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_init(struct rte_eth_dev *eth_dev);
  63 static int eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev);
  64 static int  virtio_dev_configure(struct rte_eth_dev *dev);
  65 static int  virtio_dev_start(struct rte_eth_dev *dev);
  66 static void virtio_dev_stop(struct rte_eth_dev *dev);
  67 static void virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
  68 static void virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
  69 static void virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
  70 static void virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
  71 static void virtio_dev_info_get(struct rte_eth_dev *dev,
  72                                 struct rte_eth_dev_info *dev_info);
  73 static int virtio_dev_link_update(struct rte_eth_dev *dev,
  74         __rte_unused int wait_to_complete);
  75
  76 static void virtio_set_hwaddr(struct virtio_hw *hw);
  77 static void virtio_get_hwaddr(struct virtio_hw *hw);
  78
  79 static void virtio_dev_stats_get(struct rte_eth_dev *dev,
  80                                  struct rte_eth_stats *stats);
  81 static int virtio_dev_xstats_get(struct rte_eth_dev *dev,
  82                                  struct rte_eth_xstat *xstats, unsigned n);
  83 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
  84                                        struct rte_eth_xstat_name *xstats_names,
  85                                        unsigned limit);
  86 static void virtio_dev_stats_reset(struct rte_eth_dev *dev);
  87 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
  88 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
  89                                 uint16_t vlan_id, int on);
  90 static void virtio_mac_addr_add(struct rte_eth_dev *dev,
  91                                 struct ether_addr *mac_addr,
  92                                 uint32_t index, uint32_t vmdq __rte_unused);
  93 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
  94 static void virtio_mac_addr_set(struct rte_eth_dev *dev,
  95                                 struct ether_addr *mac_addr);
  96
  97 static int virtio_dev_queue_stats_mapping_set(
  98         __rte_unused struct rte_eth_dev *eth_dev,
  99         __rte_unused uint16_t queue_id,
 100         __rte_unused uint8_t stat_idx,
 101         __rte_unused uint8_t is_rx);
 102
 103 /*
 104  * The set of PCI devices this driver supports
 105  */
 106 static const struct rte_pci_id pci_id_virtio_map[] = {
 107
 108 #define RTE_PCI_DEV_ID_DECL_VIRTIO(vend, dev) {RTE_PCI_DEVICE(vend, dev)},
 109 #include "rte_pci_dev_ids.h"
 110
 111 { .vendor_id = 0, /* sentinel */ },
 112 };
 113
 114 struct rte_virtio_xstats_name_off {
 115         char name[RTE_ETH_XSTATS_NAME_SIZE];
 116         unsigned offset;
 117 };
 118
 119 /* [rt]x_qX_ is prepended to the name string here */
 120 static const struct rte_virtio_xstats_name_off rte_virtio_q_stat_strings[] = {
 121         {"good_packets",           offsetof(struct virtqueue, packets)},
 122         {"good_bytes",             offsetof(struct virtqueue, bytes)},
 123         {"errors",                 offsetof(struct virtqueue, errors)},
 124         {"multicast_packets",      offsetof(struct virtqueue, multicast)},
 125         {"broadcast_packets",      offsetof(struct virtqueue, broadcast)},
 126         {"undersize_packets",      offsetof(struct virtqueue, size_bins[0])},
 127         {"size_64_packets",        offsetof(struct virtqueue, size_bins[1])},
 128         {"size_65_127_packets",    offsetof(struct virtqueue, size_bins[2])},
 129         {"size_128_255_packets",   offsetof(struct virtqueue, size_bins[3])},
 130         {"size_256_511_packets",   offsetof(struct virtqueue, size_bins[4])},
 131         {"size_512_1023_packets",  offsetof(struct virtqueue, size_bins[5])},
 132         {"size_1024_1517_packets", offsetof(struct virtqueue, size_bins[6])},
 133         {"size_1518_max_packets",  offsetof(struct virtqueue, size_bins[7])},
 134 };
 135
 136 #define VIRTIO_NB_Q_XSTATS (sizeof(rte_virtio_q_stat_strings) / \
 137                             sizeof(rte_virtio_q_stat_strings[0]))
 138
 139 static int
 140 virtio_send_command(struct virtqueue *vq, struct virtio_pmd_ctrl *ctrl,
 141                 int *dlen, int pkt_num)
 142 {
 143         uint32_t head, i;
 144         int k, sum = 0;
 145         virtio_net_ctrl_ack status = ~0;
 146         struct virtio_pmd_ctrl result;
 147
 148         ctrl->status = status;
 149
 150         if (!(vq && vq->hw->cvq)) {
 151                 PMD_INIT_LOG(ERR, "Control queue is not supported.");
 152                 return -1;
 153         }
 154         head = vq->vq_desc_head_idx;
 155
 156         PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
 157                 "vq->hw->cvq = %p vq = %p",
 158                 vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
 159
 160         if ((vq->vq_free_cnt < ((uint32_t)pkt_num + 2)) || (pkt_num < 1))
 161                 return -1;
 162
 163         memcpy(vq->virtio_net_hdr_mz->addr, ctrl,
 164                 sizeof(struct virtio_pmd_ctrl));
 165
 166         /*
 167          * Format is enforced in qemu code:
 168          * One TX packet for header;
 169          * At least one TX packet per argument;
 170          * One RX packet for ACK.
 171          */
 172         vq->vq_ring.desc[head].flags = VRING_DESC_F_NEXT;
 173         vq->vq_ring.desc[head].addr = vq->virtio_net_hdr_mz->phys_addr;
 174         vq->vq_ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
 175         vq->vq_free_cnt--;
 176         i = vq->vq_ring.desc[head].next;
 177
 178         for (k = 0; k < pkt_num; k++) {
 179                 vq->vq_ring.desc[i].flags = VRING_DESC_F_NEXT;
 180                 vq->vq_ring.desc[i].addr = vq->virtio_net_hdr_mz->phys_addr
 181                         + sizeof(struct virtio_net_ctrl_hdr)
 182                         + sizeof(ctrl->status) + sizeof(uint8_t)*sum;
 183                 vq->vq_ring.desc[i].len = dlen[k];
 184                 sum += dlen[k];
 185                 vq->vq_free_cnt--;
 186                 i = vq->vq_ring.desc[i].next;
 187         }
 188
 189         vq->vq_ring.desc[i].flags = VRING_DESC_F_WRITE;
 190         vq->vq_ring.desc[i].addr = vq->virtio_net_hdr_mz->phys_addr
 191                         + sizeof(struct virtio_net_ctrl_hdr);
 192         vq->vq_ring.desc[i].len = sizeof(ctrl->status);
 193         vq->vq_free_cnt--;
 194
 195         vq->vq_desc_head_idx = vq->vq_ring.desc[i].next;
 196
 197         vq_update_avail_ring(vq, head);
 198         vq_update_avail_idx(vq);
 199
 200         PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
 201
 202         virtqueue_notify(vq);
 203
 204         rte_rmb();
 205         while (vq->vq_used_cons_idx == vq->vq_ring.used->idx) {
 206                 rte_rmb();
 207                 usleep(100);
 208         }
 209
 210         while (vq->vq_used_cons_idx != vq->vq_ring.used->idx) {
 211                 uint32_t idx, desc_idx, used_idx;
 212                 struct vring_used_elem *uep;
 213
 214                 used_idx = (uint32_t)(vq->vq_used_cons_idx
 215                                 & (vq->vq_nentries - 1));
 216                 uep = &vq->vq_ring.used->ring[used_idx];
 217                 idx = (uint32_t) uep->id;
 218                 desc_idx = idx;
 219
 220                 while (vq->vq_ring.desc[desc_idx].flags & VRING_DESC_F_NEXT) {
 221                         desc_idx = vq->vq_ring.desc[desc_idx].next;
 222                         vq->vq_free_cnt++;
 223                 }
 224
 225                 vq->vq_ring.desc[desc_idx].next = vq->vq_desc_head_idx;
 226                 vq->vq_desc_head_idx = idx;
 227
 228                 vq->vq_used_cons_idx++;
 229                 vq->vq_free_cnt++;
 230         }
 231
 232         PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
 233                         vq->vq_free_cnt, vq->vq_desc_head_idx);
 234
 235         memcpy(&result, vq->virtio_net_hdr_mz->addr,
 236                         sizeof(struct virtio_pmd_ctrl));
 237
 238         return result.status;
 239 }
 240
 241 static int
 242 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
 243 {
 244         struct virtio_hw *hw = dev->data->dev_private;
 245         struct virtio_pmd_ctrl ctrl;
 246         int dlen[1];
 247         int ret;
 248
 249         ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
 250         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
 251         memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
 252
 253         dlen[0] = sizeof(uint16_t);
 254
 255         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 256         if (ret) {
 257                 PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
 258                           "failed, this is too late now...");
 259                 return -EINVAL;
 260         }
 261
 262         return 0;
 263 }
 264
 265 void
 266 virtio_dev_queue_release(struct virtqueue *vq)
 267 {
 268         struct virtio_hw *hw;
 269
 270         if (vq) {
 271                 hw = vq->hw;
 272                 if (vq->configured)
 273                         hw->vtpci_ops->del_queue(hw, vq);
 274
 275                 rte_memzone_free(vq->mz);
 276                 if (vq->virtio_net_hdr_mz)
 277                         rte_memzone_free(vq->virtio_net_hdr_mz);
 278
 279                 rte_free(vq->sw_ring);
 280                 rte_free(vq);
 281         }
 282 }
 283
 284 int virtio_dev_queue_setup(struct rte_eth_dev *dev,
 285                         int queue_type,
 286                         uint16_t queue_idx,
 287                         uint16_t vtpci_queue_idx,
 288                         uint16_t nb_desc,
 289                         unsigned int socket_id,
 290                         struct virtqueue **pvq)
 291 {
 292         char vq_name[VIRTQUEUE_MAX_NAME_SZ];
 293         const struct rte_memzone *mz;
 294         unsigned int vq_size, size;
 295         struct virtio_hw *hw = dev->data->dev_private;
 296         struct virtqueue *vq = NULL;
 297         const char *queue_names[] = {"rvq", "txq", "cvq"};
 298
 299         PMD_INIT_LOG(DEBUG, "setting up queue: %u", vtpci_queue_idx);
 300
 301         /*
 302          * Read the virtqueue size from the Queue Size field
 303          * Always power of 2 and if 0 virtqueue does not exist
 304          */
 305         vq_size = hw->vtpci_ops->get_queue_num(hw, vtpci_queue_idx);
 306         PMD_INIT_LOG(DEBUG, "vq_size: %u nb_desc:%u", vq_size, nb_desc);
 307         if (vq_size == 0) {
 308                 PMD_INIT_LOG(ERR, "virtqueue does not exist");
 309                 return -EINVAL;
 310         }
 311
 312         if (!rte_is_power_of_2(vq_size)) {
 313                 PMD_INIT_LOG(ERR, "virtqueue size is not powerof 2");
 314                 return -EINVAL;
 315         }
 316
 317         snprintf(vq_name, sizeof(vq_name), "port%d_%s%d",
 318                  dev->data->port_id, queue_names[queue_type], queue_idx);
 319         vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
 320                          vq_size * sizeof(struct vq_desc_extra),
 321                          RTE_CACHE_LINE_SIZE);
 322         if (vq == NULL) {
 323                 PMD_INIT_LOG(ERR, "Can not allocate virtqueue");
 324                 return -ENOMEM;
 325         }
 326
 327         if (queue_type == VTNET_RQ) {
 328                 size_t sz_sw;
 329
 330                 sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
 331                         sizeof(vq->sw_ring[0]);
 332                 vq->sw_ring = rte_zmalloc_socket("rxq->sw_ring", sz_sw,
 333                                                  RTE_CACHE_LINE_SIZE,
 334                                                  socket_id);
 335                 if (!vq->sw_ring) {
 336                         PMD_INIT_LOG(ERR, "Can not allocate RX soft ring");
 337                         virtio_dev_queue_release(vq);
 338                         return -ENOMEM;
 339                 }
 340         }
 341
 342         vq->hw = hw;
 343         vq->port_id = dev->data->port_id;
 344         vq->queue_id = queue_idx;
 345         vq->vq_queue_index = vtpci_queue_idx;
 346         vq->vq_nentries = vq_size;
 347
 348         if (nb_desc == 0 || nb_desc > vq_size)
 349                 nb_desc = vq_size;
 350         vq->vq_free_cnt = nb_desc;
 351
 352         /*
 353          * Reserve a memzone for vring elements
 354          */
 355         size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
 356         vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
 357         PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d", size, vq->vq_ring_size);
 358
 359         mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
 360                 socket_id, 0, VIRTIO_PCI_VRING_ALIGN);
 361         if (mz == NULL) {
 362                 if (rte_errno == EEXIST)
 363                         mz = rte_memzone_lookup(vq_name);
 364                 if (mz == NULL) {
 365                         virtio_dev_queue_release(vq);
 366                         return -ENOMEM;
 367                 }
 368         }
 369
 370         /*
 371          * Virtio PCI device VIRTIO_PCI_QUEUE_PF register is 32bit,
 372          * and only accepts 32 bit page frame number.
 373          * Check if the allocated physical memory exceeds 16TB.
 374          */
 375         if ((mz->phys_addr + vq->vq_ring_size - 1) >> (VIRTIO_PCI_QUEUE_ADDR_SHIFT + 32)) {
 376                 PMD_INIT_LOG(ERR, "vring address shouldn't be above 16TB!");
 377                 virtio_dev_queue_release(vq);
 378                 return -ENOMEM;
 379         }
 380
 381         memset(mz->addr, 0, sizeof(mz->len));
 382         vq->mz = mz;
 383         vq->vq_ring_mem = mz->phys_addr;
 384         vq->vq_ring_virt_mem = mz->addr;
 385         PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%"PRIx64, (uint64_t)mz->phys_addr);
 386         PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%"PRIx64, (uint64_t)(uintptr_t)mz->addr);
 387         vq->virtio_net_hdr_mz  = NULL;
 388         vq->virtio_net_hdr_mem = 0;
 389
 390         if (queue_type == VTNET_TQ) {
 391                 const struct rte_memzone *hdr_mz;
 392                 struct virtio_tx_region *txr;
 393                 unsigned int i;
 394
 395                 /*
 396                  * For each xmit packet, allocate a virtio_net_hdr
 397                  * and indirect ring elements
 398                  */
 399                 snprintf(vq_name, sizeof(vq_name), "port%d_tvq%d_hdrzone",
 400                          dev->data->port_id, queue_idx);
 401                 hdr_mz = rte_memzone_reserve_aligned(vq_name,
 402                                                      vq_size * sizeof(*txr),
 403                                                      socket_id, 0,
 404                                                      RTE_CACHE_LINE_SIZE);
 405                 if (hdr_mz == NULL) {
 406                         if (rte_errno == EEXIST)
 407                                 hdr_mz = rte_memzone_lookup(vq_name);
 408                         if (hdr_mz == NULL) {
 409                                 virtio_dev_queue_release(vq);
 410                                 return -ENOMEM;
 411                         }
 412                 }
 413                 vq->virtio_net_hdr_mz = hdr_mz;
 414                 vq->virtio_net_hdr_mem = hdr_mz->phys_addr;
 415
 416                 txr = hdr_mz->addr;
 417                 memset(txr, 0, vq_size * sizeof(*txr));
 418                 for (i = 0; i < vq_size; i++) {
 419                         struct vring_desc *start_dp = txr[i].tx_indir;
 420
 421                         vring_desc_init(start_dp, RTE_DIM(txr[i].tx_indir));
 422
 423                         /* first indirect descriptor is always the tx header */
 424                         start_dp->addr = vq->virtio_net_hdr_mem
 425                                 + i * sizeof(*txr)
 426                                 + offsetof(struct virtio_tx_region, tx_hdr);
 427
 428                         start_dp->len = vq->hw->vtnet_hdr_size;
 429                         start_dp->flags = VRING_DESC_F_NEXT;
 430                 }
 431
 432         } else if (queue_type == VTNET_CQ) {
 433                 /* Allocate a page for control vq command, data and status */
 434                 snprintf(vq_name, sizeof(vq_name), "port%d_cvq_hdrzone",
 435                         dev->data->port_id);
 436                 vq->virtio_net_hdr_mz = rte_memzone_reserve_aligned(vq_name,
 437                         PAGE_SIZE, socket_id, 0, RTE_CACHE_LINE_SIZE);
 438                 if (vq->virtio_net_hdr_mz == NULL) {
 439                         if (rte_errno == EEXIST)
 440                                 vq->virtio_net_hdr_mz =
 441                                         rte_memzone_lookup(vq_name);
 442                         if (vq->virtio_net_hdr_mz == NULL) {
 443                                 virtio_dev_queue_release(vq);
 444                                 return -ENOMEM;
 445                         }
 446                 }
 447                 vq->virtio_net_hdr_mem =
 448                         vq->virtio_net_hdr_mz->phys_addr;
 449                 memset(vq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
 450         }
 451
 452         hw->vtpci_ops->setup_queue(hw, vq);
 453
 454         vq->configured = 1;
 455         *pvq = vq;
 456         return 0;
 457 }
 458
 459 static int
 460 virtio_dev_cq_queue_setup(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx,
 461                 uint32_t socket_id)
 462 {
 463         struct virtqueue *vq;
 464         int ret;
 465         struct virtio_hw *hw = dev->data->dev_private;
 466
 467         PMD_INIT_FUNC_TRACE();
 468         ret = virtio_dev_queue_setup(dev, VTNET_CQ, VTNET_SQ_CQ_QUEUE_IDX,
 469                         vtpci_queue_idx, 0, socket_id, &vq);
 470         if (ret < 0) {
 471                 PMD_INIT_LOG(ERR, "control vq initialization failed");
 472                 return ret;
 473         }
 474
 475         hw->cvq = vq;
 476         return 0;
 477 }
 478
 479 static void
 480 virtio_free_queues(struct rte_eth_dev *dev)
 481 {
 482         unsigned int i;
 483
 484         for (i = 0; i < dev->data->nb_rx_queues; i++)
 485                 virtio_dev_rx_queue_release(dev->data->rx_queues[i]);
 486
 487         dev->data->nb_rx_queues = 0;
 488
 489         for (i = 0; i < dev->data->nb_tx_queues; i++)
 490                 virtio_dev_tx_queue_release(dev->data->tx_queues[i]);
 491
 492         dev->data->nb_tx_queues = 0;
 493 }
 494
 495 static void
 496 virtio_dev_close(struct rte_eth_dev *dev)
 497 {
 498         struct virtio_hw *hw = dev->data->dev_private;
 499
 500         PMD_INIT_LOG(DEBUG, "virtio_dev_close");
 501
 502         if (hw->started == 1)
 503                 virtio_dev_stop(dev);
 504
 505         /* reset the NIC */
 506         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
 507                 vtpci_irq_config(hw, VIRTIO_MSI_NO_VECTOR);
 508         vtpci_reset(hw);
 509         virtio_dev_free_mbufs(dev);
 510         virtio_free_queues(dev);
 511 }
 512
 513 static void
 514 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
 515 {
 516         struct virtio_hw *hw = dev->data->dev_private;
 517         struct virtio_pmd_ctrl ctrl;
 518         int dlen[1];
 519         int ret;
 520
 521         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 522                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 523                 return;
 524         }
 525
 526         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 527         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 528         ctrl.data[0] = 1;
 529         dlen[0] = 1;
 530
 531         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 532         if (ret)
 533                 PMD_INIT_LOG(ERR, "Failed to enable promisc");
 534 }
 535
 536 static void
 537 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
 538 {
 539         struct virtio_hw *hw = dev->data->dev_private;
 540         struct virtio_pmd_ctrl ctrl;
 541         int dlen[1];
 542         int ret;
 543
 544         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 545                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 546                 return;
 547         }
 548
 549         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 550         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 551         ctrl.data[0] = 0;
 552         dlen[0] = 1;
 553
 554         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 555         if (ret)
 556                 PMD_INIT_LOG(ERR, "Failed to disable promisc");
 557 }
 558
 559 static void
 560 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
 561 {
 562         struct virtio_hw *hw = dev->data->dev_private;
 563         struct virtio_pmd_ctrl ctrl;
 564         int dlen[1];
 565         int ret;
 566
 567         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 568                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 569                 return;
 570         }
 571
 572         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 573         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 574         ctrl.data[0] = 1;
 575         dlen[0] = 1;
 576
 577         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 578         if (ret)
 579                 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
 580 }
 581
 582 static void
 583 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
 584 {
 585         struct virtio_hw *hw = dev->data->dev_private;
 586         struct virtio_pmd_ctrl ctrl;
 587         int dlen[1];
 588         int ret;
 589
 590         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 591                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 592                 return;
 593         }
 594
 595         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 596         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 597         ctrl.data[0] = 0;
 598         dlen[0] = 1;
 599
 600         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 601         if (ret)
 602                 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
 603 }
 604
 605 /*
 606  * dev_ops for virtio, bare necessities for basic operation
 607  */
 608 static const struct eth_dev_ops virtio_eth_dev_ops = {
 609         .dev_configure           = virtio_dev_configure,
 610         .dev_start               = virtio_dev_start,
 611         .dev_stop                = virtio_dev_stop,
 612         .dev_close               = virtio_dev_close,
 613         .promiscuous_enable      = virtio_dev_promiscuous_enable,
 614         .promiscuous_disable     = virtio_dev_promiscuous_disable,
 615         .allmulticast_enable     = virtio_dev_allmulticast_enable,
 616         .allmulticast_disable    = virtio_dev_allmulticast_disable,
 617
 618         .dev_infos_get           = virtio_dev_info_get,
 619         .stats_get               = virtio_dev_stats_get,
 620         .xstats_get              = virtio_dev_xstats_get,
 621         .xstats_get_names        = virtio_dev_xstats_get_names,
 622         .stats_reset             = virtio_dev_stats_reset,
 623         .xstats_reset            = virtio_dev_stats_reset,
 624         .link_update             = virtio_dev_link_update,
 625         .rx_queue_setup          = virtio_dev_rx_queue_setup,
 626         .rx_queue_release        = virtio_dev_rx_queue_release,
 627         .tx_queue_setup          = virtio_dev_tx_queue_setup,
 628         .tx_queue_release        = virtio_dev_tx_queue_release,
 629         /* collect stats per queue */
 630         .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
 631         .vlan_filter_set         = virtio_vlan_filter_set,
 632         .mac_addr_add            = virtio_mac_addr_add,
 633         .mac_addr_remove         = virtio_mac_addr_remove,
 634         .mac_addr_set            = virtio_mac_addr_set,
 635 };
 636
 637 static inline int
 638 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
 639                                 struct rte_eth_link *link)
 640 {
 641         struct rte_eth_link *dst = link;
 642         struct rte_eth_link *src = &(dev->data->dev_link);
 643
 644         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
 645                         *(uint64_t *)src) == 0)
 646                 return -1;
 647
 648         return 0;
 649 }
 650
 651 /**
 652  * Atomically writes the link status information into global
 653  * structure rte_eth_dev.
 654  *
 655  * @param dev
 656  *   - Pointer to the structure rte_eth_dev to read from.
 657  *   - Pointer to the buffer to be saved with the link status.
 658  *
 659  * @return
 660  *   - On success, zero.
 661  *   - On failure, negative value.
 662  */
 663 static inline int
 664 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
 665                 struct rte_eth_link *link)
 666 {
 667         struct rte_eth_link *dst = &(dev->data->dev_link);
 668         struct rte_eth_link *src = link;
 669
 670         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
 671                                         *(uint64_t *)src) == 0)
 672                 return -1;
 673
 674         return 0;
 675 }
 676
 677 static void
 678 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 679 {
 680         unsigned i;
 681
 682         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 683                 const struct virtqueue *txvq = dev->data->tx_queues[i];
 684                 if (txvq == NULL)
 685                         continue;
 686
 687                 stats->opackets += txvq->packets;
 688                 stats->obytes += txvq->bytes;
 689                 stats->oerrors += txvq->errors;
 690
 691                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 692                         stats->q_opackets[i] = txvq->packets;
 693                         stats->q_obytes[i] = txvq->bytes;
 694                 }
 695         }
 696
 697         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 698                 const struct virtqueue *rxvq = dev->data->rx_queues[i];
 699                 if (rxvq == NULL)
 700                         continue;
 701
 702                 stats->ipackets += rxvq->packets;
 703                 stats->ibytes += rxvq->bytes;
 704                 stats->ierrors += rxvq->errors;
 705
 706                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 707                         stats->q_ipackets[i] = rxvq->packets;
 708                         stats->q_ibytes[i] = rxvq->bytes;
 709                 }
 710         }
 711
 712         stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
 713 }
 714
 715 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
 716                                        struct rte_eth_xstat_name *xstats_names,
 717                                        __rte_unused unsigned limit)
 718 {
 719         unsigned i;
 720         unsigned count = 0;
 721         unsigned t;
 722
 723         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_Q_XSTATS +
 724                 dev->data->nb_rx_queues * VIRTIO_NB_Q_XSTATS;
 725
 726         if (xstats_names == NULL) {
 727                 /* Note: limit checked in rte_eth_xstats_names() */
 728
 729                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
 730                         struct virtqueue *rxvq = dev->data->rx_queues[i];
 731                         if (rxvq == NULL)
 732                                 continue;
 733                         for (t = 0; t < VIRTIO_NB_Q_XSTATS; t++) {
 734                                 snprintf(xstats_names[count].name,
 735                                         sizeof(xstats_names[count].name),
 736                                         "rx_q%u_%s", i,
 737                                         rte_virtio_q_stat_strings[t].name);
 738                                 xstats_names[count].id = count;
 739                                 count++;
 740                         }
 741                 }
 742
 743                 for (i = 0; i < dev->data->nb_tx_queues; i++) {
 744                         struct virtqueue *txvq = dev->data->tx_queues[i];
 745                         if (txvq == NULL)
 746                                 continue;
 747                         for (t = 0; t < VIRTIO_NB_Q_XSTATS; t++) {
 748                                 snprintf(xstats_names[count].name,
 749                                         sizeof(xstats_names[count].name),
 750                                         "tx_q%u_%s", i,
 751                                         rte_virtio_q_stat_strings[t].name);
 752                                 xstats_names[count].id = count;
 753                                 count++;
 754                         }
 755                 }
 756                 return count;
 757         }
 758         return nstats;
 759 }
 760
 761 static int
 762 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 763                       unsigned n)
 764 {
 765         unsigned i;
 766         unsigned count = 0;
 767
 768         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_Q_XSTATS +
 769                 dev->data->nb_rx_queues * VIRTIO_NB_Q_XSTATS;
 770
 771         if (n < nstats)
 772                 return nstats;
 773
 774         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 775                 struct virtqueue *rxvq = dev->data->rx_queues[i];
 776
 777                 if (rxvq == NULL)
 778                         continue;
 779
 780                 unsigned t;
 781
 782                 for (t = 0; t < VIRTIO_NB_Q_XSTATS; t++) {
 783                         xstats[count].id = count;
 784                         xstats[count].value = *(uint64_t *)(((char *)rxvq) +
 785                                 rte_virtio_q_stat_strings[t].offset);
 786                         count++;
 787                 }
 788         }
 789
 790         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 791                 struct virtqueue *txvq = dev->data->tx_queues[i];
 792
 793                 if (txvq == NULL)
 794                         continue;
 795
 796                 unsigned t;
 797
 798                 for (t = 0; t < VIRTIO_NB_Q_XSTATS; t++) {
 799                         xstats[count].id = count;
 800                         xstats[count].value = *(uint64_t *)(((char *)txvq) +
 801                                 rte_virtio_q_stat_strings[t].offset);
 802                         count++;
 803                 }
 804         }
 805
 806         return count;
 807 }
 808
 809 static void
 810 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 811 {
 812         virtio_update_stats(dev, stats);
 813 }
 814
 815 static void
 816 virtio_dev_stats_reset(struct rte_eth_dev *dev)
 817 {
 818         unsigned int i;
 819
 820         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 821                 struct virtqueue *txvq = dev->data->tx_queues[i];
 822                 if (txvq == NULL)
 823                         continue;
 824
 825                 txvq->packets = 0;
 826                 txvq->bytes = 0;
 827                 txvq->errors = 0;
 828                 txvq->multicast = 0;
 829                 txvq->broadcast = 0;
 830                 memset(txvq->size_bins, 0, sizeof(txvq->size_bins[0]) * 8);
 831         }
 832
 833         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 834                 struct virtqueue *rxvq = dev->data->rx_queues[i];
 835                 if (rxvq == NULL)
 836                         continue;
 837
 838                 rxvq->packets = 0;
 839                 rxvq->bytes = 0;
 840                 rxvq->errors = 0;
 841                 rxvq->multicast = 0;
 842                 rxvq->broadcast = 0;
 843                 memset(rxvq->size_bins, 0, sizeof(rxvq->size_bins[0]) * 8);
 844         }
 845 }
 846
 847 static void
 848 virtio_set_hwaddr(struct virtio_hw *hw)
 849 {
 850         vtpci_write_dev_config(hw,
 851                         offsetof(struct virtio_net_config, mac),
 852                         &hw->mac_addr, ETHER_ADDR_LEN);
 853 }
 854
 855 static void
 856 virtio_get_hwaddr(struct virtio_hw *hw)
 857 {
 858         if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
 859                 vtpci_read_dev_config(hw,
 860                         offsetof(struct virtio_net_config, mac),
 861                         &hw->mac_addr, ETHER_ADDR_LEN);
 862         } else {
 863                 eth_random_addr(&hw->mac_addr[0]);
 864                 virtio_set_hwaddr(hw);
 865         }
 866 }
 867
 868 static void
 869 virtio_mac_table_set(struct virtio_hw *hw,
 870                      const struct virtio_net_ctrl_mac *uc,
 871                      const struct virtio_net_ctrl_mac *mc)
 872 {
 873         struct virtio_pmd_ctrl ctrl;
 874         int err, len[2];
 875
 876         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
 877                 PMD_DRV_LOG(INFO, "host does not support mac table");
 878                 return;
 879         }
 880
 881         ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
 882         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
 883
 884         len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
 885         memcpy(ctrl.data, uc, len[0]);
 886
 887         len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
 888         memcpy(ctrl.data + len[0], mc, len[1]);
 889
 890         err = virtio_send_command(hw->cvq, &ctrl, len, 2);
 891         if (err != 0)
 892                 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
 893 }
 894
 895 static void
 896 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
 897                     uint32_t index, uint32_t vmdq __rte_unused)
 898 {
 899         struct virtio_hw *hw = dev->data->dev_private;
 900         const struct ether_addr *addrs = dev->data->mac_addrs;
 901         unsigned int i;
 902         struct virtio_net_ctrl_mac *uc, *mc;
 903
 904         if (index >= VIRTIO_MAX_MAC_ADDRS) {
 905                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
 906                 return;
 907         }
 908
 909         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
 910         uc->entries = 0;
 911         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
 912         mc->entries = 0;
 913
 914         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
 915                 const struct ether_addr *addr
 916                         = (i == index) ? mac_addr : addrs + i;
 917                 struct virtio_net_ctrl_mac *tbl
 918                         = is_multicast_ether_addr(addr) ? mc : uc;
 919
 920                 memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
 921         }
 922
 923         virtio_mac_table_set(hw, uc, mc);
 924 }
 925
 926 static void
 927 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
 928 {
 929         struct virtio_hw *hw = dev->data->dev_private;
 930         struct ether_addr *addrs = dev->data->mac_addrs;
 931         struct virtio_net_ctrl_mac *uc, *mc;
 932         unsigned int i;
 933
 934         if (index >= VIRTIO_MAX_MAC_ADDRS) {
 935                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
 936                 return;
 937         }
 938
 939         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
 940         uc->entries = 0;
 941         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
 942         mc->entries = 0;
 943
 944         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
 945                 struct virtio_net_ctrl_mac *tbl;
 946
 947                 if (i == index || is_zero_ether_addr(addrs + i))
 948                         continue;
 949
 950                 tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
 951                 memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
 952         }
 953
 954         virtio_mac_table_set(hw, uc, mc);
 955 }
 956
 957 static void
 958 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 959 {
 960         struct virtio_hw *hw = dev->data->dev_private;
 961
 962         memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
 963
 964         /* Use atomic update if available */
 965         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
 966                 struct virtio_pmd_ctrl ctrl;
 967                 int len = ETHER_ADDR_LEN;
 968
 969                 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
 970                 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
 971
 972                 memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
 973                 virtio_send_command(hw->cvq, &ctrl, &len, 1);
 974         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
 975                 virtio_set_hwaddr(hw);
 976 }
 977
 978 static int
 979 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
 980 {
 981         struct virtio_hw *hw = dev->data->dev_private;
 982         struct virtio_pmd_ctrl ctrl;
 983         int len;
 984
 985         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
 986                 return -ENOTSUP;
 987
 988         ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
 989         ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
 990         memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
 991         len = sizeof(vlan_id);
 992
 993         return virtio_send_command(hw->cvq, &ctrl, &len, 1);
 994 }
 995
 996 static int
 997 virtio_negotiate_features(struct virtio_hw *hw)
 998 {
 999         uint64_t host_features;
1000
1001         /* Prepare guest_features: feature that driver wants to support */
1002         hw->guest_features = VIRTIO_PMD_GUEST_FEATURES;
1003         PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1004                 hw->guest_features);
1005
1006         /* Read device(host) feature bits */
1007         host_features = hw->vtpci_ops->get_features(hw);
1008         PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1009                 host_features);
1010
1011         /*
1012          * Negotiate features: Subset of device feature bits are written back
1013          * guest feature bits.
1014          */
1015         hw->guest_features = vtpci_negotiate_features(hw, host_features);
1016         PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1017                 hw->guest_features);
1018
1019         if (hw->modern) {
1020                 if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1021                         PMD_INIT_LOG(ERR,
1022                                 "VIRTIO_F_VERSION_1 features is not enabled.");
1023                         return -1;
1024                 }
1025                 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1026                 if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1027                         PMD_INIT_LOG(ERR,
1028                                 "failed to set FEATURES_OK status!");
1029                         return -1;
1030                 }
1031         }
1032
1033         return 0;
1034 }
1035
1036 /*
1037  * Process Virtio Config changed interrupt and call the callback
1038  * if link state changed.
1039  */
1040 static void
1041 virtio_interrupt_handler(__rte_unused struct rte_intr_handle *handle,
1042                          void *param)
1043 {
1044         struct rte_eth_dev *dev = param;
1045         struct virtio_hw *hw = dev->data->dev_private;
1046         uint8_t isr;
1047
1048         /* Read interrupt status which clears interrupt */
1049         isr = vtpci_isr(hw);
1050         PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1051
1052         if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0)
1053                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1054
1055         if (isr & VIRTIO_PCI_ISR_CONFIG) {
1056                 if (virtio_dev_link_update(dev, 0) == 0)
1057                         _rte_eth_dev_callback_process(dev,
1058                                                       RTE_ETH_EVENT_INTR_LSC);
1059         }
1060
1061 }
1062
1063 static void
1064 rx_func_get(struct rte_eth_dev *eth_dev)
1065 {
1066         struct virtio_hw *hw = eth_dev->data->dev_private;
1067         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF))
1068                 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1069         else
1070                 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1071 }
1072
1073 /*
1074  * This function is based on probe() function in virtio_pci.c
1075  * It returns 0 on success.
1076  */
1077 static int
1078 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1079 {
1080         struct virtio_hw *hw = eth_dev->data->dev_private;
1081         struct virtio_net_config *config;
1082         struct virtio_net_config local_config;
1083         struct rte_pci_device *pci_dev;
1084         uint32_t dev_flags = RTE_ETH_DEV_DETACHABLE;
1085         int ret;
1086
1087         RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr_mrg_rxbuf));
1088
1089         eth_dev->dev_ops = &virtio_eth_dev_ops;
1090         eth_dev->tx_pkt_burst = &virtio_xmit_pkts;
1091
1092         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1093                 rx_func_get(eth_dev);
1094                 return 0;
1095         }
1096
1097         /* Allocate memory for storing MAC addresses */
1098         eth_dev->data->mac_addrs = rte_zmalloc("virtio", VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN, 0);
1099         if (eth_dev->data->mac_addrs == NULL) {
1100                 PMD_INIT_LOG(ERR,
1101                         "Failed to allocate %d bytes needed to store MAC addresses",
1102                         VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN);
1103                 return -ENOMEM;
1104         }
1105
1106         pci_dev = eth_dev->pci_dev;
1107
1108         ret = vtpci_init(pci_dev, hw, &dev_flags);
1109         if (ret)
1110                 return ret;
1111
1112         /* Reset the device although not necessary at startup */
1113         vtpci_reset(hw);
1114
1115         /* Tell the host we've noticed this device. */
1116         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1117
1118         /* Tell the host we've known how to drive the device. */
1119         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1120         if (virtio_negotiate_features(hw) < 0)
1121                 return -1;
1122
1123         /* If host does not support status then disable LSC */
1124         if (!vtpci_with_feature(hw, VIRTIO_NET_F_STATUS))
1125                 dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1126
1127         rte_eth_copy_pci_info(eth_dev, pci_dev);
1128         eth_dev->data->dev_flags = dev_flags;
1129
1130         rx_func_get(eth_dev);
1131
1132         /* Setting up rx_header size for the device */
1133         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1134             vtpci_with_feature(hw, VIRTIO_F_VERSION_1))
1135                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1136         else
1137                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1138
1139         /* Copy the permanent MAC address to: virtio_hw */
1140         virtio_get_hwaddr(hw);
1141         ether_addr_copy((struct ether_addr *) hw->mac_addr,
1142                         &eth_dev->data->mac_addrs[0]);
1143         PMD_INIT_LOG(DEBUG,
1144                      "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1145                      hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1146                      hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1147
1148         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1149                 config = &local_config;
1150
1151                 vtpci_read_dev_config(hw,
1152                         offsetof(struct virtio_net_config, mac),
1153                         &config->mac, sizeof(config->mac));
1154
1155                 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1156                         vtpci_read_dev_config(hw,
1157                                 offsetof(struct virtio_net_config, status),
1158                                 &config->status, sizeof(config->status));
1159                 } else {
1160                         PMD_INIT_LOG(DEBUG,
1161                                      "VIRTIO_NET_F_STATUS is not supported");
1162                         config->status = 0;
1163                 }
1164
1165                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1166                         vtpci_read_dev_config(hw,
1167                                 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1168                                 &config->max_virtqueue_pairs,
1169                                 sizeof(config->max_virtqueue_pairs));
1170                 } else {
1171                         PMD_INIT_LOG(DEBUG,
1172                                      "VIRTIO_NET_F_MQ is not supported");
1173                         config->max_virtqueue_pairs = 1;
1174                 }
1175
1176                 hw->max_rx_queues =
1177                         (VIRTIO_MAX_RX_QUEUES < config->max_virtqueue_pairs) ?
1178                         VIRTIO_MAX_RX_QUEUES : config->max_virtqueue_pairs;
1179                 hw->max_tx_queues =
1180                         (VIRTIO_MAX_TX_QUEUES < config->max_virtqueue_pairs) ?
1181                         VIRTIO_MAX_TX_QUEUES : config->max_virtqueue_pairs;
1182
1183                 virtio_dev_cq_queue_setup(eth_dev,
1184                                         config->max_virtqueue_pairs * 2,
1185                                         SOCKET_ID_ANY);
1186
1187                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1188                                 config->max_virtqueue_pairs);
1189                 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1190                 PMD_INIT_LOG(DEBUG,
1191                                 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1192                                 config->mac[0], config->mac[1],
1193                                 config->mac[2], config->mac[3],
1194                                 config->mac[4], config->mac[5]);
1195         } else {
1196                 hw->max_rx_queues = 1;
1197                 hw->max_tx_queues = 1;
1198         }
1199
1200         PMD_INIT_LOG(DEBUG, "hw->max_rx_queues=%d   hw->max_tx_queues=%d",
1201                         hw->max_rx_queues, hw->max_tx_queues);
1202         PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1203                         eth_dev->data->port_id, pci_dev->id.vendor_id,
1204                         pci_dev->id.device_id);
1205
1206         /* Setup interrupt callback  */
1207         if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1208                 rte_intr_callback_register(&pci_dev->intr_handle,
1209                                    virtio_interrupt_handler, eth_dev);
1210
1211         virtio_dev_cq_start(eth_dev);
1212
1213         return 0;
1214 }
1215
1216 static int
1217 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1218 {
1219         struct rte_pci_device *pci_dev;
1220         struct virtio_hw *hw = eth_dev->data->dev_private;
1221
1222         PMD_INIT_FUNC_TRACE();
1223
1224         if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1225                 return -EPERM;
1226
1227         /* Close it anyway since there's no way to know if closed */
1228         virtio_dev_close(eth_dev);
1229
1230         pci_dev = eth_dev->pci_dev;
1231
1232         eth_dev->dev_ops = NULL;
1233         eth_dev->tx_pkt_burst = NULL;
1234         eth_dev->rx_pkt_burst = NULL;
1235
1236         virtio_dev_queue_release(hw->cvq);
1237
1238         rte_free(eth_dev->data->mac_addrs);
1239         eth_dev->data->mac_addrs = NULL;
1240
1241         /* reset interrupt callback  */
1242         if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1243                 rte_intr_callback_unregister(&pci_dev->intr_handle,
1244                                                 virtio_interrupt_handler,
1245                                                 eth_dev);
1246         rte_eal_pci_unmap_device(pci_dev);
1247
1248         PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1249
1250         return 0;
1251 }
1252
1253 static struct eth_driver rte_virtio_pmd = {
1254         .pci_drv = {
1255                 .name = "rte_virtio_pmd",
1256                 .id_table = pci_id_virtio_map,
1257                 .drv_flags = RTE_PCI_DRV_DETACHABLE,
1258         },
1259         .eth_dev_init = eth_virtio_dev_init,
1260         .eth_dev_uninit = eth_virtio_dev_uninit,
1261         .dev_private_size = sizeof(struct virtio_hw),
1262 };
1263
1264 /*
1265  * Driver initialization routine.
1266  * Invoked once at EAL init time.
1267  * Register itself as the [Poll Mode] Driver of PCI virtio devices.
1268  * Returns 0 on success.
1269  */
1270 static int
1271 rte_virtio_pmd_init(const char *name __rte_unused,
1272                     const char *param __rte_unused)
1273 {
1274         if (rte_eal_iopl_init() != 0) {
1275                 PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1276                 return -1;
1277         }
1278
1279         rte_eth_driver_register(&rte_virtio_pmd);
1280         return 0;
1281 }
1282
1283 /*
1284  * Configure virtio device
1285  * It returns 0 on success.
1286  */
1287 static int
1288 virtio_dev_configure(struct rte_eth_dev *dev)
1289 {
1290         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1291         struct virtio_hw *hw = dev->data->dev_private;
1292
1293         PMD_INIT_LOG(DEBUG, "configure");
1294
1295         if (rxmode->hw_ip_checksum) {
1296                 PMD_DRV_LOG(ERR, "HW IP checksum not supported");
1297                 return -EINVAL;
1298         }
1299
1300         hw->vlan_strip = rxmode->hw_vlan_strip;
1301
1302         if (rxmode->hw_vlan_filter
1303             && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1304                 PMD_DRV_LOG(NOTICE,
1305                             "vlan filtering not available on this host");
1306                 return -ENOTSUP;
1307         }
1308
1309         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1310                 if (vtpci_irq_config(hw, 0) == VIRTIO_MSI_NO_VECTOR) {
1311                         PMD_DRV_LOG(ERR, "failed to set config vector");
1312                         return -EBUSY;
1313                 }
1314
1315         return 0;
1316 }
1317
1318
1319 static int
1320 virtio_dev_start(struct rte_eth_dev *dev)
1321 {
1322         uint16_t nb_queues, i;
1323         struct virtio_hw *hw = dev->data->dev_private;
1324
1325         /* check if lsc interrupt feature is enabled */
1326         if (dev->data->dev_conf.intr_conf.lsc) {
1327                 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
1328                         PMD_DRV_LOG(ERR, "link status not supported by host");
1329                         return -ENOTSUP;
1330                 }
1331
1332                 if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0) {
1333                         PMD_DRV_LOG(ERR, "interrupt enable failed");
1334                         return -EIO;
1335                 }
1336         }
1337
1338         /* Initialize Link state */
1339         virtio_dev_link_update(dev, 0);
1340
1341         /* On restart after stop do not touch queues */
1342         if (hw->started)
1343                 return 0;
1344
1345         /* Do final configuration before rx/tx engine starts */
1346         virtio_dev_rxtx_start(dev);
1347         vtpci_reinit_complete(hw);
1348
1349         hw->started = 1;
1350
1351         /*Notify the backend
1352          *Otherwise the tap backend might already stop its queue due to fullness.
1353          *vhost backend will have no chance to be waked up
1354          */
1355         nb_queues = dev->data->nb_rx_queues;
1356         if (nb_queues > 1) {
1357                 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1358                         return -EINVAL;
1359         }
1360
1361         PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1362
1363         for (i = 0; i < nb_queues; i++)
1364                 virtqueue_notify(dev->data->rx_queues[i]);
1365
1366         PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1367
1368         for (i = 0; i < dev->data->nb_rx_queues; i++)
1369                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1370
1371         for (i = 0; i < dev->data->nb_tx_queues; i++)
1372                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1373
1374         return 0;
1375 }
1376
1377 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1378 {
1379         struct rte_mbuf *buf;
1380         int i, mbuf_num = 0;
1381
1382         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1383                 PMD_INIT_LOG(DEBUG,
1384                              "Before freeing rxq[%d] used and unused buf", i);
1385                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1386
1387                 PMD_INIT_LOG(DEBUG, "rx_queues[%d]=%p",
1388                                 i, dev->data->rx_queues[i]);
1389                 while ((buf = (struct rte_mbuf *)virtqueue_detatch_unused(
1390                                         dev->data->rx_queues[i])) != NULL) {
1391                         rte_pktmbuf_free(buf);
1392                         mbuf_num++;
1393                 }
1394
1395                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1396                 PMD_INIT_LOG(DEBUG,
1397                              "After freeing rxq[%d] used and unused buf", i);
1398                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1399         }
1400
1401         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1402                 PMD_INIT_LOG(DEBUG,
1403                              "Before freeing txq[%d] used and unused bufs",
1404                              i);
1405                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1406
1407                 mbuf_num = 0;
1408                 while ((buf = (struct rte_mbuf *)virtqueue_detatch_unused(
1409                                         dev->data->tx_queues[i])) != NULL) {
1410                         rte_pktmbuf_free(buf);
1411
1412                         mbuf_num++;
1413                 }
1414
1415                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1416                 PMD_INIT_LOG(DEBUG,
1417                              "After freeing txq[%d] used and unused buf", i);
1418                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1419         }
1420 }
1421
1422 /*
1423  * Stop device: disable interrupt and mark link down
1424  */
1425 static void
1426 virtio_dev_stop(struct rte_eth_dev *dev)
1427 {
1428         struct rte_eth_link link;
1429         struct virtio_hw *hw = dev->data->dev_private;
1430
1431         PMD_INIT_LOG(DEBUG, "stop");
1432
1433         hw->started = 0;
1434
1435         if (dev->data->dev_conf.intr_conf.lsc)
1436                 rte_intr_disable(&dev->pci_dev->intr_handle);
1437
1438         memset(&link, 0, sizeof(link));
1439         virtio_dev_atomic_write_link_status(dev, &link);
1440 }
1441
1442 static int
1443 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
1444 {
1445         struct rte_eth_link link, old;
1446         uint16_t status;
1447         struct virtio_hw *hw = dev->data->dev_private;
1448         memset(&link, 0, sizeof(link));
1449         virtio_dev_atomic_read_link_status(dev, &link);
1450         old = link;
1451         link.link_duplex = ETH_LINK_FULL_DUPLEX;
1452         link.link_speed  = SPEED_10G;
1453
1454         if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1455                 PMD_INIT_LOG(DEBUG, "Get link status from hw");
1456                 vtpci_read_dev_config(hw,
1457                                 offsetof(struct virtio_net_config, status),
1458                                 &status, sizeof(status));
1459                 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
1460                         link.link_status = ETH_LINK_DOWN;
1461                         PMD_INIT_LOG(DEBUG, "Port %d is down",
1462                                      dev->data->port_id);
1463                 } else {
1464                         link.link_status = ETH_LINK_UP;
1465                         PMD_INIT_LOG(DEBUG, "Port %d is up",
1466                                      dev->data->port_id);
1467                 }
1468         } else {
1469                 link.link_status = ETH_LINK_UP;
1470         }
1471         virtio_dev_atomic_write_link_status(dev, &link);
1472
1473         return (old.link_status == link.link_status) ? -1 : 0;
1474 }
1475
1476 static void
1477 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1478 {
1479         struct virtio_hw *hw = dev->data->dev_private;
1480
1481         dev_info->driver_name = dev->driver->pci_drv.name;
1482         dev_info->max_rx_queues = (uint16_t)hw->max_rx_queues;
1483         dev_info->max_tx_queues = (uint16_t)hw->max_tx_queues;
1484         dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
1485         dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
1486         dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
1487         dev_info->default_txconf = (struct rte_eth_txconf) {
1488                 .txq_flags = ETH_TXQ_FLAGS_NOOFFLOADS
1489         };
1490 }
1491
1492 /*
1493  * It enables testpmd to collect per queue stats.
1494  */
1495 static int
1496 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
1497 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
1498 __rte_unused uint8_t is_rx)
1499 {
1500         return 0;
1501 }
1502
1503 static struct rte_driver rte_virtio_driver = {
1504         .type = PMD_PDEV,
1505         .init = rte_virtio_pmd_init,
1506 };
1507
1508 PMD_REGISTER_DRIVER(rte_virtio_driver);