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