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                 const struct rte_memzone *hdr_mz;
 391                 struct virtio_tx_region *txr;
 392                 unsigned int i;
 393
 394                 /*
 395                  * For each xmit packet, allocate a virtio_net_hdr
 396                  * and indirect ring elements
 397                  */
 398                 snprintf(vq_name, sizeof(vq_name), "port%d_tvq%d_hdrzone",
 399                          dev->data->port_id, queue_idx);
 400                 hdr_mz = rte_memzone_reserve_aligned(vq_name,
 401                                                      vq_size * sizeof(*txr),
 402                                                      socket_id, 0,
 403                                                      RTE_CACHE_LINE_SIZE);
 404                 if (hdr_mz == NULL) {
 405                         if (rte_errno == EEXIST)
 406                                 hdr_mz = rte_memzone_lookup(vq_name);
 407                         if (hdr_mz == NULL) {
 408                                 rte_free(vq);
 409                                 return -ENOMEM;
 410                         }
 411                 }
 412                 vq->virtio_net_hdr_mz = hdr_mz;
 413                 vq->virtio_net_hdr_mem = hdr_mz->phys_addr;
 414
 415                 txr = hdr_mz->addr;
 416                 memset(txr, 0, vq_size * sizeof(*txr));
 417                 for (i = 0; i < vq_size; i++) {
 418                         struct vring_desc *start_dp = txr[i].tx_indir;
 419
 420                         vring_desc_init(start_dp, RTE_DIM(txr[i].tx_indir));
 421
 422                         /* first indirect descriptor is always the tx header */
 423                         start_dp->addr = vq->virtio_net_hdr_mem
 424                                 + i * sizeof(*txr)
 425                                 + offsetof(struct virtio_tx_region, tx_hdr);
 426
 427                         start_dp->len = vq->hw->vtnet_hdr_size;
 428                         start_dp->flags = VRING_DESC_F_NEXT;
 429                 }
 430
 431         } else if (queue_type == VTNET_CQ) {
 432                 /* Allocate a page for control vq command, data and status */
 433                 snprintf(vq_name, sizeof(vq_name), "port%d_cvq_hdrzone",
 434                         dev->data->port_id);
 435                 vq->virtio_net_hdr_mz = rte_memzone_reserve_aligned(vq_name,
 436                         PAGE_SIZE, socket_id, 0, RTE_CACHE_LINE_SIZE);
 437                 if (vq->virtio_net_hdr_mz == NULL) {
 438                         if (rte_errno == EEXIST)
 439                                 vq->virtio_net_hdr_mz =
 440                                         rte_memzone_lookup(vq_name);
 441                         if (vq->virtio_net_hdr_mz == NULL) {
 442                                 rte_free(vq);
 443                                 return -ENOMEM;
 444                         }
 445                 }
 446                 vq->virtio_net_hdr_mem =
 447                         vq->virtio_net_hdr_mz->phys_addr;
 448                 memset(vq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
 449         }
 450
 451         hw->vtpci_ops->setup_queue(hw, vq);
 452
 453         *pvq = vq;
 454         return 0;
 455 }
 456
 457 static int
 458 virtio_dev_cq_queue_setup(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx,
 459                 uint32_t socket_id)
 460 {
 461         struct virtqueue *vq;
 462         int ret;
 463         struct virtio_hw *hw = dev->data->dev_private;
 464
 465         PMD_INIT_FUNC_TRACE();
 466         ret = virtio_dev_queue_setup(dev, VTNET_CQ, VTNET_SQ_CQ_QUEUE_IDX,
 467                         vtpci_queue_idx, 0, socket_id, &vq);
 468         if (ret < 0) {
 469                 PMD_INIT_LOG(ERR, "control vq initialization failed");
 470                 return ret;
 471         }
 472
 473         hw->cvq = vq;
 474         return 0;
 475 }
 476
 477 static void
 478 virtio_free_queues(struct rte_eth_dev *dev)
 479 {
 480         unsigned int i;
 481
 482         for (i = 0; i < dev->data->nb_rx_queues; i++)
 483                 virtio_dev_rx_queue_release(dev->data->rx_queues[i]);
 484
 485         dev->data->nb_rx_queues = 0;
 486
 487         for (i = 0; i < dev->data->nb_tx_queues; i++)
 488                 virtio_dev_tx_queue_release(dev->data->tx_queues[i]);
 489
 490         dev->data->nb_tx_queues = 0;
 491 }
 492
 493 static void
 494 virtio_dev_close(struct rte_eth_dev *dev)
 495 {
 496         struct virtio_hw *hw = dev->data->dev_private;
 497         struct rte_pci_device *pci_dev = dev->pci_dev;
 498
 499         PMD_INIT_LOG(DEBUG, "virtio_dev_close");
 500
 501         if (hw->started == 1)
 502                 virtio_dev_stop(dev);
 503
 504         /* reset the NIC */
 505         if (pci_dev->driver->drv_flags & RTE_PCI_DRV_INTR_LSC)
 506                 vtpci_irq_config(hw, VIRTIO_MSI_NO_VECTOR);
 507         vtpci_reset(hw);
 508         virtio_dev_free_mbufs(dev);
 509         virtio_free_queues(dev);
 510 }
 511
 512 static void
 513 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
 514 {
 515         struct virtio_hw *hw = dev->data->dev_private;
 516         struct virtio_pmd_ctrl ctrl;
 517         int dlen[1];
 518         int ret;
 519
 520         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 521                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 522                 return;
 523         }
 524
 525         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 526         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 527         ctrl.data[0] = 1;
 528         dlen[0] = 1;
 529
 530         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 531         if (ret)
 532                 PMD_INIT_LOG(ERR, "Failed to enable promisc");
 533 }
 534
 535 static void
 536 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
 537 {
 538         struct virtio_hw *hw = dev->data->dev_private;
 539         struct virtio_pmd_ctrl ctrl;
 540         int dlen[1];
 541         int ret;
 542
 543         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 544                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 545                 return;
 546         }
 547
 548         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 549         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 550         ctrl.data[0] = 0;
 551         dlen[0] = 1;
 552
 553         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 554         if (ret)
 555                 PMD_INIT_LOG(ERR, "Failed to disable promisc");
 556 }
 557
 558 static void
 559 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
 560 {
 561         struct virtio_hw *hw = dev->data->dev_private;
 562         struct virtio_pmd_ctrl ctrl;
 563         int dlen[1];
 564         int ret;
 565
 566         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 567                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 568                 return;
 569         }
 570
 571         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 572         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 573         ctrl.data[0] = 1;
 574         dlen[0] = 1;
 575
 576         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 577         if (ret)
 578                 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
 579 }
 580
 581 static void
 582 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
 583 {
 584         struct virtio_hw *hw = dev->data->dev_private;
 585         struct virtio_pmd_ctrl ctrl;
 586         int dlen[1];
 587         int ret;
 588
 589         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 590                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 591                 return;
 592         }
 593
 594         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 595         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 596         ctrl.data[0] = 0;
 597         dlen[0] = 1;
 598
 599         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 600         if (ret)
 601                 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
 602 }
 603
 604 /*
 605  * dev_ops for virtio, bare necessities for basic operation
 606  */
 607 static const struct eth_dev_ops virtio_eth_dev_ops = {
 608         .dev_configure           = virtio_dev_configure,
 609         .dev_start               = virtio_dev_start,
 610         .dev_stop                = virtio_dev_stop,
 611         .dev_close               = virtio_dev_close,
 612         .promiscuous_enable      = virtio_dev_promiscuous_enable,
 613         .promiscuous_disable     = virtio_dev_promiscuous_disable,
 614         .allmulticast_enable     = virtio_dev_allmulticast_enable,
 615         .allmulticast_disable    = virtio_dev_allmulticast_disable,
 616
 617         .dev_infos_get           = virtio_dev_info_get,
 618         .stats_get               = virtio_dev_stats_get,
 619         .xstats_get              = virtio_dev_xstats_get,
 620         .stats_reset             = virtio_dev_stats_reset,
 621         .xstats_reset            = virtio_dev_stats_reset,
 622         .link_update             = virtio_dev_link_update,
 623         .rx_queue_setup          = virtio_dev_rx_queue_setup,
 624         .rx_queue_release        = virtio_dev_rx_queue_release,
 625         .tx_queue_setup          = virtio_dev_tx_queue_setup,
 626         .tx_queue_release        = virtio_dev_tx_queue_release,
 627         /* collect stats per queue */
 628         .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
 629         .vlan_filter_set         = virtio_vlan_filter_set,
 630         .mac_addr_add            = virtio_mac_addr_add,
 631         .mac_addr_remove         = virtio_mac_addr_remove,
 632         .mac_addr_set            = virtio_mac_addr_set,
 633 };
 634
 635 static inline int
 636 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
 637                                 struct rte_eth_link *link)
 638 {
 639         struct rte_eth_link *dst = link;
 640         struct rte_eth_link *src = &(dev->data->dev_link);
 641
 642         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
 643                         *(uint64_t *)src) == 0)
 644                 return -1;
 645
 646         return 0;
 647 }
 648
 649 /**
 650  * Atomically writes the link status information into global
 651  * structure rte_eth_dev.
 652  *
 653  * @param dev
 654  *   - Pointer to the structure rte_eth_dev to read from.
 655  *   - Pointer to the buffer to be saved with the link status.
 656  *
 657  * @return
 658  *   - On success, zero.
 659  *   - On failure, negative value.
 660  */
 661 static inline int
 662 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
 663                 struct rte_eth_link *link)
 664 {
 665         struct rte_eth_link *dst = &(dev->data->dev_link);
 666         struct rte_eth_link *src = link;
 667
 668         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
 669                                         *(uint64_t *)src) == 0)
 670                 return -1;
 671
 672         return 0;
 673 }
 674
 675 static void
 676 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 677 {
 678         unsigned i;
 679
 680         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 681                 const struct virtqueue *txvq = dev->data->tx_queues[i];
 682                 if (txvq == NULL)
 683                         continue;
 684
 685                 stats->opackets += txvq->packets;
 686                 stats->obytes += txvq->bytes;
 687                 stats->oerrors += txvq->errors;
 688
 689                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 690                         stats->q_opackets[i] = txvq->packets;
 691                         stats->q_obytes[i] = txvq->bytes;
 692                 }
 693         }
 694
 695         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 696                 const struct virtqueue *rxvq = dev->data->rx_queues[i];
 697                 if (rxvq == NULL)
 698                         continue;
 699
 700                 stats->ipackets += rxvq->packets;
 701                 stats->ibytes += rxvq->bytes;
 702                 stats->ierrors += rxvq->errors;
 703
 704                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 705                         stats->q_ipackets[i] = rxvq->packets;
 706                         stats->q_ibytes[i] = rxvq->bytes;
 707                 }
 708         }
 709
 710         stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
 711 }
 712
 713 static int
 714 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstats *xstats,
 715                       unsigned n)
 716 {
 717         unsigned i;
 718         unsigned count = 0;
 719
 720         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_Q_XSTATS +
 721                 dev->data->nb_rx_queues * VIRTIO_NB_Q_XSTATS;
 722
 723         if (n < nstats)
 724                 return nstats;
 725
 726         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 727                 struct virtqueue *rxvq = dev->data->rx_queues[i];
 728
 729                 if (rxvq == NULL)
 730                         continue;
 731
 732                 unsigned t;
 733
 734                 for (t = 0; t < VIRTIO_NB_Q_XSTATS; t++) {
 735                         snprintf(xstats[count].name, sizeof(xstats[count].name),
 736                                  "rx_q%u_%s", i,
 737                                  rte_virtio_q_stat_strings[t].name);
 738                         xstats[count].value = *(uint64_t *)(((char *)rxvq) +
 739                                 rte_virtio_q_stat_strings[t].offset);
 740                         count++;
 741                 }
 742         }
 743
 744         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 745                 struct virtqueue *txvq = dev->data->tx_queues[i];
 746
 747                 if (txvq == NULL)
 748                         continue;
 749
 750                 unsigned t;
 751
 752                 for (t = 0; t < VIRTIO_NB_Q_XSTATS; t++) {
 753                         snprintf(xstats[count].name, sizeof(xstats[count].name),
 754                                  "tx_q%u_%s", i,
 755                                  rte_virtio_q_stat_strings[t].name);
 756                         xstats[count].value = *(uint64_t *)(((char *)txvq) +
 757                                 rte_virtio_q_stat_strings[t].offset);
 758                         count++;
 759                 }
 760         }
 761
 762         return count;
 763 }
 764
 765 static void
 766 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 767 {
 768         virtio_update_stats(dev, stats);
 769 }
 770
 771 static void
 772 virtio_dev_stats_reset(struct rte_eth_dev *dev)
 773 {
 774         unsigned int i;
 775
 776         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 777                 struct virtqueue *txvq = dev->data->tx_queues[i];
 778                 if (txvq == NULL)
 779                         continue;
 780
 781                 txvq->packets = 0;
 782                 txvq->bytes = 0;
 783                 txvq->errors = 0;
 784                 txvq->multicast = 0;
 785                 txvq->broadcast = 0;
 786                 memset(txvq->size_bins, 0, sizeof(txvq->size_bins[0]) * 8);
 787         }
 788
 789         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 790                 struct virtqueue *rxvq = dev->data->rx_queues[i];
 791                 if (rxvq == NULL)
 792                         continue;
 793
 794                 rxvq->packets = 0;
 795                 rxvq->bytes = 0;
 796                 rxvq->errors = 0;
 797                 rxvq->multicast = 0;
 798                 rxvq->broadcast = 0;
 799                 memset(rxvq->size_bins, 0, sizeof(rxvq->size_bins[0]) * 8);
 800         }
 801 }
 802
 803 static void
 804 virtio_set_hwaddr(struct virtio_hw *hw)
 805 {
 806         vtpci_write_dev_config(hw,
 807                         offsetof(struct virtio_net_config, mac),
 808                         &hw->mac_addr, ETHER_ADDR_LEN);
 809 }
 810
 811 static void
 812 virtio_get_hwaddr(struct virtio_hw *hw)
 813 {
 814         if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
 815                 vtpci_read_dev_config(hw,
 816                         offsetof(struct virtio_net_config, mac),
 817                         &hw->mac_addr, ETHER_ADDR_LEN);
 818         } else {
 819                 eth_random_addr(&hw->mac_addr[0]);
 820                 virtio_set_hwaddr(hw);
 821         }
 822 }
 823
 824 static void
 825 virtio_mac_table_set(struct virtio_hw *hw,
 826                      const struct virtio_net_ctrl_mac *uc,
 827                      const struct virtio_net_ctrl_mac *mc)
 828 {
 829         struct virtio_pmd_ctrl ctrl;
 830         int err, len[2];
 831
 832         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
 833                 PMD_DRV_LOG(INFO, "host does not support mac table\n");
 834                 return;
 835         }
 836
 837         ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
 838         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
 839
 840         len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
 841         memcpy(ctrl.data, uc, len[0]);
 842
 843         len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
 844         memcpy(ctrl.data + len[0], mc, len[1]);
 845
 846         err = virtio_send_command(hw->cvq, &ctrl, len, 2);
 847         if (err != 0)
 848                 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
 849 }
 850
 851 static void
 852 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
 853                     uint32_t index, uint32_t vmdq __rte_unused)
 854 {
 855         struct virtio_hw *hw = dev->data->dev_private;
 856         const struct ether_addr *addrs = dev->data->mac_addrs;
 857         unsigned int i;
 858         struct virtio_net_ctrl_mac *uc, *mc;
 859
 860         if (index >= VIRTIO_MAX_MAC_ADDRS) {
 861                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
 862                 return;
 863         }
 864
 865         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
 866         uc->entries = 0;
 867         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
 868         mc->entries = 0;
 869
 870         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
 871                 const struct ether_addr *addr
 872                         = (i == index) ? mac_addr : addrs + i;
 873                 struct virtio_net_ctrl_mac *tbl
 874                         = is_multicast_ether_addr(addr) ? mc : uc;
 875
 876                 memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
 877         }
 878
 879         virtio_mac_table_set(hw, uc, mc);
 880 }
 881
 882 static void
 883 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
 884 {
 885         struct virtio_hw *hw = dev->data->dev_private;
 886         struct ether_addr *addrs = dev->data->mac_addrs;
 887         struct virtio_net_ctrl_mac *uc, *mc;
 888         unsigned int i;
 889
 890         if (index >= VIRTIO_MAX_MAC_ADDRS) {
 891                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
 892                 return;
 893         }
 894
 895         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
 896         uc->entries = 0;
 897         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
 898         mc->entries = 0;
 899
 900         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
 901                 struct virtio_net_ctrl_mac *tbl;
 902
 903                 if (i == index || is_zero_ether_addr(addrs + i))
 904                         continue;
 905
 906                 tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
 907                 memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
 908         }
 909
 910         virtio_mac_table_set(hw, uc, mc);
 911 }
 912
 913 static void
 914 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 915 {
 916         struct virtio_hw *hw = dev->data->dev_private;
 917
 918         memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
 919
 920         /* Use atomic update if available */
 921         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
 922                 struct virtio_pmd_ctrl ctrl;
 923                 int len = ETHER_ADDR_LEN;
 924
 925                 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
 926                 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
 927
 928                 memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
 929                 virtio_send_command(hw->cvq, &ctrl, &len, 1);
 930         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
 931                 virtio_set_hwaddr(hw);
 932 }
 933
 934 static int
 935 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
 936 {
 937         struct virtio_hw *hw = dev->data->dev_private;
 938         struct virtio_pmd_ctrl ctrl;
 939         int len;
 940
 941         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
 942                 return -ENOTSUP;
 943
 944         ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
 945         ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
 946         memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
 947         len = sizeof(vlan_id);
 948
 949         return virtio_send_command(hw->cvq, &ctrl, &len, 1);
 950 }
 951
 952 static int
 953 virtio_negotiate_features(struct virtio_hw *hw)
 954 {
 955         uint64_t host_features;
 956
 957         /* Prepare guest_features: feature that driver wants to support */
 958         hw->guest_features = VIRTIO_PMD_GUEST_FEATURES;
 959         PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
 960                 hw->guest_features);
 961
 962         /* Read device(host) feature bits */
 963         host_features = hw->vtpci_ops->get_features(hw);
 964         PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
 965                 host_features);
 966
 967         /*
 968          * Negotiate features: Subset of device feature bits are written back
 969          * guest feature bits.
 970          */
 971         hw->guest_features = vtpci_negotiate_features(hw, host_features);
 972         PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
 973                 hw->guest_features);
 974
 975         if (hw->modern) {
 976                 if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
 977                         PMD_INIT_LOG(ERR,
 978                                 "VIRTIO_F_VERSION_1 features is not enabled.");
 979                         return -1;
 980                 }
 981                 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
 982                 if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
 983                         PMD_INIT_LOG(ERR,
 984                                 "failed to set FEATURES_OK status!");
 985                         return -1;
 986                 }
 987         }
 988
 989         return 0;
 990 }
 991
 992 /*
 993  * Process Virtio Config changed interrupt and call the callback
 994  * if link state changed.
 995  */
 996 static void
 997 virtio_interrupt_handler(__rte_unused struct rte_intr_handle *handle,
 998                          void *param)
 999 {
1000         struct rte_eth_dev *dev = param;
1001         struct virtio_hw *hw = dev->data->dev_private;
1002         uint8_t isr;
1003
1004         /* Read interrupt status which clears interrupt */
1005         isr = vtpci_isr(hw);
1006         PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1007
1008         if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0)
1009                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1010
1011         if (isr & VIRTIO_PCI_ISR_CONFIG) {
1012                 if (virtio_dev_link_update(dev, 0) == 0)
1013                         _rte_eth_dev_callback_process(dev,
1014                                                       RTE_ETH_EVENT_INTR_LSC);
1015         }
1016
1017 }
1018
1019 static void
1020 rx_func_get(struct rte_eth_dev *eth_dev)
1021 {
1022         struct virtio_hw *hw = eth_dev->data->dev_private;
1023         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF))
1024                 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1025         else
1026                 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1027 }
1028
1029 /*
1030  * This function is based on probe() function in virtio_pci.c
1031  * It returns 0 on success.
1032  */
1033 static int
1034 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1035 {
1036         struct virtio_hw *hw = eth_dev->data->dev_private;
1037         struct virtio_net_config *config;
1038         struct virtio_net_config local_config;
1039         struct rte_pci_device *pci_dev;
1040         int ret;
1041
1042         RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr));
1043
1044         eth_dev->dev_ops = &virtio_eth_dev_ops;
1045         eth_dev->tx_pkt_burst = &virtio_xmit_pkts;
1046
1047         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1048                 rx_func_get(eth_dev);
1049                 return 0;
1050         }
1051
1052         /* Allocate memory for storing MAC addresses */
1053         eth_dev->data->mac_addrs = rte_zmalloc("virtio", VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN, 0);
1054         if (eth_dev->data->mac_addrs == NULL) {
1055                 PMD_INIT_LOG(ERR,
1056                         "Failed to allocate %d bytes needed to store MAC addresses",
1057                         VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN);
1058                 return -ENOMEM;
1059         }
1060
1061         pci_dev = eth_dev->pci_dev;
1062
1063         ret = vtpci_init(pci_dev, hw);
1064         if (ret)
1065                 return ret;
1066
1067         /* Reset the device although not necessary at startup */
1068         vtpci_reset(hw);
1069
1070         /* Tell the host we've noticed this device. */
1071         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1072
1073         /* Tell the host we've known how to drive the device. */
1074         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1075         if (virtio_negotiate_features(hw) < 0)
1076                 return -1;
1077
1078         /* If host does not support status then disable LSC */
1079         if (!vtpci_with_feature(hw, VIRTIO_NET_F_STATUS))
1080                 pci_dev->driver->drv_flags &= ~RTE_PCI_DRV_INTR_LSC;
1081
1082         rte_eth_copy_pci_info(eth_dev, pci_dev);
1083
1084         rx_func_get(eth_dev);
1085
1086         /* Setting up rx_header size for the device */
1087         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1088             vtpci_with_feature(hw, VIRTIO_F_VERSION_1))
1089                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1090         else
1091                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1092
1093         /* Copy the permanent MAC address to: virtio_hw */
1094         virtio_get_hwaddr(hw);
1095         ether_addr_copy((struct ether_addr *) hw->mac_addr,
1096                         &eth_dev->data->mac_addrs[0]);
1097         PMD_INIT_LOG(DEBUG,
1098                      "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1099                      hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1100                      hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1101
1102         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1103                 config = &local_config;
1104
1105                 vtpci_read_dev_config(hw,
1106                         offsetof(struct virtio_net_config, mac),
1107                         &config->mac, sizeof(config->mac));
1108
1109                 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1110                         vtpci_read_dev_config(hw,
1111                                 offsetof(struct virtio_net_config, status),
1112                                 &config->status, sizeof(config->status));
1113                 } else {
1114                         PMD_INIT_LOG(DEBUG,
1115                                      "VIRTIO_NET_F_STATUS is not supported");
1116                         config->status = 0;
1117                 }
1118
1119                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1120                         vtpci_read_dev_config(hw,
1121                                 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1122                                 &config->max_virtqueue_pairs,
1123                                 sizeof(config->max_virtqueue_pairs));
1124                 } else {
1125                         PMD_INIT_LOG(DEBUG,
1126                                      "VIRTIO_NET_F_MQ is not supported");
1127                         config->max_virtqueue_pairs = 1;
1128                 }
1129
1130                 hw->max_rx_queues =
1131                         (VIRTIO_MAX_RX_QUEUES < config->max_virtqueue_pairs) ?
1132                         VIRTIO_MAX_RX_QUEUES : config->max_virtqueue_pairs;
1133                 hw->max_tx_queues =
1134                         (VIRTIO_MAX_TX_QUEUES < config->max_virtqueue_pairs) ?
1135                         VIRTIO_MAX_TX_QUEUES : config->max_virtqueue_pairs;
1136
1137                 virtio_dev_cq_queue_setup(eth_dev,
1138                                         config->max_virtqueue_pairs * 2,
1139                                         SOCKET_ID_ANY);
1140
1141                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1142                                 config->max_virtqueue_pairs);
1143                 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1144                 PMD_INIT_LOG(DEBUG,
1145                                 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1146                                 config->mac[0], config->mac[1],
1147                                 config->mac[2], config->mac[3],
1148                                 config->mac[4], config->mac[5]);
1149         } else {
1150                 hw->max_rx_queues = 1;
1151                 hw->max_tx_queues = 1;
1152         }
1153
1154         PMD_INIT_LOG(DEBUG, "hw->max_rx_queues=%d   hw->max_tx_queues=%d",
1155                         hw->max_rx_queues, hw->max_tx_queues);
1156         PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1157                         eth_dev->data->port_id, pci_dev->id.vendor_id,
1158                         pci_dev->id.device_id);
1159
1160         /* Setup interrupt callback  */
1161         if (pci_dev->driver->drv_flags & RTE_PCI_DRV_INTR_LSC)
1162                 rte_intr_callback_register(&pci_dev->intr_handle,
1163                                    virtio_interrupt_handler, eth_dev);
1164
1165         virtio_dev_cq_start(eth_dev);
1166
1167         return 0;
1168 }
1169
1170 static int
1171 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1172 {
1173         struct rte_pci_device *pci_dev;
1174         struct virtio_hw *hw = eth_dev->data->dev_private;
1175
1176         PMD_INIT_FUNC_TRACE();
1177
1178         if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1179                 return -EPERM;
1180
1181         /* Close it anyway since there's no way to know if closed */
1182         virtio_dev_close(eth_dev);
1183
1184         pci_dev = eth_dev->pci_dev;
1185
1186         eth_dev->dev_ops = NULL;
1187         eth_dev->tx_pkt_burst = NULL;
1188         eth_dev->rx_pkt_burst = NULL;
1189
1190         virtio_dev_queue_release(hw->cvq);
1191
1192         rte_free(eth_dev->data->mac_addrs);
1193         eth_dev->data->mac_addrs = NULL;
1194
1195         /* reset interrupt callback  */
1196         if (pci_dev->driver->drv_flags & RTE_PCI_DRV_INTR_LSC)
1197                 rte_intr_callback_unregister(&pci_dev->intr_handle,
1198                                                 virtio_interrupt_handler,
1199                                                 eth_dev);
1200         rte_eal_pci_unmap_device(pci_dev);
1201
1202         PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1203
1204         return 0;
1205 }
1206
1207 static struct eth_driver rte_virtio_pmd = {
1208         .pci_drv = {
1209                 .name = "rte_virtio_pmd",
1210                 .id_table = pci_id_virtio_map,
1211                 .drv_flags = RTE_PCI_DRV_DETACHABLE,
1212         },
1213         .eth_dev_init = eth_virtio_dev_init,
1214         .eth_dev_uninit = eth_virtio_dev_uninit,
1215         .dev_private_size = sizeof(struct virtio_hw),
1216 };
1217
1218 /*
1219  * Driver initialization routine.
1220  * Invoked once at EAL init time.
1221  * Register itself as the [Poll Mode] Driver of PCI virtio devices.
1222  * Returns 0 on success.
1223  */
1224 static int
1225 rte_virtio_pmd_init(const char *name __rte_unused,
1226                     const char *param __rte_unused)
1227 {
1228         if (rte_eal_iopl_init() != 0) {
1229                 PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1230                 return -1;
1231         }
1232
1233         rte_eth_driver_register(&rte_virtio_pmd);
1234         return 0;
1235 }
1236
1237 /*
1238  * Configure virtio device
1239  * It returns 0 on success.
1240  */
1241 static int
1242 virtio_dev_configure(struct rte_eth_dev *dev)
1243 {
1244         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1245         struct virtio_hw *hw = dev->data->dev_private;
1246         struct rte_pci_device *pci_dev = dev->pci_dev;
1247
1248         PMD_INIT_LOG(DEBUG, "configure");
1249
1250         if (rxmode->hw_ip_checksum) {
1251                 PMD_DRV_LOG(ERR, "HW IP checksum not supported");
1252                 return -EINVAL;
1253         }
1254
1255         hw->vlan_strip = rxmode->hw_vlan_strip;
1256
1257         if (rxmode->hw_vlan_filter
1258             && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1259                 PMD_DRV_LOG(NOTICE,
1260                             "vlan filtering not available on this host");
1261                 return -ENOTSUP;
1262         }
1263
1264         if (pci_dev->driver->drv_flags & RTE_PCI_DRV_INTR_LSC)
1265                 if (vtpci_irq_config(hw, 0) == VIRTIO_MSI_NO_VECTOR) {
1266                         PMD_DRV_LOG(ERR, "failed to set config vector");
1267                         return -EBUSY;
1268                 }
1269
1270         return 0;
1271 }
1272
1273
1274 static int
1275 virtio_dev_start(struct rte_eth_dev *dev)
1276 {
1277         uint16_t nb_queues, i;
1278         struct virtio_hw *hw = dev->data->dev_private;
1279         struct rte_pci_device *pci_dev = dev->pci_dev;
1280
1281         /* check if lsc interrupt feature is enabled */
1282         if (dev->data->dev_conf.intr_conf.lsc) {
1283                 if (!(pci_dev->driver->drv_flags & RTE_PCI_DRV_INTR_LSC)) {
1284                         PMD_DRV_LOG(ERR, "link status not supported by host");
1285                         return -ENOTSUP;
1286                 }
1287
1288                 if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0) {
1289                         PMD_DRV_LOG(ERR, "interrupt enable failed");
1290                         return -EIO;
1291                 }
1292         }
1293
1294         /* Initialize Link state */
1295         virtio_dev_link_update(dev, 0);
1296
1297         /* On restart after stop do not touch queues */
1298         if (hw->started)
1299                 return 0;
1300
1301         /* Do final configuration before rx/tx engine starts */
1302         virtio_dev_rxtx_start(dev);
1303         vtpci_reinit_complete(hw);
1304
1305         hw->started = 1;
1306
1307         /*Notify the backend
1308          *Otherwise the tap backend might already stop its queue due to fullness.
1309          *vhost backend will have no chance to be waked up
1310          */
1311         nb_queues = dev->data->nb_rx_queues;
1312         if (nb_queues > 1) {
1313                 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1314                         return -EINVAL;
1315         }
1316
1317         PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1318
1319         for (i = 0; i < nb_queues; i++)
1320                 virtqueue_notify(dev->data->rx_queues[i]);
1321
1322         PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1323
1324         for (i = 0; i < dev->data->nb_rx_queues; i++)
1325                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1326
1327         for (i = 0; i < dev->data->nb_tx_queues; i++)
1328                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1329
1330         return 0;
1331 }
1332
1333 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1334 {
1335         struct rte_mbuf *buf;
1336         int i, mbuf_num = 0;
1337
1338         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1339                 PMD_INIT_LOG(DEBUG,
1340                              "Before freeing rxq[%d] used and unused buf", i);
1341                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1342
1343                 PMD_INIT_LOG(DEBUG, "rx_queues[%d]=%p",
1344                                 i, dev->data->rx_queues[i]);
1345                 while ((buf = (struct rte_mbuf *)virtqueue_detatch_unused(
1346                                         dev->data->rx_queues[i])) != NULL) {
1347                         rte_pktmbuf_free(buf);
1348                         mbuf_num++;
1349                 }
1350
1351                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1352                 PMD_INIT_LOG(DEBUG,
1353                              "After freeing rxq[%d] used and unused buf", i);
1354                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1355         }
1356
1357         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1358                 PMD_INIT_LOG(DEBUG,
1359                              "Before freeing txq[%d] used and unused bufs",
1360                              i);
1361                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1362
1363                 mbuf_num = 0;
1364                 while ((buf = (struct rte_mbuf *)virtqueue_detatch_unused(
1365                                         dev->data->tx_queues[i])) != NULL) {
1366                         rte_pktmbuf_free(buf);
1367
1368                         mbuf_num++;
1369                 }
1370
1371                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1372                 PMD_INIT_LOG(DEBUG,
1373                              "After freeing txq[%d] used and unused buf", i);
1374                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1375         }
1376 }
1377
1378 /*
1379  * Stop device: disable interrupt and mark link down
1380  */
1381 static void
1382 virtio_dev_stop(struct rte_eth_dev *dev)
1383 {
1384         struct rte_eth_link link;
1385         struct virtio_hw *hw = dev->data->dev_private;
1386
1387         PMD_INIT_LOG(DEBUG, "stop");
1388
1389         hw->started = 0;
1390
1391         if (dev->data->dev_conf.intr_conf.lsc)
1392                 rte_intr_disable(&dev->pci_dev->intr_handle);
1393
1394         memset(&link, 0, sizeof(link));
1395         virtio_dev_atomic_write_link_status(dev, &link);
1396 }
1397
1398 static int
1399 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
1400 {
1401         struct rte_eth_link link, old;
1402         uint16_t status;
1403         struct virtio_hw *hw = dev->data->dev_private;
1404         memset(&link, 0, sizeof(link));
1405         virtio_dev_atomic_read_link_status(dev, &link);
1406         old = link;
1407         link.link_duplex = FULL_DUPLEX;
1408         link.link_speed  = SPEED_10G;
1409
1410         if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1411                 PMD_INIT_LOG(DEBUG, "Get link status from hw");
1412                 vtpci_read_dev_config(hw,
1413                                 offsetof(struct virtio_net_config, status),
1414                                 &status, sizeof(status));
1415                 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
1416                         link.link_status = 0;
1417                         PMD_INIT_LOG(DEBUG, "Port %d is down",
1418                                      dev->data->port_id);
1419                 } else {
1420                         link.link_status = 1;
1421                         PMD_INIT_LOG(DEBUG, "Port %d is up",
1422                                      dev->data->port_id);
1423                 }
1424         } else {
1425                 link.link_status = 1;   /* Link up */
1426         }
1427         virtio_dev_atomic_write_link_status(dev, &link);
1428
1429         return (old.link_status == link.link_status) ? -1 : 0;
1430 }
1431
1432 static void
1433 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1434 {
1435         struct virtio_hw *hw = dev->data->dev_private;
1436
1437         dev_info->driver_name = dev->driver->pci_drv.name;
1438         dev_info->max_rx_queues = (uint16_t)hw->max_rx_queues;
1439         dev_info->max_tx_queues = (uint16_t)hw->max_tx_queues;
1440         dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
1441         dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
1442         dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
1443         dev_info->default_txconf = (struct rte_eth_txconf) {
1444                 .txq_flags = ETH_TXQ_FLAGS_NOOFFLOADS
1445         };
1446 }
1447
1448 /*
1449  * It enables testpmd to collect per queue stats.
1450  */
1451 static int
1452 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
1453 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
1454 __rte_unused uint8_t is_rx)
1455 {
1456         return 0;
1457 }
1458
1459 static struct rte_driver rte_virtio_driver = {
1460         .type = PMD_PDEV,
1461         .init = rte_virtio_pmd_init,
1462 };
1463
1464 PMD_REGISTER_DRIVER(rte_virtio_driver);