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 int
 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 0;
 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         return 0;
 748 }
 749
 750 static int
 751 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
 752 {
 753         struct virtio_hw *hw = dev->data->dev_private;
 754         struct virtio_pmd_ctrl ctrl;
 755         int dlen[1];
 756         int ret;
 757
 758         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 759                 PMD_INIT_LOG(INFO, "host does not support rx control");
 760                 return -ENOTSUP;
 761         }
 762
 763         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 764         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 765         ctrl.data[0] = 1;
 766         dlen[0] = 1;
 767
 768         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 769         if (ret) {
 770                 PMD_INIT_LOG(ERR, "Failed to enable promisc");
 771                 return -EAGAIN;
 772         }
 773
 774         return 0;
 775 }
 776
 777 static int
 778 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
 779 {
 780         struct virtio_hw *hw = dev->data->dev_private;
 781         struct virtio_pmd_ctrl ctrl;
 782         int dlen[1];
 783         int ret;
 784
 785         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 786                 PMD_INIT_LOG(INFO, "host does not support rx control");
 787                 return -ENOTSUP;
 788         }
 789
 790         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 791         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 792         ctrl.data[0] = 0;
 793         dlen[0] = 1;
 794
 795         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 796         if (ret) {
 797                 PMD_INIT_LOG(ERR, "Failed to disable promisc");
 798                 return -EAGAIN;
 799         }
 800
 801         return 0;
 802 }
 803
 804 static int
 805 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
 806 {
 807         struct virtio_hw *hw = dev->data->dev_private;
 808         struct virtio_pmd_ctrl ctrl;
 809         int dlen[1];
 810         int ret;
 811
 812         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 813                 PMD_INIT_LOG(INFO, "host does not support rx control");
 814                 return -ENOTSUP;
 815         }
 816
 817         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 818         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 819         ctrl.data[0] = 1;
 820         dlen[0] = 1;
 821
 822         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 823         if (ret) {
 824                 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
 825                 return -EAGAIN;
 826         }
 827
 828         return 0;
 829 }
 830
 831 static int
 832 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
 833 {
 834         struct virtio_hw *hw = dev->data->dev_private;
 835         struct virtio_pmd_ctrl ctrl;
 836         int dlen[1];
 837         int ret;
 838
 839         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 840                 PMD_INIT_LOG(INFO, "host does not support rx control");
 841                 return -ENOTSUP;
 842         }
 843
 844         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 845         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 846         ctrl.data[0] = 0;
 847         dlen[0] = 1;
 848
 849         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 850         if (ret) {
 851                 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
 852                 return -EAGAIN;
 853         }
 854
 855         return 0;
 856 }
 857
 858 #define VLAN_TAG_LEN           4    /* 802.3ac tag (not DMA'd) */
 859 static int
 860 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 861 {
 862         struct virtio_hw *hw = dev->data->dev_private;
 863         uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
 864                                  hw->vtnet_hdr_size;
 865         uint32_t frame_size = mtu + ether_hdr_len;
 866         uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
 867
 868         max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
 869
 870         if (mtu < RTE_ETHER_MIN_MTU || frame_size > max_frame_size) {
 871                 PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
 872                         RTE_ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
 873                 return -EINVAL;
 874         }
 875         return 0;
 876 }
 877
 878 static int
 879 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
 880 {
 881         struct virtio_hw *hw = dev->data->dev_private;
 882         struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
 883         struct virtqueue *vq = rxvq->vq;
 884
 885         virtqueue_enable_intr(vq);
 886         virtio_mb(hw->weak_barriers);
 887         return 0;
 888 }
 889
 890 static int
 891 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
 892 {
 893         struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
 894         struct virtqueue *vq = rxvq->vq;
 895
 896         virtqueue_disable_intr(vq);
 897         return 0;
 898 }
 899
 900 /*
 901  * dev_ops for virtio, bare necessities for basic operation
 902  */
 903 static const struct eth_dev_ops virtio_eth_dev_ops = {
 904         .dev_configure           = virtio_dev_configure,
 905         .dev_start               = virtio_dev_start,
 906         .dev_stop                = virtio_dev_stop,
 907         .dev_close               = virtio_dev_close,
 908         .promiscuous_enable      = virtio_dev_promiscuous_enable,
 909         .promiscuous_disable     = virtio_dev_promiscuous_disable,
 910         .allmulticast_enable     = virtio_dev_allmulticast_enable,
 911         .allmulticast_disable    = virtio_dev_allmulticast_disable,
 912         .mtu_set                 = virtio_mtu_set,
 913         .dev_infos_get           = virtio_dev_info_get,
 914         .stats_get               = virtio_dev_stats_get,
 915         .xstats_get              = virtio_dev_xstats_get,
 916         .xstats_get_names        = virtio_dev_xstats_get_names,
 917         .stats_reset             = virtio_dev_stats_reset,
 918         .xstats_reset            = virtio_dev_stats_reset,
 919         .link_update             = virtio_dev_link_update,
 920         .vlan_offload_set        = virtio_dev_vlan_offload_set,
 921         .rx_queue_setup          = virtio_dev_rx_queue_setup,
 922         .rx_queue_intr_enable    = virtio_dev_rx_queue_intr_enable,
 923         .rx_queue_intr_disable   = virtio_dev_rx_queue_intr_disable,
 924         .rx_queue_release        = virtio_dev_queue_release,
 925         .tx_queue_setup          = virtio_dev_tx_queue_setup,
 926         .tx_queue_release        = virtio_dev_queue_release,
 927         /* collect stats per queue */
 928         .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
 929         .vlan_filter_set         = virtio_vlan_filter_set,
 930         .mac_addr_add            = virtio_mac_addr_add,
 931         .mac_addr_remove         = virtio_mac_addr_remove,
 932         .mac_addr_set            = virtio_mac_addr_set,
 933 };
 934
 935 /*
 936  * dev_ops for virtio-user in secondary processes, as we just have
 937  * some limited supports currently.
 938  */
 939 const struct eth_dev_ops virtio_user_secondary_eth_dev_ops = {
 940         .dev_infos_get           = virtio_dev_info_get,
 941         .stats_get               = virtio_dev_stats_get,
 942         .xstats_get              = virtio_dev_xstats_get,
 943         .xstats_get_names        = virtio_dev_xstats_get_names,
 944         .stats_reset             = virtio_dev_stats_reset,
 945         .xstats_reset            = virtio_dev_stats_reset,
 946         /* collect stats per queue */
 947         .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
 948 };
 949
 950 static void
 951 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 952 {
 953         unsigned i;
 954
 955         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 956                 const struct virtnet_tx *txvq = dev->data->tx_queues[i];
 957                 if (txvq == NULL)
 958                         continue;
 959
 960                 stats->opackets += txvq->stats.packets;
 961                 stats->obytes += txvq->stats.bytes;
 962
 963                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 964                         stats->q_opackets[i] = txvq->stats.packets;
 965                         stats->q_obytes[i] = txvq->stats.bytes;
 966                 }
 967         }
 968
 969         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 970                 const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
 971                 if (rxvq == NULL)
 972                         continue;
 973
 974                 stats->ipackets += rxvq->stats.packets;
 975                 stats->ibytes += rxvq->stats.bytes;
 976                 stats->ierrors += rxvq->stats.errors;
 977
 978                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 979                         stats->q_ipackets[i] = rxvq->stats.packets;
 980                         stats->q_ibytes[i] = rxvq->stats.bytes;
 981                 }
 982         }
 983
 984         stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
 985 }
 986
 987 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
 988                                        struct rte_eth_xstat_name *xstats_names,
 989                                        __rte_unused unsigned limit)
 990 {
 991         unsigned i;
 992         unsigned count = 0;
 993         unsigned t;
 994
 995         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
 996                 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
 997
 998         if (xstats_names != NULL) {
 999                 /* Note: limit checked in rte_eth_xstats_names() */
1000
1001                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
1002                         struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1003                         if (rxvq == NULL)
1004                                 continue;
1005                         for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1006                                 snprintf(xstats_names[count].name,
1007                                         sizeof(xstats_names[count].name),
1008                                         "rx_q%u_%s", i,
1009                                         rte_virtio_rxq_stat_strings[t].name);
1010                                 count++;
1011                         }
1012                 }
1013
1014                 for (i = 0; i < dev->data->nb_tx_queues; i++) {
1015                         struct virtnet_tx *txvq = dev->data->tx_queues[i];
1016                         if (txvq == NULL)
1017                                 continue;
1018                         for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1019                                 snprintf(xstats_names[count].name,
1020                                         sizeof(xstats_names[count].name),
1021                                         "tx_q%u_%s", i,
1022                                         rte_virtio_txq_stat_strings[t].name);
1023                                 count++;
1024                         }
1025                 }
1026                 return count;
1027         }
1028         return nstats;
1029 }
1030
1031 static int
1032 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1033                       unsigned n)
1034 {
1035         unsigned i;
1036         unsigned count = 0;
1037
1038         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1039                 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1040
1041         if (n < nstats)
1042                 return nstats;
1043
1044         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1045                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1046
1047                 if (rxvq == NULL)
1048                         continue;
1049
1050                 unsigned t;
1051
1052                 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1053                         xstats[count].value = *(uint64_t *)(((char *)rxvq) +
1054                                 rte_virtio_rxq_stat_strings[t].offset);
1055                         xstats[count].id = count;
1056                         count++;
1057                 }
1058         }
1059
1060         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1061                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1062
1063                 if (txvq == NULL)
1064                         continue;
1065
1066                 unsigned t;
1067
1068                 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1069                         xstats[count].value = *(uint64_t *)(((char *)txvq) +
1070                                 rte_virtio_txq_stat_strings[t].offset);
1071                         xstats[count].id = count;
1072                         count++;
1073                 }
1074         }
1075
1076         return count;
1077 }
1078
1079 static int
1080 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1081 {
1082         virtio_update_stats(dev, stats);
1083
1084         return 0;
1085 }
1086
1087 static int
1088 virtio_dev_stats_reset(struct rte_eth_dev *dev)
1089 {
1090         unsigned int i;
1091
1092         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1093                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1094                 if (txvq == NULL)
1095                         continue;
1096
1097                 txvq->stats.packets = 0;
1098                 txvq->stats.bytes = 0;
1099                 txvq->stats.multicast = 0;
1100                 txvq->stats.broadcast = 0;
1101                 memset(txvq->stats.size_bins, 0,
1102                        sizeof(txvq->stats.size_bins[0]) * 8);
1103         }
1104
1105         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1106                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1107                 if (rxvq == NULL)
1108                         continue;
1109
1110                 rxvq->stats.packets = 0;
1111                 rxvq->stats.bytes = 0;
1112                 rxvq->stats.errors = 0;
1113                 rxvq->stats.multicast = 0;
1114                 rxvq->stats.broadcast = 0;
1115                 memset(rxvq->stats.size_bins, 0,
1116                        sizeof(rxvq->stats.size_bins[0]) * 8);
1117         }
1118
1119         return 0;
1120 }
1121
1122 static void
1123 virtio_set_hwaddr(struct virtio_hw *hw)
1124 {
1125         vtpci_write_dev_config(hw,
1126                         offsetof(struct virtio_net_config, mac),
1127                         &hw->mac_addr, RTE_ETHER_ADDR_LEN);
1128 }
1129
1130 static void
1131 virtio_get_hwaddr(struct virtio_hw *hw)
1132 {
1133         if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1134                 vtpci_read_dev_config(hw,
1135                         offsetof(struct virtio_net_config, mac),
1136                         &hw->mac_addr, RTE_ETHER_ADDR_LEN);
1137         } else {
1138                 rte_eth_random_addr(&hw->mac_addr[0]);
1139                 virtio_set_hwaddr(hw);
1140         }
1141 }
1142
1143 static int
1144 virtio_mac_table_set(struct virtio_hw *hw,
1145                      const struct virtio_net_ctrl_mac *uc,
1146                      const struct virtio_net_ctrl_mac *mc)
1147 {
1148         struct virtio_pmd_ctrl ctrl;
1149         int err, len[2];
1150
1151         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1152                 PMD_DRV_LOG(INFO, "host does not support mac table");
1153                 return -1;
1154         }
1155
1156         ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1157         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1158
1159         len[0] = uc->entries * RTE_ETHER_ADDR_LEN + sizeof(uc->entries);
1160         memcpy(ctrl.data, uc, len[0]);
1161
1162         len[1] = mc->entries * RTE_ETHER_ADDR_LEN + sizeof(mc->entries);
1163         memcpy(ctrl.data + len[0], mc, len[1]);
1164
1165         err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1166         if (err != 0)
1167                 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1168         return err;
1169 }
1170
1171 static int
1172 virtio_mac_addr_add(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1173                     uint32_t index, uint32_t vmdq __rte_unused)
1174 {
1175         struct virtio_hw *hw = dev->data->dev_private;
1176         const struct rte_ether_addr *addrs = dev->data->mac_addrs;
1177         unsigned int i;
1178         struct virtio_net_ctrl_mac *uc, *mc;
1179
1180         if (index >= VIRTIO_MAX_MAC_ADDRS) {
1181                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1182                 return -EINVAL;
1183         }
1184
1185         uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1186                 sizeof(uc->entries));
1187         uc->entries = 0;
1188         mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1189                 sizeof(mc->entries));
1190         mc->entries = 0;
1191
1192         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1193                 const struct rte_ether_addr *addr
1194                         = (i == index) ? mac_addr : addrs + i;
1195                 struct virtio_net_ctrl_mac *tbl
1196                         = rte_is_multicast_ether_addr(addr) ? mc : uc;
1197
1198                 memcpy(&tbl->macs[tbl->entries++], addr, RTE_ETHER_ADDR_LEN);
1199         }
1200
1201         return virtio_mac_table_set(hw, uc, mc);
1202 }
1203
1204 static void
1205 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1206 {
1207         struct virtio_hw *hw = dev->data->dev_private;
1208         struct rte_ether_addr *addrs = dev->data->mac_addrs;
1209         struct virtio_net_ctrl_mac *uc, *mc;
1210         unsigned int i;
1211
1212         if (index >= VIRTIO_MAX_MAC_ADDRS) {
1213                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1214                 return;
1215         }
1216
1217         uc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1218                 sizeof(uc->entries));
1219         uc->entries = 0;
1220         mc = alloca(VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN +
1221                 sizeof(mc->entries));
1222         mc->entries = 0;
1223
1224         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1225                 struct virtio_net_ctrl_mac *tbl;
1226
1227                 if (i == index || rte_is_zero_ether_addr(addrs + i))
1228                         continue;
1229
1230                 tbl = rte_is_multicast_ether_addr(addrs + i) ? mc : uc;
1231                 memcpy(&tbl->macs[tbl->entries++], addrs + i,
1232                         RTE_ETHER_ADDR_LEN);
1233         }
1234
1235         virtio_mac_table_set(hw, uc, mc);
1236 }
1237
1238 static int
1239 virtio_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1240 {
1241         struct virtio_hw *hw = dev->data->dev_private;
1242
1243         memcpy(hw->mac_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1244
1245         /* Use atomic update if available */
1246         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1247                 struct virtio_pmd_ctrl ctrl;
1248                 int len = RTE_ETHER_ADDR_LEN;
1249
1250                 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1251                 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1252
1253                 memcpy(ctrl.data, mac_addr, RTE_ETHER_ADDR_LEN);
1254                 return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1255         }
1256
1257         if (!vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1258                 return -ENOTSUP;
1259
1260         virtio_set_hwaddr(hw);
1261         return 0;
1262 }
1263
1264 static int
1265 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1266 {
1267         struct virtio_hw *hw = dev->data->dev_private;
1268         struct virtio_pmd_ctrl ctrl;
1269         int len;
1270
1271         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1272                 return -ENOTSUP;
1273
1274         ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1275         ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1276         memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1277         len = sizeof(vlan_id);
1278
1279         return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1280 }
1281
1282 static int
1283 virtio_intr_unmask(struct rte_eth_dev *dev)
1284 {
1285         struct virtio_hw *hw = dev->data->dev_private;
1286
1287         if (rte_intr_ack(dev->intr_handle) < 0)
1288                 return -1;
1289
1290         if (!hw->virtio_user_dev)
1291                 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1292
1293         return 0;
1294 }
1295
1296 static int
1297 virtio_intr_enable(struct rte_eth_dev *dev)
1298 {
1299         struct virtio_hw *hw = dev->data->dev_private;
1300
1301         if (rte_intr_enable(dev->intr_handle) < 0)
1302                 return -1;
1303
1304         if (!hw->virtio_user_dev)
1305                 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1306
1307         return 0;
1308 }
1309
1310 static int
1311 virtio_intr_disable(struct rte_eth_dev *dev)
1312 {
1313         struct virtio_hw *hw = dev->data->dev_private;
1314
1315         if (rte_intr_disable(dev->intr_handle) < 0)
1316                 return -1;
1317
1318         if (!hw->virtio_user_dev)
1319                 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1320
1321         return 0;
1322 }
1323
1324 static int
1325 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1326 {
1327         uint64_t host_features;
1328
1329         /* Prepare guest_features: feature that driver wants to support */
1330         PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1331                 req_features);
1332
1333         /* Read device(host) feature bits */
1334         host_features = VTPCI_OPS(hw)->get_features(hw);
1335         PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1336                 host_features);
1337
1338         /* If supported, ensure MTU value is valid before acknowledging it. */
1339         if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1340                 struct virtio_net_config config;
1341
1342                 vtpci_read_dev_config(hw,
1343                         offsetof(struct virtio_net_config, mtu),
1344                         &config.mtu, sizeof(config.mtu));
1345
1346                 if (config.mtu < RTE_ETHER_MIN_MTU)
1347                         req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1348         }
1349
1350         /*
1351          * Negotiate features: Subset of device feature bits are written back
1352          * guest feature bits.
1353          */
1354         hw->guest_features = req_features;
1355         hw->guest_features = vtpci_negotiate_features(hw, host_features);
1356         PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1357                 hw->guest_features);
1358
1359         if (hw->modern && !vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1360                 PMD_INIT_LOG(ERR,
1361                         "VIRTIO_F_VERSION_1 features is not enabled.");
1362                 return -1;
1363         }
1364
1365         if (hw->modern || hw->virtio_user_dev) {
1366                 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1367                 if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1368                         PMD_INIT_LOG(ERR,
1369                                 "failed to set FEATURES_OK status!");
1370                         return -1;
1371                 }
1372         }
1373
1374         hw->req_guest_features = req_features;
1375
1376         return 0;
1377 }
1378
1379 int
1380 virtio_dev_pause(struct rte_eth_dev *dev)
1381 {
1382         struct virtio_hw *hw = dev->data->dev_private;
1383
1384         rte_spinlock_lock(&hw->state_lock);
1385
1386         if (hw->started == 0) {
1387                 /* Device is just stopped. */
1388                 rte_spinlock_unlock(&hw->state_lock);
1389                 return -1;
1390         }
1391         hw->started = 0;
1392         /*
1393          * Prevent the worker threads from touching queues to avoid contention,
1394          * 1 ms should be enough for the ongoing Tx function to finish.
1395          */
1396         rte_delay_ms(1);
1397         return 0;
1398 }
1399
1400 /*
1401  * Recover hw state to let the worker threads continue.
1402  */
1403 void
1404 virtio_dev_resume(struct rte_eth_dev *dev)
1405 {
1406         struct virtio_hw *hw = dev->data->dev_private;
1407
1408         hw->started = 1;
1409         rte_spinlock_unlock(&hw->state_lock);
1410 }
1411
1412 /*
1413  * Should be called only after device is paused.
1414  */
1415 int
1416 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1417                 int nb_pkts)
1418 {
1419         struct virtio_hw *hw = dev->data->dev_private;
1420         struct virtnet_tx *txvq = dev->data->tx_queues[0];
1421         int ret;
1422
1423         hw->inject_pkts = tx_pkts;
1424         ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1425         hw->inject_pkts = NULL;
1426
1427         return ret;
1428 }
1429
1430 static void
1431 virtio_notify_peers(struct rte_eth_dev *dev)
1432 {
1433         struct virtio_hw *hw = dev->data->dev_private;
1434         struct virtnet_rx *rxvq;
1435         struct rte_mbuf *rarp_mbuf;
1436
1437         if (!dev->data->rx_queues)
1438                 return;
1439
1440         rxvq = dev->data->rx_queues[0];
1441         if (!rxvq)
1442                 return;
1443
1444         rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1445                         (struct rte_ether_addr *)hw->mac_addr);
1446         if (rarp_mbuf == NULL) {
1447                 PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1448                 return;
1449         }
1450
1451         /* If virtio port just stopped, no need to send RARP */
1452         if (virtio_dev_pause(dev) < 0) {
1453                 rte_pktmbuf_free(rarp_mbuf);
1454                 return;
1455         }
1456
1457         virtio_inject_pkts(dev, &rarp_mbuf, 1);
1458         virtio_dev_resume(dev);
1459 }
1460
1461 static void
1462 virtio_ack_link_announce(struct rte_eth_dev *dev)
1463 {
1464         struct virtio_hw *hw = dev->data->dev_private;
1465         struct virtio_pmd_ctrl ctrl;
1466
1467         ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1468         ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1469
1470         virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1471 }
1472
1473 /*
1474  * Process virtio config changed interrupt. Call the callback
1475  * if link state changed, generate gratuitous RARP packet if
1476  * the status indicates an ANNOUNCE.
1477  */
1478 void
1479 virtio_interrupt_handler(void *param)
1480 {
1481         struct rte_eth_dev *dev = param;
1482         struct virtio_hw *hw = dev->data->dev_private;
1483         uint8_t isr;
1484         uint16_t status;
1485
1486         /* Read interrupt status which clears interrupt */
1487         isr = vtpci_isr(hw);
1488         PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1489
1490         if (virtio_intr_unmask(dev) < 0)
1491                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1492
1493         if (isr & VIRTIO_PCI_ISR_CONFIG) {
1494                 if (virtio_dev_link_update(dev, 0) == 0)
1495                         rte_eth_dev_callback_process(dev,
1496                                                      RTE_ETH_EVENT_INTR_LSC,
1497                                                      NULL);
1498
1499                 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1500                         vtpci_read_dev_config(hw,
1501                                 offsetof(struct virtio_net_config, status),
1502                                 &status, sizeof(status));
1503                         if (status & VIRTIO_NET_S_ANNOUNCE) {
1504                                 virtio_notify_peers(dev);
1505                                 if (hw->cvq)
1506                                         virtio_ack_link_announce(dev);
1507                         }
1508                 }
1509         }
1510 }
1511
1512 /* set rx and tx handlers according to what is supported */
1513 static void
1514 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1515 {
1516         struct virtio_hw *hw = eth_dev->data->dev_private;
1517
1518         eth_dev->tx_pkt_prepare = virtio_xmit_pkts_prepare;
1519         if (vtpci_packed_queue(hw)) {
1520                 PMD_INIT_LOG(INFO,
1521                         "virtio: using packed ring %s Tx path on port %u",
1522                         hw->use_vec_tx ? "vectorized" : "standard",
1523                         eth_dev->data->port_id);
1524                 if (hw->use_vec_tx)
1525                         eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed_vec;
1526                 else
1527                         eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed;
1528         } else {
1529                 if (hw->use_inorder_tx) {
1530                         PMD_INIT_LOG(INFO, "virtio: using inorder Tx path on port %u",
1531                                 eth_dev->data->port_id);
1532                         eth_dev->tx_pkt_burst = virtio_xmit_pkts_inorder;
1533                 } else {
1534                         PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1535                                 eth_dev->data->port_id);
1536                         eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1537                 }
1538         }
1539
1540         if (vtpci_packed_queue(hw)) {
1541                 if (hw->use_vec_rx) {
1542                         PMD_INIT_LOG(INFO,
1543                                 "virtio: using packed ring vectorized Rx path on port %u",
1544                                 eth_dev->data->port_id);
1545                         eth_dev->rx_pkt_burst =
1546                                 &virtio_recv_pkts_packed_vec;
1547                 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1548                         PMD_INIT_LOG(INFO,
1549                                 "virtio: using packed ring mergeable buffer Rx path on port %u",
1550                                 eth_dev->data->port_id);
1551                         eth_dev->rx_pkt_burst =
1552                                 &virtio_recv_mergeable_pkts_packed;
1553                 } else {
1554                         PMD_INIT_LOG(INFO,
1555                                 "virtio: using packed ring standard Rx path on port %u",
1556                                 eth_dev->data->port_id);
1557                         eth_dev->rx_pkt_burst = &virtio_recv_pkts_packed;
1558                 }
1559         } else {
1560                 if (hw->use_vec_rx) {
1561                         PMD_INIT_LOG(INFO, "virtio: using vectorized Rx path on port %u",
1562                                 eth_dev->data->port_id);
1563                         eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1564                 } else if (hw->use_inorder_rx) {
1565                         PMD_INIT_LOG(INFO,
1566                                 "virtio: using inorder Rx path on port %u",
1567                                 eth_dev->data->port_id);
1568                         eth_dev->rx_pkt_burst = &virtio_recv_pkts_inorder;
1569                 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1570                         PMD_INIT_LOG(INFO,
1571                                 "virtio: using mergeable buffer Rx path on port %u",
1572                                 eth_dev->data->port_id);
1573                         eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1574                 } else {
1575                         PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1576                                 eth_dev->data->port_id);
1577                         eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1578                 }
1579         }
1580
1581 }
1582
1583 /* Only support 1:1 queue/interrupt mapping so far.
1584  * TODO: support n:1 queue/interrupt mapping when there are limited number of
1585  * interrupt vectors (<N+1).
1586  */
1587 static int
1588 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1589 {
1590         uint32_t i;
1591         struct virtio_hw *hw = dev->data->dev_private;
1592
1593         PMD_INIT_LOG(INFO, "queue/interrupt binding");
1594         for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1595                 dev->intr_handle->intr_vec[i] = i + 1;
1596                 if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1597                                                  VIRTIO_MSI_NO_VECTOR) {
1598                         PMD_DRV_LOG(ERR, "failed to set queue vector");
1599                         return -EBUSY;
1600                 }
1601         }
1602
1603         return 0;
1604 }
1605
1606 static void
1607 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1608 {
1609         uint32_t i;
1610         struct virtio_hw *hw = dev->data->dev_private;
1611
1612         PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1613         for (i = 0; i < dev->data->nb_rx_queues; ++i)
1614                 VTPCI_OPS(hw)->set_queue_irq(hw,
1615                                              hw->vqs[i * VTNET_CQ],
1616                                              VIRTIO_MSI_NO_VECTOR);
1617 }
1618
1619 static int
1620 virtio_configure_intr(struct rte_eth_dev *dev)
1621 {
1622         struct virtio_hw *hw = dev->data->dev_private;
1623
1624         if (!rte_intr_cap_multiple(dev->intr_handle)) {
1625                 PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1626                 return -ENOTSUP;
1627         }
1628
1629         if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1630                 PMD_INIT_LOG(ERR, "Fail to create eventfd");
1631                 return -1;
1632         }
1633
1634         if (!dev->intr_handle->intr_vec) {
1635                 dev->intr_handle->intr_vec =
1636                         rte_zmalloc("intr_vec",
1637                                     hw->max_queue_pairs * sizeof(int), 0);
1638                 if (!dev->intr_handle->intr_vec) {
1639                         PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1640                                      hw->max_queue_pairs);
1641                         return -ENOMEM;
1642                 }
1643         }
1644
1645         /* Re-register callback to update max_intr */
1646         rte_intr_callback_unregister(dev->intr_handle,
1647                                      virtio_interrupt_handler,
1648                                      dev);
1649         rte_intr_callback_register(dev->intr_handle,
1650                                    virtio_interrupt_handler,
1651                                    dev);
1652
1653         /* DO NOT try to remove this! This function will enable msix, or QEMU
1654          * will encounter SIGSEGV when DRIVER_OK is sent.
1655          * And for legacy devices, this should be done before queue/vec binding
1656          * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1657          * (22) will be ignored.
1658          */
1659         if (virtio_intr_enable(dev) < 0) {
1660                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1661                 return -1;
1662         }
1663
1664         if (virtio_queues_bind_intr(dev) < 0) {
1665                 PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1666                 return -1;
1667         }
1668
1669         return 0;
1670 }
1671 #define SPEED_UNKNOWN    0xffffffff
1672 #define DUPLEX_UNKNOWN   0xff
1673 /* reset device and renegotiate features if needed */
1674 static int
1675 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1676 {
1677         struct virtio_hw *hw = eth_dev->data->dev_private;
1678         struct virtio_net_config *config;
1679         struct virtio_net_config local_config;
1680         struct rte_pci_device *pci_dev = NULL;
1681         int ret;
1682
1683         /* Reset the device although not necessary at startup */
1684         vtpci_reset(hw);
1685
1686         if (hw->vqs) {
1687                 virtio_dev_free_mbufs(eth_dev);
1688                 virtio_free_queues(hw);
1689         }
1690
1691         /* Tell the host we've noticed this device. */
1692         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1693
1694         /* Tell the host we've known how to drive the device. */
1695         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1696         if (virtio_negotiate_features(hw, req_features) < 0)
1697                 return -1;
1698
1699         hw->weak_barriers = !vtpci_with_feature(hw, VIRTIO_F_ORDER_PLATFORM);
1700
1701         if (!hw->virtio_user_dev)
1702                 pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1703
1704         /* If host does not support both status and MSI-X then disable LSC */
1705         if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) &&
1706             hw->use_msix != VIRTIO_MSIX_NONE)
1707                 eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1708         else
1709                 eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1710
1711         /* Setting up rx_header size for the device */
1712         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1713             vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
1714             vtpci_with_feature(hw, VIRTIO_F_RING_PACKED))
1715                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1716         else
1717                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1718
1719         /* Copy the permanent MAC address to: virtio_hw */
1720         virtio_get_hwaddr(hw);
1721         rte_ether_addr_copy((struct rte_ether_addr *)hw->mac_addr,
1722                         &eth_dev->data->mac_addrs[0]);
1723         PMD_INIT_LOG(DEBUG,
1724                      "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1725                      hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1726                      hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1727
1728         if (hw->speed == SPEED_UNKNOWN) {
1729                 if (vtpci_with_feature(hw, VIRTIO_NET_F_SPEED_DUPLEX)) {
1730                         config = &local_config;
1731                         vtpci_read_dev_config(hw,
1732                                 offsetof(struct virtio_net_config, speed),
1733                                 &config->speed, sizeof(config->speed));
1734                         vtpci_read_dev_config(hw,
1735                                 offsetof(struct virtio_net_config, duplex),
1736                                 &config->duplex, sizeof(config->duplex));
1737                         hw->speed = config->speed;
1738                         hw->duplex = config->duplex;
1739                 }
1740         }
1741         if (hw->speed == SPEED_UNKNOWN)
1742                 hw->speed = ETH_SPEED_NUM_10G;
1743         if (hw->duplex == DUPLEX_UNKNOWN)
1744                 hw->duplex = ETH_LINK_FULL_DUPLEX;
1745         PMD_INIT_LOG(DEBUG, "link speed = %d, duplex = %d",
1746                 hw->speed, hw->duplex);
1747         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1748                 config = &local_config;
1749
1750                 vtpci_read_dev_config(hw,
1751                         offsetof(struct virtio_net_config, mac),
1752                         &config->mac, sizeof(config->mac));
1753
1754                 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1755                         vtpci_read_dev_config(hw,
1756                                 offsetof(struct virtio_net_config, status),
1757                                 &config->status, sizeof(config->status));
1758                 } else {
1759                         PMD_INIT_LOG(DEBUG,
1760                                      "VIRTIO_NET_F_STATUS is not supported");
1761                         config->status = 0;
1762                 }
1763
1764                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1765                         vtpci_read_dev_config(hw,
1766                                 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1767                                 &config->max_virtqueue_pairs,
1768                                 sizeof(config->max_virtqueue_pairs));
1769                 } else {
1770                         PMD_INIT_LOG(DEBUG,
1771                                      "VIRTIO_NET_F_MQ is not supported");
1772                         config->max_virtqueue_pairs = 1;
1773                 }
1774
1775                 hw->max_queue_pairs = config->max_virtqueue_pairs;
1776
1777                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1778                         vtpci_read_dev_config(hw,
1779                                 offsetof(struct virtio_net_config, mtu),
1780                                 &config->mtu,
1781                                 sizeof(config->mtu));
1782
1783                         /*
1784                          * MTU value has already been checked at negotiation
1785                          * time, but check again in case it has changed since
1786                          * then, which should not happen.
1787                          */
1788                         if (config->mtu < RTE_ETHER_MIN_MTU) {
1789                                 PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1790                                                 config->mtu);
1791                                 return -1;
1792                         }
1793
1794                         hw->max_mtu = config->mtu;
1795                         /* Set initial MTU to maximum one supported by vhost */
1796                         eth_dev->data->mtu = config->mtu;
1797
1798                 } else {
1799                         hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1800                                 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1801                 }
1802
1803                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1804                                 config->max_virtqueue_pairs);
1805                 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1806                 PMD_INIT_LOG(DEBUG,
1807                                 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1808                                 config->mac[0], config->mac[1],
1809                                 config->mac[2], config->mac[3],
1810                                 config->mac[4], config->mac[5]);
1811         } else {
1812                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1813                 hw->max_queue_pairs = 1;
1814                 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1815                         VLAN_TAG_LEN - hw->vtnet_hdr_size;
1816         }
1817
1818         ret = virtio_alloc_queues(eth_dev);
1819         if (ret < 0)
1820                 return ret;
1821
1822         if (eth_dev->data->dev_conf.intr_conf.rxq) {
1823                 if (virtio_configure_intr(eth_dev) < 0) {
1824                         PMD_INIT_LOG(ERR, "failed to configure interrupt");
1825                         virtio_free_queues(hw);
1826                         return -1;
1827                 }
1828         }
1829
1830         vtpci_reinit_complete(hw);
1831
1832         if (pci_dev)
1833                 PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1834                         eth_dev->data->port_id, pci_dev->id.vendor_id,
1835                         pci_dev->id.device_id);
1836
1837         return 0;
1838 }
1839
1840 /*
1841  * Remap the PCI device again (IO port map for legacy device and
1842  * memory map for modern device), so that the secondary process
1843  * could have the PCI initiated correctly.
1844  */
1845 static int
1846 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1847 {
1848         if (hw->modern) {
1849                 /*
1850                  * We don't have to re-parse the PCI config space, since
1851                  * rte_pci_map_device() makes sure the mapped address
1852                  * in secondary process would equal to the one mapped in
1853                  * the primary process: error will be returned if that
1854                  * requirement is not met.
1855                  *
1856                  * That said, we could simply reuse all cap pointers
1857                  * (such as dev_cfg, common_cfg, etc.) parsed from the
1858                  * primary process, which is stored in shared memory.
1859                  */
1860                 if (rte_pci_map_device(pci_dev)) {
1861                         PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1862                         return -1;
1863                 }
1864         } else {
1865                 if (rte_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1866                         return -1;
1867         }
1868
1869         return 0;
1870 }
1871
1872 static void
1873 virtio_set_vtpci_ops(struct virtio_hw *hw)
1874 {
1875 #ifdef RTE_VIRTIO_USER
1876         if (hw->virtio_user_dev)
1877                 VTPCI_OPS(hw) = &virtio_user_ops;
1878         else
1879 #endif
1880         if (hw->modern)
1881                 VTPCI_OPS(hw) = &modern_ops;
1882         else
1883                 VTPCI_OPS(hw) = &legacy_ops;
1884 }
1885
1886 /*
1887  * This function is based on probe() function in virtio_pci.c
1888  * It returns 0 on success.
1889  */
1890 int
1891 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1892 {
1893         struct virtio_hw *hw = eth_dev->data->dev_private;
1894         uint32_t speed = SPEED_UNKNOWN;
1895         int vectorized = 0;
1896         int ret;
1897
1898         if (sizeof(struct virtio_net_hdr_mrg_rxbuf) > RTE_PKTMBUF_HEADROOM) {
1899                 PMD_INIT_LOG(ERR,
1900                         "Not sufficient headroom required = %d, avail = %d",
1901                         (int)sizeof(struct virtio_net_hdr_mrg_rxbuf),
1902                         RTE_PKTMBUF_HEADROOM);
1903
1904                 return -1;
1905         }
1906
1907         eth_dev->dev_ops = &virtio_eth_dev_ops;
1908         eth_dev->rx_descriptor_done = virtio_dev_rx_queue_done;
1909
1910         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1911                 if (!hw->virtio_user_dev) {
1912                         ret = virtio_remap_pci(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1913                         if (ret)
1914                                 return ret;
1915                 }
1916
1917                 virtio_set_vtpci_ops(hw);
1918                 set_rxtx_funcs(eth_dev);
1919
1920                 return 0;
1921         }
1922         ret = virtio_dev_devargs_parse(eth_dev->device->devargs,
1923                  NULL, &speed, &vectorized);
1924         if (ret < 0)
1925                 return ret;
1926         hw->speed = speed;
1927
1928         /* Allocate memory for storing MAC addresses */
1929         eth_dev->data->mac_addrs = rte_zmalloc("virtio",
1930                                 VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN, 0);
1931         if (eth_dev->data->mac_addrs == NULL) {
1932                 PMD_INIT_LOG(ERR,
1933                         "Failed to allocate %d bytes needed to store MAC addresses",
1934                         VIRTIO_MAX_MAC_ADDRS * RTE_ETHER_ADDR_LEN);
1935                 return -ENOMEM;
1936         }
1937
1938         hw->port_id = eth_dev->data->port_id;
1939         /* For virtio_user case the hw->virtio_user_dev is populated by
1940          * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1941          */
1942         if (!hw->virtio_user_dev) {
1943                 ret = vtpci_init(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1944                 if (ret)
1945                         goto err_vtpci_init;
1946         }
1947
1948         rte_spinlock_init(&hw->state_lock);
1949
1950         /* reset device and negotiate default features */
1951         ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1952         if (ret < 0)
1953                 goto err_virtio_init;
1954
1955         if (vectorized) {
1956                 if (!vtpci_packed_queue(hw)) {
1957                         hw->use_vec_rx = 1;
1958                 } else {
1959 #if !defined(CC_AVX512_SUPPORT)
1960                         PMD_DRV_LOG(INFO,
1961                                 "building environment do not support packed ring vectorized");
1962 #else
1963                         hw->use_vec_rx = 1;
1964                         hw->use_vec_tx = 1;
1965 #endif
1966                 }
1967         }
1968
1969         hw->opened = true;
1970
1971         return 0;
1972
1973 err_virtio_init:
1974         if (!hw->virtio_user_dev) {
1975                 rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev));
1976                 if (!hw->modern)
1977                         rte_pci_ioport_unmap(VTPCI_IO(hw));
1978         }
1979 err_vtpci_init:
1980         rte_free(eth_dev->data->mac_addrs);
1981         eth_dev->data->mac_addrs = NULL;
1982         return ret;
1983 }
1984
1985 static int
1986 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1987 {
1988         PMD_INIT_FUNC_TRACE();
1989
1990         if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1991                 return 0;
1992
1993         virtio_dev_stop(eth_dev);
1994         virtio_dev_close(eth_dev);
1995
1996         eth_dev->dev_ops = NULL;
1997         eth_dev->tx_pkt_burst = NULL;
1998         eth_dev->rx_pkt_burst = NULL;
1999
2000         PMD_INIT_LOG(DEBUG, "dev_uninit completed");
2001
2002         return 0;
2003 }
2004
2005
2006 static int vdpa_check_handler(__rte_unused const char *key,
2007                 const char *value, void *ret_val)
2008 {
2009         if (strcmp(value, "1") == 0)
2010                 *(int *)ret_val = 1;
2011         else
2012                 *(int *)ret_val = 0;
2013
2014         return 0;
2015 }
2016
2017
2018 static uint32_t
2019 virtio_dev_speed_capa_get(uint32_t speed)
2020 {
2021         switch (speed) {
2022         case ETH_SPEED_NUM_10G:
2023                 return ETH_LINK_SPEED_10G;
2024         case ETH_SPEED_NUM_20G:
2025                 return ETH_LINK_SPEED_20G;
2026         case ETH_SPEED_NUM_25G:
2027                 return ETH_LINK_SPEED_25G;
2028         case ETH_SPEED_NUM_40G:
2029                 return ETH_LINK_SPEED_40G;
2030         case ETH_SPEED_NUM_50G:
2031                 return ETH_LINK_SPEED_50G;
2032         case ETH_SPEED_NUM_56G:
2033                 return ETH_LINK_SPEED_56G;
2034         case ETH_SPEED_NUM_100G:
2035                 return ETH_LINK_SPEED_100G;
2036         default:
2037                 return 0;
2038         }
2039 }
2040
2041 static int vectorized_check_handler(__rte_unused const char *key,
2042                 const char *value, void *ret_val)
2043 {
2044         if (strcmp(value, "1") == 0)
2045                 *(int *)ret_val = 1;
2046         else
2047                 *(int *)ret_val = 0;
2048
2049         return 0;
2050 }
2051
2052 #define VIRTIO_ARG_SPEED      "speed"
2053 #define VIRTIO_ARG_VDPA       "vdpa"
2054 #define VIRTIO_ARG_VECTORIZED "vectorized"
2055
2056
2057 static int
2058 link_speed_handler(const char *key __rte_unused,
2059                 const char *value, void *ret_val)
2060 {
2061         uint32_t val;
2062         if (!value || !ret_val)
2063                 return -EINVAL;
2064         val = strtoul(value, NULL, 0);
2065         /* validate input */
2066         if (virtio_dev_speed_capa_get(val) == 0)
2067                 return -EINVAL;
2068         *(uint32_t *)ret_val = val;
2069
2070         return 0;
2071 }
2072
2073
2074 static int
2075 virtio_dev_devargs_parse(struct rte_devargs *devargs, int *vdpa,
2076         uint32_t *speed, int *vectorized)
2077 {
2078         struct rte_kvargs *kvlist;
2079         int ret = 0;
2080
2081         if (devargs == NULL)
2082                 return 0;
2083
2084         kvlist = rte_kvargs_parse(devargs->args, NULL);
2085         if (kvlist == NULL) {
2086                 PMD_INIT_LOG(ERR, "error when parsing param");
2087                 return 0;
2088         }
2089         if (vdpa && rte_kvargs_count(kvlist, VIRTIO_ARG_VDPA) == 1) {
2090                 /* vdpa mode selected when there's a key-value pair:
2091                  * vdpa=1
2092                  */
2093                 ret = rte_kvargs_process(kvlist, VIRTIO_ARG_VDPA,
2094                                 vdpa_check_handler, vdpa);
2095                 if (ret < 0) {
2096                         PMD_INIT_LOG(ERR, "Failed to parse %s",
2097                                 VIRTIO_ARG_VDPA);
2098                         goto exit;
2099                 }
2100         }
2101         if (speed && rte_kvargs_count(kvlist, VIRTIO_ARG_SPEED) == 1) {
2102                 ret = rte_kvargs_process(kvlist,
2103                                         VIRTIO_ARG_SPEED,
2104                                         link_speed_handler, speed);
2105                 if (ret < 0) {
2106                         PMD_INIT_LOG(ERR, "Failed to parse %s",
2107                                         VIRTIO_ARG_SPEED);
2108                         goto exit;
2109                 }
2110         }
2111
2112         if (vectorized &&
2113                 rte_kvargs_count(kvlist, VIRTIO_ARG_VECTORIZED) == 1) {
2114                 ret = rte_kvargs_process(kvlist,
2115                                 VIRTIO_ARG_VECTORIZED,
2116                                 vectorized_check_handler, vectorized);
2117                 if (ret < 0) {
2118                         PMD_INIT_LOG(ERR, "Failed to parse %s",
2119                                         VIRTIO_ARG_VECTORIZED);
2120                         goto exit;
2121                 }
2122         }
2123
2124 exit:
2125         rte_kvargs_free(kvlist);
2126         return ret;
2127 }
2128
2129 static int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2130         struct rte_pci_device *pci_dev)
2131 {
2132         int vdpa = 0;
2133         int ret = 0;
2134
2135         ret = virtio_dev_devargs_parse(pci_dev->device.devargs, &vdpa, NULL,
2136                 NULL);
2137         if (ret < 0) {
2138                 PMD_INIT_LOG(ERR, "devargs parsing is failed");
2139                 return ret;
2140         }
2141         /* virtio pmd skips probe if device needs to work in vdpa mode */
2142         if (vdpa == 1)
2143                 return 1;
2144
2145         return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw),
2146                 eth_virtio_dev_init);
2147 }
2148
2149 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev)
2150 {
2151         int ret;
2152
2153         ret = rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
2154         /* Port has already been released by close. */
2155         if (ret == -ENODEV)
2156                 ret = 0;
2157         return ret;
2158 }
2159
2160 static struct rte_pci_driver rte_virtio_pmd = {
2161         .driver = {
2162                 .name = "net_virtio",
2163         },
2164         .id_table = pci_id_virtio_map,
2165         .drv_flags = 0,
2166         .probe = eth_virtio_pci_probe,
2167         .remove = eth_virtio_pci_remove,
2168 };
2169
2170 RTE_INIT(rte_virtio_pmd_init)
2171 {
2172         rte_eal_iopl_init();
2173         rte_pci_register(&rte_virtio_pmd);
2174 }
2175
2176 static bool
2177 rx_offload_enabled(struct virtio_hw *hw)
2178 {
2179         return vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM) ||
2180                 vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2181                 vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6);
2182 }
2183
2184 static bool
2185 tx_offload_enabled(struct virtio_hw *hw)
2186 {
2187         return vtpci_with_feature(hw, VIRTIO_NET_F_CSUM) ||
2188                 vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO4) ||
2189                 vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO6);
2190 }
2191
2192 /*
2193  * Configure virtio device
2194  * It returns 0 on success.
2195  */
2196 static int
2197 virtio_dev_configure(struct rte_eth_dev *dev)
2198 {
2199         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2200         const struct rte_eth_txmode *txmode = &dev->data->dev_conf.txmode;
2201         struct virtio_hw *hw = dev->data->dev_private;
2202         uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
2203                 hw->vtnet_hdr_size;
2204         uint64_t rx_offloads = rxmode->offloads;
2205         uint64_t tx_offloads = txmode->offloads;
2206         uint64_t req_features;
2207         int ret;
2208
2209         PMD_INIT_LOG(DEBUG, "configure");
2210         req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
2211
2212         if (rxmode->mq_mode != ETH_MQ_RX_NONE) {
2213                 PMD_DRV_LOG(ERR,
2214                         "Unsupported Rx multi queue mode %d",
2215                         rxmode->mq_mode);
2216                 return -EINVAL;
2217         }
2218
2219         if (txmode->mq_mode != ETH_MQ_TX_NONE) {
2220                 PMD_DRV_LOG(ERR,
2221                         "Unsupported Tx multi queue mode %d",
2222                         txmode->mq_mode);
2223                 return -EINVAL;
2224         }
2225
2226         if (dev->data->dev_conf.intr_conf.rxq) {
2227                 ret = virtio_init_device(dev, hw->req_guest_features);
2228                 if (ret < 0)
2229                         return ret;
2230         }
2231
2232         if (rxmode->max_rx_pkt_len > hw->max_mtu + ether_hdr_len)
2233                 req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
2234
2235         if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2236                            DEV_RX_OFFLOAD_TCP_CKSUM))
2237                 req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
2238
2239         if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
2240                 req_features |=
2241                         (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2242                         (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2243
2244         if (tx_offloads & (DEV_TX_OFFLOAD_UDP_CKSUM |
2245                            DEV_TX_OFFLOAD_TCP_CKSUM))
2246                 req_features |= (1ULL << VIRTIO_NET_F_CSUM);
2247
2248         if (tx_offloads & DEV_TX_OFFLOAD_TCP_TSO)
2249                 req_features |=
2250                         (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2251                         (1ULL << VIRTIO_NET_F_HOST_TSO6);
2252
2253         /* if request features changed, reinit the device */
2254         if (req_features != hw->req_guest_features) {
2255                 ret = virtio_init_device(dev, req_features);
2256                 if (ret < 0)
2257                         return ret;
2258         }
2259
2260         if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2261                             DEV_RX_OFFLOAD_TCP_CKSUM)) &&
2262                 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
2263                 PMD_DRV_LOG(ERR,
2264                         "rx checksum not available on this host");
2265                 return -ENOTSUP;
2266         }
2267
2268         if ((rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) &&
2269                 (!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2270                  !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
2271                 PMD_DRV_LOG(ERR,
2272                         "Large Receive Offload not available on this host");
2273                 return -ENOTSUP;
2274         }
2275
2276         /* start control queue */
2277         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
2278                 virtio_dev_cq_start(dev);
2279
2280         if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2281                 hw->vlan_strip = 1;
2282
2283         if ((rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
2284             && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2285                 PMD_DRV_LOG(ERR,
2286                             "vlan filtering not available on this host");
2287                 return -ENOTSUP;
2288         }
2289
2290         hw->has_tx_offload = tx_offload_enabled(hw);
2291         hw->has_rx_offload = rx_offload_enabled(hw);
2292
2293         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2294                 /* Enable vector (0) for Link State Intrerrupt */
2295                 if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
2296                                 VIRTIO_MSI_NO_VECTOR) {
2297                         PMD_DRV_LOG(ERR, "failed to set config vector");
2298                         return -EBUSY;
2299                 }
2300
2301         if (vtpci_packed_queue(hw)) {
2302 #if defined(RTE_ARCH_X86_64) && defined(CC_AVX512_SUPPORT)
2303                 if ((hw->use_vec_rx || hw->use_vec_tx) &&
2304                     (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F) ||
2305                      !vtpci_with_feature(hw, VIRTIO_F_IN_ORDER) ||
2306                      !vtpci_with_feature(hw, VIRTIO_F_VERSION_1))) {
2307                         PMD_DRV_LOG(INFO,
2308                                 "disabled packed ring vectorized path for requirements not met");
2309                         hw->use_vec_rx = 0;
2310                         hw->use_vec_tx = 0;
2311                 }
2312 #else
2313                 hw->use_vec_rx = 0;
2314                 hw->use_vec_tx = 0;
2315 #endif
2316
2317                 if (hw->use_vec_rx) {
2318                         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2319                                 PMD_DRV_LOG(INFO,
2320                                         "disabled packed ring vectorized rx for mrg_rxbuf enabled");
2321                                 hw->use_vec_rx = 0;
2322                         }
2323
2324                         if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) {
2325                                 PMD_DRV_LOG(INFO,
2326                                         "disabled packed ring vectorized rx for TCP_LRO enabled");
2327                                 hw->use_vec_rx = 0;
2328                         }
2329                 }
2330         } else {
2331                 if (vtpci_with_feature(hw, VIRTIO_F_IN_ORDER)) {
2332                         hw->use_inorder_tx = 1;
2333                         hw->use_inorder_rx = 1;
2334                         hw->use_vec_rx = 0;
2335                 }
2336
2337                 if (hw->use_vec_rx) {
2338 #if defined RTE_ARCH_ARM64 || defined RTE_ARCH_ARM
2339                         if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
2340                                 PMD_DRV_LOG(INFO,
2341                                         "disabled split ring vectorized path for requirement not met");
2342                                 hw->use_vec_rx = 0;
2343                         }
2344 #endif
2345                         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2346                                 PMD_DRV_LOG(INFO,
2347                                         "disabled split ring vectorized rx for mrg_rxbuf enabled");
2348                                 hw->use_vec_rx = 0;
2349                         }
2350
2351                         if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2352                                            DEV_RX_OFFLOAD_TCP_CKSUM |
2353                                            DEV_RX_OFFLOAD_TCP_LRO |
2354                                            DEV_RX_OFFLOAD_VLAN_STRIP)) {
2355                                 PMD_DRV_LOG(INFO,
2356                                         "disabled split ring vectorized rx for offloading enabled");
2357                                 hw->use_vec_rx = 0;
2358                         }
2359                 }
2360         }
2361
2362         return 0;
2363 }
2364
2365
2366 static int
2367 virtio_dev_start(struct rte_eth_dev *dev)
2368 {
2369         uint16_t nb_queues, i;
2370         struct virtnet_rx *rxvq;
2371         struct virtnet_tx *txvq __rte_unused;
2372         struct virtio_hw *hw = dev->data->dev_private;
2373         int ret;
2374
2375         /* Finish the initialization of the queues */
2376         for (i = 0; i < dev->data->nb_rx_queues; i++) {
2377                 ret = virtio_dev_rx_queue_setup_finish(dev, i);
2378                 if (ret < 0)
2379                         return ret;
2380         }
2381         for (i = 0; i < dev->data->nb_tx_queues; i++) {
2382                 ret = virtio_dev_tx_queue_setup_finish(dev, i);
2383                 if (ret < 0)
2384                         return ret;
2385         }
2386
2387         /* check if lsc interrupt feature is enabled */
2388         if (dev->data->dev_conf.intr_conf.lsc) {
2389                 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
2390                         PMD_DRV_LOG(ERR, "link status not supported by host");
2391                         return -ENOTSUP;
2392                 }
2393         }
2394
2395         /* Enable uio/vfio intr/eventfd mapping: althrough we already did that
2396          * in device configure, but it could be unmapped  when device is
2397          * stopped.
2398          */
2399         if (dev->data->dev_conf.intr_conf.lsc ||
2400             dev->data->dev_conf.intr_conf.rxq) {
2401                 virtio_intr_disable(dev);
2402
2403                 /* Setup interrupt callback  */
2404                 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2405                         rte_intr_callback_register(dev->intr_handle,
2406                                                    virtio_interrupt_handler,
2407                                                    dev);
2408
2409                 if (virtio_intr_enable(dev) < 0) {
2410                         PMD_DRV_LOG(ERR, "interrupt enable failed");
2411                         return -EIO;
2412                 }
2413         }
2414
2415         /*Notify the backend
2416          *Otherwise the tap backend might already stop its queue due to fullness.
2417          *vhost backend will have no chance to be waked up
2418          */
2419         nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
2420         if (hw->max_queue_pairs > 1) {
2421                 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
2422                         return -EINVAL;
2423         }
2424
2425         PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
2426
2427         for (i = 0; i < dev->data->nb_rx_queues; i++) {
2428                 rxvq = dev->data->rx_queues[i];
2429                 /* Flush the old packets */
2430                 virtqueue_rxvq_flush(rxvq->vq);
2431                 virtqueue_notify(rxvq->vq);
2432         }
2433
2434         for (i = 0; i < dev->data->nb_tx_queues; i++) {
2435                 txvq = dev->data->tx_queues[i];
2436                 virtqueue_notify(txvq->vq);
2437         }
2438
2439         PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
2440
2441         for (i = 0; i < dev->data->nb_rx_queues; i++) {
2442                 rxvq = dev->data->rx_queues[i];
2443                 VIRTQUEUE_DUMP(rxvq->vq);
2444         }
2445
2446         for (i = 0; i < dev->data->nb_tx_queues; i++) {
2447                 txvq = dev->data->tx_queues[i];
2448                 VIRTQUEUE_DUMP(txvq->vq);
2449         }
2450
2451         set_rxtx_funcs(dev);
2452         hw->started = true;
2453
2454         /* Initialize Link state */
2455         virtio_dev_link_update(dev, 0);
2456
2457         return 0;
2458 }
2459
2460 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
2461 {
2462         struct virtio_hw *hw = dev->data->dev_private;
2463         uint16_t nr_vq = virtio_get_nr_vq(hw);
2464         const char *type __rte_unused;
2465         unsigned int i, mbuf_num = 0;
2466         struct virtqueue *vq;
2467         struct rte_mbuf *buf;
2468         int queue_type;
2469
2470         if (hw->vqs == NULL)
2471                 return;
2472
2473         for (i = 0; i < nr_vq; i++) {
2474                 vq = hw->vqs[i];
2475                 if (!vq)
2476                         continue;
2477
2478                 queue_type = virtio_get_queue_type(hw, i);
2479                 if (queue_type == VTNET_RQ)
2480                         type = "rxq";
2481                 else if (queue_type == VTNET_TQ)
2482                         type = "txq";
2483                 else
2484                         continue;
2485
2486                 PMD_INIT_LOG(DEBUG,
2487                         "Before freeing %s[%d] used and unused buf",
2488                         type, i);
2489                 VIRTQUEUE_DUMP(vq);
2490
2491                 while ((buf = virtqueue_detach_unused(vq)) != NULL) {
2492                         rte_pktmbuf_free(buf);
2493                         mbuf_num++;
2494                 }
2495
2496                 PMD_INIT_LOG(DEBUG,
2497                         "After freeing %s[%d] used and unused buf",
2498                         type, i);
2499                 VIRTQUEUE_DUMP(vq);
2500         }
2501
2502         PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
2503 }
2504
2505 /*
2506  * Stop device: disable interrupt and mark link down
2507  */
2508 static void
2509 virtio_dev_stop(struct rte_eth_dev *dev)
2510 {
2511         struct virtio_hw *hw = dev->data->dev_private;
2512         struct rte_eth_link link;
2513         struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
2514
2515         PMD_INIT_LOG(DEBUG, "stop");
2516
2517         rte_spinlock_lock(&hw->state_lock);
2518         if (!hw->started)
2519                 goto out_unlock;
2520         hw->started = false;
2521
2522         if (intr_conf->lsc || intr_conf->rxq) {
2523                 virtio_intr_disable(dev);
2524
2525                 /* Reset interrupt callback  */
2526                 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
2527                         rte_intr_callback_unregister(dev->intr_handle,
2528                                                      virtio_interrupt_handler,
2529                                                      dev);
2530                 }
2531         }
2532
2533         memset(&link, 0, sizeof(link));
2534         rte_eth_linkstatus_set(dev, &link);
2535 out_unlock:
2536         rte_spinlock_unlock(&hw->state_lock);
2537 }
2538
2539 static int
2540 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2541 {
2542         struct rte_eth_link link;
2543         uint16_t status;
2544         struct virtio_hw *hw = dev->data->dev_private;
2545
2546         memset(&link, 0, sizeof(link));
2547         link.link_duplex = hw->duplex;
2548         link.link_speed  = hw->speed;
2549         link.link_autoneg = ETH_LINK_AUTONEG;
2550
2551         if (!hw->started) {
2552                 link.link_status = ETH_LINK_DOWN;
2553         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2554                 PMD_INIT_LOG(DEBUG, "Get link status from hw");
2555                 vtpci_read_dev_config(hw,
2556                                 offsetof(struct virtio_net_config, status),
2557                                 &status, sizeof(status));
2558                 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2559                         link.link_status = ETH_LINK_DOWN;
2560                         PMD_INIT_LOG(DEBUG, "Port %d is down",
2561                                      dev->data->port_id);
2562                 } else {
2563                         link.link_status = ETH_LINK_UP;
2564                         PMD_INIT_LOG(DEBUG, "Port %d is up",
2565                                      dev->data->port_id);
2566                 }
2567         } else {
2568                 link.link_status = ETH_LINK_UP;
2569         }
2570
2571         return rte_eth_linkstatus_set(dev, &link);
2572 }
2573
2574 static int
2575 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2576 {
2577         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2578         struct virtio_hw *hw = dev->data->dev_private;
2579         uint64_t offloads = rxmode->offloads;
2580
2581         if (mask & ETH_VLAN_FILTER_MASK) {
2582                 if ((offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2583                                 !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2584
2585                         PMD_DRV_LOG(NOTICE,
2586                                 "vlan filtering not available on this host");
2587
2588                         return -ENOTSUP;
2589                 }
2590         }
2591
2592         if (mask & ETH_VLAN_STRIP_MASK)
2593                 hw->vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
2594
2595         return 0;
2596 }
2597
2598 static int
2599 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2600 {
2601         uint64_t tso_mask, host_features;
2602         struct virtio_hw *hw = dev->data->dev_private;
2603         dev_info->speed_capa = virtio_dev_speed_capa_get(hw->speed);
2604
2605         dev_info->max_rx_queues =
2606                 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2607         dev_info->max_tx_queues =
2608                 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2609         dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2610         dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2611         dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2612
2613         host_features = VTPCI_OPS(hw)->get_features(hw);
2614         dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
2615         dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2616         if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2617                 dev_info->rx_offload_capa |=
2618                         DEV_RX_OFFLOAD_TCP_CKSUM |
2619                         DEV_RX_OFFLOAD_UDP_CKSUM;
2620         }
2621         if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN))
2622                 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_FILTER;
2623         tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2624                 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2625         if ((host_features & tso_mask) == tso_mask)
2626                 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2627
2628         dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
2629                                     DEV_TX_OFFLOAD_VLAN_INSERT;
2630         if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2631                 dev_info->tx_offload_capa |=
2632                         DEV_TX_OFFLOAD_UDP_CKSUM |
2633                         DEV_TX_OFFLOAD_TCP_CKSUM;
2634         }
2635         tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2636                 (1ULL << VIRTIO_NET_F_HOST_TSO6);
2637         if ((host_features & tso_mask) == tso_mask)
2638                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2639
2640         return 0;
2641 }
2642
2643 /*
2644  * It enables testpmd to collect per queue stats.
2645  */
2646 static int
2647 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2648 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2649 __rte_unused uint8_t is_rx)
2650 {
2651         return 0;
2652 }
2653
2654 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
2655 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
2656 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio-pci");
2657 RTE_LOG_REGISTER(virtio_logtype_init, pmd.net.virtio.init, NOTICE);
2658 RTE_LOG_REGISTER(virtio_logtype_driver, pmd.net.virtio.driver, NOTICE);