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