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_atomic.h>
  18 #include <rte_branch_prediction.h>
  19 #include <rte_pci.h>
  20 #include <rte_bus_pci.h>
  21 #include <rte_ether.h>
  22 #include <rte_ip.h>
  23 #include <rte_arp.h>
  24 #include <rte_common.h>
  25 #include <rte_errno.h>
  26 #include <rte_cpuflags.h>
  27
  28 #include <rte_memory.h>
  29 #include <rte_eal.h>
  30 #include <rte_dev.h>
  31 #include <rte_cycles.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 void virtio_mac_addr_set(struct rte_eth_dev *dev,
  72                                 struct ether_addr *mac_addr);
  73
  74 static int virtio_intr_enable(struct rte_eth_dev *dev);
  75 static int virtio_intr_disable(struct rte_eth_dev *dev);
  76
  77 static int virtio_dev_queue_stats_mapping_set(
  78         struct rte_eth_dev *eth_dev,
  79         uint16_t queue_id,
  80         uint8_t stat_idx,
  81         uint8_t is_rx);
  82
  83 int virtio_logtype_init;
  84 int virtio_logtype_driver;
  85
  86 static void virtio_notify_peers(struct rte_eth_dev *dev);
  87 static void virtio_ack_link_announce(struct rte_eth_dev *dev);
  88
  89 /*
  90  * The set of PCI devices this driver supports
  91  */
  92 static const struct rte_pci_id pci_id_virtio_map[] = {
  93         { RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_LEGACY_DEVICEID_NET) },
  94         { RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_MODERN_DEVICEID_NET) },
  95         { .vendor_id = 0, /* sentinel */ },
  96 };
  97
  98 struct rte_virtio_xstats_name_off {
  99         char name[RTE_ETH_XSTATS_NAME_SIZE];
 100         unsigned offset;
 101 };
 102
 103 /* [rt]x_qX_ is prepended to the name string here */
 104 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
 105         {"good_packets",           offsetof(struct virtnet_rx, stats.packets)},
 106         {"good_bytes",             offsetof(struct virtnet_rx, stats.bytes)},
 107         {"errors",                 offsetof(struct virtnet_rx, stats.errors)},
 108         {"multicast_packets",      offsetof(struct virtnet_rx, stats.multicast)},
 109         {"broadcast_packets",      offsetof(struct virtnet_rx, stats.broadcast)},
 110         {"undersize_packets",      offsetof(struct virtnet_rx, stats.size_bins[0])},
 111         {"size_64_packets",        offsetof(struct virtnet_rx, stats.size_bins[1])},
 112         {"size_65_127_packets",    offsetof(struct virtnet_rx, stats.size_bins[2])},
 113         {"size_128_255_packets",   offsetof(struct virtnet_rx, stats.size_bins[3])},
 114         {"size_256_511_packets",   offsetof(struct virtnet_rx, stats.size_bins[4])},
 115         {"size_512_1023_packets",  offsetof(struct virtnet_rx, stats.size_bins[5])},
 116         {"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
 117         {"size_1519_max_packets",  offsetof(struct virtnet_rx, stats.size_bins[7])},
 118 };
 119
 120 /* [rt]x_qX_ is prepended to the name string here */
 121 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
 122         {"good_packets",           offsetof(struct virtnet_tx, stats.packets)},
 123         {"good_bytes",             offsetof(struct virtnet_tx, stats.bytes)},
 124         {"errors",                 offsetof(struct virtnet_tx, stats.errors)},
 125         {"multicast_packets",      offsetof(struct virtnet_tx, stats.multicast)},
 126         {"broadcast_packets",      offsetof(struct virtnet_tx, stats.broadcast)},
 127         {"undersize_packets",      offsetof(struct virtnet_tx, stats.size_bins[0])},
 128         {"size_64_packets",        offsetof(struct virtnet_tx, stats.size_bins[1])},
 129         {"size_65_127_packets",    offsetof(struct virtnet_tx, stats.size_bins[2])},
 130         {"size_128_255_packets",   offsetof(struct virtnet_tx, stats.size_bins[3])},
 131         {"size_256_511_packets",   offsetof(struct virtnet_tx, stats.size_bins[4])},
 132         {"size_512_1023_packets",  offsetof(struct virtnet_tx, stats.size_bins[5])},
 133         {"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
 134         {"size_1519_max_packets",  offsetof(struct virtnet_tx, stats.size_bins[7])},
 135 };
 136
 137 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
 138                             sizeof(rte_virtio_rxq_stat_strings[0]))
 139 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
 140                             sizeof(rte_virtio_txq_stat_strings[0]))
 141
 142 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
 143
 144 static int
 145 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
 146                 int *dlen, int pkt_num)
 147 {
 148         uint32_t head, i;
 149         int k, sum = 0;
 150         virtio_net_ctrl_ack status = ~0;
 151         struct virtio_pmd_ctrl *result;
 152         struct virtqueue *vq;
 153
 154         ctrl->status = status;
 155
 156         if (!cvq || !cvq->vq) {
 157                 PMD_INIT_LOG(ERR, "Control queue is not supported.");
 158                 return -1;
 159         }
 160
 161         rte_spinlock_lock(&cvq->lock);
 162         vq = cvq->vq;
 163         head = vq->vq_desc_head_idx;
 164
 165         PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
 166                 "vq->hw->cvq = %p vq = %p",
 167                 vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
 168
 169         if (vq->vq_free_cnt < pkt_num + 2 || pkt_num < 1) {
 170                 rte_spinlock_unlock(&cvq->lock);
 171                 return -1;
 172         }
 173
 174         memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
 175                 sizeof(struct virtio_pmd_ctrl));
 176
 177         /*
 178          * Format is enforced in qemu code:
 179          * One TX packet for header;
 180          * At least one TX packet per argument;
 181          * One RX packet for ACK.
 182          */
 183         vq->vq_ring.desc[head].flags = VRING_DESC_F_NEXT;
 184         vq->vq_ring.desc[head].addr = cvq->virtio_net_hdr_mem;
 185         vq->vq_ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
 186         vq->vq_free_cnt--;
 187         i = vq->vq_ring.desc[head].next;
 188
 189         for (k = 0; k < pkt_num; k++) {
 190                 vq->vq_ring.desc[i].flags = VRING_DESC_F_NEXT;
 191                 vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
 192                         + sizeof(struct virtio_net_ctrl_hdr)
 193                         + sizeof(ctrl->status) + sizeof(uint8_t)*sum;
 194                 vq->vq_ring.desc[i].len = dlen[k];
 195                 sum += dlen[k];
 196                 vq->vq_free_cnt--;
 197                 i = vq->vq_ring.desc[i].next;
 198         }
 199
 200         vq->vq_ring.desc[i].flags = VRING_DESC_F_WRITE;
 201         vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
 202                         + sizeof(struct virtio_net_ctrl_hdr);
 203         vq->vq_ring.desc[i].len = sizeof(ctrl->status);
 204         vq->vq_free_cnt--;
 205
 206         vq->vq_desc_head_idx = vq->vq_ring.desc[i].next;
 207
 208         vq_update_avail_ring(vq, head);
 209         vq_update_avail_idx(vq);
 210
 211         PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
 212
 213         virtqueue_notify(vq);
 214
 215         rte_rmb();
 216         while (VIRTQUEUE_NUSED(vq) == 0) {
 217                 rte_rmb();
 218                 usleep(100);
 219         }
 220
 221         while (VIRTQUEUE_NUSED(vq)) {
 222                 uint32_t idx, desc_idx, used_idx;
 223                 struct vring_used_elem *uep;
 224
 225                 used_idx = (uint32_t)(vq->vq_used_cons_idx
 226                                 & (vq->vq_nentries - 1));
 227                 uep = &vq->vq_ring.used->ring[used_idx];
 228                 idx = (uint32_t) uep->id;
 229                 desc_idx = idx;
 230
 231                 while (vq->vq_ring.desc[desc_idx].flags & VRING_DESC_F_NEXT) {
 232                         desc_idx = vq->vq_ring.desc[desc_idx].next;
 233                         vq->vq_free_cnt++;
 234                 }
 235
 236                 vq->vq_ring.desc[desc_idx].next = vq->vq_desc_head_idx;
 237                 vq->vq_desc_head_idx = idx;
 238
 239                 vq->vq_used_cons_idx++;
 240                 vq->vq_free_cnt++;
 241         }
 242
 243         PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
 244                         vq->vq_free_cnt, vq->vq_desc_head_idx);
 245
 246         result = cvq->virtio_net_hdr_mz->addr;
 247
 248         rte_spinlock_unlock(&cvq->lock);
 249         return result->status;
 250 }
 251
 252 static int
 253 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
 254 {
 255         struct virtio_hw *hw = dev->data->dev_private;
 256         struct virtio_pmd_ctrl ctrl;
 257         int dlen[1];
 258         int ret;
 259
 260         ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
 261         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
 262         memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
 263
 264         dlen[0] = sizeof(uint16_t);
 265
 266         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 267         if (ret) {
 268                 PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
 269                           "failed, this is too late now...");
 270                 return -EINVAL;
 271         }
 272
 273         return 0;
 274 }
 275
 276 static void
 277 virtio_dev_queue_release(void *queue __rte_unused)
 278 {
 279         /* do nothing */
 280 }
 281
 282 static int
 283 virtio_get_queue_type(struct virtio_hw *hw, uint16_t vtpci_queue_idx)
 284 {
 285         if (vtpci_queue_idx == hw->max_queue_pairs * 2)
 286                 return VTNET_CQ;
 287         else if (vtpci_queue_idx % 2 == 0)
 288                 return VTNET_RQ;
 289         else
 290                 return VTNET_TQ;
 291 }
 292
 293 static uint16_t
 294 virtio_get_nr_vq(struct virtio_hw *hw)
 295 {
 296         uint16_t nr_vq = hw->max_queue_pairs * 2;
 297
 298         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
 299                 nr_vq += 1;
 300
 301         return nr_vq;
 302 }
 303
 304 static void
 305 virtio_init_vring(struct virtqueue *vq)
 306 {
 307         int size = vq->vq_nentries;
 308         struct vring *vr = &vq->vq_ring;
 309         uint8_t *ring_mem = vq->vq_ring_virt_mem;
 310
 311         PMD_INIT_FUNC_TRACE();
 312
 313         /*
 314          * Reinitialise since virtio port might have been stopped and restarted
 315          */
 316         memset(ring_mem, 0, vq->vq_ring_size);
 317         vring_init(vr, size, ring_mem, VIRTIO_PCI_VRING_ALIGN);
 318         vq->vq_used_cons_idx = 0;
 319         vq->vq_desc_head_idx = 0;
 320         vq->vq_avail_idx = 0;
 321         vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
 322         vq->vq_free_cnt = vq->vq_nentries;
 323         memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
 324
 325         vring_desc_init(vr->desc, size);
 326
 327         /*
 328          * Disable device(host) interrupting guest
 329          */
 330         virtqueue_disable_intr(vq);
 331 }
 332
 333 static int
 334 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
 335 {
 336         char vq_name[VIRTQUEUE_MAX_NAME_SZ];
 337         char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
 338         const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
 339         unsigned int vq_size, size;
 340         struct virtio_hw *hw = dev->data->dev_private;
 341         struct virtnet_rx *rxvq = NULL;
 342         struct virtnet_tx *txvq = NULL;
 343         struct virtnet_ctl *cvq = NULL;
 344         struct virtqueue *vq;
 345         size_t sz_hdr_mz = 0;
 346         void *sw_ring = NULL;
 347         int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
 348         int ret;
 349
 350         PMD_INIT_LOG(DEBUG, "setting up queue: %u", vtpci_queue_idx);
 351
 352         /*
 353          * Read the virtqueue size from the Queue Size field
 354          * Always power of 2 and if 0 virtqueue does not exist
 355          */
 356         vq_size = VTPCI_OPS(hw)->get_queue_num(hw, vtpci_queue_idx);
 357         PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
 358         if (vq_size == 0) {
 359                 PMD_INIT_LOG(ERR, "virtqueue does not exist");
 360                 return -EINVAL;
 361         }
 362
 363         if (!rte_is_power_of_2(vq_size)) {
 364                 PMD_INIT_LOG(ERR, "virtqueue size is not powerof 2");
 365                 return -EINVAL;
 366         }
 367
 368         snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
 369                  dev->data->port_id, vtpci_queue_idx);
 370
 371         size = RTE_ALIGN_CEIL(sizeof(*vq) +
 372                                 vq_size * sizeof(struct vq_desc_extra),
 373                                 RTE_CACHE_LINE_SIZE);
 374         if (queue_type == VTNET_TQ) {
 375                 /*
 376                  * For each xmit packet, allocate a virtio_net_hdr
 377                  * and indirect ring elements
 378                  */
 379                 sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
 380         } else if (queue_type == VTNET_CQ) {
 381                 /* Allocate a page for control vq command, data and status */
 382                 sz_hdr_mz = PAGE_SIZE;
 383         }
 384
 385         vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
 386                                 SOCKET_ID_ANY);
 387         if (vq == NULL) {
 388                 PMD_INIT_LOG(ERR, "can not allocate vq");
 389                 return -ENOMEM;
 390         }
 391         hw->vqs[vtpci_queue_idx] = vq;
 392
 393         vq->hw = hw;
 394         vq->vq_queue_index = vtpci_queue_idx;
 395         vq->vq_nentries = vq_size;
 396
 397         /*
 398          * Reserve a memzone for vring elements
 399          */
 400         size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
 401         vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
 402         PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
 403                      size, vq->vq_ring_size);
 404
 405         mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
 406                                          SOCKET_ID_ANY,
 407                                          0, VIRTIO_PCI_VRING_ALIGN);
 408         if (mz == NULL) {
 409                 if (rte_errno == EEXIST)
 410                         mz = rte_memzone_lookup(vq_name);
 411                 if (mz == NULL) {
 412                         ret = -ENOMEM;
 413                         goto fail_q_alloc;
 414                 }
 415         }
 416
 417         memset(mz->addr, 0, mz->len);
 418
 419         vq->vq_ring_mem = mz->iova;
 420         vq->vq_ring_virt_mem = mz->addr;
 421         PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%" PRIx64,
 422                      (uint64_t)mz->iova);
 423         PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
 424                      (uint64_t)(uintptr_t)mz->addr);
 425
 426         virtio_init_vring(vq);
 427
 428         if (sz_hdr_mz) {
 429                 snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
 430                          dev->data->port_id, vtpci_queue_idx);
 431                 hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
 432                                                      SOCKET_ID_ANY, 0,
 433                                                      RTE_CACHE_LINE_SIZE);
 434                 if (hdr_mz == NULL) {
 435                         if (rte_errno == EEXIST)
 436                                 hdr_mz = rte_memzone_lookup(vq_hdr_name);
 437                         if (hdr_mz == NULL) {
 438                                 ret = -ENOMEM;
 439                                 goto fail_q_alloc;
 440                         }
 441                 }
 442         }
 443
 444         if (queue_type == VTNET_RQ) {
 445                 size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
 446                                sizeof(vq->sw_ring[0]);
 447
 448                 sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
 449                                 RTE_CACHE_LINE_SIZE, SOCKET_ID_ANY);
 450                 if (!sw_ring) {
 451                         PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
 452                         ret = -ENOMEM;
 453                         goto fail_q_alloc;
 454                 }
 455
 456                 vq->sw_ring = sw_ring;
 457                 rxvq = &vq->rxq;
 458                 rxvq->vq = vq;
 459                 rxvq->port_id = dev->data->port_id;
 460                 rxvq->mz = mz;
 461         } else if (queue_type == VTNET_TQ) {
 462                 txvq = &vq->txq;
 463                 txvq->vq = vq;
 464                 txvq->port_id = dev->data->port_id;
 465                 txvq->mz = mz;
 466                 txvq->virtio_net_hdr_mz = hdr_mz;
 467                 txvq->virtio_net_hdr_mem = hdr_mz->iova;
 468         } else if (queue_type == VTNET_CQ) {
 469                 cvq = &vq->cq;
 470                 cvq->vq = vq;
 471                 cvq->mz = mz;
 472                 cvq->virtio_net_hdr_mz = hdr_mz;
 473                 cvq->virtio_net_hdr_mem = hdr_mz->iova;
 474                 memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
 475
 476                 hw->cvq = cvq;
 477         }
 478
 479         /* For virtio_user case (that is when hw->dev is NULL), we use
 480          * virtual address. And we need properly set _offset_, please see
 481          * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
 482          */
 483         if (!hw->virtio_user_dev)
 484                 vq->offset = offsetof(struct rte_mbuf, buf_iova);
 485         else {
 486                 vq->vq_ring_mem = (uintptr_t)mz->addr;
 487                 vq->offset = offsetof(struct rte_mbuf, buf_addr);
 488                 if (queue_type == VTNET_TQ)
 489                         txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
 490                 else if (queue_type == VTNET_CQ)
 491                         cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
 492         }
 493
 494         if (queue_type == VTNET_TQ) {
 495                 struct virtio_tx_region *txr;
 496                 unsigned int i;
 497
 498                 txr = hdr_mz->addr;
 499                 memset(txr, 0, vq_size * sizeof(*txr));
 500                 for (i = 0; i < vq_size; i++) {
 501                         struct vring_desc *start_dp = txr[i].tx_indir;
 502
 503                         vring_desc_init(start_dp, RTE_DIM(txr[i].tx_indir));
 504
 505                         /* first indirect descriptor is always the tx header */
 506                         start_dp->addr = txvq->virtio_net_hdr_mem
 507                                 + i * sizeof(*txr)
 508                                 + offsetof(struct virtio_tx_region, tx_hdr);
 509
 510                         start_dp->len = hw->vtnet_hdr_size;
 511                         start_dp->flags = VRING_DESC_F_NEXT;
 512                 }
 513         }
 514
 515         if (VTPCI_OPS(hw)->setup_queue(hw, vq) < 0) {
 516                 PMD_INIT_LOG(ERR, "setup_queue failed");
 517                 return -EINVAL;
 518         }
 519
 520         return 0;
 521
 522 fail_q_alloc:
 523         rte_free(sw_ring);
 524         rte_memzone_free(hdr_mz);
 525         rte_memzone_free(mz);
 526         rte_free(vq);
 527
 528         return ret;
 529 }
 530
 531 static void
 532 virtio_free_queues(struct virtio_hw *hw)
 533 {
 534         uint16_t nr_vq = virtio_get_nr_vq(hw);
 535         struct virtqueue *vq;
 536         int queue_type;
 537         uint16_t i;
 538
 539         if (hw->vqs == NULL)
 540                 return;
 541
 542         for (i = 0; i < nr_vq; i++) {
 543                 vq = hw->vqs[i];
 544                 if (!vq)
 545                         continue;
 546
 547                 queue_type = virtio_get_queue_type(hw, i);
 548                 if (queue_type == VTNET_RQ) {
 549                         rte_free(vq->sw_ring);
 550                         rte_memzone_free(vq->rxq.mz);
 551                 } else if (queue_type == VTNET_TQ) {
 552                         rte_memzone_free(vq->txq.mz);
 553                         rte_memzone_free(vq->txq.virtio_net_hdr_mz);
 554                 } else {
 555                         rte_memzone_free(vq->cq.mz);
 556                         rte_memzone_free(vq->cq.virtio_net_hdr_mz);
 557                 }
 558
 559                 rte_free(vq);
 560                 hw->vqs[i] = NULL;
 561         }
 562
 563         rte_free(hw->vqs);
 564         hw->vqs = NULL;
 565 }
 566
 567 static int
 568 virtio_alloc_queues(struct rte_eth_dev *dev)
 569 {
 570         struct virtio_hw *hw = dev->data->dev_private;
 571         uint16_t nr_vq = virtio_get_nr_vq(hw);
 572         uint16_t i;
 573         int ret;
 574
 575         hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
 576         if (!hw->vqs) {
 577                 PMD_INIT_LOG(ERR, "failed to allocate vqs");
 578                 return -ENOMEM;
 579         }
 580
 581         for (i = 0; i < nr_vq; i++) {
 582                 ret = virtio_init_queue(dev, i);
 583                 if (ret < 0) {
 584                         virtio_free_queues(hw);
 585                         return ret;
 586                 }
 587         }
 588
 589         return 0;
 590 }
 591
 592 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
 593
 594 static void
 595 virtio_dev_close(struct rte_eth_dev *dev)
 596 {
 597         struct virtio_hw *hw = dev->data->dev_private;
 598         struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
 599
 600         PMD_INIT_LOG(DEBUG, "virtio_dev_close");
 601
 602         /* reset the NIC */
 603         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
 604                 VTPCI_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
 605         if (intr_conf->rxq)
 606                 virtio_queues_unbind_intr(dev);
 607
 608         if (intr_conf->lsc || intr_conf->rxq) {
 609                 virtio_intr_disable(dev);
 610                 rte_intr_efd_disable(dev->intr_handle);
 611                 rte_free(dev->intr_handle->intr_vec);
 612                 dev->intr_handle->intr_vec = NULL;
 613         }
 614
 615         vtpci_reset(hw);
 616         virtio_dev_free_mbufs(dev);
 617         virtio_free_queues(hw);
 618 }
 619
 620 static void
 621 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
 622 {
 623         struct virtio_hw *hw = dev->data->dev_private;
 624         struct virtio_pmd_ctrl ctrl;
 625         int dlen[1];
 626         int ret;
 627
 628         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 629                 PMD_INIT_LOG(INFO, "host does not support rx control");
 630                 return;
 631         }
 632
 633         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 634         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 635         ctrl.data[0] = 1;
 636         dlen[0] = 1;
 637
 638         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 639         if (ret)
 640                 PMD_INIT_LOG(ERR, "Failed to enable promisc");
 641 }
 642
 643 static void
 644 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
 645 {
 646         struct virtio_hw *hw = dev->data->dev_private;
 647         struct virtio_pmd_ctrl ctrl;
 648         int dlen[1];
 649         int ret;
 650
 651         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 652                 PMD_INIT_LOG(INFO, "host does not support rx control");
 653                 return;
 654         }
 655
 656         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 657         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 658         ctrl.data[0] = 0;
 659         dlen[0] = 1;
 660
 661         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 662         if (ret)
 663                 PMD_INIT_LOG(ERR, "Failed to disable promisc");
 664 }
 665
 666 static void
 667 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
 668 {
 669         struct virtio_hw *hw = dev->data->dev_private;
 670         struct virtio_pmd_ctrl ctrl;
 671         int dlen[1];
 672         int ret;
 673
 674         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 675                 PMD_INIT_LOG(INFO, "host does not support rx control");
 676                 return;
 677         }
 678
 679         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 680         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 681         ctrl.data[0] = 1;
 682         dlen[0] = 1;
 683
 684         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 685         if (ret)
 686                 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
 687 }
 688
 689 static void
 690 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
 691 {
 692         struct virtio_hw *hw = dev->data->dev_private;
 693         struct virtio_pmd_ctrl ctrl;
 694         int dlen[1];
 695         int ret;
 696
 697         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 698                 PMD_INIT_LOG(INFO, "host does not support rx control");
 699                 return;
 700         }
 701
 702         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 703         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 704         ctrl.data[0] = 0;
 705         dlen[0] = 1;
 706
 707         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 708         if (ret)
 709                 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
 710 }
 711
 712 #define VLAN_TAG_LEN           4    /* 802.3ac tag (not DMA'd) */
 713 static int
 714 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 715 {
 716         struct virtio_hw *hw = dev->data->dev_private;
 717         uint32_t ether_hdr_len = ETHER_HDR_LEN + VLAN_TAG_LEN +
 718                                  hw->vtnet_hdr_size;
 719         uint32_t frame_size = mtu + ether_hdr_len;
 720         uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
 721
 722         max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
 723
 724         if (mtu < ETHER_MIN_MTU || frame_size > max_frame_size) {
 725                 PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
 726                         ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
 727                 return -EINVAL;
 728         }
 729         return 0;
 730 }
 731
 732 static int
 733 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
 734 {
 735         struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
 736         struct virtqueue *vq = rxvq->vq;
 737
 738         virtqueue_enable_intr(vq);
 739         return 0;
 740 }
 741
 742 static int
 743 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
 744 {
 745         struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
 746         struct virtqueue *vq = rxvq->vq;
 747
 748         virtqueue_disable_intr(vq);
 749         return 0;
 750 }
 751
 752 /*
 753  * dev_ops for virtio, bare necessities for basic operation
 754  */
 755 static const struct eth_dev_ops virtio_eth_dev_ops = {
 756         .dev_configure           = virtio_dev_configure,
 757         .dev_start               = virtio_dev_start,
 758         .dev_stop                = virtio_dev_stop,
 759         .dev_close               = virtio_dev_close,
 760         .promiscuous_enable      = virtio_dev_promiscuous_enable,
 761         .promiscuous_disable     = virtio_dev_promiscuous_disable,
 762         .allmulticast_enable     = virtio_dev_allmulticast_enable,
 763         .allmulticast_disable    = virtio_dev_allmulticast_disable,
 764         .mtu_set                 = virtio_mtu_set,
 765         .dev_infos_get           = virtio_dev_info_get,
 766         .stats_get               = virtio_dev_stats_get,
 767         .xstats_get              = virtio_dev_xstats_get,
 768         .xstats_get_names        = virtio_dev_xstats_get_names,
 769         .stats_reset             = virtio_dev_stats_reset,
 770         .xstats_reset            = virtio_dev_stats_reset,
 771         .link_update             = virtio_dev_link_update,
 772         .vlan_offload_set        = virtio_dev_vlan_offload_set,
 773         .rx_queue_setup          = virtio_dev_rx_queue_setup,
 774         .rx_queue_intr_enable    = virtio_dev_rx_queue_intr_enable,
 775         .rx_queue_intr_disable   = virtio_dev_rx_queue_intr_disable,
 776         .rx_queue_release        = virtio_dev_queue_release,
 777         .rx_descriptor_done      = virtio_dev_rx_queue_done,
 778         .tx_queue_setup          = virtio_dev_tx_queue_setup,
 779         .tx_queue_release        = virtio_dev_queue_release,
 780         /* collect stats per queue */
 781         .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
 782         .vlan_filter_set         = virtio_vlan_filter_set,
 783         .mac_addr_add            = virtio_mac_addr_add,
 784         .mac_addr_remove         = virtio_mac_addr_remove,
 785         .mac_addr_set            = virtio_mac_addr_set,
 786 };
 787
 788 static inline int
 789 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
 790                                 struct rte_eth_link *link)
 791 {
 792         struct rte_eth_link *dst = link;
 793         struct rte_eth_link *src = &(dev->data->dev_link);
 794
 795         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
 796                         *(uint64_t *)src) == 0)
 797                 return -1;
 798
 799         return 0;
 800 }
 801
 802 /**
 803  * Atomically writes the link status information into global
 804  * structure rte_eth_dev.
 805  *
 806  * @param dev
 807  *   - Pointer to the structure rte_eth_dev to read from.
 808  *   - Pointer to the buffer to be saved with the link status.
 809  *
 810  * @return
 811  *   - On success, zero.
 812  *   - On failure, negative value.
 813  */
 814 static inline int
 815 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
 816                 struct rte_eth_link *link)
 817 {
 818         struct rte_eth_link *dst = &(dev->data->dev_link);
 819         struct rte_eth_link *src = link;
 820
 821         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
 822                                         *(uint64_t *)src) == 0)
 823                 return -1;
 824
 825         return 0;
 826 }
 827
 828 static void
 829 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 830 {
 831         unsigned i;
 832
 833         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 834                 const struct virtnet_tx *txvq = dev->data->tx_queues[i];
 835                 if (txvq == NULL)
 836                         continue;
 837
 838                 stats->opackets += txvq->stats.packets;
 839                 stats->obytes += txvq->stats.bytes;
 840                 stats->oerrors += txvq->stats.errors;
 841
 842                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 843                         stats->q_opackets[i] = txvq->stats.packets;
 844                         stats->q_obytes[i] = txvq->stats.bytes;
 845                 }
 846         }
 847
 848         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 849                 const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
 850                 if (rxvq == NULL)
 851                         continue;
 852
 853                 stats->ipackets += rxvq->stats.packets;
 854                 stats->ibytes += rxvq->stats.bytes;
 855                 stats->ierrors += rxvq->stats.errors;
 856
 857                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 858                         stats->q_ipackets[i] = rxvq->stats.packets;
 859                         stats->q_ibytes[i] = rxvq->stats.bytes;
 860                 }
 861         }
 862
 863         stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
 864 }
 865
 866 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
 867                                        struct rte_eth_xstat_name *xstats_names,
 868                                        __rte_unused unsigned limit)
 869 {
 870         unsigned i;
 871         unsigned count = 0;
 872         unsigned t;
 873
 874         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
 875                 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
 876
 877         if (xstats_names != NULL) {
 878                 /* Note: limit checked in rte_eth_xstats_names() */
 879
 880                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
 881                         struct virtnet_rx *rxvq = dev->data->rx_queues[i];
 882                         if (rxvq == NULL)
 883                                 continue;
 884                         for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
 885                                 snprintf(xstats_names[count].name,
 886                                         sizeof(xstats_names[count].name),
 887                                         "rx_q%u_%s", i,
 888                                         rte_virtio_rxq_stat_strings[t].name);
 889                                 count++;
 890                         }
 891                 }
 892
 893                 for (i = 0; i < dev->data->nb_tx_queues; i++) {
 894                         struct virtnet_tx *txvq = dev->data->tx_queues[i];
 895                         if (txvq == NULL)
 896                                 continue;
 897                         for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
 898                                 snprintf(xstats_names[count].name,
 899                                         sizeof(xstats_names[count].name),
 900                                         "tx_q%u_%s", i,
 901                                         rte_virtio_txq_stat_strings[t].name);
 902                                 count++;
 903                         }
 904                 }
 905                 return count;
 906         }
 907         return nstats;
 908 }
 909
 910 static int
 911 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 912                       unsigned n)
 913 {
 914         unsigned i;
 915         unsigned count = 0;
 916
 917         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
 918                 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
 919
 920         if (n < nstats)
 921                 return nstats;
 922
 923         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 924                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
 925
 926                 if (rxvq == NULL)
 927                         continue;
 928
 929                 unsigned t;
 930
 931                 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
 932                         xstats[count].value = *(uint64_t *)(((char *)rxvq) +
 933                                 rte_virtio_rxq_stat_strings[t].offset);
 934                         xstats[count].id = count;
 935                         count++;
 936                 }
 937         }
 938
 939         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 940                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
 941
 942                 if (txvq == NULL)
 943                         continue;
 944
 945                 unsigned t;
 946
 947                 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
 948                         xstats[count].value = *(uint64_t *)(((char *)txvq) +
 949                                 rte_virtio_txq_stat_strings[t].offset);
 950                         xstats[count].id = count;
 951                         count++;
 952                 }
 953         }
 954
 955         return count;
 956 }
 957
 958 static int
 959 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 960 {
 961         virtio_update_stats(dev, stats);
 962
 963         return 0;
 964 }
 965
 966 static void
 967 virtio_dev_stats_reset(struct rte_eth_dev *dev)
 968 {
 969         unsigned int i;
 970
 971         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 972                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
 973                 if (txvq == NULL)
 974                         continue;
 975
 976                 txvq->stats.packets = 0;
 977                 txvq->stats.bytes = 0;
 978                 txvq->stats.errors = 0;
 979                 txvq->stats.multicast = 0;
 980                 txvq->stats.broadcast = 0;
 981                 memset(txvq->stats.size_bins, 0,
 982                        sizeof(txvq->stats.size_bins[0]) * 8);
 983         }
 984
 985         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 986                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
 987                 if (rxvq == NULL)
 988                         continue;
 989
 990                 rxvq->stats.packets = 0;
 991                 rxvq->stats.bytes = 0;
 992                 rxvq->stats.errors = 0;
 993                 rxvq->stats.multicast = 0;
 994                 rxvq->stats.broadcast = 0;
 995                 memset(rxvq->stats.size_bins, 0,
 996                        sizeof(rxvq->stats.size_bins[0]) * 8);
 997         }
 998 }
 999
1000 static void
1001 virtio_set_hwaddr(struct virtio_hw *hw)
1002 {
1003         vtpci_write_dev_config(hw,
1004                         offsetof(struct virtio_net_config, mac),
1005                         &hw->mac_addr, ETHER_ADDR_LEN);
1006 }
1007
1008 static void
1009 virtio_get_hwaddr(struct virtio_hw *hw)
1010 {
1011         if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1012                 vtpci_read_dev_config(hw,
1013                         offsetof(struct virtio_net_config, mac),
1014                         &hw->mac_addr, ETHER_ADDR_LEN);
1015         } else {
1016                 eth_random_addr(&hw->mac_addr[0]);
1017                 virtio_set_hwaddr(hw);
1018         }
1019 }
1020
1021 static int
1022 virtio_mac_table_set(struct virtio_hw *hw,
1023                      const struct virtio_net_ctrl_mac *uc,
1024                      const struct virtio_net_ctrl_mac *mc)
1025 {
1026         struct virtio_pmd_ctrl ctrl;
1027         int err, len[2];
1028
1029         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1030                 PMD_DRV_LOG(INFO, "host does not support mac table");
1031                 return -1;
1032         }
1033
1034         ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1035         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1036
1037         len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
1038         memcpy(ctrl.data, uc, len[0]);
1039
1040         len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
1041         memcpy(ctrl.data + len[0], mc, len[1]);
1042
1043         err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1044         if (err != 0)
1045                 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1046         return err;
1047 }
1048
1049 static int
1050 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
1051                     uint32_t index, uint32_t vmdq __rte_unused)
1052 {
1053         struct virtio_hw *hw = dev->data->dev_private;
1054         const struct ether_addr *addrs = dev->data->mac_addrs;
1055         unsigned int i;
1056         struct virtio_net_ctrl_mac *uc, *mc;
1057
1058         if (index >= VIRTIO_MAX_MAC_ADDRS) {
1059                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1060                 return -EINVAL;
1061         }
1062
1063         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1064         uc->entries = 0;
1065         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1066         mc->entries = 0;
1067
1068         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1069                 const struct ether_addr *addr
1070                         = (i == index) ? mac_addr : addrs + i;
1071                 struct virtio_net_ctrl_mac *tbl
1072                         = is_multicast_ether_addr(addr) ? mc : uc;
1073
1074                 memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
1075         }
1076
1077         return virtio_mac_table_set(hw, uc, mc);
1078 }
1079
1080 static void
1081 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1082 {
1083         struct virtio_hw *hw = dev->data->dev_private;
1084         struct ether_addr *addrs = dev->data->mac_addrs;
1085         struct virtio_net_ctrl_mac *uc, *mc;
1086         unsigned int i;
1087
1088         if (index >= VIRTIO_MAX_MAC_ADDRS) {
1089                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1090                 return;
1091         }
1092
1093         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1094         uc->entries = 0;
1095         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1096         mc->entries = 0;
1097
1098         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1099                 struct virtio_net_ctrl_mac *tbl;
1100
1101                 if (i == index || is_zero_ether_addr(addrs + i))
1102                         continue;
1103
1104                 tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
1105                 memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
1106         }
1107
1108         virtio_mac_table_set(hw, uc, mc);
1109 }
1110
1111 static void
1112 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1113 {
1114         struct virtio_hw *hw = dev->data->dev_private;
1115
1116         memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
1117
1118         /* Use atomic update if available */
1119         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1120                 struct virtio_pmd_ctrl ctrl;
1121                 int len = ETHER_ADDR_LEN;
1122
1123                 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1124                 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1125
1126                 memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
1127                 virtio_send_command(hw->cvq, &ctrl, &len, 1);
1128         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1129                 virtio_set_hwaddr(hw);
1130 }
1131
1132 static int
1133 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1134 {
1135         struct virtio_hw *hw = dev->data->dev_private;
1136         struct virtio_pmd_ctrl ctrl;
1137         int len;
1138
1139         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1140                 return -ENOTSUP;
1141
1142         ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1143         ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1144         memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1145         len = sizeof(vlan_id);
1146
1147         return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1148 }
1149
1150 static int
1151 virtio_intr_enable(struct rte_eth_dev *dev)
1152 {
1153         struct virtio_hw *hw = dev->data->dev_private;
1154
1155         if (rte_intr_enable(dev->intr_handle) < 0)
1156                 return -1;
1157
1158         if (!hw->virtio_user_dev)
1159                 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1160
1161         return 0;
1162 }
1163
1164 static int
1165 virtio_intr_disable(struct rte_eth_dev *dev)
1166 {
1167         struct virtio_hw *hw = dev->data->dev_private;
1168
1169         if (rte_intr_disable(dev->intr_handle) < 0)
1170                 return -1;
1171
1172         if (!hw->virtio_user_dev)
1173                 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1174
1175         return 0;
1176 }
1177
1178 static int
1179 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1180 {
1181         uint64_t host_features;
1182
1183         /* Prepare guest_features: feature that driver wants to support */
1184         PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1185                 req_features);
1186
1187         /* Read device(host) feature bits */
1188         host_features = VTPCI_OPS(hw)->get_features(hw);
1189         PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1190                 host_features);
1191
1192         /* If supported, ensure MTU value is valid before acknowledging it. */
1193         if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1194                 struct virtio_net_config config;
1195
1196                 vtpci_read_dev_config(hw,
1197                         offsetof(struct virtio_net_config, mtu),
1198                         &config.mtu, sizeof(config.mtu));
1199
1200                 if (config.mtu < ETHER_MIN_MTU)
1201                         req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1202         }
1203
1204         /*
1205          * Negotiate features: Subset of device feature bits are written back
1206          * guest feature bits.
1207          */
1208         hw->guest_features = req_features;
1209         hw->guest_features = vtpci_negotiate_features(hw, host_features);
1210         PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1211                 hw->guest_features);
1212
1213         if (hw->modern) {
1214                 if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1215                         PMD_INIT_LOG(ERR,
1216                                 "VIRTIO_F_VERSION_1 features is not enabled.");
1217                         return -1;
1218                 }
1219                 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1220                 if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1221                         PMD_INIT_LOG(ERR,
1222                                 "failed to set FEATURES_OK status!");
1223                         return -1;
1224                 }
1225         }
1226
1227         hw->req_guest_features = req_features;
1228
1229         return 0;
1230 }
1231
1232 int
1233 virtio_dev_pause(struct rte_eth_dev *dev)
1234 {
1235         struct virtio_hw *hw = dev->data->dev_private;
1236
1237         rte_spinlock_lock(&hw->state_lock);
1238
1239         if (hw->started == 0) {
1240                 /* Device is just stopped. */
1241                 rte_spinlock_unlock(&hw->state_lock);
1242                 return -1;
1243         }
1244         hw->started = 0;
1245         /*
1246          * Prevent the worker threads from touching queues to avoid contention,
1247          * 1 ms should be enough for the ongoing Tx function to finish.
1248          */
1249         rte_delay_ms(1);
1250         return 0;
1251 }
1252
1253 /*
1254  * Recover hw state to let the worker threads continue.
1255  */
1256 void
1257 virtio_dev_resume(struct rte_eth_dev *dev)
1258 {
1259         struct virtio_hw *hw = dev->data->dev_private;
1260
1261         hw->started = 1;
1262         rte_spinlock_unlock(&hw->state_lock);
1263 }
1264
1265 /*
1266  * Should be called only after device is paused.
1267  */
1268 int
1269 virtio_inject_pkts(struct rte_eth_dev *dev, struct rte_mbuf **tx_pkts,
1270                 int nb_pkts)
1271 {
1272         struct virtio_hw *hw = dev->data->dev_private;
1273         struct virtnet_tx *txvq = dev->data->tx_queues[0];
1274         int ret;
1275
1276         hw->inject_pkts = tx_pkts;
1277         ret = dev->tx_pkt_burst(txvq, tx_pkts, nb_pkts);
1278         hw->inject_pkts = NULL;
1279
1280         return ret;
1281 }
1282
1283 static void
1284 virtio_notify_peers(struct rte_eth_dev *dev)
1285 {
1286         struct virtio_hw *hw = dev->data->dev_private;
1287         struct virtnet_rx *rxvq = dev->data->rx_queues[0];
1288         struct rte_mbuf *rarp_mbuf;
1289
1290         rarp_mbuf = rte_net_make_rarp_packet(rxvq->mpool,
1291                         (struct ether_addr *)hw->mac_addr);
1292         if (rarp_mbuf == NULL) {
1293                 PMD_DRV_LOG(ERR, "failed to make RARP packet.");
1294                 return;
1295         }
1296
1297         /* If virtio port just stopped, no need to send RARP */
1298         if (virtio_dev_pause(dev) < 0) {
1299                 rte_pktmbuf_free(rarp_mbuf);
1300                 return;
1301         }
1302
1303         virtio_inject_pkts(dev, &rarp_mbuf, 1);
1304         virtio_dev_resume(dev);
1305 }
1306
1307 static void
1308 virtio_ack_link_announce(struct rte_eth_dev *dev)
1309 {
1310         struct virtio_hw *hw = dev->data->dev_private;
1311         struct virtio_pmd_ctrl ctrl;
1312
1313         ctrl.hdr.class = VIRTIO_NET_CTRL_ANNOUNCE;
1314         ctrl.hdr.cmd = VIRTIO_NET_CTRL_ANNOUNCE_ACK;
1315
1316         virtio_send_command(hw->cvq, &ctrl, NULL, 0);
1317 }
1318
1319 /*
1320  * Process virtio config changed interrupt. Call the callback
1321  * if link state changed, generate gratuitous RARP packet if
1322  * the status indicates an ANNOUNCE.
1323  */
1324 void
1325 virtio_interrupt_handler(void *param)
1326 {
1327         struct rte_eth_dev *dev = param;
1328         struct virtio_hw *hw = dev->data->dev_private;
1329         uint8_t isr;
1330
1331         /* Read interrupt status which clears interrupt */
1332         isr = vtpci_isr(hw);
1333         PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1334
1335         if (virtio_intr_enable(dev) < 0)
1336                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1337
1338         if (isr & VIRTIO_PCI_ISR_CONFIG) {
1339                 if (virtio_dev_link_update(dev, 0) == 0)
1340                         _rte_eth_dev_callback_process(dev,
1341                                                       RTE_ETH_EVENT_INTR_LSC,
1342                                                       NULL);
1343         }
1344
1345         if (isr & VIRTIO_NET_S_ANNOUNCE) {
1346                 virtio_notify_peers(dev);
1347                 virtio_ack_link_announce(dev);
1348         }
1349 }
1350
1351 /* set rx and tx handlers according to what is supported */
1352 static void
1353 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1354 {
1355         struct virtio_hw *hw = eth_dev->data->dev_private;
1356
1357         if (hw->use_simple_rx) {
1358                 PMD_INIT_LOG(INFO, "virtio: using simple Rx path on port %u",
1359                         eth_dev->data->port_id);
1360                 eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1361         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1362                 PMD_INIT_LOG(INFO,
1363                         "virtio: using mergeable buffer Rx path on port %u",
1364                         eth_dev->data->port_id);
1365                 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1366         } else {
1367                 PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1368                         eth_dev->data->port_id);
1369                 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1370         }
1371
1372         if (hw->use_simple_tx) {
1373                 PMD_INIT_LOG(INFO, "virtio: using simple Tx path on port %u",
1374                         eth_dev->data->port_id);
1375                 eth_dev->tx_pkt_burst = virtio_xmit_pkts_simple;
1376         } else {
1377                 PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1378                         eth_dev->data->port_id);
1379                 eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1380         }
1381 }
1382
1383 /* Only support 1:1 queue/interrupt mapping so far.
1384  * TODO: support n:1 queue/interrupt mapping when there are limited number of
1385  * interrupt vectors (<N+1).
1386  */
1387 static int
1388 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1389 {
1390         uint32_t i;
1391         struct virtio_hw *hw = dev->data->dev_private;
1392
1393         PMD_INIT_LOG(INFO, "queue/interrupt binding");
1394         for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1395                 dev->intr_handle->intr_vec[i] = i + 1;
1396                 if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1397                                                  VIRTIO_MSI_NO_VECTOR) {
1398                         PMD_DRV_LOG(ERR, "failed to set queue vector");
1399                         return -EBUSY;
1400                 }
1401         }
1402
1403         return 0;
1404 }
1405
1406 static void
1407 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1408 {
1409         uint32_t i;
1410         struct virtio_hw *hw = dev->data->dev_private;
1411
1412         PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1413         for (i = 0; i < dev->data->nb_rx_queues; ++i)
1414                 VTPCI_OPS(hw)->set_queue_irq(hw,
1415                                              hw->vqs[i * VTNET_CQ],
1416                                              VIRTIO_MSI_NO_VECTOR);
1417 }
1418
1419 static int
1420 virtio_configure_intr(struct rte_eth_dev *dev)
1421 {
1422         struct virtio_hw *hw = dev->data->dev_private;
1423
1424         if (!rte_intr_cap_multiple(dev->intr_handle)) {
1425                 PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1426                 return -ENOTSUP;
1427         }
1428
1429         if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1430                 PMD_INIT_LOG(ERR, "Fail to create eventfd");
1431                 return -1;
1432         }
1433
1434         if (!dev->intr_handle->intr_vec) {
1435                 dev->intr_handle->intr_vec =
1436                         rte_zmalloc("intr_vec",
1437                                     hw->max_queue_pairs * sizeof(int), 0);
1438                 if (!dev->intr_handle->intr_vec) {
1439                         PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1440                                      hw->max_queue_pairs);
1441                         return -ENOMEM;
1442                 }
1443         }
1444
1445         /* Re-register callback to update max_intr */
1446         rte_intr_callback_unregister(dev->intr_handle,
1447                                      virtio_interrupt_handler,
1448                                      dev);
1449         rte_intr_callback_register(dev->intr_handle,
1450                                    virtio_interrupt_handler,
1451                                    dev);
1452
1453         /* DO NOT try to remove this! This function will enable msix, or QEMU
1454          * will encounter SIGSEGV when DRIVER_OK is sent.
1455          * And for legacy devices, this should be done before queue/vec binding
1456          * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1457          * (22) will be ignored.
1458          */
1459         if (virtio_intr_enable(dev) < 0) {
1460                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1461                 return -1;
1462         }
1463
1464         if (virtio_queues_bind_intr(dev) < 0) {
1465                 PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1466                 return -1;
1467         }
1468
1469         return 0;
1470 }
1471
1472 /* reset device and renegotiate features if needed */
1473 static int
1474 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1475 {
1476         struct virtio_hw *hw = eth_dev->data->dev_private;
1477         struct virtio_net_config *config;
1478         struct virtio_net_config local_config;
1479         struct rte_pci_device *pci_dev = NULL;
1480         int ret;
1481
1482         /* Reset the device although not necessary at startup */
1483         vtpci_reset(hw);
1484
1485         /* Tell the host we've noticed this device. */
1486         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1487
1488         /* Tell the host we've known how to drive the device. */
1489         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1490         if (virtio_negotiate_features(hw, req_features) < 0)
1491                 return -1;
1492
1493         if (!hw->virtio_user_dev) {
1494                 pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1495                 rte_eth_copy_pci_info(eth_dev, pci_dev);
1496         }
1497
1498         /* If host does not support both status and MSI-X then disable LSC */
1499         if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) &&
1500             hw->use_msix != VIRTIO_MSIX_NONE)
1501                 eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1502         else
1503                 eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1504
1505         /* Setting up rx_header size for the device */
1506         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1507             vtpci_with_feature(hw, VIRTIO_F_VERSION_1))
1508                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1509         else
1510                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1511
1512         /* Copy the permanent MAC address to: virtio_hw */
1513         virtio_get_hwaddr(hw);
1514         ether_addr_copy((struct ether_addr *) hw->mac_addr,
1515                         &eth_dev->data->mac_addrs[0]);
1516         PMD_INIT_LOG(DEBUG,
1517                      "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1518                      hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1519                      hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1520
1521         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1522                 config = &local_config;
1523
1524                 vtpci_read_dev_config(hw,
1525                         offsetof(struct virtio_net_config, mac),
1526                         &config->mac, sizeof(config->mac));
1527
1528                 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1529                         vtpci_read_dev_config(hw,
1530                                 offsetof(struct virtio_net_config, status),
1531                                 &config->status, sizeof(config->status));
1532                 } else {
1533                         PMD_INIT_LOG(DEBUG,
1534                                      "VIRTIO_NET_F_STATUS is not supported");
1535                         config->status = 0;
1536                 }
1537
1538                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1539                         vtpci_read_dev_config(hw,
1540                                 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1541                                 &config->max_virtqueue_pairs,
1542                                 sizeof(config->max_virtqueue_pairs));
1543                 } else {
1544                         PMD_INIT_LOG(DEBUG,
1545                                      "VIRTIO_NET_F_MQ is not supported");
1546                         config->max_virtqueue_pairs = 1;
1547                 }
1548
1549                 hw->max_queue_pairs = config->max_virtqueue_pairs;
1550
1551                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1552                         vtpci_read_dev_config(hw,
1553                                 offsetof(struct virtio_net_config, mtu),
1554                                 &config->mtu,
1555                                 sizeof(config->mtu));
1556
1557                         /*
1558                          * MTU value has already been checked at negotiation
1559                          * time, but check again in case it has changed since
1560                          * then, which should not happen.
1561                          */
1562                         if (config->mtu < ETHER_MIN_MTU) {
1563                                 PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1564                                                 config->mtu);
1565                                 return -1;
1566                         }
1567
1568                         hw->max_mtu = config->mtu;
1569                         /* Set initial MTU to maximum one supported by vhost */
1570                         eth_dev->data->mtu = config->mtu;
1571
1572                 } else {
1573                         hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - ETHER_HDR_LEN -
1574                                 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1575                 }
1576
1577                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1578                                 config->max_virtqueue_pairs);
1579                 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1580                 PMD_INIT_LOG(DEBUG,
1581                                 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1582                                 config->mac[0], config->mac[1],
1583                                 config->mac[2], config->mac[3],
1584                                 config->mac[4], config->mac[5]);
1585         } else {
1586                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1587                 hw->max_queue_pairs = 1;
1588                 hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - ETHER_HDR_LEN -
1589                         VLAN_TAG_LEN - hw->vtnet_hdr_size;
1590         }
1591
1592         ret = virtio_alloc_queues(eth_dev);
1593         if (ret < 0)
1594                 return ret;
1595
1596         if (eth_dev->data->dev_conf.intr_conf.rxq) {
1597                 if (virtio_configure_intr(eth_dev) < 0) {
1598                         PMD_INIT_LOG(ERR, "failed to configure interrupt");
1599                         return -1;
1600                 }
1601         }
1602
1603         vtpci_reinit_complete(hw);
1604
1605         if (pci_dev)
1606                 PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1607                         eth_dev->data->port_id, pci_dev->id.vendor_id,
1608                         pci_dev->id.device_id);
1609
1610         return 0;
1611 }
1612
1613 /*
1614  * Remap the PCI device again (IO port map for legacy device and
1615  * memory map for modern device), so that the secondary process
1616  * could have the PCI initiated correctly.
1617  */
1618 static int
1619 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1620 {
1621         if (hw->modern) {
1622                 /*
1623                  * We don't have to re-parse the PCI config space, since
1624                  * rte_pci_map_device() makes sure the mapped address
1625                  * in secondary process would equal to the one mapped in
1626                  * the primary process: error will be returned if that
1627                  * requirement is not met.
1628                  *
1629                  * That said, we could simply reuse all cap pointers
1630                  * (such as dev_cfg, common_cfg, etc.) parsed from the
1631                  * primary process, which is stored in shared memory.
1632                  */
1633                 if (rte_pci_map_device(pci_dev)) {
1634                         PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1635                         return -1;
1636                 }
1637         } else {
1638                 if (rte_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1639                         return -1;
1640         }
1641
1642         return 0;
1643 }
1644
1645 static void
1646 virtio_set_vtpci_ops(struct virtio_hw *hw)
1647 {
1648 #ifdef RTE_VIRTIO_USER
1649         if (hw->virtio_user_dev)
1650                 VTPCI_OPS(hw) = &virtio_user_ops;
1651         else
1652 #endif
1653         if (hw->modern)
1654                 VTPCI_OPS(hw) = &modern_ops;
1655         else
1656                 VTPCI_OPS(hw) = &legacy_ops;
1657 }
1658
1659 /*
1660  * This function is based on probe() function in virtio_pci.c
1661  * It returns 0 on success.
1662  */
1663 int
1664 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1665 {
1666         struct virtio_hw *hw = eth_dev->data->dev_private;
1667         int ret;
1668
1669         RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr_mrg_rxbuf));
1670
1671         eth_dev->dev_ops = &virtio_eth_dev_ops;
1672
1673         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1674                 if (!hw->virtio_user_dev) {
1675                         ret = virtio_remap_pci(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1676                         if (ret)
1677                                 return ret;
1678                 }
1679
1680                 virtio_set_vtpci_ops(hw);
1681                 set_rxtx_funcs(eth_dev);
1682
1683                 return 0;
1684         }
1685
1686         /* Allocate memory for storing MAC addresses */
1687         eth_dev->data->mac_addrs = rte_zmalloc("virtio", VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN, 0);
1688         if (eth_dev->data->mac_addrs == NULL) {
1689                 PMD_INIT_LOG(ERR,
1690                         "Failed to allocate %d bytes needed to store MAC addresses",
1691                         VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN);
1692                 return -ENOMEM;
1693         }
1694
1695         hw->port_id = eth_dev->data->port_id;
1696         /* For virtio_user case the hw->virtio_user_dev is populated by
1697          * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1698          */
1699         if (!hw->virtio_user_dev) {
1700                 ret = vtpci_init(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1701                 if (ret)
1702                         goto out;
1703         }
1704
1705         /* reset device and negotiate default features */
1706         ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1707         if (ret < 0)
1708                 goto out;
1709
1710         /* Setup interrupt callback  */
1711         if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1712                 rte_intr_callback_register(eth_dev->intr_handle,
1713                         virtio_interrupt_handler, eth_dev);
1714
1715         return 0;
1716
1717 out:
1718         rte_free(eth_dev->data->mac_addrs);
1719         return ret;
1720 }
1721
1722 static int
1723 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1724 {
1725         PMD_INIT_FUNC_TRACE();
1726
1727         if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1728                 return -EPERM;
1729
1730         virtio_dev_stop(eth_dev);
1731         virtio_dev_close(eth_dev);
1732
1733         eth_dev->dev_ops = NULL;
1734         eth_dev->tx_pkt_burst = NULL;
1735         eth_dev->rx_pkt_burst = NULL;
1736
1737         rte_free(eth_dev->data->mac_addrs);
1738         eth_dev->data->mac_addrs = NULL;
1739
1740         /* reset interrupt callback  */
1741         if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1742                 rte_intr_callback_unregister(eth_dev->intr_handle,
1743                                                 virtio_interrupt_handler,
1744                                                 eth_dev);
1745         if (eth_dev->device)
1746                 rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev));
1747
1748         PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1749
1750         return 0;
1751 }
1752
1753 static int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1754         struct rte_pci_device *pci_dev)
1755 {
1756         return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw),
1757                 eth_virtio_dev_init);
1758 }
1759
1760 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev)
1761 {
1762         return rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
1763 }
1764
1765 static struct rte_pci_driver rte_virtio_pmd = {
1766         .driver = {
1767                 .name = "net_virtio",
1768         },
1769         .id_table = pci_id_virtio_map,
1770         .drv_flags = 0,
1771         .probe = eth_virtio_pci_probe,
1772         .remove = eth_virtio_pci_remove,
1773 };
1774
1775 RTE_INIT(rte_virtio_pmd_init);
1776 static void
1777 rte_virtio_pmd_init(void)
1778 {
1779         if (rte_eal_iopl_init() != 0) {
1780                 PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1781                 return;
1782         }
1783
1784         rte_pci_register(&rte_virtio_pmd);
1785 }
1786
1787 /*
1788  * Configure virtio device
1789  * It returns 0 on success.
1790  */
1791 static int
1792 virtio_dev_configure(struct rte_eth_dev *dev)
1793 {
1794         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1795         struct virtio_hw *hw = dev->data->dev_private;
1796         uint64_t req_features;
1797         int ret;
1798
1799         PMD_INIT_LOG(DEBUG, "configure");
1800         req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
1801
1802         if (dev->data->dev_conf.intr_conf.rxq) {
1803                 ret = virtio_init_device(dev, hw->req_guest_features);
1804                 if (ret < 0)
1805                         return ret;
1806         }
1807
1808         /* The name hw_ip_checksum is a bit confusing since it can be
1809          * set by the application to request L3 and/or L4 checksums. In
1810          * case of virtio, only L4 checksum is supported.
1811          */
1812         if (rxmode->hw_ip_checksum)
1813                 req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
1814
1815         if (rxmode->enable_lro)
1816                 req_features |=
1817                         (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
1818                         (1ULL << VIRTIO_NET_F_GUEST_TSO6);
1819
1820         /* if request features changed, reinit the device */
1821         if (req_features != hw->req_guest_features) {
1822                 ret = virtio_init_device(dev, req_features);
1823                 if (ret < 0)
1824                         return ret;
1825         }
1826
1827         if (rxmode->hw_ip_checksum &&
1828                 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
1829                 PMD_DRV_LOG(ERR,
1830                         "rx checksum not available on this host");
1831                 return -ENOTSUP;
1832         }
1833
1834         if (rxmode->enable_lro &&
1835                 (!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
1836                  !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO6))) {
1837                 PMD_DRV_LOG(ERR,
1838                         "Large Receive Offload not available on this host");
1839                 return -ENOTSUP;
1840         }
1841
1842         /* start control queue */
1843         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
1844                 virtio_dev_cq_start(dev);
1845
1846         hw->vlan_strip = rxmode->hw_vlan_strip;
1847
1848         if (rxmode->hw_vlan_filter
1849             && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1850                 PMD_DRV_LOG(ERR,
1851                             "vlan filtering not available on this host");
1852                 return -ENOTSUP;
1853         }
1854
1855         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1856                 /* Enable vector (0) for Link State Intrerrupt */
1857                 if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
1858                                 VIRTIO_MSI_NO_VECTOR) {
1859                         PMD_DRV_LOG(ERR, "failed to set config vector");
1860                         return -EBUSY;
1861                 }
1862
1863         rte_spinlock_init(&hw->state_lock);
1864
1865         hw->use_simple_rx = 1;
1866         hw->use_simple_tx = 1;
1867
1868 #if defined RTE_ARCH_ARM64 || defined RTE_ARCH_ARM
1869         if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
1870                 hw->use_simple_rx = 0;
1871                 hw->use_simple_tx = 0;
1872         }
1873 #endif
1874         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1875                 hw->use_simple_rx = 0;
1876                 hw->use_simple_tx = 0;
1877         }
1878
1879         if (rxmode->hw_ip_checksum)
1880                 hw->use_simple_rx = 0;
1881
1882         return 0;
1883 }
1884
1885
1886 static int
1887 virtio_dev_start(struct rte_eth_dev *dev)
1888 {
1889         uint16_t nb_queues, i;
1890         struct virtnet_rx *rxvq;
1891         struct virtnet_tx *txvq __rte_unused;
1892         struct virtio_hw *hw = dev->data->dev_private;
1893         int ret;
1894
1895         /* Finish the initialization of the queues */
1896         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1897                 ret = virtio_dev_rx_queue_setup_finish(dev, i);
1898                 if (ret < 0)
1899                         return ret;
1900         }
1901         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1902                 ret = virtio_dev_tx_queue_setup_finish(dev, i);
1903                 if (ret < 0)
1904                         return ret;
1905         }
1906
1907         /* check if lsc interrupt feature is enabled */
1908         if (dev->data->dev_conf.intr_conf.lsc) {
1909                 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
1910                         PMD_DRV_LOG(ERR, "link status not supported by host");
1911                         return -ENOTSUP;
1912                 }
1913         }
1914
1915         /* Enable uio/vfio intr/eventfd mapping: althrough we already did that
1916          * in device configure, but it could be unmapped  when device is
1917          * stopped.
1918          */
1919         if (dev->data->dev_conf.intr_conf.lsc ||
1920             dev->data->dev_conf.intr_conf.rxq) {
1921                 virtio_intr_disable(dev);
1922
1923                 if (virtio_intr_enable(dev) < 0) {
1924                         PMD_DRV_LOG(ERR, "interrupt enable failed");
1925                         return -EIO;
1926                 }
1927         }
1928
1929         /*Notify the backend
1930          *Otherwise the tap backend might already stop its queue due to fullness.
1931          *vhost backend will have no chance to be waked up
1932          */
1933         nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
1934         if (hw->max_queue_pairs > 1) {
1935                 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1936                         return -EINVAL;
1937         }
1938
1939         PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1940
1941         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1942                 rxvq = dev->data->rx_queues[i];
1943                 /* Flush the old packets */
1944                 virtqueue_rxvq_flush(rxvq->vq);
1945                 virtqueue_notify(rxvq->vq);
1946         }
1947
1948         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1949                 txvq = dev->data->tx_queues[i];
1950                 virtqueue_notify(txvq->vq);
1951         }
1952
1953         PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1954
1955         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1956                 rxvq = dev->data->rx_queues[i];
1957                 VIRTQUEUE_DUMP(rxvq->vq);
1958         }
1959
1960         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1961                 txvq = dev->data->tx_queues[i];
1962                 VIRTQUEUE_DUMP(txvq->vq);
1963         }
1964
1965         set_rxtx_funcs(dev);
1966         hw->started = 1;
1967
1968         /* Initialize Link state */
1969         virtio_dev_link_update(dev, 0);
1970
1971         return 0;
1972 }
1973
1974 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1975 {
1976         struct rte_mbuf *buf;
1977         int i, mbuf_num = 0;
1978
1979         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1980                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1981
1982                 PMD_INIT_LOG(DEBUG,
1983                              "Before freeing rxq[%d] used and unused buf", i);
1984                 VIRTQUEUE_DUMP(rxvq->vq);
1985
1986                 PMD_INIT_LOG(DEBUG, "rx_queues[%d]=%p", i, rxvq);
1987                 while ((buf = virtqueue_detatch_unused(rxvq->vq)) != NULL) {
1988                         rte_pktmbuf_free(buf);
1989                         mbuf_num++;
1990                 }
1991
1992                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1993                 PMD_INIT_LOG(DEBUG,
1994                              "After freeing rxq[%d] used and unused buf", i);
1995                 VIRTQUEUE_DUMP(rxvq->vq);
1996         }
1997
1998         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1999                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
2000
2001                 PMD_INIT_LOG(DEBUG,
2002                              "Before freeing txq[%d] used and unused bufs",
2003                              i);
2004                 VIRTQUEUE_DUMP(txvq->vq);
2005
2006                 mbuf_num = 0;
2007                 while ((buf = virtqueue_detatch_unused(txvq->vq)) != NULL) {
2008                         rte_pktmbuf_free(buf);
2009                         mbuf_num++;
2010                 }
2011
2012                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
2013                 PMD_INIT_LOG(DEBUG,
2014                              "After freeing txq[%d] used and unused buf", i);
2015                 VIRTQUEUE_DUMP(txvq->vq);
2016         }
2017 }
2018
2019 /*
2020  * Stop device: disable interrupt and mark link down
2021  */
2022 static void
2023 virtio_dev_stop(struct rte_eth_dev *dev)
2024 {
2025         struct virtio_hw *hw = dev->data->dev_private;
2026         struct rte_eth_link link;
2027         struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
2028
2029         PMD_INIT_LOG(DEBUG, "stop");
2030
2031         rte_spinlock_lock(&hw->state_lock);
2032         if (intr_conf->lsc || intr_conf->rxq)
2033                 virtio_intr_disable(dev);
2034
2035         hw->started = 0;
2036         memset(&link, 0, sizeof(link));
2037         virtio_dev_atomic_write_link_status(dev, &link);
2038         rte_spinlock_unlock(&hw->state_lock);
2039 }
2040
2041 static int
2042 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
2043 {
2044         struct rte_eth_link link, old;
2045         uint16_t status;
2046         struct virtio_hw *hw = dev->data->dev_private;
2047         memset(&link, 0, sizeof(link));
2048         virtio_dev_atomic_read_link_status(dev, &link);
2049         old = link;
2050         link.link_duplex = ETH_LINK_FULL_DUPLEX;
2051         link.link_speed  = ETH_SPEED_NUM_10G;
2052
2053         if (hw->started == 0) {
2054                 link.link_status = ETH_LINK_DOWN;
2055         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
2056                 PMD_INIT_LOG(DEBUG, "Get link status from hw");
2057                 vtpci_read_dev_config(hw,
2058                                 offsetof(struct virtio_net_config, status),
2059                                 &status, sizeof(status));
2060                 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
2061                         link.link_status = ETH_LINK_DOWN;
2062                         PMD_INIT_LOG(DEBUG, "Port %d is down",
2063                                      dev->data->port_id);
2064                 } else {
2065                         link.link_status = ETH_LINK_UP;
2066                         PMD_INIT_LOG(DEBUG, "Port %d is up",
2067                                      dev->data->port_id);
2068                 }
2069         } else {
2070                 link.link_status = ETH_LINK_UP;
2071         }
2072         virtio_dev_atomic_write_link_status(dev, &link);
2073
2074         return (old.link_status == link.link_status) ? -1 : 0;
2075 }
2076
2077 static int
2078 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2079 {
2080         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
2081         struct virtio_hw *hw = dev->data->dev_private;
2082
2083         if (mask & ETH_VLAN_FILTER_MASK) {
2084                 if (rxmode->hw_vlan_filter &&
2085                                 !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2086
2087                         PMD_DRV_LOG(NOTICE,
2088                                 "vlan filtering not available on this host");
2089
2090                         return -ENOTSUP;
2091                 }
2092         }
2093
2094         if (mask & ETH_VLAN_STRIP_MASK)
2095                 hw->vlan_strip = rxmode->hw_vlan_strip;
2096
2097         return 0;
2098 }
2099
2100 static void
2101 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2102 {
2103         uint64_t tso_mask, host_features;
2104         struct virtio_hw *hw = dev->data->dev_private;
2105
2106         dev_info->speed_capa = ETH_LINK_SPEED_10G; /* fake value */
2107
2108         dev_info->pci_dev = dev->device ? RTE_ETH_DEV_TO_PCI(dev) : NULL;
2109         dev_info->max_rx_queues =
2110                 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2111         dev_info->max_tx_queues =
2112                 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2113         dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2114         dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2115         dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2116         dev_info->default_txconf = (struct rte_eth_txconf) {
2117                 .txq_flags = ETH_TXQ_FLAGS_NOOFFLOADS
2118         };
2119
2120         host_features = VTPCI_OPS(hw)->get_features(hw);
2121         dev_info->rx_offload_capa = 0;
2122         if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2123                 dev_info->rx_offload_capa |=
2124                         DEV_RX_OFFLOAD_TCP_CKSUM |
2125                         DEV_RX_OFFLOAD_UDP_CKSUM;
2126         }
2127         tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2128                 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2129         if ((host_features & tso_mask) == tso_mask)
2130                 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2131
2132         dev_info->tx_offload_capa = 0;
2133         if (hw->guest_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2134                 dev_info->tx_offload_capa |=
2135                         DEV_TX_OFFLOAD_UDP_CKSUM |
2136                         DEV_TX_OFFLOAD_TCP_CKSUM;
2137         }
2138         tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2139                 (1ULL << VIRTIO_NET_F_HOST_TSO6);
2140         if ((hw->guest_features & tso_mask) == tso_mask)
2141                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2142 }
2143
2144 /*
2145  * It enables testpmd to collect per queue stats.
2146  */
2147 static int
2148 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2149 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2150 __rte_unused uint8_t is_rx)
2151 {
2152         return 0;
2153 }
2154
2155 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
2156 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
2157 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio-pci");
2158
2159 RTE_INIT(virtio_init_log);
2160 static void
2161 virtio_init_log(void)
2162 {
2163         virtio_logtype_init = rte_log_register("pmd.virtio.init");
2164         if (virtio_logtype_init >= 0)
2165                 rte_log_set_level(virtio_logtype_init, RTE_LOG_NOTICE);
2166         virtio_logtype_driver = rte_log_register("pmd.virtio.driver");
2167         if (virtio_logtype_driver >= 0)
2168                 rte_log_set_level(virtio_logtype_driver, RTE_LOG_NOTICE);
2169 }