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 #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 #include <rte_errno.h>
  55
  56 #include <rte_memory.h>
  57 #include <rte_eal.h>
  58 #include <rte_dev.h>
  59
  60 #include "virtio_ethdev.h"
  61 #include "virtio_pci.h"
  62 #include "virtio_logs.h"
  63 #include "virtqueue.h"
  64
  65
  66 static int eth_virtio_dev_init(struct rte_eth_dev *eth_dev);
  67 static int eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev);
  68 static int  virtio_dev_configure(struct rte_eth_dev *dev);
  69 static int  virtio_dev_start(struct rte_eth_dev *dev);
  70 static void virtio_dev_stop(struct rte_eth_dev *dev);
  71 static void virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
  72 static void virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
  73 static void virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
  74 static void virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
  75 static void virtio_dev_info_get(struct rte_eth_dev *dev,
  76                                 struct rte_eth_dev_info *dev_info);
  77 static int virtio_dev_link_update(struct rte_eth_dev *dev,
  78         __rte_unused int wait_to_complete);
  79
  80 static void virtio_set_hwaddr(struct virtio_hw *hw);
  81 static void virtio_get_hwaddr(struct virtio_hw *hw);
  82
  83 static void virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats);
  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 static int
 113 virtio_send_command(struct virtqueue *vq, struct virtio_pmd_ctrl *ctrl,
 114                 int *dlen, int pkt_num)
 115 {
 116         uint32_t head, i;
 117         int k, sum = 0;
 118         virtio_net_ctrl_ack status = ~0;
 119         struct virtio_pmd_ctrl result;
 120
 121         ctrl->status = status;
 122
 123         if (!(vq && vq->hw->cvq)) {
 124                 PMD_INIT_LOG(ERR,
 125                              "%s(): Control queue is not supported.",
 126                              __func__);
 127                 return -1;
 128         }
 129         head = vq->vq_desc_head_idx;
 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         if (ret) {
 232                 PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
 233                           "failed, this is too late now...");
 234                 return -EINVAL;
 235         }
 236
 237         return 0;
 238 }
 239
 240 void
 241 virtio_dev_queue_release(struct virtqueue *vq) {
 242         struct virtio_hw *hw = vq->hw;
 243
 244         if (vq) {
 245                 /* Select and deactivate the queue */
 246                 VIRTIO_WRITE_REG_2(hw, VIRTIO_PCI_QUEUE_SEL, vq->queue_id);
 247                 VIRTIO_WRITE_REG_4(hw, VIRTIO_PCI_QUEUE_PFN, 0);
 248
 249                 rte_free(vq);
 250                 vq = NULL;
 251         }
 252 }
 253
 254 int virtio_dev_queue_setup(struct rte_eth_dev *dev,
 255                         int queue_type,
 256                         uint16_t queue_idx,
 257                         uint16_t  vtpci_queue_idx,
 258                         uint16_t nb_desc,
 259                         unsigned int socket_id,
 260                         struct virtqueue **pvq)
 261 {
 262         char vq_name[VIRTQUEUE_MAX_NAME_SZ];
 263         const struct rte_memzone *mz;
 264         uint16_t vq_size;
 265         int size;
 266         struct virtio_hw *hw = dev->data->dev_private;
 267         struct virtqueue  *vq = NULL;
 268
 269         /* Write the virtqueue index to the Queue Select Field */
 270         VIRTIO_WRITE_REG_2(hw, VIRTIO_PCI_QUEUE_SEL, vtpci_queue_idx);
 271         PMD_INIT_LOG(DEBUG, "selecting queue: %d", vtpci_queue_idx);
 272
 273         /*
 274          * Read the virtqueue size from the Queue Size field
 275          * Always power of 2 and if 0 virtqueue does not exist
 276          */
 277         vq_size = VIRTIO_READ_REG_2(hw, VIRTIO_PCI_QUEUE_NUM);
 278         PMD_INIT_LOG(DEBUG, "vq_size: %d nb_desc:%d", vq_size, nb_desc);
 279         if (nb_desc == 0)
 280                 nb_desc = vq_size;
 281         if (vq_size == 0) {
 282                 PMD_INIT_LOG(ERR, "%s: virtqueue does not exist", __func__);
 283                 return -EINVAL;
 284         }
 285
 286         if (!rte_is_power_of_2(vq_size)) {
 287                 PMD_INIT_LOG(ERR, "%s: virtqueue size is not powerof 2", __func__);
 288                 return -EINVAL;
 289         }
 290
 291         if (nb_desc < vq_size) {
 292                 if (!rte_is_power_of_2(nb_desc)) {
 293                         PMD_INIT_LOG(ERR,
 294                                      "nb_desc(%u) size is not powerof 2",
 295                                      nb_desc);
 296                         return -EINVAL;
 297                 }
 298                 vq_size = nb_desc;
 299         }
 300
 301         if (queue_type == VTNET_RQ) {
 302                 snprintf(vq_name, sizeof(vq_name), "port%d_rvq%d",
 303                         dev->data->port_id, queue_idx);
 304                 vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
 305                         vq_size * sizeof(struct vq_desc_extra), RTE_CACHE_LINE_SIZE);
 306         } else if (queue_type == VTNET_TQ) {
 307                 snprintf(vq_name, sizeof(vq_name), "port%d_tvq%d",
 308                         dev->data->port_id, queue_idx);
 309                 vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
 310                         vq_size * sizeof(struct vq_desc_extra), RTE_CACHE_LINE_SIZE);
 311         } else if (queue_type == VTNET_CQ) {
 312                 snprintf(vq_name, sizeof(vq_name), "port%d_cvq",
 313                         dev->data->port_id);
 314                 vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
 315                         vq_size * sizeof(struct vq_desc_extra),
 316                         RTE_CACHE_LINE_SIZE);
 317         }
 318         if (vq == NULL) {
 319                 PMD_INIT_LOG(ERR, "%s: Can not allocate virtqueue", __func__);
 320                 return (-ENOMEM);
 321         }
 322
 323         vq->hw = hw;
 324         vq->port_id = dev->data->port_id;
 325         vq->queue_id = queue_idx;
 326         vq->vq_queue_index = vtpci_queue_idx;
 327         vq->vq_nentries = vq_size;
 328         vq->vq_free_cnt = vq_size;
 329
 330         /*
 331          * Reserve a memzone for vring elements
 332          */
 333         size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
 334         vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
 335         PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d", size, vq->vq_ring_size);
 336
 337         mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
 338                 socket_id, 0, VIRTIO_PCI_VRING_ALIGN);
 339         if (mz == NULL) {
 340                 if (rte_errno == EEXIST)
 341                         mz = rte_memzone_lookup(vq_name);
 342                 if (mz == NULL) {
 343                         rte_free(vq);
 344                         return -ENOMEM;
 345                 }
 346         }
 347
 348         /*
 349          * Virtio PCI device VIRTIO_PCI_QUEUE_PF register is 32bit,
 350          * and only accepts 32 bit page frame number.
 351          * Check if the allocated physical memory exceeds 16TB.
 352          */
 353         if ((mz->phys_addr + vq->vq_ring_size - 1) >> (VIRTIO_PCI_QUEUE_ADDR_SHIFT + 32)) {
 354                 PMD_INIT_LOG(ERR, "vring address shouldn't be above 16TB!");
 355                 rte_free(vq);
 356                 return -ENOMEM;
 357         }
 358
 359         memset(mz->addr, 0, sizeof(mz->len));
 360         vq->mz = mz;
 361         vq->vq_ring_mem = mz->phys_addr;
 362         vq->vq_ring_virt_mem = mz->addr;
 363         PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%"PRIx64, (uint64_t)mz->phys_addr);
 364         PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%"PRIx64, (uint64_t)(uintptr_t)mz->addr);
 365         vq->virtio_net_hdr_mz  = NULL;
 366         vq->virtio_net_hdr_mem = 0;
 367
 368         if (queue_type == VTNET_TQ) {
 369                 /*
 370                  * For each xmit packet, allocate a virtio_net_hdr
 371                  */
 372                 snprintf(vq_name, sizeof(vq_name), "port%d_tvq%d_hdrzone",
 373                         dev->data->port_id, queue_idx);
 374                 vq->virtio_net_hdr_mz = rte_memzone_reserve_aligned(vq_name,
 375                         vq_size * hw->vtnet_hdr_size,
 376                         socket_id, 0, RTE_CACHE_LINE_SIZE);
 377                 if (vq->virtio_net_hdr_mz == NULL) {
 378                         if (rte_errno == EEXIST)
 379                                 vq->virtio_net_hdr_mz =
 380                                         rte_memzone_lookup(vq_name);
 381                         if (vq->virtio_net_hdr_mz == NULL) {
 382                                 rte_free(vq);
 383                                 return -ENOMEM;
 384                         }
 385                 }
 386                 vq->virtio_net_hdr_mem =
 387                         vq->virtio_net_hdr_mz->phys_addr;
 388                 memset(vq->virtio_net_hdr_mz->addr, 0,
 389                         vq_size * hw->vtnet_hdr_size);
 390         } else if (queue_type == VTNET_CQ) {
 391                 /* Allocate a page for control vq command, data and status */
 392                 snprintf(vq_name, sizeof(vq_name), "port%d_cvq_hdrzone",
 393                         dev->data->port_id);
 394                 vq->virtio_net_hdr_mz = rte_memzone_reserve_aligned(vq_name,
 395                         PAGE_SIZE, socket_id, 0, RTE_CACHE_LINE_SIZE);
 396                 if (vq->virtio_net_hdr_mz == NULL) {
 397                         if (rte_errno == EEXIST)
 398                                 vq->virtio_net_hdr_mz =
 399                                         rte_memzone_lookup(vq_name);
 400                         if (vq->virtio_net_hdr_mz == NULL) {
 401                                 rte_free(vq);
 402                                 return -ENOMEM;
 403                         }
 404                 }
 405                 vq->virtio_net_hdr_mem =
 406                         vq->virtio_net_hdr_mz->phys_addr;
 407                 memset(vq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
 408         }
 409
 410         /*
 411          * Set guest physical address of the virtqueue
 412          * in VIRTIO_PCI_QUEUE_PFN config register of device
 413          */
 414         VIRTIO_WRITE_REG_4(hw, VIRTIO_PCI_QUEUE_PFN,
 415                         mz->phys_addr >> VIRTIO_PCI_QUEUE_ADDR_SHIFT);
 416         *pvq = vq;
 417         return 0;
 418 }
 419
 420 static int
 421 virtio_dev_cq_queue_setup(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx,
 422                 uint32_t socket_id)
 423 {
 424         struct virtqueue *vq;
 425         uint16_t nb_desc = 0;
 426         int ret;
 427         struct virtio_hw *hw = dev->data->dev_private;
 428
 429         PMD_INIT_FUNC_TRACE();
 430         ret = virtio_dev_queue_setup(dev, VTNET_CQ, VTNET_SQ_CQ_QUEUE_IDX,
 431                         vtpci_queue_idx, nb_desc, socket_id, &vq);
 432         if (ret < 0) {
 433                 PMD_INIT_LOG(ERR, "control vq initialization failed");
 434                 return ret;
 435         }
 436
 437         hw->cvq = vq;
 438         return 0;
 439 }
 440
 441 static void
 442 virtio_free_queues(struct rte_eth_dev *dev)
 443 {
 444         unsigned int i;
 445
 446         for (i = 0; i < dev->data->nb_rx_queues; i++)
 447                 virtio_dev_rx_queue_release(dev->data->rx_queues[i]);
 448
 449         dev->data->nb_rx_queues = 0;
 450
 451         for (i = 0; i < dev->data->nb_tx_queues; i++)
 452                 virtio_dev_tx_queue_release(dev->data->tx_queues[i]);
 453
 454         dev->data->nb_tx_queues = 0;
 455 }
 456
 457 static void
 458 virtio_dev_close(struct rte_eth_dev *dev)
 459 {
 460         struct virtio_hw *hw = dev->data->dev_private;
 461         struct rte_pci_device *pci_dev = dev->pci_dev;
 462
 463         PMD_INIT_LOG(DEBUG, "virtio_dev_close");
 464
 465         /* reset the NIC */
 466         if (pci_dev->driver->drv_flags & RTE_PCI_DRV_INTR_LSC)
 467                 vtpci_irq_config(hw, VIRTIO_MSI_NO_VECTOR);
 468         vtpci_reset(hw);
 469         hw->started = 0;
 470         virtio_dev_free_mbufs(dev);
 471         virtio_free_queues(dev);
 472 }
 473
 474 static void
 475 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
 476 {
 477         struct virtio_hw *hw = dev->data->dev_private;
 478         struct virtio_pmd_ctrl ctrl;
 479         int dlen[1];
 480         int ret;
 481
 482         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 483                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 484                 return;
 485         }
 486
 487         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 488         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 489         ctrl.data[0] = 1;
 490         dlen[0] = 1;
 491
 492         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 493         if (ret)
 494                 PMD_INIT_LOG(ERR, "Failed to enable promisc");
 495 }
 496
 497 static void
 498 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
 499 {
 500         struct virtio_hw *hw = dev->data->dev_private;
 501         struct virtio_pmd_ctrl ctrl;
 502         int dlen[1];
 503         int ret;
 504
 505         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 506                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 507                 return;
 508         }
 509
 510         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 511         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 512         ctrl.data[0] = 0;
 513         dlen[0] = 1;
 514
 515         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 516         if (ret)
 517                 PMD_INIT_LOG(ERR, "Failed to disable promisc");
 518 }
 519
 520 static void
 521 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
 522 {
 523         struct virtio_hw *hw = dev->data->dev_private;
 524         struct virtio_pmd_ctrl ctrl;
 525         int dlen[1];
 526         int ret;
 527
 528         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 529                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 530                 return;
 531         }
 532
 533         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 534         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 535         ctrl.data[0] = 1;
 536         dlen[0] = 1;
 537
 538         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 539         if (ret)
 540                 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
 541 }
 542
 543 static void
 544 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
 545 {
 546         struct virtio_hw *hw = dev->data->dev_private;
 547         struct virtio_pmd_ctrl ctrl;
 548         int dlen[1];
 549         int ret;
 550
 551         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 552                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 553                 return;
 554         }
 555
 556         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 557         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 558         ctrl.data[0] = 0;
 559         dlen[0] = 1;
 560
 561         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 562         if (ret)
 563                 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
 564 }
 565
 566 /*
 567  * dev_ops for virtio, bare necessities for basic operation
 568  */
 569 static const struct eth_dev_ops virtio_eth_dev_ops = {
 570         .dev_configure           = virtio_dev_configure,
 571         .dev_start               = virtio_dev_start,
 572         .dev_stop                = virtio_dev_stop,
 573         .dev_close               = virtio_dev_close,
 574         .promiscuous_enable      = virtio_dev_promiscuous_enable,
 575         .promiscuous_disable     = virtio_dev_promiscuous_disable,
 576         .allmulticast_enable     = virtio_dev_allmulticast_enable,
 577         .allmulticast_disable    = virtio_dev_allmulticast_disable,
 578
 579         .dev_infos_get           = virtio_dev_info_get,
 580         .stats_get               = virtio_dev_stats_get,
 581         .stats_reset             = virtio_dev_stats_reset,
 582         .link_update             = virtio_dev_link_update,
 583         .rx_queue_setup          = virtio_dev_rx_queue_setup,
 584         .rx_queue_release        = virtio_dev_rx_queue_release,
 585         .tx_queue_setup          = virtio_dev_tx_queue_setup,
 586         .tx_queue_release        = virtio_dev_tx_queue_release,
 587         /* collect stats per queue */
 588         .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
 589         .vlan_filter_set         = virtio_vlan_filter_set,
 590         .mac_addr_add            = virtio_mac_addr_add,
 591         .mac_addr_remove         = virtio_mac_addr_remove,
 592         .mac_addr_set            = virtio_mac_addr_set,
 593 };
 594
 595 static inline int
 596 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
 597                                 struct rte_eth_link *link)
 598 {
 599         struct rte_eth_link *dst = link;
 600         struct rte_eth_link *src = &(dev->data->dev_link);
 601
 602         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
 603                         *(uint64_t *)src) == 0)
 604                 return -1;
 605
 606         return 0;
 607 }
 608
 609 /**
 610  * Atomically writes the link status information into global
 611  * structure rte_eth_dev.
 612  *
 613  * @param dev
 614  *   - Pointer to the structure rte_eth_dev to read from.
 615  *   - Pointer to the buffer to be saved with the link status.
 616  *
 617  * @return
 618  *   - On success, zero.
 619  *   - On failure, negative value.
 620  */
 621 static inline int
 622 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
 623                 struct rte_eth_link *link)
 624 {
 625         struct rte_eth_link *dst = &(dev->data->dev_link);
 626         struct rte_eth_link *src = link;
 627
 628         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
 629                                         *(uint64_t *)src) == 0)
 630                 return -1;
 631
 632         return 0;
 633 }
 634
 635 static void
 636 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 637 {
 638         unsigned i;
 639
 640         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 641                 const struct virtqueue *txvq = dev->data->tx_queues[i];
 642                 if (txvq == NULL)
 643                         continue;
 644
 645                 stats->opackets += txvq->packets;
 646                 stats->obytes += txvq->bytes;
 647                 stats->oerrors += txvq->errors;
 648
 649                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 650                         stats->q_opackets[i] = txvq->packets;
 651                         stats->q_obytes[i] = txvq->bytes;
 652                 }
 653         }
 654
 655         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 656                 const struct virtqueue *rxvq = dev->data->rx_queues[i];
 657                 if (rxvq == NULL)
 658                         continue;
 659
 660                 stats->ipackets += rxvq->packets;
 661                 stats->ibytes += rxvq->bytes;
 662                 stats->ierrors += rxvq->errors;
 663
 664                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 665                         stats->q_ipackets[i] = rxvq->packets;
 666                         stats->q_ibytes[i] = rxvq->bytes;
 667                 }
 668         }
 669
 670         stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
 671 }
 672
 673 static void
 674 virtio_dev_stats_reset(struct rte_eth_dev *dev)
 675 {
 676         unsigned int i;
 677
 678         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 679                 struct virtqueue *txvq = dev->data->tx_queues[i];
 680                 if (txvq == NULL)
 681                         continue;
 682
 683                 txvq->packets = 0;
 684                 txvq->bytes = 0;
 685                 txvq->errors = 0;
 686         }
 687
 688         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 689                 struct virtqueue *rxvq = dev->data->rx_queues[i];
 690                 if (rxvq == NULL)
 691                         continue;
 692
 693                 rxvq->packets = 0;
 694                 rxvq->bytes = 0;
 695                 rxvq->errors = 0;
 696         }
 697
 698         dev->data->rx_mbuf_alloc_failed = 0;
 699 }
 700
 701 static void
 702 virtio_set_hwaddr(struct virtio_hw *hw)
 703 {
 704         vtpci_write_dev_config(hw,
 705                         offsetof(struct virtio_net_config, mac),
 706                         &hw->mac_addr, ETHER_ADDR_LEN);
 707 }
 708
 709 static void
 710 virtio_get_hwaddr(struct virtio_hw *hw)
 711 {
 712         if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
 713                 vtpci_read_dev_config(hw,
 714                         offsetof(struct virtio_net_config, mac),
 715                         &hw->mac_addr, ETHER_ADDR_LEN);
 716         } else {
 717                 eth_random_addr(&hw->mac_addr[0]);
 718                 virtio_set_hwaddr(hw);
 719         }
 720 }
 721
 722 static void
 723 virtio_mac_table_set(struct virtio_hw *hw,
 724                      const struct virtio_net_ctrl_mac *uc,
 725                      const struct virtio_net_ctrl_mac *mc)
 726 {
 727         struct virtio_pmd_ctrl ctrl;
 728         int err, len[2];
 729
 730         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
 731                 PMD_DRV_LOG(INFO, "host does not support mac table\n");
 732                 return;
 733         }
 734
 735         ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
 736         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
 737
 738         len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
 739         memcpy(ctrl.data, uc, len[0]);
 740
 741         len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
 742         memcpy(ctrl.data + len[0], mc, len[1]);
 743
 744         err = virtio_send_command(hw->cvq, &ctrl, len, 2);
 745         if (err != 0)
 746                 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
 747 }
 748
 749 static void
 750 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
 751                     uint32_t index, uint32_t vmdq __rte_unused)
 752 {
 753         struct virtio_hw *hw = dev->data->dev_private;
 754         const struct ether_addr *addrs = dev->data->mac_addrs;
 755         unsigned int i;
 756         struct virtio_net_ctrl_mac *uc, *mc;
 757
 758         if (index >= VIRTIO_MAX_MAC_ADDRS) {
 759                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
 760                 return;
 761         }
 762
 763         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
 764         uc->entries = 0;
 765         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
 766         mc->entries = 0;
 767
 768         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
 769                 const struct ether_addr *addr
 770                         = (i == index) ? mac_addr : addrs + i;
 771                 struct virtio_net_ctrl_mac *tbl
 772                         = is_multicast_ether_addr(addr) ? mc : uc;
 773
 774                 memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
 775         }
 776
 777         virtio_mac_table_set(hw, uc, mc);
 778 }
 779
 780 static void
 781 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
 782 {
 783         struct virtio_hw *hw = dev->data->dev_private;
 784         struct ether_addr *addrs = dev->data->mac_addrs;
 785         struct virtio_net_ctrl_mac *uc, *mc;
 786         unsigned int i;
 787
 788         if (index >= VIRTIO_MAX_MAC_ADDRS) {
 789                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
 790                 return;
 791         }
 792
 793         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
 794         uc->entries = 0;
 795         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
 796         mc->entries = 0;
 797
 798         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
 799                 struct virtio_net_ctrl_mac *tbl;
 800
 801                 if (i == index || is_zero_ether_addr(addrs + i))
 802                         continue;
 803
 804                 tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
 805                 memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
 806         }
 807
 808         virtio_mac_table_set(hw, uc, mc);
 809 }
 810
 811 static void
 812 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 813 {
 814         struct virtio_hw *hw = dev->data->dev_private;
 815
 816         memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
 817
 818         /* Use atomic update if available */
 819         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
 820                 struct virtio_pmd_ctrl ctrl;
 821                 int len = ETHER_ADDR_LEN;
 822
 823                 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
 824                 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
 825
 826                 memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
 827                 virtio_send_command(hw->cvq, &ctrl, &len, 1);
 828         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
 829                 virtio_set_hwaddr(hw);
 830 }
 831
 832 static int
 833 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
 834 {
 835         struct virtio_hw *hw = dev->data->dev_private;
 836         struct virtio_pmd_ctrl ctrl;
 837         int len;
 838
 839         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
 840                 return -ENOTSUP;
 841
 842         ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
 843         ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
 844         memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
 845         len = sizeof(vlan_id);
 846
 847         return virtio_send_command(hw->cvq, &ctrl, &len, 1);
 848 }
 849
 850 static void
 851 virtio_negotiate_features(struct virtio_hw *hw)
 852 {
 853         uint32_t host_features;
 854
 855         /* Prepare guest_features: feature that driver wants to support */
 856         hw->guest_features = VIRTIO_PMD_GUEST_FEATURES;
 857         PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %x",
 858                 hw->guest_features);
 859
 860         /* Read device(host) feature bits */
 861         host_features = VIRTIO_READ_REG_4(hw, VIRTIO_PCI_HOST_FEATURES);
 862         PMD_INIT_LOG(DEBUG, "host_features before negotiate = %x",
 863                 host_features);
 864
 865         /*
 866          * Negotiate features: Subset of device feature bits are written back
 867          * guest feature bits.
 868          */
 869         hw->guest_features = vtpci_negotiate_features(hw, host_features);
 870         PMD_INIT_LOG(DEBUG, "features after negotiate = %x",
 871                 hw->guest_features);
 872 }
 873
 874 #ifdef RTE_EXEC_ENV_LINUXAPP
 875 static int
 876 parse_sysfs_value(const char *filename, unsigned long *val)
 877 {
 878         FILE *f;
 879         char buf[BUFSIZ];
 880         char *end = NULL;
 881
 882         f = fopen(filename, "r");
 883         if (f == NULL) {
 884                 PMD_INIT_LOG(ERR, "%s(): cannot open sysfs value %s",
 885                              __func__, filename);
 886                 return -1;
 887         }
 888
 889         if (fgets(buf, sizeof(buf), f) == NULL) {
 890                 PMD_INIT_LOG(ERR, "%s(): cannot read sysfs value %s",
 891                              __func__, filename);
 892                 fclose(f);
 893                 return -1;
 894         }
 895         *val = strtoul(buf, &end, 0);
 896         if ((buf[0] == '\0') || (end == NULL) || (*end != '\n')) {
 897                 PMD_INIT_LOG(ERR, "%s(): cannot parse sysfs value %s",
 898                              __func__, filename);
 899                 fclose(f);
 900                 return -1;
 901         }
 902         fclose(f);
 903         return 0;
 904 }
 905
 906 static int get_uio_dev(struct rte_pci_addr *loc, char *buf, unsigned int buflen,
 907                         unsigned int *uio_num)
 908 {
 909         struct dirent *e;
 910         DIR *dir;
 911         char dirname[PATH_MAX];
 912
 913         /* depending on kernel version, uio can be located in uio/uioX
 914          * or uio:uioX */
 915         snprintf(dirname, sizeof(dirname),
 916                      SYSFS_PCI_DEVICES "/" PCI_PRI_FMT "/uio",
 917                      loc->domain, loc->bus, loc->devid, loc->function);
 918         dir = opendir(dirname);
 919         if (dir == NULL) {
 920                 /* retry with the parent directory */
 921                 snprintf(dirname, sizeof(dirname),
 922                              SYSFS_PCI_DEVICES "/" PCI_PRI_FMT,
 923                              loc->domain, loc->bus, loc->devid, loc->function);
 924                 dir = opendir(dirname);
 925
 926                 if (dir == NULL) {
 927                         PMD_INIT_LOG(ERR, "Cannot opendir %s", dirname);
 928                         return -1;
 929                 }
 930         }
 931
 932         /* take the first file starting with "uio" */
 933         while ((e = readdir(dir)) != NULL) {
 934                 /* format could be uio%d ...*/
 935                 int shortprefix_len = sizeof("uio") - 1;
 936                 /* ... or uio:uio%d */
 937                 int longprefix_len = sizeof("uio:uio") - 1;
 938                 char *endptr;
 939
 940                 if (strncmp(e->d_name, "uio", 3) != 0)
 941                         continue;
 942
 943                 /* first try uio%d */
 944                 errno = 0;
 945                 *uio_num = strtoull(e->d_name + shortprefix_len, &endptr, 10);
 946                 if (errno == 0 && endptr != (e->d_name + shortprefix_len)) {
 947                         snprintf(buf, buflen, "%s/uio%u", dirname, *uio_num);
 948                         break;
 949                 }
 950
 951                 /* then try uio:uio%d */
 952                 errno = 0;
 953                 *uio_num = strtoull(e->d_name + longprefix_len, &endptr, 10);
 954                 if (errno == 0 && endptr != (e->d_name + longprefix_len)) {
 955                         snprintf(buf, buflen, "%s/uio:uio%u", dirname,
 956                                      *uio_num);
 957                         break;
 958                 }
 959         }
 960         closedir(dir);
 961
 962         /* No uio resource found */
 963         if (e == NULL) {
 964                 PMD_INIT_LOG(ERR, "Could not find uio resource");
 965                 return -1;
 966         }
 967
 968         return 0;
 969 }
 970
 971 static int
 972 virtio_has_msix(const struct rte_pci_addr *loc)
 973 {
 974         DIR *d;
 975         char dirname[PATH_MAX];
 976
 977         snprintf(dirname, sizeof(dirname),
 978                      SYSFS_PCI_DEVICES "/" PCI_PRI_FMT "/msi_irqs",
 979                      loc->domain, loc->bus, loc->devid, loc->function);
 980
 981         d = opendir(dirname);
 982         if (d)
 983                 closedir(d);
 984
 985         return (d != NULL);
 986 }
 987
 988 /* Extract I/O port numbers from sysfs */
 989 static int virtio_resource_init_by_uio(struct rte_pci_device *pci_dev)
 990 {
 991         char dirname[PATH_MAX];
 992         char filename[PATH_MAX];
 993         unsigned long start, size;
 994         unsigned int uio_num;
 995
 996         if (get_uio_dev(&pci_dev->addr, dirname, sizeof(dirname), &uio_num) < 0)
 997                 return -1;
 998
 999         /* get portio size */
1000         snprintf(filename, sizeof(filename),
1001                      "%s/portio/port0/size", dirname);
1002         if (parse_sysfs_value(filename, &size) < 0) {
1003                 PMD_INIT_LOG(ERR, "%s(): cannot parse size",
1004                              __func__);
1005                 return -1;
1006         }
1007
1008         /* get portio start */
1009         snprintf(filename, sizeof(filename),
1010                  "%s/portio/port0/start", dirname);
1011         if (parse_sysfs_value(filename, &start) < 0) {
1012                 PMD_INIT_LOG(ERR, "%s(): cannot parse portio start",
1013                              __func__);
1014                 return -1;
1015         }
1016         pci_dev->mem_resource[0].addr = (void *)(uintptr_t)start;
1017         pci_dev->mem_resource[0].len =  (uint64_t)size;
1018         PMD_INIT_LOG(DEBUG,
1019                      "PCI Port IO found start=0x%lx with size=0x%lx",
1020                      start, size);
1021
1022         /* save fd */
1023         memset(dirname, 0, sizeof(dirname));
1024         snprintf(dirname, sizeof(dirname), "/dev/uio%u", uio_num);
1025         pci_dev->intr_handle.fd = open(dirname, O_RDWR);
1026         if (pci_dev->intr_handle.fd < 0) {
1027                 PMD_INIT_LOG(ERR, "Cannot open %s: %s\n",
1028                         dirname, strerror(errno));
1029                 return -1;
1030         }
1031
1032         pci_dev->intr_handle.type = RTE_INTR_HANDLE_UIO;
1033         pci_dev->driver->drv_flags |= RTE_PCI_DRV_INTR_LSC;
1034
1035         return 0;
1036 }
1037
1038 /* Extract port I/O numbers from proc/ioports */
1039 static int virtio_resource_init_by_ioports(struct rte_pci_device *pci_dev)
1040 {
1041         uint16_t start, end;
1042         int size;
1043         FILE *fp;
1044         char *line = NULL;
1045         char pci_id[16];
1046         int found = 0;
1047         size_t linesz;
1048
1049         snprintf(pci_id, sizeof(pci_id), PCI_PRI_FMT,
1050                  pci_dev->addr.domain,
1051                  pci_dev->addr.bus,
1052                  pci_dev->addr.devid,
1053                  pci_dev->addr.function);
1054
1055         fp = fopen("/proc/ioports", "r");
1056         if (fp == NULL) {
1057                 PMD_INIT_LOG(ERR, "%s(): can't open ioports", __func__);
1058                 return -1;
1059         }
1060
1061         while (getdelim(&line, &linesz, '\n', fp) > 0) {
1062                 char *ptr = line;
1063                 char *left;
1064                 int n;
1065
1066                 n = strcspn(ptr, ":");
1067                 ptr[n] = 0;
1068                 left = &ptr[n+1];
1069
1070                 while (*left && isspace(*left))
1071                         left++;
1072
1073                 if (!strncmp(left, pci_id, strlen(pci_id))) {
1074                         found = 1;
1075
1076                         while (*ptr && isspace(*ptr))
1077                                 ptr++;
1078
1079                         sscanf(ptr, "%04hx-%04hx", &start, &end);
1080                         size = end - start + 1;
1081
1082                         break;
1083                 }
1084         }
1085
1086         free(line);
1087         fclose(fp);
1088
1089         if (!found)
1090                 return -1;
1091
1092         pci_dev->mem_resource[0].addr = (void *)(uintptr_t)(uint32_t)start;
1093         pci_dev->mem_resource[0].len =  (uint64_t)size;
1094         PMD_INIT_LOG(DEBUG,
1095                 "PCI Port IO found start=0x%x with size=0x%x",
1096                 start, size);
1097
1098         /* can't support lsc interrupt without uio */
1099         pci_dev->driver->drv_flags &= ~RTE_PCI_DRV_INTR_LSC;
1100
1101         return 0;
1102 }
1103
1104 /* Extract I/O port numbers from sysfs */
1105 static int virtio_resource_init(struct rte_pci_device *pci_dev)
1106 {
1107         if (virtio_resource_init_by_uio(pci_dev) == 0)
1108                 return 0;
1109         else
1110                 return virtio_resource_init_by_ioports(pci_dev);
1111 }
1112
1113 #else
1114 static int
1115 virtio_has_msix(const struct rte_pci_addr *loc __rte_unused)
1116 {
1117         /* nic_uio does not enable interrupts, return 0 (false). */
1118         return 0;
1119 }
1120
1121 static int virtio_resource_init(struct rte_pci_device *pci_dev __rte_unused)
1122 {
1123         /* no setup required */
1124         return 0;
1125 }
1126 #endif
1127
1128 /*
1129  * Process Virtio Config changed interrupt and call the callback
1130  * if link state changed.
1131  */
1132 static void
1133 virtio_interrupt_handler(__rte_unused struct rte_intr_handle *handle,
1134                          void *param)
1135 {
1136         struct rte_eth_dev *dev = param;
1137         struct virtio_hw *hw = dev->data->dev_private;
1138         uint8_t isr;
1139
1140         /* Read interrupt status which clears interrupt */
1141         isr = vtpci_isr(hw);
1142         PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1143
1144         if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0)
1145                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1146
1147         if (isr & VIRTIO_PCI_ISR_CONFIG) {
1148                 if (virtio_dev_link_update(dev, 0) == 0)
1149                         _rte_eth_dev_callback_process(dev,
1150                                                       RTE_ETH_EVENT_INTR_LSC);
1151         }
1152
1153 }
1154
1155 static void
1156 rx_func_get(struct rte_eth_dev *eth_dev)
1157 {
1158         struct virtio_hw *hw = eth_dev->data->dev_private;
1159         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF))
1160                 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1161         else
1162                 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1163 }
1164
1165 /*
1166  * This function is based on probe() function in virtio_pci.c
1167  * It returns 0 on success.
1168  */
1169 static int
1170 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1171 {
1172         struct virtio_hw *hw = eth_dev->data->dev_private;
1173         struct virtio_net_config *config;
1174         struct virtio_net_config local_config;
1175         uint32_t offset_conf = sizeof(config->mac);
1176         struct rte_pci_device *pci_dev;
1177
1178         RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr));
1179
1180         eth_dev->dev_ops = &virtio_eth_dev_ops;
1181         eth_dev->tx_pkt_burst = &virtio_xmit_pkts;
1182
1183         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1184                 rx_func_get(eth_dev);
1185                 return 0;
1186         }
1187
1188         /* Allocate memory for storing MAC addresses */
1189         eth_dev->data->mac_addrs = rte_zmalloc("virtio", ETHER_ADDR_LEN, 0);
1190         if (eth_dev->data->mac_addrs == NULL) {
1191                 PMD_INIT_LOG(ERR,
1192                         "Failed to allocate %d bytes needed to store MAC addresses",
1193                         ETHER_ADDR_LEN);
1194                 return -ENOMEM;
1195         }
1196
1197         pci_dev = eth_dev->pci_dev;
1198         if (virtio_resource_init(pci_dev) < 0)
1199                 return -1;
1200
1201         hw->use_msix = virtio_has_msix(&pci_dev->addr);
1202         hw->io_base = (uint32_t)(uintptr_t)pci_dev->mem_resource[0].addr;
1203
1204         /* Reset the device although not necessary at startup */
1205         vtpci_reset(hw);
1206
1207         /* Tell the host we've noticed this device. */
1208         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1209
1210         /* Tell the host we've known how to drive the device. */
1211         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1212         virtio_negotiate_features(hw);
1213
1214         rx_func_get(eth_dev);
1215
1216         /* Setting up rx_header size for the device */
1217         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF))
1218                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1219         else
1220                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1221
1222         /* Copy the permanent MAC address to: virtio_hw */
1223         virtio_get_hwaddr(hw);
1224         ether_addr_copy((struct ether_addr *) hw->mac_addr,
1225                         &eth_dev->data->mac_addrs[0]);
1226         PMD_INIT_LOG(DEBUG,
1227                      "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1228                      hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1229                      hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1230
1231         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1232                 config = &local_config;
1233
1234                 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1235                         offset_conf += sizeof(config->status);
1236                 } else {
1237                         PMD_INIT_LOG(DEBUG,
1238                                      "VIRTIO_NET_F_STATUS is not supported");
1239                         config->status = 0;
1240                 }
1241
1242                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1243                         offset_conf += sizeof(config->max_virtqueue_pairs);
1244                 } else {
1245                         PMD_INIT_LOG(DEBUG,
1246                                      "VIRTIO_NET_F_MQ is not supported");
1247                         config->max_virtqueue_pairs = 1;
1248                 }
1249
1250                 vtpci_read_dev_config(hw, 0, (uint8_t *)config, offset_conf);
1251
1252                 hw->max_rx_queues =
1253                         (VIRTIO_MAX_RX_QUEUES < config->max_virtqueue_pairs) ?
1254                         VIRTIO_MAX_RX_QUEUES : config->max_virtqueue_pairs;
1255                 hw->max_tx_queues =
1256                         (VIRTIO_MAX_TX_QUEUES < config->max_virtqueue_pairs) ?
1257                         VIRTIO_MAX_TX_QUEUES : config->max_virtqueue_pairs;
1258
1259                 virtio_dev_cq_queue_setup(eth_dev,
1260                                         config->max_virtqueue_pairs * 2,
1261                                         SOCKET_ID_ANY);
1262
1263                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1264                                 config->max_virtqueue_pairs);
1265                 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1266                 PMD_INIT_LOG(DEBUG,
1267                                 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1268                                 config->mac[0], config->mac[1],
1269                                 config->mac[2], config->mac[3],
1270                                 config->mac[4], config->mac[5]);
1271         } else {
1272                 hw->max_rx_queues = 1;
1273                 hw->max_tx_queues = 1;
1274         }
1275
1276         eth_dev->data->nb_rx_queues = hw->max_rx_queues;
1277         eth_dev->data->nb_tx_queues = hw->max_tx_queues;
1278
1279         PMD_INIT_LOG(DEBUG, "hw->max_rx_queues=%d   hw->max_tx_queues=%d",
1280                         hw->max_rx_queues, hw->max_tx_queues);
1281         PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1282                         eth_dev->data->port_id, pci_dev->id.vendor_id,
1283                         pci_dev->id.device_id);
1284
1285         /* Setup interrupt callback  */
1286         if (pci_dev->driver->drv_flags & RTE_PCI_DRV_INTR_LSC)
1287                 rte_intr_callback_register(&pci_dev->intr_handle,
1288                                    virtio_interrupt_handler, eth_dev);
1289
1290         virtio_dev_cq_start(eth_dev);
1291
1292         return 0;
1293 }
1294
1295 static int
1296 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1297 {
1298         struct rte_pci_device *pci_dev;
1299         struct virtio_hw *hw = eth_dev->data->dev_private;
1300
1301         PMD_INIT_FUNC_TRACE();
1302
1303         if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1304                 return -EPERM;
1305
1306         if (hw->started == 1) {
1307                 virtio_dev_stop(eth_dev);
1308                 virtio_dev_close(eth_dev);
1309         }
1310         pci_dev = eth_dev->pci_dev;
1311
1312         eth_dev->dev_ops = NULL;
1313         eth_dev->tx_pkt_burst = NULL;
1314         eth_dev->rx_pkt_burst = NULL;
1315
1316         virtio_dev_queue_release(hw->cvq);
1317
1318         rte_free(eth_dev->data->mac_addrs);
1319         eth_dev->data->mac_addrs = NULL;
1320
1321         /* reset interrupt callback  */
1322         if (pci_dev->driver->drv_flags & RTE_PCI_DRV_INTR_LSC)
1323                 rte_intr_callback_unregister(&pci_dev->intr_handle,
1324                                                 virtio_interrupt_handler,
1325                                                 eth_dev);
1326
1327         PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1328
1329         return 0;
1330 }
1331
1332 static struct eth_driver rte_virtio_pmd = {
1333         .pci_drv = {
1334                 .name = "rte_virtio_pmd",
1335                 .id_table = pci_id_virtio_map,
1336                 .drv_flags = RTE_PCI_DRV_DETACHABLE,
1337         },
1338         .eth_dev_init = eth_virtio_dev_init,
1339         .eth_dev_uninit = eth_virtio_dev_uninit,
1340         .dev_private_size = sizeof(struct virtio_hw),
1341 };
1342
1343 /*
1344  * Driver initialization routine.
1345  * Invoked once at EAL init time.
1346  * Register itself as the [Poll Mode] Driver of PCI virtio devices.
1347  * Returns 0 on success.
1348  */
1349 static int
1350 rte_virtio_pmd_init(const char *name __rte_unused,
1351                     const char *param __rte_unused)
1352 {
1353         if (rte_eal_iopl_init() != 0) {
1354                 PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1355                 return -1;
1356         }
1357
1358         rte_eth_driver_register(&rte_virtio_pmd);
1359         return 0;
1360 }
1361
1362 /*
1363  * Configure virtio device
1364  * It returns 0 on success.
1365  */
1366 static int
1367 virtio_dev_configure(struct rte_eth_dev *dev)
1368 {
1369         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1370         struct virtio_hw *hw = dev->data->dev_private;
1371         struct rte_pci_device *pci_dev = dev->pci_dev;
1372
1373         PMD_INIT_LOG(DEBUG, "configure");
1374
1375         if (rxmode->hw_ip_checksum) {
1376                 PMD_DRV_LOG(ERR, "HW IP checksum not supported");
1377                 return (-EINVAL);
1378         }
1379
1380         hw->vlan_strip = rxmode->hw_vlan_strip;
1381
1382         if (rxmode->hw_vlan_filter
1383             && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1384                 PMD_DRV_LOG(NOTICE,
1385                             "vlan filtering not available on this host");
1386                 return -ENOTSUP;
1387         }
1388
1389         if (pci_dev->driver->drv_flags & RTE_PCI_DRV_INTR_LSC)
1390                 if (vtpci_irq_config(hw, 0) == VIRTIO_MSI_NO_VECTOR) {
1391                         PMD_DRV_LOG(ERR, "failed to set config vector");
1392                         return -EBUSY;
1393                 }
1394
1395         return 0;
1396 }
1397
1398
1399 static int
1400 virtio_dev_start(struct rte_eth_dev *dev)
1401 {
1402         uint16_t nb_queues, i;
1403         struct virtio_hw *hw = dev->data->dev_private;
1404         struct rte_pci_device *pci_dev = dev->pci_dev;
1405
1406         /* check if lsc interrupt feature is enabled */
1407         if ((dev->data->dev_conf.intr_conf.lsc) &&
1408                 (pci_dev->driver->drv_flags & RTE_PCI_DRV_INTR_LSC)) {
1409                 if (!vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1410                         PMD_DRV_LOG(ERR, "link status not supported by host");
1411                         return -ENOTSUP;
1412                 }
1413
1414                 if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0) {
1415                         PMD_DRV_LOG(ERR, "interrupt enable failed");
1416                         return -EIO;
1417                 }
1418         }
1419
1420         /* Initialize Link state */
1421         virtio_dev_link_update(dev, 0);
1422
1423         /* On restart after stop do not touch queues */
1424         if (hw->started)
1425                 return 0;
1426
1427         /* Do final configuration before rx/tx engine starts */
1428         virtio_dev_rxtx_start(dev);
1429         vtpci_reinit_complete(hw);
1430
1431         hw->started = 1;
1432
1433         /*Notify the backend
1434          *Otherwise the tap backend might already stop its queue due to fullness.
1435          *vhost backend will have no chance to be waked up
1436          */
1437         nb_queues = dev->data->nb_rx_queues;
1438         if (nb_queues > 1) {
1439                 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1440                         return -EINVAL;
1441         }
1442
1443         PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1444
1445         for (i = 0; i < nb_queues; i++)
1446                 virtqueue_notify(dev->data->rx_queues[i]);
1447
1448         PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1449
1450         for (i = 0; i < dev->data->nb_rx_queues; i++)
1451                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1452
1453         for (i = 0; i < dev->data->nb_tx_queues; i++)
1454                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1455
1456         return 0;
1457 }
1458
1459 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1460 {
1461         struct rte_mbuf *buf;
1462         int i, mbuf_num = 0;
1463
1464         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1465                 PMD_INIT_LOG(DEBUG,
1466                              "Before freeing rxq[%d] used and unused buf", i);
1467                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1468
1469                 PMD_INIT_LOG(DEBUG, "rx_queues[%d]=%p",
1470                                 i, dev->data->rx_queues[i]);
1471                 while ((buf = (struct rte_mbuf *)virtqueue_detatch_unused(
1472                                         dev->data->rx_queues[i])) != NULL) {
1473                         rte_pktmbuf_free(buf);
1474                         mbuf_num++;
1475                 }
1476
1477                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1478                 PMD_INIT_LOG(DEBUG,
1479                              "After freeing rxq[%d] used and unused buf", i);
1480                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1481         }
1482
1483         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1484                 PMD_INIT_LOG(DEBUG,
1485                              "Before freeing txq[%d] used and unused bufs",
1486                              i);
1487                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1488
1489                 mbuf_num = 0;
1490                 while ((buf = (struct rte_mbuf *)virtqueue_detatch_unused(
1491                                         dev->data->tx_queues[i])) != NULL) {
1492                         rte_pktmbuf_free(buf);
1493
1494                         mbuf_num++;
1495                 }
1496
1497                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1498                 PMD_INIT_LOG(DEBUG,
1499                              "After freeing txq[%d] used and unused buf", i);
1500                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1501         }
1502 }
1503
1504 /*
1505  * Stop device: disable interrupt and mark link down
1506  */
1507 static void
1508 virtio_dev_stop(struct rte_eth_dev *dev)
1509 {
1510         struct rte_eth_link link;
1511
1512         PMD_INIT_LOG(DEBUG, "stop");
1513
1514         if (dev->data->dev_conf.intr_conf.lsc)
1515                 rte_intr_disable(&dev->pci_dev->intr_handle);
1516
1517         memset(&link, 0, sizeof(link));
1518         virtio_dev_atomic_write_link_status(dev, &link);
1519 }
1520
1521 static int
1522 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
1523 {
1524         struct rte_eth_link link, old;
1525         uint16_t status;
1526         struct virtio_hw *hw = dev->data->dev_private;
1527         memset(&link, 0, sizeof(link));
1528         virtio_dev_atomic_read_link_status(dev, &link);
1529         old = link;
1530         link.link_duplex = FULL_DUPLEX;
1531         link.link_speed  = SPEED_10G;
1532
1533         if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1534                 PMD_INIT_LOG(DEBUG, "Get link status from hw");
1535                 vtpci_read_dev_config(hw,
1536                                 offsetof(struct virtio_net_config, status),
1537                                 &status, sizeof(status));
1538                 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
1539                         link.link_status = 0;
1540                         PMD_INIT_LOG(DEBUG, "Port %d is down",
1541                                      dev->data->port_id);
1542                 } else {
1543                         link.link_status = 1;
1544                         PMD_INIT_LOG(DEBUG, "Port %d is up",
1545                                      dev->data->port_id);
1546                 }
1547         } else {
1548                 link.link_status = 1;   /* Link up */
1549         }
1550         virtio_dev_atomic_write_link_status(dev, &link);
1551
1552         return (old.link_status == link.link_status) ? -1 : 0;
1553 }
1554
1555 static void
1556 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1557 {
1558         struct virtio_hw *hw = dev->data->dev_private;
1559
1560         dev_info->driver_name = dev->driver->pci_drv.name;
1561         dev_info->max_rx_queues = (uint16_t)hw->max_rx_queues;
1562         dev_info->max_tx_queues = (uint16_t)hw->max_tx_queues;
1563         dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
1564         dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
1565         dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
1566         dev_info->default_txconf = (struct rte_eth_txconf) {
1567                 .txq_flags = ETH_TXQ_FLAGS_NOOFFLOADS
1568         };
1569 }
1570
1571 /*
1572  * It enables testpmd to collect per queue stats.
1573  */
1574 static int
1575 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
1576 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
1577 __rte_unused uint8_t is_rx)
1578 {
1579         return 0;
1580 }
1581
1582 static struct rte_driver rte_virtio_driver = {
1583         .type = PMD_PDEV,
1584         .init = rte_virtio_pmd_init,
1585 };
1586
1587 PMD_REGISTER_DRIVER(rte_virtio_driver);