drivers/net/virtio/virtio_ethdev.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2016 Intel Corporation
   3  */
   4
   5 #include <stdint.h>
   6 #include <string.h>
   7 #include <stdio.h>
   8 #include <errno.h>
   9 #include <unistd.h>
  10
  11 #include <rte_ethdev_driver.h>
  12 #include <rte_ethdev_pci.h>
  13 #include <rte_memcpy.h>
  14 #include <rte_string_fns.h>
  15 #include <rte_memzone.h>
  16 #include <rte_malloc.h>
  17 #include <rte_branch_prediction.h>
  18 #include <rte_pci.h>
  19 #include <rte_bus_pci.h>
  20 #include <rte_ether.h>
  21 #include <rte_ip.h>
  22 #include <rte_arp.h>
  23 #include <rte_common.h>
  24 #include <rte_errno.h>
  25 #include <rte_cpuflags.h>
  26
  27 #include <rte_memory.h>
  28 #include <rte_eal.h>
  29 #include <rte_dev.h>
  30 #include <rte_cycles.h>
  31 #include <rte_kvargs.h>
  32
  33 #include "virtio_ethdev.h"
  34 #include "virtio_pci.h"
  35 #include "virtio_logs.h"
  36 #include "virtqueue.h"
  37 #include "virtio_rxtx.h"
  38
  39 static int eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev);
  40 static int  virtio_dev_configure(struct rte_eth_dev *dev);
  41 static int  virtio_dev_start(struct rte_eth_dev *dev);
  42 static void virtio_dev_stop(struct rte_eth_dev *dev);
  43 static void virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
  44 static void virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
  45 static void virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
  46 static void virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
  47 static void virtio_dev_info_get(struct rte_eth_dev *dev,
  48                                 struct rte_eth_dev_info *dev_info);
  49 static int virtio_dev_link_update(struct rte_eth_dev *dev,
  50         int wait_to_complete);
  51 static int virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask);
  52
  53 static void virtio_set_hwaddr(struct virtio_hw *hw);
  54 static void virtio_get_hwaddr(struct virtio_hw *hw);
  55
  56 static int virtio_dev_stats_get(struct rte_eth_dev *dev,
  57                                  struct rte_eth_stats *stats);
  58 static int virtio_dev_xstats_get(struct rte_eth_dev *dev,
  59                                  struct rte_eth_xstat *xstats, unsigned n);
  60 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
  61                                        struct rte_eth_xstat_name *xstats_names,
  62                                        unsigned limit);
  63 static void virtio_dev_stats_reset(struct rte_eth_dev *dev);
  64 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
  65 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
  66                                 uint16_t vlan_id, int on);
  67 static int virtio_mac_addr_add(struct rte_eth_dev *dev,
  68                                 struct ether_addr *mac_addr,
  69                                 uint32_t index, uint32_t vmdq);
  70 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
  71 static int virtio_mac_addr_set(struct rte_eth_dev *dev,
  72                                 struct ether_addr *mac_addr);
  73
  74 static int virtio_intr_disable(struct rte_eth_dev *dev);
  75
  76 static int virtio_dev_queue_stats_mapping_set(
  77         struct rte_eth_dev *eth_dev,
  78         uint16_t queue_id,
  79         uint8_t stat_idx,
  80         uint8_t is_rx);
  81
  82 int virtio_logtype_init;
  83 int virtio_logtype_driver;
  84
  85 static void virtio_notify_peers(struct rte_eth_dev *dev);
  86 static void virtio_ack_link_announce(struct rte_eth_dev *dev);
  87
  88 /*
  89  * The set of PCI devices this driver supports
  90  */
  91 static const struct rte_pci_id pci_id_virtio_map[] = {
  92         { RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_LEGACY_DEVICEID_NET) },
  93         { RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_MODERN_DEVICEID_NET) },
  94         { .vendor_id = 0, /* sentinel */ },
  95 };
  96
  97 struct rte_virtio_xstats_name_off {
  98         char name[RTE_ETH_XSTATS_NAME_SIZE];
  99         unsigned offset;
 100 };
 101
 102 /* [rt]x_qX_ is prepended to the name string here */
 103 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
 104         {"good_packets",           offsetof(struct virtnet_rx, stats.packets)},
 105         {"good_bytes",             offsetof(struct virtnet_rx, stats.bytes)},
 106         {"errors",                 offsetof(struct virtnet_rx, stats.errors)},
 107         {"multicast_packets",      offsetof(struct virtnet_rx, stats.multicast)},
 108         {"broadcast_packets",      offsetof(struct virtnet_rx, stats.broadcast)},
 109         {"undersize_packets",      offsetof(struct virtnet_rx, stats.size_bins[0])},
 110         {"size_64_packets",        offsetof(struct virtnet_rx, stats.size_bins[1])},
 111         {"size_65_127_packets",    offsetof(struct virtnet_rx, stats.size_bins[2])},
 112         {"size_128_255_packets",   offsetof(struct virtnet_rx, stats.size_bins[3])},
 113         {"size_256_511_packets",   offsetof(struct virtnet_rx, stats.size_bins[4])},
 114         {"size_512_1023_packets",  offsetof(struct virtnet_rx, stats.size_bins[5])},
 115         {"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
 116         {"size_1519_max_packets",  offsetof(struct virtnet_rx, stats.size_bins[7])},
 117 };
 118
 119 /* [rt]x_qX_ is prepended to the name string here */
 120 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
 121         {"good_packets",           offsetof(struct virtnet_tx, stats.packets)},
 122         {"good_bytes",             offsetof(struct virtnet_tx, stats.bytes)},
 123         {"errors",                 offsetof(struct virtnet_tx, stats.errors)},
 124         {"multicast_packets",      offsetof(struct virtnet_tx, stats.multicast)},
 125         {"broadcast_packets",      offsetof(struct virtnet_tx, stats.broadcast)},
 126         {"undersize_packets",      offsetof(struct virtnet_tx, stats.size_bins[0])},
 127         {"size_64_packets",        offsetof(struct virtnet_tx, stats.size_bins[1])},
 128         {"size_65_127_packets",    offsetof(struct virtnet_tx, stats.size_bins[2])},
 129         {"size_128_255_packets",   offsetof(struct virtnet_tx, stats.size_bins[3])},
 130         {"size_256_511_packets",   offsetof(struct virtnet_tx, stats.size_bins[4])},
 131         {"size_512_1023_packets",  offsetof(struct virtnet_tx, stats.size_bins[5])},
 132         {"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
 133         {"size_1519_max_packets",  offsetof(struct virtnet_tx, stats.size_bins[7])},
 134 };
 135
 136 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
 137                             sizeof(rte_virtio_rxq_stat_strings[0]))
 138 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
 139                             sizeof(rte_virtio_txq_stat_strings[0]))
 140
 141 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
 142
 143 static struct virtio_pmd_ctrl *
 144 virtio_send_command_packed(struct virtnet_ctl *cvq,
 145                            struct virtio_pmd_ctrl *ctrl,
 146                            int *dlen, int pkt_num)
 147 {
 148         struct virtqueue *vq = cvq->vq;
 149         int head;
 150         struct vring_packed_desc *desc = vq->vq_packed.ring.desc;
 151         struct virtio_pmd_ctrl *result;
 152         uint16_t flags;
 153         int sum = 0;
 154         int nb_descs = 0;
 155         int k;
 156
 157         /*
 158          * Format is enforced in qemu code:
 159          * One TX packet for header;
 160          * At least one TX packet per argument;
 161          * One RX packet for ACK.
 162          */
 163         head = vq->vq_avail_idx;
 164         flags = vq->vq_packed.cached_flags;
 165         desc[head].addr = cvq->virtio_net_hdr_mem;
 166         desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
 167         vq->vq_free_cnt--;
 168         nb_descs++;
 169         if (++vq->vq_avail_idx >= vq->vq_nentries) {
 170                 vq->vq_avail_idx -= vq->vq_nentries;
 171                 vq->vq_packed.cached_flags ^=
 172                         VRING_DESC_F_AVAIL(1) | VRING_DESC_F_USED(1);
 173         }
 174
 175         for (k = 0; k < pkt_num; k++) {
 176                 desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
 177                         + sizeof(struct virtio_net_ctrl_hdr)
 178                         + sizeof(ctrl->status) + sizeof(uint8_t) * sum;
 179                 desc[vq->vq_avail_idx].len = dlen[k];
 180                 desc[vq->vq_avail_idx].flags = VRING_DESC_F_NEXT |
 181                         vq->vq_packed.cached_flags;
 182                 sum += dlen[k];
 183                 vq->vq_free_cnt--;
 184                 nb_descs++;
 185                 if (++vq->vq_avail_idx >= vq->vq_nentries) {
 186                         vq->vq_avail_idx -= vq->vq_nentries;
 187                         vq->vq_packed.cached_flags ^=
 188                                 VRING_DESC_F_AVAIL(1) | VRING_DESC_F_USED(1);
 189                 }
 190         }
 191
 192         desc[vq->vq_avail_idx].addr = cvq->virtio_net_hdr_mem
 193                 + sizeof(struct virtio_net_ctrl_hdr);
 194         desc[vq->vq_avail_idx].len = sizeof(ctrl->status);
 195         desc[vq->vq_avail_idx].flags = VRING_DESC_F_WRITE |
 196                 vq->vq_packed.cached_flags;
 197         vq->vq_free_cnt--;
 198         nb_descs++;
 199         if (++vq->vq_avail_idx >= vq->vq_nentries) {
 200                 vq->vq_avail_idx -= vq->vq_nentries;
 201                 vq->vq_packed.cached_flags ^=
 202                         VRING_DESC_F_AVAIL(1) | VRING_DESC_F_USED(1);
 203         }
 204
 205         virtio_wmb(vq->hw->weak_barriers);
 206         desc[head].flags = VRING_DESC_F_NEXT | flags;
 207
 208         virtio_wmb(vq->hw->weak_barriers);
 209         virtqueue_notify(vq);
 210
 211         /* wait for used descriptors in virtqueue */
 212         while (!desc_is_used(&desc[head], vq))
 213                 usleep(100);
 214
 215         virtio_rmb(vq->hw->weak_barriers);
 216
 217         /* now get used descriptors */
 218         vq->vq_free_cnt += nb_descs;
 219         vq->vq_used_cons_idx += nb_descs;
 220         if (vq->vq_used_cons_idx >= vq->vq_nentries) {
 221                 vq->vq_used_cons_idx -= vq->vq_nentries;
 222                 vq->vq_packed.used_wrap_counter ^= 1;
 223         }
 224
 225         PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\n"
 226                         "vq->vq_avail_idx=%d\n"
 227                         "vq->vq_used_cons_idx=%d\n"
 228                         "vq->vq_packed.cached_flags=0x%x\n"
 229                         "vq->vq_packed.used_wrap_counter=%d\n",
 230                         vq->vq_free_cnt,
 231                         vq->vq_avail_idx,
 232                         vq->vq_used_cons_idx,
 233                         vq->vq_packed.cached_flags,
 234                         vq->vq_packed.used_wrap_counter);
 235
 236         result = cvq->virtio_net_hdr_mz->addr;
 237         return result;
 238 }
 239
 240 static struct virtio_pmd_ctrl *
 241 virtio_send_command_split(struct virtnet_ctl *cvq,
 242                           struct virtio_pmd_ctrl *ctrl,
 243                           int *dlen, int pkt_num)
 244 {
 245         struct virtio_pmd_ctrl *result;
 246         struct virtqueue *vq = cvq->vq;
 247         uint32_t head, i;
 248         int k, sum = 0;
 249
 250         head = vq->vq_desc_head_idx;
 251
 252         /*
 253          * Format is enforced in qemu code:
 254          * One TX packet for header;
 255          * At least one TX packet per argument;
 256          * One RX packet for ACK.
 257          */
 258         vq->vq_split.ring.desc[head].flags = VRING_DESC_F_NEXT;
 259         vq->vq_split.ring.desc[head].addr = cvq->virtio_net_hdr_mem;
 260         vq->vq_split.ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
 261         vq->vq_free_cnt--;
 262         i = vq->vq_split.ring.desc[head].next;
 263
 264         for (k = 0; k < pkt_num; k++) {
 265                 vq->vq_split.ring.desc[i].flags = VRING_DESC_F_NEXT;
 266                 vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
 267                         + sizeof(struct virtio_net_ctrl_hdr)
 268                         + sizeof(ctrl->status) + sizeof(uint8_t)*sum;
 269                 vq->vq_split.ring.desc[i].len = dlen[k];
 270                 sum += dlen[k];
 271                 vq->vq_free_cnt--;
 272                 i = vq->vq_split.ring.desc[i].next;
 273         }
 274
 275         vq->vq_split.ring.desc[i].flags = VRING_DESC_F_WRITE;
 276         vq->vq_split.ring.desc[i].addr = cvq->virtio_net_hdr_mem
 277                         + sizeof(struct virtio_net_ctrl_hdr);
 278         vq->vq_split.ring.desc[i].len = sizeof(ctrl->status);
 279         vq->vq_free_cnt--;
 280
 281         vq->vq_desc_head_idx = vq->vq_split.ring.desc[i].next;
 282
 283         vq_update_avail_ring(vq, head);
 284         vq_update_avail_idx(vq);
 285
 286         PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
 287
 288         virtqueue_notify(vq);
 289
 290         rte_rmb();
 291         while (VIRTQUEUE_NUSED(vq) == 0) {
 292                 rte_rmb();
 293                 usleep(100);
 294         }
 295
 296         while (VIRTQUEUE_NUSED(vq)) {
 297                 uint32_t idx, desc_idx, used_idx;
 298                 struct vring_used_elem *uep;
 299
 300                 used_idx = (uint32_t)(vq->vq_used_cons_idx
 301                                 & (vq->vq_nentries - 1));
 302                 uep = &vq->vq_split.ring.used->ring[used_idx];
 303                 idx = (uint32_t) uep->id;
 304                 desc_idx = idx;
 305
 306                 while (vq->vq_split.ring.desc[desc_idx].flags &
 307                                 VRING_DESC_F_NEXT) {
 308                         desc_idx = vq->vq_split.ring.desc[desc_idx].next;
 309                         vq->vq_free_cnt++;
 310                 }
 311
 312                 vq->vq_split.ring.desc[desc_idx].next = vq->vq_desc_head_idx;
 313                 vq->vq_desc_head_idx = idx;
 314
 315                 vq->vq_used_cons_idx++;
 316                 vq->vq_free_cnt++;
 317         }
 318
 319         PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
 320                         vq->vq_free_cnt, vq->vq_desc_head_idx);
 321
 322         result = cvq->virtio_net_hdr_mz->addr;
 323         return result;
 324 }
 325
 326 static int
 327 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
 328                     int *dlen, int pkt_num)
 329 {
 330         virtio_net_ctrl_ack status = ~0;
 331         struct virtio_pmd_ctrl *result;
 332         struct virtqueue *vq;
 333
 334         ctrl->status = status;
 335
 336         if (!cvq || !cvq->vq) {
 337                 PMD_INIT_LOG(ERR, "Control queue is not supported.");
 338                 return -1;
 339         }
 340
 341         rte_spinlock_lock(&cvq->lock);
 342         vq = cvq->vq;
 343
 344         PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
 345                 "vq->hw->cvq = %p vq = %p",
 346                 vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
 347
 348         if (vq->vq_free_cnt < pkt_num + 2 || pkt_num < 1) {
 349                 rte_spinlock_unlock(&cvq->lock);
 350                 return -1;
 351         }
 352
 353         memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
 354                 sizeof(struct virtio_pmd_ctrl));
 355
 356         if (vtpci_packed_queue(vq->hw))
 357                 result = virtio_send_command_packed(cvq, ctrl, dlen, pkt_num);
 358         else
 359                 result = virtio_send_command_split(cvq, ctrl, dlen, pkt_num);
 360
 361         rte_spinlock_unlock(&cvq->lock);
 362         return result->status;
 363 }
 364
 365 static int
 366 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
 367 {
 368         struct virtio_hw *hw = dev->data->dev_private;
 369         struct virtio_pmd_ctrl ctrl;
 370         int dlen[1];
 371         int ret;
 372
 373         ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
 374         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
 375         memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
 376
 377         dlen[0] = sizeof(uint16_t);
 378
 379         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 380         if (ret) {
 381                 PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
 382                           "failed, this is too late now...");
 383                 return -EINVAL;
 384         }
 385
 386         return 0;
 387 }
 388
 389 static void
 390 virtio_dev_queue_release(void *queue __rte_unused)
 391 {
 392         /* do nothing */
 393 }
 394
 395 static uint16_t
 396 virtio_get_nr_vq(struct virtio_hw *hw)
 397 {
 398         uint16_t nr_vq = hw->max_queue_pairs * 2;
 399
 400         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
 401                 nr_vq += 1;
 402
 403         return nr_vq;
 404 }
 405
 406 static void
 407 virtio_init_vring(struct virtqueue *vq)
 408 {
 409         int size = vq->vq_nentries;
 410         uint8_t *ring_mem = vq->vq_ring_virt_mem;
 411
 412         PMD_INIT_FUNC_TRACE();
 413
 414         memset(ring_mem, 0, vq->vq_ring_size);
 415
 416         vq->vq_used_cons_idx = 0;
 417         vq->vq_desc_head_idx = 0;
 418         vq->vq_avail_idx = 0;
 419         vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
 420         vq->vq_free_cnt = vq->vq_nentries;
 421         memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
 422         if (vtpci_packed_queue(vq->hw)) {
 423                 vring_init_packed(&vq->vq_packed.ring, ring_mem,
 424                                   VIRTIO_PCI_VRING_ALIGN, size);
 425                 vring_desc_init_packed(vq, size);
 426         } else {
 427                 struct vring *vr = &vq->vq_split.ring;
 428
 429                 vring_init_split(vr, ring_mem, VIRTIO_PCI_VRING_ALIGN, size);
 430                 vring_desc_init_split(vr->desc, size);
 431         }
 432         /*
 433          * Disable device(host) interrupting guest
 434          */
 435         virtqueue_disable_intr(vq);
 436 }
 437
 438 static int
 439 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
 440 {
 441         char vq_name[VIRTQUEUE_MAX_NAME_SZ];
 442         char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
 443         const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
 444         unsigned int vq_size, size;
 445         struct virtio_hw *hw = dev->data->dev_private;
 446         struct virtnet_rx *rxvq = NULL;
 447         struct virtnet_tx *txvq = NULL;
 448         struct virtnet_ctl *cvq = NULL;
 449         struct virtqueue *vq;
 450         size_t sz_hdr_mz = 0;
 451         void *sw_ring = NULL;
 452         int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
 453         int ret;
 454         int numa_node = dev->device->numa_node;
 455
 456         PMD_INIT_LOG(INFO, "setting up queue: %u on NUMA node %d",
 457                         vtpci_queue_idx, numa_node);
 458
 459         /*
 460          * Read the virtqueue size from the Queue Size field
 461          * Always power of 2 and if 0 virtqueue does not exist
 462          */
 463         vq_size = VTPCI_OPS(hw)->get_queue_num(hw, vtpci_queue_idx);
 464         PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
 465         if (vq_size == 0) {
 466                 PMD_INIT_LOG(ERR, "virtqueue does not exist");
 467                 return -EINVAL;
 468         }
 469
 470         if (!rte_is_power_of_2(vq_size)) {
 471                 PMD_INIT_LOG(ERR, "virtqueue size is not powerof 2");
 472                 return -EINVAL;
 473         }
 474
 475         snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
 476                  dev->data->port_id, vtpci_queue_idx);
 477
 478         size = RTE_ALIGN_CEIL(sizeof(*vq) +
 479                                 vq_size * sizeof(struct vq_desc_extra),
 480                                 RTE_CACHE_LINE_SIZE);
 481         if (queue_type == VTNET_TQ) {
 482                 /*
 483                  * For each xmit packet, allocate a virtio_net_hdr
 484                  * and indirect ring elements
 485                  */
 486                 sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
 487         } else if (queue_type == VTNET_CQ) {
 488                 /* Allocate a page for control vq command, data and status */
 489                 sz_hdr_mz = PAGE_SIZE;
 490         }
 491
 492         vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
 493                                 numa_node);
 494         if (vq == NULL) {
 495                 PMD_INIT_LOG(ERR, "can not allocate vq");
 496                 return -ENOMEM;
 497         }
 498         hw->vqs[vtpci_queue_idx] = vq;
 499
 500         vq->hw = hw;
 501         vq->vq_queue_index = vtpci_queue_idx;
 502         vq->vq_nentries = vq_size;
 503         if (vtpci_packed_queue(hw)) {
 504                 vq->vq_packed.used_wrap_counter = 1;
 505                 vq->vq_packed.cached_flags = VRING_DESC_F_AVAIL(1);
 506                 vq->vq_packed.event_flags_shadow = 0;
 507                 if (queue_type == VTNET_RQ)
 508                         vq->vq_packed.cached_flags |= VRING_DESC_F_WRITE;
 509         }
 510
 511         /*
 512          * Reserve a memzone for vring elements
 513          */
 514         size = vring_size(hw, vq_size, VIRTIO_PCI_VRING_ALIGN);
 515         vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
 516         PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
 517                      size, vq->vq_ring_size);
 518
 519         mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
 520                         numa_node, RTE_MEMZONE_IOVA_CONTIG,
 521                         VIRTIO_PCI_VRING_ALIGN);
 522         if (mz == NULL) {
 523                 if (rte_errno == EEXIST)
 524                         mz = rte_memzone_lookup(vq_name);
 525                 if (mz == NULL) {
 526                         ret = -ENOMEM;
 527                         goto fail_q_alloc;
 528                 }
 529         }
 530
 531         memset(mz->addr, 0, mz->len);
 532
 533         vq->vq_ring_mem = mz->iova;
 534         vq->vq_ring_virt_mem = mz->addr;
 535         PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%" PRIx64,
 536                      (uint64_t)mz->iova);
 537         PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
 538                      (uint64_t)(uintptr_t)mz->addr);
 539
 540         virtio_init_vring(vq);
 541
 542         if (sz_hdr_mz) {
 543                 snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
 544                          dev->data->port_id, vtpci_queue_idx);
 545                 hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
 546                                 numa_node, RTE_MEMZONE_IOVA_CONTIG,
 547                                 RTE_CACHE_LINE_SIZE);
 548                 if (hdr_mz == NULL) {
 549                         if (rte_errno == EEXIST)
 550                                 hdr_mz = rte_memzone_lookup(vq_hdr_name);
 551                         if (hdr_mz == NULL) {
 552                                 ret = -ENOMEM;
 553                                 goto fail_q_alloc;
 554                         }
 555                 }
 556         }
 557
 558         if (queue_type == VTNET_RQ) {
 559                 size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
 560                                sizeof(vq->sw_ring[0]);
 561
 562                 sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
 563                                 RTE_CACHE_LINE_SIZE, numa_node);
 564                 if (!sw_ring) {
 565                         PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
 566                         ret = -ENOMEM;
 567                         goto fail_q_alloc;
 568                 }
 569
 570                 vq->sw_ring = sw_ring;
 571                 rxvq = &vq->rxq;
 572                 rxvq->vq = vq;
 573                 rxvq->port_id = dev->data->port_id;
 574                 rxvq->mz = mz;
 575         } else if (queue_type == VTNET_TQ) {
 576                 txvq = &vq->txq;
 577                 txvq->vq = vq;
 578                 txvq->port_id = dev->data->port_id;
 579                 txvq->mz = mz;
 580                 txvq->virtio_net_hdr_mz = hdr_mz;
 581                 txvq->virtio_net_hdr_mem = hdr_mz->iova;
 582         } else if (queue_type == VTNET_CQ) {
 583                 cvq = &vq->cq;
 584                 cvq->vq = vq;
 585                 cvq->mz = mz;
 586                 cvq->virtio_net_hdr_mz = hdr_mz;
 587                 cvq->virtio_net_hdr_mem = hdr_mz->iova;
 588                 memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
 589
 590                 hw->cvq = cvq;
 591         }
 592
 593         /* For virtio_user case (that is when hw->dev is NULL), we use
 594          * virtual address. And we need properly set _offset_, please see
 595          * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
 596          */
 597         if (!hw->virtio_user_dev)
 598                 vq->offset = offsetof(struct rte_mbuf, buf_iova);
 599         else {
 600                 vq->vq_ring_mem = (uintptr_t)mz->addr;
 601                 vq->offset = offsetof(struct rte_mbuf, buf_addr);
 602                 if (queue_type == VTNET_TQ)
 603                         txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
 604                 else if (queue_type == VTNET_CQ)
 605                         cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
 606         }
 607
 608         if (queue_type == VTNET_TQ) {
 609                 struct virtio_tx_region *txr;
 610                 unsigned int i;
 611
 612                 txr = hdr_mz->addr;
 613                 memset(txr, 0, vq_size * sizeof(*txr));
 614                 for (i = 0; i < vq_size; i++) {
 615                         struct vring_desc *start_dp = txr[i].tx_indir;
 616
 617                         /* first indirect descriptor is always the tx header */
 618                         if (!vtpci_packed_queue(hw)) {
 619                                 vring_desc_init_split(start_dp,
 620                                                       RTE_DIM(txr[i].tx_indir));
 621                                 start_dp->addr = txvq->virtio_net_hdr_mem
 622                                         + i * sizeof(*txr)
 623                                         + offsetof(struct virtio_tx_region,
 624                                                    tx_hdr);
 625                                 start_dp->len = hw->vtnet_hdr_size;
 626                                 start_dp->flags = VRING_DESC_F_NEXT;
 627                         }
 628                 }
 629         }
 630
 631         if (VTPCI_OPS(hw)->setup_queue(hw, vq) < 0) {
 632                 PMD_INIT_LOG(ERR, "setup_queue failed");
 633                 return -EINVAL;
 634         }
 635
 636         return 0;
 637
 638 fail_q_alloc:
 639         rte_free(sw_ring);
 640         rte_memzone_free(hdr_mz);
 641         rte_memzone_free(mz);
 642         rte_free(vq);
 643
 644         return ret;
 645 }
 646
 647 static void
 648 virtio_free_queues(struct virtio_hw *hw)
 649 {
 650         uint16_t nr_vq = virtio_get_nr_vq(hw);
 651         struct virtqueue *vq;
 652         int queue_type;
 653         uint16_t i;
 654
 655         if (hw->vqs == NULL)
 656                 return;
 657
 658         for (i = 0; i < nr_vq; i++) {
 659                 vq = hw->vqs[i];
 660                 if (!vq)
 661                         continue;
 662
 663                 queue_type = virtio_get_queue_type(hw, i);
 664                 if (queue_type == VTNET_RQ) {
 665                         rte_free(vq->sw_ring);
 666                         rte_memzone_free(vq->rxq.mz);
 667                 } else if (queue_type == VTNET_TQ) {
 668                         rte_memzone_free(vq->txq.mz);
 669                         rte_memzone_free(vq->txq.virtio_net_hdr_mz);
 670                 } else {
 671                         rte_memzone_free(vq->cq.mz);
 672                         rte_memzone_free(vq->cq.virtio_net_hdr_mz);
 673                 }
 674
 675                 rte_free(vq);
 676                 hw->vqs[i] = NULL;
 677         }
 678
 679         rte_free(hw->vqs);
 680         hw->vqs = NULL;
 681 }
 682
 683 static int
 684 virtio_alloc_queues(struct rte_eth_dev *dev)
 685 {
 686         struct virtio_hw *hw = dev->data->dev_private;
 687         uint16_t nr_vq = virtio_get_nr_vq(hw);
 688         uint16_t i;
 689         int ret;
 690
 691         hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
 692         if (!hw->vqs) {
 693                 PMD_INIT_LOG(ERR, "failed to allocate vqs");
 694                 return -ENOMEM;
 695         }
 696
 697         for (i = 0; i < nr_vq; i++) {
 698                 ret = virtio_init_queue(dev, i);
 699                 if (ret < 0) {
 700                         virtio_free_queues(hw);
 701                         return ret;
 702                 }
 703         }
 704
 705         return 0;
 706 }
 707
 708 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
 709
 710 static void
 711 virtio_dev_close(struct rte_eth_dev *dev)
 712 {
 713         struct virtio_hw *hw = dev->data->dev_private;
 714         struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
 715
 716         PMD_INIT_LOG(DEBUG, "virtio_dev_close");
 717
 718         if (!hw->opened)
 719                 return;
 720         hw->opened = false;
 721
 722         /* reset the NIC */
 723         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
 724                 VTPCI_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
 725         if (intr_conf->rxq)
 726                 virtio_queues_unbind_intr(dev);
 727
 728         if (intr_conf->lsc || intr_conf->rxq) {
 729                 virtio_intr_disable(dev);
 730                 rte_intr_efd_disable(dev->intr_handle);
 731                 rte_free(dev->intr_handle->intr_vec);
 732                 dev->intr_handle->intr_vec = NULL;
 733         }
 734
 735         vtpci_reset(hw);
 736         virtio_dev_free_mbufs(dev);
 737         virtio_free_queues(hw);
 738 }
 739
 740 static void
 741 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
 742 {
 743         struct virtio_hw *hw = dev->data->dev_private;
 744         struct virtio_pmd_ctrl ctrl;
 745         int dlen[1];
 746         int ret;
 747
 748         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 749                 PMD_INIT_LOG(INFO, "host does not support rx control");
 750                 return;
 751         }
 752
 753         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 754         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 755         ctrl.data[0] = 1;
 756         dlen[0] = 1;
 757
 758         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 759         if (ret)
 760                 PMD_INIT_LOG(ERR, "Failed to enable promisc");
 761 }
 762
 763 static void
 764 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
 765 {
 766         struct virtio_hw *hw = dev->data->dev_private;
 767         struct virtio_pmd_ctrl ctrl;
 768         int dlen[1];
 769         int ret;
 770
 771         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 772                 PMD_INIT_LOG(INFO, "host does not support rx control");
 773                 return;
 774         }
 775
 776         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 777         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 778         ctrl.data[0] = 0;
 779         dlen[0] = 1;
 780
 781         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 782         if (ret)
 783                 PMD_INIT_LOG(ERR, "Failed to disable promisc");
 784 }
 785
 786 static void
 787 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
 788 {
 789         struct virtio_hw *hw = dev->data->dev_private;
 790         struct virtio_pmd_ctrl ctrl;
 791         int dlen[1];
 792         int ret;
 793
 794         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 795                 PMD_INIT_LOG(INFO, "host does not support rx control");
 796                 return;
 797         }
 798
 799         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 800         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 801         ctrl.data[0] = 1;
 802         dlen[0] = 1;
 803
 804         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 805         if (ret)
 806                 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
 807 }
 808
 809 static void
 810 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
 811 {
 812         struct virtio_hw *hw = dev->data->dev_private;
 813         struct virtio_pmd_ctrl ctrl;
 814         int dlen[1];
 815         int ret;
 816
 817         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 818                 PMD_INIT_LOG(INFO, "host does not support rx control");
 819                 return;
 820         }
 821
 822         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 823         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 824         ctrl.data[0] = 0;
 825         dlen[0] = 1;
 826
 827         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 828         if (ret)
 829                 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
 830 }
 831
 832 #define VLAN_TAG_LEN           4    /* 802.3ac tag (not DMA'd) */
 833 static int
 834 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 835 {
 836         struct virtio_hw *hw = dev->data->dev_private;
 837         uint32_t ether_hdr_len = ETHER_HDR_LEN + VLAN_TAG_LEN +
 838                                  hw->vtnet_hdr_size;
 839         uint32_t frame_size = mtu + ether_hdr_len;
 840         uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
 841
 842         max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
 843
 844         if (mtu < ETHER_MIN_MTU || frame_size > max_frame_size) {
 845                 PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
 846                         ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
 847                 return -EINVAL;
 848         }
 849         return 0;
 850 }
 851
 852 static int
 853 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
 854 {
 855         struct virtio_hw *hw = dev->data->dev_private;
 856         struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
 857         struct virtqueue *vq = rxvq->vq;
 858
 859         virtqueue_enable_intr(vq);
 860         virtio_mb(hw->weak_barriers);
 861         return 0;
 862 }
 863
 864 static int
 865 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
 866 {
 867         struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
 868         struct virtqueue *vq = rxvq->vq;
 869
 870         virtqueue_disable_intr(vq);
 871         return 0;
 872 }
 873
 874 /*
 875  * dev_ops for virtio, bare necessities for basic operation
 876  */
 877 static const struct eth_dev_ops virtio_eth_dev_ops = {
 878         .dev_configure           = virtio_dev_configure,
 879         .dev_start               = virtio_dev_start,
 880         .dev_stop                = virtio_dev_stop,
 881         .dev_close               = virtio_dev_close,
 882         .promiscuous_enable      = virtio_dev_promiscuous_enable,
 883         .promiscuous_disable     = virtio_dev_promiscuous_disable,
 884         .allmulticast_enable     = virtio_dev_allmulticast_enable,
 885         .allmulticast_disable    = virtio_dev_allmulticast_disable,
 886         .mtu_set                 = virtio_mtu_set,
 887         .dev_infos_get           = virtio_dev_info_get,
 888         .stats_get               = virtio_dev_stats_get,
 889         .xstats_get              = virtio_dev_xstats_get,
 890         .xstats_get_names        = virtio_dev_xstats_get_names,
 891         .stats_reset             = virtio_dev_stats_reset,
 892         .xstats_reset            = virtio_dev_stats_reset,
 893         .link_update             = virtio_dev_link_update,
 894         .vlan_offload_set        = virtio_dev_vlan_offload_set,
 895         .rx_queue_setup          = virtio_dev_rx_queue_setup,
 896         .rx_queue_intr_enable    = virtio_dev_rx_queue_intr_enable,
 897         .rx_queue_intr_disable   = virtio_dev_rx_queue_intr_disable,
 898         .rx_queue_release        = virtio_dev_queue_release,
 899         .rx_descriptor_done      = virtio_dev_rx_queue_done,
 900         .tx_queue_setup          = virtio_dev_tx_queue_setup,
 901         .tx_queue_release        = virtio_dev_queue_release,
 902         /* collect stats per queue */
 903         .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
 904         .vlan_filter_set         = virtio_vlan_filter_set,
 905         .mac_addr_add            = virtio_mac_addr_add,
 906         .mac_addr_remove         = virtio_mac_addr_remove,
 907         .mac_addr_set            = virtio_mac_addr_set,
 908 };
 909
 910 /*
 911  * dev_ops for virtio-user in secondary processes, as we just have
 912  * some limited supports currently.
 913  */
 914 const struct eth_dev_ops virtio_user_secondary_eth_dev_ops = {
 915         .dev_infos_get           = virtio_dev_info_get,
 916         .stats_get               = virtio_dev_stats_get,
 917         .xstats_get              = virtio_dev_xstats_get,
 918         .xstats_get_names        = virtio_dev_xstats_get_names,
 919         .stats_reset             = virtio_dev_stats_reset,
 920         .xstats_reset            = virtio_dev_stats_reset,
 921         /* collect stats per queue */
 922         .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
 923 };
 924
 925 static void
 926 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 927 {
 928         unsigned i;
 929
 930         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 931                 const struct virtnet_tx *txvq = dev->data->tx_queues[i];
 932                 if (txvq == NULL)
 933                         continue;
 934
 935                 stats->opackets += txvq->stats.packets;
 936                 stats->obytes += txvq->stats.bytes;
 937                 stats->oerrors += txvq->stats.errors;
 938
 939                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 940                         stats->q_opackets[i] = txvq->stats.packets;
 941                         stats->q_obytes[i] = txvq->stats.bytes;
 942                 }
 943         }
 944
 945         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 946                 const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
 947                 if (rxvq == NULL)
 948                         continue;
 949
 950                 stats->ipackets += rxvq->stats.packets;
 951                 stats->ibytes += rxvq->stats.bytes;
 952                 stats->ierrors += rxvq->stats.errors;
 953
 954                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 955                         stats->q_ipackets[i] = rxvq->stats.packets;
 956                         stats->q_ibytes[i] = rxvq->stats.bytes;
 957                 }
 958         }
 959
 960         stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
 961 }
 962
 963 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
 964                                        struct rte_eth_xstat_name *xstats_names,
 965                                        __rte_unused unsigned limit)
 966 {
 967         unsigned i;
 968         unsigned count = 0;
 969         unsigned t;
 970
 971         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
 972                 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
 973
 974         if (xstats_names != NULL) {
 975                 /* Note: limit checked in rte_eth_xstats_names() */
 976
 977                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
 978                         struct virtnet_rx *rxvq = dev->data->rx_queues[i];
 979                         if (rxvq == NULL)
 980                                 continue;
 981                         for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
 982                                 snprintf(xstats_names[count].name,
 983                                         sizeof(xstats_names[count].name),
 984                                         "rx_q%u_%s", i,
 985                                         rte_virtio_rxq_stat_strings[t].name);
 986                                 count++;
 987                         }
 988                 }
 989
 990                 for (i = 0; i < dev->data->nb_tx_queues; i++) {
 991                         struct virtnet_tx *txvq = dev->data->tx_queues[i];
 992                         if (txvq == NULL)
 993                                 continue;
 994                         for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
 995                                 snprintf(xstats_names[count].name,
 996                                         sizeof(xstats_names[count].name),
 997                                         "tx_q%u_%s", i,
 998                                         rte_virtio_txq_stat_strings[t].name);
 999                                 count++;
1000                         }
1001                 }
1002                 return count;
1003         }
1004         return nstats;
1005 }
1006
1007 static int
1008 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1009                       unsigned n)
1010 {
1011         unsigned i;
1012         unsigned count = 0;
1013
1014         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
1015                 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
1016
1017         if (n < nstats)
1018                 return nstats;
1019
1020         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1021                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1022
1023                 if (rxvq == NULL)
1024                         continue;
1025
1026                 unsigned t;
1027
1028                 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
1029                         xstats[count].value = *(uint64_t *)(((char *)rxvq) +
1030                                 rte_virtio_rxq_stat_strings[t].offset);
1031                         xstats[count].id = count;
1032                         count++;
1033                 }
1034         }
1035
1036         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1037                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1038
1039                 if (txvq == NULL)
1040                         continue;
1041
1042                 unsigned t;
1043
1044                 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
1045                         xstats[count].value = *(uint64_t *)(((char *)txvq) +
1046                                 rte_virtio_txq_stat_strings[t].offset);
1047                         xstats[count].id = count;
1048                         count++;
1049                 }
1050         }
1051
1052         return count;
1053 }
1054
1055 static int
1056 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1057 {
1058         virtio_update_stats(dev, stats);
1059
1060         return 0;
1061 }
1062
1063 static void
1064 virtio_dev_stats_reset(struct rte_eth_dev *dev)
1065 {
1066         unsigned int i;
1067
1068         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1069                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1070                 if (txvq == NULL)
1071                         continue;
1072
1073                 txvq->stats.packets = 0;
1074                 txvq->stats.bytes = 0;
1075                 txvq->stats.errors = 0;
1076                 txvq->stats.multicast = 0;
1077                 txvq->stats.broadcast = 0;
1078                 memset(txvq->stats.size_bins, 0,
1079                        sizeof(txvq->stats.size_bins[0]) * 8);
1080         }
1081
1082         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1083                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1084                 if (rxvq == NULL)
1085                         continue;
1086
1087                 rxvq->stats.packets = 0;
1088                 rxvq->stats.bytes = 0;
1089                 rxvq->stats.errors = 0;
1090                 rxvq->stats.multicast = 0;
1091                 rxvq->stats.broadcast = 0;
1092                 memset(rxvq->stats.size_bins, 0,
1093                        sizeof(rxvq->stats.size_bins[0]) * 8);
1094         }
1095 }
1096
1097 static void
1098 virtio_set_hwaddr(struct virtio_hw *hw)
1099 {
1100         vtpci_write_dev_config(hw,
1101                         offsetof(struct virtio_net_config, mac),
1102                         &hw->mac_addr, ETHER_ADDR_LEN);
1103 }
1104
1105 static void
1106 virtio_get_hwaddr(struct virtio_hw *hw)
1107 {
1108         if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1109                 vtpci_read_dev_config(hw,
1110                         offsetof(struct virtio_net_config, mac),
1111                         &hw->mac_addr, ETHER_ADDR_LEN);
1112         } else {
1113                 eth_random_addr(&hw->mac_addr[0]);
1114                 virtio_set_hwaddr(hw);
1115         }
1116 }
1117
1118 static int
1119 virtio_mac_table_set(struct virtio_hw *hw,
1120                      const struct virtio_net_ctrl_mac *uc,
1121                      const struct virtio_net_ctrl_mac *mc)
1122 {
1123         struct virtio_pmd_ctrl ctrl;
1124         int err, len[2];
1125
1126         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1127                 PMD_DRV_LOG(INFO, "host does not support mac table");
1128                 return -1;
1129         }
1130
1131         ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1132         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1133
1134         len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
1135         memcpy(ctrl.data, uc, len[0]);
1136
1137         len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
1138         memcpy(ctrl.data + len[0], mc, len[1]);
1139
1140         err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1141         if (err != 0)
1142                 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1143         return err;
1144 }
1145
1146 static int
1147 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
1148                     uint32_t index, uint32_t vmdq __rte_unused)
1149 {
1150         struct virtio_hw *hw = dev->data->dev_private;
1151         const struct ether_addr *addrs = dev->data->mac_addrs;
1152         unsigned int i;
1153         struct virtio_net_ctrl_mac *uc, *mc;
1154
1155         if (index >= VIRTIO_MAX_MAC_ADDRS) {
1156                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1157                 return -EINVAL;
1158         }
1159
1160         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1161         uc->entries = 0;
1162         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1163         mc->entries = 0;
1164
1165         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1166                 const struct ether_addr *addr
1167                         = (i == index) ? mac_addr : addrs + i;
1168                 struct virtio_net_ctrl_mac *tbl
1169                         = is_multicast_ether_addr(addr) ? mc : uc;
1170
1171                 memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
1172         }
1173
1174         return virtio_mac_table_set(hw, uc, mc);
1175 }
1176
1177 static void
1178 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1179 {
1180         struct virtio_hw *hw = dev->data->dev_private;
1181         struct ether_addr *addrs = dev->data->mac_addrs;
1182         struct virtio_net_ctrl_mac *uc, *mc;
1183         unsigned int i;
1184
1185         if (index >= VIRTIO_MAX_MAC_ADDRS) {
1186                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1187                 return;
1188         }
1189
1190         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1191         uc->entries = 0;
1192         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1193         mc->entries = 0;
1194
1195         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1196                 struct virtio_net_ctrl_mac *tbl;
1197
1198                 if (i == index || is_zero_ether_addr(addrs + i))
1199                         continue;
1200
1201                 tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
1202                 memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
1203         }
1204
1205         virtio_mac_table_set(hw, uc, mc);
1206 }
1207
1208 static int
1209 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1210 {
1211         struct virtio_hw *hw = dev->data->dev_private;
1212
1213         memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
1214
1215         /* Use atomic update if available */
1216         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1217                 struct virtio_pmd_ctrl ctrl;
1218                 int len = ETHER_ADDR_LEN;
1219
1220                 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1221                 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1222
1223                 memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
1224                 return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1225         }
1226
1227         if (!vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1228                 return -ENOTSUP;
1229
1230         virtio_set_hwaddr(hw);
1231         return 0;
1232 }
1233
1234 static int
1235 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1236 {
1237         struct virtio_hw *hw = dev->data->dev_private;
1238         struct virtio_pmd_ctrl ctrl;
1239         int len;
1240
1241         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1242                 return -ENOTSUP;
1243
1244         ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1245         ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1246         memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1247         len = sizeof(vlan_id);
1248
1249         return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1250 }
1251
1252 static int
1253 virtio_intr_enable(struct rte_eth_dev *dev)
1254 {
1255         struct virtio_hw *hw = dev->data->dev_private;
1256
1257         if (rte_intr_enable(dev->intr_handle) < 0)
1258                 return -1;
1259
1260         if (!hw->virtio_user_dev)
1261                 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1262
1263         return 0;
1264 }
1265
1266 static int
1267 virtio_intr_disable(struct rte_eth_dev *dev)
1268 {
1269         struct virtio_hw *hw = dev->data->dev_private;
1270
1271         if (rte_intr_disable(dev->intr_handle) < 0)
1272                 return -1;
1273
1274         if (!hw->virtio_user_dev)
1275                 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1276
1277         return 0;
1278 }
1279
1280 static int
1281 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1282 {
1283         uint64_t host_features;
1284
1285         /* Prepare guest_features: feature that driver wants to support */
1286         PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1287                 req_features);
1288
1289         /* Read device(host) feature bits */
1290         host_features = VTPCI_OPS(hw)->get_features(hw);
1291         PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1292                 host_features);
1293
1294         /* If supported, ensure MTU value is valid before acknowledging it. */
1295         if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1296                 struct virtio_net_config config;
1297
1298                 vtpci_read_dev_config(hw,
1299                         offsetof(struct virtio_net_config, mtu),
1300                         &config.mtu, sizeof(config.mtu));
1301
1302                 if (config.mtu < ETHER_MIN_MTU)
1303                         req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1304         }
1305
1306         /*
1307          * Negotiate features: Subset of device feature bits are written back
1308          * guest feature bits.
1309          */
1310         hw->guest_features = req_features;
1311         hw->guest_features = vtpci_negotiate_features(hw, host_features);
1312         PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1313                 hw->guest_features);
1314
1315         if (hw->modern) {
1316                 if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1317                         PMD_INIT_LOG(ERR,
1318                                 "VIRTIO_F_VERSION_1 features is not enabled.");
1319                         return -1;
1320                 }
1321                 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1322                 if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1323                         PMD_INIT_LOG(ERR,
1324                                 "failed to set FEATURES_OK status!");
1325                         return -1;
1326                 }
1327         }
1328
1329         hw->req_guest_features = req_features;
1330
1331         return 0;
1332 }
1333
1334 int
1335 virtio_dev_pause(struct rte_eth_dev *dev)
1336 {
1337         struct virtio_hw *hw = dev->data->dev_private;
1338
1339         rte_spinlock_lock(&hw->state_lock);
1340
1341         if (hw->started == 0) {
1342                 /* Device is just stopped. */
1343                 rte_spinlock_unlock(&hw->state_lock);
1344                 return -1;
1345         }
1346         hw->started = 0;
1347         /*
1348          * Prevent the worker threads from touching queues to avoid contention,
1349          * 1 ms should be enough for the ongoing Tx function to finish.
1350          */
1351         rte_delay_ms(1);
1352         return 0;
1353 }
1354
1355 /*
1356  * Recover hw state to let the worker threads continue.
1357  */
1358 void
1359 virtio_dev_resume(struct rte_eth_dev *dev)
1360 {
1361         struct virtio_hw *hw = dev->data->dev_private;
1362
1363         hw->started = 1;
1364         rte_spinlock_unlock(&hw->state_lock);
1365 }
1366
1367 /*
1368  * Should be called only after device is paused.
1369  */
1370 int
1371 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1372                 int nb_pkts)
1373 {
1374         struct virtio_hw *hw = dev->data->dev_private;
1375         struct virtnet_tx *txvq = dev->data->tx_queues[0];
1376         int ret;
1377
1378         hw->inject_pkts = tx_pkts;
1379         ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1380         hw->inject_pkts = NULL;
1381
1382         return ret;
1383 }
1384
1385 static void
1386 virtio_notify_peers(struct rte_eth_dev *dev)
1387 {
1388         struct virtio_hw *hw = dev->data->dev_private;
1389         struct virtnet_rx *rxvq;
1390         struct rte_mbuf *rarp_mbuf;
1391
1392         if (!dev->data->rx_queues)
1393                 return;
1394
1395         rxvq = dev->data->rx_queues[0];
1396         if (!rxvq)
1397                 return;
1398
1399         rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1400                         (struct ether_addr *)hw->mac_addr);
1401         if (rarp_mbuf == NULL) {
1402                 PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1403                 return;
1404         }
1405
1406         /* If virtio port just stopped, no need to send RARP */
1407         if (virtio_dev_pause(dev) < 0) {
1408                 rte_pktmbuf_free(rarp_mbuf);
1409                 return;
1410         }
1411
1412         virtio_inject_pkts(dev, &rarp_mbuf, 1);
1413         virtio_dev_resume(dev);
1414 }
1415
1416 static void
1417 virtio_ack_link_announce(struct rte_eth_dev *dev)
1418 {
1419         struct virtio_hw *hw = dev->data->dev_private;
1420         struct virtio_pmd_ctrl ctrl;
1421
1422         ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1423         ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1424
1425         virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1426 }
1427
1428 /*
1429  * Process virtio config changed interrupt. Call the callback
1430  * if link state changed, generate gratuitous RARP packet if
1431  * the status indicates an ANNOUNCE.
1432  */
1433 void
1434 virtio_interrupt_handler(void *param)
1435 {
1436         struct rte_eth_dev *dev = param;
1437         struct virtio_hw *hw = dev->data->dev_private;
1438         uint8_t isr;
1439         uint16_t status;
1440
1441         /* Read interrupt status which clears interrupt */
1442         isr = vtpci_isr(hw);
1443         PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1444
1445         if (virtio_intr_enable(dev) < 0)
1446                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1447
1448         if (isr & VIRTIO_PCI_ISR_CONFIG) {
1449                 if (virtio_dev_link_update(dev, 0) == 0)
1450                         _rte_eth_dev_callback_process(dev,
1451                                                       RTE_ETH_EVENT_INTR_LSC,
1452                                                       NULL);
1453
1454                 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1455                         vtpci_read_dev_config(hw,
1456                                 offsetof(struct virtio_net_config, status),
1457                                 &status, sizeof(status));
1458                         if (status & VIRTIO_NET_S_ANNOUNCE) {
1459                                 virtio_notify_peers(dev);
1460                                 if (hw->cvq)
1461                                         virtio_ack_link_announce(dev);
1462                         }
1463                 }
1464         }
1465 }
1466
1467 /* set rx and tx handlers according to what is supported */
1468 static void
1469 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1470 {
1471         struct virtio_hw *hw = eth_dev->data->dev_private;
1472
1473         if (vtpci_packed_queue(hw)) {
1474                 PMD_INIT_LOG(INFO,
1475                         "virtio: using packed ring %s Tx path on port %u",
1476                         hw->use_inorder_tx ? "inorder" : "standard",
1477                         eth_dev->data->port_id);
1478                 eth_dev->tx_pkt_burst = virtio_xmit_pkts_packed;
1479         } else {
1480                 if (hw->use_inorder_tx) {
1481                         PMD_INIT_LOG(INFO, "virtio: using inorder Tx path on port %u",
1482                                 eth_dev->data->port_id);
1483                         eth_dev->tx_pkt_burst = virtio_xmit_pkts_inorder;
1484                 } else {
1485                         PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1486                                 eth_dev->data->port_id);
1487                         eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1488                 }
1489         }
1490
1491         if (vtpci_packed_queue(hw)) {
1492                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1493                         PMD_INIT_LOG(INFO,
1494                                 "virtio: using packed ring mergeable buffer Rx path on port %u",
1495                                 eth_dev->data->port_id);
1496                         eth_dev->rx_pkt_burst =
1497                                 &virtio_recv_mergeable_pkts_packed;
1498                 } else {
1499                         PMD_INIT_LOG(INFO,
1500                                 "virtio: using packed ring standard Rx path on port %u",
1501                                 eth_dev->data->port_id);
1502                         eth_dev->rx_pkt_burst = &virtio_recv_pkts_packed;
1503                 }
1504         } else {
1505                 if (hw->use_simple_rx) {
1506                         PMD_INIT_LOG(INFO, "virtio: using simple Rx path on port %u",
1507                                 eth_dev->data->port_id);
1508                         eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1509                 } else if (hw->use_inorder_rx) {
1510                         PMD_INIT_LOG(INFO,
1511                                 "virtio: using inorder Rx path on port %u",
1512                                 eth_dev->data->port_id);
1513                         eth_dev->rx_pkt_burst = &virtio_recv_pkts_inorder;
1514                 } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1515                         PMD_INIT_LOG(INFO,
1516                                 "virtio: using mergeable buffer Rx path on port %u",
1517                                 eth_dev->data->port_id);
1518                         eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1519                 } else {
1520                         PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1521                                 eth_dev->data->port_id);
1522                         eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1523                 }
1524         }
1525
1526 }
1527
1528 /* Only support 1:1 queue/interrupt mapping so far.
1529  * TODO: support n:1 queue/interrupt mapping when there are limited number of
1530  * interrupt vectors (<N+1).
1531  */
1532 static int
1533 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1534 {
1535         uint32_t i;
1536         struct virtio_hw *hw = dev->data->dev_private;
1537
1538         PMD_INIT_LOG(INFO, "queue/interrupt binding");
1539         for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1540                 dev->intr_handle->intr_vec[i] = i + 1;
1541                 if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1542                                                  VIRTIO_MSI_NO_VECTOR) {
1543                         PMD_DRV_LOG(ERR, "failed to set queue vector");
1544                         return -EBUSY;
1545                 }
1546         }
1547
1548         return 0;
1549 }
1550
1551 static void
1552 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1553 {
1554         uint32_t i;
1555         struct virtio_hw *hw = dev->data->dev_private;
1556
1557         PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1558         for (i = 0; i < dev->data->nb_rx_queues; ++i)
1559                 VTPCI_OPS(hw)->set_queue_irq(hw,
1560                                              hw->vqs[i * VTNET_CQ],
1561                                              VIRTIO_MSI_NO_VECTOR);
1562 }
1563
1564 static int
1565 virtio_configure_intr(struct rte_eth_dev *dev)
1566 {
1567         struct virtio_hw *hw = dev->data->dev_private;
1568
1569         if (!rte_intr_cap_multiple(dev->intr_handle)) {
1570                 PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1571                 return -ENOTSUP;
1572         }
1573
1574         if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1575                 PMD_INIT_LOG(ERR, "Fail to create eventfd");
1576                 return -1;
1577         }
1578
1579         if (!dev->intr_handle->intr_vec) {
1580                 dev->intr_handle->intr_vec =
1581                         rte_zmalloc("intr_vec",
1582                                     hw->max_queue_pairs * sizeof(int), 0);
1583                 if (!dev->intr_handle->intr_vec) {
1584                         PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1585                                      hw->max_queue_pairs);
1586                         return -ENOMEM;
1587                 }
1588         }
1589
1590         /* Re-register callback to update max_intr */
1591         rte_intr_callback_unregister(dev->intr_handle,
1592                                      virtio_interrupt_handler,
1593                                      dev);
1594         rte_intr_callback_register(dev->intr_handle,
1595                                    virtio_interrupt_handler,
1596                                    dev);
1597
1598         /* DO NOT try to remove this! This function will enable msix, or QEMU
1599          * will encounter SIGSEGV when DRIVER_OK is sent.
1600          * And for legacy devices, this should be done before queue/vec binding
1601          * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1602          * (22) will be ignored.
1603          */
1604         if (virtio_intr_enable(dev) < 0) {
1605                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1606                 return -1;
1607         }
1608
1609         if (virtio_queues_bind_intr(dev) < 0) {
1610                 PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1611                 return -1;
1612         }
1613
1614         return 0;
1615 }
1616
1617 /* reset device and renegotiate features if needed */
1618 static int
1619 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1620 {
1621         struct virtio_hw *hw = eth_dev->data->dev_private;
1622         struct virtio_net_config *config;
1623         struct virtio_net_config local_config;
1624         struct rte_pci_device *pci_dev = NULL;
1625         int ret;
1626
1627         /* Reset the device although not necessary at startup */
1628         vtpci_reset(hw);
1629
1630         if (hw->vqs) {
1631                 virtio_dev_free_mbufs(eth_dev);
1632                 virtio_free_queues(hw);
1633         }
1634
1635         /* Tell the host we've noticed this device. */
1636         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1637
1638         /* Tell the host we've known how to drive the device. */
1639         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1640         if (virtio_negotiate_features(hw, req_features) < 0)
1641                 return -1;
1642
1643         hw->weak_barriers = !vtpci_with_feature(hw, VIRTIO_F_ORDER_PLATFORM);
1644
1645         if (!hw->virtio_user_dev) {
1646                 pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1647                 rte_eth_copy_pci_info(eth_dev, pci_dev);
1648         }
1649
1650         /* If host does not support both status and MSI-X then disable LSC */
1651         if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) &&
1652             hw->use_msix != VIRTIO_MSIX_NONE)
1653                 eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1654         else
1655                 eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1656
1657         /* Setting up rx_header size for the device */
1658         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1659             vtpci_with_feature(hw, VIRTIO_F_VERSION_1) ||
1660             vtpci_with_feature(hw, VIRTIO_F_RING_PACKED))
1661                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1662         else
1663                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1664
1665         /* Copy the permanent MAC address to: virtio_hw */
1666         virtio_get_hwaddr(hw);
1667         ether_addr_copy((struct ether_addr *) hw->mac_addr,
1668                         &eth_dev->data->mac_addrs[0]);
1669         PMD_INIT_LOG(DEBUG,
1670                      "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1671                      hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1672                      hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1673
1674         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1675                 config = &local_config;
1676
1677                 vtpci_read_dev_config(hw,
1678                         offsetof(struct virtio_net_config, mac),
1679                         &config->mac, sizeof(config->mac));
1680
1681                 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1682                         vtpci_read_dev_config(hw,
1683                                 offsetof(struct virtio_net_config, status),
1684                                 &config->status, sizeof(config->status));
1685                 } else {
1686                         PMD_INIT_LOG(DEBUG,
1687                                      "VIRTIO_NET_F_STATUS is not supported");
1688                         config->status = 0;
1689                 }
1690
1691                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1692                         vtpci_read_dev_config(hw,
1693                                 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1694                                 &config->max_virtqueue_pairs,
1695                                 sizeof(config->max_virtqueue_pairs));
1696                 } else {
1697                         PMD_INIT_LOG(DEBUG,
1698                                      "VIRTIO_NET_F_MQ is not supported");
1699                         config->max_virtqueue_pairs = 1;
1700                 }
1701
1702                 hw->max_queue_pairs = config->max_virtqueue_pairs;
1703
1704                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1705                         vtpci_read_dev_config(hw,
1706                                 offsetof(struct virtio_net_config, mtu),
1707                                 &config->mtu,
1708                                 sizeof(config->mtu));
1709
1710                         /*
1711                          * MTU value has already been checked at negotiation
1712                          * time, but check again in case it has changed since
1713                          * then, which should not happen.
1714                          */
1715                         if (config->mtu < ETHER_MIN_MTU) {
1716                                 PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1717                                                 config->mtu);
1718                                 return -1;
1719                         }
1720
1721                         hw->max_mtu = config->mtu;
1722                         /* Set initial MTU to maximum one supported by vhost */
1723                         eth_dev->data->mtu = config->mtu;
1724
1725                 } else {
1726                         hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - ETHER_HDR_LEN -
1727                                 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1728                 }
1729
1730                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1731                                 config->max_virtqueue_pairs);
1732                 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1733                 PMD_INIT_LOG(DEBUG,
1734                                 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1735                                 config->mac[0], config->mac[1],
1736                                 config->mac[2], config->mac[3],
1737                                 config->mac[4], config->mac[5]);
1738         } else {
1739                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1740                 hw->max_queue_pairs = 1;
1741                 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - ETHER_HDR_LEN -
1742                         VLAN_TAG_LEN - hw->vtnet_hdr_size;
1743         }
1744
1745         ret = virtio_alloc_queues(eth_dev);
1746         if (ret < 0)
1747                 return ret;
1748
1749         if (eth_dev->data->dev_conf.intr_conf.rxq) {
1750                 if (virtio_configure_intr(eth_dev) < 0) {
1751                         PMD_INIT_LOG(ERR, "failed to configure interrupt");
1752                         return -1;
1753                 }
1754         }
1755
1756         vtpci_reinit_complete(hw);
1757
1758         if (pci_dev)
1759                 PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1760                         eth_dev->data->port_id, pci_dev->id.vendor_id,
1761                         pci_dev->id.device_id);
1762
1763         return 0;
1764 }
1765
1766 /*
1767  * Remap the PCI device again (IO port map for legacy device and
1768  * memory map for modern device), so that the secondary process
1769  * could have the PCI initiated correctly.
1770  */
1771 static int
1772 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1773 {
1774         if (hw->modern) {
1775                 /*
1776                  * We don't have to re-parse the PCI config space, since
1777                  * rte_pci_map_device() makes sure the mapped address
1778                  * in secondary process would equal to the one mapped in
1779                  * the primary process: error will be returned if that
1780                  * requirement is not met.
1781                  *
1782                  * That said, we could simply reuse all cap pointers
1783                  * (such as dev_cfg, common_cfg, etc.) parsed from the
1784                  * primary process, which is stored in shared memory.
1785                  */
1786                 if (rte_pci_map_device(pci_dev)) {
1787                         PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1788                         return -1;
1789                 }
1790         } else {
1791                 if (rte_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1792                         return -1;
1793         }
1794
1795         return 0;
1796 }
1797
1798 static void
1799 virtio_set_vtpci_ops(struct virtio_hw *hw)
1800 {
1801 #ifdef RTE_VIRTIO_USER
1802         if (hw->virtio_user_dev)
1803                 VTPCI_OPS(hw) = &virtio_user_ops;
1804         else
1805 #endif
1806         if (hw->modern)
1807                 VTPCI_OPS(hw) = &modern_ops;
1808         else
1809                 VTPCI_OPS(hw) = &legacy_ops;
1810 }
1811
1812 /*
1813  * This function is based on probe() function in virtio_pci.c
1814  * It returns 0 on success.
1815  */
1816 int
1817 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1818 {
1819         struct virtio_hw *hw = eth_dev->data->dev_private;
1820         int ret;
1821
1822         RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr_mrg_rxbuf));
1823
1824         eth_dev->dev_ops = &virtio_eth_dev_ops;
1825
1826         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1827                 if (!hw->virtio_user_dev) {
1828                         ret = virtio_remap_pci(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1829                         if (ret)
1830                                 return ret;
1831                 }
1832
1833                 virtio_set_vtpci_ops(hw);
1834                 set_rxtx_funcs(eth_dev);
1835
1836                 return 0;
1837         }
1838
1839         /* Allocate memory for storing MAC addresses */
1840         eth_dev->data->mac_addrs = rte_zmalloc("virtio", VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN, 0);
1841         if (eth_dev->data->mac_addrs == NULL) {
1842                 PMD_INIT_LOG(ERR,
1843                         "Failed to allocate %d bytes needed to store MAC addresses",
1844                         VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN);
1845                 return -ENOMEM;
1846         }
1847
1848         hw->port_id = eth_dev->data->port_id;
1849         /* For virtio_user case the hw->virtio_user_dev is populated by
1850          * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1851          */
1852         if (!hw->virtio_user_dev) {
1853                 ret = vtpci_init(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1854                 if (ret)
1855                         goto out;
1856         }
1857
1858         /* reset device and negotiate default features */
1859         ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1860         if (ret < 0)
1861                 goto out;
1862
1863         return 0;
1864
1865 out:
1866         rte_free(eth_dev->data->mac_addrs);
1867         return ret;
1868 }
1869
1870 static int
1871 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1872 {
1873         PMD_INIT_FUNC_TRACE();
1874
1875         if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1876                 return 0;
1877
1878         virtio_dev_stop(eth_dev);
1879         virtio_dev_close(eth_dev);
1880
1881         eth_dev->dev_ops = NULL;
1882         eth_dev->tx_pkt_burst = NULL;
1883         eth_dev->rx_pkt_burst = NULL;
1884
1885         if (eth_dev->device)
1886                 rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev));
1887
1888         PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1889
1890         return 0;
1891 }
1892
1893 static int vdpa_check_handler(__rte_unused const char *key,
1894                 const char *value, __rte_unused void *opaque)
1895 {
1896         if (strcmp(value, "1"))
1897                 return -1;
1898
1899         return 0;
1900 }
1901
1902 static int
1903 vdpa_mode_selected(struct rte_devargs *devargs)
1904 {
1905         struct rte_kvargs *kvlist;
1906         const char *key = "vdpa";
1907         int ret = 0;
1908
1909         if (devargs == NULL)
1910                 return 0;
1911
1912         kvlist = rte_kvargs_parse(devargs->args, NULL);
1913         if (kvlist == NULL)
1914                 return 0;
1915
1916         if (!rte_kvargs_count(kvlist, key))
1917                 goto exit;
1918
1919         /* vdpa mode selected when there's a key-value pair: vdpa=1 */
1920         if (rte_kvargs_process(kvlist, key,
1921                                 vdpa_check_handler, NULL) < 0) {
1922                 goto exit;
1923         }
1924         ret = 1;
1925
1926 exit:
1927         rte_kvargs_free(kvlist);
1928         return ret;
1929 }
1930
1931 static int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1932         struct rte_pci_device *pci_dev)
1933 {
1934         if (rte_eal_iopl_init() != 0) {
1935                 PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1936                 return 1;
1937         }
1938
1939         /* virtio pmd skips probe if device needs to work in vdpa mode */
1940         if (vdpa_mode_selected(pci_dev->device.devargs))
1941                 return 1;
1942
1943         return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw),
1944                 eth_virtio_dev_init);
1945 }
1946
1947 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev)
1948 {
1949         return rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
1950 }
1951
1952 static struct rte_pci_driver rte_virtio_pmd = {
1953         .driver = {
1954                 .name = "net_virtio",
1955         },
1956         .id_table = pci_id_virtio_map,
1957         .drv_flags = 0,
1958         .probe = eth_virtio_pci_probe,
1959         .remove = eth_virtio_pci_remove,
1960 };
1961
1962 RTE_INIT(rte_virtio_pmd_init)
1963 {
1964         rte_eal_iopl_init();
1965         rte_pci_register(&rte_virtio_pmd);
1966 }
1967
1968 static bool
1969 rx_offload_enabled(struct virtio_hw *hw)
1970 {
1971         return vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM) ||
1972                 vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
1973                 vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6);
1974 }
1975
1976 static bool
1977 tx_offload_enabled(struct virtio_hw *hw)
1978 {
1979         return vtpci_with_feature(hw, VIRTIO_NET_F_CSUM) ||
1980                 vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO4) ||
1981                 vtpci_with_feature(hw, VIRTIO_NET_F_HOST_TSO6);
1982 }
1983
1984 /*
1985  * Configure virtio device
1986  * It returns 0 on success.
1987  */
1988 static int
1989 virtio_dev_configure(struct rte_eth_dev *dev)
1990 {
1991         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1992         const struct rte_eth_txmode *txmode = &dev->data->dev_conf.txmode;
1993         struct virtio_hw *hw = dev->data->dev_private;
1994         uint32_t ether_hdr_len = ETHER_HDR_LEN + VLAN_TAG_LEN +
1995                 hw->vtnet_hdr_size;
1996         uint64_t rx_offloads = rxmode->offloads;
1997         uint64_t tx_offloads = txmode->offloads;
1998         uint64_t req_features;
1999         int ret;
2000
2001         PMD_INIT_LOG(DEBUG, "configure");
2002         req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
2003
2004         if (dev->data->dev_conf.intr_conf.rxq) {
2005                 ret = virtio_init_device(dev, hw->req_guest_features);
2006                 if (ret < 0)
2007                         return ret;
2008         }
2009
2010         if (rxmode->max_rx_pkt_len > hw->max_mtu + ether_hdr_len)
2011                 req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
2012
2013         if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2014                            DEV_RX_OFFLOAD_TCP_CKSUM))
2015                 req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
2016
2017         if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO)
2018                 req_features |=
2019                         (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2020                         (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2021
2022         if (tx_offloads & (DEV_TX_OFFLOAD_UDP_CKSUM |
2023                            DEV_TX_OFFLOAD_TCP_CKSUM))
2024                 req_features |= (1ULL << VIRTIO_NET_F_CSUM);
2025
2026         if (tx_offloads & DEV_TX_OFFLOAD_TCP_TSO)
2027                 req_features |=
2028                         (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2029                         (1ULL << VIRTIO_NET_F_HOST_TSO6);
2030
2031         /* if request features changed, reinit the device */
2032         if (req_features != hw->req_guest_features) {
2033                 ret = virtio_init_device(dev, req_features);
2034                 if (ret < 0)
2035                         return ret;
2036         }
2037
2038         if ((rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2039                             DEV_RX_OFFLOAD_TCP_CKSUM)) &&
2040                 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
2041                 PMD_DRV_LOG(ERR,
2042                         "rx checksum not available on this host");
2043                 return -ENOTSUP;
2044         }
2045
2046         if ((rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) &&
2047                 (!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
2048                  !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
2049                 PMD_DRV_LOG(ERR,
2050                         "Large Receive Offload not available on this host");
2051                 return -ENOTSUP;
2052         }
2053
2054         /* start control queue */
2055         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
2056                 virtio_dev_cq_start(dev);
2057
2058         if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2059                 hw->vlan_strip = 1;
2060
2061         if ((rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
2062             && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2063                 PMD_DRV_LOG(ERR,
2064                             "vlan filtering not available on this host");
2065                 return -ENOTSUP;
2066         }
2067
2068         hw->has_tx_offload = tx_offload_enabled(hw);
2069         hw->has_rx_offload = rx_offload_enabled(hw);
2070
2071         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2072                 /* Enable vector (0) for Link State Intrerrupt */
2073                 if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
2074                                 VIRTIO_MSI_NO_VECTOR) {
2075                         PMD_DRV_LOG(ERR, "failed to set config vector");
2076                         return -EBUSY;
2077                 }
2078
2079         rte_spinlock_init(&hw->state_lock);
2080
2081         hw->use_simple_rx = 1;
2082
2083         if (vtpci_with_feature(hw, VIRTIO_F_IN_ORDER)) {
2084                 hw->use_inorder_tx = 1;
2085                 hw->use_inorder_rx = 1;
2086                 hw->use_simple_rx = 0;
2087         }
2088
2089         if (vtpci_packed_queue(hw)) {
2090                 hw->use_simple_rx = 0;
2091                 hw->use_inorder_rx = 0;
2092         }
2093
2094 #if defined RTE_ARCH_ARM64 || defined RTE_ARCH_ARM
2095         if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
2096                 hw->use_simple_rx = 0;
2097         }
2098 #endif
2099         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
2100                  hw->use_simple_rx = 0;
2101         }
2102
2103         if (rx_offloads & (DEV_RX_OFFLOAD_UDP_CKSUM |
2104                            DEV_RX_OFFLOAD_TCP_CKSUM |
2105                            DEV_RX_OFFLOAD_TCP_LRO |
2106                            DEV_RX_OFFLOAD_VLAN_STRIP))
2107                 hw->use_simple_rx = 0;
2108
2109         hw->opened = true;
2110
2111         return 0;
2112 }
2113
2114
2115 static int
2116 virtio_dev_start(struct rte_eth_dev *dev)
2117 {
2118         uint16_t nb_queues, i;
2119         struct virtnet_rx *rxvq;
2120         struct virtnet_tx *txvq __rte_unused;
2121         struct virtio_hw *hw = dev->data->dev_private;
2122         int ret;
2123
2124         /* Finish the initialization of the queues */
2125         for (i = 0; i < dev->data->nb_rx_queues; i++) {
2126                 ret = virtio_dev_rx_queue_setup_finish(dev, i);
2127                 if (ret < 0)
2128                         return ret;
2129         }
2130         for (i = 0; i < dev->data->nb_tx_queues; i++) {
2131                 ret = virtio_dev_tx_queue_setup_finish(dev, i);
2132                 if (ret < 0)
2133                         return ret;
2134         }
2135
2136         /* check if lsc interrupt feature is enabled */
2137         if (dev->data->dev_conf.intr_conf.lsc) {
2138                 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
2139                         PMD_DRV_LOG(ERR, "link status not supported by host");
2140                         return -ENOTSUP;
2141                 }
2142         }
2143
2144         /* Enable uio/vfio intr/eventfd mapping: althrough we already did that
2145          * in device configure, but it could be unmapped  when device is
2146          * stopped.
2147          */
2148         if (dev->data->dev_conf.intr_conf.lsc ||
2149             dev->data->dev_conf.intr_conf.rxq) {
2150                 virtio_intr_disable(dev);
2151
2152                 /* Setup interrupt callback  */
2153                 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
2154                         rte_intr_callback_register(dev->intr_handle,
2155                                                    virtio_interrupt_handler,
2156                                                    dev);
2157
2158                 if (virtio_intr_enable(dev) < 0) {
2159                         PMD_DRV_LOG(ERR, "interrupt enable failed");
2160                         return -EIO;
2161                 }
2162         }
2163
2164         /*Notify the backend
2165          *Otherwise the tap backend might already stop its queue due to fullness.
2166          *vhost backend will have no chance to be waked up
2167          */
2168         nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
2169         if (hw->max_queue_pairs > 1) {
2170                 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
2171                         return -EINVAL;
2172         }
2173
2174         PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
2175
2176         for (i = 0; i < dev->data->nb_rx_queues; i++) {
2177                 rxvq = dev->data->rx_queues[i];
2178                 /* Flush the old packets */
2179                 virtqueue_rxvq_flush(rxvq->vq);
2180                 virtqueue_notify(rxvq->vq);
2181         }
2182
2183         for (i = 0; i < dev->data->nb_tx_queues; i++) {
2184                 txvq = dev->data->tx_queues[i];
2185                 virtqueue_notify(txvq->vq);
2186         }
2187
2188         PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
2189
2190         for (i = 0; i < dev->data->nb_rx_queues; i++) {
2191                 rxvq = dev->data->rx_queues[i];
2192                 VIRTQUEUE_DUMP(rxvq->vq);
2193         }
2194
2195         for (i = 0; i < dev->data->nb_tx_queues; i++) {
2196                 txvq = dev->data->tx_queues[i];
2197                 VIRTQUEUE_DUMP(txvq->vq);
2198         }
2199
2200         set_rxtx_funcs(dev);
2201         hw->started = true;
2202
2203         /* Initialize Link state */
2204         virtio_dev_link_update(dev, 0);
2205
2206         return 0;
2207 }
2208
2209 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
2210 {
2211         struct virtio_hw *hw = dev->data->dev_private;
2212         uint16_t nr_vq = virtio_get_nr_vq(hw);
2213         const char *type __rte_unused;
2214         unsigned int i, mbuf_num = 0;
2215         struct virtqueue *vq;
2216         struct rte_mbuf *buf;
2217         int queue_type;
2218
2219         if (hw->vqs == NULL)
2220                 return;
2221
2222         for (i = 0; i < nr_vq; i++) {
2223                 vq = hw->vqs[i];
2224                 if (!vq)
2225                         continue;
2226
2227                 queue_type = virtio_get_queue_type(hw, i);
2228                 if (queue_type == VTNET_RQ)
2229                         type = "rxq";
2230                 else if (queue_type == VTNET_TQ)
2231                         type = "txq";
2232                 else
2233                         continue;
2234
2235                 PMD_INIT_LOG(DEBUG,
2236                         "Before freeing %s[%d] used and unused buf",
2237                         type, i);
2238                 VIRTQUEUE_DUMP(vq);
2239
2240                 while ((buf = virtqueue_detach_unused(vq)) != NULL) {
2241                         rte_pktmbuf_free(buf);
2242                         mbuf_num++;
2243                 }
2244
2245                 PMD_INIT_LOG(DEBUG,
2246                         "After freeing %s[%d] used and unused buf",
2247                         type, i);
2248                 VIRTQUEUE_DUMP(vq);
2249         }
2250
2251         PMD_INIT_LOG(DEBUG, "%d mbufs freed", mbuf_num);
2252 }
2253
2254 /*
2255  * Stop device: disable interrupt and mark link down
2256  */
2257 static void
2258 virtio_dev_stop(struct rte_eth_dev *dev)
2259 {
2260         struct virtio_hw *hw = dev->data->dev_private;
2261         struct rte_eth_link link;
2262         struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
2263
2264         PMD_INIT_LOG(DEBUG, "stop");
2265
2266         rte_spinlock_lock(&hw->state_lock);
2267         if (!hw->started)
2268                 goto out_unlock;
2269         hw->started = false;
2270
2271         if (intr_conf->lsc || intr_conf->rxq) {
2272                 virtio_intr_disable(dev);
2273
2274                 /* Reset interrupt callback  */
2275                 if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC) {
2276                         rte_intr_callback_unregister(dev->intr_handle,
2277                                                      virtio_interrupt_handler,
2278                                                      dev);
2279                 }
2280         }
2281
2282         memset(&link, 0, sizeof(link));
2283         rte_eth_linkstatus_set(dev, &link);
2284 out_unlock:
2285         rte_spinlock_unlock(&hw->state_lock);
2286 }
2287
2288 static int
2289 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2290 {
2291         struct rte_eth_link link;
2292         uint16_t status;
2293         struct virtio_hw *hw = dev->data->dev_private;
2294
2295         memset(&link, 0, sizeof(link));
2296         link.link_duplex = ETH_LINK_FULL_DUPLEX;
2297         link.link_speed  = ETH_SPEED_NUM_10G;
2298         link.link_autoneg = ETH_LINK_FIXED;
2299
2300         if (!hw->started) {
2301                 link.link_status = ETH_LINK_DOWN;
2302         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2303                 PMD_INIT_LOG(DEBUG, "Get link status from hw");
2304                 vtpci_read_dev_config(hw,
2305                                 offsetof(struct virtio_net_config, status),
2306                                 &status, sizeof(status));
2307                 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2308                         link.link_status = ETH_LINK_DOWN;
2309                         PMD_INIT_LOG(DEBUG, "Port %d is down",
2310                                      dev->data->port_id);
2311                 } else {
2312                         link.link_status = ETH_LINK_UP;
2313                         PMD_INIT_LOG(DEBUG, "Port %d is up",
2314                                      dev->data->port_id);
2315                 }
2316         } else {
2317                 link.link_status = ETH_LINK_UP;
2318         }
2319
2320         return rte_eth_linkstatus_set(dev, &link);
2321 }
2322
2323 static int
2324 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2325 {
2326         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2327         struct virtio_hw *hw = dev->data->dev_private;
2328         uint64_t offloads = rxmode->offloads;
2329
2330         if (mask & ETH_VLAN_FILTER_MASK) {
2331                 if ((offloads & DEV_RX_OFFLOAD_VLAN_FILTER) &&
2332                                 !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2333
2334                         PMD_DRV_LOG(NOTICE,
2335                                 "vlan filtering not available on this host");
2336
2337                         return -ENOTSUP;
2338                 }
2339         }
2340
2341         if (mask & ETH_VLAN_STRIP_MASK)
2342                 hw->vlan_strip = !!(offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
2343
2344         return 0;
2345 }
2346
2347 static void
2348 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2349 {
2350         uint64_t tso_mask, host_features;
2351         struct virtio_hw *hw = dev->data->dev_private;
2352
2353         dev_info->speed_capa = ETH_LINK_SPEED_10G; /* fake value */
2354
2355         dev_info->max_rx_queues =
2356                 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2357         dev_info->max_tx_queues =
2358                 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2359         dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2360         dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2361         dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2362
2363         host_features = VTPCI_OPS(hw)->get_features(hw);
2364         dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
2365         dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2366         if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2367                 dev_info->rx_offload_capa |=
2368                         DEV_RX_OFFLOAD_TCP_CKSUM |
2369                         DEV_RX_OFFLOAD_UDP_CKSUM;
2370         }
2371         if (host_features & (1ULL << VIRTIO_NET_F_CTRL_VLAN))
2372                 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_VLAN_FILTER;
2373         tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2374                 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2375         if ((host_features & tso_mask) == tso_mask)
2376                 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2377
2378         dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
2379                                     DEV_TX_OFFLOAD_VLAN_INSERT;
2380         if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2381                 dev_info->tx_offload_capa |=
2382                         DEV_TX_OFFLOAD_UDP_CKSUM |
2383                         DEV_TX_OFFLOAD_TCP_CKSUM;
2384         }
2385         tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2386                 (1ULL << VIRTIO_NET_F_HOST_TSO6);
2387         if ((host_features & tso_mask) == tso_mask)
2388                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2389 }
2390
2391 /*
2392  * It enables testpmd to collect per queue stats.
2393  */
2394 static int
2395 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2396 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2397 __rte_unused uint8_t is_rx)
2398 {
2399         return 0;
2400 }
2401
2402 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
2403 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
2404 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio-pci");
2405
2406 RTE_INIT(virtio_init_log)
2407 {
2408         virtio_logtype_init = rte_log_register("pmd.net.virtio.init");
2409         if (virtio_logtype_init >= 0)
2410                 rte_log_set_level(virtio_logtype_init, RTE_LOG_NOTICE);
2411         virtio_logtype_driver = rte_log_register("pmd.net.virtio.driver");
2412         if (virtio_logtype_driver >= 0)
2413                 rte_log_set_level(virtio_logtype_driver, RTE_LOG_NOTICE);
2414 }