drivers/net/virtio/virtio_ethdev.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2016 Intel Corporation. All rights reserved.
   5  *   All rights reserved.
   6  *
   7  *   Redistribution and use in source and binary forms, with or without
   8  *   modification, are permitted provided that the following conditions
   9  *   are met:
  10  *
  11  *     * Redistributions of source code must retain the above copyright
  12  *       notice, this list of conditions and the following disclaimer.
  13  *     * Redistributions in binary form must reproduce the above copyright
  14  *       notice, this list of conditions and the following disclaimer in
  15  *       the documentation and/or other materials provided with the
  16  *       distribution.
  17  *     * Neither the name of Intel Corporation nor the names of its
  18  *       contributors may be used to endorse or promote products derived
  19  *       from this software without specific prior written permission.
  20  *
  21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  */
  33
  34 #include <stdint.h>
  35 #include <string.h>
  36 #include <stdio.h>
  37 #include <errno.h>
  38 #include <unistd.h>
  39
  40 #include <rte_ethdev.h>
  41 #include <rte_ethdev_pci.h>
  42 #include <rte_memcpy.h>
  43 #include <rte_string_fns.h>
  44 #include <rte_memzone.h>
  45 #include <rte_malloc.h>
  46 #include <rte_atomic.h>
  47 #include <rte_branch_prediction.h>
  48 #include <rte_pci.h>
  49 #include <rte_bus_pci.h>
  50 #include <rte_ether.h>
  51 #include <rte_common.h>
  52 #include <rte_errno.h>
  53 #include <rte_cpuflags.h>
  54
  55 #include <rte_memory.h>
  56 #include <rte_eal.h>
  57 #include <rte_dev.h>
  58
  59 #include "virtio_ethdev.h"
  60 #include "virtio_pci.h"
  61 #include "virtio_logs.h"
  62 #include "virtqueue.h"
  63 #include "virtio_rxtx.h"
  64
  65 static int eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev);
  66 static int  virtio_dev_configure(struct rte_eth_dev *dev);
  67 static int  virtio_dev_start(struct rte_eth_dev *dev);
  68 static void virtio_dev_stop(struct rte_eth_dev *dev);
  69 static void virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
  70 static void virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
  71 static void virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
  72 static void virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
  73 static void virtio_dev_info_get(struct rte_eth_dev *dev,
  74                                 struct rte_eth_dev_info *dev_info);
  75 static int virtio_dev_link_update(struct rte_eth_dev *dev,
  76         int wait_to_complete);
  77 static int virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask);
  78
  79 static void virtio_set_hwaddr(struct virtio_hw *hw);
  80 static void virtio_get_hwaddr(struct virtio_hw *hw);
  81
  82 static int virtio_dev_stats_get(struct rte_eth_dev *dev,
  83                                  struct rte_eth_stats *stats);
  84 static int virtio_dev_xstats_get(struct rte_eth_dev *dev,
  85                                  struct rte_eth_xstat *xstats, unsigned n);
  86 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
  87                                        struct rte_eth_xstat_name *xstats_names,
  88                                        unsigned limit);
  89 static void virtio_dev_stats_reset(struct rte_eth_dev *dev);
  90 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
  91 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
  92                                 uint16_t vlan_id, int on);
  93 static int virtio_mac_addr_add(struct rte_eth_dev *dev,
  94                                 struct ether_addr *mac_addr,
  95                                 uint32_t index, uint32_t vmdq);
  96 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
  97 static void virtio_mac_addr_set(struct rte_eth_dev *dev,
  98                                 struct ether_addr *mac_addr);
  99
 100 static int virtio_intr_enable(struct rte_eth_dev *dev);
 101 static int virtio_intr_disable(struct rte_eth_dev *dev);
 102
 103 static int virtio_dev_queue_stats_mapping_set(
 104         struct rte_eth_dev *eth_dev,
 105         uint16_t queue_id,
 106         uint8_t stat_idx,
 107         uint8_t is_rx);
 108
 109 /*
 110  * The set of PCI devices this driver supports
 111  */
 112 static const struct rte_pci_id pci_id_virtio_map[] = {
 113         { RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_LEGACY_DEVICEID_NET) },
 114         { RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_MODERN_DEVICEID_NET) },
 115         { .vendor_id = 0, /* sentinel */ },
 116 };
 117
 118 struct rte_virtio_xstats_name_off {
 119         char name[RTE_ETH_XSTATS_NAME_SIZE];
 120         unsigned offset;
 121 };
 122
 123 /* [rt]x_qX_ is prepended to the name string here */
 124 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
 125         {"good_packets",           offsetof(struct virtnet_rx, stats.packets)},
 126         {"good_bytes",             offsetof(struct virtnet_rx, stats.bytes)},
 127         {"errors",                 offsetof(struct virtnet_rx, stats.errors)},
 128         {"multicast_packets",      offsetof(struct virtnet_rx, stats.multicast)},
 129         {"broadcast_packets",      offsetof(struct virtnet_rx, stats.broadcast)},
 130         {"undersize_packets",      offsetof(struct virtnet_rx, stats.size_bins[0])},
 131         {"size_64_packets",        offsetof(struct virtnet_rx, stats.size_bins[1])},
 132         {"size_65_127_packets",    offsetof(struct virtnet_rx, stats.size_bins[2])},
 133         {"size_128_255_packets",   offsetof(struct virtnet_rx, stats.size_bins[3])},
 134         {"size_256_511_packets",   offsetof(struct virtnet_rx, stats.size_bins[4])},
 135         {"size_512_1023_packets",  offsetof(struct virtnet_rx, stats.size_bins[5])},
 136         {"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
 137         {"size_1519_max_packets",  offsetof(struct virtnet_rx, stats.size_bins[7])},
 138 };
 139
 140 /* [rt]x_qX_ is prepended to the name string here */
 141 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
 142         {"good_packets",           offsetof(struct virtnet_tx, stats.packets)},
 143         {"good_bytes",             offsetof(struct virtnet_tx, stats.bytes)},
 144         {"errors",                 offsetof(struct virtnet_tx, stats.errors)},
 145         {"multicast_packets",      offsetof(struct virtnet_tx, stats.multicast)},
 146         {"broadcast_packets",      offsetof(struct virtnet_tx, stats.broadcast)},
 147         {"undersize_packets",      offsetof(struct virtnet_tx, stats.size_bins[0])},
 148         {"size_64_packets",        offsetof(struct virtnet_tx, stats.size_bins[1])},
 149         {"size_65_127_packets",    offsetof(struct virtnet_tx, stats.size_bins[2])},
 150         {"size_128_255_packets",   offsetof(struct virtnet_tx, stats.size_bins[3])},
 151         {"size_256_511_packets",   offsetof(struct virtnet_tx, stats.size_bins[4])},
 152         {"size_512_1023_packets",  offsetof(struct virtnet_tx, stats.size_bins[5])},
 153         {"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
 154         {"size_1519_max_packets",  offsetof(struct virtnet_tx, stats.size_bins[7])},
 155 };
 156
 157 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
 158                             sizeof(rte_virtio_rxq_stat_strings[0]))
 159 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
 160                             sizeof(rte_virtio_txq_stat_strings[0]))
 161
 162 struct virtio_hw_internal virtio_hw_internal[RTE_MAX_ETHPORTS];
 163
 164 static int
 165 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
 166                 int *dlen, int pkt_num)
 167 {
 168         uint32_t head, i;
 169         int k, sum = 0;
 170         virtio_net_ctrl_ack status = ~0;
 171         struct virtio_pmd_ctrl *result;
 172         struct virtqueue *vq;
 173
 174         ctrl->status = status;
 175
 176         if (!cvq || !cvq->vq) {
 177                 PMD_INIT_LOG(ERR, "Control queue is not supported.");
 178                 return -1;
 179         }
 180         vq = cvq->vq;
 181         head = vq->vq_desc_head_idx;
 182
 183         PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
 184                 "vq->hw->cvq = %p vq = %p",
 185                 vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
 186
 187         if ((vq->vq_free_cnt < ((uint32_t)pkt_num + 2)) || (pkt_num < 1))
 188                 return -1;
 189
 190         memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
 191                 sizeof(struct virtio_pmd_ctrl));
 192
 193         /*
 194          * Format is enforced in qemu code:
 195          * One TX packet for header;
 196          * At least one TX packet per argument;
 197          * One RX packet for ACK.
 198          */
 199         vq->vq_ring.desc[head].flags = VRING_DESC_F_NEXT;
 200         vq->vq_ring.desc[head].addr = cvq->virtio_net_hdr_mem;
 201         vq->vq_ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
 202         vq->vq_free_cnt--;
 203         i = vq->vq_ring.desc[head].next;
 204
 205         for (k = 0; k < pkt_num; k++) {
 206                 vq->vq_ring.desc[i].flags = VRING_DESC_F_NEXT;
 207                 vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
 208                         + sizeof(struct virtio_net_ctrl_hdr)
 209                         + sizeof(ctrl->status) + sizeof(uint8_t)*sum;
 210                 vq->vq_ring.desc[i].len = dlen[k];
 211                 sum += dlen[k];
 212                 vq->vq_free_cnt--;
 213                 i = vq->vq_ring.desc[i].next;
 214         }
 215
 216         vq->vq_ring.desc[i].flags = VRING_DESC_F_WRITE;
 217         vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
 218                         + sizeof(struct virtio_net_ctrl_hdr);
 219         vq->vq_ring.desc[i].len = sizeof(ctrl->status);
 220         vq->vq_free_cnt--;
 221
 222         vq->vq_desc_head_idx = vq->vq_ring.desc[i].next;
 223
 224         vq_update_avail_ring(vq, head);
 225         vq_update_avail_idx(vq);
 226
 227         PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
 228
 229         virtqueue_notify(vq);
 230
 231         rte_rmb();
 232         while (VIRTQUEUE_NUSED(vq) == 0) {
 233                 rte_rmb();
 234                 usleep(100);
 235         }
 236
 237         while (VIRTQUEUE_NUSED(vq)) {
 238                 uint32_t idx, desc_idx, used_idx;
 239                 struct vring_used_elem *uep;
 240
 241                 used_idx = (uint32_t)(vq->vq_used_cons_idx
 242                                 & (vq->vq_nentries - 1));
 243                 uep = &vq->vq_ring.used->ring[used_idx];
 244                 idx = (uint32_t) uep->id;
 245                 desc_idx = idx;
 246
 247                 while (vq->vq_ring.desc[desc_idx].flags & VRING_DESC_F_NEXT) {
 248                         desc_idx = vq->vq_ring.desc[desc_idx].next;
 249                         vq->vq_free_cnt++;
 250                 }
 251
 252                 vq->vq_ring.desc[desc_idx].next = vq->vq_desc_head_idx;
 253                 vq->vq_desc_head_idx = idx;
 254
 255                 vq->vq_used_cons_idx++;
 256                 vq->vq_free_cnt++;
 257         }
 258
 259         PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
 260                         vq->vq_free_cnt, vq->vq_desc_head_idx);
 261
 262         result = cvq->virtio_net_hdr_mz->addr;
 263
 264         return result->status;
 265 }
 266
 267 static int
 268 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
 269 {
 270         struct virtio_hw *hw = dev->data->dev_private;
 271         struct virtio_pmd_ctrl ctrl;
 272         int dlen[1];
 273         int ret;
 274
 275         ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
 276         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
 277         memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
 278
 279         dlen[0] = sizeof(uint16_t);
 280
 281         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 282         if (ret) {
 283                 PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
 284                           "failed, this is too late now...");
 285                 return -EINVAL;
 286         }
 287
 288         return 0;
 289 }
 290
 291 static void
 292 virtio_dev_queue_release(void *queue __rte_unused)
 293 {
 294         /* do nothing */
 295 }
 296
 297 static int
 298 virtio_get_queue_type(struct virtio_hw *hw, uint16_t vtpci_queue_idx)
 299 {
 300         if (vtpci_queue_idx == hw->max_queue_pairs * 2)
 301                 return VTNET_CQ;
 302         else if (vtpci_queue_idx % 2 == 0)
 303                 return VTNET_RQ;
 304         else
 305                 return VTNET_TQ;
 306 }
 307
 308 static uint16_t
 309 virtio_get_nr_vq(struct virtio_hw *hw)
 310 {
 311         uint16_t nr_vq = hw->max_queue_pairs * 2;
 312
 313         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
 314                 nr_vq += 1;
 315
 316         return nr_vq;
 317 }
 318
 319 static void
 320 virtio_init_vring(struct virtqueue *vq)
 321 {
 322         int size = vq->vq_nentries;
 323         struct vring *vr = &vq->vq_ring;
 324         uint8_t *ring_mem = vq->vq_ring_virt_mem;
 325
 326         PMD_INIT_FUNC_TRACE();
 327
 328         /*
 329          * Reinitialise since virtio port might have been stopped and restarted
 330          */
 331         memset(ring_mem, 0, vq->vq_ring_size);
 332         vring_init(vr, size, ring_mem, VIRTIO_PCI_VRING_ALIGN);
 333         vq->vq_used_cons_idx = 0;
 334         vq->vq_desc_head_idx = 0;
 335         vq->vq_avail_idx = 0;
 336         vq->vq_desc_tail_idx = (uint16_t)(vq->vq_nentries - 1);
 337         vq->vq_free_cnt = vq->vq_nentries;
 338         memset(vq->vq_descx, 0, sizeof(struct vq_desc_extra) * vq->vq_nentries);
 339
 340         vring_desc_init(vr->desc, size);
 341
 342         /*
 343          * Disable device(host) interrupting guest
 344          */
 345         virtqueue_disable_intr(vq);
 346 }
 347
 348 static int
 349 virtio_init_queue(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx)
 350 {
 351         char vq_name[VIRTQUEUE_MAX_NAME_SZ];
 352         char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
 353         const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
 354         unsigned int vq_size, size;
 355         struct virtio_hw *hw = dev->data->dev_private;
 356         struct virtnet_rx *rxvq = NULL;
 357         struct virtnet_tx *txvq = NULL;
 358         struct virtnet_ctl *cvq = NULL;
 359         struct virtqueue *vq;
 360         size_t sz_hdr_mz = 0;
 361         void *sw_ring = NULL;
 362         int queue_type = virtio_get_queue_type(hw, vtpci_queue_idx);
 363         int ret;
 364
 365         PMD_INIT_LOG(DEBUG, "setting up queue: %u", vtpci_queue_idx);
 366
 367         /*
 368          * Read the virtqueue size from the Queue Size field
 369          * Always power of 2 and if 0 virtqueue does not exist
 370          */
 371         vq_size = VTPCI_OPS(hw)->get_queue_num(hw, vtpci_queue_idx);
 372         PMD_INIT_LOG(DEBUG, "vq_size: %u", vq_size);
 373         if (vq_size == 0) {
 374                 PMD_INIT_LOG(ERR, "virtqueue does not exist");
 375                 return -EINVAL;
 376         }
 377
 378         if (!rte_is_power_of_2(vq_size)) {
 379                 PMD_INIT_LOG(ERR, "virtqueue size is not powerof 2");
 380                 return -EINVAL;
 381         }
 382
 383         snprintf(vq_name, sizeof(vq_name), "port%d_vq%d",
 384                  dev->data->port_id, vtpci_queue_idx);
 385
 386         size = RTE_ALIGN_CEIL(sizeof(*vq) +
 387                                 vq_size * sizeof(struct vq_desc_extra),
 388                                 RTE_CACHE_LINE_SIZE);
 389         if (queue_type == VTNET_TQ) {
 390                 /*
 391                  * For each xmit packet, allocate a virtio_net_hdr
 392                  * and indirect ring elements
 393                  */
 394                 sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
 395         } else if (queue_type == VTNET_CQ) {
 396                 /* Allocate a page for control vq command, data and status */
 397                 sz_hdr_mz = PAGE_SIZE;
 398         }
 399
 400         vq = rte_zmalloc_socket(vq_name, size, RTE_CACHE_LINE_SIZE,
 401                                 SOCKET_ID_ANY);
 402         if (vq == NULL) {
 403                 PMD_INIT_LOG(ERR, "can not allocate vq");
 404                 return -ENOMEM;
 405         }
 406         hw->vqs[vtpci_queue_idx] = vq;
 407
 408         vq->hw = hw;
 409         vq->vq_queue_index = vtpci_queue_idx;
 410         vq->vq_nentries = vq_size;
 411
 412         /*
 413          * Reserve a memzone for vring elements
 414          */
 415         size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
 416         vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
 417         PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
 418                      size, vq->vq_ring_size);
 419
 420         mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
 421                                          SOCKET_ID_ANY,
 422                                          0, VIRTIO_PCI_VRING_ALIGN);
 423         if (mz == NULL) {
 424                 if (rte_errno == EEXIST)
 425                         mz = rte_memzone_lookup(vq_name);
 426                 if (mz == NULL) {
 427                         ret = -ENOMEM;
 428                         goto fail_q_alloc;
 429                 }
 430         }
 431
 432         memset(mz->addr, 0, mz->len);
 433
 434         vq->vq_ring_mem = mz->iova;
 435         vq->vq_ring_virt_mem = mz->addr;
 436         PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%" PRIx64,
 437                      (uint64_t)mz->iova);
 438         PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
 439                      (uint64_t)(uintptr_t)mz->addr);
 440
 441         virtio_init_vring(vq);
 442
 443         if (sz_hdr_mz) {
 444                 snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_vq%d_hdr",
 445                          dev->data->port_id, vtpci_queue_idx);
 446                 hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
 447                                                      SOCKET_ID_ANY, 0,
 448                                                      RTE_CACHE_LINE_SIZE);
 449                 if (hdr_mz == NULL) {
 450                         if (rte_errno == EEXIST)
 451                                 hdr_mz = rte_memzone_lookup(vq_hdr_name);
 452                         if (hdr_mz == NULL) {
 453                                 ret = -ENOMEM;
 454                                 goto fail_q_alloc;
 455                         }
 456                 }
 457         }
 458
 459         if (queue_type == VTNET_RQ) {
 460                 size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
 461                                sizeof(vq->sw_ring[0]);
 462
 463                 sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
 464                                 RTE_CACHE_LINE_SIZE, SOCKET_ID_ANY);
 465                 if (!sw_ring) {
 466                         PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
 467                         ret = -ENOMEM;
 468                         goto fail_q_alloc;
 469                 }
 470
 471                 vq->sw_ring = sw_ring;
 472                 rxvq = &vq->rxq;
 473                 rxvq->vq = vq;
 474                 rxvq->port_id = dev->data->port_id;
 475                 rxvq->mz = mz;
 476         } else if (queue_type == VTNET_TQ) {
 477                 txvq = &vq->txq;
 478                 txvq->vq = vq;
 479                 txvq->port_id = dev->data->port_id;
 480                 txvq->mz = mz;
 481                 txvq->virtio_net_hdr_mz = hdr_mz;
 482                 txvq->virtio_net_hdr_mem = hdr_mz->iova;
 483         } else if (queue_type == VTNET_CQ) {
 484                 cvq = &vq->cq;
 485                 cvq->vq = vq;
 486                 cvq->mz = mz;
 487                 cvq->virtio_net_hdr_mz = hdr_mz;
 488                 cvq->virtio_net_hdr_mem = hdr_mz->iova;
 489                 memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
 490
 491                 hw->cvq = cvq;
 492         }
 493
 494         /* For virtio_user case (that is when hw->dev is NULL), we use
 495          * virtual address. And we need properly set _offset_, please see
 496          * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
 497          */
 498         if (!hw->virtio_user_dev)
 499                 vq->offset = offsetof(struct rte_mbuf, buf_iova);
 500         else {
 501                 vq->vq_ring_mem = (uintptr_t)mz->addr;
 502                 vq->offset = offsetof(struct rte_mbuf, buf_addr);
 503                 if (queue_type == VTNET_TQ)
 504                         txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
 505                 else if (queue_type == VTNET_CQ)
 506                         cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
 507         }
 508
 509         if (queue_type == VTNET_TQ) {
 510                 struct virtio_tx_region *txr;
 511                 unsigned int i;
 512
 513                 txr = hdr_mz->addr;
 514                 memset(txr, 0, vq_size * sizeof(*txr));
 515                 for (i = 0; i < vq_size; i++) {
 516                         struct vring_desc *start_dp = txr[i].tx_indir;
 517
 518                         vring_desc_init(start_dp, RTE_DIM(txr[i].tx_indir));
 519
 520                         /* first indirect descriptor is always the tx header */
 521                         start_dp->addr = txvq->virtio_net_hdr_mem
 522                                 + i * sizeof(*txr)
 523                                 + offsetof(struct virtio_tx_region, tx_hdr);
 524
 525                         start_dp->len = hw->vtnet_hdr_size;
 526                         start_dp->flags = VRING_DESC_F_NEXT;
 527                 }
 528         }
 529
 530         if (VTPCI_OPS(hw)->setup_queue(hw, vq) < 0) {
 531                 PMD_INIT_LOG(ERR, "setup_queue failed");
 532                 return -EINVAL;
 533         }
 534
 535         return 0;
 536
 537 fail_q_alloc:
 538         rte_free(sw_ring);
 539         rte_memzone_free(hdr_mz);
 540         rte_memzone_free(mz);
 541         rte_free(vq);
 542
 543         return ret;
 544 }
 545
 546 static void
 547 virtio_free_queues(struct virtio_hw *hw)
 548 {
 549         uint16_t nr_vq = virtio_get_nr_vq(hw);
 550         struct virtqueue *vq;
 551         int queue_type;
 552         uint16_t i;
 553
 554         if (hw->vqs == NULL)
 555                 return;
 556
 557         for (i = 0; i < nr_vq; i++) {
 558                 vq = hw->vqs[i];
 559                 if (!vq)
 560                         continue;
 561
 562                 queue_type = virtio_get_queue_type(hw, i);
 563                 if (queue_type == VTNET_RQ) {
 564                         rte_free(vq->sw_ring);
 565                         rte_memzone_free(vq->rxq.mz);
 566                 } else if (queue_type == VTNET_TQ) {
 567                         rte_memzone_free(vq->txq.mz);
 568                         rte_memzone_free(vq->txq.virtio_net_hdr_mz);
 569                 } else {
 570                         rte_memzone_free(vq->cq.mz);
 571                         rte_memzone_free(vq->cq.virtio_net_hdr_mz);
 572                 }
 573
 574                 rte_free(vq);
 575                 hw->vqs[i] = NULL;
 576         }
 577
 578         rte_free(hw->vqs);
 579         hw->vqs = NULL;
 580 }
 581
 582 static int
 583 virtio_alloc_queues(struct rte_eth_dev *dev)
 584 {
 585         struct virtio_hw *hw = dev->data->dev_private;
 586         uint16_t nr_vq = virtio_get_nr_vq(hw);
 587         uint16_t i;
 588         int ret;
 589
 590         hw->vqs = rte_zmalloc(NULL, sizeof(struct virtqueue *) * nr_vq, 0);
 591         if (!hw->vqs) {
 592                 PMD_INIT_LOG(ERR, "failed to allocate vqs");
 593                 return -ENOMEM;
 594         }
 595
 596         for (i = 0; i < nr_vq; i++) {
 597                 ret = virtio_init_queue(dev, i);
 598                 if (ret < 0) {
 599                         virtio_free_queues(hw);
 600                         return ret;
 601                 }
 602         }
 603
 604         return 0;
 605 }
 606
 607 static void virtio_queues_unbind_intr(struct rte_eth_dev *dev);
 608
 609 static void
 610 virtio_dev_close(struct rte_eth_dev *dev)
 611 {
 612         struct virtio_hw *hw = dev->data->dev_private;
 613         struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
 614
 615         PMD_INIT_LOG(DEBUG, "virtio_dev_close");
 616
 617         /* reset the NIC */
 618         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
 619                 VTPCI_OPS(hw)->set_config_irq(hw, VIRTIO_MSI_NO_VECTOR);
 620         if (intr_conf->rxq)
 621                 virtio_queues_unbind_intr(dev);
 622
 623         if (intr_conf->lsc || intr_conf->rxq) {
 624                 virtio_intr_disable(dev);
 625                 rte_intr_efd_disable(dev->intr_handle);
 626                 rte_free(dev->intr_handle->intr_vec);
 627                 dev->intr_handle->intr_vec = NULL;
 628         }
 629
 630         vtpci_reset(hw);
 631         virtio_dev_free_mbufs(dev);
 632         virtio_free_queues(hw);
 633 }
 634
 635 static void
 636 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
 637 {
 638         struct virtio_hw *hw = dev->data->dev_private;
 639         struct virtio_pmd_ctrl ctrl;
 640         int dlen[1];
 641         int ret;
 642
 643         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 644                 PMD_INIT_LOG(INFO, "host does not support rx control");
 645                 return;
 646         }
 647
 648         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 649         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 650         ctrl.data[0] = 1;
 651         dlen[0] = 1;
 652
 653         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 654         if (ret)
 655                 PMD_INIT_LOG(ERR, "Failed to enable promisc");
 656 }
 657
 658 static void
 659 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
 660 {
 661         struct virtio_hw *hw = dev->data->dev_private;
 662         struct virtio_pmd_ctrl ctrl;
 663         int dlen[1];
 664         int ret;
 665
 666         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 667                 PMD_INIT_LOG(INFO, "host does not support rx control");
 668                 return;
 669         }
 670
 671         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 672         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 673         ctrl.data[0] = 0;
 674         dlen[0] = 1;
 675
 676         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 677         if (ret)
 678                 PMD_INIT_LOG(ERR, "Failed to disable promisc");
 679 }
 680
 681 static void
 682 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
 683 {
 684         struct virtio_hw *hw = dev->data->dev_private;
 685         struct virtio_pmd_ctrl ctrl;
 686         int dlen[1];
 687         int ret;
 688
 689         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 690                 PMD_INIT_LOG(INFO, "host does not support rx control");
 691                 return;
 692         }
 693
 694         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 695         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 696         ctrl.data[0] = 1;
 697         dlen[0] = 1;
 698
 699         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 700         if (ret)
 701                 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
 702 }
 703
 704 static void
 705 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
 706 {
 707         struct virtio_hw *hw = dev->data->dev_private;
 708         struct virtio_pmd_ctrl ctrl;
 709         int dlen[1];
 710         int ret;
 711
 712         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 713                 PMD_INIT_LOG(INFO, "host does not support rx control");
 714                 return;
 715         }
 716
 717         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 718         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 719         ctrl.data[0] = 0;
 720         dlen[0] = 1;
 721
 722         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 723         if (ret)
 724                 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
 725 }
 726
 727 #define VLAN_TAG_LEN           4    /* 802.3ac tag (not DMA'd) */
 728 static int
 729 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 730 {
 731         struct virtio_hw *hw = dev->data->dev_private;
 732         uint32_t ether_hdr_len = ETHER_HDR_LEN + VLAN_TAG_LEN +
 733                                  hw->vtnet_hdr_size;
 734         uint32_t frame_size = mtu + ether_hdr_len;
 735         uint32_t max_frame_size = hw->max_mtu + ether_hdr_len;
 736
 737         max_frame_size = RTE_MIN(max_frame_size, VIRTIO_MAX_RX_PKTLEN);
 738
 739         if (mtu < ETHER_MIN_MTU || frame_size > max_frame_size) {
 740                 PMD_INIT_LOG(ERR, "MTU should be between %d and %d",
 741                         ETHER_MIN_MTU, max_frame_size - ether_hdr_len);
 742                 return -EINVAL;
 743         }
 744         return 0;
 745 }
 746
 747 static int
 748 virtio_dev_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
 749 {
 750         struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
 751         struct virtqueue *vq = rxvq->vq;
 752
 753         virtqueue_enable_intr(vq);
 754         return 0;
 755 }
 756
 757 static int
 758 virtio_dev_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
 759 {
 760         struct virtnet_rx *rxvq = dev->data->rx_queues[queue_id];
 761         struct virtqueue *vq = rxvq->vq;
 762
 763         virtqueue_disable_intr(vq);
 764         return 0;
 765 }
 766
 767 /*
 768  * dev_ops for virtio, bare necessities for basic operation
 769  */
 770 static const struct eth_dev_ops virtio_eth_dev_ops = {
 771         .dev_configure           = virtio_dev_configure,
 772         .dev_start               = virtio_dev_start,
 773         .dev_stop                = virtio_dev_stop,
 774         .dev_close               = virtio_dev_close,
 775         .promiscuous_enable      = virtio_dev_promiscuous_enable,
 776         .promiscuous_disable     = virtio_dev_promiscuous_disable,
 777         .allmulticast_enable     = virtio_dev_allmulticast_enable,
 778         .allmulticast_disable    = virtio_dev_allmulticast_disable,
 779         .mtu_set                 = virtio_mtu_set,
 780         .dev_infos_get           = virtio_dev_info_get,
 781         .stats_get               = virtio_dev_stats_get,
 782         .xstats_get              = virtio_dev_xstats_get,
 783         .xstats_get_names        = virtio_dev_xstats_get_names,
 784         .stats_reset             = virtio_dev_stats_reset,
 785         .xstats_reset            = virtio_dev_stats_reset,
 786         .link_update             = virtio_dev_link_update,
 787         .vlan_offload_set        = virtio_dev_vlan_offload_set,
 788         .rx_queue_setup          = virtio_dev_rx_queue_setup,
 789         .rx_queue_intr_enable    = virtio_dev_rx_queue_intr_enable,
 790         .rx_queue_intr_disable   = virtio_dev_rx_queue_intr_disable,
 791         .rx_queue_release        = virtio_dev_queue_release,
 792         .rx_descriptor_done      = virtio_dev_rx_queue_done,
 793         .tx_queue_setup          = virtio_dev_tx_queue_setup,
 794         .tx_queue_release        = virtio_dev_queue_release,
 795         /* collect stats per queue */
 796         .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
 797         .vlan_filter_set         = virtio_vlan_filter_set,
 798         .mac_addr_add            = virtio_mac_addr_add,
 799         .mac_addr_remove         = virtio_mac_addr_remove,
 800         .mac_addr_set            = virtio_mac_addr_set,
 801 };
 802
 803 static inline int
 804 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
 805                                 struct rte_eth_link *link)
 806 {
 807         struct rte_eth_link *dst = link;
 808         struct rte_eth_link *src = &(dev->data->dev_link);
 809
 810         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
 811                         *(uint64_t *)src) == 0)
 812                 return -1;
 813
 814         return 0;
 815 }
 816
 817 /**
 818  * Atomically writes the link status information into global
 819  * structure rte_eth_dev.
 820  *
 821  * @param dev
 822  *   - Pointer to the structure rte_eth_dev to read from.
 823  *   - Pointer to the buffer to be saved with the link status.
 824  *
 825  * @return
 826  *   - On success, zero.
 827  *   - On failure, negative value.
 828  */
 829 static inline int
 830 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
 831                 struct rte_eth_link *link)
 832 {
 833         struct rte_eth_link *dst = &(dev->data->dev_link);
 834         struct rte_eth_link *src = link;
 835
 836         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
 837                                         *(uint64_t *)src) == 0)
 838                 return -1;
 839
 840         return 0;
 841 }
 842
 843 static void
 844 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 845 {
 846         unsigned i;
 847
 848         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 849                 const struct virtnet_tx *txvq = dev->data->tx_queues[i];
 850                 if (txvq == NULL)
 851                         continue;
 852
 853                 stats->opackets += txvq->stats.packets;
 854                 stats->obytes += txvq->stats.bytes;
 855                 stats->oerrors += txvq->stats.errors;
 856
 857                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 858                         stats->q_opackets[i] = txvq->stats.packets;
 859                         stats->q_obytes[i] = txvq->stats.bytes;
 860                 }
 861         }
 862
 863         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 864                 const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
 865                 if (rxvq == NULL)
 866                         continue;
 867
 868                 stats->ipackets += rxvq->stats.packets;
 869                 stats->ibytes += rxvq->stats.bytes;
 870                 stats->ierrors += rxvq->stats.errors;
 871
 872                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 873                         stats->q_ipackets[i] = rxvq->stats.packets;
 874                         stats->q_ibytes[i] = rxvq->stats.bytes;
 875                 }
 876         }
 877
 878         stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
 879 }
 880
 881 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
 882                                        struct rte_eth_xstat_name *xstats_names,
 883                                        __rte_unused unsigned limit)
 884 {
 885         unsigned i;
 886         unsigned count = 0;
 887         unsigned t;
 888
 889         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
 890                 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
 891
 892         if (xstats_names != NULL) {
 893                 /* Note: limit checked in rte_eth_xstats_names() */
 894
 895                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
 896                         struct virtqueue *rxvq = dev->data->rx_queues[i];
 897                         if (rxvq == NULL)
 898                                 continue;
 899                         for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
 900                                 snprintf(xstats_names[count].name,
 901                                         sizeof(xstats_names[count].name),
 902                                         "rx_q%u_%s", i,
 903                                         rte_virtio_rxq_stat_strings[t].name);
 904                                 count++;
 905                         }
 906                 }
 907
 908                 for (i = 0; i < dev->data->nb_tx_queues; i++) {
 909                         struct virtqueue *txvq = dev->data->tx_queues[i];
 910                         if (txvq == NULL)
 911                                 continue;
 912                         for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
 913                                 snprintf(xstats_names[count].name,
 914                                         sizeof(xstats_names[count].name),
 915                                         "tx_q%u_%s", i,
 916                                         rte_virtio_txq_stat_strings[t].name);
 917                                 count++;
 918                         }
 919                 }
 920                 return count;
 921         }
 922         return nstats;
 923 }
 924
 925 static int
 926 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 927                       unsigned n)
 928 {
 929         unsigned i;
 930         unsigned count = 0;
 931
 932         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
 933                 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
 934
 935         if (n < nstats)
 936                 return nstats;
 937
 938         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 939                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
 940
 941                 if (rxvq == NULL)
 942                         continue;
 943
 944                 unsigned t;
 945
 946                 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
 947                         xstats[count].value = *(uint64_t *)(((char *)rxvq) +
 948                                 rte_virtio_rxq_stat_strings[t].offset);
 949                         xstats[count].id = count;
 950                         count++;
 951                 }
 952         }
 953
 954         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 955                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
 956
 957                 if (txvq == NULL)
 958                         continue;
 959
 960                 unsigned t;
 961
 962                 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
 963                         xstats[count].value = *(uint64_t *)(((char *)txvq) +
 964                                 rte_virtio_txq_stat_strings[t].offset);
 965                         xstats[count].id = count;
 966                         count++;
 967                 }
 968         }
 969
 970         return count;
 971 }
 972
 973 static int
 974 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 975 {
 976         virtio_update_stats(dev, stats);
 977
 978         return 0;
 979 }
 980
 981 static void
 982 virtio_dev_stats_reset(struct rte_eth_dev *dev)
 983 {
 984         unsigned int i;
 985
 986         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 987                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
 988                 if (txvq == NULL)
 989                         continue;
 990
 991                 txvq->stats.packets = 0;
 992                 txvq->stats.bytes = 0;
 993                 txvq->stats.errors = 0;
 994                 txvq->stats.multicast = 0;
 995                 txvq->stats.broadcast = 0;
 996                 memset(txvq->stats.size_bins, 0,
 997                        sizeof(txvq->stats.size_bins[0]) * 8);
 998         }
 999
1000         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1001                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1002                 if (rxvq == NULL)
1003                         continue;
1004
1005                 rxvq->stats.packets = 0;
1006                 rxvq->stats.bytes = 0;
1007                 rxvq->stats.errors = 0;
1008                 rxvq->stats.multicast = 0;
1009                 rxvq->stats.broadcast = 0;
1010                 memset(rxvq->stats.size_bins, 0,
1011                        sizeof(rxvq->stats.size_bins[0]) * 8);
1012         }
1013 }
1014
1015 static void
1016 virtio_set_hwaddr(struct virtio_hw *hw)
1017 {
1018         vtpci_write_dev_config(hw,
1019                         offsetof(struct virtio_net_config, mac),
1020                         &hw->mac_addr, ETHER_ADDR_LEN);
1021 }
1022
1023 static void
1024 virtio_get_hwaddr(struct virtio_hw *hw)
1025 {
1026         if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
1027                 vtpci_read_dev_config(hw,
1028                         offsetof(struct virtio_net_config, mac),
1029                         &hw->mac_addr, ETHER_ADDR_LEN);
1030         } else {
1031                 eth_random_addr(&hw->mac_addr[0]);
1032                 virtio_set_hwaddr(hw);
1033         }
1034 }
1035
1036 static int
1037 virtio_mac_table_set(struct virtio_hw *hw,
1038                      const struct virtio_net_ctrl_mac *uc,
1039                      const struct virtio_net_ctrl_mac *mc)
1040 {
1041         struct virtio_pmd_ctrl ctrl;
1042         int err, len[2];
1043
1044         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1045                 PMD_DRV_LOG(INFO, "host does not support mac table");
1046                 return -1;
1047         }
1048
1049         ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1050         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
1051
1052         len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
1053         memcpy(ctrl.data, uc, len[0]);
1054
1055         len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
1056         memcpy(ctrl.data + len[0], mc, len[1]);
1057
1058         err = virtio_send_command(hw->cvq, &ctrl, len, 2);
1059         if (err != 0)
1060                 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
1061         return err;
1062 }
1063
1064 static int
1065 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
1066                     uint32_t index, uint32_t vmdq __rte_unused)
1067 {
1068         struct virtio_hw *hw = dev->data->dev_private;
1069         const struct ether_addr *addrs = dev->data->mac_addrs;
1070         unsigned int i;
1071         struct virtio_net_ctrl_mac *uc, *mc;
1072
1073         if (index >= VIRTIO_MAX_MAC_ADDRS) {
1074                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1075                 return -EINVAL;
1076         }
1077
1078         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1079         uc->entries = 0;
1080         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1081         mc->entries = 0;
1082
1083         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1084                 const struct ether_addr *addr
1085                         = (i == index) ? mac_addr : addrs + i;
1086                 struct virtio_net_ctrl_mac *tbl
1087                         = is_multicast_ether_addr(addr) ? mc : uc;
1088
1089                 memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
1090         }
1091
1092         return virtio_mac_table_set(hw, uc, mc);
1093 }
1094
1095 static void
1096 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
1097 {
1098         struct virtio_hw *hw = dev->data->dev_private;
1099         struct ether_addr *addrs = dev->data->mac_addrs;
1100         struct virtio_net_ctrl_mac *uc, *mc;
1101         unsigned int i;
1102
1103         if (index >= VIRTIO_MAX_MAC_ADDRS) {
1104                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1105                 return;
1106         }
1107
1108         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1109         uc->entries = 0;
1110         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1111         mc->entries = 0;
1112
1113         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1114                 struct virtio_net_ctrl_mac *tbl;
1115
1116                 if (i == index || is_zero_ether_addr(addrs + i))
1117                         continue;
1118
1119                 tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
1120                 memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
1121         }
1122
1123         virtio_mac_table_set(hw, uc, mc);
1124 }
1125
1126 static void
1127 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1128 {
1129         struct virtio_hw *hw = dev->data->dev_private;
1130
1131         memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
1132
1133         /* Use atomic update if available */
1134         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1135                 struct virtio_pmd_ctrl ctrl;
1136                 int len = ETHER_ADDR_LEN;
1137
1138                 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1139                 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1140
1141                 memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
1142                 virtio_send_command(hw->cvq, &ctrl, &len, 1);
1143         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1144                 virtio_set_hwaddr(hw);
1145 }
1146
1147 static int
1148 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1149 {
1150         struct virtio_hw *hw = dev->data->dev_private;
1151         struct virtio_pmd_ctrl ctrl;
1152         int len;
1153
1154         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1155                 return -ENOTSUP;
1156
1157         ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1158         ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1159         memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1160         len = sizeof(vlan_id);
1161
1162         return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1163 }
1164
1165 static int
1166 virtio_intr_enable(struct rte_eth_dev *dev)
1167 {
1168         struct virtio_hw *hw = dev->data->dev_private;
1169
1170         if (rte_intr_enable(dev->intr_handle) < 0)
1171                 return -1;
1172
1173         if (!hw->virtio_user_dev)
1174                 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1175
1176         return 0;
1177 }
1178
1179 static int
1180 virtio_intr_disable(struct rte_eth_dev *dev)
1181 {
1182         struct virtio_hw *hw = dev->data->dev_private;
1183
1184         if (rte_intr_disable(dev->intr_handle) < 0)
1185                 return -1;
1186
1187         if (!hw->virtio_user_dev)
1188                 hw->use_msix = vtpci_msix_detect(RTE_ETH_DEV_TO_PCI(dev));
1189
1190         return 0;
1191 }
1192
1193 static int
1194 virtio_negotiate_features(struct virtio_hw *hw, uint64_t req_features)
1195 {
1196         uint64_t host_features;
1197
1198         /* Prepare guest_features: feature that driver wants to support */
1199         PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1200                 req_features);
1201
1202         /* Read device(host) feature bits */
1203         host_features = VTPCI_OPS(hw)->get_features(hw);
1204         PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1205                 host_features);
1206
1207         /* If supported, ensure MTU value is valid before acknowledging it. */
1208         if (host_features & req_features & (1ULL << VIRTIO_NET_F_MTU)) {
1209                 struct virtio_net_config config;
1210
1211                 vtpci_read_dev_config(hw,
1212                         offsetof(struct virtio_net_config, mtu),
1213                         &config.mtu, sizeof(config.mtu));
1214
1215                 if (config.mtu < ETHER_MIN_MTU)
1216                         req_features &= ~(1ULL << VIRTIO_NET_F_MTU);
1217         }
1218
1219         /*
1220          * Negotiate features: Subset of device feature bits are written back
1221          * guest feature bits.
1222          */
1223         hw->guest_features = req_features;
1224         hw->guest_features = vtpci_negotiate_features(hw, host_features);
1225         PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1226                 hw->guest_features);
1227
1228         if (hw->modern) {
1229                 if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1230                         PMD_INIT_LOG(ERR,
1231                                 "VIRTIO_F_VERSION_1 features is not enabled.");
1232                         return -1;
1233                 }
1234                 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1235                 if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1236                         PMD_INIT_LOG(ERR,
1237                                 "failed to set FEATURES_OK status!");
1238                         return -1;
1239                 }
1240         }
1241
1242         hw->req_guest_features = req_features;
1243
1244         return 0;
1245 }
1246
1247 /*
1248  * Process Virtio Config changed interrupt and call the callback
1249  * if link state changed.
1250  */
1251 void
1252 virtio_interrupt_handler(void *param)
1253 {
1254         struct rte_eth_dev *dev = param;
1255         struct virtio_hw *hw = dev->data->dev_private;
1256         uint8_t isr;
1257
1258         /* Read interrupt status which clears interrupt */
1259         isr = vtpci_isr(hw);
1260         PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1261
1262         if (virtio_intr_enable(dev) < 0)
1263                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1264
1265         if (isr & VIRTIO_PCI_ISR_CONFIG) {
1266                 if (virtio_dev_link_update(dev, 0) == 0)
1267                         _rte_eth_dev_callback_process(dev,
1268                                                       RTE_ETH_EVENT_INTR_LSC,
1269                                                       NULL, NULL);
1270         }
1271
1272 }
1273
1274 /* set rx and tx handlers according to what is supported */
1275 static void
1276 set_rxtx_funcs(struct rte_eth_dev *eth_dev)
1277 {
1278         struct virtio_hw *hw = eth_dev->data->dev_private;
1279
1280         if (hw->use_simple_rx) {
1281                 PMD_INIT_LOG(INFO, "virtio: using simple Rx path on port %u",
1282                         eth_dev->data->port_id);
1283                 eth_dev->rx_pkt_burst = virtio_recv_pkts_vec;
1284         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1285                 PMD_INIT_LOG(INFO,
1286                         "virtio: using mergeable buffer Rx path on port %u",
1287                         eth_dev->data->port_id);
1288                 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1289         } else {
1290                 PMD_INIT_LOG(INFO, "virtio: using standard Rx path on port %u",
1291                         eth_dev->data->port_id);
1292                 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1293         }
1294
1295         if (hw->use_simple_tx) {
1296                 PMD_INIT_LOG(INFO, "virtio: using simple Tx path on port %u",
1297                         eth_dev->data->port_id);
1298                 eth_dev->tx_pkt_burst = virtio_xmit_pkts_simple;
1299         } else {
1300                 PMD_INIT_LOG(INFO, "virtio: using standard Tx path on port %u",
1301                         eth_dev->data->port_id);
1302                 eth_dev->tx_pkt_burst = virtio_xmit_pkts;
1303         }
1304 }
1305
1306 /* Only support 1:1 queue/interrupt mapping so far.
1307  * TODO: support n:1 queue/interrupt mapping when there are limited number of
1308  * interrupt vectors (<N+1).
1309  */
1310 static int
1311 virtio_queues_bind_intr(struct rte_eth_dev *dev)
1312 {
1313         uint32_t i;
1314         struct virtio_hw *hw = dev->data->dev_private;
1315
1316         PMD_INIT_LOG(INFO, "queue/interrupt binding");
1317         for (i = 0; i < dev->data->nb_rx_queues; ++i) {
1318                 dev->intr_handle->intr_vec[i] = i + 1;
1319                 if (VTPCI_OPS(hw)->set_queue_irq(hw, hw->vqs[i * 2], i + 1) ==
1320                                                  VIRTIO_MSI_NO_VECTOR) {
1321                         PMD_DRV_LOG(ERR, "failed to set queue vector");
1322                         return -EBUSY;
1323                 }
1324         }
1325
1326         return 0;
1327 }
1328
1329 static void
1330 virtio_queues_unbind_intr(struct rte_eth_dev *dev)
1331 {
1332         uint32_t i;
1333         struct virtio_hw *hw = dev->data->dev_private;
1334
1335         PMD_INIT_LOG(INFO, "queue/interrupt unbinding");
1336         for (i = 0; i < dev->data->nb_rx_queues; ++i)
1337                 VTPCI_OPS(hw)->set_queue_irq(hw,
1338                                              hw->vqs[i * VTNET_CQ],
1339                                              VIRTIO_MSI_NO_VECTOR);
1340 }
1341
1342 static int
1343 virtio_configure_intr(struct rte_eth_dev *dev)
1344 {
1345         struct virtio_hw *hw = dev->data->dev_private;
1346
1347         if (!rte_intr_cap_multiple(dev->intr_handle)) {
1348                 PMD_INIT_LOG(ERR, "Multiple intr vector not supported");
1349                 return -ENOTSUP;
1350         }
1351
1352         if (rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues)) {
1353                 PMD_INIT_LOG(ERR, "Fail to create eventfd");
1354                 return -1;
1355         }
1356
1357         if (!dev->intr_handle->intr_vec) {
1358                 dev->intr_handle->intr_vec =
1359                         rte_zmalloc("intr_vec",
1360                                     hw->max_queue_pairs * sizeof(int), 0);
1361                 if (!dev->intr_handle->intr_vec) {
1362                         PMD_INIT_LOG(ERR, "Failed to allocate %u rxq vectors",
1363                                      hw->max_queue_pairs);
1364                         return -ENOMEM;
1365                 }
1366         }
1367
1368         /* Re-register callback to update max_intr */
1369         rte_intr_callback_unregister(dev->intr_handle,
1370                                      virtio_interrupt_handler,
1371                                      dev);
1372         rte_intr_callback_register(dev->intr_handle,
1373                                    virtio_interrupt_handler,
1374                                    dev);
1375
1376         /* DO NOT try to remove this! This function will enable msix, or QEMU
1377          * will encounter SIGSEGV when DRIVER_OK is sent.
1378          * And for legacy devices, this should be done before queue/vec binding
1379          * to change the config size from 20 to 24, or VIRTIO_MSI_QUEUE_VECTOR
1380          * (22) will be ignored.
1381          */
1382         if (virtio_intr_enable(dev) < 0) {
1383                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1384                 return -1;
1385         }
1386
1387         if (virtio_queues_bind_intr(dev) < 0) {
1388                 PMD_INIT_LOG(ERR, "Failed to bind queue/interrupt");
1389                 return -1;
1390         }
1391
1392         return 0;
1393 }
1394
1395 /* reset device and renegotiate features if needed */
1396 static int
1397 virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
1398 {
1399         struct virtio_hw *hw = eth_dev->data->dev_private;
1400         struct virtio_net_config *config;
1401         struct virtio_net_config local_config;
1402         struct rte_pci_device *pci_dev = NULL;
1403         int ret;
1404
1405         /* Reset the device although not necessary at startup */
1406         vtpci_reset(hw);
1407
1408         /* Tell the host we've noticed this device. */
1409         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1410
1411         /* Tell the host we've known how to drive the device. */
1412         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1413         if (virtio_negotiate_features(hw, req_features) < 0)
1414                 return -1;
1415
1416         if (!hw->virtio_user_dev) {
1417                 pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1418                 rte_eth_copy_pci_info(eth_dev, pci_dev);
1419         }
1420
1421         /* If host does not support both status and MSI-X then disable LSC */
1422         if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS) &&
1423             hw->use_msix != VIRTIO_MSIX_NONE)
1424                 eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
1425         else
1426                 eth_dev->data->dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1427
1428         /* Setting up rx_header size for the device */
1429         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1430             vtpci_with_feature(hw, VIRTIO_F_VERSION_1))
1431                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1432         else
1433                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1434
1435         /* Copy the permanent MAC address to: virtio_hw */
1436         virtio_get_hwaddr(hw);
1437         ether_addr_copy((struct ether_addr *) hw->mac_addr,
1438                         &eth_dev->data->mac_addrs[0]);
1439         PMD_INIT_LOG(DEBUG,
1440                      "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1441                      hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1442                      hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1443
1444         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1445                 config = &local_config;
1446
1447                 vtpci_read_dev_config(hw,
1448                         offsetof(struct virtio_net_config, mac),
1449                         &config->mac, sizeof(config->mac));
1450
1451                 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1452                         vtpci_read_dev_config(hw,
1453                                 offsetof(struct virtio_net_config, status),
1454                                 &config->status, sizeof(config->status));
1455                 } else {
1456                         PMD_INIT_LOG(DEBUG,
1457                                      "VIRTIO_NET_F_STATUS is not supported");
1458                         config->status = 0;
1459                 }
1460
1461                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1462                         vtpci_read_dev_config(hw,
1463                                 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1464                                 &config->max_virtqueue_pairs,
1465                                 sizeof(config->max_virtqueue_pairs));
1466                 } else {
1467                         PMD_INIT_LOG(DEBUG,
1468                                      "VIRTIO_NET_F_MQ is not supported");
1469                         config->max_virtqueue_pairs = 1;
1470                 }
1471
1472                 hw->max_queue_pairs = config->max_virtqueue_pairs;
1473
1474                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MTU)) {
1475                         vtpci_read_dev_config(hw,
1476                                 offsetof(struct virtio_net_config, mtu),
1477                                 &config->mtu,
1478                                 sizeof(config->mtu));
1479
1480                         /*
1481                          * MTU value has already been checked at negotiation
1482                          * time, but check again in case it has changed since
1483                          * then, which should not happen.
1484                          */
1485                         if (config->mtu < ETHER_MIN_MTU) {
1486                                 PMD_INIT_LOG(ERR, "invalid max MTU value (%u)",
1487                                                 config->mtu);
1488                                 return -1;
1489                         }
1490
1491                         hw->max_mtu = config->mtu;
1492                         /* Set initial MTU to maximum one supported by vhost */
1493                         eth_dev->data->mtu = config->mtu;
1494
1495                 } else {
1496                         hw->max_mtu = VIRTIO_MAX_RX_PKTLEN - ETHER_HDR_LEN -
1497                                 VLAN_TAG_LEN - hw->vtnet_hdr_size;
1498                 }
1499
1500                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1501                                 config->max_virtqueue_pairs);
1502                 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1503                 PMD_INIT_LOG(DEBUG,
1504                                 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1505                                 config->mac[0], config->mac[1],
1506                                 config->mac[2], config->mac[3],
1507                                 config->mac[4], config->mac[5]);
1508         } else {
1509                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=1");
1510                 hw->max_queue_pairs = 1;
1511         }
1512
1513         ret = virtio_alloc_queues(eth_dev);
1514         if (ret < 0)
1515                 return ret;
1516
1517         if (eth_dev->data->dev_conf.intr_conf.rxq) {
1518                 if (virtio_configure_intr(eth_dev) < 0) {
1519                         PMD_INIT_LOG(ERR, "failed to configure interrupt");
1520                         return -1;
1521                 }
1522         }
1523
1524         vtpci_reinit_complete(hw);
1525
1526         if (pci_dev)
1527                 PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1528                         eth_dev->data->port_id, pci_dev->id.vendor_id,
1529                         pci_dev->id.device_id);
1530
1531         return 0;
1532 }
1533
1534 /*
1535  * Remap the PCI device again (IO port map for legacy device and
1536  * memory map for modern device), so that the secondary process
1537  * could have the PCI initiated correctly.
1538  */
1539 static int
1540 virtio_remap_pci(struct rte_pci_device *pci_dev, struct virtio_hw *hw)
1541 {
1542         if (hw->modern) {
1543                 /*
1544                  * We don't have to re-parse the PCI config space, since
1545                  * rte_pci_map_device() makes sure the mapped address
1546                  * in secondary process would equal to the one mapped in
1547                  * the primary process: error will be returned if that
1548                  * requirement is not met.
1549                  *
1550                  * That said, we could simply reuse all cap pointers
1551                  * (such as dev_cfg, common_cfg, etc.) parsed from the
1552                  * primary process, which is stored in shared memory.
1553                  */
1554                 if (rte_pci_map_device(pci_dev)) {
1555                         PMD_INIT_LOG(DEBUG, "failed to map pci device!");
1556                         return -1;
1557                 }
1558         } else {
1559                 if (rte_pci_ioport_map(pci_dev, 0, VTPCI_IO(hw)) < 0)
1560                         return -1;
1561         }
1562
1563         return 0;
1564 }
1565
1566 static void
1567 virtio_set_vtpci_ops(struct virtio_hw *hw)
1568 {
1569 #ifdef RTE_VIRTIO_USER
1570         if (hw->virtio_user_dev)
1571                 VTPCI_OPS(hw) = &virtio_user_ops;
1572         else
1573 #endif
1574         if (hw->modern)
1575                 VTPCI_OPS(hw) = &modern_ops;
1576         else
1577                 VTPCI_OPS(hw) = &legacy_ops;
1578 }
1579
1580 /*
1581  * This function is based on probe() function in virtio_pci.c
1582  * It returns 0 on success.
1583  */
1584 int
1585 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1586 {
1587         struct virtio_hw *hw = eth_dev->data->dev_private;
1588         int ret;
1589
1590         RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr_mrg_rxbuf));
1591
1592         eth_dev->dev_ops = &virtio_eth_dev_ops;
1593
1594         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1595                 if (!hw->virtio_user_dev) {
1596                         ret = virtio_remap_pci(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1597                         if (ret)
1598                                 return ret;
1599                 }
1600
1601                 virtio_set_vtpci_ops(hw);
1602                 set_rxtx_funcs(eth_dev);
1603
1604                 return 0;
1605         }
1606
1607         /* Allocate memory for storing MAC addresses */
1608         eth_dev->data->mac_addrs = rte_zmalloc("virtio", VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN, 0);
1609         if (eth_dev->data->mac_addrs == NULL) {
1610                 PMD_INIT_LOG(ERR,
1611                         "Failed to allocate %d bytes needed to store MAC addresses",
1612                         VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN);
1613                 return -ENOMEM;
1614         }
1615
1616         hw->port_id = eth_dev->data->port_id;
1617         /* For virtio_user case the hw->virtio_user_dev is populated by
1618          * virtio_user_eth_dev_alloc() before eth_virtio_dev_init() is called.
1619          */
1620         if (!hw->virtio_user_dev) {
1621                 ret = vtpci_init(RTE_ETH_DEV_TO_PCI(eth_dev), hw);
1622                 if (ret)
1623                         goto out;
1624         }
1625
1626         /* reset device and negotiate default features */
1627         ret = virtio_init_device(eth_dev, VIRTIO_PMD_DEFAULT_GUEST_FEATURES);
1628         if (ret < 0)
1629                 goto out;
1630
1631         /* Setup interrupt callback  */
1632         if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1633                 rte_intr_callback_register(eth_dev->intr_handle,
1634                         virtio_interrupt_handler, eth_dev);
1635
1636         return 0;
1637
1638 out:
1639         rte_free(eth_dev->data->mac_addrs);
1640         return ret;
1641 }
1642
1643 static int
1644 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1645 {
1646         PMD_INIT_FUNC_TRACE();
1647
1648         if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1649                 return -EPERM;
1650
1651         virtio_dev_stop(eth_dev);
1652         virtio_dev_close(eth_dev);
1653
1654         eth_dev->dev_ops = NULL;
1655         eth_dev->tx_pkt_burst = NULL;
1656         eth_dev->rx_pkt_burst = NULL;
1657
1658         rte_free(eth_dev->data->mac_addrs);
1659         eth_dev->data->mac_addrs = NULL;
1660
1661         /* reset interrupt callback  */
1662         if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1663                 rte_intr_callback_unregister(eth_dev->intr_handle,
1664                                                 virtio_interrupt_handler,
1665                                                 eth_dev);
1666         if (eth_dev->device)
1667                 rte_pci_unmap_device(RTE_ETH_DEV_TO_PCI(eth_dev));
1668
1669         PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1670
1671         return 0;
1672 }
1673
1674 static int eth_virtio_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1675         struct rte_pci_device *pci_dev)
1676 {
1677         return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct virtio_hw),
1678                 eth_virtio_dev_init);
1679 }
1680
1681 static int eth_virtio_pci_remove(struct rte_pci_device *pci_dev)
1682 {
1683         return rte_eth_dev_pci_generic_remove(pci_dev, eth_virtio_dev_uninit);
1684 }
1685
1686 static struct rte_pci_driver rte_virtio_pmd = {
1687         .driver = {
1688                 .name = "net_virtio",
1689         },
1690         .id_table = pci_id_virtio_map,
1691         .drv_flags = 0,
1692         .probe = eth_virtio_pci_probe,
1693         .remove = eth_virtio_pci_remove,
1694 };
1695
1696 RTE_INIT(rte_virtio_pmd_init);
1697 static void
1698 rte_virtio_pmd_init(void)
1699 {
1700         if (rte_eal_iopl_init() != 0) {
1701                 PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1702                 return;
1703         }
1704
1705         rte_pci_register(&rte_virtio_pmd);
1706 }
1707
1708 /*
1709  * Configure virtio device
1710  * It returns 0 on success.
1711  */
1712 static int
1713 virtio_dev_configure(struct rte_eth_dev *dev)
1714 {
1715         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1716         struct virtio_hw *hw = dev->data->dev_private;
1717         uint64_t req_features;
1718         int ret;
1719
1720         PMD_INIT_LOG(DEBUG, "configure");
1721         req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
1722
1723         if (dev->data->dev_conf.intr_conf.rxq) {
1724                 ret = virtio_init_device(dev, hw->req_guest_features);
1725                 if (ret < 0)
1726                         return ret;
1727         }
1728
1729         /* The name hw_ip_checksum is a bit confusing since it can be
1730          * set by the application to request L3 and/or L4 checksums. In
1731          * case of virtio, only L4 checksum is supported.
1732          */
1733         if (rxmode->hw_ip_checksum)
1734                 req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
1735
1736         if (rxmode->enable_lro)
1737                 req_features |=
1738                         (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
1739                         (1ULL << VIRTIO_NET_F_GUEST_TSO6);
1740
1741         /* if request features changed, reinit the device */
1742         if (req_features != hw->req_guest_features) {
1743                 ret = virtio_init_device(dev, req_features);
1744                 if (ret < 0)
1745                         return ret;
1746         }
1747
1748         if (rxmode->hw_ip_checksum &&
1749                 !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
1750                 PMD_DRV_LOG(ERR,
1751                         "rx checksum not available on this host");
1752                 return -ENOTSUP;
1753         }
1754
1755         if (rxmode->enable_lro &&
1756                 (!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
1757                         !vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4))) {
1758                 PMD_DRV_LOG(ERR,
1759                         "Large Receive Offload not available on this host");
1760                 return -ENOTSUP;
1761         }
1762
1763         /* start control queue */
1764         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ))
1765                 virtio_dev_cq_start(dev);
1766
1767         hw->vlan_strip = rxmode->hw_vlan_strip;
1768
1769         if (rxmode->hw_vlan_filter
1770             && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1771                 PMD_DRV_LOG(ERR,
1772                             "vlan filtering not available on this host");
1773                 return -ENOTSUP;
1774         }
1775
1776         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1777                 /* Enable vector (0) for Link State Intrerrupt */
1778                 if (VTPCI_OPS(hw)->set_config_irq(hw, 0) ==
1779                                 VIRTIO_MSI_NO_VECTOR) {
1780                         PMD_DRV_LOG(ERR, "failed to set config vector");
1781                         return -EBUSY;
1782                 }
1783
1784         hw->use_simple_rx = 1;
1785         hw->use_simple_tx = 1;
1786
1787 #if defined RTE_ARCH_ARM64 || defined RTE_ARCH_ARM
1788         if (!rte_cpu_get_flag_enabled(RTE_CPUFLAG_NEON)) {
1789                 hw->use_simple_rx = 0;
1790                 hw->use_simple_tx = 0;
1791         }
1792 #endif
1793         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF)) {
1794                 hw->use_simple_rx = 0;
1795                 hw->use_simple_tx = 0;
1796         }
1797
1798         if (rxmode->hw_ip_checksum)
1799                 hw->use_simple_rx = 0;
1800
1801         return 0;
1802 }
1803
1804
1805 static int
1806 virtio_dev_start(struct rte_eth_dev *dev)
1807 {
1808         uint16_t nb_queues, i;
1809         struct virtnet_rx *rxvq;
1810         struct virtnet_tx *txvq __rte_unused;
1811         struct virtio_hw *hw = dev->data->dev_private;
1812         int ret;
1813
1814         /* Finish the initialization of the queues */
1815         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1816                 ret = virtio_dev_rx_queue_setup_finish(dev, i);
1817                 if (ret < 0)
1818                         return ret;
1819         }
1820         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1821                 ret = virtio_dev_tx_queue_setup_finish(dev, i);
1822                 if (ret < 0)
1823                         return ret;
1824         }
1825
1826         /* check if lsc interrupt feature is enabled */
1827         if (dev->data->dev_conf.intr_conf.lsc) {
1828                 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
1829                         PMD_DRV_LOG(ERR, "link status not supported by host");
1830                         return -ENOTSUP;
1831                 }
1832         }
1833
1834         /* Enable uio/vfio intr/eventfd mapping: althrough we already did that
1835          * in device configure, but it could be unmapped  when device is
1836          * stopped.
1837          */
1838         if (dev->data->dev_conf.intr_conf.lsc ||
1839             dev->data->dev_conf.intr_conf.rxq) {
1840                 virtio_intr_disable(dev);
1841
1842                 if (virtio_intr_enable(dev) < 0) {
1843                         PMD_DRV_LOG(ERR, "interrupt enable failed");
1844                         return -EIO;
1845                 }
1846         }
1847
1848         /*Notify the backend
1849          *Otherwise the tap backend might already stop its queue due to fullness.
1850          *vhost backend will have no chance to be waked up
1851          */
1852         nb_queues = RTE_MAX(dev->data->nb_rx_queues, dev->data->nb_tx_queues);
1853         if (hw->max_queue_pairs > 1) {
1854                 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1855                         return -EINVAL;
1856         }
1857
1858         PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1859
1860         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1861                 rxvq = dev->data->rx_queues[i];
1862                 /* Flush the old packets */
1863                 virtqueue_flush(rxvq->vq);
1864                 virtqueue_notify(rxvq->vq);
1865         }
1866
1867         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1868                 txvq = dev->data->tx_queues[i];
1869                 virtqueue_notify(txvq->vq);
1870         }
1871
1872         PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1873
1874         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1875                 rxvq = dev->data->rx_queues[i];
1876                 VIRTQUEUE_DUMP(rxvq->vq);
1877         }
1878
1879         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1880                 txvq = dev->data->tx_queues[i];
1881                 VIRTQUEUE_DUMP(txvq->vq);
1882         }
1883
1884         set_rxtx_funcs(dev);
1885         hw->started = 1;
1886
1887         /* Initialize Link state */
1888         virtio_dev_link_update(dev, 0);
1889
1890         return 0;
1891 }
1892
1893 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1894 {
1895         struct rte_mbuf *buf;
1896         int i, mbuf_num = 0;
1897
1898         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1899                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1900
1901                 PMD_INIT_LOG(DEBUG,
1902                              "Before freeing rxq[%d] used and unused buf", i);
1903                 VIRTQUEUE_DUMP(rxvq->vq);
1904
1905                 PMD_INIT_LOG(DEBUG, "rx_queues[%d]=%p", i, rxvq);
1906                 while ((buf = virtqueue_detatch_unused(rxvq->vq)) != NULL) {
1907                         rte_pktmbuf_free(buf);
1908                         mbuf_num++;
1909                 }
1910
1911                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1912                 PMD_INIT_LOG(DEBUG,
1913                              "After freeing rxq[%d] used and unused buf", i);
1914                 VIRTQUEUE_DUMP(rxvq->vq);
1915         }
1916
1917         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1918                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1919
1920                 PMD_INIT_LOG(DEBUG,
1921                              "Before freeing txq[%d] used and unused bufs",
1922                              i);
1923                 VIRTQUEUE_DUMP(txvq->vq);
1924
1925                 mbuf_num = 0;
1926                 while ((buf = virtqueue_detatch_unused(txvq->vq)) != NULL) {
1927                         rte_pktmbuf_free(buf);
1928                         mbuf_num++;
1929                 }
1930
1931                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1932                 PMD_INIT_LOG(DEBUG,
1933                              "After freeing txq[%d] used and unused buf", i);
1934                 VIRTQUEUE_DUMP(txvq->vq);
1935         }
1936 }
1937
1938 /*
1939  * Stop device: disable interrupt and mark link down
1940  */
1941 static void
1942 virtio_dev_stop(struct rte_eth_dev *dev)
1943 {
1944         struct virtio_hw *hw = dev->data->dev_private;
1945         struct rte_eth_link link;
1946         struct rte_intr_conf *intr_conf = &dev->data->dev_conf.intr_conf;
1947
1948         PMD_INIT_LOG(DEBUG, "stop");
1949
1950         if (intr_conf->lsc || intr_conf->rxq)
1951                 virtio_intr_disable(dev);
1952
1953         hw->started = 0;
1954         memset(&link, 0, sizeof(link));
1955         virtio_dev_atomic_write_link_status(dev, &link);
1956 }
1957
1958 static int
1959 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
1960 {
1961         struct rte_eth_link link, old;
1962         uint16_t status;
1963         struct virtio_hw *hw = dev->data->dev_private;
1964         memset(&link, 0, sizeof(link));
1965         virtio_dev_atomic_read_link_status(dev, &link);
1966         old = link;
1967         link.link_duplex = ETH_LINK_FULL_DUPLEX;
1968         link.link_speed  = SPEED_10G;
1969
1970         if (hw->started == 0) {
1971                 link.link_status = ETH_LINK_DOWN;
1972         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1973                 PMD_INIT_LOG(DEBUG, "Get link status from hw");
1974                 vtpci_read_dev_config(hw,
1975                                 offsetof(struct virtio_net_config, status),
1976                                 &status, sizeof(status));
1977                 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
1978                         link.link_status = ETH_LINK_DOWN;
1979                         PMD_INIT_LOG(DEBUG, "Port %d is down",
1980                                      dev->data->port_id);
1981                 } else {
1982                         link.link_status = ETH_LINK_UP;
1983                         PMD_INIT_LOG(DEBUG, "Port %d is up",
1984                                      dev->data->port_id);
1985                 }
1986         } else {
1987                 link.link_status = ETH_LINK_UP;
1988         }
1989         virtio_dev_atomic_write_link_status(dev, &link);
1990
1991         return (old.link_status == link.link_status) ? -1 : 0;
1992 }
1993
1994 static int
1995 virtio_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
1996 {
1997         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1998         struct virtio_hw *hw = dev->data->dev_private;
1999
2000         if (mask & ETH_VLAN_FILTER_MASK) {
2001                 if (rxmode->hw_vlan_filter &&
2002                                 !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
2003
2004                         PMD_DRV_LOG(NOTICE,
2005                                 "vlan filtering not available on this host");
2006
2007                         return -ENOTSUP;
2008                 }
2009         }
2010
2011         if (mask & ETH_VLAN_STRIP_MASK)
2012                 hw->vlan_strip = rxmode->hw_vlan_strip;
2013
2014         return 0;
2015 }
2016
2017 static void
2018 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
2019 {
2020         uint64_t tso_mask, host_features;
2021         struct virtio_hw *hw = dev->data->dev_private;
2022
2023         dev_info->speed_capa = ETH_LINK_SPEED_10G; /* fake value */
2024
2025         dev_info->pci_dev = dev->device ? RTE_ETH_DEV_TO_PCI(dev) : NULL;
2026         dev_info->max_rx_queues =
2027                 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_RX_QUEUES);
2028         dev_info->max_tx_queues =
2029                 RTE_MIN(hw->max_queue_pairs, VIRTIO_MAX_TX_QUEUES);
2030         dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
2031         dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
2032         dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
2033         dev_info->default_txconf = (struct rte_eth_txconf) {
2034                 .txq_flags = ETH_TXQ_FLAGS_NOOFFLOADS
2035         };
2036
2037         host_features = VTPCI_OPS(hw)->get_features(hw);
2038         dev_info->rx_offload_capa = 0;
2039         if (host_features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
2040                 dev_info->rx_offload_capa |=
2041                         DEV_RX_OFFLOAD_TCP_CKSUM |
2042                         DEV_RX_OFFLOAD_UDP_CKSUM;
2043         }
2044         tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
2045                 (1ULL << VIRTIO_NET_F_GUEST_TSO6);
2046         if ((host_features & tso_mask) == tso_mask)
2047                 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
2048
2049         dev_info->tx_offload_capa = 0;
2050         if (hw->guest_features & (1ULL << VIRTIO_NET_F_CSUM)) {
2051                 dev_info->tx_offload_capa |=
2052                         DEV_TX_OFFLOAD_UDP_CKSUM |
2053                         DEV_TX_OFFLOAD_TCP_CKSUM;
2054         }
2055         tso_mask = (1ULL << VIRTIO_NET_F_HOST_TSO4) |
2056                 (1ULL << VIRTIO_NET_F_HOST_TSO6);
2057         if ((hw->guest_features & tso_mask) == tso_mask)
2058                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
2059 }
2060
2061 /*
2062  * It enables testpmd to collect per queue stats.
2063  */
2064 static int
2065 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
2066 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
2067 __rte_unused uint8_t is_rx)
2068 {
2069         return 0;
2070 }
2071
2072 RTE_PMD_EXPORT_NAME(net_virtio, __COUNTER__);
2073 RTE_PMD_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);
2074 RTE_PMD_REGISTER_KMOD_DEP(net_virtio, "* igb_uio | uio_pci_generic | vfio-pci");