drivers/net/virtio/virtio_ethdev.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2014 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 #ifdef RTE_EXEC_ENV_LINUXAPP
  40 #include <dirent.h>
  41 #include <fcntl.h>
  42 #endif
  43
  44 #include <rte_ethdev.h>
  45 #include <rte_memcpy.h>
  46 #include <rte_string_fns.h>
  47 #include <rte_memzone.h>
  48 #include <rte_malloc.h>
  49 #include <rte_atomic.h>
  50 #include <rte_branch_prediction.h>
  51 #include <rte_pci.h>
  52 #include <rte_ether.h>
  53 #include <rte_common.h>
  54
  55 #include <rte_memory.h>
  56 #include <rte_eal.h>
  57 #include <rte_dev.h>
  58
  59 #include "virtio_ethdev.h"
  60 #include "virtio_pci.h"
  61 #include "virtio_logs.h"
  62 #include "virtqueue.h"
  63
  64
  65 static int eth_virtio_dev_init(struct rte_eth_dev *eth_dev);
  66 static int  virtio_dev_configure(struct rte_eth_dev *dev);
  67 static int  virtio_dev_start(struct rte_eth_dev *dev);
  68 static void virtio_dev_stop(struct rte_eth_dev *dev);
  69 static void virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
  70 static void virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
  71 static void virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
  72 static void virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
  73 static void virtio_dev_info_get(struct rte_eth_dev *dev,
  74                                 struct rte_eth_dev_info *dev_info);
  75 static int virtio_dev_link_update(struct rte_eth_dev *dev,
  76         __rte_unused int wait_to_complete);
  77
  78 static void virtio_set_hwaddr(struct virtio_hw *hw);
  79 static void virtio_get_hwaddr(struct virtio_hw *hw);
  80
  81 static void virtio_dev_rx_queue_release(__rte_unused void *rxq);
  82 static void virtio_dev_tx_queue_release(__rte_unused void *txq);
  83
  84 static void virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats);
  85 static void virtio_dev_stats_reset(struct rte_eth_dev *dev);
  86 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
  87 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
  88                                 uint16_t vlan_id, int on);
  89 static void virtio_mac_addr_add(struct rte_eth_dev *dev,
  90                                 struct ether_addr *mac_addr,
  91                                 uint32_t index, uint32_t vmdq __rte_unused);
  92 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
  93 static void virtio_mac_addr_set(struct rte_eth_dev *dev,
  94                                 struct ether_addr *mac_addr);
  95
  96 static int virtio_dev_queue_stats_mapping_set(
  97         __rte_unused struct rte_eth_dev *eth_dev,
  98         __rte_unused uint16_t queue_id,
  99         __rte_unused uint8_t stat_idx,
 100         __rte_unused uint8_t is_rx);
 101
 102 /*
 103  * The set of PCI devices this driver supports
 104  */
 105 static const struct rte_pci_id pci_id_virtio_map[] = {
 106
 107 #define RTE_PCI_DEV_ID_DECL_VIRTIO(vend, dev) {RTE_PCI_DEVICE(vend, dev)},
 108 #include "rte_pci_dev_ids.h"
 109
 110 { .vendor_id = 0, /* sentinel */ },
 111 };
 112
 113 static int
 114 virtio_send_command(struct virtqueue *vq, struct virtio_pmd_ctrl *ctrl,
 115                 int *dlen, int pkt_num)
 116 {
 117         uint16_t head = vq->vq_desc_head_idx, i;
 118         int k, sum = 0;
 119         virtio_net_ctrl_ack status = ~0;
 120         struct virtio_pmd_ctrl result;
 121
 122         ctrl->status = status;
 123
 124         if (!vq->hw->cvq) {
 125                 PMD_INIT_LOG(ERR,
 126                              "%s(): Control queue is not supported.",
 127                              __func__);
 128                 return -1;
 129         }
 130
 131         PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
 132                 "vq->hw->cvq = %p vq = %p",
 133                 vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
 134
 135         if ((vq->vq_free_cnt < ((uint32_t)pkt_num + 2)) || (pkt_num < 1))
 136                 return -1;
 137
 138         memcpy(vq->virtio_net_hdr_mz->addr, ctrl,
 139                 sizeof(struct virtio_pmd_ctrl));
 140
 141         /*
 142          * Format is enforced in qemu code:
 143          * One TX packet for header;
 144          * At least one TX packet per argument;
 145          * One RX packet for ACK.
 146          */
 147         vq->vq_ring.desc[head].flags = VRING_DESC_F_NEXT;
 148         vq->vq_ring.desc[head].addr = vq->virtio_net_hdr_mz->phys_addr;
 149         vq->vq_ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
 150         vq->vq_free_cnt--;
 151         i = vq->vq_ring.desc[head].next;
 152
 153         for (k = 0; k < pkt_num; k++) {
 154                 vq->vq_ring.desc[i].flags = VRING_DESC_F_NEXT;
 155                 vq->vq_ring.desc[i].addr = vq->virtio_net_hdr_mz->phys_addr
 156                         + sizeof(struct virtio_net_ctrl_hdr)
 157                         + sizeof(ctrl->status) + sizeof(uint8_t)*sum;
 158                 vq->vq_ring.desc[i].len = dlen[k];
 159                 sum += dlen[k];
 160                 vq->vq_free_cnt--;
 161                 i = vq->vq_ring.desc[i].next;
 162         }
 163
 164         vq->vq_ring.desc[i].flags = VRING_DESC_F_WRITE;
 165         vq->vq_ring.desc[i].addr = vq->virtio_net_hdr_mz->phys_addr
 166                         + sizeof(struct virtio_net_ctrl_hdr);
 167         vq->vq_ring.desc[i].len = sizeof(ctrl->status);
 168         vq->vq_free_cnt--;
 169
 170         vq->vq_desc_head_idx = vq->vq_ring.desc[i].next;
 171
 172         vq_update_avail_ring(vq, head);
 173         vq_update_avail_idx(vq);
 174
 175         PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
 176
 177         virtqueue_notify(vq);
 178
 179         rte_rmb();
 180         while (vq->vq_used_cons_idx == vq->vq_ring.used->idx) {
 181                 rte_rmb();
 182                 usleep(100);
 183         }
 184
 185         while (vq->vq_used_cons_idx != vq->vq_ring.used->idx) {
 186                 uint32_t idx, desc_idx, used_idx;
 187                 struct vring_used_elem *uep;
 188
 189                 used_idx = (uint32_t)(vq->vq_used_cons_idx
 190                                 & (vq->vq_nentries - 1));
 191                 uep = &vq->vq_ring.used->ring[used_idx];
 192                 idx = (uint32_t) uep->id;
 193                 desc_idx = idx;
 194
 195                 while (vq->vq_ring.desc[desc_idx].flags & VRING_DESC_F_NEXT) {
 196                         desc_idx = vq->vq_ring.desc[desc_idx].next;
 197                         vq->vq_free_cnt++;
 198                 }
 199
 200                 vq->vq_ring.desc[desc_idx].next = vq->vq_desc_head_idx;
 201                 vq->vq_desc_head_idx = idx;
 202
 203                 vq->vq_used_cons_idx++;
 204                 vq->vq_free_cnt++;
 205         }
 206
 207         PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
 208                         vq->vq_free_cnt, vq->vq_desc_head_idx);
 209
 210         memcpy(&result, vq->virtio_net_hdr_mz->addr,
 211                         sizeof(struct virtio_pmd_ctrl));
 212
 213         return result.status;
 214 }
 215
 216 static int
 217 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
 218 {
 219         struct virtio_hw *hw = dev->data->dev_private;
 220         struct virtio_pmd_ctrl ctrl;
 221         int dlen[1];
 222         int ret;
 223
 224         ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
 225         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
 226         memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
 227
 228         dlen[0] = sizeof(uint16_t);
 229
 230         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 231
 232         if (ret) {
 233                 PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
 234                           "failed, this is too late now...");
 235                 return -EINVAL;
 236         }
 237
 238         return 0;
 239 }
 240
 241 int virtio_dev_queue_setup(struct rte_eth_dev *dev,
 242                         int queue_type,
 243                         uint16_t queue_idx,
 244                         uint16_t  vtpci_queue_idx,
 245                         uint16_t nb_desc,
 246                         unsigned int socket_id,
 247                         struct virtqueue **pvq)
 248 {
 249         char vq_name[VIRTQUEUE_MAX_NAME_SZ];
 250         const struct rte_memzone *mz;
 251         uint16_t vq_size;
 252         int size;
 253         struct virtio_hw *hw = dev->data->dev_private;
 254         struct virtqueue  *vq = NULL;
 255
 256         /* Write the virtqueue index to the Queue Select Field */
 257         VIRTIO_WRITE_REG_2(hw, VIRTIO_PCI_QUEUE_SEL, vtpci_queue_idx);
 258         PMD_INIT_LOG(DEBUG, "selecting queue: %d", vtpci_queue_idx);
 259
 260         /*
 261          * Read the virtqueue size from the Queue Size field
 262          * Always power of 2 and if 0 virtqueue does not exist
 263          */
 264         vq_size = VIRTIO_READ_REG_2(hw, VIRTIO_PCI_QUEUE_NUM);
 265         PMD_INIT_LOG(DEBUG, "vq_size: %d nb_desc:%d", vq_size, nb_desc);
 266         if (nb_desc == 0)
 267                 nb_desc = vq_size;
 268         if (vq_size == 0) {
 269                 PMD_INIT_LOG(ERR, "%s: virtqueue does not exist", __func__);
 270                 return -EINVAL;
 271         } else if (!rte_is_power_of_2(vq_size)) {
 272                 PMD_INIT_LOG(ERR, "%s: virtqueue size is not powerof 2", __func__);
 273                 return -EINVAL;
 274         } else if (nb_desc != vq_size) {
 275                 PMD_INIT_LOG(ERR, "Warning: nb_desc(%d) is not equal to vq size (%d), fall to vq size",
 276                         nb_desc, vq_size);
 277                 nb_desc = vq_size;
 278         }
 279
 280         if (queue_type == VTNET_RQ) {
 281                 snprintf(vq_name, sizeof(vq_name), "port%d_rvq%d",
 282                         dev->data->port_id, queue_idx);
 283                 vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
 284                         vq_size * sizeof(struct vq_desc_extra), RTE_CACHE_LINE_SIZE);
 285         } else if (queue_type == VTNET_TQ) {
 286                 snprintf(vq_name, sizeof(vq_name), "port%d_tvq%d",
 287                         dev->data->port_id, queue_idx);
 288                 vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
 289                         vq_size * sizeof(struct vq_desc_extra), RTE_CACHE_LINE_SIZE);
 290         } else if (queue_type == VTNET_CQ) {
 291                 snprintf(vq_name, sizeof(vq_name), "port%d_cvq",
 292                         dev->data->port_id);
 293                 vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
 294                         vq_size * sizeof(struct vq_desc_extra),
 295                         RTE_CACHE_LINE_SIZE);
 296         }
 297         if (vq == NULL) {
 298                 PMD_INIT_LOG(ERR, "%s: Can not allocate virtqueue", __func__);
 299                 return (-ENOMEM);
 300         }
 301
 302         vq->hw = hw;
 303         vq->port_id = dev->data->port_id;
 304         vq->queue_id = queue_idx;
 305         vq->vq_queue_index = vtpci_queue_idx;
 306         vq->vq_nentries = vq_size;
 307         vq->vq_free_cnt = vq_size;
 308
 309         /*
 310          * Reserve a memzone for vring elements
 311          */
 312         size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
 313         vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
 314         PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d", size, vq->vq_ring_size);
 315
 316         mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
 317                 socket_id, 0, VIRTIO_PCI_VRING_ALIGN);
 318         if (mz == NULL) {
 319                 rte_free(vq);
 320                 return -ENOMEM;
 321         }
 322
 323         /*
 324          * Virtio PCI device VIRTIO_PCI_QUEUE_PF register is 32bit,
 325          * and only accepts 32 bit page frame number.
 326          * Check if the allocated physical memory exceeds 16TB.
 327          */
 328         if ((mz->phys_addr + vq->vq_ring_size - 1) >> (VIRTIO_PCI_QUEUE_ADDR_SHIFT + 32)) {
 329                 PMD_INIT_LOG(ERR, "vring address shouldn't be above 16TB!");
 330                 rte_free(vq);
 331                 return -ENOMEM;
 332         }
 333
 334         memset(mz->addr, 0, sizeof(mz->len));
 335         vq->mz = mz;
 336         vq->vq_ring_mem = mz->phys_addr;
 337         vq->vq_ring_virt_mem = mz->addr;
 338         PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%"PRIx64, (uint64_t)mz->phys_addr);
 339         PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%"PRIx64, (uint64_t)(uintptr_t)mz->addr);
 340         vq->virtio_net_hdr_mz  = NULL;
 341         vq->virtio_net_hdr_mem = 0;
 342
 343         if (queue_type == VTNET_TQ) {
 344                 /*
 345                  * For each xmit packet, allocate a virtio_net_hdr
 346                  */
 347                 snprintf(vq_name, sizeof(vq_name), "port%d_tvq%d_hdrzone",
 348                         dev->data->port_id, queue_idx);
 349                 vq->virtio_net_hdr_mz = rte_memzone_reserve_aligned(vq_name,
 350                         vq_size * hw->vtnet_hdr_size,
 351                         socket_id, 0, RTE_CACHE_LINE_SIZE);
 352                 if (vq->virtio_net_hdr_mz == NULL) {
 353                         rte_free(vq);
 354                         return -ENOMEM;
 355                 }
 356                 vq->virtio_net_hdr_mem =
 357                         vq->virtio_net_hdr_mz->phys_addr;
 358                 memset(vq->virtio_net_hdr_mz->addr, 0,
 359                         vq_size * hw->vtnet_hdr_size);
 360         } else if (queue_type == VTNET_CQ) {
 361                 /* Allocate a page for control vq command, data and status */
 362                 snprintf(vq_name, sizeof(vq_name), "port%d_cvq_hdrzone",
 363                         dev->data->port_id);
 364                 vq->virtio_net_hdr_mz = rte_memzone_reserve_aligned(vq_name,
 365                         PAGE_SIZE, socket_id, 0, RTE_CACHE_LINE_SIZE);
 366                 if (vq->virtio_net_hdr_mz == NULL) {
 367                         rte_free(vq);
 368                         return -ENOMEM;
 369                 }
 370                 vq->virtio_net_hdr_mem =
 371                         vq->virtio_net_hdr_mz->phys_addr;
 372                 memset(vq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
 373         }
 374
 375         /*
 376          * Set guest physical address of the virtqueue
 377          * in VIRTIO_PCI_QUEUE_PFN config register of device
 378          */
 379         VIRTIO_WRITE_REG_4(hw, VIRTIO_PCI_QUEUE_PFN,
 380                         mz->phys_addr >> VIRTIO_PCI_QUEUE_ADDR_SHIFT);
 381         *pvq = vq;
 382         return 0;
 383 }
 384
 385 static int
 386 virtio_dev_cq_queue_setup(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx,
 387                 uint32_t socket_id)
 388 {
 389         struct virtqueue *vq;
 390         uint16_t nb_desc = 0;
 391         int ret;
 392         struct virtio_hw *hw = dev->data->dev_private;
 393
 394         PMD_INIT_FUNC_TRACE();
 395         ret = virtio_dev_queue_setup(dev, VTNET_CQ, VTNET_SQ_CQ_QUEUE_IDX,
 396                         vtpci_queue_idx, nb_desc, socket_id, &vq);
 397
 398         if (ret < 0) {
 399                 PMD_INIT_LOG(ERR, "control vq initialization failed");
 400                 return ret;
 401         }
 402
 403         hw->cvq = vq;
 404         return 0;
 405 }
 406
 407 static void
 408 virtio_dev_close(struct rte_eth_dev *dev)
 409 {
 410         struct virtio_hw *hw = dev->data->dev_private;
 411         struct rte_pci_device *pci_dev = dev->pci_dev;
 412
 413         PMD_INIT_LOG(DEBUG, "virtio_dev_close");
 414
 415         /* reset the NIC */
 416         if (pci_dev->driver->drv_flags & RTE_PCI_DRV_INTR_LSC)
 417                 vtpci_irq_config(hw, VIRTIO_MSI_NO_VECTOR);
 418         vtpci_reset(hw);
 419         hw->started = 0;
 420         virtio_dev_free_mbufs(dev);
 421 }
 422
 423 static void
 424 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
 425 {
 426         struct virtio_hw *hw = dev->data->dev_private;
 427         struct virtio_pmd_ctrl ctrl;
 428         int dlen[1];
 429         int ret;
 430
 431         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 432         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 433         ctrl.data[0] = 1;
 434         dlen[0] = 1;
 435
 436         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 437
 438         if (ret)
 439                 PMD_INIT_LOG(ERR, "Failed to enable promisc");
 440 }
 441
 442 static void
 443 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
 444 {
 445         struct virtio_hw *hw = dev->data->dev_private;
 446         struct virtio_pmd_ctrl ctrl;
 447         int dlen[1];
 448         int ret;
 449
 450         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 451         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 452         ctrl.data[0] = 0;
 453         dlen[0] = 1;
 454
 455         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 456
 457         if (ret)
 458                 PMD_INIT_LOG(ERR, "Failed to disable promisc");
 459 }
 460
 461 static void
 462 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
 463 {
 464         struct virtio_hw *hw = dev->data->dev_private;
 465         struct virtio_pmd_ctrl ctrl;
 466         int dlen[1];
 467         int ret;
 468
 469         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 470         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 471         ctrl.data[0] = 1;
 472         dlen[0] = 1;
 473
 474         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 475
 476         if (ret)
 477                 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
 478 }
 479
 480 static void
 481 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
 482 {
 483         struct virtio_hw *hw = dev->data->dev_private;
 484         struct virtio_pmd_ctrl ctrl;
 485         int dlen[1];
 486         int ret;
 487
 488         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 489         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 490         ctrl.data[0] = 0;
 491         dlen[0] = 1;
 492
 493         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 494
 495         if (ret)
 496                 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
 497 }
 498
 499 /*
 500  * dev_ops for virtio, bare necessities for basic operation
 501  */
 502 static const struct eth_dev_ops virtio_eth_dev_ops = {
 503         .dev_configure           = virtio_dev_configure,
 504         .dev_start               = virtio_dev_start,
 505         .dev_stop                = virtio_dev_stop,
 506         .dev_close               = virtio_dev_close,
 507         .promiscuous_enable      = virtio_dev_promiscuous_enable,
 508         .promiscuous_disable     = virtio_dev_promiscuous_disable,
 509         .allmulticast_enable     = virtio_dev_allmulticast_enable,
 510         .allmulticast_disable    = virtio_dev_allmulticast_disable,
 511
 512         .dev_infos_get           = virtio_dev_info_get,
 513         .stats_get               = virtio_dev_stats_get,
 514         .stats_reset             = virtio_dev_stats_reset,
 515         .link_update             = virtio_dev_link_update,
 516         .rx_queue_setup          = virtio_dev_rx_queue_setup,
 517         /* meaningfull only to multiple queue */
 518         .rx_queue_release        = virtio_dev_rx_queue_release,
 519         .tx_queue_setup          = virtio_dev_tx_queue_setup,
 520         /* meaningfull only to multiple queue */
 521         .tx_queue_release        = virtio_dev_tx_queue_release,
 522         /* collect stats per queue */
 523         .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
 524         .vlan_filter_set         = virtio_vlan_filter_set,
 525         .mac_addr_add            = virtio_mac_addr_add,
 526         .mac_addr_remove         = virtio_mac_addr_remove,
 527         .mac_addr_set            = virtio_mac_addr_set,
 528 };
 529
 530 static inline int
 531 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
 532                                 struct rte_eth_link *link)
 533 {
 534         struct rte_eth_link *dst = link;
 535         struct rte_eth_link *src = &(dev->data->dev_link);
 536
 537         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
 538                         *(uint64_t *)src) == 0)
 539                 return -1;
 540
 541         return 0;
 542 }
 543
 544 /**
 545  * Atomically writes the link status information into global
 546  * structure rte_eth_dev.
 547  *
 548  * @param dev
 549  *   - Pointer to the structure rte_eth_dev to read from.
 550  *   - Pointer to the buffer to be saved with the link status.
 551  *
 552  * @return
 553  *   - On success, zero.
 554  *   - On failure, negative value.
 555  */
 556 static inline int
 557 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
 558                 struct rte_eth_link *link)
 559 {
 560         struct rte_eth_link *dst = &(dev->data->dev_link);
 561         struct rte_eth_link *src = link;
 562
 563         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
 564                                         *(uint64_t *)src) == 0)
 565                 return -1;
 566
 567         return 0;
 568 }
 569
 570 static void
 571 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 572 {
 573         unsigned i;
 574
 575         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 576                 const struct virtqueue *txvq = dev->data->tx_queues[i];
 577                 if (txvq == NULL)
 578                         continue;
 579
 580                 stats->opackets += txvq->packets;
 581                 stats->obytes += txvq->bytes;
 582                 stats->oerrors += txvq->errors;
 583
 584                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 585                         stats->q_opackets[i] = txvq->packets;
 586                         stats->q_obytes[i] = txvq->bytes;
 587                 }
 588         }
 589
 590         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 591                 const struct virtqueue *rxvq = dev->data->rx_queues[i];
 592                 if (rxvq == NULL)
 593                         continue;
 594
 595                 stats->ipackets += rxvq->packets;
 596                 stats->ibytes += rxvq->bytes;
 597                 stats->ierrors += rxvq->errors;
 598
 599                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 600                         stats->q_ipackets[i] = rxvq->packets;
 601                         stats->q_ibytes[i] = rxvq->bytes;
 602                 }
 603         }
 604
 605         stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
 606 }
 607
 608 static void
 609 virtio_dev_stats_reset(struct rte_eth_dev *dev)
 610 {
 611         unsigned int i;
 612
 613         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 614                 struct virtqueue *txvq = dev->data->tx_queues[i];
 615                 if (txvq == NULL)
 616                         continue;
 617
 618                 txvq->packets = 0;
 619                 txvq->bytes = 0;
 620                 txvq->errors = 0;
 621         }
 622
 623         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 624                 struct virtqueue *rxvq = dev->data->rx_queues[i];
 625                 if (rxvq == NULL)
 626                         continue;
 627
 628                 rxvq->packets = 0;
 629                 rxvq->bytes = 0;
 630                 rxvq->errors = 0;
 631         }
 632
 633         dev->data->rx_mbuf_alloc_failed = 0;
 634 }
 635
 636 static void
 637 virtio_set_hwaddr(struct virtio_hw *hw)
 638 {
 639         vtpci_write_dev_config(hw,
 640                         offsetof(struct virtio_net_config, mac),
 641                         &hw->mac_addr, ETHER_ADDR_LEN);
 642 }
 643
 644 static void
 645 virtio_get_hwaddr(struct virtio_hw *hw)
 646 {
 647         if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
 648                 vtpci_read_dev_config(hw,
 649                         offsetof(struct virtio_net_config, mac),
 650                         &hw->mac_addr, ETHER_ADDR_LEN);
 651         } else {
 652                 eth_random_addr(&hw->mac_addr[0]);
 653                 virtio_set_hwaddr(hw);
 654         }
 655 }
 656
 657 static int
 658 virtio_mac_table_set(struct virtio_hw *hw,
 659                      const struct virtio_net_ctrl_mac *uc,
 660                      const struct virtio_net_ctrl_mac *mc)
 661 {
 662         struct virtio_pmd_ctrl ctrl;
 663         int err, len[2];
 664
 665         ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
 666         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
 667
 668         len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
 669         memcpy(ctrl.data, uc, len[0]);
 670
 671         len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
 672         memcpy(ctrl.data + len[0], mc, len[1]);
 673
 674         err = virtio_send_command(hw->cvq, &ctrl, len, 2);
 675         if (err != 0)
 676                 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
 677
 678         return err;
 679 }
 680
 681 static void
 682 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
 683                     uint32_t index, uint32_t vmdq __rte_unused)
 684 {
 685         struct virtio_hw *hw = dev->data->dev_private;
 686         const struct ether_addr *addrs = dev->data->mac_addrs;
 687         unsigned int i;
 688         struct virtio_net_ctrl_mac *uc, *mc;
 689
 690         if (index >= VIRTIO_MAX_MAC_ADDRS) {
 691                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
 692                 return;
 693         }
 694
 695         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
 696         uc->entries = 0;
 697         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
 698         mc->entries = 0;
 699
 700         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
 701                 const struct ether_addr *addr
 702                         = (i == index) ? mac_addr : addrs + i;
 703                 struct virtio_net_ctrl_mac *tbl
 704                         = is_multicast_ether_addr(addr) ? mc : uc;
 705
 706                 memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
 707         }
 708
 709         virtio_mac_table_set(hw, uc, mc);
 710 }
 711
 712 static void
 713 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
 714 {
 715         struct virtio_hw *hw = dev->data->dev_private;
 716         struct ether_addr *addrs = dev->data->mac_addrs;
 717         struct virtio_net_ctrl_mac *uc, *mc;
 718         unsigned int i;
 719
 720         if (index >= VIRTIO_MAX_MAC_ADDRS) {
 721                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
 722                 return;
 723         }
 724
 725         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
 726         uc->entries = 0;
 727         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
 728         mc->entries = 0;
 729
 730         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
 731                 struct virtio_net_ctrl_mac *tbl;
 732
 733                 if (i == index || is_zero_ether_addr(addrs + i))
 734                         continue;
 735
 736                 tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
 737                 memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
 738         }
 739
 740         virtio_mac_table_set(hw, uc, mc);
 741 }
 742
 743 static void
 744 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 745 {
 746         struct virtio_hw *hw = dev->data->dev_private;
 747
 748         memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
 749
 750         /* Use atomic update if available */
 751         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
 752                 struct virtio_pmd_ctrl ctrl;
 753                 int len = ETHER_ADDR_LEN;
 754
 755                 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
 756                 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
 757
 758                 memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
 759                 virtio_send_command(hw->cvq, &ctrl, &len, 1);
 760         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
 761                 virtio_set_hwaddr(hw);
 762 }
 763
 764 static int
 765 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
 766 {
 767         struct virtio_hw *hw = dev->data->dev_private;
 768         struct virtio_pmd_ctrl ctrl;
 769         int len;
 770
 771         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
 772                 return -ENOTSUP;
 773
 774         ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
 775         ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
 776         memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
 777         len = sizeof(vlan_id);
 778
 779         return virtio_send_command(hw->cvq, &ctrl, &len, 1);
 780 }
 781
 782 static void
 783 virtio_negotiate_features(struct virtio_hw *hw)
 784 {
 785         uint32_t host_features, mask;
 786
 787         /* checksum offload not implemented */
 788         mask = VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM;
 789
 790         /* TSO and LRO are only available when their corresponding
 791          * checksum offload feature is also negotiated.
 792          */
 793         mask |= VIRTIO_NET_F_HOST_TSO4 | VIRTIO_NET_F_HOST_TSO6 | VIRTIO_NET_F_HOST_ECN;
 794         mask |= VIRTIO_NET_F_GUEST_TSO4 | VIRTIO_NET_F_GUEST_TSO6 | VIRTIO_NET_F_GUEST_ECN;
 795         mask |= VTNET_LRO_FEATURES;
 796
 797         /* not negotiating INDIRECT descriptor table support */
 798         mask |= VIRTIO_RING_F_INDIRECT_DESC;
 799
 800         /* Prepare guest_features: feature that driver wants to support */
 801         hw->guest_features = VTNET_FEATURES & ~mask;
 802         PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %x",
 803                 hw->guest_features);
 804
 805         /* Read device(host) feature bits */
 806         host_features = VIRTIO_READ_REG_4(hw, VIRTIO_PCI_HOST_FEATURES);
 807         PMD_INIT_LOG(DEBUG, "host_features before negotiate = %x",
 808                 host_features);
 809
 810         /*
 811          * Negotiate features: Subset of device feature bits are written back
 812          * guest feature bits.
 813          */
 814         hw->guest_features = vtpci_negotiate_features(hw, host_features);
 815         PMD_INIT_LOG(DEBUG, "features after negotiate = %x",
 816                 hw->guest_features);
 817 }
 818
 819 #ifdef RTE_EXEC_ENV_LINUXAPP
 820 static int
 821 parse_sysfs_value(const char *filename, unsigned long *val)
 822 {
 823         FILE *f;
 824         char buf[BUFSIZ];
 825         char *end = NULL;
 826
 827         f = fopen(filename, "r");
 828         if (f == NULL) {
 829                 PMD_INIT_LOG(ERR, "%s(): cannot open sysfs value %s",
 830                              __func__, filename);
 831                 return -1;
 832         }
 833
 834         if (fgets(buf, sizeof(buf), f) == NULL) {
 835                 PMD_INIT_LOG(ERR, "%s(): cannot read sysfs value %s",
 836                              __func__, filename);
 837                 fclose(f);
 838                 return -1;
 839         }
 840         *val = strtoul(buf, &end, 0);
 841         if ((buf[0] == '\0') || (end == NULL) || (*end != '\n')) {
 842                 PMD_INIT_LOG(ERR, "%s(): cannot parse sysfs value %s",
 843                              __func__, filename);
 844                 fclose(f);
 845                 return -1;
 846         }
 847         fclose(f);
 848         return 0;
 849 }
 850
 851 static int get_uio_dev(struct rte_pci_addr *loc, char *buf, unsigned int buflen,
 852                         unsigned int *uio_num)
 853 {
 854         struct dirent *e;
 855         DIR *dir;
 856         char dirname[PATH_MAX];
 857
 858         /* depending on kernel version, uio can be located in uio/uioX
 859          * or uio:uioX */
 860         snprintf(dirname, sizeof(dirname),
 861                      SYSFS_PCI_DEVICES "/" PCI_PRI_FMT "/uio",
 862                      loc->domain, loc->bus, loc->devid, loc->function);
 863         dir = opendir(dirname);
 864         if (dir == NULL) {
 865                 /* retry with the parent directory */
 866                 snprintf(dirname, sizeof(dirname),
 867                              SYSFS_PCI_DEVICES "/" PCI_PRI_FMT,
 868                              loc->domain, loc->bus, loc->devid, loc->function);
 869                 dir = opendir(dirname);
 870
 871                 if (dir == NULL) {
 872                         PMD_INIT_LOG(ERR, "Cannot opendir %s", dirname);
 873                         return -1;
 874                 }
 875         }
 876
 877         /* take the first file starting with "uio" */
 878         while ((e = readdir(dir)) != NULL) {
 879                 /* format could be uio%d ...*/
 880                 int shortprefix_len = sizeof("uio") - 1;
 881                 /* ... or uio:uio%d */
 882                 int longprefix_len = sizeof("uio:uio") - 1;
 883                 char *endptr;
 884
 885                 if (strncmp(e->d_name, "uio", 3) != 0)
 886                         continue;
 887
 888                 /* first try uio%d */
 889                 errno = 0;
 890                 *uio_num = strtoull(e->d_name + shortprefix_len, &endptr, 10);
 891                 if (errno == 0 && endptr != (e->d_name + shortprefix_len)) {
 892                         snprintf(buf, buflen, "%s/uio%u", dirname, *uio_num);
 893                         break;
 894                 }
 895
 896                 /* then try uio:uio%d */
 897                 errno = 0;
 898                 *uio_num = strtoull(e->d_name + longprefix_len, &endptr, 10);
 899                 if (errno == 0 && endptr != (e->d_name + longprefix_len)) {
 900                         snprintf(buf, buflen, "%s/uio:uio%u", dirname,
 901                                      *uio_num);
 902                         break;
 903                 }
 904         }
 905         closedir(dir);
 906
 907         /* No uio resource found */
 908         if (e == NULL) {
 909                 PMD_INIT_LOG(ERR, "Could not find uio resource");
 910                 return -1;
 911         }
 912
 913         return 0;
 914 }
 915
 916 static int
 917 virtio_has_msix(const struct rte_pci_addr *loc)
 918 {
 919         DIR *d;
 920         char dirname[PATH_MAX];
 921
 922         snprintf(dirname, sizeof(dirname),
 923                      SYSFS_PCI_DEVICES "/" PCI_PRI_FMT "/msi_irqs",
 924                      loc->domain, loc->bus, loc->devid, loc->function);
 925
 926         d = opendir(dirname);
 927         if (d)
 928                 closedir(d);
 929
 930         return (d != NULL);
 931 }
 932
 933 /* Extract I/O port numbers from sysfs */
 934 static int virtio_resource_init_by_uio(struct rte_pci_device *pci_dev)
 935 {
 936         char dirname[PATH_MAX];
 937         char filename[PATH_MAX];
 938         unsigned long start, size;
 939         unsigned int uio_num;
 940
 941         if (get_uio_dev(&pci_dev->addr, dirname, sizeof(dirname), &uio_num) < 0)
 942                 return -1;
 943
 944         /* get portio size */
 945         snprintf(filename, sizeof(filename),
 946                      "%s/portio/port0/size", dirname);
 947         if (parse_sysfs_value(filename, &size) < 0) {
 948                 PMD_INIT_LOG(ERR, "%s(): cannot parse size",
 949                              __func__);
 950                 return -1;
 951         }
 952
 953         /* get portio start */
 954         snprintf(filename, sizeof(filename),
 955                  "%s/portio/port0/start", dirname);
 956         if (parse_sysfs_value(filename, &start) < 0) {
 957                 PMD_INIT_LOG(ERR, "%s(): cannot parse portio start",
 958                              __func__);
 959                 return -1;
 960         }
 961         pci_dev->mem_resource[0].addr = (void *)(uintptr_t)start;
 962         pci_dev->mem_resource[0].len =  (uint64_t)size;
 963         PMD_INIT_LOG(DEBUG,
 964                      "PCI Port IO found start=0x%lx with size=0x%lx",
 965                      start, size);
 966
 967         /* save fd */
 968         memset(dirname, 0, sizeof(dirname));
 969         snprintf(dirname, sizeof(dirname), "/dev/uio%u", uio_num);
 970         pci_dev->intr_handle.fd = open(dirname, O_RDWR);
 971         if (pci_dev->intr_handle.fd < 0) {
 972                 PMD_INIT_LOG(ERR, "Cannot open %s: %s\n",
 973                         dirname, strerror(errno));
 974                 return -1;
 975         }
 976
 977         pci_dev->intr_handle.type = RTE_INTR_HANDLE_UIO;
 978         pci_dev->driver->drv_flags |= RTE_PCI_DRV_INTR_LSC;
 979
 980         return 0;
 981 }
 982
 983 /* Extract port I/O numbers from proc/ioports */
 984 static int virtio_resource_init_by_ioports(struct rte_pci_device *pci_dev)
 985 {
 986         uint16_t start, end;
 987         int size;
 988         FILE *fp;
 989         char *line = NULL;
 990         char pci_id[16];
 991         int found = 0;
 992         size_t linesz;
 993
 994         snprintf(pci_id, sizeof(pci_id), PCI_PRI_FMT,
 995                  pci_dev->addr.domain,
 996                  pci_dev->addr.bus,
 997                  pci_dev->addr.devid,
 998                  pci_dev->addr.function);
 999
1000         fp = fopen("/proc/ioports", "r");
1001         if (fp == NULL) {
1002                 PMD_INIT_LOG(ERR, "%s(): can't open ioports", __func__);
1003                 return -1;
1004         }
1005
1006         while (getdelim(&line, &linesz, '\n', fp) > 0) {
1007                 char *ptr = line;
1008                 char *left;
1009                 int n;
1010
1011                 n = strcspn(ptr, ":");
1012                 ptr[n] = 0;
1013                 left = &ptr[n+1];
1014
1015                 while (*left && isspace(*left))
1016                         left++;
1017
1018                 if (!strncmp(left, pci_id, strlen(pci_id))) {
1019                         found = 1;
1020
1021                         while (*ptr && isspace(*ptr))
1022                                 ptr++;
1023
1024                         sscanf(ptr, "%04hx-%04hx", &start, &end);
1025                         size = end - start + 1;
1026
1027                         break;
1028                 }
1029         }
1030
1031         free(line);
1032         fclose(fp);
1033
1034         if (!found)
1035                 return -1;
1036
1037         pci_dev->mem_resource[0].addr = (void *)(uintptr_t)(uint32_t)start;
1038         pci_dev->mem_resource[0].len =  (uint64_t)size;
1039         PMD_INIT_LOG(DEBUG,
1040                 "PCI Port IO found start=0x%x with size=0x%x",
1041                 start, size);
1042
1043         /* can't support lsc interrupt without uio */
1044         pci_dev->driver->drv_flags &= ~RTE_PCI_DRV_INTR_LSC;
1045
1046         return 0;
1047 }
1048
1049 /* Extract I/O port numbers from sysfs */
1050 static int virtio_resource_init(struct rte_pci_device *pci_dev)
1051 {
1052         if (virtio_resource_init_by_uio(pci_dev) == 0)
1053                 return 0;
1054         else
1055                 return virtio_resource_init_by_ioports(pci_dev);
1056 }
1057
1058 #else
1059 static int
1060 virtio_has_msix(const struct rte_pci_addr *loc __rte_unused)
1061 {
1062         /* nic_uio does not enable interrupts, return 0 (false). */
1063         return 0;
1064 }
1065
1066 static int virtio_resource_init(struct rte_pci_device *pci_dev __rte_unused)
1067 {
1068         /* no setup required */
1069         return 0;
1070 }
1071 #endif
1072
1073 /*
1074  * Process Virtio Config changed interrupt and call the callback
1075  * if link state changed.
1076  */
1077 static void
1078 virtio_interrupt_handler(__rte_unused struct rte_intr_handle *handle,
1079                          void *param)
1080 {
1081         struct rte_eth_dev *dev = param;
1082         struct virtio_hw *hw = dev->data->dev_private;
1083         uint8_t isr;
1084
1085         /* Read interrupt status which clears interrupt */
1086         isr = vtpci_isr(hw);
1087         PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1088
1089         if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0)
1090                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1091
1092         if (isr & VIRTIO_PCI_ISR_CONFIG) {
1093                 if (virtio_dev_link_update(dev, 0) == 0)
1094                         _rte_eth_dev_callback_process(dev,
1095                                                       RTE_ETH_EVENT_INTR_LSC);
1096         }
1097
1098 }
1099
1100 static void
1101 rx_func_get(struct rte_eth_dev *eth_dev)
1102 {
1103         struct virtio_hw *hw = eth_dev->data->dev_private;
1104         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF))
1105                 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1106         else
1107                 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1108 }
1109
1110 /*
1111  * This function is based on probe() function in virtio_pci.c
1112  * It returns 0 on success.
1113  */
1114 static int
1115 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1116 {
1117         struct virtio_hw *hw = eth_dev->data->dev_private;
1118         struct virtio_net_config *config;
1119         struct virtio_net_config local_config;
1120         uint32_t offset_conf = sizeof(config->mac);
1121         struct rte_pci_device *pci_dev;
1122
1123         RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr));
1124
1125         eth_dev->dev_ops = &virtio_eth_dev_ops;
1126         eth_dev->tx_pkt_burst = &virtio_xmit_pkts;
1127
1128         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1129                 rx_func_get(eth_dev);
1130                 return 0;
1131         }
1132
1133         /* Allocate memory for storing MAC addresses */
1134         eth_dev->data->mac_addrs = rte_zmalloc("virtio", ETHER_ADDR_LEN, 0);
1135         if (eth_dev->data->mac_addrs == NULL) {
1136                 PMD_INIT_LOG(ERR,
1137                         "Failed to allocate %d bytes needed to store MAC addresses",
1138                         ETHER_ADDR_LEN);
1139                 return -ENOMEM;
1140         }
1141
1142         pci_dev = eth_dev->pci_dev;
1143         if (virtio_resource_init(pci_dev) < 0)
1144                 return -1;
1145
1146         hw->use_msix = virtio_has_msix(&pci_dev->addr);
1147         hw->io_base = (uint32_t)(uintptr_t)pci_dev->mem_resource[0].addr;
1148
1149         /* Reset the device although not necessary at startup */
1150         vtpci_reset(hw);
1151
1152         /* Tell the host we've noticed this device. */
1153         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1154
1155         /* Tell the host we've known how to drive the device. */
1156         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1157         virtio_negotiate_features(hw);
1158
1159         rx_func_get(eth_dev);
1160
1161         /* Setting up rx_header size for the device */
1162         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF))
1163                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1164         else
1165                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1166
1167         /* Copy the permanent MAC address to: virtio_hw */
1168         virtio_get_hwaddr(hw);
1169         ether_addr_copy((struct ether_addr *) hw->mac_addr,
1170                         &eth_dev->data->mac_addrs[0]);
1171         PMD_INIT_LOG(DEBUG,
1172                      "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1173                      hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1174                      hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1175
1176         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1177                 config = &local_config;
1178
1179                 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1180                         offset_conf += sizeof(config->status);
1181                 } else {
1182                         PMD_INIT_LOG(DEBUG,
1183                                      "VIRTIO_NET_F_STATUS is not supported");
1184                         config->status = 0;
1185                 }
1186
1187                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1188                         offset_conf += sizeof(config->max_virtqueue_pairs);
1189                 } else {
1190                         PMD_INIT_LOG(DEBUG,
1191                                      "VIRTIO_NET_F_MQ is not supported");
1192                         config->max_virtqueue_pairs = 1;
1193                 }
1194
1195                 vtpci_read_dev_config(hw, 0, (uint8_t *)config, offset_conf);
1196
1197                 hw->max_rx_queues =
1198                         (VIRTIO_MAX_RX_QUEUES < config->max_virtqueue_pairs) ?
1199                         VIRTIO_MAX_RX_QUEUES : config->max_virtqueue_pairs;
1200                 hw->max_tx_queues =
1201                         (VIRTIO_MAX_TX_QUEUES < config->max_virtqueue_pairs) ?
1202                         VIRTIO_MAX_TX_QUEUES : config->max_virtqueue_pairs;
1203
1204                 virtio_dev_cq_queue_setup(eth_dev,
1205                                         config->max_virtqueue_pairs * 2,
1206                                         SOCKET_ID_ANY);
1207
1208                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1209                                 config->max_virtqueue_pairs);
1210                 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1211                 PMD_INIT_LOG(DEBUG,
1212                                 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1213                                 config->mac[0], config->mac[1],
1214                                 config->mac[2], config->mac[3],
1215                                 config->mac[4], config->mac[5]);
1216         } else {
1217                 hw->max_rx_queues = 1;
1218                 hw->max_tx_queues = 1;
1219         }
1220
1221         eth_dev->data->nb_rx_queues = hw->max_rx_queues;
1222         eth_dev->data->nb_tx_queues = hw->max_tx_queues;
1223
1224         PMD_INIT_LOG(DEBUG, "hw->max_rx_queues=%d   hw->max_tx_queues=%d",
1225                         hw->max_rx_queues, hw->max_tx_queues);
1226         PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1227                         eth_dev->data->port_id, pci_dev->id.vendor_id,
1228                         pci_dev->id.device_id);
1229
1230         /* Setup interrupt callback  */
1231         if (pci_dev->driver->drv_flags & RTE_PCI_DRV_INTR_LSC)
1232                 rte_intr_callback_register(&pci_dev->intr_handle,
1233                                    virtio_interrupt_handler, eth_dev);
1234
1235         virtio_dev_cq_start(eth_dev);
1236
1237         return 0;
1238 }
1239
1240 static struct eth_driver rte_virtio_pmd = {
1241         .pci_drv = {
1242                 .name = "rte_virtio_pmd",
1243                 .id_table = pci_id_virtio_map,
1244         },
1245         .eth_dev_init = eth_virtio_dev_init,
1246         .dev_private_size = sizeof(struct virtio_hw),
1247 };
1248
1249 /*
1250  * Driver initialization routine.
1251  * Invoked once at EAL init time.
1252  * Register itself as the [Poll Mode] Driver of PCI virtio devices.
1253  * Returns 0 on success.
1254  */
1255 static int
1256 rte_virtio_pmd_init(const char *name __rte_unused,
1257                     const char *param __rte_unused)
1258 {
1259         if (rte_eal_iopl_init() != 0) {
1260                 PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1261                 return -1;
1262         }
1263
1264         rte_eth_driver_register(&rte_virtio_pmd);
1265         return 0;
1266 }
1267
1268 /*
1269  * Only 1 queue is supported, no queue release related operation
1270  */
1271 static void
1272 virtio_dev_rx_queue_release(__rte_unused void *rxq)
1273 {
1274 }
1275
1276 static void
1277 virtio_dev_tx_queue_release(__rte_unused void *txq)
1278 {
1279 }
1280
1281 /*
1282  * Configure virtio device
1283  * It returns 0 on success.
1284  */
1285 static int
1286 virtio_dev_configure(struct rte_eth_dev *dev)
1287 {
1288         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1289         struct virtio_hw *hw = dev->data->dev_private;
1290         struct rte_pci_device *pci_dev = dev->pci_dev;
1291
1292         PMD_INIT_LOG(DEBUG, "configure");
1293
1294         if (rxmode->hw_ip_checksum) {
1295                 PMD_DRV_LOG(ERR, "HW IP checksum not supported");
1296                 return (-EINVAL);
1297         }
1298
1299         hw->vlan_strip = rxmode->hw_vlan_strip;
1300
1301         if (rxmode->hw_vlan_filter
1302             && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1303                 PMD_DRV_LOG(NOTICE,
1304                             "vlan filtering not available on this host");
1305                 return -ENOTSUP;
1306         }
1307
1308         if (pci_dev->driver->drv_flags & RTE_PCI_DRV_INTR_LSC)
1309                 if (vtpci_irq_config(hw, 0) == VIRTIO_MSI_NO_VECTOR) {
1310                         PMD_DRV_LOG(ERR, "failed to set config vector");
1311                         return -EBUSY;
1312                 }
1313
1314         return 0;
1315 }
1316
1317
1318 static int
1319 virtio_dev_start(struct rte_eth_dev *dev)
1320 {
1321         uint16_t nb_queues, i;
1322         struct virtio_hw *hw = dev->data->dev_private;
1323         struct rte_pci_device *pci_dev = dev->pci_dev;
1324
1325         /* check if lsc interrupt feature is enabled */
1326         if ((dev->data->dev_conf.intr_conf.lsc) &&
1327                 (pci_dev->driver->drv_flags & RTE_PCI_DRV_INTR_LSC)) {
1328                 if (!vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1329                         PMD_DRV_LOG(ERR, "link status not supported by host");
1330                         return -ENOTSUP;
1331                 }
1332
1333                 if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0) {
1334                         PMD_DRV_LOG(ERR, "interrupt enable failed");
1335                         return -EIO;
1336                 }
1337         }
1338
1339         /* Initialize Link state */
1340         virtio_dev_link_update(dev, 0);
1341
1342         /* On restart after stop do not touch queues */
1343         if (hw->started)
1344                 return 0;
1345
1346         /* Do final configuration before rx/tx engine starts */
1347         virtio_dev_rxtx_start(dev);
1348         vtpci_reinit_complete(hw);
1349
1350         hw->started = 1;
1351
1352         /*Notify the backend
1353          *Otherwise the tap backend might already stop its queue due to fullness.
1354          *vhost backend will have no chance to be waked up
1355          */
1356         nb_queues = dev->data->nb_rx_queues;
1357         if (nb_queues > 1) {
1358                 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1359                         return -EINVAL;
1360         }
1361
1362         PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1363
1364         for (i = 0; i < nb_queues; i++)
1365                 virtqueue_notify(dev->data->rx_queues[i]);
1366
1367         PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1368
1369         for (i = 0; i < dev->data->nb_rx_queues; i++)
1370                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1371
1372         for (i = 0; i < dev->data->nb_tx_queues; i++)
1373                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1374
1375         return 0;
1376 }
1377
1378 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1379 {
1380         struct rte_mbuf *buf;
1381         int i, mbuf_num = 0;
1382
1383         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1384                 PMD_INIT_LOG(DEBUG,
1385                              "Before freeing rxq[%d] used and unused buf", i);
1386                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1387
1388                 while ((buf = (struct rte_mbuf *)virtqueue_detatch_unused(
1389                                         dev->data->rx_queues[i])) != NULL) {
1390                         rte_pktmbuf_free(buf);
1391                         mbuf_num++;
1392                 }
1393
1394                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1395                 PMD_INIT_LOG(DEBUG,
1396                              "After freeing rxq[%d] used and unused buf", i);
1397                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1398         }
1399
1400         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1401                 PMD_INIT_LOG(DEBUG,
1402                              "Before freeing txq[%d] used and unused bufs",
1403                              i);
1404                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1405
1406                 mbuf_num = 0;
1407                 while ((buf = (struct rte_mbuf *)virtqueue_detatch_unused(
1408                                         dev->data->tx_queues[i])) != NULL) {
1409                         rte_pktmbuf_free(buf);
1410
1411                         mbuf_num++;
1412                 }
1413
1414                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1415                 PMD_INIT_LOG(DEBUG,
1416                              "After freeing txq[%d] used and unused buf", i);
1417                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1418         }
1419 }
1420
1421 /*
1422  * Stop device: disable interrupt and mark link down
1423  */
1424 static void
1425 virtio_dev_stop(struct rte_eth_dev *dev)
1426 {
1427         struct rte_eth_link link;
1428
1429         PMD_INIT_LOG(DEBUG, "stop");
1430
1431         if (dev->data->dev_conf.intr_conf.lsc)
1432                 rte_intr_disable(&dev->pci_dev->intr_handle);
1433
1434         memset(&link, 0, sizeof(link));
1435         virtio_dev_atomic_write_link_status(dev, &link);
1436 }
1437
1438 static int
1439 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
1440 {
1441         struct rte_eth_link link, old;
1442         uint16_t status;
1443         struct virtio_hw *hw = dev->data->dev_private;
1444         memset(&link, 0, sizeof(link));
1445         virtio_dev_atomic_read_link_status(dev, &link);
1446         old = link;
1447         link.link_duplex = FULL_DUPLEX;
1448         link.link_speed  = SPEED_10G;
1449
1450         if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1451                 PMD_INIT_LOG(DEBUG, "Get link status from hw");
1452                 vtpci_read_dev_config(hw,
1453                                 offsetof(struct virtio_net_config, status),
1454                                 &status, sizeof(status));
1455                 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
1456                         link.link_status = 0;
1457                         PMD_INIT_LOG(DEBUG, "Port %d is down",
1458                                      dev->data->port_id);
1459                 } else {
1460                         link.link_status = 1;
1461                         PMD_INIT_LOG(DEBUG, "Port %d is up",
1462                                      dev->data->port_id);
1463                 }
1464         } else {
1465                 link.link_status = 1;   /* Link up */
1466         }
1467         virtio_dev_atomic_write_link_status(dev, &link);
1468
1469         return (old.link_status == link.link_status) ? -1 : 0;
1470 }
1471
1472 static void
1473 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1474 {
1475         struct virtio_hw *hw = dev->data->dev_private;
1476
1477         dev_info->driver_name = dev->driver->pci_drv.name;
1478         dev_info->max_rx_queues = (uint16_t)hw->max_rx_queues;
1479         dev_info->max_tx_queues = (uint16_t)hw->max_tx_queues;
1480         dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
1481         dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
1482         dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
1483         dev_info->default_txconf = (struct rte_eth_txconf) {
1484                 .txq_flags = ETH_TXQ_FLAGS_NOOFFLOADS
1485         };
1486 }
1487
1488 /*
1489  * It enables testpmd to collect per queue stats.
1490  */
1491 static int
1492 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
1493 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
1494 __rte_unused uint8_t is_rx)
1495 {
1496         return 0;
1497 }
1498
1499 static struct rte_driver rte_virtio_driver = {
1500         .type = PMD_PDEV,
1501         .init = rte_virtio_pmd_init,
1502 };
1503
1504 PMD_REGISTER_DRIVER(rte_virtio_driver);