drivers/net/virtio/virtio_ethdev.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2015 Intel Corporation. All rights reserved.
   5  *   All rights reserved.
   6  *
   7  *   Redistribution and use in source and binary forms, with or without
   8  *   modification, are permitted provided that the following conditions
   9  *   are met:
  10  *
  11  *     * Redistributions of source code must retain the above copyright
  12  *       notice, this list of conditions and the following disclaimer.
  13  *     * Redistributions in binary form must reproduce the above copyright
  14  *       notice, this list of conditions and the following disclaimer in
  15  *       the documentation and/or other materials provided with the
  16  *       distribution.
  17  *     * Neither the name of Intel Corporation nor the names of its
  18  *       contributors may be used to endorse or promote products derived
  19  *       from this software without specific prior written permission.
  20  *
  21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  */
  33
  34 #include <stdint.h>
  35 #include <string.h>
  36 #include <stdio.h>
  37 #include <errno.h>
  38 #include <unistd.h>
  39 #ifdef RTE_EXEC_ENV_LINUXAPP
  40 #include <dirent.h>
  41 #include <fcntl.h>
  42 #endif
  43
  44 #include <rte_ethdev.h>
  45 #include <rte_memcpy.h>
  46 #include <rte_string_fns.h>
  47 #include <rte_memzone.h>
  48 #include <rte_malloc.h>
  49 #include <rte_atomic.h>
  50 #include <rte_branch_prediction.h>
  51 #include <rte_pci.h>
  52 #include <rte_ether.h>
  53 #include <rte_common.h>
  54 #include <rte_errno.h>
  55
  56 #include <rte_memory.h>
  57 #include <rte_eal.h>
  58 #include <rte_dev.h>
  59
  60 #include "virtio_ethdev.h"
  61 #include "virtio_pci.h"
  62 #include "virtio_logs.h"
  63 #include "virtqueue.h"
  64
  65
  66 static int eth_virtio_dev_init(struct rte_eth_dev *eth_dev);
  67 static int eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev);
  68 static int  virtio_dev_configure(struct rte_eth_dev *dev);
  69 static int  virtio_dev_start(struct rte_eth_dev *dev);
  70 static void virtio_dev_stop(struct rte_eth_dev *dev);
  71 static void virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
  72 static void virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
  73 static void virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
  74 static void virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
  75 static void virtio_dev_info_get(struct rte_eth_dev *dev,
  76                                 struct rte_eth_dev_info *dev_info);
  77 static int virtio_dev_link_update(struct rte_eth_dev *dev,
  78         __rte_unused int wait_to_complete);
  79
  80 static void virtio_set_hwaddr(struct virtio_hw *hw);
  81 static void virtio_get_hwaddr(struct virtio_hw *hw);
  82
  83 static void virtio_dev_rx_queue_release(__rte_unused void *rxq);
  84 static void virtio_dev_tx_queue_release(__rte_unused void *txq);
  85
  86 static void virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats);
  87 static void virtio_dev_stats_reset(struct rte_eth_dev *dev);
  88 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
  89 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
  90                                 uint16_t vlan_id, int on);
  91 static void virtio_mac_addr_add(struct rte_eth_dev *dev,
  92                                 struct ether_addr *mac_addr,
  93                                 uint32_t index, uint32_t vmdq __rte_unused);
  94 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
  95 static void virtio_mac_addr_set(struct rte_eth_dev *dev,
  96                                 struct ether_addr *mac_addr);
  97
  98 static int virtio_dev_queue_stats_mapping_set(
  99         __rte_unused struct rte_eth_dev *eth_dev,
 100         __rte_unused uint16_t queue_id,
 101         __rte_unused uint8_t stat_idx,
 102         __rte_unused uint8_t is_rx);
 103
 104 /*
 105  * The set of PCI devices this driver supports
 106  */
 107 static const struct rte_pci_id pci_id_virtio_map[] = {
 108
 109 #define RTE_PCI_DEV_ID_DECL_VIRTIO(vend, dev) {RTE_PCI_DEVICE(vend, dev)},
 110 #include "rte_pci_dev_ids.h"
 111
 112 { .vendor_id = 0, /* sentinel */ },
 113 };
 114
 115 static int
 116 virtio_send_command(struct virtqueue *vq, struct virtio_pmd_ctrl *ctrl,
 117                 int *dlen, int pkt_num)
 118 {
 119         uint32_t head, i;
 120         int k, sum = 0;
 121         virtio_net_ctrl_ack status = ~0;
 122         struct virtio_pmd_ctrl result;
 123
 124         ctrl->status = status;
 125
 126         if (!(vq && vq->hw->cvq)) {
 127                 PMD_INIT_LOG(ERR,
 128                              "%s(): Control queue is not supported.",
 129                              __func__);
 130                 return -1;
 131         }
 132         head = vq->vq_desc_head_idx;
 133
 134         PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
 135                 "vq->hw->cvq = %p vq = %p",
 136                 vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
 137
 138         if ((vq->vq_free_cnt < ((uint32_t)pkt_num + 2)) || (pkt_num < 1))
 139                 return -1;
 140
 141         memcpy(vq->virtio_net_hdr_mz->addr, ctrl,
 142                 sizeof(struct virtio_pmd_ctrl));
 143
 144         /*
 145          * Format is enforced in qemu code:
 146          * One TX packet for header;
 147          * At least one TX packet per argument;
 148          * One RX packet for ACK.
 149          */
 150         vq->vq_ring.desc[head].flags = VRING_DESC_F_NEXT;
 151         vq->vq_ring.desc[head].addr = vq->virtio_net_hdr_mz->phys_addr;
 152         vq->vq_ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
 153         vq->vq_free_cnt--;
 154         i = vq->vq_ring.desc[head].next;
 155
 156         for (k = 0; k < pkt_num; k++) {
 157                 vq->vq_ring.desc[i].flags = VRING_DESC_F_NEXT;
 158                 vq->vq_ring.desc[i].addr = vq->virtio_net_hdr_mz->phys_addr
 159                         + sizeof(struct virtio_net_ctrl_hdr)
 160                         + sizeof(ctrl->status) + sizeof(uint8_t)*sum;
 161                 vq->vq_ring.desc[i].len = dlen[k];
 162                 sum += dlen[k];
 163                 vq->vq_free_cnt--;
 164                 i = vq->vq_ring.desc[i].next;
 165         }
 166
 167         vq->vq_ring.desc[i].flags = VRING_DESC_F_WRITE;
 168         vq->vq_ring.desc[i].addr = vq->virtio_net_hdr_mz->phys_addr
 169                         + sizeof(struct virtio_net_ctrl_hdr);
 170         vq->vq_ring.desc[i].len = sizeof(ctrl->status);
 171         vq->vq_free_cnt--;
 172
 173         vq->vq_desc_head_idx = vq->vq_ring.desc[i].next;
 174
 175         vq_update_avail_ring(vq, head);
 176         vq_update_avail_idx(vq);
 177
 178         PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
 179
 180         virtqueue_notify(vq);
 181
 182         rte_rmb();
 183         while (vq->vq_used_cons_idx == vq->vq_ring.used->idx) {
 184                 rte_rmb();
 185                 usleep(100);
 186         }
 187
 188         while (vq->vq_used_cons_idx != vq->vq_ring.used->idx) {
 189                 uint32_t idx, desc_idx, used_idx;
 190                 struct vring_used_elem *uep;
 191
 192                 used_idx = (uint32_t)(vq->vq_used_cons_idx
 193                                 & (vq->vq_nentries - 1));
 194                 uep = &vq->vq_ring.used->ring[used_idx];
 195                 idx = (uint32_t) uep->id;
 196                 desc_idx = idx;
 197
 198                 while (vq->vq_ring.desc[desc_idx].flags & VRING_DESC_F_NEXT) {
 199                         desc_idx = vq->vq_ring.desc[desc_idx].next;
 200                         vq->vq_free_cnt++;
 201                 }
 202
 203                 vq->vq_ring.desc[desc_idx].next = vq->vq_desc_head_idx;
 204                 vq->vq_desc_head_idx = idx;
 205
 206                 vq->vq_used_cons_idx++;
 207                 vq->vq_free_cnt++;
 208         }
 209
 210         PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
 211                         vq->vq_free_cnt, vq->vq_desc_head_idx);
 212
 213         memcpy(&result, vq->virtio_net_hdr_mz->addr,
 214                         sizeof(struct virtio_pmd_ctrl));
 215
 216         return result.status;
 217 }
 218
 219 static int
 220 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
 221 {
 222         struct virtio_hw *hw = dev->data->dev_private;
 223         struct virtio_pmd_ctrl ctrl;
 224         int dlen[1];
 225         int ret;
 226
 227         ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
 228         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
 229         memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
 230
 231         dlen[0] = sizeof(uint16_t);
 232
 233         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 234         if (ret) {
 235                 PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
 236                           "failed, this is too late now...");
 237                 return -EINVAL;
 238         }
 239
 240         return 0;
 241 }
 242
 243 int virtio_dev_queue_setup(struct rte_eth_dev *dev,
 244                         int queue_type,
 245                         uint16_t queue_idx,
 246                         uint16_t  vtpci_queue_idx,
 247                         uint16_t nb_desc,
 248                         unsigned int socket_id,
 249                         struct virtqueue **pvq)
 250 {
 251         char vq_name[VIRTQUEUE_MAX_NAME_SZ];
 252         const struct rte_memzone *mz;
 253         uint16_t vq_size;
 254         int size;
 255         struct virtio_hw *hw = dev->data->dev_private;
 256         struct virtqueue  *vq = NULL;
 257
 258         /* Write the virtqueue index to the Queue Select Field */
 259         VIRTIO_WRITE_REG_2(hw, VIRTIO_PCI_QUEUE_SEL, vtpci_queue_idx);
 260         PMD_INIT_LOG(DEBUG, "selecting queue: %d", vtpci_queue_idx);
 261
 262         /*
 263          * Read the virtqueue size from the Queue Size field
 264          * Always power of 2 and if 0 virtqueue does not exist
 265          */
 266         vq_size = VIRTIO_READ_REG_2(hw, VIRTIO_PCI_QUEUE_NUM);
 267         PMD_INIT_LOG(DEBUG, "vq_size: %d nb_desc:%d", vq_size, nb_desc);
 268         if (nb_desc == 0)
 269                 nb_desc = vq_size;
 270         if (vq_size == 0) {
 271                 PMD_INIT_LOG(ERR, "%s: virtqueue does not exist", __func__);
 272                 return -EINVAL;
 273         }
 274
 275         if (!rte_is_power_of_2(vq_size)) {
 276                 PMD_INIT_LOG(ERR, "%s: virtqueue size is not powerof 2", __func__);
 277                 return -EINVAL;
 278         }
 279
 280         if (nb_desc < vq_size) {
 281                 if (!rte_is_power_of_2(nb_desc)) {
 282                         PMD_INIT_LOG(ERR,
 283                                      "nb_desc(%u) size is not powerof 2",
 284                                      nb_desc);
 285                         return -EINVAL;
 286                 }
 287                 vq_size = nb_desc;
 288         }
 289
 290         if (queue_type == VTNET_RQ) {
 291                 snprintf(vq_name, sizeof(vq_name), "port%d_rvq%d",
 292                         dev->data->port_id, queue_idx);
 293                 vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
 294                         vq_size * sizeof(struct vq_desc_extra), RTE_CACHE_LINE_SIZE);
 295         } else if (queue_type == VTNET_TQ) {
 296                 snprintf(vq_name, sizeof(vq_name), "port%d_tvq%d",
 297                         dev->data->port_id, queue_idx);
 298                 vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
 299                         vq_size * sizeof(struct vq_desc_extra), RTE_CACHE_LINE_SIZE);
 300         } else if (queue_type == VTNET_CQ) {
 301                 snprintf(vq_name, sizeof(vq_name), "port%d_cvq",
 302                         dev->data->port_id);
 303                 vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
 304                         vq_size * sizeof(struct vq_desc_extra),
 305                         RTE_CACHE_LINE_SIZE);
 306         }
 307         if (vq == NULL) {
 308                 PMD_INIT_LOG(ERR, "%s: Can not allocate virtqueue", __func__);
 309                 return (-ENOMEM);
 310         }
 311
 312         vq->hw = hw;
 313         vq->port_id = dev->data->port_id;
 314         vq->queue_id = queue_idx;
 315         vq->vq_queue_index = vtpci_queue_idx;
 316         vq->vq_nentries = vq_size;
 317         vq->vq_free_cnt = vq_size;
 318
 319         /*
 320          * Reserve a memzone for vring elements
 321          */
 322         size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
 323         vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
 324         PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d", size, vq->vq_ring_size);
 325
 326         mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
 327                 socket_id, 0, VIRTIO_PCI_VRING_ALIGN);
 328         if (mz == NULL) {
 329                 if (rte_errno == EEXIST)
 330                         mz = rte_memzone_lookup(vq_name);
 331                 if (mz == NULL) {
 332                         rte_free(vq);
 333                         return -ENOMEM;
 334                 }
 335         }
 336
 337         /*
 338          * Virtio PCI device VIRTIO_PCI_QUEUE_PF register is 32bit,
 339          * and only accepts 32 bit page frame number.
 340          * Check if the allocated physical memory exceeds 16TB.
 341          */
 342         if ((mz->phys_addr + vq->vq_ring_size - 1) >> (VIRTIO_PCI_QUEUE_ADDR_SHIFT + 32)) {
 343                 PMD_INIT_LOG(ERR, "vring address shouldn't be above 16TB!");
 344                 rte_free(vq);
 345                 return -ENOMEM;
 346         }
 347
 348         memset(mz->addr, 0, sizeof(mz->len));
 349         vq->mz = mz;
 350         vq->vq_ring_mem = mz->phys_addr;
 351         vq->vq_ring_virt_mem = mz->addr;
 352         PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%"PRIx64, (uint64_t)mz->phys_addr);
 353         PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%"PRIx64, (uint64_t)(uintptr_t)mz->addr);
 354         vq->virtio_net_hdr_mz  = NULL;
 355         vq->virtio_net_hdr_mem = 0;
 356
 357         if (queue_type == VTNET_TQ) {
 358                 /*
 359                  * For each xmit packet, allocate a virtio_net_hdr
 360                  */
 361                 snprintf(vq_name, sizeof(vq_name), "port%d_tvq%d_hdrzone",
 362                         dev->data->port_id, queue_idx);
 363                 vq->virtio_net_hdr_mz = rte_memzone_reserve_aligned(vq_name,
 364                         vq_size * hw->vtnet_hdr_size,
 365                         socket_id, 0, RTE_CACHE_LINE_SIZE);
 366                 if (vq->virtio_net_hdr_mz == NULL) {
 367                         if (rte_errno == EEXIST)
 368                                 vq->virtio_net_hdr_mz =
 369                                         rte_memzone_lookup(vq_name);
 370                         if (vq->virtio_net_hdr_mz == NULL) {
 371                                 rte_free(vq);
 372                                 return -ENOMEM;
 373                         }
 374                 }
 375                 vq->virtio_net_hdr_mem =
 376                         vq->virtio_net_hdr_mz->phys_addr;
 377                 memset(vq->virtio_net_hdr_mz->addr, 0,
 378                         vq_size * hw->vtnet_hdr_size);
 379         } else if (queue_type == VTNET_CQ) {
 380                 /* Allocate a page for control vq command, data and status */
 381                 snprintf(vq_name, sizeof(vq_name), "port%d_cvq_hdrzone",
 382                         dev->data->port_id);
 383                 vq->virtio_net_hdr_mz = rte_memzone_reserve_aligned(vq_name,
 384                         PAGE_SIZE, socket_id, 0, RTE_CACHE_LINE_SIZE);
 385                 if (vq->virtio_net_hdr_mz == NULL) {
 386                         if (rte_errno == EEXIST)
 387                                 vq->virtio_net_hdr_mz =
 388                                         rte_memzone_lookup(vq_name);
 389                         if (vq->virtio_net_hdr_mz == NULL) {
 390                                 rte_free(vq);
 391                                 return -ENOMEM;
 392                         }
 393                 }
 394                 vq->virtio_net_hdr_mem =
 395                         vq->virtio_net_hdr_mz->phys_addr;
 396                 memset(vq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
 397         }
 398
 399         /*
 400          * Set guest physical address of the virtqueue
 401          * in VIRTIO_PCI_QUEUE_PFN config register of device
 402          */
 403         VIRTIO_WRITE_REG_4(hw, VIRTIO_PCI_QUEUE_PFN,
 404                         mz->phys_addr >> VIRTIO_PCI_QUEUE_ADDR_SHIFT);
 405         *pvq = vq;
 406         return 0;
 407 }
 408
 409 static int
 410 virtio_dev_cq_queue_setup(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx,
 411                 uint32_t socket_id)
 412 {
 413         struct virtqueue *vq;
 414         uint16_t nb_desc = 0;
 415         int ret;
 416         struct virtio_hw *hw = dev->data->dev_private;
 417
 418         PMD_INIT_FUNC_TRACE();
 419         ret = virtio_dev_queue_setup(dev, VTNET_CQ, VTNET_SQ_CQ_QUEUE_IDX,
 420                         vtpci_queue_idx, nb_desc, socket_id, &vq);
 421         if (ret < 0) {
 422                 PMD_INIT_LOG(ERR, "control vq initialization failed");
 423                 return ret;
 424         }
 425
 426         hw->cvq = vq;
 427         return 0;
 428 }
 429
 430 static void
 431 virtio_dev_close(struct rte_eth_dev *dev)
 432 {
 433         struct virtio_hw *hw = dev->data->dev_private;
 434         struct rte_pci_device *pci_dev = dev->pci_dev;
 435
 436         PMD_INIT_LOG(DEBUG, "virtio_dev_close");
 437
 438         /* reset the NIC */
 439         if (pci_dev->driver->drv_flags & RTE_PCI_DRV_INTR_LSC)
 440                 vtpci_irq_config(hw, VIRTIO_MSI_NO_VECTOR);
 441         vtpci_reset(hw);
 442         hw->started = 0;
 443         virtio_dev_free_mbufs(dev);
 444 }
 445
 446 static void
 447 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
 448 {
 449         struct virtio_hw *hw = dev->data->dev_private;
 450         struct virtio_pmd_ctrl ctrl;
 451         int dlen[1];
 452         int ret;
 453
 454         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 455                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 456                 return;
 457         }
 458
 459         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 460         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 461         ctrl.data[0] = 1;
 462         dlen[0] = 1;
 463
 464         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 465         if (ret)
 466                 PMD_INIT_LOG(ERR, "Failed to enable promisc");
 467 }
 468
 469 static void
 470 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
 471 {
 472         struct virtio_hw *hw = dev->data->dev_private;
 473         struct virtio_pmd_ctrl ctrl;
 474         int dlen[1];
 475         int ret;
 476
 477         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 478                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 479                 return;
 480         }
 481
 482         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 483         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 484         ctrl.data[0] = 0;
 485         dlen[0] = 1;
 486
 487         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 488         if (ret)
 489                 PMD_INIT_LOG(ERR, "Failed to disable promisc");
 490 }
 491
 492 static void
 493 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
 494 {
 495         struct virtio_hw *hw = dev->data->dev_private;
 496         struct virtio_pmd_ctrl ctrl;
 497         int dlen[1];
 498         int ret;
 499
 500         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 501                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 502                 return;
 503         }
 504
 505         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 506         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 507         ctrl.data[0] = 1;
 508         dlen[0] = 1;
 509
 510         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 511         if (ret)
 512                 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
 513 }
 514
 515 static void
 516 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
 517 {
 518         struct virtio_hw *hw = dev->data->dev_private;
 519         struct virtio_pmd_ctrl ctrl;
 520         int dlen[1];
 521         int ret;
 522
 523         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 524                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 525                 return;
 526         }
 527
 528         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 529         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 530         ctrl.data[0] = 0;
 531         dlen[0] = 1;
 532
 533         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 534         if (ret)
 535                 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
 536 }
 537
 538 /*
 539  * dev_ops for virtio, bare necessities for basic operation
 540  */
 541 static const struct eth_dev_ops virtio_eth_dev_ops = {
 542         .dev_configure           = virtio_dev_configure,
 543         .dev_start               = virtio_dev_start,
 544         .dev_stop                = virtio_dev_stop,
 545         .dev_close               = virtio_dev_close,
 546         .promiscuous_enable      = virtio_dev_promiscuous_enable,
 547         .promiscuous_disable     = virtio_dev_promiscuous_disable,
 548         .allmulticast_enable     = virtio_dev_allmulticast_enable,
 549         .allmulticast_disable    = virtio_dev_allmulticast_disable,
 550
 551         .dev_infos_get           = virtio_dev_info_get,
 552         .stats_get               = virtio_dev_stats_get,
 553         .stats_reset             = virtio_dev_stats_reset,
 554         .link_update             = virtio_dev_link_update,
 555         .rx_queue_setup          = virtio_dev_rx_queue_setup,
 556         /* meaningfull only to multiple queue */
 557         .rx_queue_release        = virtio_dev_rx_queue_release,
 558         .tx_queue_setup          = virtio_dev_tx_queue_setup,
 559         /* meaningfull only to multiple queue */
 560         .tx_queue_release        = virtio_dev_tx_queue_release,
 561         /* collect stats per queue */
 562         .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
 563         .vlan_filter_set         = virtio_vlan_filter_set,
 564         .mac_addr_add            = virtio_mac_addr_add,
 565         .mac_addr_remove         = virtio_mac_addr_remove,
 566         .mac_addr_set            = virtio_mac_addr_set,
 567 };
 568
 569 static inline int
 570 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
 571                                 struct rte_eth_link *link)
 572 {
 573         struct rte_eth_link *dst = link;
 574         struct rte_eth_link *src = &(dev->data->dev_link);
 575
 576         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
 577                         *(uint64_t *)src) == 0)
 578                 return -1;
 579
 580         return 0;
 581 }
 582
 583 /**
 584  * Atomically writes the link status information into global
 585  * structure rte_eth_dev.
 586  *
 587  * @param dev
 588  *   - Pointer to the structure rte_eth_dev to read from.
 589  *   - Pointer to the buffer to be saved with the link status.
 590  *
 591  * @return
 592  *   - On success, zero.
 593  *   - On failure, negative value.
 594  */
 595 static inline int
 596 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
 597                 struct rte_eth_link *link)
 598 {
 599         struct rte_eth_link *dst = &(dev->data->dev_link);
 600         struct rte_eth_link *src = link;
 601
 602         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
 603                                         *(uint64_t *)src) == 0)
 604                 return -1;
 605
 606         return 0;
 607 }
 608
 609 static void
 610 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 611 {
 612         unsigned i;
 613
 614         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 615                 const struct virtqueue *txvq = dev->data->tx_queues[i];
 616                 if (txvq == NULL)
 617                         continue;
 618
 619                 stats->opackets += txvq->packets;
 620                 stats->obytes += txvq->bytes;
 621                 stats->oerrors += txvq->errors;
 622
 623                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 624                         stats->q_opackets[i] = txvq->packets;
 625                         stats->q_obytes[i] = txvq->bytes;
 626                 }
 627         }
 628
 629         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 630                 const struct virtqueue *rxvq = dev->data->rx_queues[i];
 631                 if (rxvq == NULL)
 632                         continue;
 633
 634                 stats->ipackets += rxvq->packets;
 635                 stats->ibytes += rxvq->bytes;
 636                 stats->ierrors += rxvq->errors;
 637
 638                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 639                         stats->q_ipackets[i] = rxvq->packets;
 640                         stats->q_ibytes[i] = rxvq->bytes;
 641                 }
 642         }
 643
 644         stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
 645 }
 646
 647 static void
 648 virtio_dev_stats_reset(struct rte_eth_dev *dev)
 649 {
 650         unsigned int i;
 651
 652         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 653                 struct virtqueue *txvq = dev->data->tx_queues[i];
 654                 if (txvq == NULL)
 655                         continue;
 656
 657                 txvq->packets = 0;
 658                 txvq->bytes = 0;
 659                 txvq->errors = 0;
 660         }
 661
 662         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 663                 struct virtqueue *rxvq = dev->data->rx_queues[i];
 664                 if (rxvq == NULL)
 665                         continue;
 666
 667                 rxvq->packets = 0;
 668                 rxvq->bytes = 0;
 669                 rxvq->errors = 0;
 670         }
 671
 672         dev->data->rx_mbuf_alloc_failed = 0;
 673 }
 674
 675 static void
 676 virtio_set_hwaddr(struct virtio_hw *hw)
 677 {
 678         vtpci_write_dev_config(hw,
 679                         offsetof(struct virtio_net_config, mac),
 680                         &hw->mac_addr, ETHER_ADDR_LEN);
 681 }
 682
 683 static void
 684 virtio_get_hwaddr(struct virtio_hw *hw)
 685 {
 686         if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
 687                 vtpci_read_dev_config(hw,
 688                         offsetof(struct virtio_net_config, mac),
 689                         &hw->mac_addr, ETHER_ADDR_LEN);
 690         } else {
 691                 eth_random_addr(&hw->mac_addr[0]);
 692                 virtio_set_hwaddr(hw);
 693         }
 694 }
 695
 696 static void
 697 virtio_mac_table_set(struct virtio_hw *hw,
 698                      const struct virtio_net_ctrl_mac *uc,
 699                      const struct virtio_net_ctrl_mac *mc)
 700 {
 701         struct virtio_pmd_ctrl ctrl;
 702         int err, len[2];
 703
 704         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
 705                 PMD_DRV_LOG(INFO, "host does not support mac table\n");
 706                 return;
 707         }
 708
 709         ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
 710         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
 711
 712         len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
 713         memcpy(ctrl.data, uc, len[0]);
 714
 715         len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
 716         memcpy(ctrl.data + len[0], mc, len[1]);
 717
 718         err = virtio_send_command(hw->cvq, &ctrl, len, 2);
 719         if (err != 0)
 720                 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
 721 }
 722
 723 static void
 724 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
 725                     uint32_t index, uint32_t vmdq __rte_unused)
 726 {
 727         struct virtio_hw *hw = dev->data->dev_private;
 728         const struct ether_addr *addrs = dev->data->mac_addrs;
 729         unsigned int i;
 730         struct virtio_net_ctrl_mac *uc, *mc;
 731
 732         if (index >= VIRTIO_MAX_MAC_ADDRS) {
 733                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
 734                 return;
 735         }
 736
 737         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
 738         uc->entries = 0;
 739         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
 740         mc->entries = 0;
 741
 742         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
 743                 const struct ether_addr *addr
 744                         = (i == index) ? mac_addr : addrs + i;
 745                 struct virtio_net_ctrl_mac *tbl
 746                         = is_multicast_ether_addr(addr) ? mc : uc;
 747
 748                 memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
 749         }
 750
 751         virtio_mac_table_set(hw, uc, mc);
 752 }
 753
 754 static void
 755 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
 756 {
 757         struct virtio_hw *hw = dev->data->dev_private;
 758         struct ether_addr *addrs = dev->data->mac_addrs;
 759         struct virtio_net_ctrl_mac *uc, *mc;
 760         unsigned int i;
 761
 762         if (index >= VIRTIO_MAX_MAC_ADDRS) {
 763                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
 764                 return;
 765         }
 766
 767         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
 768         uc->entries = 0;
 769         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
 770         mc->entries = 0;
 771
 772         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
 773                 struct virtio_net_ctrl_mac *tbl;
 774
 775                 if (i == index || is_zero_ether_addr(addrs + i))
 776                         continue;
 777
 778                 tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
 779                 memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
 780         }
 781
 782         virtio_mac_table_set(hw, uc, mc);
 783 }
 784
 785 static void
 786 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 787 {
 788         struct virtio_hw *hw = dev->data->dev_private;
 789
 790         memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
 791
 792         /* Use atomic update if available */
 793         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
 794                 struct virtio_pmd_ctrl ctrl;
 795                 int len = ETHER_ADDR_LEN;
 796
 797                 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
 798                 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
 799
 800                 memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
 801                 virtio_send_command(hw->cvq, &ctrl, &len, 1);
 802         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
 803                 virtio_set_hwaddr(hw);
 804 }
 805
 806 static int
 807 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
 808 {
 809         struct virtio_hw *hw = dev->data->dev_private;
 810         struct virtio_pmd_ctrl ctrl;
 811         int len;
 812
 813         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
 814                 return -ENOTSUP;
 815
 816         ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
 817         ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
 818         memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
 819         len = sizeof(vlan_id);
 820
 821         return virtio_send_command(hw->cvq, &ctrl, &len, 1);
 822 }
 823
 824 static void
 825 virtio_negotiate_features(struct virtio_hw *hw)
 826 {
 827         uint32_t host_features;
 828
 829         /* Prepare guest_features: feature that driver wants to support */
 830         hw->guest_features = VIRTIO_PMD_GUEST_FEATURES;
 831         PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %x",
 832                 hw->guest_features);
 833
 834         /* Read device(host) feature bits */
 835         host_features = VIRTIO_READ_REG_4(hw, VIRTIO_PCI_HOST_FEATURES);
 836         PMD_INIT_LOG(DEBUG, "host_features before negotiate = %x",
 837                 host_features);
 838
 839         /*
 840          * Negotiate features: Subset of device feature bits are written back
 841          * guest feature bits.
 842          */
 843         hw->guest_features = vtpci_negotiate_features(hw, host_features);
 844         PMD_INIT_LOG(DEBUG, "features after negotiate = %x",
 845                 hw->guest_features);
 846 }
 847
 848 #ifdef RTE_EXEC_ENV_LINUXAPP
 849 static int
 850 parse_sysfs_value(const char *filename, unsigned long *val)
 851 {
 852         FILE *f;
 853         char buf[BUFSIZ];
 854         char *end = NULL;
 855
 856         f = fopen(filename, "r");
 857         if (f == NULL) {
 858                 PMD_INIT_LOG(ERR, "%s(): cannot open sysfs value %s",
 859                              __func__, filename);
 860                 return -1;
 861         }
 862
 863         if (fgets(buf, sizeof(buf), f) == NULL) {
 864                 PMD_INIT_LOG(ERR, "%s(): cannot read sysfs value %s",
 865                              __func__, filename);
 866                 fclose(f);
 867                 return -1;
 868         }
 869         *val = strtoul(buf, &end, 0);
 870         if ((buf[0] == '\0') || (end == NULL) || (*end != '\n')) {
 871                 PMD_INIT_LOG(ERR, "%s(): cannot parse sysfs value %s",
 872                              __func__, filename);
 873                 fclose(f);
 874                 return -1;
 875         }
 876         fclose(f);
 877         return 0;
 878 }
 879
 880 static int get_uio_dev(struct rte_pci_addr *loc, char *buf, unsigned int buflen,
 881                         unsigned int *uio_num)
 882 {
 883         struct dirent *e;
 884         DIR *dir;
 885         char dirname[PATH_MAX];
 886
 887         /* depending on kernel version, uio can be located in uio/uioX
 888          * or uio:uioX */
 889         snprintf(dirname, sizeof(dirname),
 890                      SYSFS_PCI_DEVICES "/" PCI_PRI_FMT "/uio",
 891                      loc->domain, loc->bus, loc->devid, loc->function);
 892         dir = opendir(dirname);
 893         if (dir == NULL) {
 894                 /* retry with the parent directory */
 895                 snprintf(dirname, sizeof(dirname),
 896                              SYSFS_PCI_DEVICES "/" PCI_PRI_FMT,
 897                              loc->domain, loc->bus, loc->devid, loc->function);
 898                 dir = opendir(dirname);
 899
 900                 if (dir == NULL) {
 901                         PMD_INIT_LOG(ERR, "Cannot opendir %s", dirname);
 902                         return -1;
 903                 }
 904         }
 905
 906         /* take the first file starting with "uio" */
 907         while ((e = readdir(dir)) != NULL) {
 908                 /* format could be uio%d ...*/
 909                 int shortprefix_len = sizeof("uio") - 1;
 910                 /* ... or uio:uio%d */
 911                 int longprefix_len = sizeof("uio:uio") - 1;
 912                 char *endptr;
 913
 914                 if (strncmp(e->d_name, "uio", 3) != 0)
 915                         continue;
 916
 917                 /* first try uio%d */
 918                 errno = 0;
 919                 *uio_num = strtoull(e->d_name + shortprefix_len, &endptr, 10);
 920                 if (errno == 0 && endptr != (e->d_name + shortprefix_len)) {
 921                         snprintf(buf, buflen, "%s/uio%u", dirname, *uio_num);
 922                         break;
 923                 }
 924
 925                 /* then try uio:uio%d */
 926                 errno = 0;
 927                 *uio_num = strtoull(e->d_name + longprefix_len, &endptr, 10);
 928                 if (errno == 0 && endptr != (e->d_name + longprefix_len)) {
 929                         snprintf(buf, buflen, "%s/uio:uio%u", dirname,
 930                                      *uio_num);
 931                         break;
 932                 }
 933         }
 934         closedir(dir);
 935
 936         /* No uio resource found */
 937         if (e == NULL) {
 938                 PMD_INIT_LOG(ERR, "Could not find uio resource");
 939                 return -1;
 940         }
 941
 942         return 0;
 943 }
 944
 945 static int
 946 virtio_has_msix(const struct rte_pci_addr *loc)
 947 {
 948         DIR *d;
 949         char dirname[PATH_MAX];
 950
 951         snprintf(dirname, sizeof(dirname),
 952                      SYSFS_PCI_DEVICES "/" PCI_PRI_FMT "/msi_irqs",
 953                      loc->domain, loc->bus, loc->devid, loc->function);
 954
 955         d = opendir(dirname);
 956         if (d)
 957                 closedir(d);
 958
 959         return (d != NULL);
 960 }
 961
 962 /* Extract I/O port numbers from sysfs */
 963 static int virtio_resource_init_by_uio(struct rte_pci_device *pci_dev)
 964 {
 965         char dirname[PATH_MAX];
 966         char filename[PATH_MAX];
 967         unsigned long start, size;
 968         unsigned int uio_num;
 969
 970         if (get_uio_dev(&pci_dev->addr, dirname, sizeof(dirname), &uio_num) < 0)
 971                 return -1;
 972
 973         /* get portio size */
 974         snprintf(filename, sizeof(filename),
 975                      "%s/portio/port0/size", dirname);
 976         if (parse_sysfs_value(filename, &size) < 0) {
 977                 PMD_INIT_LOG(ERR, "%s(): cannot parse size",
 978                              __func__);
 979                 return -1;
 980         }
 981
 982         /* get portio start */
 983         snprintf(filename, sizeof(filename),
 984                  "%s/portio/port0/start", dirname);
 985         if (parse_sysfs_value(filename, &start) < 0) {
 986                 PMD_INIT_LOG(ERR, "%s(): cannot parse portio start",
 987                              __func__);
 988                 return -1;
 989         }
 990         pci_dev->mem_resource[0].addr = (void *)(uintptr_t)start;
 991         pci_dev->mem_resource[0].len =  (uint64_t)size;
 992         PMD_INIT_LOG(DEBUG,
 993                      "PCI Port IO found start=0x%lx with size=0x%lx",
 994                      start, size);
 995
 996         /* save fd */
 997         memset(dirname, 0, sizeof(dirname));
 998         snprintf(dirname, sizeof(dirname), "/dev/uio%u", uio_num);
 999         pci_dev->intr_handle.fd = open(dirname, O_RDWR);
1000         if (pci_dev->intr_handle.fd < 0) {
1001                 PMD_INIT_LOG(ERR, "Cannot open %s: %s\n",
1002                         dirname, strerror(errno));
1003                 return -1;
1004         }
1005
1006         pci_dev->intr_handle.type = RTE_INTR_HANDLE_UIO;
1007         pci_dev->driver->drv_flags |= RTE_PCI_DRV_INTR_LSC;
1008
1009         return 0;
1010 }
1011
1012 /* Extract port I/O numbers from proc/ioports */
1013 static int virtio_resource_init_by_ioports(struct rte_pci_device *pci_dev)
1014 {
1015         uint16_t start, end;
1016         int size;
1017         FILE *fp;
1018         char *line = NULL;
1019         char pci_id[16];
1020         int found = 0;
1021         size_t linesz;
1022
1023         snprintf(pci_id, sizeof(pci_id), PCI_PRI_FMT,
1024                  pci_dev->addr.domain,
1025                  pci_dev->addr.bus,
1026                  pci_dev->addr.devid,
1027                  pci_dev->addr.function);
1028
1029         fp = fopen("/proc/ioports", "r");
1030         if (fp == NULL) {
1031                 PMD_INIT_LOG(ERR, "%s(): can't open ioports", __func__);
1032                 return -1;
1033         }
1034
1035         while (getdelim(&line, &linesz, '\n', fp) > 0) {
1036                 char *ptr = line;
1037                 char *left;
1038                 int n;
1039
1040                 n = strcspn(ptr, ":");
1041                 ptr[n] = 0;
1042                 left = &ptr[n+1];
1043
1044                 while (*left && isspace(*left))
1045                         left++;
1046
1047                 if (!strncmp(left, pci_id, strlen(pci_id))) {
1048                         found = 1;
1049
1050                         while (*ptr && isspace(*ptr))
1051                                 ptr++;
1052
1053                         sscanf(ptr, "%04hx-%04hx", &start, &end);
1054                         size = end - start + 1;
1055
1056                         break;
1057                 }
1058         }
1059
1060         free(line);
1061         fclose(fp);
1062
1063         if (!found)
1064                 return -1;
1065
1066         pci_dev->mem_resource[0].addr = (void *)(uintptr_t)(uint32_t)start;
1067         pci_dev->mem_resource[0].len =  (uint64_t)size;
1068         PMD_INIT_LOG(DEBUG,
1069                 "PCI Port IO found start=0x%x with size=0x%x",
1070                 start, size);
1071
1072         /* can't support lsc interrupt without uio */
1073         pci_dev->driver->drv_flags &= ~RTE_PCI_DRV_INTR_LSC;
1074
1075         return 0;
1076 }
1077
1078 /* Extract I/O port numbers from sysfs */
1079 static int virtio_resource_init(struct rte_pci_device *pci_dev)
1080 {
1081         if (virtio_resource_init_by_uio(pci_dev) == 0)
1082                 return 0;
1083         else
1084                 return virtio_resource_init_by_ioports(pci_dev);
1085 }
1086
1087 #else
1088 static int
1089 virtio_has_msix(const struct rte_pci_addr *loc __rte_unused)
1090 {
1091         /* nic_uio does not enable interrupts, return 0 (false). */
1092         return 0;
1093 }
1094
1095 static int virtio_resource_init(struct rte_pci_device *pci_dev __rte_unused)
1096 {
1097         /* no setup required */
1098         return 0;
1099 }
1100 #endif
1101
1102 /*
1103  * Process Virtio Config changed interrupt and call the callback
1104  * if link state changed.
1105  */
1106 static void
1107 virtio_interrupt_handler(__rte_unused struct rte_intr_handle *handle,
1108                          void *param)
1109 {
1110         struct rte_eth_dev *dev = param;
1111         struct virtio_hw *hw = dev->data->dev_private;
1112         uint8_t isr;
1113
1114         /* Read interrupt status which clears interrupt */
1115         isr = vtpci_isr(hw);
1116         PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1117
1118         if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0)
1119                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1120
1121         if (isr & VIRTIO_PCI_ISR_CONFIG) {
1122                 if (virtio_dev_link_update(dev, 0) == 0)
1123                         _rte_eth_dev_callback_process(dev,
1124                                                       RTE_ETH_EVENT_INTR_LSC);
1125         }
1126
1127 }
1128
1129 static void
1130 rx_func_get(struct rte_eth_dev *eth_dev)
1131 {
1132         struct virtio_hw *hw = eth_dev->data->dev_private;
1133         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF))
1134                 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1135         else
1136                 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1137 }
1138
1139 /*
1140  * This function is based on probe() function in virtio_pci.c
1141  * It returns 0 on success.
1142  */
1143 static int
1144 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1145 {
1146         struct virtio_hw *hw = eth_dev->data->dev_private;
1147         struct virtio_net_config *config;
1148         struct virtio_net_config local_config;
1149         uint32_t offset_conf = sizeof(config->mac);
1150         struct rte_pci_device *pci_dev;
1151
1152         RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr));
1153
1154         eth_dev->dev_ops = &virtio_eth_dev_ops;
1155         eth_dev->tx_pkt_burst = &virtio_xmit_pkts;
1156
1157         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1158                 rx_func_get(eth_dev);
1159                 return 0;
1160         }
1161
1162         /* Allocate memory for storing MAC addresses */
1163         eth_dev->data->mac_addrs = rte_zmalloc("virtio", ETHER_ADDR_LEN, 0);
1164         if (eth_dev->data->mac_addrs == NULL) {
1165                 PMD_INIT_LOG(ERR,
1166                         "Failed to allocate %d bytes needed to store MAC addresses",
1167                         ETHER_ADDR_LEN);
1168                 return -ENOMEM;
1169         }
1170
1171         pci_dev = eth_dev->pci_dev;
1172         if (virtio_resource_init(pci_dev) < 0)
1173                 return -1;
1174
1175         hw->use_msix = virtio_has_msix(&pci_dev->addr);
1176         hw->io_base = (uint32_t)(uintptr_t)pci_dev->mem_resource[0].addr;
1177
1178         /* Reset the device although not necessary at startup */
1179         vtpci_reset(hw);
1180
1181         /* Tell the host we've noticed this device. */
1182         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1183
1184         /* Tell the host we've known how to drive the device. */
1185         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1186         virtio_negotiate_features(hw);
1187
1188         rx_func_get(eth_dev);
1189
1190         /* Setting up rx_header size for the device */
1191         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF))
1192                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1193         else
1194                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1195
1196         /* Copy the permanent MAC address to: virtio_hw */
1197         virtio_get_hwaddr(hw);
1198         ether_addr_copy((struct ether_addr *) hw->mac_addr,
1199                         &eth_dev->data->mac_addrs[0]);
1200         PMD_INIT_LOG(DEBUG,
1201                      "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1202                      hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1203                      hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1204
1205         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1206                 config = &local_config;
1207
1208                 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1209                         offset_conf += sizeof(config->status);
1210                 } else {
1211                         PMD_INIT_LOG(DEBUG,
1212                                      "VIRTIO_NET_F_STATUS is not supported");
1213                         config->status = 0;
1214                 }
1215
1216                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1217                         offset_conf += sizeof(config->max_virtqueue_pairs);
1218                 } else {
1219                         PMD_INIT_LOG(DEBUG,
1220                                      "VIRTIO_NET_F_MQ is not supported");
1221                         config->max_virtqueue_pairs = 1;
1222                 }
1223
1224                 vtpci_read_dev_config(hw, 0, (uint8_t *)config, offset_conf);
1225
1226                 hw->max_rx_queues =
1227                         (VIRTIO_MAX_RX_QUEUES < config->max_virtqueue_pairs) ?
1228                         VIRTIO_MAX_RX_QUEUES : config->max_virtqueue_pairs;
1229                 hw->max_tx_queues =
1230                         (VIRTIO_MAX_TX_QUEUES < config->max_virtqueue_pairs) ?
1231                         VIRTIO_MAX_TX_QUEUES : config->max_virtqueue_pairs;
1232
1233                 virtio_dev_cq_queue_setup(eth_dev,
1234                                         config->max_virtqueue_pairs * 2,
1235                                         SOCKET_ID_ANY);
1236
1237                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1238                                 config->max_virtqueue_pairs);
1239                 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1240                 PMD_INIT_LOG(DEBUG,
1241                                 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1242                                 config->mac[0], config->mac[1],
1243                                 config->mac[2], config->mac[3],
1244                                 config->mac[4], config->mac[5]);
1245         } else {
1246                 hw->max_rx_queues = 1;
1247                 hw->max_tx_queues = 1;
1248         }
1249
1250         eth_dev->data->nb_rx_queues = hw->max_rx_queues;
1251         eth_dev->data->nb_tx_queues = hw->max_tx_queues;
1252
1253         PMD_INIT_LOG(DEBUG, "hw->max_rx_queues=%d   hw->max_tx_queues=%d",
1254                         hw->max_rx_queues, hw->max_tx_queues);
1255         PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1256                         eth_dev->data->port_id, pci_dev->id.vendor_id,
1257                         pci_dev->id.device_id);
1258
1259         /* Setup interrupt callback  */
1260         if (pci_dev->driver->drv_flags & RTE_PCI_DRV_INTR_LSC)
1261                 rte_intr_callback_register(&pci_dev->intr_handle,
1262                                    virtio_interrupt_handler, eth_dev);
1263
1264         virtio_dev_cq_start(eth_dev);
1265
1266         return 0;
1267 }
1268
1269 static int
1270 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1271 {
1272         struct rte_pci_device *pci_dev;
1273         struct virtio_hw *hw = eth_dev->data->dev_private;
1274
1275         PMD_INIT_FUNC_TRACE();
1276
1277         if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1278                 return -EPERM;
1279
1280         if (hw->started == 1) {
1281                 virtio_dev_stop(eth_dev);
1282                 virtio_dev_close(eth_dev);
1283         }
1284         pci_dev = eth_dev->pci_dev;
1285
1286         eth_dev->dev_ops = NULL;
1287         eth_dev->tx_pkt_burst = NULL;
1288         eth_dev->rx_pkt_burst = NULL;
1289
1290         rte_free(hw->cvq);
1291         hw->cvq = NULL;
1292
1293         rte_free(eth_dev->data->mac_addrs);
1294         eth_dev->data->mac_addrs = NULL;
1295
1296         /* reset interrupt callback  */
1297         if (pci_dev->driver->drv_flags & RTE_PCI_DRV_INTR_LSC)
1298                 rte_intr_callback_unregister(&pci_dev->intr_handle,
1299                                                 virtio_interrupt_handler,
1300                                                 eth_dev);
1301
1302         PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1303
1304         return 0;
1305 }
1306
1307 static struct eth_driver rte_virtio_pmd = {
1308         .pci_drv = {
1309                 .name = "rte_virtio_pmd",
1310                 .id_table = pci_id_virtio_map,
1311                 .drv_flags = RTE_PCI_DRV_DETACHABLE,
1312         },
1313         .eth_dev_init = eth_virtio_dev_init,
1314         .eth_dev_uninit = eth_virtio_dev_uninit,
1315         .dev_private_size = sizeof(struct virtio_hw),
1316 };
1317
1318 /*
1319  * Driver initialization routine.
1320  * Invoked once at EAL init time.
1321  * Register itself as the [Poll Mode] Driver of PCI virtio devices.
1322  * Returns 0 on success.
1323  */
1324 static int
1325 rte_virtio_pmd_init(const char *name __rte_unused,
1326                     const char *param __rte_unused)
1327 {
1328         if (rte_eal_iopl_init() != 0) {
1329                 PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1330                 return -1;
1331         }
1332
1333         rte_eth_driver_register(&rte_virtio_pmd);
1334         return 0;
1335 }
1336
1337 /*
1338  * Only 1 queue is supported, no queue release related operation
1339  */
1340 static void
1341 virtio_dev_rx_queue_release(__rte_unused void *rxq)
1342 {
1343 }
1344
1345 static void
1346 virtio_dev_tx_queue_release(__rte_unused void *txq)
1347 {
1348 }
1349
1350 /*
1351  * Configure virtio device
1352  * It returns 0 on success.
1353  */
1354 static int
1355 virtio_dev_configure(struct rte_eth_dev *dev)
1356 {
1357         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1358         struct virtio_hw *hw = dev->data->dev_private;
1359         struct rte_pci_device *pci_dev = dev->pci_dev;
1360
1361         PMD_INIT_LOG(DEBUG, "configure");
1362
1363         if (rxmode->hw_ip_checksum) {
1364                 PMD_DRV_LOG(ERR, "HW IP checksum not supported");
1365                 return (-EINVAL);
1366         }
1367
1368         hw->vlan_strip = rxmode->hw_vlan_strip;
1369
1370         if (rxmode->hw_vlan_filter
1371             && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1372                 PMD_DRV_LOG(NOTICE,
1373                             "vlan filtering not available on this host");
1374                 return -ENOTSUP;
1375         }
1376
1377         if (pci_dev->driver->drv_flags & RTE_PCI_DRV_INTR_LSC)
1378                 if (vtpci_irq_config(hw, 0) == VIRTIO_MSI_NO_VECTOR) {
1379                         PMD_DRV_LOG(ERR, "failed to set config vector");
1380                         return -EBUSY;
1381                 }
1382
1383         return 0;
1384 }
1385
1386
1387 static int
1388 virtio_dev_start(struct rte_eth_dev *dev)
1389 {
1390         uint16_t nb_queues, i;
1391         struct virtio_hw *hw = dev->data->dev_private;
1392         struct rte_pci_device *pci_dev = dev->pci_dev;
1393
1394         /* check if lsc interrupt feature is enabled */
1395         if ((dev->data->dev_conf.intr_conf.lsc) &&
1396                 (pci_dev->driver->drv_flags & RTE_PCI_DRV_INTR_LSC)) {
1397                 if (!vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1398                         PMD_DRV_LOG(ERR, "link status not supported by host");
1399                         return -ENOTSUP;
1400                 }
1401
1402                 if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0) {
1403                         PMD_DRV_LOG(ERR, "interrupt enable failed");
1404                         return -EIO;
1405                 }
1406         }
1407
1408         /* Initialize Link state */
1409         virtio_dev_link_update(dev, 0);
1410
1411         /* On restart after stop do not touch queues */
1412         if (hw->started)
1413                 return 0;
1414
1415         /* Do final configuration before rx/tx engine starts */
1416         virtio_dev_rxtx_start(dev);
1417         vtpci_reinit_complete(hw);
1418
1419         hw->started = 1;
1420
1421         /*Notify the backend
1422          *Otherwise the tap backend might already stop its queue due to fullness.
1423          *vhost backend will have no chance to be waked up
1424          */
1425         nb_queues = dev->data->nb_rx_queues;
1426         if (nb_queues > 1) {
1427                 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1428                         return -EINVAL;
1429         }
1430
1431         PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1432
1433         for (i = 0; i < nb_queues; i++)
1434                 virtqueue_notify(dev->data->rx_queues[i]);
1435
1436         PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1437
1438         for (i = 0; i < dev->data->nb_rx_queues; i++)
1439                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1440
1441         for (i = 0; i < dev->data->nb_tx_queues; i++)
1442                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1443
1444         return 0;
1445 }
1446
1447 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1448 {
1449         struct rte_mbuf *buf;
1450         int i, mbuf_num = 0;
1451
1452         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1453                 PMD_INIT_LOG(DEBUG,
1454                              "Before freeing rxq[%d] used and unused buf", i);
1455                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1456
1457                 PMD_INIT_LOG(DEBUG, "rx_queues[%d]=%p",
1458                                 i, dev->data->rx_queues[i]);
1459                 while ((buf = (struct rte_mbuf *)virtqueue_detatch_unused(
1460                                         dev->data->rx_queues[i])) != NULL) {
1461                         rte_pktmbuf_free(buf);
1462                         mbuf_num++;
1463                 }
1464
1465                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1466                 PMD_INIT_LOG(DEBUG,
1467                              "After freeing rxq[%d] used and unused buf", i);
1468                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1469         }
1470
1471         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1472                 PMD_INIT_LOG(DEBUG,
1473                              "Before freeing txq[%d] used and unused bufs",
1474                              i);
1475                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1476
1477                 mbuf_num = 0;
1478                 while ((buf = (struct rte_mbuf *)virtqueue_detatch_unused(
1479                                         dev->data->tx_queues[i])) != NULL) {
1480                         rte_pktmbuf_free(buf);
1481
1482                         mbuf_num++;
1483                 }
1484
1485                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1486                 PMD_INIT_LOG(DEBUG,
1487                              "After freeing txq[%d] used and unused buf", i);
1488                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1489         }
1490 }
1491
1492 /*
1493  * Stop device: disable interrupt and mark link down
1494  */
1495 static void
1496 virtio_dev_stop(struct rte_eth_dev *dev)
1497 {
1498         struct rte_eth_link link;
1499
1500         PMD_INIT_LOG(DEBUG, "stop");
1501
1502         if (dev->data->dev_conf.intr_conf.lsc)
1503                 rte_intr_disable(&dev->pci_dev->intr_handle);
1504
1505         memset(&link, 0, sizeof(link));
1506         virtio_dev_atomic_write_link_status(dev, &link);
1507 }
1508
1509 static int
1510 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
1511 {
1512         struct rte_eth_link link, old;
1513         uint16_t status;
1514         struct virtio_hw *hw = dev->data->dev_private;
1515         memset(&link, 0, sizeof(link));
1516         virtio_dev_atomic_read_link_status(dev, &link);
1517         old = link;
1518         link.link_duplex = FULL_DUPLEX;
1519         link.link_speed  = SPEED_10G;
1520
1521         if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1522                 PMD_INIT_LOG(DEBUG, "Get link status from hw");
1523                 vtpci_read_dev_config(hw,
1524                                 offsetof(struct virtio_net_config, status),
1525                                 &status, sizeof(status));
1526                 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
1527                         link.link_status = 0;
1528                         PMD_INIT_LOG(DEBUG, "Port %d is down",
1529                                      dev->data->port_id);
1530                 } else {
1531                         link.link_status = 1;
1532                         PMD_INIT_LOG(DEBUG, "Port %d is up",
1533                                      dev->data->port_id);
1534                 }
1535         } else {
1536                 link.link_status = 1;   /* Link up */
1537         }
1538         virtio_dev_atomic_write_link_status(dev, &link);
1539
1540         return (old.link_status == link.link_status) ? -1 : 0;
1541 }
1542
1543 static void
1544 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1545 {
1546         struct virtio_hw *hw = dev->data->dev_private;
1547
1548         dev_info->driver_name = dev->driver->pci_drv.name;
1549         dev_info->max_rx_queues = (uint16_t)hw->max_rx_queues;
1550         dev_info->max_tx_queues = (uint16_t)hw->max_tx_queues;
1551         dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
1552         dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
1553         dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
1554         dev_info->default_txconf = (struct rte_eth_txconf) {
1555                 .txq_flags = ETH_TXQ_FLAGS_NOOFFLOADS
1556         };
1557 }
1558
1559 /*
1560  * It enables testpmd to collect per queue stats.
1561  */
1562 static int
1563 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
1564 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
1565 __rte_unused uint8_t is_rx)
1566 {
1567         return 0;
1568 }
1569
1570 static struct rte_driver rte_virtio_driver = {
1571         .type = PMD_PDEV,
1572         .init = rte_virtio_pmd_init,
1573 };
1574
1575 PMD_REGISTER_DRIVER(rte_virtio_driver);