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