drivers/net/virtio/virtio_ethdev.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2015 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
  63 static int eth_virtio_dev_init(struct rte_eth_dev *eth_dev);
  64 static int eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev);
  65 static int  virtio_dev_configure(struct rte_eth_dev *dev);
  66 static int  virtio_dev_start(struct rte_eth_dev *dev);
  67 static void virtio_dev_stop(struct rte_eth_dev *dev);
  68 static void virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
  69 static void virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
  70 static void virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
  71 static void virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
  72 static void virtio_dev_info_get(struct rte_eth_dev *dev,
  73                                 struct rte_eth_dev_info *dev_info);
  74 static int virtio_dev_link_update(struct rte_eth_dev *dev,
  75         __rte_unused int wait_to_complete);
  76
  77 static void virtio_set_hwaddr(struct virtio_hw *hw);
  78 static void virtio_get_hwaddr(struct virtio_hw *hw);
  79
  80 static void virtio_dev_stats_get(struct rte_eth_dev *dev,
  81                                  struct rte_eth_stats *stats);
  82 static int virtio_dev_xstats_get(struct rte_eth_dev *dev,
  83                                  struct rte_eth_xstats *xstats, unsigned n);
  84 static void virtio_dev_stats_reset(struct rte_eth_dev *dev);
  85 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
  86 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
  87                                 uint16_t vlan_id, int on);
  88 static void virtio_mac_addr_add(struct rte_eth_dev *dev,
  89                                 struct ether_addr *mac_addr,
  90                                 uint32_t index, uint32_t vmdq __rte_unused);
  91 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
  92 static void virtio_mac_addr_set(struct rte_eth_dev *dev,
  93                                 struct ether_addr *mac_addr);
  94
  95 static int virtio_dev_queue_stats_mapping_set(
  96         __rte_unused struct rte_eth_dev *eth_dev,
  97         __rte_unused uint16_t queue_id,
  98         __rte_unused uint8_t stat_idx,
  99         __rte_unused uint8_t is_rx);
 100
 101 /*
 102  * The set of PCI devices this driver supports
 103  */
 104 static const struct rte_pci_id pci_id_virtio_map[] = {
 105
 106 #define RTE_PCI_DEV_ID_DECL_VIRTIO(vend, dev) {RTE_PCI_DEVICE(vend, dev)},
 107 #include "rte_pci_dev_ids.h"
 108
 109 { .vendor_id = 0, /* sentinel */ },
 110 };
 111
 112 struct rte_virtio_xstats_name_off {
 113         char name[RTE_ETH_XSTATS_NAME_SIZE];
 114         unsigned offset;
 115 };
 116
 117 /* [rt]x_qX_ is prepended to the name string here */
 118 static const struct rte_virtio_xstats_name_off rte_virtio_q_stat_strings[] = {
 119         {"good_packets",           offsetof(struct virtqueue, packets)},
 120         {"good_bytes",             offsetof(struct virtqueue, bytes)},
 121         {"errors",                 offsetof(struct virtqueue, errors)},
 122         {"multicast_packets",      offsetof(struct virtqueue, multicast)},
 123         {"broadcast_packets",      offsetof(struct virtqueue, broadcast)},
 124         {"undersize_packets",      offsetof(struct virtqueue, size_bins[0])},
 125         {"size_64_packets",        offsetof(struct virtqueue, size_bins[1])},
 126         {"size_65_127_packets",    offsetof(struct virtqueue, size_bins[2])},
 127         {"size_128_255_packets",   offsetof(struct virtqueue, size_bins[3])},
 128         {"size_256_511_packets",   offsetof(struct virtqueue, size_bins[4])},
 129         {"size_512_1023_packets",  offsetof(struct virtqueue, size_bins[5])},
 130         {"size_1024_1517_packets", offsetof(struct virtqueue, size_bins[6])},
 131         {"size_1518_max_packets",  offsetof(struct virtqueue, size_bins[7])},
 132 };
 133
 134 #define VIRTIO_NB_Q_XSTATS (sizeof(rte_virtio_q_stat_strings) / \
 135                             sizeof(rte_virtio_q_stat_strings[0]))
 136
 137 static int
 138 virtio_send_command(struct virtqueue *vq, struct virtio_pmd_ctrl *ctrl,
 139                 int *dlen, int pkt_num)
 140 {
 141         uint32_t head, i;
 142         int k, sum = 0;
 143         virtio_net_ctrl_ack status = ~0;
 144         struct virtio_pmd_ctrl result;
 145
 146         ctrl->status = status;
 147
 148         if (!(vq && vq->hw->cvq)) {
 149                 PMD_INIT_LOG(ERR,
 150                              "%s(): Control queue is not supported.",
 151                              __func__);
 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         struct virtio_hw *hw;
 268
 269         if (vq) {
 270                 hw = vq->hw;
 271                 hw->vtpci_ops->del_queue(hw, vq);
 272
 273                 rte_free(vq->sw_ring);
 274                 rte_free(vq);
 275         }
 276 }
 277
 278 int virtio_dev_queue_setup(struct rte_eth_dev *dev,
 279                         int queue_type,
 280                         uint16_t queue_idx,
 281                         uint16_t vtpci_queue_idx,
 282                         uint16_t nb_desc,
 283                         unsigned int socket_id,
 284                         struct virtqueue **pvq)
 285 {
 286         char vq_name[VIRTQUEUE_MAX_NAME_SZ];
 287         const struct rte_memzone *mz;
 288         unsigned int vq_size, size;
 289         struct virtio_hw *hw = dev->data->dev_private;
 290         struct virtqueue *vq = NULL;
 291
 292         PMD_INIT_LOG(DEBUG, "setting up queue: %u", vtpci_queue_idx);
 293
 294         /*
 295          * Read the virtqueue size from the Queue Size field
 296          * Always power of 2 and if 0 virtqueue does not exist
 297          */
 298         vq_size = hw->vtpci_ops->get_queue_num(hw, vtpci_queue_idx);
 299         PMD_INIT_LOG(DEBUG, "vq_size: %u nb_desc:%u", vq_size, nb_desc);
 300         if (vq_size == 0) {
 301                 PMD_INIT_LOG(ERR, "%s: virtqueue does not exist", __func__);
 302                 return -EINVAL;
 303         }
 304
 305         if (!rte_is_power_of_2(vq_size)) {
 306                 PMD_INIT_LOG(ERR, "%s: virtqueue size is not powerof 2", __func__);
 307                 return -EINVAL;
 308         }
 309
 310         if (queue_type == VTNET_RQ) {
 311                 snprintf(vq_name, sizeof(vq_name), "port%d_rvq%d",
 312                         dev->data->port_id, queue_idx);
 313                 vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
 314                         vq_size * sizeof(struct vq_desc_extra), RTE_CACHE_LINE_SIZE);
 315                 vq->sw_ring = rte_zmalloc_socket("rxq->sw_ring",
 316                         (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
 317                         sizeof(vq->sw_ring[0]), RTE_CACHE_LINE_SIZE, socket_id);
 318         } else if (queue_type == VTNET_TQ) {
 319                 snprintf(vq_name, sizeof(vq_name), "port%d_tvq%d",
 320                         dev->data->port_id, queue_idx);
 321                 vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
 322                         vq_size * sizeof(struct vq_desc_extra), RTE_CACHE_LINE_SIZE);
 323         } else if (queue_type == VTNET_CQ) {
 324                 snprintf(vq_name, sizeof(vq_name), "port%d_cvq",
 325                         dev->data->port_id);
 326                 vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
 327                         vq_size * sizeof(struct vq_desc_extra),
 328                         RTE_CACHE_LINE_SIZE);
 329         }
 330         if (vq == NULL) {
 331                 PMD_INIT_LOG(ERR, "%s: Can not allocate virtqueue", __func__);
 332                 return -ENOMEM;
 333         }
 334         if (queue_type == VTNET_RQ && vq->sw_ring == NULL) {
 335                 PMD_INIT_LOG(ERR, "%s: Can not allocate RX soft ring",
 336                         __func__);
 337                 rte_free(vq);
 338                 return -ENOMEM;
 339         }
 340
 341         vq->hw = hw;
 342         vq->port_id = dev->data->port_id;
 343         vq->queue_id = queue_idx;
 344         vq->vq_queue_index = vtpci_queue_idx;
 345         vq->vq_nentries = vq_size;
 346
 347         if (nb_desc == 0 || nb_desc > vq_size)
 348                 nb_desc = vq_size;
 349         vq->vq_free_cnt = nb_desc;
 350
 351         /*
 352          * Reserve a memzone for vring elements
 353          */
 354         size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
 355         vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
 356         PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d", size, vq->vq_ring_size);
 357
 358         mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
 359                 socket_id, 0, VIRTIO_PCI_VRING_ALIGN);
 360         if (mz == NULL) {
 361                 if (rte_errno == EEXIST)
 362                         mz = rte_memzone_lookup(vq_name);
 363                 if (mz == NULL) {
 364                         rte_free(vq);
 365                         return -ENOMEM;
 366                 }
 367         }
 368
 369         /*
 370          * Virtio PCI device VIRTIO_PCI_QUEUE_PF register is 32bit,
 371          * and only accepts 32 bit page frame number.
 372          * Check if the allocated physical memory exceeds 16TB.
 373          */
 374         if ((mz->phys_addr + vq->vq_ring_size - 1) >> (VIRTIO_PCI_QUEUE_ADDR_SHIFT + 32)) {
 375                 PMD_INIT_LOG(ERR, "vring address shouldn't be above 16TB!");
 376                 rte_free(vq);
 377                 return -ENOMEM;
 378         }
 379
 380         memset(mz->addr, 0, sizeof(mz->len));
 381         vq->mz = mz;
 382         vq->vq_ring_mem = mz->phys_addr;
 383         vq->vq_ring_virt_mem = mz->addr;
 384         PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%"PRIx64, (uint64_t)mz->phys_addr);
 385         PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%"PRIx64, (uint64_t)(uintptr_t)mz->addr);
 386         vq->virtio_net_hdr_mz  = NULL;
 387         vq->virtio_net_hdr_mem = 0;
 388
 389         if (queue_type == VTNET_TQ) {
 390                 /*
 391                  * For each xmit packet, allocate a virtio_net_hdr
 392                  */
 393                 snprintf(vq_name, sizeof(vq_name), "port%d_tvq%d_hdrzone",
 394                         dev->data->port_id, queue_idx);
 395                 vq->virtio_net_hdr_mz = rte_memzone_reserve_aligned(vq_name,
 396                         vq_size * hw->vtnet_hdr_size,
 397                         socket_id, 0, RTE_CACHE_LINE_SIZE);
 398                 if (vq->virtio_net_hdr_mz == NULL) {
 399                         if (rte_errno == EEXIST)
 400                                 vq->virtio_net_hdr_mz =
 401                                         rte_memzone_lookup(vq_name);
 402                         if (vq->virtio_net_hdr_mz == NULL) {
 403                                 rte_free(vq);
 404                                 return -ENOMEM;
 405                         }
 406                 }
 407                 vq->virtio_net_hdr_mem =
 408                         vq->virtio_net_hdr_mz->phys_addr;
 409                 memset(vq->virtio_net_hdr_mz->addr, 0,
 410                         vq_size * hw->vtnet_hdr_size);
 411         } else if (queue_type == VTNET_CQ) {
 412                 /* Allocate a page for control vq command, data and status */
 413                 snprintf(vq_name, sizeof(vq_name), "port%d_cvq_hdrzone",
 414                         dev->data->port_id);
 415                 vq->virtio_net_hdr_mz = rte_memzone_reserve_aligned(vq_name,
 416                         PAGE_SIZE, socket_id, 0, RTE_CACHE_LINE_SIZE);
 417                 if (vq->virtio_net_hdr_mz == NULL) {
 418                         if (rte_errno == EEXIST)
 419                                 vq->virtio_net_hdr_mz =
 420                                         rte_memzone_lookup(vq_name);
 421                         if (vq->virtio_net_hdr_mz == NULL) {
 422                                 rte_free(vq);
 423                                 return -ENOMEM;
 424                         }
 425                 }
 426                 vq->virtio_net_hdr_mem =
 427                         vq->virtio_net_hdr_mz->phys_addr;
 428                 memset(vq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
 429         }
 430
 431         hw->vtpci_ops->setup_queue(hw, vq);
 432
 433         *pvq = vq;
 434         return 0;
 435 }
 436
 437 static int
 438 virtio_dev_cq_queue_setup(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx,
 439                 uint32_t socket_id)
 440 {
 441         struct virtqueue *vq;
 442         int ret;
 443         struct virtio_hw *hw = dev->data->dev_private;
 444
 445         PMD_INIT_FUNC_TRACE();
 446         ret = virtio_dev_queue_setup(dev, VTNET_CQ, VTNET_SQ_CQ_QUEUE_IDX,
 447                         vtpci_queue_idx, 0, socket_id, &vq);
 448         if (ret < 0) {
 449                 PMD_INIT_LOG(ERR, "control vq initialization failed");
 450                 return ret;
 451         }
 452
 453         hw->cvq = vq;
 454         return 0;
 455 }
 456
 457 static void
 458 virtio_free_queues(struct rte_eth_dev *dev)
 459 {
 460         unsigned int i;
 461
 462         for (i = 0; i < dev->data->nb_rx_queues; i++)
 463                 virtio_dev_rx_queue_release(dev->data->rx_queues[i]);
 464
 465         dev->data->nb_rx_queues = 0;
 466
 467         for (i = 0; i < dev->data->nb_tx_queues; i++)
 468                 virtio_dev_tx_queue_release(dev->data->tx_queues[i]);
 469
 470         dev->data->nb_tx_queues = 0;
 471 }
 472
 473 static void
 474 virtio_dev_close(struct rte_eth_dev *dev)
 475 {
 476         struct virtio_hw *hw = dev->data->dev_private;
 477         struct rte_pci_device *pci_dev = dev->pci_dev;
 478
 479         PMD_INIT_LOG(DEBUG, "virtio_dev_close");
 480
 481         if (hw->started == 1)
 482                 virtio_dev_stop(dev);
 483
 484         /* reset the NIC */
 485         if (pci_dev->driver->drv_flags & RTE_PCI_DRV_INTR_LSC)
 486                 vtpci_irq_config(hw, VIRTIO_MSI_NO_VECTOR);
 487         vtpci_reset(hw);
 488         virtio_dev_free_mbufs(dev);
 489         virtio_free_queues(dev);
 490 }
 491
 492 static void
 493 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
 494 {
 495         struct virtio_hw *hw = dev->data->dev_private;
 496         struct virtio_pmd_ctrl ctrl;
 497         int dlen[1];
 498         int ret;
 499
 500         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 501                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 502                 return;
 503         }
 504
 505         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 506         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 507         ctrl.data[0] = 1;
 508         dlen[0] = 1;
 509
 510         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 511         if (ret)
 512                 PMD_INIT_LOG(ERR, "Failed to enable promisc");
 513 }
 514
 515 static void
 516 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
 517 {
 518         struct virtio_hw *hw = dev->data->dev_private;
 519         struct virtio_pmd_ctrl ctrl;
 520         int dlen[1];
 521         int ret;
 522
 523         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 524                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 525                 return;
 526         }
 527
 528         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 529         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 530         ctrl.data[0] = 0;
 531         dlen[0] = 1;
 532
 533         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 534         if (ret)
 535                 PMD_INIT_LOG(ERR, "Failed to disable promisc");
 536 }
 537
 538 static void
 539 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
 540 {
 541         struct virtio_hw *hw = dev->data->dev_private;
 542         struct virtio_pmd_ctrl ctrl;
 543         int dlen[1];
 544         int ret;
 545
 546         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 547                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 548                 return;
 549         }
 550
 551         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 552         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 553         ctrl.data[0] = 1;
 554         dlen[0] = 1;
 555
 556         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 557         if (ret)
 558                 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
 559 }
 560
 561 static void
 562 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
 563 {
 564         struct virtio_hw *hw = dev->data->dev_private;
 565         struct virtio_pmd_ctrl ctrl;
 566         int dlen[1];
 567         int ret;
 568
 569         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 570                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 571                 return;
 572         }
 573
 574         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 575         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 576         ctrl.data[0] = 0;
 577         dlen[0] = 1;
 578
 579         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 580         if (ret)
 581                 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
 582 }
 583
 584 /*
 585  * dev_ops for virtio, bare necessities for basic operation
 586  */
 587 static const struct eth_dev_ops virtio_eth_dev_ops = {
 588         .dev_configure           = virtio_dev_configure,
 589         .dev_start               = virtio_dev_start,
 590         .dev_stop                = virtio_dev_stop,
 591         .dev_close               = virtio_dev_close,
 592         .promiscuous_enable      = virtio_dev_promiscuous_enable,
 593         .promiscuous_disable     = virtio_dev_promiscuous_disable,
 594         .allmulticast_enable     = virtio_dev_allmulticast_enable,
 595         .allmulticast_disable    = virtio_dev_allmulticast_disable,
 596
 597         .dev_infos_get           = virtio_dev_info_get,
 598         .stats_get               = virtio_dev_stats_get,
 599         .xstats_get              = virtio_dev_xstats_get,
 600         .stats_reset             = virtio_dev_stats_reset,
 601         .xstats_reset            = virtio_dev_stats_reset,
 602         .link_update             = virtio_dev_link_update,
 603         .rx_queue_setup          = virtio_dev_rx_queue_setup,
 604         .rx_queue_release        = virtio_dev_rx_queue_release,
 605         .tx_queue_setup          = virtio_dev_tx_queue_setup,
 606         .tx_queue_release        = virtio_dev_tx_queue_release,
 607         /* collect stats per queue */
 608         .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
 609         .vlan_filter_set         = virtio_vlan_filter_set,
 610         .mac_addr_add            = virtio_mac_addr_add,
 611         .mac_addr_remove         = virtio_mac_addr_remove,
 612         .mac_addr_set            = virtio_mac_addr_set,
 613 };
 614
 615 static inline int
 616 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
 617                                 struct rte_eth_link *link)
 618 {
 619         struct rte_eth_link *dst = link;
 620         struct rte_eth_link *src = &(dev->data->dev_link);
 621
 622         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
 623                         *(uint64_t *)src) == 0)
 624                 return -1;
 625
 626         return 0;
 627 }
 628
 629 /**
 630  * Atomically writes the link status information into global
 631  * structure rte_eth_dev.
 632  *
 633  * @param dev
 634  *   - Pointer to the structure rte_eth_dev to read from.
 635  *   - Pointer to the buffer to be saved with the link status.
 636  *
 637  * @return
 638  *   - On success, zero.
 639  *   - On failure, negative value.
 640  */
 641 static inline int
 642 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
 643                 struct rte_eth_link *link)
 644 {
 645         struct rte_eth_link *dst = &(dev->data->dev_link);
 646         struct rte_eth_link *src = link;
 647
 648         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
 649                                         *(uint64_t *)src) == 0)
 650                 return -1;
 651
 652         return 0;
 653 }
 654
 655 static void
 656 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 657 {
 658         unsigned i;
 659
 660         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 661                 const struct virtqueue *txvq = dev->data->tx_queues[i];
 662                 if (txvq == NULL)
 663                         continue;
 664
 665                 stats->opackets += txvq->packets;
 666                 stats->obytes += txvq->bytes;
 667                 stats->oerrors += txvq->errors;
 668
 669                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 670                         stats->q_opackets[i] = txvq->packets;
 671                         stats->q_obytes[i] = txvq->bytes;
 672                 }
 673         }
 674
 675         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 676                 const struct virtqueue *rxvq = dev->data->rx_queues[i];
 677                 if (rxvq == NULL)
 678                         continue;
 679
 680                 stats->ipackets += rxvq->packets;
 681                 stats->ibytes += rxvq->bytes;
 682                 stats->ierrors += rxvq->errors;
 683
 684                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 685                         stats->q_ipackets[i] = rxvq->packets;
 686                         stats->q_ibytes[i] = rxvq->bytes;
 687                 }
 688         }
 689
 690         stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
 691 }
 692
 693 static int
 694 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstats *xstats,
 695                       unsigned n)
 696 {
 697         unsigned i;
 698         unsigned count = 0;
 699
 700         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_Q_XSTATS +
 701                 dev->data->nb_rx_queues * VIRTIO_NB_Q_XSTATS;
 702
 703         if (n < nstats)
 704                 return nstats;
 705
 706         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 707                 struct virtqueue *rxvq = dev->data->rx_queues[i];
 708
 709                 if (rxvq == NULL)
 710                         continue;
 711
 712                 unsigned t;
 713
 714                 for (t = 0; t < VIRTIO_NB_Q_XSTATS; t++) {
 715                         snprintf(xstats[count].name, sizeof(xstats[count].name),
 716                                  "rx_q%u_%s", i,
 717                                  rte_virtio_q_stat_strings[t].name);
 718                         xstats[count].value = *(uint64_t *)(((char *)rxvq) +
 719                                 rte_virtio_q_stat_strings[t].offset);
 720                         count++;
 721                 }
 722         }
 723
 724         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 725                 struct virtqueue *txvq = dev->data->tx_queues[i];
 726
 727                 if (txvq == NULL)
 728                         continue;
 729
 730                 unsigned t;
 731
 732                 for (t = 0; t < VIRTIO_NB_Q_XSTATS; t++) {
 733                         snprintf(xstats[count].name, sizeof(xstats[count].name),
 734                                  "tx_q%u_%s", i,
 735                                  rte_virtio_q_stat_strings[t].name);
 736                         xstats[count].value = *(uint64_t *)(((char *)txvq) +
 737                                 rte_virtio_q_stat_strings[t].offset);
 738                         count++;
 739                 }
 740         }
 741
 742         return count;
 743 }
 744
 745 static void
 746 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 747 {
 748         virtio_update_stats(dev, stats);
 749 }
 750
 751 static void
 752 virtio_dev_stats_reset(struct rte_eth_dev *dev)
 753 {
 754         unsigned int i;
 755
 756         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 757                 struct virtqueue *txvq = dev->data->tx_queues[i];
 758                 if (txvq == NULL)
 759                         continue;
 760
 761                 txvq->packets = 0;
 762                 txvq->bytes = 0;
 763                 txvq->errors = 0;
 764                 txvq->multicast = 0;
 765                 txvq->broadcast = 0;
 766                 memset(txvq->size_bins, 0, sizeof(txvq->size_bins[0]) * 8);
 767         }
 768
 769         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 770                 struct virtqueue *rxvq = dev->data->rx_queues[i];
 771                 if (rxvq == NULL)
 772                         continue;
 773
 774                 rxvq->packets = 0;
 775                 rxvq->bytes = 0;
 776                 rxvq->errors = 0;
 777                 rxvq->multicast = 0;
 778                 rxvq->broadcast = 0;
 779                 memset(rxvq->size_bins, 0, sizeof(rxvq->size_bins[0]) * 8);
 780         }
 781 }
 782
 783 static void
 784 virtio_set_hwaddr(struct virtio_hw *hw)
 785 {
 786         vtpci_write_dev_config(hw,
 787                         offsetof(struct virtio_net_config, mac),
 788                         &hw->mac_addr, ETHER_ADDR_LEN);
 789 }
 790
 791 static void
 792 virtio_get_hwaddr(struct virtio_hw *hw)
 793 {
 794         if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
 795                 vtpci_read_dev_config(hw,
 796                         offsetof(struct virtio_net_config, mac),
 797                         &hw->mac_addr, ETHER_ADDR_LEN);
 798         } else {
 799                 eth_random_addr(&hw->mac_addr[0]);
 800                 virtio_set_hwaddr(hw);
 801         }
 802 }
 803
 804 static void
 805 virtio_mac_table_set(struct virtio_hw *hw,
 806                      const struct virtio_net_ctrl_mac *uc,
 807                      const struct virtio_net_ctrl_mac *mc)
 808 {
 809         struct virtio_pmd_ctrl ctrl;
 810         int err, len[2];
 811
 812         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
 813                 PMD_DRV_LOG(INFO, "host does not support mac table\n");
 814                 return;
 815         }
 816
 817         ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
 818         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
 819
 820         len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
 821         memcpy(ctrl.data, uc, len[0]);
 822
 823         len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
 824         memcpy(ctrl.data + len[0], mc, len[1]);
 825
 826         err = virtio_send_command(hw->cvq, &ctrl, len, 2);
 827         if (err != 0)
 828                 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
 829 }
 830
 831 static void
 832 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
 833                     uint32_t index, uint32_t vmdq __rte_unused)
 834 {
 835         struct virtio_hw *hw = dev->data->dev_private;
 836         const struct ether_addr *addrs = dev->data->mac_addrs;
 837         unsigned int i;
 838         struct virtio_net_ctrl_mac *uc, *mc;
 839
 840         if (index >= VIRTIO_MAX_MAC_ADDRS) {
 841                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
 842                 return;
 843         }
 844
 845         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
 846         uc->entries = 0;
 847         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
 848         mc->entries = 0;
 849
 850         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
 851                 const struct ether_addr *addr
 852                         = (i == index) ? mac_addr : addrs + i;
 853                 struct virtio_net_ctrl_mac *tbl
 854                         = is_multicast_ether_addr(addr) ? mc : uc;
 855
 856                 memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
 857         }
 858
 859         virtio_mac_table_set(hw, uc, mc);
 860 }
 861
 862 static void
 863 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
 864 {
 865         struct virtio_hw *hw = dev->data->dev_private;
 866         struct ether_addr *addrs = dev->data->mac_addrs;
 867         struct virtio_net_ctrl_mac *uc, *mc;
 868         unsigned int i;
 869
 870         if (index >= VIRTIO_MAX_MAC_ADDRS) {
 871                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
 872                 return;
 873         }
 874
 875         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
 876         uc->entries = 0;
 877         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
 878         mc->entries = 0;
 879
 880         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
 881                 struct virtio_net_ctrl_mac *tbl;
 882
 883                 if (i == index || is_zero_ether_addr(addrs + i))
 884                         continue;
 885
 886                 tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
 887                 memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
 888         }
 889
 890         virtio_mac_table_set(hw, uc, mc);
 891 }
 892
 893 static void
 894 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 895 {
 896         struct virtio_hw *hw = dev->data->dev_private;
 897
 898         memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
 899
 900         /* Use atomic update if available */
 901         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
 902                 struct virtio_pmd_ctrl ctrl;
 903                 int len = ETHER_ADDR_LEN;
 904
 905                 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
 906                 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
 907
 908                 memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
 909                 virtio_send_command(hw->cvq, &ctrl, &len, 1);
 910         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
 911                 virtio_set_hwaddr(hw);
 912 }
 913
 914 static int
 915 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
 916 {
 917         struct virtio_hw *hw = dev->data->dev_private;
 918         struct virtio_pmd_ctrl ctrl;
 919         int len;
 920
 921         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
 922                 return -ENOTSUP;
 923
 924         ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
 925         ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
 926         memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
 927         len = sizeof(vlan_id);
 928
 929         return virtio_send_command(hw->cvq, &ctrl, &len, 1);
 930 }
 931
 932 static int
 933 virtio_negotiate_features(struct virtio_hw *hw)
 934 {
 935         uint64_t host_features;
 936
 937         /* Prepare guest_features: feature that driver wants to support */
 938         hw->guest_features = VIRTIO_PMD_GUEST_FEATURES;
 939         PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
 940                 hw->guest_features);
 941
 942         /* Read device(host) feature bits */
 943         host_features = hw->vtpci_ops->get_features(hw);
 944         PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
 945                 host_features);
 946
 947         /*
 948          * Negotiate features: Subset of device feature bits are written back
 949          * guest feature bits.
 950          */
 951         hw->guest_features = vtpci_negotiate_features(hw, host_features);
 952         PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
 953                 hw->guest_features);
 954
 955         if (hw->modern) {
 956                 if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
 957                         PMD_INIT_LOG(ERR,
 958                                 "VIRTIO_F_VERSION_1 features is not enabled.");
 959                         return -1;
 960                 }
 961                 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
 962                 if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
 963                         PMD_INIT_LOG(ERR,
 964                                 "failed to set FEATURES_OK status!");
 965                         return -1;
 966                 }
 967         }
 968
 969         return 0;
 970 }
 971
 972 /*
 973  * Process Virtio Config changed interrupt and call the callback
 974  * if link state changed.
 975  */
 976 static void
 977 virtio_interrupt_handler(__rte_unused struct rte_intr_handle *handle,
 978                          void *param)
 979 {
 980         struct rte_eth_dev *dev = param;
 981         struct virtio_hw *hw = dev->data->dev_private;
 982         uint8_t isr;
 983
 984         /* Read interrupt status which clears interrupt */
 985         isr = vtpci_isr(hw);
 986         PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
 987
 988         if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0)
 989                 PMD_DRV_LOG(ERR, "interrupt enable failed");
 990
 991         if (isr & VIRTIO_PCI_ISR_CONFIG) {
 992                 if (virtio_dev_link_update(dev, 0) == 0)
 993                         _rte_eth_dev_callback_process(dev,
 994                                                       RTE_ETH_EVENT_INTR_LSC);
 995         }
 996
 997 }
 998
 999 static void
1000 rx_func_get(struct rte_eth_dev *eth_dev)
1001 {
1002         struct virtio_hw *hw = eth_dev->data->dev_private;
1003         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF))
1004                 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1005         else
1006                 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1007 }
1008
1009 /*
1010  * This function is based on probe() function in virtio_pci.c
1011  * It returns 0 on success.
1012  */
1013 static int
1014 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1015 {
1016         struct virtio_hw *hw = eth_dev->data->dev_private;
1017         struct virtio_net_config *config;
1018         struct virtio_net_config local_config;
1019         struct rte_pci_device *pci_dev;
1020         int ret;
1021
1022         RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr));
1023
1024         eth_dev->dev_ops = &virtio_eth_dev_ops;
1025         eth_dev->tx_pkt_burst = &virtio_xmit_pkts;
1026
1027         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1028                 rx_func_get(eth_dev);
1029                 return 0;
1030         }
1031
1032         /* Allocate memory for storing MAC addresses */
1033         eth_dev->data->mac_addrs = rte_zmalloc("virtio", VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN, 0);
1034         if (eth_dev->data->mac_addrs == NULL) {
1035                 PMD_INIT_LOG(ERR,
1036                         "Failed to allocate %d bytes needed to store MAC addresses",
1037                         VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN);
1038                 return -ENOMEM;
1039         }
1040
1041         pci_dev = eth_dev->pci_dev;
1042
1043         ret = vtpci_init(pci_dev, hw);
1044         if (ret)
1045                 return ret;
1046
1047         /* Reset the device although not necessary at startup */
1048         vtpci_reset(hw);
1049
1050         /* Tell the host we've noticed this device. */
1051         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1052
1053         /* Tell the host we've known how to drive the device. */
1054         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1055         if (virtio_negotiate_features(hw) < 0)
1056                 return -1;
1057
1058         /* If host does not support status then disable LSC */
1059         if (!vtpci_with_feature(hw, VIRTIO_NET_F_STATUS))
1060                 pci_dev->driver->drv_flags &= ~RTE_PCI_DRV_INTR_LSC;
1061
1062         rte_eth_copy_pci_info(eth_dev, pci_dev);
1063
1064         rx_func_get(eth_dev);
1065
1066         /* Setting up rx_header size for the device */
1067         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1068             vtpci_with_feature(hw, VIRTIO_F_VERSION_1))
1069                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1070         else
1071                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1072
1073         /* Copy the permanent MAC address to: virtio_hw */
1074         virtio_get_hwaddr(hw);
1075         ether_addr_copy((struct ether_addr *) hw->mac_addr,
1076                         &eth_dev->data->mac_addrs[0]);
1077         PMD_INIT_LOG(DEBUG,
1078                      "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1079                      hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1080                      hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1081
1082         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1083                 config = &local_config;
1084
1085                 vtpci_read_dev_config(hw,
1086                         offsetof(struct virtio_net_config, mac),
1087                         &config->mac, sizeof(config->mac));
1088
1089                 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1090                         vtpci_read_dev_config(hw,
1091                                 offsetof(struct virtio_net_config, status),
1092                                 &config->status, sizeof(config->status));
1093                 } else {
1094                         PMD_INIT_LOG(DEBUG,
1095                                      "VIRTIO_NET_F_STATUS is not supported");
1096                         config->status = 0;
1097                 }
1098
1099                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1100                         vtpci_read_dev_config(hw,
1101                                 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1102                                 &config->max_virtqueue_pairs,
1103                                 sizeof(config->max_virtqueue_pairs));
1104                 } else {
1105                         PMD_INIT_LOG(DEBUG,
1106                                      "VIRTIO_NET_F_MQ is not supported");
1107                         config->max_virtqueue_pairs = 1;
1108                 }
1109
1110                 hw->max_rx_queues =
1111                         (VIRTIO_MAX_RX_QUEUES < config->max_virtqueue_pairs) ?
1112                         VIRTIO_MAX_RX_QUEUES : config->max_virtqueue_pairs;
1113                 hw->max_tx_queues =
1114                         (VIRTIO_MAX_TX_QUEUES < config->max_virtqueue_pairs) ?
1115                         VIRTIO_MAX_TX_QUEUES : config->max_virtqueue_pairs;
1116
1117                 virtio_dev_cq_queue_setup(eth_dev,
1118                                         config->max_virtqueue_pairs * 2,
1119                                         SOCKET_ID_ANY);
1120
1121                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1122                                 config->max_virtqueue_pairs);
1123                 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1124                 PMD_INIT_LOG(DEBUG,
1125                                 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1126                                 config->mac[0], config->mac[1],
1127                                 config->mac[2], config->mac[3],
1128                                 config->mac[4], config->mac[5]);
1129         } else {
1130                 hw->max_rx_queues = 1;
1131                 hw->max_tx_queues = 1;
1132         }
1133
1134         PMD_INIT_LOG(DEBUG, "hw->max_rx_queues=%d   hw->max_tx_queues=%d",
1135                         hw->max_rx_queues, hw->max_tx_queues);
1136         PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1137                         eth_dev->data->port_id, pci_dev->id.vendor_id,
1138                         pci_dev->id.device_id);
1139
1140         /* Setup interrupt callback  */
1141         if (pci_dev->driver->drv_flags & RTE_PCI_DRV_INTR_LSC)
1142                 rte_intr_callback_register(&pci_dev->intr_handle,
1143                                    virtio_interrupt_handler, eth_dev);
1144
1145         virtio_dev_cq_start(eth_dev);
1146
1147         return 0;
1148 }
1149
1150 static int
1151 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1152 {
1153         struct rte_pci_device *pci_dev;
1154         struct virtio_hw *hw = eth_dev->data->dev_private;
1155
1156         PMD_INIT_FUNC_TRACE();
1157
1158         if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1159                 return -EPERM;
1160
1161         /* Close it anyway since there's no way to know if closed */
1162         virtio_dev_close(eth_dev);
1163
1164         pci_dev = eth_dev->pci_dev;
1165
1166         eth_dev->dev_ops = NULL;
1167         eth_dev->tx_pkt_burst = NULL;
1168         eth_dev->rx_pkt_burst = NULL;
1169
1170         virtio_dev_queue_release(hw->cvq);
1171
1172         rte_free(eth_dev->data->mac_addrs);
1173         eth_dev->data->mac_addrs = NULL;
1174
1175         /* reset interrupt callback  */
1176         if (pci_dev->driver->drv_flags & RTE_PCI_DRV_INTR_LSC)
1177                 rte_intr_callback_unregister(&pci_dev->intr_handle,
1178                                                 virtio_interrupt_handler,
1179                                                 eth_dev);
1180         rte_eal_pci_unmap_device(pci_dev);
1181
1182         PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1183
1184         return 0;
1185 }
1186
1187 static struct eth_driver rte_virtio_pmd = {
1188         .pci_drv = {
1189                 .name = "rte_virtio_pmd",
1190                 .id_table = pci_id_virtio_map,
1191                 .drv_flags = RTE_PCI_DRV_DETACHABLE,
1192         },
1193         .eth_dev_init = eth_virtio_dev_init,
1194         .eth_dev_uninit = eth_virtio_dev_uninit,
1195         .dev_private_size = sizeof(struct virtio_hw),
1196 };
1197
1198 /*
1199  * Driver initialization routine.
1200  * Invoked once at EAL init time.
1201  * Register itself as the [Poll Mode] Driver of PCI virtio devices.
1202  * Returns 0 on success.
1203  */
1204 static int
1205 rte_virtio_pmd_init(const char *name __rte_unused,
1206                     const char *param __rte_unused)
1207 {
1208         if (rte_eal_iopl_init() != 0) {
1209                 PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1210                 return -1;
1211         }
1212
1213         rte_eth_driver_register(&rte_virtio_pmd);
1214         return 0;
1215 }
1216
1217 /*
1218  * Configure virtio device
1219  * It returns 0 on success.
1220  */
1221 static int
1222 virtio_dev_configure(struct rte_eth_dev *dev)
1223 {
1224         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1225         struct virtio_hw *hw = dev->data->dev_private;
1226         struct rte_pci_device *pci_dev = dev->pci_dev;
1227
1228         PMD_INIT_LOG(DEBUG, "configure");
1229
1230         if (rxmode->hw_ip_checksum) {
1231                 PMD_DRV_LOG(ERR, "HW IP checksum not supported");
1232                 return -EINVAL;
1233         }
1234
1235         hw->vlan_strip = rxmode->hw_vlan_strip;
1236
1237         if (rxmode->hw_vlan_filter
1238             && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1239                 PMD_DRV_LOG(NOTICE,
1240                             "vlan filtering not available on this host");
1241                 return -ENOTSUP;
1242         }
1243
1244         if (pci_dev->driver->drv_flags & RTE_PCI_DRV_INTR_LSC)
1245                 if (vtpci_irq_config(hw, 0) == VIRTIO_MSI_NO_VECTOR) {
1246                         PMD_DRV_LOG(ERR, "failed to set config vector");
1247                         return -EBUSY;
1248                 }
1249
1250         return 0;
1251 }
1252
1253
1254 static int
1255 virtio_dev_start(struct rte_eth_dev *dev)
1256 {
1257         uint16_t nb_queues, i;
1258         struct virtio_hw *hw = dev->data->dev_private;
1259         struct rte_pci_device *pci_dev = dev->pci_dev;
1260
1261         /* check if lsc interrupt feature is enabled */
1262         if (dev->data->dev_conf.intr_conf.lsc) {
1263                 if (!(pci_dev->driver->drv_flags & RTE_PCI_DRV_INTR_LSC)) {
1264                         PMD_DRV_LOG(ERR, "link status not supported by host");
1265                         return -ENOTSUP;
1266                 }
1267
1268                 if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0) {
1269                         PMD_DRV_LOG(ERR, "interrupt enable failed");
1270                         return -EIO;
1271                 }
1272         }
1273
1274         /* Initialize Link state */
1275         virtio_dev_link_update(dev, 0);
1276
1277         /* On restart after stop do not touch queues */
1278         if (hw->started)
1279                 return 0;
1280
1281         /* Do final configuration before rx/tx engine starts */
1282         virtio_dev_rxtx_start(dev);
1283         vtpci_reinit_complete(hw);
1284
1285         hw->started = 1;
1286
1287         /*Notify the backend
1288          *Otherwise the tap backend might already stop its queue due to fullness.
1289          *vhost backend will have no chance to be waked up
1290          */
1291         nb_queues = dev->data->nb_rx_queues;
1292         if (nb_queues > 1) {
1293                 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1294                         return -EINVAL;
1295         }
1296
1297         PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1298
1299         for (i = 0; i < nb_queues; i++)
1300                 virtqueue_notify(dev->data->rx_queues[i]);
1301
1302         PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1303
1304         for (i = 0; i < dev->data->nb_rx_queues; i++)
1305                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1306
1307         for (i = 0; i < dev->data->nb_tx_queues; i++)
1308                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1309
1310         return 0;
1311 }
1312
1313 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1314 {
1315         struct rte_mbuf *buf;
1316         int i, mbuf_num = 0;
1317
1318         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1319                 PMD_INIT_LOG(DEBUG,
1320                              "Before freeing rxq[%d] used and unused buf", i);
1321                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1322
1323                 PMD_INIT_LOG(DEBUG, "rx_queues[%d]=%p",
1324                                 i, dev->data->rx_queues[i]);
1325                 while ((buf = (struct rte_mbuf *)virtqueue_detatch_unused(
1326                                         dev->data->rx_queues[i])) != NULL) {
1327                         rte_pktmbuf_free(buf);
1328                         mbuf_num++;
1329                 }
1330
1331                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1332                 PMD_INIT_LOG(DEBUG,
1333                              "After freeing rxq[%d] used and unused buf", i);
1334                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1335         }
1336
1337         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1338                 PMD_INIT_LOG(DEBUG,
1339                              "Before freeing txq[%d] used and unused bufs",
1340                              i);
1341                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1342
1343                 mbuf_num = 0;
1344                 while ((buf = (struct rte_mbuf *)virtqueue_detatch_unused(
1345                                         dev->data->tx_queues[i])) != NULL) {
1346                         rte_pktmbuf_free(buf);
1347
1348                         mbuf_num++;
1349                 }
1350
1351                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1352                 PMD_INIT_LOG(DEBUG,
1353                              "After freeing txq[%d] used and unused buf", i);
1354                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1355         }
1356 }
1357
1358 /*
1359  * Stop device: disable interrupt and mark link down
1360  */
1361 static void
1362 virtio_dev_stop(struct rte_eth_dev *dev)
1363 {
1364         struct rte_eth_link link;
1365         struct virtio_hw *hw = dev->data->dev_private;
1366
1367         PMD_INIT_LOG(DEBUG, "stop");
1368
1369         hw->started = 0;
1370
1371         if (dev->data->dev_conf.intr_conf.lsc)
1372                 rte_intr_disable(&dev->pci_dev->intr_handle);
1373
1374         memset(&link, 0, sizeof(link));
1375         virtio_dev_atomic_write_link_status(dev, &link);
1376 }
1377
1378 static int
1379 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
1380 {
1381         struct rte_eth_link link, old;
1382         uint16_t status;
1383         struct virtio_hw *hw = dev->data->dev_private;
1384         memset(&link, 0, sizeof(link));
1385         virtio_dev_atomic_read_link_status(dev, &link);
1386         old = link;
1387         link.link_duplex = FULL_DUPLEX;
1388         link.link_speed  = SPEED_10G;
1389
1390         if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1391                 PMD_INIT_LOG(DEBUG, "Get link status from hw");
1392                 vtpci_read_dev_config(hw,
1393                                 offsetof(struct virtio_net_config, status),
1394                                 &status, sizeof(status));
1395                 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
1396                         link.link_status = 0;
1397                         PMD_INIT_LOG(DEBUG, "Port %d is down",
1398                                      dev->data->port_id);
1399                 } else {
1400                         link.link_status = 1;
1401                         PMD_INIT_LOG(DEBUG, "Port %d is up",
1402                                      dev->data->port_id);
1403                 }
1404         } else {
1405                 link.link_status = 1;   /* Link up */
1406         }
1407         virtio_dev_atomic_write_link_status(dev, &link);
1408
1409         return (old.link_status == link.link_status) ? -1 : 0;
1410 }
1411
1412 static void
1413 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1414 {
1415         struct virtio_hw *hw = dev->data->dev_private;
1416
1417         dev_info->driver_name = dev->driver->pci_drv.name;
1418         dev_info->max_rx_queues = (uint16_t)hw->max_rx_queues;
1419         dev_info->max_tx_queues = (uint16_t)hw->max_tx_queues;
1420         dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
1421         dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
1422         dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
1423         dev_info->default_txconf = (struct rte_eth_txconf) {
1424                 .txq_flags = ETH_TXQ_FLAGS_NOOFFLOADS
1425         };
1426 }
1427
1428 /*
1429  * It enables testpmd to collect per queue stats.
1430  */
1431 static int
1432 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
1433 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
1434 __rte_unused uint8_t is_rx)
1435 {
1436         return 0;
1437 }
1438
1439 static struct rte_driver rte_virtio_driver = {
1440         .type = PMD_PDEV,
1441         .init = rte_virtio_pmd_init,
1442 };
1443
1444 PMD_REGISTER_DRIVER(rte_virtio_driver);