2244                         (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2245                         (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2246
2247         if (tx_offloads & (DEV_TX_OFFLOAD_UDP_CKSUM |
2248                            DEV_TX_OFFLOAD_TCP_CKSUM))
2249                 req_features |= (1ULL << VIRTIO_NET_F_CSUM);
2250
2251         if (tx_offloads & DEV_TX_OFFLOAD_TCP_TSO)
2252                 req_features |=
2253                         (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2254                         (1ULL << VIRTIO_NET_F_HOST_TSO6);
2255
2256         /* if request features changed, reinit the device */
2257         if (req_features != hw->req_guest_features) {
2258                 ret = virtio_init_device(dev, req_features);
2259                 if (ret < 0)
2260                         return ret;
2261         }
2262
2263         if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2264                             DEV_RX_OFFLOAD_TCP_CKSUM)) &&
2265                 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
2266                 PMD_DRV_LOG(ERR,
2267                         "rx checksum not available on this host");
2268                 return -ENOTSUP;
2269         }
2270
2271         if ((rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) &&
2272                 (!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2273                  !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
2274                 PMD_DRV_LOG(ERR,
2275                         "Large Receive Offload not available on this host");
2276                 return -ENOTSUP;
2277         }
2278
2279         /* start control queue */
2280         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
2281                 virtio_dev_cq_start(dev);
2282
2283         if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2284                 hw->vlan_strip = 1;
2285
2286         if ((rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
2287             && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2288                 PMD_DRV_LOG(ERR,
2289                             "vlan filtering not available on this host");
2290                 return -ENOTSUP;
2291         }
2292
2293         hw->has_tx_offload = tx_offload_enabled(hw);
2294         hw->has_rx_offload = rx_offload_enabled(hw);
2295
2296         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2297                 /* Enable vector (0) for Link State Intrerrupt */
2298                 if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
2299                                 VIRTIO_MSI_NO_VECTOR) {
2300                         PMD_DRV_LOG(ERR, "failed to set config vector");
2301                         return -EBUSY;
2302                 }
2303
2304         if (vtpci_packed_queue(hw)) {
2305 #if defined(RTE_ARCH_X86_64) && defined(CC_AVX512_SUPPORT)
2306                 if ((hw->use_vec_rx || hw->use_vec_tx) &&
2307                     (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F) ||
2308                      !vtpci_with_feature(hw, VIRTIO_F_IN_ORDER) ||
2309                      !vtpci_with_feature(hw, VIRTIO_F_VERSION_1))) {
2310                         PMD_DRV_LOG(INFO,
2311                                 "disabled packed ring vectorized path for requirements not met");
2312                         hw->use_vec_rx = 0;
2313                         hw->use_vec_tx = 0;
2314                 }
2315 #else
2316                 hw->use_vec_rx = 0;
2317                 hw->use_vec_tx = 0;
2318 #endif
2319
2320                 if (hw->use_vec_rx) {
2321                         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2322                                 PMD_DRV_LOG(INFO,
2323                                         "disabled packed ring vectorized rx for mrg_rxbuf enabled");
2324                                 hw->use_vec_rx = 0;
2325                         }
2326
2327                         if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) {
2328                                 PMD_DRV_LOG(INFO,
2329                                         "disabled packed ring vectorized rx for TCP_LRO enabled");
2330                                 hw->use_vec_rx = 0;
2331                         }
2332                 }
2333         } else {
2334                 if (vtpci_with_feature(hw, VIRTIO_F_IN_ORDER)) {
2335                         hw->use_inorder_tx = 1;
2336                         hw->use_inorder_rx = 1;
2337                         hw->use_vec_rx = 0;
2338                 }
2339
2340                 if (hw->use_vec_rx) {
2341 #if defined RTE_ARCH_ARM64 || defined RTE_ARCH_ARM
2342                         if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
2343                                 PMD_DRV_LOG(INFO,
2344                                         "disabled split ring vectorized path for requirement not met");
2345                                 hw->use_vec_rx = 0;
2346                         }
2347 #endif
2348                         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2349                                 PMD_DRV_LOG(INFO,
2350                                         "disabled split ring vectorized rx for mrg_rxbuf enabled");
2351                                 hw->use_vec_rx = 0;
2352                         }
2353
2354                         if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2355                                            DEV_RX_OFFLOAD_TCP_CKSUM |
2356                                            DEV_RX_OFFLOAD_TCP_LRO |
2357                                            DEV_RX_OFFLOAD_VLAN_STRIP)) {
2358                                 PMD_DRV_LOG(INFO,
2359                                         "disabled split ring vectorized rx for offloading enabled");
2360                                 hw->use_vec_rx = 0;
2361                         }
2362                 }
2363         }
2364
2365         return 0;
2366 }
2367
2368
2369 static int
2370 virtio_dev_start(struct rte_eth_dev *dev)
2371 {
2372         uint16_t nb_queues, i;
2373         struct virtnet_rx *rxvq;
2374         struct virtnet_tx *txvq __rte_unused;
2375         struct virtio_hw *hw = dev->data->dev_private;
2376         int ret;
2377
2378         /* Finish the initialization of the queues */
2379         for (i = 0; i < dev->data->nb_rx_queues; i++) {
2380                 ret = virtio_dev_rx_queue_setup_finish(dev, i);
2381                 if (ret < 0)
2382                         return ret;
2383         }
2384         for (i = 0; i < dev->data->nb_tx_queues; i++) {
2385                 ret = virtio_dev_tx_queue_setup_finish(dev, i);
2386                 if (ret < 0)
2387                         return ret;
2388         }
2389
2390         /* check if lsc interrupt feature is enabled */
2391         if (dev->data->dev_conf.intr_conf.lsc) {
2392                 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
2393                         PMD_DRV_LOG(ERR, "link status not supported by host");
2394                         return -ENOTSUP;
2395                 }
2396         }
2397
2398         /* Enable uio/vfio intr/eventfd mapping: althrough we already did that
2399          * in device configure, but it could be unmapped  when device is
2400          * stopped.
2401          */
2402         if (dev->data->dev_conf.intr_conf.lsc ||
2403             dev->data->dev_conf.intr_conf.rxq) {
2404                 virtio_intr_disable(dev);
2405
2406                 /* Setup interrupt callback  */
2407                 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2408                         rte_intr_callback_register(dev->intr_handle,
2409                                                    virtio_interrupt_handler,
2410                                                    dev);
2411
2412                 if (virtio_intr_enable(dev) < 0) {
2413                         PMD_DRV_LOG(ERR, "interrupt enable failed");
2414                         return -EIO;
2415                 }
2416         }
2417
2418         /*Notify the backend
2419          *Otherwise the tap backend might already stop its queue due to fullness.
2420          *vhost backend will have no chance to be waked up
2421          */
2422         nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
2423         if (hw->max_queue_pairs > 1) {
2424                 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
2425                         return -EINVAL;
2426         }
2427
2428         PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
2429
2430         for (i = 0; i < dev->data->nb_rx_queues; i++) {
2431                 rxvq = dev->data->rx_queues[i];
2432                 /* Flush the old packets */
2433                 virtqueue_rxvq_flush(rxvq->vq);
2434                 virtqueue_notify(rxvq->vq);
2435         }
2436
2437         for (i = 0; i < dev->data->nb_tx_queues; i++) {
2438                 txvq = dev->data->tx_queues[i];
2439                 virtqueue_notify(txvq->vq);
2440         }
2441
2442         PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
2443
2444         for (i = 0; i < dev->data->nb_rx_queues; i++) {
2445                 rxvq = dev->data->rx_queues[i];
2446                 VIRTQUEUE_DUMP(rxvq->vq);
2447         }
2448
2449         for (i = 0; i < dev->data->nb_tx_queues; i++) {
2450                 txvq = dev->data->tx_queues[i];
2451                 VIRTQUEUE_DUMP(txvq->vq);
2452         }
2453
2454         set_rxtx_funcs(dev);
2455         hw->started = true;
2456
2457         /* Initialize Link state */
2458         virtio_dev_link_update(dev, 0);
2459
2460         return 0;
2461 }
2462
2463 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
2464 {
2465         struct virtio_hw *hw = dev->data->dev_private;
2466         uint16_t nr_vq = virtio_get_nr_vq(hw);
2467         const char *type __rte_unused;
2468         unsigned int i, mbuf_num = 0;
2469         struct virtqueue *vq;
2470         struct rte_mbuf *buf;
2471         int queue_type;
2472
2473         if (hw->vqs == NULL)
2474                 return;
2475
2476         for (i = 0; i < nr_vq; i++) {
2477                 vq = hw->vqs[i];
2478                 if (!vq)
2479                         continue;
2480
2481                 queue_type = virtio_get_queue_type(hw, i);
2482                 if (queue_type == VTNET_RQ)
2483                         type = "rxq";
2484                 else if (queue_type == VTNET_TQ)
2485                         type = "txq";
2486                 else
2487                         continue;
2488
2489                 PMD_INIT_LOG(DEBUG,
2490                         "Before freeing %s[%d] used and unused buf",
2491                         type, i);
2492                 VIRTQUEUE_DUMP(vq);
2493
2494                 while ((buf = virtqueue_detach_unused(vq)) != NULL) {
2495                         rte_pktmbuf_free(buf);
2496                         mbuf_num++;
2497                 }
2498
2499                 PMD_INIT_LOG(DEBUG,
2500                         "After freeing %s[%d] used and unused buf",
2501                         type, i);
2502                 VIRTQUEUE_DUMP(vq);
2503         }
2504
2505         PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
2506 }
2507
2508 /*
2509  * Stop device: disable interrupt and mark link down
2510  */
2511 static void
2512 virtio_dev_stop(struct rte_eth_dev *dev)
2513 {
2514         struct virtio_hw *hw = dev->data->dev_private;
2515         struct rte_eth_link link;
2516         struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
2517
2518         PMD_INIT_LOG(DEBUG, "stop");
2519
2520         rte_spinlock_lock(&hw->state_lock);
2521         if (!hw->started)
2522                 goto out_unlock;
2523         hw->started = false;
2524
2525         if (intr_conf->lsc || intr_conf->rxq) {
2526                 virtio_intr_disable(dev);
2527
2528                 /* Reset interrupt callback  */
2529                 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
2530                         rte_intr_callback_unregister(dev->intr_handle,
2531                                                      virtio_interrupt_handler,
2532                                                      dev);
2533                 }
2534         }
2535
2536         memset(&link, 0, sizeof(link));
2537         rte_eth_linkstatus_set(dev, &link);
2538 out_unlock:
2539         rte_spinlock_unlock(&hw->state_lock);
2540 }
2541
2542 static int
2543 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2544 {
2545         struct rte_eth_link link;
2546         uint16_t status;
2547         struct virtio_hw *hw = dev->data->dev_private;
2548
2549         memset(&link, 0, sizeof(link));
2550         link.link_duplex = hw->duplex;
2551         link.link_speed  = hw->speed;
2552         link.link_autoneg = ETH_LINK_AUTONEG;
2553
2554         if (!hw->started) {
2555                 link.link_status = ETH_LINK_DOWN;
2556         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2557                 PMD_INIT_LOG(DEBUG, "Get link status from hw");
2558                 vtpci_read_dev_config(hw,
2559                                 offsetof(struct virtio_net_config, status),
2560                                 &status, sizeof(status));
2561                 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2562                         link.link_status = ETH_LINK_DOWN;
2563                         PMD_INIT_LOG(DEBUG, "Port %d is down",
2564                                      dev->data->port_id);
2565                 } else {
2566                         link.link_status = ETH_LINK_UP;
2567                         PMD_INIT_LOG(DEBUG, "Port %d is up",
2568                                      dev->data->port_id);
2569                 }
2570         } else {
2571                 link.link_status = ETH_LINK_UP;
2572         }
2573
2574         return rte_eth_linkstatus_set(dev, &link);
2575 }
2576
2577 static int
2578 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2579 {
2580         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2581         struct virtio_hw *hw = dev->data->dev_private;
2582         uint64_t offloads = rxmode->offloads;
2583
2584         if (mask & ETH_VLAN_FILTER_MASK) {
2585                 if ((offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2586                                 !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2587
2588                         PMD_DRV_LOG(NOTICE,
2589                                 "vlan filtering not available on this host");
2590
2591                         return -ENOTSUP;
2592                 }
2593         }
2594
2595         if (mask & ETH_VLAN_STRIP_MASK)
2596                 hw->vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
2597
2598         return 0;
2599 }
2600
2601 static int
2602 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2603 {
2604         uint64_t tso_mask, host_features;
2605         struct virtio_hw *hw = dev->data->dev_private;
2606         dev_info->speed_capa = virtio_dev_speed_capa_get(hw->speed);
2607
2608         dev_info->max_rx_queues =
2609                 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2610         dev_info->max_tx_queues =
2611                 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2612         dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2613         dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2614         dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2615
2616         host_features = VTPCI_OPS(hw)->get_features(hw);
2617         dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
2618         dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2619         if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2620                 dev_info->rx_offload_capa |=
2621                         DEV_RX_OFFLOAD_TCP_CKSUM |
2622                         DEV_RX_OFFLOAD_UDP_CKSUM;
2623         }
2624         if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN))
2625                 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_FILTER;
2626         tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2627                 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2628         if ((host_features & tso_mask) == tso_mask)
2629                 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2630
2631         dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
2632                                     DEV_TX_OFFLOAD_VLAN_INSERT;
2633         if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2634                 dev_info->tx_offload_capa |=
2635                         DEV_TX_OFFLOAD_UDP_CKSUM |
2636                         DEV_TX_OFFLOAD_TCP_CKSUM;
2637         }
2638         tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2639                 (1ULL << VIRTIO_NET_F_HOST_TSO6);
2640         if ((host_features & tso_mask) == tso_mask)
2641                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2642
2643         return 0;
2644 }
2645
2646 /*
2647  * It enables testpmd to collect per queue stats.
2648  */
2649 static int
2650 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2651 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2652 __rte_unused uint8_t is_rx)
2653 {
2654         return 0;
2655 }
2656
2657 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
2658 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
2659 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio-pci");
2660 RTE_LOG_REGISTER(virtio_logtype_init, pmd.net.virtio.init, NOTICE);
2661 RTE_LOG_REGISTER(virtio_logtype_driver, pmd.net.virtio.driver, NOTICE);