drivers/net/virtio/virtio_ethdev.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
   5  *   All rights reserved.
   6  *
   7  *   Redistribution and use in source and binary forms, with or without
   8  *   modification, are permitted provided that the following conditions
   9  *   are met:
  10  *
  11  *     * Redistributions of source code must retain the above copyright
  12  *       notice, this list of conditions and the following disclaimer.
  13  *     * Redistributions in binary form must reproduce the above copyright
  14  *       notice, this list of conditions and the following disclaimer in
  15  *       the documentation and/or other materials provided with the
  16  *       distribution.
  17  *     * Neither the name of Intel Corporation nor the names of its
  18  *       contributors may be used to endorse or promote products derived
  19  *       from this software without specific prior written permission.
  20  *
  21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  */
  33
  34 #include <stdint.h>
  35 #include <string.h>
  36 #include <stdio.h>
  37 #include <errno.h>
  38 #include <unistd.h>
  39 #ifdef RTE_EXEC_ENV_LINUXAPP
  40 #include <dirent.h>
  41 #include <fcntl.h>
  42 #endif
  43
  44 #include <rte_ethdev.h>
  45 #include <rte_memcpy.h>
  46 #include <rte_string_fns.h>
  47 #include <rte_memzone.h>
  48 #include <rte_malloc.h>
  49 #include <rte_atomic.h>
  50 #include <rte_branch_prediction.h>
  51 #include <rte_pci.h>
  52 #include <rte_ether.h>
  53 #include <rte_common.h>
  54
  55 #include <rte_memory.h>
  56 #include <rte_eal.h>
  57 #include <rte_dev.h>
  58
  59 #include "virtio_ethdev.h"
  60 #include "virtio_pci.h"
  61 #include "virtio_logs.h"
  62 #include "virtqueue.h"
  63
  64
  65 static int eth_virtio_dev_init(struct rte_eth_dev *eth_dev);
  66 static int  virtio_dev_configure(struct rte_eth_dev *dev);
  67 static int  virtio_dev_start(struct rte_eth_dev *dev);
  68 static void virtio_dev_stop(struct rte_eth_dev *dev);
  69 static void virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
  70 static void virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
  71 static void virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
  72 static void virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
  73 static void virtio_dev_info_get(struct rte_eth_dev *dev,
  74                                 struct rte_eth_dev_info *dev_info);
  75 static int virtio_dev_link_update(struct rte_eth_dev *dev,
  76         __rte_unused int wait_to_complete);
  77
  78 static void virtio_set_hwaddr(struct virtio_hw *hw);
  79 static void virtio_get_hwaddr(struct virtio_hw *hw);
  80
  81 static void virtio_dev_rx_queue_release(__rte_unused void *rxq);
  82 static void virtio_dev_tx_queue_release(__rte_unused void *txq);
  83
  84 static void virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats);
  85 static void virtio_dev_stats_reset(struct rte_eth_dev *dev);
  86 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
  87 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
  88                                 uint16_t vlan_id, int on);
  89 static void virtio_mac_addr_add(struct rte_eth_dev *dev,
  90                                 struct ether_addr *mac_addr,
  91                                 uint32_t index, uint32_t vmdq __rte_unused);
  92 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
  93 static void virtio_mac_addr_set(struct rte_eth_dev *dev,
  94                                 struct ether_addr *mac_addr);
  95
  96 static int virtio_dev_queue_stats_mapping_set(
  97         __rte_unused struct rte_eth_dev *eth_dev,
  98         __rte_unused uint16_t queue_id,
  99         __rte_unused uint8_t stat_idx,
 100         __rte_unused uint8_t is_rx);
 101
 102 /*
 103  * The set of PCI devices this driver supports
 104  */
 105 static const struct rte_pci_id pci_id_virtio_map[] = {
 106
 107 #define RTE_PCI_DEV_ID_DECL_VIRTIO(vend, dev) {RTE_PCI_DEVICE(vend, dev)},
 108 #include "rte_pci_dev_ids.h"
 109
 110 { .vendor_id = 0, /* sentinel */ },
 111 };
 112
 113 static int
 114 virtio_send_command(struct virtqueue *vq, struct virtio_pmd_ctrl *ctrl,
 115                 int *dlen, int pkt_num)
 116 {
 117         uint16_t head = vq->vq_desc_head_idx, i;
 118         int k, sum = 0;
 119         virtio_net_ctrl_ack status = ~0;
 120         struct virtio_pmd_ctrl result;
 121
 122         ctrl->status = status;
 123
 124         if (!vq->hw->cvq) {
 125                 PMD_INIT_LOG(ERR,
 126                              "%s(): Control queue is not supported.",
 127                              __func__);
 128                 return -1;
 129         }
 130
 131         PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
 132                 "vq->hw->cvq = %p vq = %p",
 133                 vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
 134
 135         if ((vq->vq_free_cnt < ((uint32_t)pkt_num + 2)) || (pkt_num < 1))
 136                 return -1;
 137
 138         memcpy(vq->virtio_net_hdr_mz->addr, ctrl,
 139                 sizeof(struct virtio_pmd_ctrl));
 140
 141         /*
 142          * Format is enforced in qemu code:
 143          * One TX packet for header;
 144          * At least one TX packet per argument;
 145          * One RX packet for ACK.
 146          */
 147         vq->vq_ring.desc[head].flags = VRING_DESC_F_NEXT;
 148         vq->vq_ring.desc[head].addr = vq->virtio_net_hdr_mz->phys_addr;
 149         vq->vq_ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
 150         vq->vq_free_cnt--;
 151         i = vq->vq_ring.desc[head].next;
 152
 153         for (k = 0; k < pkt_num; k++) {
 154                 vq->vq_ring.desc[i].flags = VRING_DESC_F_NEXT;
 155                 vq->vq_ring.desc[i].addr = vq->virtio_net_hdr_mz->phys_addr
 156                         + sizeof(struct virtio_net_ctrl_hdr)
 157                         + sizeof(ctrl->status) + sizeof(uint8_t)*sum;
 158                 vq->vq_ring.desc[i].len = dlen[k];
 159                 sum += dlen[k];
 160                 vq->vq_free_cnt--;
 161                 i = vq->vq_ring.desc[i].next;
 162         }
 163
 164         vq->vq_ring.desc[i].flags = VRING_DESC_F_WRITE;
 165         vq->vq_ring.desc[i].addr = vq->virtio_net_hdr_mz->phys_addr
 166                         + sizeof(struct virtio_net_ctrl_hdr);
 167         vq->vq_ring.desc[i].len = sizeof(ctrl->status);
 168         vq->vq_free_cnt--;
 169
 170         vq->vq_desc_head_idx = vq->vq_ring.desc[i].next;
 171
 172         vq_update_avail_ring(vq, head);
 173         vq_update_avail_idx(vq);
 174
 175         PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
 176
 177         virtqueue_notify(vq);
 178
 179         rte_rmb();
 180         while (vq->vq_used_cons_idx == vq->vq_ring.used->idx) {
 181                 rte_rmb();
 182                 usleep(100);
 183         }
 184
 185         while (vq->vq_used_cons_idx != vq->vq_ring.used->idx) {
 186                 uint32_t idx, desc_idx, used_idx;
 187                 struct vring_used_elem *uep;
 188
 189                 used_idx = (uint32_t)(vq->vq_used_cons_idx
 190                                 & (vq->vq_nentries - 1));
 191                 uep = &vq->vq_ring.used->ring[used_idx];
 192                 idx = (uint32_t) uep->id;
 193                 desc_idx = idx;
 194
 195                 while (vq->vq_ring.desc[desc_idx].flags & VRING_DESC_F_NEXT) {
 196                         desc_idx = vq->vq_ring.desc[desc_idx].next;
 197                         vq->vq_free_cnt++;
 198                 }
 199
 200                 vq->vq_ring.desc[desc_idx].next = vq->vq_desc_head_idx;
 201                 vq->vq_desc_head_idx = idx;
 202
 203                 vq->vq_used_cons_idx++;
 204                 vq->vq_free_cnt++;
 205         }
 206
 207         PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
 208                         vq->vq_free_cnt, vq->vq_desc_head_idx);
 209
 210         memcpy(&result, vq->virtio_net_hdr_mz->addr,
 211                         sizeof(struct virtio_pmd_ctrl));
 212
 213         return result.status;
 214 }
 215
 216 static int
 217 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
 218 {
 219         struct virtio_hw *hw = dev->data->dev_private;
 220         struct virtio_pmd_ctrl ctrl;
 221         int dlen[1];
 222         int ret;
 223
 224         ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
 225         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
 226         memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
 227
 228         dlen[0] = sizeof(uint16_t);
 229
 230         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 231         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 int virtio_dev_queue_setup(struct rte_eth_dev *dev,
 241                         int queue_type,
 242                         uint16_t queue_idx,
 243                         uint16_t  vtpci_queue_idx,
 244                         uint16_t nb_desc,
 245                         unsigned int socket_id,
 246                         struct virtqueue **pvq)
 247 {
 248         char vq_name[VIRTQUEUE_MAX_NAME_SZ];
 249         const struct rte_memzone *mz;
 250         uint16_t vq_size;
 251         int size;
 252         struct virtio_hw *hw = dev->data->dev_private;
 253         struct virtqueue  *vq = NULL;
 254
 255         /* Write the virtqueue index to the Queue Select Field */
 256         VIRTIO_WRITE_REG_2(hw, VIRTIO_PCI_QUEUE_SEL, vtpci_queue_idx);
 257         PMD_INIT_LOG(DEBUG, "selecting queue: %d", vtpci_queue_idx);
 258
 259         /*
 260          * Read the virtqueue size from the Queue Size field
 261          * Always power of 2 and if 0 virtqueue does not exist
 262          */
 263         vq_size = VIRTIO_READ_REG_2(hw, VIRTIO_PCI_QUEUE_NUM);
 264         PMD_INIT_LOG(DEBUG, "vq_size: %d nb_desc:%d", vq_size, nb_desc);
 265         if (nb_desc == 0)
 266                 nb_desc = vq_size;
 267         if (vq_size == 0) {
 268                 PMD_INIT_LOG(ERR, "%s: virtqueue does not exist", __func__);
 269                 return -EINVAL;
 270         } else if (!rte_is_power_of_2(vq_size)) {
 271                 PMD_INIT_LOG(ERR, "%s: virtqueue size is not powerof 2", __func__);
 272                 return -EINVAL;
 273         } else if (nb_desc != vq_size) {
 274                 PMD_INIT_LOG(ERR, "Warning: nb_desc(%d) is not equal to vq size (%d), fall to vq size",
 275                         nb_desc, vq_size);
 276                 nb_desc = vq_size;
 277         }
 278
 279         if (queue_type == VTNET_RQ) {
 280                 snprintf(vq_name, sizeof(vq_name), "port%d_rvq%d",
 281                         dev->data->port_id, queue_idx);
 282                 vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
 283                         vq_size * sizeof(struct vq_desc_extra), RTE_CACHE_LINE_SIZE);
 284         } else if (queue_type == VTNET_TQ) {
 285                 snprintf(vq_name, sizeof(vq_name), "port%d_tvq%d",
 286                         dev->data->port_id, queue_idx);
 287                 vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
 288                         vq_size * sizeof(struct vq_desc_extra), RTE_CACHE_LINE_SIZE);
 289         } else if (queue_type == VTNET_CQ) {
 290                 snprintf(vq_name, sizeof(vq_name), "port%d_cvq",
 291                         dev->data->port_id);
 292                 vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
 293                         vq_size * sizeof(struct vq_desc_extra),
 294                         RTE_CACHE_LINE_SIZE);
 295         }
 296         if (vq == NULL) {
 297                 PMD_INIT_LOG(ERR, "%s: Can not allocate virtqueue", __func__);
 298                 return (-ENOMEM);
 299         }
 300
 301         vq->hw = hw;
 302         vq->port_id = dev->data->port_id;
 303         vq->queue_id = queue_idx;
 304         vq->vq_queue_index = vtpci_queue_idx;
 305         vq->vq_nentries = vq_size;
 306         vq->vq_free_cnt = vq_size;
 307
 308         /*
 309          * Reserve a memzone for vring elements
 310          */
 311         size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
 312         vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
 313         PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d", size, vq->vq_ring_size);
 314
 315         mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
 316                 socket_id, 0, VIRTIO_PCI_VRING_ALIGN);
 317         if (mz == NULL) {
 318                 rte_free(vq);
 319                 return -ENOMEM;
 320         }
 321
 322         /*
 323          * Virtio PCI device VIRTIO_PCI_QUEUE_PF register is 32bit,
 324          * and only accepts 32 bit page frame number.
 325          * Check if the allocated physical memory exceeds 16TB.
 326          */
 327         if ((mz->phys_addr + vq->vq_ring_size - 1) >> (VIRTIO_PCI_QUEUE_ADDR_SHIFT + 32)) {
 328                 PMD_INIT_LOG(ERR, "vring address shouldn't be above 16TB!");
 329                 rte_free(vq);
 330                 return -ENOMEM;
 331         }
 332
 333         memset(mz->addr, 0, sizeof(mz->len));
 334         vq->mz = mz;
 335         vq->vq_ring_mem = mz->phys_addr;
 336         vq->vq_ring_virt_mem = mz->addr;
 337         PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%"PRIx64, (uint64_t)mz->phys_addr);
 338         PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%"PRIx64, (uint64_t)(uintptr_t)mz->addr);
 339         vq->virtio_net_hdr_mz  = NULL;
 340         vq->virtio_net_hdr_mem = 0;
 341
 342         if (queue_type == VTNET_TQ) {
 343                 /*
 344                  * For each xmit packet, allocate a virtio_net_hdr
 345                  */
 346                 snprintf(vq_name, sizeof(vq_name), "port%d_tvq%d_hdrzone",
 347                         dev->data->port_id, queue_idx);
 348                 vq->virtio_net_hdr_mz = rte_memzone_reserve_aligned(vq_name,
 349                         vq_size * hw->vtnet_hdr_size,
 350                         socket_id, 0, RTE_CACHE_LINE_SIZE);
 351                 if (vq->virtio_net_hdr_mz == NULL) {
 352                         rte_free(vq);
 353                         return -ENOMEM;
 354                 }
 355                 vq->virtio_net_hdr_mem =
 356                         vq->virtio_net_hdr_mz->phys_addr;
 357                 memset(vq->virtio_net_hdr_mz->addr, 0,
 358                         vq_size * hw->vtnet_hdr_size);
 359         } else if (queue_type == VTNET_CQ) {
 360                 /* Allocate a page for control vq command, data and status */
 361                 snprintf(vq_name, sizeof(vq_name), "port%d_cvq_hdrzone",
 362                         dev->data->port_id);
 363                 vq->virtio_net_hdr_mz = rte_memzone_reserve_aligned(vq_name,
 364                         PAGE_SIZE, socket_id, 0, RTE_CACHE_LINE_SIZE);
 365                 if (vq->virtio_net_hdr_mz == NULL) {
 366                         rte_free(vq);
 367                         return -ENOMEM;
 368                 }
 369                 vq->virtio_net_hdr_mem =
 370                         vq->virtio_net_hdr_mz->phys_addr;
 371                 memset(vq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
 372         }
 373
 374         /*
 375          * Set guest physical address of the virtqueue
 376          * in VIRTIO_PCI_QUEUE_PFN config register of device
 377          */
 378         VIRTIO_WRITE_REG_4(hw, VIRTIO_PCI_QUEUE_PFN,
 379                         mz->phys_addr >> VIRTIO_PCI_QUEUE_ADDR_SHIFT);
 380         *pvq = vq;
 381         return 0;
 382 }
 383
 384 static int
 385 virtio_dev_cq_queue_setup(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx,
 386                 uint32_t socket_id)
 387 {
 388         struct virtqueue *vq;
 389         uint16_t nb_desc = 0;
 390         int ret;
 391         struct virtio_hw *hw = dev->data->dev_private;
 392
 393         PMD_INIT_FUNC_TRACE();
 394         ret = virtio_dev_queue_setup(dev, VTNET_CQ, VTNET_SQ_CQ_QUEUE_IDX,
 395                         vtpci_queue_idx, nb_desc, socket_id, &vq);
 396         if (ret < 0) {
 397                 PMD_INIT_LOG(ERR, "control vq initialization failed");
 398                 return ret;
 399         }
 400
 401         hw->cvq = vq;
 402         return 0;
 403 }
 404
 405 static void
 406 virtio_dev_close(struct rte_eth_dev *dev)
 407 {
 408         struct virtio_hw *hw = dev->data->dev_private;
 409         struct rte_pci_device *pci_dev = dev->pci_dev;
 410
 411         PMD_INIT_LOG(DEBUG, "virtio_dev_close");
 412
 413         /* reset the NIC */
 414         if (pci_dev->driver->drv_flags & RTE_PCI_DRV_INTR_LSC)
 415                 vtpci_irq_config(hw, VIRTIO_MSI_NO_VECTOR);
 416         vtpci_reset(hw);
 417         hw->started = 0;
 418         virtio_dev_free_mbufs(dev);
 419 }
 420
 421 static void
 422 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
 423 {
 424         struct virtio_hw *hw = dev->data->dev_private;
 425         struct virtio_pmd_ctrl ctrl;
 426         int dlen[1];
 427         int ret;
 428
 429         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 430                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 431                 return;
 432         }
 433
 434         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 435         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 436         ctrl.data[0] = 1;
 437         dlen[0] = 1;
 438
 439         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 440         if (ret)
 441                 PMD_INIT_LOG(ERR, "Failed to enable promisc");
 442 }
 443
 444 static void
 445 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
 446 {
 447         struct virtio_hw *hw = dev->data->dev_private;
 448         struct virtio_pmd_ctrl ctrl;
 449         int dlen[1];
 450         int ret;
 451
 452         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 453                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 454                 return;
 455         }
 456
 457         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 458         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 459         ctrl.data[0] = 0;
 460         dlen[0] = 1;
 461
 462         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 463         if (ret)
 464                 PMD_INIT_LOG(ERR, "Failed to disable promisc");
 465 }
 466
 467 static void
 468 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
 469 {
 470         struct virtio_hw *hw = dev->data->dev_private;
 471         struct virtio_pmd_ctrl ctrl;
 472         int dlen[1];
 473         int ret;
 474
 475         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 476                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 477                 return;
 478         }
 479
 480         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 481         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 482         ctrl.data[0] = 1;
 483         dlen[0] = 1;
 484
 485         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 486         if (ret)
 487                 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
 488 }
 489
 490 static void
 491 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
 492 {
 493         struct virtio_hw *hw = dev->data->dev_private;
 494         struct virtio_pmd_ctrl ctrl;
 495         int dlen[1];
 496         int ret;
 497
 498         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 499                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 500                 return;
 501         }
 502
 503         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 504         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 505         ctrl.data[0] = 0;
 506         dlen[0] = 1;
 507
 508         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 509         if (ret)
 510                 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
 511 }
 512
 513 /*
 514  * dev_ops for virtio, bare necessities for basic operation
 515  */
 516 static const struct eth_dev_ops virtio_eth_dev_ops = {
 517         .dev_configure           = virtio_dev_configure,
 518         .dev_start               = virtio_dev_start,
 519         .dev_stop                = virtio_dev_stop,
 520         .dev_close               = virtio_dev_close,
 521         .promiscuous_enable      = virtio_dev_promiscuous_enable,
 522         .promiscuous_disable     = virtio_dev_promiscuous_disable,
 523         .allmulticast_enable     = virtio_dev_allmulticast_enable,
 524         .allmulticast_disable    = virtio_dev_allmulticast_disable,
 525
 526         .dev_infos_get           = virtio_dev_info_get,
 527         .stats_get               = virtio_dev_stats_get,
 528         .stats_reset             = virtio_dev_stats_reset,
 529         .link_update             = virtio_dev_link_update,
 530         .rx_queue_setup          = virtio_dev_rx_queue_setup,
 531         /* meaningfull only to multiple queue */
 532         .rx_queue_release        = virtio_dev_rx_queue_release,
 533         .tx_queue_setup          = virtio_dev_tx_queue_setup,
 534         /* meaningfull only to multiple queue */
 535         .tx_queue_release        = virtio_dev_tx_queue_release,
 536         /* collect stats per queue */
 537         .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
 538         .vlan_filter_set         = virtio_vlan_filter_set,
 539         .mac_addr_add            = virtio_mac_addr_add,
 540         .mac_addr_remove         = virtio_mac_addr_remove,
 541         .mac_addr_set            = virtio_mac_addr_set,
 542 };
 543
 544 static inline int
 545 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
 546                                 struct rte_eth_link *link)
 547 {
 548         struct rte_eth_link *dst = link;
 549         struct rte_eth_link *src = &(dev->data->dev_link);
 550
 551         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
 552                         *(uint64_t *)src) == 0)
 553                 return -1;
 554
 555         return 0;
 556 }
 557
 558 /**
 559  * Atomically writes the link status information into global
 560  * structure rte_eth_dev.
 561  *
 562  * @param dev
 563  *   - Pointer to the structure rte_eth_dev to read from.
 564  *   - Pointer to the buffer to be saved with the link status.
 565  *
 566  * @return
 567  *   - On success, zero.
 568  *   - On failure, negative value.
 569  */
 570 static inline int
 571 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
 572                 struct rte_eth_link *link)
 573 {
 574         struct rte_eth_link *dst = &(dev->data->dev_link);
 575         struct rte_eth_link *src = link;
 576
 577         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
 578                                         *(uint64_t *)src) == 0)
 579                 return -1;
 580
 581         return 0;
 582 }
 583
 584 static void
 585 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 586 {
 587         unsigned i;
 588
 589         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 590                 const struct virtqueue *txvq = dev->data->tx_queues[i];
 591                 if (txvq == NULL)
 592                         continue;
 593
 594                 stats->opackets += txvq->packets;
 595                 stats->obytes += txvq->bytes;
 596                 stats->oerrors += txvq->errors;
 597
 598                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 599                         stats->q_opackets[i] = txvq->packets;
 600                         stats->q_obytes[i] = txvq->bytes;
 601                 }
 602         }
 603
 604         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 605                 const struct virtqueue *rxvq = dev->data->rx_queues[i];
 606                 if (rxvq == NULL)
 607                         continue;
 608
 609                 stats->ipackets += rxvq->packets;
 610                 stats->ibytes += rxvq->bytes;
 611                 stats->ierrors += rxvq->errors;
 612
 613                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 614                         stats->q_ipackets[i] = rxvq->packets;
 615                         stats->q_ibytes[i] = rxvq->bytes;
 616                 }
 617         }
 618
 619         stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
 620 }
 621
 622 static void
 623 virtio_dev_stats_reset(struct rte_eth_dev *dev)
 624 {
 625         unsigned int i;
 626
 627         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 628                 struct virtqueue *txvq = dev->data->tx_queues[i];
 629                 if (txvq == NULL)
 630                         continue;
 631
 632                 txvq->packets = 0;
 633                 txvq->bytes = 0;
 634                 txvq->errors = 0;
 635         }
 636
 637         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 638                 struct virtqueue *rxvq = dev->data->rx_queues[i];
 639                 if (rxvq == NULL)
 640                         continue;
 641
 642                 rxvq->packets = 0;
 643                 rxvq->bytes = 0;
 644                 rxvq->errors = 0;
 645         }
 646
 647         dev->data->rx_mbuf_alloc_failed = 0;
 648 }
 649
 650 static void
 651 virtio_set_hwaddr(struct virtio_hw *hw)
 652 {
 653         vtpci_write_dev_config(hw,
 654                         offsetof(struct virtio_net_config, mac),
 655                         &hw->mac_addr, ETHER_ADDR_LEN);
 656 }
 657
 658 static void
 659 virtio_get_hwaddr(struct virtio_hw *hw)
 660 {
 661         if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
 662                 vtpci_read_dev_config(hw,
 663                         offsetof(struct virtio_net_config, mac),
 664                         &hw->mac_addr, ETHER_ADDR_LEN);
 665         } else {
 666                 eth_random_addr(&hw->mac_addr[0]);
 667                 virtio_set_hwaddr(hw);
 668         }
 669 }
 670
 671 static int
 672 virtio_mac_table_set(struct virtio_hw *hw,
 673                      const struct virtio_net_ctrl_mac *uc,
 674                      const struct virtio_net_ctrl_mac *mc)
 675 {
 676         struct virtio_pmd_ctrl ctrl;
 677         int err, len[2];
 678
 679         ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
 680         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
 681
 682         len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
 683         memcpy(ctrl.data, uc, len[0]);
 684
 685         len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
 686         memcpy(ctrl.data + len[0], mc, len[1]);
 687
 688         err = virtio_send_command(hw->cvq, &ctrl, len, 2);
 689         if (err != 0)
 690                 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
 691
 692         return err;
 693 }
 694
 695 static void
 696 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
 697                     uint32_t index, uint32_t vmdq __rte_unused)
 698 {
 699         struct virtio_hw *hw = dev->data->dev_private;
 700         const struct ether_addr *addrs = dev->data->mac_addrs;
 701         unsigned int i;
 702         struct virtio_net_ctrl_mac *uc, *mc;
 703
 704         if (index >= VIRTIO_MAX_MAC_ADDRS) {
 705                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
 706                 return;
 707         }
 708
 709         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
 710         uc->entries = 0;
 711         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
 712         mc->entries = 0;
 713
 714         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
 715                 const struct ether_addr *addr
 716                         = (i == index) ? mac_addr : addrs + i;
 717                 struct virtio_net_ctrl_mac *tbl
 718                         = is_multicast_ether_addr(addr) ? mc : uc;
 719
 720                 memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
 721         }
 722
 723         virtio_mac_table_set(hw, uc, mc);
 724 }
 725
 726 static void
 727 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
 728 {
 729         struct virtio_hw *hw = dev->data->dev_private;
 730         struct ether_addr *addrs = dev->data->mac_addrs;
 731         struct virtio_net_ctrl_mac *uc, *mc;
 732         unsigned int i;
 733
 734         if (index >= VIRTIO_MAX_MAC_ADDRS) {
 735                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
 736                 return;
 737         }
 738
 739         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
 740         uc->entries = 0;
 741         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
 742         mc->entries = 0;
 743
 744         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
 745                 struct virtio_net_ctrl_mac *tbl;
 746
 747                 if (i == index || is_zero_ether_addr(addrs + i))
 748                         continue;
 749
 750                 tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
 751                 memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
 752         }
 753
 754         virtio_mac_table_set(hw, uc, mc);
 755 }
 756
 757 static void
 758 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 759 {
 760         struct virtio_hw *hw = dev->data->dev_private;
 761
 762         memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
 763
 764         /* Use atomic update if available */
 765         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
 766                 struct virtio_pmd_ctrl ctrl;
 767                 int len = ETHER_ADDR_LEN;
 768
 769                 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
 770                 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
 771
 772                 memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
 773                 virtio_send_command(hw->cvq, &ctrl, &len, 1);
 774         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
 775                 virtio_set_hwaddr(hw);
 776 }
 777
 778 static int
 779 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
 780 {
 781         struct virtio_hw *hw = dev->data->dev_private;
 782         struct virtio_pmd_ctrl ctrl;
 783         int len;
 784
 785         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
 786                 return -ENOTSUP;
 787
 788         ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
 789         ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
 790         memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
 791         len = sizeof(vlan_id);
 792
 793         return virtio_send_command(hw->cvq, &ctrl, &len, 1);
 794 }
 795
 796 static void
 797 virtio_negotiate_features(struct virtio_hw *hw)
 798 {
 799         uint32_t host_features, mask;
 800
 801         /* checksum offload not implemented */
 802         mask = VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM;
 803
 804         /* TSO and LRO are only available when their corresponding
 805          * checksum offload feature is also negotiated.
 806          */
 807         mask |= VIRTIO_NET_F_HOST_TSO4 | VIRTIO_NET_F_HOST_TSO6 | VIRTIO_NET_F_HOST_ECN;
 808         mask |= VIRTIO_NET_F_GUEST_TSO4 | VIRTIO_NET_F_GUEST_TSO6 | VIRTIO_NET_F_GUEST_ECN;
 809         mask |= VTNET_LRO_FEATURES;
 810
 811         /* not negotiating INDIRECT descriptor table support */
 812         mask |= VIRTIO_RING_F_INDIRECT_DESC;
 813
 814         /* Prepare guest_features: feature that driver wants to support */
 815         hw->guest_features = VTNET_FEATURES & ~mask;
 816         PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %x",
 817                 hw->guest_features);
 818
 819         /* Read device(host) feature bits */
 820         host_features = VIRTIO_READ_REG_4(hw, VIRTIO_PCI_HOST_FEATURES);
 821         PMD_INIT_LOG(DEBUG, "host_features before negotiate = %x",
 822                 host_features);
 823
 824         /*
 825          * Negotiate features: Subset of device feature bits are written back
 826          * guest feature bits.
 827          */
 828         hw->guest_features = vtpci_negotiate_features(hw, host_features);
 829         PMD_INIT_LOG(DEBUG, "features after negotiate = %x",
 830                 hw->guest_features);
 831 }
 832
 833 #ifdef RTE_EXEC_ENV_LINUXAPP
 834 static int
 835 parse_sysfs_value(const char *filename, unsigned long *val)
 836 {
 837         FILE *f;
 838         char buf[BUFSIZ];
 839         char *end = NULL;
 840
 841         f = fopen(filename, "r");
 842         if (f == NULL) {
 843                 PMD_INIT_LOG(ERR, "%s(): cannot open sysfs value %s",
 844                              __func__, filename);
 845                 return -1;
 846         }
 847
 848         if (fgets(buf, sizeof(buf), f) == NULL) {
 849                 PMD_INIT_LOG(ERR, "%s(): cannot read sysfs value %s",
 850                              __func__, filename);
 851                 fclose(f);
 852                 return -1;
 853         }
 854         *val = strtoul(buf, &end, 0);
 855         if ((buf[0] == '\0') || (end == NULL) || (*end != '\n')) {
 856                 PMD_INIT_LOG(ERR, "%s(): cannot parse sysfs value %s",
 857                              __func__, filename);
 858                 fclose(f);
 859                 return -1;
 860         }
 861         fclose(f);
 862         return 0;
 863 }
 864
 865 static int get_uio_dev(struct rte_pci_addr *loc, char *buf, unsigned int buflen,
 866                         unsigned int *uio_num)
 867 {
 868         struct dirent *e;
 869         DIR *dir;
 870         char dirname[PATH_MAX];
 871
 872         /* depending on kernel version, uio can be located in uio/uioX
 873          * or uio:uioX */
 874         snprintf(dirname, sizeof(dirname),
 875                      SYSFS_PCI_DEVICES "/" PCI_PRI_FMT "/uio",
 876                      loc->domain, loc->bus, loc->devid, loc->function);
 877         dir = opendir(dirname);
 878         if (dir == NULL) {
 879                 /* retry with the parent directory */
 880                 snprintf(dirname, sizeof(dirname),
 881                              SYSFS_PCI_DEVICES "/" PCI_PRI_FMT,
 882                              loc->domain, loc->bus, loc->devid, loc->function);
 883                 dir = opendir(dirname);
 884
 885                 if (dir == NULL) {
 886                         PMD_INIT_LOG(ERR, "Cannot opendir %s", dirname);
 887                         return -1;
 888                 }
 889         }
 890
 891         /* take the first file starting with "uio" */
 892         while ((e = readdir(dir)) != NULL) {
 893                 /* format could be uio%d ...*/
 894                 int shortprefix_len = sizeof("uio") - 1;
 895                 /* ... or uio:uio%d */
 896                 int longprefix_len = sizeof("uio:uio") - 1;
 897                 char *endptr;
 898
 899                 if (strncmp(e->d_name, "uio", 3) != 0)
 900                         continue;
 901
 902                 /* first try uio%d */
 903                 errno = 0;
 904                 *uio_num = strtoull(e->d_name + shortprefix_len, &endptr, 10);
 905                 if (errno == 0 && endptr != (e->d_name + shortprefix_len)) {
 906                         snprintf(buf, buflen, "%s/uio%u", dirname, *uio_num);
 907                         break;
 908                 }
 909
 910                 /* then try uio:uio%d */
 911                 errno = 0;
 912                 *uio_num = strtoull(e->d_name + longprefix_len, &endptr, 10);
 913                 if (errno == 0 && endptr != (e->d_name + longprefix_len)) {
 914                         snprintf(buf, buflen, "%s/uio:uio%u", dirname,
 915                                      *uio_num);
 916                         break;
 917                 }
 918         }
 919         closedir(dir);
 920
 921         /* No uio resource found */
 922         if (e == NULL) {
 923                 PMD_INIT_LOG(ERR, "Could not find uio resource");
 924                 return -1;
 925         }
 926
 927         return 0;
 928 }
 929
 930 static int
 931 virtio_has_msix(const struct rte_pci_addr *loc)
 932 {
 933         DIR *d;
 934         char dirname[PATH_MAX];
 935
 936         snprintf(dirname, sizeof(dirname),
 937                      SYSFS_PCI_DEVICES "/" PCI_PRI_FMT "/msi_irqs",
 938                      loc->domain, loc->bus, loc->devid, loc->function);
 939
 940         d = opendir(dirname);
 941         if (d)
 942                 closedir(d);
 943
 944         return (d != NULL);
 945 }
 946
 947 /* Extract I/O port numbers from sysfs */
 948 static int virtio_resource_init_by_uio(struct rte_pci_device *pci_dev)
 949 {
 950         char dirname[PATH_MAX];
 951         char filename[PATH_MAX];
 952         unsigned long start, size;
 953         unsigned int uio_num;
 954
 955         if (get_uio_dev(&pci_dev->addr, dirname, sizeof(dirname), &uio_num) < 0)
 956                 return -1;
 957
 958         /* get portio size */
 959         snprintf(filename, sizeof(filename),
 960                      "%s/portio/port0/size", dirname);
 961         if (parse_sysfs_value(filename, &size) < 0) {
 962                 PMD_INIT_LOG(ERR, "%s(): cannot parse size",
 963                              __func__);
 964                 return -1;
 965         }
 966
 967         /* get portio start */
 968         snprintf(filename, sizeof(filename),
 969                  "%s/portio/port0/start", dirname);
 970         if (parse_sysfs_value(filename, &start) < 0) {
 971                 PMD_INIT_LOG(ERR, "%s(): cannot parse portio start",
 972                              __func__);
 973                 return -1;
 974         }
 975         pci_dev->mem_resource[0].addr = (void *)(uintptr_t)start;
 976         pci_dev->mem_resource[0].len =  (uint64_t)size;
 977         PMD_INIT_LOG(DEBUG,
 978                      "PCI Port IO found start=0x%lx with size=0x%lx",
 979                      start, size);
 980
 981         /* save fd */
 982         memset(dirname, 0, sizeof(dirname));
 983         snprintf(dirname, sizeof(dirname), "/dev/uio%u", uio_num);
 984         pci_dev->intr_handle.fd = open(dirname, O_RDWR);
 985         if (pci_dev->intr_handle.fd < 0) {
 986                 PMD_INIT_LOG(ERR, "Cannot open %s: %s\n",
 987                         dirname, strerror(errno));
 988                 return -1;
 989         }
 990
 991         pci_dev->intr_handle.type = RTE_INTR_HANDLE_UIO;
 992         pci_dev->driver->drv_flags |= RTE_PCI_DRV_INTR_LSC;
 993
 994         return 0;
 995 }
 996
 997 /* Extract port I/O numbers from proc/ioports */
 998 static int virtio_resource_init_by_ioports(struct rte_pci_device *pci_dev)
 999 {
1000         uint16_t start, end;
1001         int size;
1002         FILE *fp;
1003         char *line = NULL;
1004         char pci_id[16];
1005         int found = 0;
1006         size_t linesz;
1007
1008         snprintf(pci_id, sizeof(pci_id), PCI_PRI_FMT,
1009                  pci_dev->addr.domain,
1010                  pci_dev->addr.bus,
1011                  pci_dev->addr.devid,
1012                  pci_dev->addr.function);
1013
1014         fp = fopen("/proc/ioports", "r");
1015         if (fp == NULL) {
1016                 PMD_INIT_LOG(ERR, "%s(): can't open ioports", __func__);
1017                 return -1;
1018         }
1019
1020         while (getdelim(&line, &linesz, '\n', fp) > 0) {
1021                 char *ptr = line;
1022                 char *left;
1023                 int n;
1024
1025                 n = strcspn(ptr, ":");
1026                 ptr[n] = 0;
1027                 left = &ptr[n+1];
1028
1029                 while (*left && isspace(*left))
1030                         left++;
1031
1032                 if (!strncmp(left, pci_id, strlen(pci_id))) {
1033                         found = 1;
1034
1035                         while (*ptr && isspace(*ptr))
1036                                 ptr++;
1037
1038                         sscanf(ptr, "%04hx-%04hx", &start, &end);
1039                         size = end - start + 1;
1040
1041                         break;
1042                 }
1043         }
1044
1045         free(line);
1046         fclose(fp);
1047
1048         if (!found)
1049                 return -1;
1050
1051         pci_dev->mem_resource[0].addr = (void *)(uintptr_t)(uint32_t)start;
1052         pci_dev->mem_resource[0].len =  (uint64_t)size;
1053         PMD_INIT_LOG(DEBUG,
1054                 "PCI Port IO found start=0x%x with size=0x%x",
1055                 start, size);
1056
1057         /* can't support lsc interrupt without uio */
1058         pci_dev->driver->drv_flags &= ~RTE_PCI_DRV_INTR_LSC;
1059
1060         return 0;
1061 }
1062
1063 /* Extract I/O port numbers from sysfs */
1064 static int virtio_resource_init(struct rte_pci_device *pci_dev)
1065 {
1066         if (virtio_resource_init_by_uio(pci_dev) == 0)
1067                 return 0;
1068         else
1069                 return virtio_resource_init_by_ioports(pci_dev);
1070 }
1071
1072 #else
1073 static int
1074 virtio_has_msix(const struct rte_pci_addr *loc __rte_unused)
1075 {
1076         /* nic_uio does not enable interrupts, return 0 (false). */
1077         return 0;
1078 }
1079
1080 static int virtio_resource_init(struct rte_pci_device *pci_dev __rte_unused)
1081 {
1082         /* no setup required */
1083         return 0;
1084 }
1085 #endif
1086
1087 /*
1088  * Process Virtio Config changed interrupt and call the callback
1089  * if link state changed.
1090  */
1091 static void
1092 virtio_interrupt_handler(__rte_unused struct rte_intr_handle *handle,
1093                          void *param)
1094 {
1095         struct rte_eth_dev *dev = param;
1096         struct virtio_hw *hw = dev->data->dev_private;
1097         uint8_t isr;
1098
1099         /* Read interrupt status which clears interrupt */
1100         isr = vtpci_isr(hw);
1101         PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1102
1103         if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0)
1104                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1105
1106         if (isr & VIRTIO_PCI_ISR_CONFIG) {
1107                 if (virtio_dev_link_update(dev, 0) == 0)
1108                         _rte_eth_dev_callback_process(dev,
1109                                                       RTE_ETH_EVENT_INTR_LSC);
1110         }
1111
1112 }
1113
1114 static void
1115 rx_func_get(struct rte_eth_dev *eth_dev)
1116 {
1117         struct virtio_hw *hw = eth_dev->data->dev_private;
1118         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF))
1119                 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1120         else
1121                 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1122 }
1123
1124 /*
1125  * This function is based on probe() function in virtio_pci.c
1126  * It returns 0 on success.
1127  */
1128 static int
1129 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1130 {
1131         struct virtio_hw *hw = eth_dev->data->dev_private;
1132         struct virtio_net_config *config;
1133         struct virtio_net_config local_config;
1134         uint32_t offset_conf = sizeof(config->mac);
1135         struct rte_pci_device *pci_dev;
1136
1137         RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr));
1138
1139         eth_dev->dev_ops = &virtio_eth_dev_ops;
1140         eth_dev->tx_pkt_burst = &virtio_xmit_pkts;
1141
1142         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1143                 rx_func_get(eth_dev);
1144                 return 0;
1145         }
1146
1147         /* Allocate memory for storing MAC addresses */
1148         eth_dev->data->mac_addrs = rte_zmalloc("virtio", ETHER_ADDR_LEN, 0);
1149         if (eth_dev->data->mac_addrs == NULL) {
1150                 PMD_INIT_LOG(ERR,
1151                         "Failed to allocate %d bytes needed to store MAC addresses",
1152                         ETHER_ADDR_LEN);
1153                 return -ENOMEM;
1154         }
1155
1156         pci_dev = eth_dev->pci_dev;
1157         if (virtio_resource_init(pci_dev) < 0)
1158                 return -1;
1159
1160         hw->use_msix = virtio_has_msix(&pci_dev->addr);
1161         hw->io_base = (uint32_t)(uintptr_t)pci_dev->mem_resource[0].addr;
1162
1163         /* Reset the device although not necessary at startup */
1164         vtpci_reset(hw);
1165
1166         /* Tell the host we've noticed this device. */
1167         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1168
1169         /* Tell the host we've known how to drive the device. */
1170         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1171         virtio_negotiate_features(hw);
1172
1173         rx_func_get(eth_dev);
1174
1175         /* Setting up rx_header size for the device */
1176         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF))
1177                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1178         else
1179                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1180
1181         /* Copy the permanent MAC address to: virtio_hw */
1182         virtio_get_hwaddr(hw);
1183         ether_addr_copy((struct ether_addr *) hw->mac_addr,
1184                         &eth_dev->data->mac_addrs[0]);
1185         PMD_INIT_LOG(DEBUG,
1186                      "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1187                      hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1188                      hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1189
1190         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1191                 config = &local_config;
1192
1193                 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1194                         offset_conf += sizeof(config->status);
1195                 } else {
1196                         PMD_INIT_LOG(DEBUG,
1197                                      "VIRTIO_NET_F_STATUS is not supported");
1198                         config->status = 0;
1199                 }
1200
1201                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1202                         offset_conf += sizeof(config->max_virtqueue_pairs);
1203                 } else {
1204                         PMD_INIT_LOG(DEBUG,
1205                                      "VIRTIO_NET_F_MQ is not supported");
1206                         config->max_virtqueue_pairs = 1;
1207                 }
1208
1209                 vtpci_read_dev_config(hw, 0, (uint8_t *)config, offset_conf);
1210
1211                 hw->max_rx_queues =
1212                         (VIRTIO_MAX_RX_QUEUES < config->max_virtqueue_pairs) ?
1213                         VIRTIO_MAX_RX_QUEUES : config->max_virtqueue_pairs;
1214                 hw->max_tx_queues =
1215                         (VIRTIO_MAX_TX_QUEUES < config->max_virtqueue_pairs) ?
1216                         VIRTIO_MAX_TX_QUEUES : config->max_virtqueue_pairs;
1217
1218                 virtio_dev_cq_queue_setup(eth_dev,
1219                                         config->max_virtqueue_pairs * 2,
1220                                         SOCKET_ID_ANY);
1221
1222                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1223                                 config->max_virtqueue_pairs);
1224                 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1225                 PMD_INIT_LOG(DEBUG,
1226                                 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1227                                 config->mac[0], config->mac[1],
1228                                 config->mac[2], config->mac[3],
1229                                 config->mac[4], config->mac[5]);
1230         } else {
1231                 hw->max_rx_queues = 1;
1232                 hw->max_tx_queues = 1;
1233         }
1234
1235         eth_dev->data->nb_rx_queues = hw->max_rx_queues;
1236         eth_dev->data->nb_tx_queues = hw->max_tx_queues;
1237
1238         PMD_INIT_LOG(DEBUG, "hw->max_rx_queues=%d   hw->max_tx_queues=%d",
1239                         hw->max_rx_queues, hw->max_tx_queues);
1240         PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1241                         eth_dev->data->port_id, pci_dev->id.vendor_id,
1242                         pci_dev->id.device_id);
1243
1244         /* Setup interrupt callback  */
1245         if (pci_dev->driver->drv_flags & RTE_PCI_DRV_INTR_LSC)
1246                 rte_intr_callback_register(&pci_dev->intr_handle,
1247                                    virtio_interrupt_handler, eth_dev);
1248
1249         virtio_dev_cq_start(eth_dev);
1250
1251         return 0;
1252 }
1253
1254 static struct eth_driver rte_virtio_pmd = {
1255         .pci_drv = {
1256                 .name = "rte_virtio_pmd",
1257                 .id_table = pci_id_virtio_map,
1258         },
1259         .eth_dev_init = eth_virtio_dev_init,
1260         .dev_private_size = sizeof(struct virtio_hw),
1261 };
1262
1263 /*
1264  * Driver initialization routine.
1265  * Invoked once at EAL init time.
1266  * Register itself as the [Poll Mode] Driver of PCI virtio devices.
1267  * Returns 0 on success.
1268  */
1269 static int
1270 rte_virtio_pmd_init(const char *name __rte_unused,
1271                     const char *param __rte_unused)
1272 {
1273         if (rte_eal_iopl_init() != 0) {
1274                 PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1275                 return -1;
1276         }
1277
1278         rte_eth_driver_register(&rte_virtio_pmd);
1279         return 0;
1280 }
1281
1282 /*
1283  * Only 1 queue is supported, no queue release related operation
1284  */
1285 static void
1286 virtio_dev_rx_queue_release(__rte_unused void *rxq)
1287 {
1288 }
1289
1290 static void
1291 virtio_dev_tx_queue_release(__rte_unused void *txq)
1292 {
1293 }
1294
1295 /*
1296  * Configure virtio device
1297  * It returns 0 on success.
1298  */
1299 static int
1300 virtio_dev_configure(struct rte_eth_dev *dev)
1301 {
1302         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1303         struct virtio_hw *hw = dev->data->dev_private;
1304         struct rte_pci_device *pci_dev = dev->pci_dev;
1305
1306         PMD_INIT_LOG(DEBUG, "configure");
1307
1308         if (rxmode->hw_ip_checksum) {
1309                 PMD_DRV_LOG(ERR, "HW IP checksum not supported");
1310                 return (-EINVAL);
1311         }
1312
1313         hw->vlan_strip = rxmode->hw_vlan_strip;
1314
1315         if (rxmode->hw_vlan_filter
1316             && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1317                 PMD_DRV_LOG(NOTICE,
1318                             "vlan filtering not available on this host");
1319                 return -ENOTSUP;
1320         }
1321
1322         if (pci_dev->driver->drv_flags & RTE_PCI_DRV_INTR_LSC)
1323                 if (vtpci_irq_config(hw, 0) == VIRTIO_MSI_NO_VECTOR) {
1324                         PMD_DRV_LOG(ERR, "failed to set config vector");
1325                         return -EBUSY;
1326                 }
1327
1328         return 0;
1329 }
1330
1331
1332 static int
1333 virtio_dev_start(struct rte_eth_dev *dev)
1334 {
1335         uint16_t nb_queues, i;
1336         struct virtio_hw *hw = dev->data->dev_private;
1337         struct rte_pci_device *pci_dev = dev->pci_dev;
1338
1339         /* check if lsc interrupt feature is enabled */
1340         if ((dev->data->dev_conf.intr_conf.lsc) &&
1341                 (pci_dev->driver->drv_flags & RTE_PCI_DRV_INTR_LSC)) {
1342                 if (!vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1343                         PMD_DRV_LOG(ERR, "link status not supported by host");
1344                         return -ENOTSUP;
1345                 }
1346
1347                 if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0) {
1348                         PMD_DRV_LOG(ERR, "interrupt enable failed");
1349                         return -EIO;
1350                 }
1351         }
1352
1353         /* Initialize Link state */
1354         virtio_dev_link_update(dev, 0);
1355
1356         /* On restart after stop do not touch queues */
1357         if (hw->started)
1358                 return 0;
1359
1360         /* Do final configuration before rx/tx engine starts */
1361         virtio_dev_rxtx_start(dev);
1362         vtpci_reinit_complete(hw);
1363
1364         hw->started = 1;
1365
1366         /*Notify the backend
1367          *Otherwise the tap backend might already stop its queue due to fullness.
1368          *vhost backend will have no chance to be waked up
1369          */
1370         nb_queues = dev->data->nb_rx_queues;
1371         if (nb_queues > 1) {
1372                 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1373                         return -EINVAL;
1374         }
1375
1376         PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1377
1378         for (i = 0; i < nb_queues; i++)
1379                 virtqueue_notify(dev->data->rx_queues[i]);
1380
1381         PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1382
1383         for (i = 0; i < dev->data->nb_rx_queues; i++)
1384                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1385
1386         for (i = 0; i < dev->data->nb_tx_queues; i++)
1387                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1388
1389         return 0;
1390 }
1391
1392 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1393 {
1394         struct rte_mbuf *buf;
1395         int i, mbuf_num = 0;
1396
1397         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1398                 PMD_INIT_LOG(DEBUG,
1399                              "Before freeing rxq[%d] used and unused buf", i);
1400                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1401
1402                 while ((buf = (struct rte_mbuf *)virtqueue_detatch_unused(
1403                                         dev->data->rx_queues[i])) != NULL) {
1404                         rte_pktmbuf_free(buf);
1405                         mbuf_num++;
1406                 }
1407
1408                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1409                 PMD_INIT_LOG(DEBUG,
1410                              "After freeing rxq[%d] used and unused buf", i);
1411                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1412         }
1413
1414         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1415                 PMD_INIT_LOG(DEBUG,
1416                              "Before freeing txq[%d] used and unused bufs",
1417                              i);
1418                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1419
1420                 mbuf_num = 0;
1421                 while ((buf = (struct rte_mbuf *)virtqueue_detatch_unused(
1422                                         dev->data->tx_queues[i])) != NULL) {
1423                         rte_pktmbuf_free(buf);
1424
1425                         mbuf_num++;
1426                 }
1427
1428                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1429                 PMD_INIT_LOG(DEBUG,
1430                              "After freeing txq[%d] used and unused buf", i);
1431                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1432         }
1433 }
1434
1435 /*
1436  * Stop device: disable interrupt and mark link down
1437  */
1438 static void
1439 virtio_dev_stop(struct rte_eth_dev *dev)
1440 {
1441         struct rte_eth_link link;
1442
1443         PMD_INIT_LOG(DEBUG, "stop");
1444
1445         if (dev->data->dev_conf.intr_conf.lsc)
1446                 rte_intr_disable(&dev->pci_dev->intr_handle);
1447
1448         memset(&link, 0, sizeof(link));
1449         virtio_dev_atomic_write_link_status(dev, &link);
1450 }
1451
1452 static int
1453 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
1454 {
1455         struct rte_eth_link link, old;
1456         uint16_t status;
1457         struct virtio_hw *hw = dev->data->dev_private;
1458         memset(&link, 0, sizeof(link));
1459         virtio_dev_atomic_read_link_status(dev, &link);
1460         old = link;
1461         link.link_duplex = FULL_DUPLEX;
1462         link.link_speed  = SPEED_10G;
1463
1464         if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1465                 PMD_INIT_LOG(DEBUG, "Get link status from hw");
1466                 vtpci_read_dev_config(hw,
1467                                 offsetof(struct virtio_net_config, status),
1468                                 &status, sizeof(status));
1469                 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
1470                         link.link_status = 0;
1471                         PMD_INIT_LOG(DEBUG, "Port %d is down",
1472                                      dev->data->port_id);
1473                 } else {
1474                         link.link_status = 1;
1475                         PMD_INIT_LOG(DEBUG, "Port %d is up",
1476                                      dev->data->port_id);
1477                 }
1478         } else {
1479                 link.link_status = 1;   /* Link up */
1480         }
1481         virtio_dev_atomic_write_link_status(dev, &link);
1482
1483         return (old.link_status == link.link_status) ? -1 : 0;
1484 }
1485
1486 static void
1487 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1488 {
1489         struct virtio_hw *hw = dev->data->dev_private;
1490
1491         dev_info->driver_name = dev->driver->pci_drv.name;
1492         dev_info->max_rx_queues = (uint16_t)hw->max_rx_queues;
1493         dev_info->max_tx_queues = (uint16_t)hw->max_tx_queues;
1494         dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
1495         dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
1496         dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
1497         dev_info->default_txconf = (struct rte_eth_txconf) {
1498                 .txq_flags = ETH_TXQ_FLAGS_NOOFFLOADS
1499         };
1500 }
1501
1502 /*
1503  * It enables testpmd to collect per queue stats.
1504  */
1505 static int
1506 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
1507 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
1508 __rte_unused uint8_t is_rx)
1509 {
1510         return 0;
1511 }
1512
1513 static struct rte_driver rte_virtio_driver = {
1514         .type = PMD_PDEV,
1515         .init = rte_virtio_pmd_init,
1516 };
1517
1518 PMD_REGISTER_DRIVER(rte_virtio_driver);