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 <ethdev_driver.h>
  12 #include <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 #include <rte_vect.h>
  27
  28 #include <rte_memory.h>
  29 #include <rte_eal.h>
  30 #include <rte_dev.h>
  31 #include <rte_cycles.h>
  32 #include <rte_kvargs.h>
  33
  34 #include "virtio_ethdev.h"
  35 #include "virtio_pci.h"
  36 #include "virtio_logs.h"
  37 #include "virtqueue.h"
  38 #include "virtio_rxtx.h"
  39 #include "virtio_user/virtio_user_dev.h"
  40
  41 static int  virtio_dev_configure(struct rte_eth_dev *dev);
  42 static int  virtio_dev_start(struct rte_eth_dev *dev);
  43 static int virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
  44 static int virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
  45 static int virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
  46 static int virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
  47 static uint32_t virtio_dev_speed_capa_get(uint32_t speed);
  48 static int virtio_dev_devargs_parse(struct rte_devargs *devargs,
  49         uint32_t *speed,
  50         int *vectorized);
  51 static int virtio_dev_info_get(struct rte_eth_dev *dev,
  52                                 struct rte_eth_dev_info *dev_info);
  53 static int virtio_dev_link_update(struct rte_eth_dev *dev,
  54         int wait_to_complete);
  55 static int virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask);
  56
  57 static void virtio_set_hwaddr(struct virtio_hw *hw);
  58 static void virtio_get_hwaddr(struct virtio_hw *hw);
  59
  60 static int virtio_dev_stats_get(struct rte_eth_dev *dev,
  61                                  struct rte_eth_stats *stats);
  62 static int virtio_dev_xstats_get(struct rte_eth_dev *dev,
  63                                  struct rte_eth_xstat *xstats, unsigned n);
  64 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
  65                                        struct rte_eth_xstat_name *xstats_names,
  66                                        unsigned limit);
  67 static int virtio_dev_stats_reset(struct rte_eth_dev *dev);
  68 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
  69 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
  70                                 uint16_t vlan_id, int on);
  71 static int virtio_mac_addr_add(struct rte_eth_dev *dev,
  72                                 struct rte_ether_addr *mac_addr,
  73                                 uint32_t index, uint32_t vmdq);
  74 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
  75 static int virtio_mac_addr_set(struct rte_eth_dev *dev,
  76                                 struct rte_ether_addr *mac_addr);
  77
  78 static int virtio_intr_disable(struct rte_eth_dev *dev);
  79
  80 static int virtio_dev_queue_stats_mapping_set(
  81         struct rte_eth_dev *eth_dev,
  82         uint16_t queue_id,
  83         uint8_t stat_idx,
  84         uint8_t is_rx);
  85
  86 static void virtio_notify_peers(struct rte_eth_dev *dev);
  87 static void virtio_ack_link_announce(struct rte_eth_dev *dev);
  88
  89 struct rte_virtio_xstats_name_off {
  90         char name[RTE_ETH_XSTATS_NAME_SIZE];
  91         unsigned offset;
  92 };
  93
  94 /* [rt]x_qX_ is prepended to the name string here */
  95 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
  96         {"good_packets",           offsetof(struct virtnet_rx, stats.packets)},
  97         {"good_bytes",             offsetof(struct virtnet_rx, stats.bytes)},
  98         {"errors",                 offsetof(struct virtnet_rx, stats.errors)},
  99         {"multicast_packets",      offsetof(struct virtnet_rx, stats.multicast)},
 100         {"broadcast_packets",      offsetof(struct virtnet_rx, stats.broadcast)},
 101         {"undersize_packets",      offsetof(struct virtnet_rx, stats.size_bins[0])},
 102         {"size_64_packets",        offsetof(struct virtnet_rx, stats.size_bins[1])},
 103         {"size_65_127_packets",    offsetof(struct virtnet_rx, stats.size_bins[2])},
 104         {"size_128_255_packets",   offsetof(struct virtnet_rx, stats.size_bins[3])},
 105         {"size_256_511_packets",   offsetof(struct virtnet_rx, stats.size_bins[4])},
 106         {"size_512_1023_packets",  offsetof(struct virtnet_rx, stats.size_bins[5])},
 107         {"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
 108         {"size_1519_max_packets",  offsetof(struct virtnet_rx, stats.size_bins[7])},
 109 };
 110
 111 /* [rt]x_qX_ is prepended to the name string here */
 112 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
 113         {"good_packets",           offsetof(struct virtnet_tx, stats.packets)},
 114         {"good_bytes",             offsetof(struct virtnet_tx, stats.bytes)},
 115         {"multicast_packets",      offsetof(struct virtnet_tx, stats.multicast)},
 116         {"broadcast_packets",      offsetof(struct virtnet_tx, stats.broadcast)},
 117         {"undersize_packets",      offsetof(struct virtnet_tx, stats.size_bins[0])},
 118         {"size_64_packets",        offsetof(struct virtnet_tx, stats.size_bins[1])},
 119         {"size_65_127_packets",    offsetof(struct virtnet_tx, stats.size_bins[2])},
 120         {"size_128_255_packets",   offsetof(struct virtnet_tx, stats.size_bins[3])},
 121         {"size_256_511_packets",   offsetof(struct virtnet_tx, stats.size_bins[4])},
 122         {"size_512_1023_packets",  offsetof(struct virtnet_tx, stats.size_bins[5])},
 123         {"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
 124         {"size_1519_max_packets",  offsetof(struct virtnet_tx, stats.size_bins[7])},
 125 };
 126
 127 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
 128                             sizeof(rte_virtio_rxq_stat_strings[0]))
 129 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
 130                             sizeof(rte_virtio_txq_stat_strings[0]))
 131
 132 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
 133
 134 static struct virtio_pmd_ctrl *
 135 virtio_send_command_packed(struct virtnet_ctl *cvq,
 136                            struct virtio_pmd_ctrl *ctrl,
 137                            int *dlen, int pkt_num)
 138 {
 139         struct virtqueue *vq = cvq->vq;
 140         int head;
 141         struct vring_packed_desc *desc = vq->vq_packed.ring.desc;
 142         struct virtio_pmd_ctrl *result;
 143         uint16_t flags;
 144         int sum = 0;
 145         int nb_descs = 0;
 146         int k;
 147
 148         /*
 149          * Format is enforced in qemu code:
 150          * One TX packet for header;
 151          * At least one TX packet per argument;
 152          * One RX packet for ACK.
 153          */
 154         head = vq->vq_avail_idx;
 155         flags = vq->vq_packed.cached_flags;
 156         desc[head].addr = cvq->virtio_net_hdr_mem;
 157         desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
 158         vq->vq_free_cnt--;
 159         nb_descs++;
 160         if (++vq->vq_avail_idx >= vq->vq_nentries) {
 161                 vq->vq_avail_idx -= vq->vq_nentries;
 162                 vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
 163         }
 164
 165         for (k = 0; k < pkt_num; k++) {
 166                 desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
 167                         + sizeof(struct virtio_net_ctrl_hdr)
 168                         + sizeof(ctrl->status) + sizeof(uint8_t) * sum;
 169                 desc[vq->vq_avail_idx].len = dlen[k];
 170                 desc[vq->vq_avail_idx].flags = VRING_DESC_F_NEXT |
 171                         vq->vq_packed.cached_flags;
 172                 sum += dlen[k];
 173                 vq->vq_free_cnt--;
 174                 nb_descs++;
 175                 if (++vq->vq_avail_idx >= vq->vq_nentries) {
 176                         vq->vq_avail_idx -= vq->vq_nentries;
 177                         vq->vq_packed.cached_flags ^=
 178                                 VRING_PACKED_DESC_F_AVAIL_USED;
 179                 }
 180         }
 181
 182         desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
 183                 + sizeof(struct virtio_net_ctrl_hdr);
 184         desc[vq->vq_avail_idx].len = sizeof(ctrl->status);
 185         desc[vq->vq_avail_idx].flags = VRING_DESC_F_WRITE |
 186                 vq->vq_packed.cached_flags;
 187         vq->vq_free_cnt--;
 188         nb_descs++;
 189         if (++vq->vq_avail_idx >= vq->vq_nentries) {
 190                 vq->vq_avail_idx -= vq->vq_nentries;
 191                 vq->vq_packed.cached_flags ^= VRING_PACKED_DESC_F_AVAIL_USED;
 192         }
 193
 194         virtqueue_store_flags_packed(&desc[head], VRING_DESC_F_NEXT | flags,
 195                         vq->hw->weak_barriers);
 196
 197         virtio_wmb(vq->hw->weak_barriers);
 198         virtqueue_notify(vq);
 199
 200         /* wait for used desc in virtqueue
 201          * desc_is_used has a load-acquire or rte_io_rmb inside
 202          */
 203         while (!desc_is_used(&desc[head], vq))
 204                 usleep(100);
 205
 206         /* now get used descriptors */
 207         vq->vq_free_cnt += nb_descs;
 208         vq->vq_used_cons_idx += nb_descs;
 209         if (vq->vq_used_cons_idx >= vq->vq_nentries) {
 210                 vq->vq_used_cons_idx -= vq->vq_nentries;
 211                 vq->vq_packed.used_wrap_counter ^= 1;
 212         }
 213
 214         PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\n"
 215                         "vq->vq_avail_idx=%d\n"
 216                         "vq->vq_used_cons_idx=%d\n"
 217                         "vq->vq_packed.cached_flags=0x%x\n"
 218                         "vq->vq_packed.used_wrap_counter=%d\n",
 219                         vq->vq_free_cnt,
 220                         vq->vq_avail_idx,
 221                         vq->vq_used_cons_idx,
 222                         vq->vq_packed.cached_flags,
 223                         vq->vq_packed.used_wrap_counter);
 224
 225         result = cvq->virtio_net_hdr_mz->addr;
 226         return result;
 227 }
 228
 229 static struct virtio_pmd_ctrl *
 230 virtio_send_command_split(struct virtnet_ctl *cvq,
 231                           struct virtio_pmd_ctrl *ctrl,
 232                           int *dlen, int pkt_num)
 233 {
 234         struct virtio_pmd_ctrl *result;
 235         struct virtqueue *vq = cvq->vq;
 236         uint32_t head, i;
 237         int k, sum = 0;
 238
 239         head = vq->vq_desc_head_idx;
 240
 241         /*
 242          * Format is enforced in qemu code:
 243          * One TX packet for header;
 244          * At least one TX packet per argument;
 245          * One RX packet for ACK.
 246          */
 247         vq->vq_split.ring.desc[head].flags = VRING_DESC_F_NEXT;
 248         vq->vq_split.ring.desc[head].addr = cvq->virtio_net_hdr_mem;
 249         vq->vq_split.ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
 250         vq->vq_free_cnt--;
 251         i = vq->vq_split.ring.desc[head].next;
 252
 253         for (k = 0; k < pkt_num; k++) {
 254                 vq->vq_split.ring.desc[i].flags = VRING_DESC_F_NEXT;
 255                 vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
 256                         + sizeof(struct virtio_net_ctrl_hdr)
 257                         + sizeof(ctrl->status) + sizeof(uint8_t)*sum;
 258                 vq->vq_split.ring.desc[i].len = dlen[k];
 259                 sum += dlen[k];
 260                 vq->vq_free_cnt--;
 261                 i = vq->vq_split.ring.desc[i].next;
 262         }
 263
 264         vq->vq_split.ring.desc[i].flags = VRING_DESC_F_WRITE;
 265         vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
 266                         + sizeof(struct virtio_net_ctrl_hdr);
 267         vq->vq_split.ring.desc[i].len = sizeof(ctrl->status);
 268         vq->vq_free_cnt--;
 269
 270         vq->vq_desc_head_idx = vq->vq_split.ring.desc[i].next;
 271
 272         vq_update_avail_ring(vq, head);
 273         vq_update_avail_idx(vq);
 274
 275         PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
 276
 277         virtqueue_notify(vq);
 278
 279         while (virtqueue_nused(vq) == 0)
 280                 usleep(100);
 281
 282         while (virtqueue_nused(vq)) {
 283                 uint32_t idx, desc_idx, used_idx;
 284                 struct vring_used_elem *uep;
 285
 286                 used_idx = (uint32_t)(vq->vq_used_cons_idx
 287                                 & (vq->vq_nentries - 1));
 288                 uep = &vq->vq_split.ring.used->ring[used_idx];
 289                 idx = (uint32_t) uep->id;
 290                 desc_idx = idx;
 291
 292                 while (vq->vq_split.ring.desc[desc_idx].flags &
 293                                 VRING_DESC_F_NEXT) {
 294                         desc_idx = vq->vq_split.ring.desc[desc_idx].next;
 295                         vq->vq_free_cnt++;
 296                 }
 297
 298                 vq->vq_split.ring.desc[desc_idx].next = vq->vq_desc_head_idx;
 299                 vq->vq_desc_head_idx = idx;
 300
 301                 vq->vq_used_cons_idx++;
 302                 vq->vq_free_cnt++;
 303         }
 304
 305         PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
 306                         vq->vq_free_cnt, vq->vq_desc_head_idx);
 307
 308         result = cvq->virtio_net_hdr_mz->addr;
 309         return result;
 310 }
 311
 312 static int
 313 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
 314                     int *dlen, int pkt_num)
 315 {
 316         virtio_net_ctrl_ack status = ~0;
 317         struct virtio_pmd_ctrl *result;
 318         struct virtqueue *vq;
 319
 320         ctrl->status = status;
 321
 322         if (!cvq || !cvq->vq) {
 323                 PMD_INIT_LOG(ERR, "Control queue is not supported.");
 324                 return -1;
 325         }
 326
 327         rte_spinlock_lock(&cvq->lock);
 328         vq = cvq->vq;
 329
 330         PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
 331                 "vq->hw->cvq = %p vq = %p",
 332                 vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
 333
 334         if (vq->vq_free_cnt < pkt_num + 2 || pkt_num < 1) {
 335                 rte_spinlock_unlock(&cvq->lock);
 336                 return -1;
 337         }
 338
 339         memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
 340                 sizeof(struct virtio_pmd_ctrl));
 341
 342         if (vtpci_packed_queue(vq->hw))
 343                 result = virtio_send_command_packed(cvq, ctrl, dlen, pkt_num);
 344         else
 345                 result = virtio_send_command_split(cvq, ctrl, dlen, pkt_num);
 346
 347         rte_spinlock_unlock(&cvq->lock);
 348         return result->status;
 349 }
 350
 351 static int
 352 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
 353 {
 354         struct virtio_hw *hw = dev->data->dev_private;
 355         struct virtio_pmd_ctrl ctrl;
 356         int dlen[1];
 357         int ret;
 358
 359         ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
 360         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
 361         memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
 362
 363         dlen[0] = sizeof(uint16_t);
 364
 365         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 366         if (ret) {
 367                 PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
 368                           "failed, this is too late now...");
 369                 return -EINVAL;
 370         }
 371
 372         return 0;
 373 }
 374
 375 static void
 376 virtio_dev_queue_release(void *queue __rte_unused)
 377 {
 378         /* do nothing */
 379 }
 380
 381 static uint16_t
 382 virtio_get_nr_vq(struct virtio_hw *hw)
 383 {
 384         uint16_t nr_vq = hw->max_queue_pairs * 2;
 385
 386         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
 387                 nr_vq += 1;
 388
 389         return nr_vq;
 390 }
 391
 392 static void
 393 virtio_init_vring(struct virtqueue *vq)
 394 {
 395         int size = vq->vq_nentries;
 396         uint8_t *ring_mem = vq->vq_ring_virt_mem;
 397
 398         PMD_INIT_FUNC_TRACE();
 399
 400         memset(ring_mem, 0, vq->vq_ring_size);
 401
 402         vq->vq_used_cons_idx = 0;
 403         vq->vq_desc_head_idx = 0;
 404         vq->vq_avail_idx = 0;
 405         vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
 406         vq->vq_free_cnt = vq->vq_nentries;
 407         memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
 408         if (vtpci_packed_queue(vq->hw)) {
 409                 vring_init_packed(&vq->vq_packed.ring, ring_mem,
 410                                   VIRTIO_PCI_VRING_ALIGN, size);
 411                 vring_desc_init_packed(vq, size);
 412         } else {
 413                 struct vring *vr = &vq->vq_split.ring;
 414
 415                 vring_init_split(vr, ring_mem, VIRTIO_PCI_VRING_ALIGN, size);
 416                 vring_desc_init_split(vr->desc, size);
 417         }
 418         /*
 419          * Disable device(host) interrupting guest
 420          */
 421         virtqueue_disable_intr(vq);
 422 }
 423
 424 static int
 425 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
 426 {
 427         char vq_name[VIRTQUEUE_MAX_NAME_SZ];
 428         char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
 429         const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
 430         unsigned int vq_size, size;
 431         struct virtio_hw *hw = dev->data->dev_private;
 432         struct virtnet_rx *rxvq = NULL;
 433         struct virtnet_tx *txvq = NULL;
 434         struct virtnet_ctl *cvq = NULL;
 435         struct virtqueue *vq;
 436         size_t sz_hdr_mz = 0;
 437         void *sw_ring = NULL;
 438         int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
 439         int ret;
 440         int numa_node = dev->device->numa_node;
 441
 442         PMD_INIT_LOG(INFO, "setting up queue: %u on NUMA node %d",
 443                         vtpci_queue_idx, numa_node);
 444
 445         /*
 446          * Read the virtqueue size from the Queue Size field
 447          * Always power of 2 and if 0 virtqueue does not exist
 448          */
 449         vq_size = VTPCI_OPS(hw)->get_queue_num(hw, vtpci_queue_idx);
 450         PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
 451         if (vq_size == 0) {
 452                 PMD_INIT_LOG(ERR, "virtqueue does not exist");
 453                 return -EINVAL;
 454         }
 455
 456         if (!vtpci_packed_queue(hw) && !rte_is_power_of_2(vq_size)) {
 457                 PMD_INIT_LOG(ERR, "split virtqueue size is not power of 2");
 458                 return -EINVAL;
 459         }
 460
 461         snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
 462                  dev->data->port_id, vtpci_queue_idx);
 463
 464         size = RTE_ALIGN_CEIL(sizeof(*vq) +
 465                                 vq_size * sizeof(struct vq_desc_extra),
 466                                 RTE_CACHE_LINE_SIZE);
 467         if (queue_type == VTNET_TQ) {
 468                 /*
 469                  * For each xmit packet, allocate a virtio_net_hdr
 470                  * and indirect ring elements
 471                  */
 472                 sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
 473         } else if (queue_type == VTNET_CQ) {
 474                 /* Allocate a page for control vq command, data and status */
 475                 sz_hdr_mz = PAGE_SIZE;
 476         }
 477
 478         vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
 479                                 numa_node);
 480         if (vq == NULL) {
 481                 PMD_INIT_LOG(ERR, "can not allocate vq");
 482                 return -ENOMEM;
 483         }
 484         hw->vqs[vtpci_queue_idx] = vq;
 485
 486         vq->hw = hw;
 487         vq->vq_queue_index = vtpci_queue_idx;
 488         vq->vq_nentries = vq_size;
 489         if (vtpci_packed_queue(hw)) {
 490                 vq->vq_packed.used_wrap_counter = 1;
 491                 vq->vq_packed.cached_flags = VRING_PACKED_DESC_F_AVAIL;
 492                 vq->vq_packed.event_flags_shadow = 0;
 493                 if (queue_type == VTNET_RQ)
 494                         vq->vq_packed.cached_flags |= VRING_DESC_F_WRITE;
 495         }
 496
 497         /*
 498          * Reserve a memzone for vring elements
 499          */
 500         size = vring_size(hw, vq_size, VIRTIO_PCI_VRING_ALIGN);
 501         vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
 502         PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
 503                      size, vq->vq_ring_size);
 504
 505         mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
 506                         numa_node, RTE_MEMZONE_IOVA_CONTIG,
 507                         VIRTIO_PCI_VRING_ALIGN);
 508         if (mz == NULL) {
 509                 if (rte_errno == EEXIST)
 510                         mz = rte_memzone_lookup(vq_name);
 511                 if (mz == NULL) {
 512                         ret = -ENOMEM;
 513                         goto fail_q_alloc;
 514                 }
 515         }
 516
 517         memset(mz->addr, 0, mz->len);
 518
 519         vq->vq_ring_mem = mz->iova;
 520         vq->vq_ring_virt_mem = mz->addr;
 521         PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%" PRIx64,
 522                      (uint64_t)mz->iova);
 523         PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
 524                      (uint64_t)(uintptr_t)mz->addr);
 525
 526         virtio_init_vring(vq);
 527
 528         if (sz_hdr_mz) {
 529                 snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
 530                          dev->data->port_id, vtpci_queue_idx);
 531                 hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
 532                                 numa_node, RTE_MEMZONE_IOVA_CONTIG,
 533                                 RTE_CACHE_LINE_SIZE);
 534                 if (hdr_mz == NULL) {
 535                         if (rte_errno == EEXIST)
 536                                 hdr_mz = rte_memzone_lookup(vq_hdr_name);
 537                         if (hdr_mz == NULL) {
 538                                 ret = -ENOMEM;
 539                                 goto fail_q_alloc;
 540                         }
 541                 }
 542         }
 543
 544         if (queue_type == VTNET_RQ) {
 545                 size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
 546                                sizeof(vq->sw_ring[0]);
 547
 548                 sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
 549                                 RTE_CACHE_LINE_SIZE, numa_node);
 550                 if (!sw_ring) {
 551                         PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
 552                         ret = -ENOMEM;
 553                         goto fail_q_alloc;
 554                 }
 555
 556                 vq->sw_ring = sw_ring;
 557                 rxvq = &vq->rxq;
 558                 rxvq->vq = vq;
 559                 rxvq->port_id = dev->data->port_id;
 560                 rxvq->mz = mz;
 561         } else if (queue_type == VTNET_TQ) {
 562                 txvq = &vq->txq;
 563                 txvq->vq = vq;
 564                 txvq->port_id = dev->data->port_id;
 565                 txvq->mz = mz;
 566                 txvq->virtio_net_hdr_mz = hdr_mz;
 567                 txvq->virtio_net_hdr_mem = hdr_mz->iova;
 568         } else if (queue_type == VTNET_CQ) {
 569                 cvq = &vq->cq;
 570                 cvq->vq = vq;
 571                 cvq->mz = mz;
 572                 cvq->virtio_net_hdr_mz = hdr_mz;
 573                 cvq->virtio_net_hdr_mem = hdr_mz->iova;
 574                 memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
 575
 576                 hw->cvq = cvq;
 577         }
 578
 579         /* For virtio_user case (that is when hw->virtio_user_dev is not NULL),
 580          * we use virtual address. And we need properly set _offset_, please see
 581          * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
 582          */
 583         if (hw->bus_type == VIRTIO_BUS_PCI_LEGACY || hw->bus_type == VIRTIO_BUS_PCI_MODERN) {
 584                 vq->offset = offsetof(struct rte_mbuf, buf_iova);
 585         } else if (hw->bus_type == VIRTIO_BUS_USER) {
 586                 vq->vq_ring_mem = (uintptr_t)mz->addr;
 587                 vq->offset = offsetof(struct rte_mbuf, buf_addr);
 588                 if (queue_type == VTNET_TQ)
 589                         txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
 590                 else if (queue_type == VTNET_CQ)
 591                         cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
 592         }
 593
 594         if (queue_type == VTNET_TQ) {
 595                 struct virtio_tx_region *txr;
 596                 unsigned int i;
 597
 598                 txr = hdr_mz->addr;
 599                 memset(txr, 0, vq_size * sizeof(*txr));
 600                 for (i = 0; i < vq_size; i++) {
 601                         /* first indirect descriptor is always the tx header */
 602                         if (!vtpci_packed_queue(hw)) {
 603                                 struct vring_desc *start_dp = txr[i].tx_indir;
 604                                 vring_desc_init_split(start_dp,
 605                                                       RTE_DIM(txr[i].tx_indir));
 606                                 start_dp->addr = txvq->virtio_net_hdr_mem
 607                                         + i * sizeof(*txr)
 608                                         + offsetof(struct virtio_tx_region,
 609                                                    tx_hdr);
 610                                 start_dp->len = hw->vtnet_hdr_size;
 611                                 start_dp->flags = VRING_DESC_F_NEXT;
 612                         } else {
 613                                 struct vring_packed_desc *start_dp =
 614                                         txr[i].tx_packed_indir;
 615                                 vring_desc_init_indirect_packed(start_dp,
 616                                       RTE_DIM(txr[i].tx_packed_indir));
 617                                 start_dp->addr = txvq->virtio_net_hdr_mem
 618                                         + i * sizeof(*txr)
 619                                         + offsetof(struct virtio_tx_region,
 620                                                    tx_hdr);
 621                                 start_dp->len = hw->vtnet_hdr_size;
 622                         }
 623                 }
 624         }
 625
 626         if (VTPCI_OPS(hw)->setup_queue(hw, vq) < 0) {
 627                 PMD_INIT_LOG(ERR, "setup_queue failed");
 628                 return -EINVAL;
 629         }
 630
 631         return 0;
 632
 633 fail_q_alloc:
 634         rte_free(sw_ring);
 635         rte_memzone_free(hdr_mz);
 636         rte_memzone_free(mz);
 637         rte_free(vq);
 638
 639         return ret;
 640 }
 641
 642 static void
 643 virtio_free_queues(struct virtio_hw *hw)
 644 {
 645         uint16_t nr_vq = virtio_get_nr_vq(hw);
 646         struct virtqueue *vq;
 647         int queue_type;
 648         uint16_t i;
 649
 650         if (hw->vqs == NULL)
 651                 return;
 652
 653         for (i = 0; i < nr_vq; i++) {
 654                 vq = hw->vqs[i];
 655                 if (!vq)
 656                         continue;
 657
 658                 queue_type = virtio_get_queue_type(hw, i);
 659                 if (queue_type == VTNET_RQ) {
 660                         rte_free(vq->sw_ring);
 661                         rte_memzone_free(vq->rxq.mz);
 662                 } else if (queue_type == VTNET_TQ) {
 663                         rte_memzone_free(vq->txq.mz);
 664                         rte_memzone_free(vq->txq.virtio_net_hdr_mz);
 665                 } else {
 666                         rte_memzone_free(vq->cq.mz);
 667                         rte_memzone_free(vq->cq.virtio_net_hdr_mz);
 668                 }
 669
 670                 rte_free(vq);
 671                 hw->vqs[i] = NULL;
 672         }
 673
 674         rte_free(hw->vqs);
 675         hw->vqs = NULL;
 676 }
 677
 678 static int
 679 virtio_alloc_queues(struct rte_eth_dev *dev)
 680 {
 681         struct virtio_hw *hw = dev->data->dev_private;
 682         uint16_t nr_vq = virtio_get_nr_vq(hw);
 683         uint16_t i;
 684         int ret;
 685
 686         hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
 687         if (!hw->vqs) {
 688                 PMD_INIT_LOG(ERR, "failed to allocate vqs");
 689                 return -ENOMEM;
 690         }
 691
 692         for (i = 0; i < nr_vq; i++) {
 693                 ret = virtio_init_queue(dev, i);
 694                 if (ret < 0) {
 695                         virtio_free_queues(hw);
 696                         return ret;
 697                 }
 698         }
 699
 700         return 0;
 701 }
 702
 703 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
 704
 705 int
 706 virtio_dev_close(struct rte_eth_dev *dev)
 707 {
 708         struct virtio_hw *hw = dev->data->dev_private;
 709         struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
 710
 711         PMD_INIT_LOG(DEBUG, "virtio_dev_close");
 712         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 713                 return 0;
 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->bus_type == VIRTIO_BUS_USER)
 738                 virtio_user_dev_uninit(dev->data->dev_private);
 739         else
 740 #endif
 741         if (dev->device) {
 742                 rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(dev));
 743                 if (hw->bus_type == VIRTIO_BUS_PCI_LEGACY)
 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->bus_type == VIRTIO_BUS_PCI_LEGACY || hw->bus_type == VIRTIO_BUS_PCI_MODERN)
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->bus_type == VIRTIO_BUS_PCI_LEGACY || hw->bus_type == VIRTIO_BUS_PCI_MODERN)
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->bus_type == VIRTIO_BUS_PCI_LEGACY || hw->bus_type == VIRTIO_BUS_PCI_MODERN)
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->bus_type == VIRTIO_BUS_PCI_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->bus_type == VIRTIO_BUS_PCI_MODERN || hw->bus_type == VIRTIO_BUS_USER) {
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 DUPLEX_UNKNOWN   0xff
1672 /* reset device and renegotiate features if needed */
1673 static int
1674 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1675 {
1676         struct virtio_hw *hw = eth_dev->data->dev_private;
1677         struct virtio_net_config *config;
1678         struct virtio_net_config local_config;
1679         struct rte_pci_device *pci_dev = NULL;
1680         int ret;
1681
1682         /* Reset the device although not necessary at startup */
1683         vtpci_reset(hw);
1684
1685         if (hw->vqs) {
1686                 virtio_dev_free_mbufs(eth_dev);
1687                 virtio_free_queues(hw);
1688         }
1689
1690         /* Tell the host we've noticed this device. */
1691         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1692
1693         /* Tell the host we've known how to drive the device. */
1694         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1695         if (virtio_negotiate_features(hw, req_features) < 0)
1696                 return -1;
1697
1698         hw->weak_barriers = !vtpci_with_feature(hw, VIRTIO_F_ORDER_PLATFORM);
1699
1700         if (hw->bus_type == VIRTIO_BUS_PCI_LEGACY || hw->bus_type == VIRTIO_BUS_PCI_MODERN)
1701                 pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1702
1703         /* If host does not support both status and MSI-X then disable LSC */
1704         if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) &&
1705             hw->use_msix != VIRTIO_MSIX_NONE)
1706                 eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1707         else
1708                 eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1709
1710         eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
1711
1712         /* Setting up rx_header size for the device */
1713         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1714             vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
1715             vtpci_with_feature(hw, VIRTIO_F_RING_PACKED))
1716                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1717         else
1718                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1719
1720         /* Copy the permanent MAC address to: virtio_hw */
1721         virtio_get_hwaddr(hw);
1722         rte_ether_addr_copy((struct rte_ether_addr *)hw->mac_addr,
1723                         &eth_dev->data->mac_addrs[0]);
1724         PMD_INIT_LOG(DEBUG,
1725                      "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1726                      hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1727                      hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1728
1729         if (hw->speed == ETH_SPEED_NUM_UNKNOWN) {
1730                 if (vtpci_with_feature(hw, VIRTIO_NET_F_SPEED_DUPLEX)) {
1731                         config = &local_config;
1732                         vtpci_read_dev_config(hw,
1733                                 offsetof(struct virtio_net_config, speed),
1734                                 &config->speed, sizeof(config->speed));
1735                         vtpci_read_dev_config(hw,
1736                                 offsetof(struct virtio_net_config, duplex),
1737                                 &config->duplex, sizeof(config->duplex));
1738                         hw->speed = config->speed;
1739                         hw->duplex = config->duplex;
1740                 }
1741         }
1742         if (hw->duplex == DUPLEX_UNKNOWN)
1743                 hw->duplex = ETH_LINK_FULL_DUPLEX;
1744         PMD_INIT_LOG(DEBUG, "link speed = %d, duplex = %d",
1745                 hw->speed, hw->duplex);
1746         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1747                 config = &local_config;
1748
1749                 vtpci_read_dev_config(hw,
1750                         offsetof(struct virtio_net_config, mac),
1751                         &config->mac, sizeof(config->mac));
1752
1753                 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1754                         vtpci_read_dev_config(hw,
1755                                 offsetof(struct virtio_net_config, status),
1756                                 &config->status, sizeof(config->status));
1757                 } else {
1758                         PMD_INIT_LOG(DEBUG,
1759                                      "VIRTIO_NET_F_STATUS is not supported");
1760                         config->status = 0;
1761                 }
1762
1763                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1764                         vtpci_read_dev_config(hw,
1765                                 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1766                                 &config->max_virtqueue_pairs,
1767                                 sizeof(config->max_virtqueue_pairs));
1768                 } else {
1769                         PMD_INIT_LOG(DEBUG,
1770                                      "VIRTIO_NET_F_MQ is not supported");
1771                         config->max_virtqueue_pairs = 1;
1772                 }
1773
1774                 hw->max_queue_pairs = config->max_virtqueue_pairs;
1775
1776                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1777                         vtpci_read_dev_config(hw,
1778                                 offsetof(struct virtio_net_config, mtu),
1779                                 &config->mtu,
1780                                 sizeof(config->mtu));
1781
1782                         /*
1783                          * MTU value has already been checked at negotiation
1784                          * time, but check again in case it has changed since
1785                          * then, which should not happen.
1786                          */
1787                         if (config->mtu < RTE_ETHER_MIN_MTU) {
1788                                 PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1789                                                 config->mtu);
1790                                 return -1;
1791                         }
1792
1793                         hw->max_mtu = config->mtu;
1794                         /* Set initial MTU to maximum one supported by vhost */
1795                         eth_dev->data->mtu = config->mtu;
1796
1797                 } else {
1798                         hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1799                                 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1800                 }
1801
1802                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1803                                 config->max_virtqueue_pairs);
1804                 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1805                 PMD_INIT_LOG(DEBUG,
1806                                 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1807                                 config->mac[0], config->mac[1],
1808                                 config->mac[2], config->mac[3],
1809                                 config->mac[4], config->mac[5]);
1810         } else {
1811                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1812                 hw->max_queue_pairs = 1;
1813                 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - RTE_ETHER_HDR_LEN -
1814                         VLAN_TAG_LEN - hw->vtnet_hdr_size;
1815         }
1816
1817         ret = virtio_alloc_queues(eth_dev);
1818         if (ret < 0)
1819                 return ret;
1820
1821         if (eth_dev->data->dev_conf.intr_conf.rxq) {
1822                 if (virtio_configure_intr(eth_dev) < 0) {
1823                         PMD_INIT_LOG(ERR, "failed to configure interrupt");
1824                         virtio_free_queues(hw);
1825                         return -1;
1826                 }
1827         }
1828
1829         vtpci_reinit_complete(hw);
1830
1831         if (pci_dev)
1832                 PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1833                         eth_dev->data->port_id, pci_dev->id.vendor_id,
1834                         pci_dev->id.device_id);
1835
1836         return 0;
1837 }
1838
1839 /*
1840  * Remap the PCI device again (IO port map for legacy device and
1841  * memory map for modern device), so that the secondary process
1842  * could have the PCI initiated correctly.
1843  */
1844 static int
1845 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1846 {
1847         if (hw->bus_type == VIRTIO_BUS_PCI_MODERN) {
1848                 /*
1849                  * We don't have to re-parse the PCI config space, since
1850                  * rte_pci_map_device() makes sure the mapped address
1851                  * in secondary process would equal to the one mapped in
1852                  * the primary process: error will be returned if that
1853                  * requirement is not met.
1854                  *
1855                  * That said, we could simply reuse all cap pointers
1856                  * (such as dev_cfg, common_cfg, etc.) parsed from the
1857                  * primary process, which is stored in shared memory.
1858                  */
1859                 if (rte_pci_map_device(pci_dev)) {
1860                         PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1861                         return -1;
1862                 }
1863         } else if (hw->bus_type == VIRTIO_BUS_PCI_LEGACY) {
1864                 if (rte_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1865                         return -1;
1866         }
1867
1868         return 0;
1869 }
1870
1871 static void
1872 virtio_set_vtpci_ops(struct virtio_hw *hw)
1873 {
1874 #ifdef RTE_VIRTIO_USER
1875         if (hw->bus_type == VIRTIO_BUS_USER)
1876                 VTPCI_OPS(hw) = &virtio_user_ops;
1877         else
1878 #endif
1879         if (hw->bus_type == VIRTIO_BUS_PCI_MODERN)
1880                 VTPCI_OPS(hw) = &modern_ops;
1881         else if (hw->bus_type == VIRTIO_BUS_PCI_LEGACY)
1882                 VTPCI_OPS(hw) = &legacy_ops;
1883
1884         return;
1885 }
1886
1887 /*
1888  * This function is based on probe() function in virtio_pci.c
1889  * It returns 0 on success.
1890  */
1891 int
1892 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1893 {
1894         struct virtio_hw *hw = eth_dev->data->dev_private;
1895         uint32_t speed = ETH_SPEED_NUM_UNKNOWN;
1896         int vectorized = 0;
1897         int ret;
1898
1899         if (sizeof(struct virtio_net_hdr_mrg_rxbuf) > RTE_PKTMBUF_HEADROOM) {
1900                 PMD_INIT_LOG(ERR,
1901                         "Not sufficient headroom required = %d, avail = %d",
1902                         (int)sizeof(struct virtio_net_hdr_mrg_rxbuf),
1903                         RTE_PKTMBUF_HEADROOM);
1904
1905                 return -1;
1906         }
1907
1908         eth_dev->dev_ops = &virtio_eth_dev_ops;
1909         eth_dev->rx_descriptor_done = virtio_dev_rx_queue_done;
1910
1911         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1912                 if (hw->bus_type != VIRTIO_BUS_USER) {
1913                         ret = virtio_remap_pci(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1914                         if (ret)
1915                                 return ret;
1916                 }
1917
1918                 virtio_set_vtpci_ops(hw);
1919                 set_rxtx_funcs(eth_dev);
1920
1921                 return 0;
1922         }
1923         ret = virtio_dev_devargs_parse(eth_dev->device->devargs, &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->bus_type != VIRTIO_BUS_USER) {
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) || defined(RTE_ARCH_ARM)
1960                         hw->use_vec_rx = 1;
1961                         hw->use_vec_tx = 1;
1962 #else
1963                         PMD_DRV_LOG(INFO,
1964                                 "building environment do not support packed ring vectorized");
1965 #endif
1966                 }
1967         }
1968
1969         hw->opened = true;
1970
1971         return 0;
1972
1973 err_virtio_init:
1974         if (hw->bus_type == VIRTIO_BUS_PCI_MODERN || hw->bus_type == VIRTIO_BUS_PCI_LEGACY) {
1975                 rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev));
1976                 if (hw->bus_type == VIRTIO_BUS_PCI_LEGACY)
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 uint32_t
1986 virtio_dev_speed_capa_get(uint32_t speed)
1987 {
1988         switch (speed) {
1989         case ETH_SPEED_NUM_10G:
1990                 return ETH_LINK_SPEED_10G;
1991         case ETH_SPEED_NUM_20G:
1992                 return ETH_LINK_SPEED_20G;
1993         case ETH_SPEED_NUM_25G:
1994                 return ETH_LINK_SPEED_25G;
1995         case ETH_SPEED_NUM_40G:
1996                 return ETH_LINK_SPEED_40G;
1997         case ETH_SPEED_NUM_50G:
1998                 return ETH_LINK_SPEED_50G;
1999         case ETH_SPEED_NUM_56G:
2000                 return ETH_LINK_SPEED_56G;
2001         case ETH_SPEED_NUM_100G:
2002                 return ETH_LINK_SPEED_100G;
2003         case ETH_SPEED_NUM_200G:
2004                 return ETH_LINK_SPEED_200G;
2005         default:
2006                 return 0;
2007         }
2008 }
2009
2010 static int vectorized_check_handler(__rte_unused const char *key,
2011                 const char *value, void *ret_val)
2012 {
2013         if (strcmp(value, "1") == 0)
2014                 *(int *)ret_val = 1;
2015         else
2016                 *(int *)ret_val = 0;
2017
2018         return 0;
2019 }
2020
2021 #define VIRTIO_ARG_SPEED      "speed"
2022 #define VIRTIO_ARG_VECTORIZED "vectorized"
2023
2024 static int
2025 link_speed_handler(const char *key __rte_unused,
2026                 const char *value, void *ret_val)
2027 {
2028         uint32_t val;
2029         if (!value || !ret_val)
2030                 return -EINVAL;
2031         val = strtoul(value, NULL, 0);
2032         /* validate input */
2033         if (virtio_dev_speed_capa_get(val) == 0)
2034                 return -EINVAL;
2035         *(uint32_t *)ret_val = val;
2036
2037         return 0;
2038 }
2039
2040
2041 static int
2042 virtio_dev_devargs_parse(struct rte_devargs *devargs, uint32_t *speed, int *vectorized)
2043 {
2044         struct rte_kvargs *kvlist;
2045         int ret = 0;
2046
2047         if (devargs == NULL)
2048                 return 0;
2049
2050         kvlist = rte_kvargs_parse(devargs->args, NULL);
2051         if (kvlist == NULL) {
2052                 PMD_INIT_LOG(ERR, "error when parsing param");
2053                 return 0;
2054         }
2055
2056         if (speed && rte_kvargs_count(kvlist, VIRTIO_ARG_SPEED) == 1) {
2057                 ret = rte_kvargs_process(kvlist,
2058                                         VIRTIO_ARG_SPEED,
2059                                         link_speed_handler, speed);
2060                 if (ret < 0) {
2061                         PMD_INIT_LOG(ERR, "Failed to parse %s",
2062                                         VIRTIO_ARG_SPEED);
2063                         goto exit;
2064                 }
2065         }
2066
2067         if (vectorized &&
2068                 rte_kvargs_count(kvlist, VIRTIO_ARG_VECTORIZED) == 1) {
2069                 ret = rte_kvargs_process(kvlist,
2070                                 VIRTIO_ARG_VECTORIZED,
2071                                 vectorized_check_handler, vectorized);
2072                 if (ret < 0) {
2073                         PMD_INIT_LOG(ERR, "Failed to parse %s",
2074                                         VIRTIO_ARG_VECTORIZED);
2075                         goto exit;
2076                 }
2077         }
2078
2079 exit:
2080         rte_kvargs_free(kvlist);
2081         return ret;
2082 }
2083
2084 static bool
2085 rx_offload_enabled(struct virtio_hw *hw)
2086 {
2087         return vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM) ||
2088                 vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2089                 vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6);
2090 }
2091
2092 static bool
2093 tx_offload_enabled(struct virtio_hw *hw)
2094 {
2095         return vtpci_with_feature(hw, VIRTIO_NET_F_CSUM) ||
2096                 vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO4) ||
2097                 vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO6);
2098 }
2099
2100 /*
2101  * Configure virtio device
2102  * It returns 0 on success.
2103  */
2104 static int
2105 virtio_dev_configure(struct rte_eth_dev *dev)
2106 {
2107         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2108         const struct rte_eth_txmode *txmode = &dev->data->dev_conf.txmode;
2109         struct virtio_hw *hw = dev->data->dev_private;
2110         uint32_t ether_hdr_len = RTE_ETHER_HDR_LEN + VLAN_TAG_LEN +
2111                 hw->vtnet_hdr_size;
2112         uint64_t rx_offloads = rxmode->offloads;
2113         uint64_t tx_offloads = txmode->offloads;
2114         uint64_t req_features;
2115         int ret;
2116
2117         PMD_INIT_LOG(DEBUG, "configure");
2118         req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
2119
2120         if (rxmode->mq_mode != ETH_MQ_RX_NONE) {
2121                 PMD_DRV_LOG(ERR,
2122                         "Unsupported Rx multi queue mode %d",
2123                         rxmode->mq_mode);
2124                 return -EINVAL;
2125         }
2126
2127         if (txmode->mq_mode != ETH_MQ_TX_NONE) {
2128                 PMD_DRV_LOG(ERR,
2129                         "Unsupported Tx multi queue mode %d",
2130                         txmode->mq_mode);
2131                 return -EINVAL;
2132         }
2133
2134         if (dev->data->dev_conf.intr_conf.rxq) {
2135                 ret = virtio_init_device(dev, hw->req_guest_features);
2136                 if (ret < 0)
2137                         return ret;
2138         }
2139
2140         if (rxmode->max_rx_pkt_len > hw->max_mtu + ether_hdr_len)
2141                 req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
2142
2143         if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2144                            DEV_RX_OFFLOAD_TCP_CKSUM))
2145                 req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
2146
2147         if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
2148                 req_features |=
2149                         (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2150                         (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2151
2152         if (tx_offloads & (DEV_TX_OFFLOAD_UDP_CKSUM |
2153                            DEV_TX_OFFLOAD_TCP_CKSUM))
2154                 req_features |= (1ULL << VIRTIO_NET_F_CSUM);
2155
2156         if (tx_offloads & DEV_TX_OFFLOAD_TCP_TSO)
2157                 req_features |=
2158                         (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2159                         (1ULL << VIRTIO_NET_F_HOST_TSO6);
2160
2161         /* if request features changed, reinit the device */
2162         if (req_features != hw->req_guest_features) {
2163                 ret = virtio_init_device(dev, req_features);
2164                 if (ret < 0)
2165                         return ret;
2166         }
2167
2168         if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2169                             DEV_RX_OFFLOAD_TCP_CKSUM)) &&
2170                 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
2171                 PMD_DRV_LOG(ERR,
2172                         "rx checksum not available on this host");
2173                 return -ENOTSUP;
2174         }
2175
2176         if ((rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) &&
2177                 (!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2178                  !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
2179                 PMD_DRV_LOG(ERR,
2180                         "Large Receive Offload not available on this host");
2181                 return -ENOTSUP;
2182         }
2183
2184         /* start control queue */
2185         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
2186                 virtio_dev_cq_start(dev);
2187
2188         if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2189                 hw->vlan_strip = 1;
2190
2191         if ((rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
2192             && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2193                 PMD_DRV_LOG(ERR,
2194                             "vlan filtering not available on this host");
2195                 return -ENOTSUP;
2196         }
2197
2198         hw->has_tx_offload = tx_offload_enabled(hw);
2199         hw->has_rx_offload = rx_offload_enabled(hw);
2200
2201         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2202                 /* Enable vector (0) for Link State Intrerrupt */
2203                 if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
2204                                 VIRTIO_MSI_NO_VECTOR) {
2205                         PMD_DRV_LOG(ERR, "failed to set config vector");
2206                         return -EBUSY;
2207                 }
2208
2209         if (vtpci_packed_queue(hw)) {
2210 #if defined(RTE_ARCH_X86_64) && defined(CC_AVX512_SUPPORT)
2211                 if ((hw->use_vec_rx || hw->use_vec_tx) &&
2212                     (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F) ||
2213                      !vtpci_with_feature(hw, VIRTIO_F_IN_ORDER) ||
2214                      !vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
2215                      rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_512)) {
2216                         PMD_DRV_LOG(INFO,
2217                                 "disabled packed ring vectorized path for requirements not met");
2218                         hw->use_vec_rx = 0;
2219                         hw->use_vec_tx = 0;
2220                 }
2221 #elif defined(RTE_ARCH_ARM)
2222                 if ((hw->use_vec_rx || hw->use_vec_tx) &&
2223                     (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON) ||
2224                      !vtpci_with_feature(hw, VIRTIO_F_IN_ORDER) ||
2225                      !vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
2226                      rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_128)) {
2227                         PMD_DRV_LOG(INFO,
2228                                 "disabled packed ring vectorized path for requirements not met");
2229                         hw->use_vec_rx = 0;
2230                         hw->use_vec_tx = 0;
2231                 }
2232 #else
2233                 hw->use_vec_rx = 0;
2234                 hw->use_vec_tx = 0;
2235 #endif
2236
2237                 if (hw->use_vec_rx) {
2238                         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2239                                 PMD_DRV_LOG(INFO,
2240                                         "disabled packed ring vectorized rx for mrg_rxbuf enabled");
2241                                 hw->use_vec_rx = 0;
2242                         }
2243
2244                         if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) {
2245                                 PMD_DRV_LOG(INFO,
2246                                         "disabled packed ring vectorized rx for TCP_LRO enabled");
2247                                 hw->use_vec_rx = 0;
2248                         }
2249                 }
2250         } else {
2251                 if (vtpci_with_feature(hw, VIRTIO_F_IN_ORDER)) {
2252                         hw->use_inorder_tx = 1;
2253                         hw->use_inorder_rx = 1;
2254                         hw->use_vec_rx = 0;
2255                 }
2256
2257                 if (hw->use_vec_rx) {
2258 #if defined RTE_ARCH_ARM
2259                         if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
2260                                 PMD_DRV_LOG(INFO,
2261                                         "disabled split ring vectorized path for requirement not met");
2262                                 hw->use_vec_rx = 0;
2263                         }
2264 #endif
2265                         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2266                                 PMD_DRV_LOG(INFO,
2267                                         "disabled split ring vectorized rx for mrg_rxbuf enabled");
2268                                 hw->use_vec_rx = 0;
2269                         }
2270
2271                         if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2272                                            DEV_RX_OFFLOAD_TCP_CKSUM |
2273                                            DEV_RX_OFFLOAD_TCP_LRO |
2274                                            DEV_RX_OFFLOAD_VLAN_STRIP)) {
2275                                 PMD_DRV_LOG(INFO,
2276                                         "disabled split ring vectorized rx for offloading enabled");
2277                                 hw->use_vec_rx = 0;
2278                         }
2279
2280                         if (rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_128) {
2281                                 PMD_DRV_LOG(INFO,
2282                                         "disabled split ring vectorized rx, max SIMD bitwidth too low");
2283                                 hw->use_vec_rx = 0;
2284                         }
2285                 }
2286         }
2287
2288         return 0;
2289 }
2290
2291
2292 static int
2293 virtio_dev_start(struct rte_eth_dev *dev)
2294 {
2295         uint16_t nb_queues, i;
2296         struct virtnet_rx *rxvq;
2297         struct virtnet_tx *txvq __rte_unused;
2298         struct virtio_hw *hw = dev->data->dev_private;
2299         int ret;
2300
2301         /* Finish the initialization of the queues */
2302         for (i = 0; i < dev->data->nb_rx_queues; i++) {
2303                 ret = virtio_dev_rx_queue_setup_finish(dev, i);
2304                 if (ret < 0)
2305                         return ret;
2306         }
2307         for (i = 0; i < dev->data->nb_tx_queues; i++) {
2308                 ret = virtio_dev_tx_queue_setup_finish(dev, i);
2309                 if (ret < 0)
2310                         return ret;
2311         }
2312
2313         /* check if lsc interrupt feature is enabled */
2314         if (dev->data->dev_conf.intr_conf.lsc) {
2315                 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
2316                         PMD_DRV_LOG(ERR, "link status not supported by host");
2317                         return -ENOTSUP;
2318                 }
2319         }
2320
2321         /* Enable uio/vfio intr/eventfd mapping: althrough we already did that
2322          * in device configure, but it could be unmapped  when device is
2323          * stopped.
2324          */
2325         if (dev->data->dev_conf.intr_conf.lsc ||
2326             dev->data->dev_conf.intr_conf.rxq) {
2327                 virtio_intr_disable(dev);
2328
2329                 /* Setup interrupt callback  */
2330                 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2331                         rte_intr_callback_register(dev->intr_handle,
2332                                                    virtio_interrupt_handler,
2333                                                    dev);
2334
2335                 if (virtio_intr_enable(dev) < 0) {
2336                         PMD_DRV_LOG(ERR, "interrupt enable failed");
2337                         return -EIO;
2338                 }
2339         }
2340
2341         /*Notify the backend
2342          *Otherwise the tap backend might already stop its queue due to fullness.
2343          *vhost backend will have no chance to be waked up
2344          */
2345         nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
2346         if (hw->max_queue_pairs > 1) {
2347                 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
2348                         return -EINVAL;
2349         }
2350
2351         PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
2352
2353         for (i = 0; i < dev->data->nb_rx_queues; i++) {
2354                 rxvq = dev->data->rx_queues[i];
2355                 /* Flush the old packets */
2356                 virtqueue_rxvq_flush(rxvq->vq);
2357                 virtqueue_notify(rxvq->vq);
2358         }
2359
2360         for (i = 0; i < dev->data->nb_tx_queues; i++) {
2361                 txvq = dev->data->tx_queues[i];
2362                 virtqueue_notify(txvq->vq);
2363         }
2364
2365         PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
2366
2367         for (i = 0; i < dev->data->nb_rx_queues; i++) {
2368                 rxvq = dev->data->rx_queues[i];
2369                 VIRTQUEUE_DUMP(rxvq->vq);
2370         }
2371
2372         for (i = 0; i < dev->data->nb_tx_queues; i++) {
2373                 txvq = dev->data->tx_queues[i];
2374                 VIRTQUEUE_DUMP(txvq->vq);
2375         }
2376
2377         set_rxtx_funcs(dev);
2378         hw->started = true;
2379
2380         /* Initialize Link state */
2381         virtio_dev_link_update(dev, 0);
2382
2383         return 0;
2384 }
2385
2386 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
2387 {
2388         struct virtio_hw *hw = dev->data->dev_private;
2389         uint16_t nr_vq = virtio_get_nr_vq(hw);
2390         const char *type __rte_unused;
2391         unsigned int i, mbuf_num = 0;
2392         struct virtqueue *vq;
2393         struct rte_mbuf *buf;
2394         int queue_type;
2395
2396         if (hw->vqs == NULL)
2397                 return;
2398
2399         for (i = 0; i < nr_vq; i++) {
2400                 vq = hw->vqs[i];
2401                 if (!vq)
2402                         continue;
2403
2404                 queue_type = virtio_get_queue_type(hw, i);
2405                 if (queue_type == VTNET_RQ)
2406                         type = "rxq";
2407                 else if (queue_type == VTNET_TQ)
2408                         type = "txq";
2409                 else
2410                         continue;
2411
2412                 PMD_INIT_LOG(DEBUG,
2413                         "Before freeing %s[%d] used and unused buf",
2414                         type, i);
2415                 VIRTQUEUE_DUMP(vq);
2416
2417                 while ((buf = virtqueue_detach_unused(vq)) != NULL) {
2418                         rte_pktmbuf_free(buf);
2419                         mbuf_num++;
2420                 }
2421
2422                 PMD_INIT_LOG(DEBUG,
2423                         "After freeing %s[%d] used and unused buf",
2424                         type, i);
2425                 VIRTQUEUE_DUMP(vq);
2426         }
2427
2428         PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
2429 }
2430
2431 /*
2432  * Stop device: disable interrupt and mark link down
2433  */
2434 int
2435 virtio_dev_stop(struct rte_eth_dev *dev)
2436 {
2437         struct virtio_hw *hw = dev->data->dev_private;
2438         struct rte_eth_link link;
2439         struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
2440
2441         PMD_INIT_LOG(DEBUG, "stop");
2442         dev->data->dev_started = 0;
2443
2444         rte_spinlock_lock(&hw->state_lock);
2445         if (!hw->started)
2446                 goto out_unlock;
2447         hw->started = false;
2448
2449         if (intr_conf->lsc || intr_conf->rxq) {
2450                 virtio_intr_disable(dev);
2451
2452                 /* Reset interrupt callback  */
2453                 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
2454                         rte_intr_callback_unregister(dev->intr_handle,
2455                                                      virtio_interrupt_handler,
2456                                                      dev);
2457                 }
2458         }
2459
2460         memset(&link, 0, sizeof(link));
2461         rte_eth_linkstatus_set(dev, &link);
2462 out_unlock:
2463         rte_spinlock_unlock(&hw->state_lock);
2464
2465         return 0;
2466 }
2467
2468 static int
2469 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2470 {
2471         struct rte_eth_link link;
2472         uint16_t status;
2473         struct virtio_hw *hw = dev->data->dev_private;
2474
2475         memset(&link, 0, sizeof(link));
2476         link.link_duplex = hw->duplex;
2477         link.link_speed  = hw->speed;
2478         link.link_autoneg = ETH_LINK_AUTONEG;
2479
2480         if (!hw->started) {
2481                 link.link_status = ETH_LINK_DOWN;
2482                 link.link_speed = ETH_SPEED_NUM_NONE;
2483         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2484                 PMD_INIT_LOG(DEBUG, "Get link status from hw");
2485                 vtpci_read_dev_config(hw,
2486                                 offsetof(struct virtio_net_config, status),
2487                                 &status, sizeof(status));
2488                 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2489                         link.link_status = ETH_LINK_DOWN;
2490                         link.link_speed = ETH_SPEED_NUM_NONE;
2491                         PMD_INIT_LOG(DEBUG, "Port %d is down",
2492                                      dev->data->port_id);
2493                 } else {
2494                         link.link_status = ETH_LINK_UP;
2495                         PMD_INIT_LOG(DEBUG, "Port %d is up",
2496                                      dev->data->port_id);
2497                 }
2498         } else {
2499                 link.link_status = ETH_LINK_UP;
2500         }
2501
2502         return rte_eth_linkstatus_set(dev, &link);
2503 }
2504
2505 static int
2506 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2507 {
2508         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2509         struct virtio_hw *hw = dev->data->dev_private;
2510         uint64_t offloads = rxmode->offloads;
2511
2512         if (mask & ETH_VLAN_FILTER_MASK) {
2513                 if ((offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2514                                 !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2515
2516                         PMD_DRV_LOG(NOTICE,
2517                                 "vlan filtering not available on this host");
2518
2519                         return -ENOTSUP;
2520                 }
2521         }
2522
2523         if (mask & ETH_VLAN_STRIP_MASK)
2524                 hw->vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
2525
2526         return 0;
2527 }
2528
2529 static int
2530 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2531 {
2532         uint64_t tso_mask, host_features;
2533         struct virtio_hw *hw = dev->data->dev_private;
2534         dev_info->speed_capa = virtio_dev_speed_capa_get(hw->speed);
2535
2536         dev_info->max_rx_queues =
2537                 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2538         dev_info->max_tx_queues =
2539                 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2540         dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2541         dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2542         dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2543
2544         host_features = VTPCI_OPS(hw)->get_features(hw);
2545         dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
2546         dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2547         if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2548                 dev_info->rx_offload_capa |=
2549                         DEV_RX_OFFLOAD_TCP_CKSUM |
2550                         DEV_RX_OFFLOAD_UDP_CKSUM;
2551         }
2552         if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN))
2553                 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_FILTER;
2554         tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2555                 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2556         if ((host_features & tso_mask) == tso_mask)
2557                 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2558
2559         dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
2560                                     DEV_TX_OFFLOAD_VLAN_INSERT;
2561         if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2562                 dev_info->tx_offload_capa |=
2563                         DEV_TX_OFFLOAD_UDP_CKSUM |
2564                         DEV_TX_OFFLOAD_TCP_CKSUM;
2565         }
2566         tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2567                 (1ULL << VIRTIO_NET_F_HOST_TSO6);
2568         if ((host_features & tso_mask) == tso_mask)
2569                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2570
2571         return 0;
2572 }
2573
2574 /*
2575  * It enables testpmd to collect per queue stats.
2576  */
2577 static int
2578 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2579 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2580 __rte_unused uint8_t is_rx)
2581 {
2582         return 0;
2583 }
2584
2585 RTE_LOG_REGISTER(virtio_logtype_init, pmd.net.virtio.init, NOTICE);
2586 RTE_LOG_REGISTER(virtio_logtype_driver, pmd.net.virtio.driver, NOTICE);