drivers/net/virtio/virtio_ethdev.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2015 Intel Corporation. All rights reserved.
   5  *   All rights reserved.
   6  *
   7  *   Redistribution and use in source and binary forms, with or without
   8  *   modification, are permitted provided that the following conditions
   9  *   are met:
  10  *
  11  *     * Redistributions of source code must retain the above copyright
  12  *       notice, this list of conditions and the following disclaimer.
  13  *     * Redistributions in binary form must reproduce the above copyright
  14  *       notice, this list of conditions and the following disclaimer in
  15  *       the documentation and/or other materials provided with the
  16  *       distribution.
  17  *     * Neither the name of Intel Corporation nor the names of its
  18  *       contributors may be used to endorse or promote products derived
  19  *       from this software without specific prior written permission.
  20  *
  21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  */
  33
  34 #include <stdint.h>
  35 #include <string.h>
  36 #include <stdio.h>
  37 #include <errno.h>
  38 #include <unistd.h>
  39 #ifdef RTE_EXEC_ENV_LINUXAPP
  40 #include <dirent.h>
  41 #include <fcntl.h>
  42 #endif
  43
  44 #include <rte_ethdev.h>
  45 #include <rte_memcpy.h>
  46 #include <rte_string_fns.h>
  47 #include <rte_memzone.h>
  48 #include <rte_malloc.h>
  49 #include <rte_atomic.h>
  50 #include <rte_branch_prediction.h>
  51 #include <rte_pci.h>
  52 #include <rte_ether.h>
  53 #include <rte_common.h>
  54 #include <rte_errno.h>
  55
  56 #include <rte_memory.h>
  57 #include <rte_eal.h>
  58 #include <rte_dev.h>
  59
  60 #include "virtio_ethdev.h"
  61 #include "virtio_pci.h"
  62 #include "virtio_logs.h"
  63 #include "virtqueue.h"
  64
  65
  66 static int eth_virtio_dev_init(struct rte_eth_dev *eth_dev);
  67 static int eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev);
  68 static int  virtio_dev_configure(struct rte_eth_dev *dev);
  69 static int  virtio_dev_start(struct rte_eth_dev *dev);
  70 static void virtio_dev_stop(struct rte_eth_dev *dev);
  71 static void virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
  72 static void virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
  73 static void virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
  74 static void virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
  75 static void virtio_dev_info_get(struct rte_eth_dev *dev,
  76                                 struct rte_eth_dev_info *dev_info);
  77 static int virtio_dev_link_update(struct rte_eth_dev *dev,
  78         __rte_unused int wait_to_complete);
  79
  80 static void virtio_set_hwaddr(struct virtio_hw *hw);
  81 static void virtio_get_hwaddr(struct virtio_hw *hw);
  82
  83 static void virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats);
  84 static void virtio_dev_stats_reset(struct rte_eth_dev *dev);
  85 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
  86 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
  87                                 uint16_t vlan_id, int on);
  88 static void virtio_mac_addr_add(struct rte_eth_dev *dev,
  89                                 struct ether_addr *mac_addr,
  90                                 uint32_t index, uint32_t vmdq __rte_unused);
  91 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
  92 static void virtio_mac_addr_set(struct rte_eth_dev *dev,
  93                                 struct ether_addr *mac_addr);
  94
  95 static int virtio_dev_queue_stats_mapping_set(
  96         __rte_unused struct rte_eth_dev *eth_dev,
  97         __rte_unused uint16_t queue_id,
  98         __rte_unused uint8_t stat_idx,
  99         __rte_unused uint8_t is_rx);
 100
 101 /*
 102  * The set of PCI devices this driver supports
 103  */
 104 static const struct rte_pci_id pci_id_virtio_map[] = {
 105
 106 #define RTE_PCI_DEV_ID_DECL_VIRTIO(vend, dev) {RTE_PCI_DEVICE(vend, dev)},
 107 #include "rte_pci_dev_ids.h"
 108
 109 { .vendor_id = 0, /* sentinel */ },
 110 };
 111
 112 static int
 113 virtio_send_command(struct virtqueue *vq, struct virtio_pmd_ctrl *ctrl,
 114                 int *dlen, int pkt_num)
 115 {
 116         uint32_t head, i;
 117         int k, sum = 0;
 118         virtio_net_ctrl_ack status = ~0;
 119         struct virtio_pmd_ctrl result;
 120
 121         ctrl->status = status;
 122
 123         if (!(vq && vq->hw->cvq)) {
 124                 PMD_INIT_LOG(ERR,
 125                              "%s(): Control queue is not supported.",
 126                              __func__);
 127                 return -1;
 128         }
 129         head = vq->vq_desc_head_idx;
 130
 131         PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
 132                 "vq->hw->cvq = %p vq = %p",
 133                 vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
 134
 135         if ((vq->vq_free_cnt < ((uint32_t)pkt_num + 2)) || (pkt_num < 1))
 136                 return -1;
 137
 138         memcpy(vq->virtio_net_hdr_mz->addr, ctrl,
 139                 sizeof(struct virtio_pmd_ctrl));
 140
 141         /*
 142          * Format is enforced in qemu code:
 143          * One TX packet for header;
 144          * At least one TX packet per argument;
 145          * One RX packet for ACK.
 146          */
 147         vq->vq_ring.desc[head].flags = VRING_DESC_F_NEXT;
 148         vq->vq_ring.desc[head].addr = vq->virtio_net_hdr_mz->phys_addr;
 149         vq->vq_ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
 150         vq->vq_free_cnt--;
 151         i = vq->vq_ring.desc[head].next;
 152
 153         for (k = 0; k < pkt_num; k++) {
 154                 vq->vq_ring.desc[i].flags = VRING_DESC_F_NEXT;
 155                 vq->vq_ring.desc[i].addr = vq->virtio_net_hdr_mz->phys_addr
 156                         + sizeof(struct virtio_net_ctrl_hdr)
 157                         + sizeof(ctrl->status) + sizeof(uint8_t)*sum;
 158                 vq->vq_ring.desc[i].len = dlen[k];
 159                 sum += dlen[k];
 160                 vq->vq_free_cnt--;
 161                 i = vq->vq_ring.desc[i].next;
 162         }
 163
 164         vq->vq_ring.desc[i].flags = VRING_DESC_F_WRITE;
 165         vq->vq_ring.desc[i].addr = vq->virtio_net_hdr_mz->phys_addr
 166                         + sizeof(struct virtio_net_ctrl_hdr);
 167         vq->vq_ring.desc[i].len = sizeof(ctrl->status);
 168         vq->vq_free_cnt--;
 169
 170         vq->vq_desc_head_idx = vq->vq_ring.desc[i].next;
 171
 172         vq_update_avail_ring(vq, head);
 173         vq_update_avail_idx(vq);
 174
 175         PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
 176
 177         virtqueue_notify(vq);
 178
 179         rte_rmb();
 180         while (vq->vq_used_cons_idx == vq->vq_ring.used->idx) {
 181                 rte_rmb();
 182                 usleep(100);
 183         }
 184
 185         while (vq->vq_used_cons_idx != vq->vq_ring.used->idx) {
 186                 uint32_t idx, desc_idx, used_idx;
 187                 struct vring_used_elem *uep;
 188
 189                 used_idx = (uint32_t)(vq->vq_used_cons_idx
 190                                 & (vq->vq_nentries - 1));
 191                 uep = &vq->vq_ring.used->ring[used_idx];
 192                 idx = (uint32_t) uep->id;
 193                 desc_idx = idx;
 194
 195                 while (vq->vq_ring.desc[desc_idx].flags & VRING_DESC_F_NEXT) {
 196                         desc_idx = vq->vq_ring.desc[desc_idx].next;
 197                         vq->vq_free_cnt++;
 198                 }
 199
 200                 vq->vq_ring.desc[desc_idx].next = vq->vq_desc_head_idx;
 201                 vq->vq_desc_head_idx = idx;
 202
 203                 vq->vq_used_cons_idx++;
 204                 vq->vq_free_cnt++;
 205         }
 206
 207         PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
 208                         vq->vq_free_cnt, vq->vq_desc_head_idx);
 209
 210         memcpy(&result, vq->virtio_net_hdr_mz->addr,
 211                         sizeof(struct virtio_pmd_ctrl));
 212
 213         return result.status;
 214 }
 215
 216 static int
 217 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
 218 {
 219         struct virtio_hw *hw = dev->data->dev_private;
 220         struct virtio_pmd_ctrl ctrl;
 221         int dlen[1];
 222         int ret;
 223
 224         ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
 225         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
 226         memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
 227
 228         dlen[0] = sizeof(uint16_t);
 229
 230         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 231         if (ret) {
 232                 PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
 233                           "failed, this is too late now...");
 234                 return -EINVAL;
 235         }
 236
 237         return 0;
 238 }
 239
 240 void
 241 virtio_dev_queue_release(struct virtqueue *vq) {
 242         struct virtio_hw *hw;
 243
 244         if (vq) {
 245                 hw = vq->hw;
 246                 /* Select and deactivate the queue */
 247                 VIRTIO_WRITE_REG_2(hw, VIRTIO_PCI_QUEUE_SEL, vq->queue_id);
 248                 VIRTIO_WRITE_REG_4(hw, VIRTIO_PCI_QUEUE_PFN, 0);
 249
 250                 rte_free(vq);
 251                 vq = NULL;
 252         }
 253 }
 254
 255 int virtio_dev_queue_setup(struct rte_eth_dev *dev,
 256                         int queue_type,
 257                         uint16_t queue_idx,
 258                         uint16_t vtpci_queue_idx,
 259                         uint16_t nb_desc,
 260                         unsigned int socket_id,
 261                         struct virtqueue **pvq)
 262 {
 263         char vq_name[VIRTQUEUE_MAX_NAME_SZ];
 264         const struct rte_memzone *mz;
 265         uint16_t vq_size;
 266         int size;
 267         struct virtio_hw *hw = dev->data->dev_private;
 268         struct virtqueue *vq = NULL;
 269
 270         /* Write the virtqueue index to the Queue Select Field */
 271         VIRTIO_WRITE_REG_2(hw, VIRTIO_PCI_QUEUE_SEL, vtpci_queue_idx);
 272         PMD_INIT_LOG(DEBUG, "selecting queue: %d", vtpci_queue_idx);
 273
 274         /*
 275          * Read the virtqueue size from the Queue Size field
 276          * Always power of 2 and if 0 virtqueue does not exist
 277          */
 278         vq_size = VIRTIO_READ_REG_2(hw, VIRTIO_PCI_QUEUE_NUM);
 279         PMD_INIT_LOG(DEBUG, "vq_size: %d nb_desc:%d", vq_size, nb_desc);
 280         if (vq_size == 0) {
 281                 PMD_INIT_LOG(ERR, "%s: virtqueue does not exist", __func__);
 282                 return -EINVAL;
 283         }
 284
 285         if (!rte_is_power_of_2(vq_size)) {
 286                 PMD_INIT_LOG(ERR, "%s: virtqueue size is not powerof 2", __func__);
 287                 return -EINVAL;
 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
 318         if (nb_desc == 0 || nb_desc > vq_size)
 319                 nb_desc = vq_size;
 320         vq->vq_free_cnt = nb_desc;
 321
 322         /*
 323          * Reserve a memzone for vring elements
 324          */
 325         size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
 326         vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
 327         PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d", size, vq->vq_ring_size);
 328
 329         mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
 330                 socket_id, 0, VIRTIO_PCI_VRING_ALIGN);
 331         if (mz == NULL) {
 332                 if (rte_errno == EEXIST)
 333                         mz = rte_memzone_lookup(vq_name);
 334                 if (mz == NULL) {
 335                         rte_free(vq);
 336                         return -ENOMEM;
 337                 }
 338         }
 339
 340         /*
 341          * Virtio PCI device VIRTIO_PCI_QUEUE_PF register is 32bit,
 342          * and only accepts 32 bit page frame number.
 343          * Check if the allocated physical memory exceeds 16TB.
 344          */
 345         if ((mz->phys_addr + vq->vq_ring_size - 1) >> (VIRTIO_PCI_QUEUE_ADDR_SHIFT + 32)) {
 346                 PMD_INIT_LOG(ERR, "vring address shouldn't be above 16TB!");
 347                 rte_free(vq);
 348                 return -ENOMEM;
 349         }
 350
 351         memset(mz->addr, 0, sizeof(mz->len));
 352         vq->mz = mz;
 353         vq->vq_ring_mem = mz->phys_addr;
 354         vq->vq_ring_virt_mem = mz->addr;
 355         PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%"PRIx64, (uint64_t)mz->phys_addr);
 356         PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%"PRIx64, (uint64_t)(uintptr_t)mz->addr);
 357         vq->virtio_net_hdr_mz  = NULL;
 358         vq->virtio_net_hdr_mem = 0;
 359
 360         if (queue_type == VTNET_TQ) {
 361                 /*
 362                  * For each xmit packet, allocate a virtio_net_hdr
 363                  */
 364                 snprintf(vq_name, sizeof(vq_name), "port%d_tvq%d_hdrzone",
 365                         dev->data->port_id, queue_idx);
 366                 vq->virtio_net_hdr_mz = rte_memzone_reserve_aligned(vq_name,
 367                         vq_size * hw->vtnet_hdr_size,
 368                         socket_id, 0, RTE_CACHE_LINE_SIZE);
 369                 if (vq->virtio_net_hdr_mz == NULL) {
 370                         if (rte_errno == EEXIST)
 371                                 vq->virtio_net_hdr_mz =
 372                                         rte_memzone_lookup(vq_name);
 373                         if (vq->virtio_net_hdr_mz == NULL) {
 374                                 rte_free(vq);
 375                                 return -ENOMEM;
 376                         }
 377                 }
 378                 vq->virtio_net_hdr_mem =
 379                         vq->virtio_net_hdr_mz->phys_addr;
 380                 memset(vq->virtio_net_hdr_mz->addr, 0,
 381                         vq_size * hw->vtnet_hdr_size);
 382         } else if (queue_type == VTNET_CQ) {
 383                 /* Allocate a page for control vq command, data and status */
 384                 snprintf(vq_name, sizeof(vq_name), "port%d_cvq_hdrzone",
 385                         dev->data->port_id);
 386                 vq->virtio_net_hdr_mz = rte_memzone_reserve_aligned(vq_name,
 387                         PAGE_SIZE, socket_id, 0, RTE_CACHE_LINE_SIZE);
 388                 if (vq->virtio_net_hdr_mz == NULL) {
 389                         if (rte_errno == EEXIST)
 390                                 vq->virtio_net_hdr_mz =
 391                                         rte_memzone_lookup(vq_name);
 392                         if (vq->virtio_net_hdr_mz == NULL) {
 393                                 rte_free(vq);
 394                                 return -ENOMEM;
 395                         }
 396                 }
 397                 vq->virtio_net_hdr_mem =
 398                         vq->virtio_net_hdr_mz->phys_addr;
 399                 memset(vq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
 400         }
 401
 402         /*
 403          * Set guest physical address of the virtqueue
 404          * in VIRTIO_PCI_QUEUE_PFN config register of device
 405          */
 406         VIRTIO_WRITE_REG_4(hw, VIRTIO_PCI_QUEUE_PFN,
 407                         mz->phys_addr >> VIRTIO_PCI_QUEUE_ADDR_SHIFT);
 408         *pvq = vq;
 409         return 0;
 410 }
 411
 412 static int
 413 virtio_dev_cq_queue_setup(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx,
 414                 uint32_t socket_id)
 415 {
 416         struct virtqueue *vq;
 417         int ret;
 418         struct virtio_hw *hw = dev->data->dev_private;
 419
 420         PMD_INIT_FUNC_TRACE();
 421         ret = virtio_dev_queue_setup(dev, VTNET_CQ, VTNET_SQ_CQ_QUEUE_IDX,
 422                         vtpci_queue_idx, 0, socket_id, &vq);
 423         if (ret < 0) {
 424                 PMD_INIT_LOG(ERR, "control vq initialization failed");
 425                 return ret;
 426         }
 427
 428         hw->cvq = vq;
 429         return 0;
 430 }
 431
 432 static void
 433 virtio_free_queues(struct rte_eth_dev *dev)
 434 {
 435         unsigned int i;
 436
 437         for (i = 0; i < dev->data->nb_rx_queues; i++)
 438                 virtio_dev_rx_queue_release(dev->data->rx_queues[i]);
 439
 440         dev->data->nb_rx_queues = 0;
 441
 442         for (i = 0; i < dev->data->nb_tx_queues; i++)
 443                 virtio_dev_tx_queue_release(dev->data->tx_queues[i]);
 444
 445         dev->data->nb_tx_queues = 0;
 446 }
 447
 448 static void
 449 virtio_dev_close(struct rte_eth_dev *dev)
 450 {
 451         struct virtio_hw *hw = dev->data->dev_private;
 452         struct rte_pci_device *pci_dev = dev->pci_dev;
 453
 454         PMD_INIT_LOG(DEBUG, "virtio_dev_close");
 455
 456         /* reset the NIC */
 457         if (pci_dev->driver->drv_flags & RTE_PCI_DRV_INTR_LSC)
 458                 vtpci_irq_config(hw, VIRTIO_MSI_NO_VECTOR);
 459         vtpci_reset(hw);
 460         hw->started = 0;
 461         virtio_dev_free_mbufs(dev);
 462         virtio_free_queues(dev);
 463 }
 464
 465 static void
 466 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
 467 {
 468         struct virtio_hw *hw = dev->data->dev_private;
 469         struct virtio_pmd_ctrl ctrl;
 470         int dlen[1];
 471         int ret;
 472
 473         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 474                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 475                 return;
 476         }
 477
 478         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 479         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 480         ctrl.data[0] = 1;
 481         dlen[0] = 1;
 482
 483         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 484         if (ret)
 485                 PMD_INIT_LOG(ERR, "Failed to enable promisc");
 486 }
 487
 488 static void
 489 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
 490 {
 491         struct virtio_hw *hw = dev->data->dev_private;
 492         struct virtio_pmd_ctrl ctrl;
 493         int dlen[1];
 494         int ret;
 495
 496         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 497                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 498                 return;
 499         }
 500
 501         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 502         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 503         ctrl.data[0] = 0;
 504         dlen[0] = 1;
 505
 506         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 507         if (ret)
 508                 PMD_INIT_LOG(ERR, "Failed to disable promisc");
 509 }
 510
 511 static void
 512 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
 513 {
 514         struct virtio_hw *hw = dev->data->dev_private;
 515         struct virtio_pmd_ctrl ctrl;
 516         int dlen[1];
 517         int ret;
 518
 519         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 520                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 521                 return;
 522         }
 523
 524         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 525         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 526         ctrl.data[0] = 1;
 527         dlen[0] = 1;
 528
 529         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 530         if (ret)
 531                 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
 532 }
 533
 534 static void
 535 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
 536 {
 537         struct virtio_hw *hw = dev->data->dev_private;
 538         struct virtio_pmd_ctrl ctrl;
 539         int dlen[1];
 540         int ret;
 541
 542         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 543                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 544                 return;
 545         }
 546
 547         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 548         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 549         ctrl.data[0] = 0;
 550         dlen[0] = 1;
 551
 552         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 553         if (ret)
 554                 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
 555 }
 556
 557 /*
 558  * dev_ops for virtio, bare necessities for basic operation
 559  */
 560 static const struct eth_dev_ops virtio_eth_dev_ops = {
 561         .dev_configure           = virtio_dev_configure,
 562         .dev_start               = virtio_dev_start,
 563         .dev_stop                = virtio_dev_stop,
 564         .dev_close               = virtio_dev_close,
 565         .promiscuous_enable      = virtio_dev_promiscuous_enable,
 566         .promiscuous_disable     = virtio_dev_promiscuous_disable,
 567         .allmulticast_enable     = virtio_dev_allmulticast_enable,
 568         .allmulticast_disable    = virtio_dev_allmulticast_disable,
 569
 570         .dev_infos_get           = virtio_dev_info_get,
 571         .stats_get               = virtio_dev_stats_get,
 572         .stats_reset             = virtio_dev_stats_reset,
 573         .link_update             = virtio_dev_link_update,
 574         .rx_queue_setup          = virtio_dev_rx_queue_setup,
 575         .rx_queue_release        = virtio_dev_rx_queue_release,
 576         .tx_queue_setup          = virtio_dev_tx_queue_setup,
 577         .tx_queue_release        = virtio_dev_tx_queue_release,
 578         /* collect stats per queue */
 579         .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
 580         .vlan_filter_set         = virtio_vlan_filter_set,
 581         .mac_addr_add            = virtio_mac_addr_add,
 582         .mac_addr_remove         = virtio_mac_addr_remove,
 583         .mac_addr_set            = virtio_mac_addr_set,
 584 };
 585
 586 static inline int
 587 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
 588                                 struct rte_eth_link *link)
 589 {
 590         struct rte_eth_link *dst = link;
 591         struct rte_eth_link *src = &(dev->data->dev_link);
 592
 593         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
 594                         *(uint64_t *)src) == 0)
 595                 return -1;
 596
 597         return 0;
 598 }
 599
 600 /**
 601  * Atomically writes the link status information into global
 602  * structure rte_eth_dev.
 603  *
 604  * @param dev
 605  *   - Pointer to the structure rte_eth_dev to read from.
 606  *   - Pointer to the buffer to be saved with the link status.
 607  *
 608  * @return
 609  *   - On success, zero.
 610  *   - On failure, negative value.
 611  */
 612 static inline int
 613 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
 614                 struct rte_eth_link *link)
 615 {
 616         struct rte_eth_link *dst = &(dev->data->dev_link);
 617         struct rte_eth_link *src = link;
 618
 619         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
 620                                         *(uint64_t *)src) == 0)
 621                 return -1;
 622
 623         return 0;
 624 }
 625
 626 static void
 627 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 628 {
 629         unsigned i;
 630
 631         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 632                 const struct virtqueue *txvq = dev->data->tx_queues[i];
 633                 if (txvq == NULL)
 634                         continue;
 635
 636                 stats->opackets += txvq->packets;
 637                 stats->obytes += txvq->bytes;
 638                 stats->oerrors += txvq->errors;
 639
 640                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 641                         stats->q_opackets[i] = txvq->packets;
 642                         stats->q_obytes[i] = txvq->bytes;
 643                 }
 644         }
 645
 646         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 647                 const struct virtqueue *rxvq = dev->data->rx_queues[i];
 648                 if (rxvq == NULL)
 649                         continue;
 650
 651                 stats->ipackets += rxvq->packets;
 652                 stats->ibytes += rxvq->bytes;
 653                 stats->ierrors += rxvq->errors;
 654
 655                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 656                         stats->q_ipackets[i] = rxvq->packets;
 657                         stats->q_ibytes[i] = rxvq->bytes;
 658                 }
 659         }
 660
 661         stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
 662 }
 663
 664 static void
 665 virtio_dev_stats_reset(struct rte_eth_dev *dev)
 666 {
 667         unsigned int i;
 668
 669         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 670                 struct virtqueue *txvq = dev->data->tx_queues[i];
 671                 if (txvq == NULL)
 672                         continue;
 673
 674                 txvq->packets = 0;
 675                 txvq->bytes = 0;
 676                 txvq->errors = 0;
 677         }
 678
 679         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 680                 struct virtqueue *rxvq = dev->data->rx_queues[i];
 681                 if (rxvq == NULL)
 682                         continue;
 683
 684                 rxvq->packets = 0;
 685                 rxvq->bytes = 0;
 686                 rxvq->errors = 0;
 687         }
 688
 689         dev->data->rx_mbuf_alloc_failed = 0;
 690 }
 691
 692 static void
 693 virtio_set_hwaddr(struct virtio_hw *hw)
 694 {
 695         vtpci_write_dev_config(hw,
 696                         offsetof(struct virtio_net_config, mac),
 697                         &hw->mac_addr, ETHER_ADDR_LEN);
 698 }
 699
 700 static void
 701 virtio_get_hwaddr(struct virtio_hw *hw)
 702 {
 703         if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
 704                 vtpci_read_dev_config(hw,
 705                         offsetof(struct virtio_net_config, mac),
 706                         &hw->mac_addr, ETHER_ADDR_LEN);
 707         } else {
 708                 eth_random_addr(&hw->mac_addr[0]);
 709                 virtio_set_hwaddr(hw);
 710         }
 711 }
 712
 713 static void
 714 virtio_mac_table_set(struct virtio_hw *hw,
 715                      const struct virtio_net_ctrl_mac *uc,
 716                      const struct virtio_net_ctrl_mac *mc)
 717 {
 718         struct virtio_pmd_ctrl ctrl;
 719         int err, len[2];
 720
 721         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
 722                 PMD_DRV_LOG(INFO, "host does not support mac table\n");
 723                 return;
 724         }
 725
 726         ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
 727         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
 728
 729         len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
 730         memcpy(ctrl.data, uc, len[0]);
 731
 732         len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
 733         memcpy(ctrl.data + len[0], mc, len[1]);
 734
 735         err = virtio_send_command(hw->cvq, &ctrl, len, 2);
 736         if (err != 0)
 737                 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
 738 }
 739
 740 static void
 741 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
 742                     uint32_t index, uint32_t vmdq __rte_unused)
 743 {
 744         struct virtio_hw *hw = dev->data->dev_private;
 745         const struct ether_addr *addrs = dev->data->mac_addrs;
 746         unsigned int i;
 747         struct virtio_net_ctrl_mac *uc, *mc;
 748
 749         if (index >= VIRTIO_MAX_MAC_ADDRS) {
 750                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
 751                 return;
 752         }
 753
 754         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
 755         uc->entries = 0;
 756         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
 757         mc->entries = 0;
 758
 759         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
 760                 const struct ether_addr *addr
 761                         = (i == index) ? mac_addr : addrs + i;
 762                 struct virtio_net_ctrl_mac *tbl
 763                         = is_multicast_ether_addr(addr) ? mc : uc;
 764
 765                 memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
 766         }
 767
 768         virtio_mac_table_set(hw, uc, mc);
 769 }
 770
 771 static void
 772 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
 773 {
 774         struct virtio_hw *hw = dev->data->dev_private;
 775         struct ether_addr *addrs = dev->data->mac_addrs;
 776         struct virtio_net_ctrl_mac *uc, *mc;
 777         unsigned int i;
 778
 779         if (index >= VIRTIO_MAX_MAC_ADDRS) {
 780                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
 781                 return;
 782         }
 783
 784         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
 785         uc->entries = 0;
 786         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
 787         mc->entries = 0;
 788
 789         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
 790                 struct virtio_net_ctrl_mac *tbl;
 791
 792                 if (i == index || is_zero_ether_addr(addrs + i))
 793                         continue;
 794
 795                 tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
 796                 memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
 797         }
 798
 799         virtio_mac_table_set(hw, uc, mc);
 800 }
 801
 802 static void
 803 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 804 {
 805         struct virtio_hw *hw = dev->data->dev_private;
 806
 807         memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
 808
 809         /* Use atomic update if available */
 810         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
 811                 struct virtio_pmd_ctrl ctrl;
 812                 int len = ETHER_ADDR_LEN;
 813
 814                 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
 815                 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
 816
 817                 memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
 818                 virtio_send_command(hw->cvq, &ctrl, &len, 1);
 819         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
 820                 virtio_set_hwaddr(hw);
 821 }
 822
 823 static int
 824 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
 825 {
 826         struct virtio_hw *hw = dev->data->dev_private;
 827         struct virtio_pmd_ctrl ctrl;
 828         int len;
 829
 830         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
 831                 return -ENOTSUP;
 832
 833         ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
 834         ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
 835         memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
 836         len = sizeof(vlan_id);
 837
 838         return virtio_send_command(hw->cvq, &ctrl, &len, 1);
 839 }
 840
 841 static void
 842 virtio_negotiate_features(struct virtio_hw *hw)
 843 {
 844         uint32_t host_features;
 845
 846         /* Prepare guest_features: feature that driver wants to support */
 847         hw->guest_features = VIRTIO_PMD_GUEST_FEATURES;
 848         PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %x",
 849                 hw->guest_features);
 850
 851         /* Read device(host) feature bits */
 852         host_features = VIRTIO_READ_REG_4(hw, VIRTIO_PCI_HOST_FEATURES);
 853         PMD_INIT_LOG(DEBUG, "host_features before negotiate = %x",
 854                 host_features);
 855
 856         /*
 857          * Negotiate features: Subset of device feature bits are written back
 858          * guest feature bits.
 859          */
 860         hw->guest_features = vtpci_negotiate_features(hw, host_features);
 861         PMD_INIT_LOG(DEBUG, "features after negotiate = %x",
 862                 hw->guest_features);
 863 }
 864
 865 #ifdef RTE_EXEC_ENV_LINUXAPP
 866 static int
 867 parse_sysfs_value(const char *filename, unsigned long *val)
 868 {
 869         FILE *f;
 870         char buf[BUFSIZ];
 871         char *end = NULL;
 872
 873         f = fopen(filename, "r");
 874         if (f == NULL) {
 875                 PMD_INIT_LOG(ERR, "%s(): cannot open sysfs value %s",
 876                              __func__, filename);
 877                 return -1;
 878         }
 879
 880         if (fgets(buf, sizeof(buf), f) == NULL) {
 881                 PMD_INIT_LOG(ERR, "%s(): cannot read sysfs value %s",
 882                              __func__, filename);
 883                 fclose(f);
 884                 return -1;
 885         }
 886         *val = strtoul(buf, &end, 0);
 887         if ((buf[0] == '\0') || (end == NULL) || (*end != '\n')) {
 888                 PMD_INIT_LOG(ERR, "%s(): cannot parse sysfs value %s",
 889                              __func__, filename);
 890                 fclose(f);
 891                 return -1;
 892         }
 893         fclose(f);
 894         return 0;
 895 }
 896
 897 static int get_uio_dev(struct rte_pci_addr *loc, char *buf, unsigned int buflen,
 898                         unsigned int *uio_num)
 899 {
 900         struct dirent *e;
 901         DIR *dir;
 902         char dirname[PATH_MAX];
 903
 904         /* depending on kernel version, uio can be located in uio/uioX
 905          * or uio:uioX */
 906         snprintf(dirname, sizeof(dirname),
 907                      SYSFS_PCI_DEVICES "/" PCI_PRI_FMT "/uio",
 908                      loc->domain, loc->bus, loc->devid, loc->function);
 909         dir = opendir(dirname);
 910         if (dir == NULL) {
 911                 /* retry with the parent directory */
 912                 snprintf(dirname, sizeof(dirname),
 913                              SYSFS_PCI_DEVICES "/" PCI_PRI_FMT,
 914                              loc->domain, loc->bus, loc->devid, loc->function);
 915                 dir = opendir(dirname);
 916
 917                 if (dir == NULL) {
 918                         PMD_INIT_LOG(ERR, "Cannot opendir %s", dirname);
 919                         return -1;
 920                 }
 921         }
 922
 923         /* take the first file starting with "uio" */
 924         while ((e = readdir(dir)) != NULL) {
 925                 /* format could be uio%d ...*/
 926                 int shortprefix_len = sizeof("uio") - 1;
 927                 /* ... or uio:uio%d */
 928                 int longprefix_len = sizeof("uio:uio") - 1;
 929                 char *endptr;
 930
 931                 if (strncmp(e->d_name, "uio", 3) != 0)
 932                         continue;
 933
 934                 /* first try uio%d */
 935                 errno = 0;
 936                 *uio_num = strtoull(e->d_name + shortprefix_len, &endptr, 10);
 937                 if (errno == 0 && endptr != (e->d_name + shortprefix_len)) {
 938                         snprintf(buf, buflen, "%s/uio%u", dirname, *uio_num);
 939                         break;
 940                 }
 941
 942                 /* then try uio:uio%d */
 943                 errno = 0;
 944                 *uio_num = strtoull(e->d_name + longprefix_len, &endptr, 10);
 945                 if (errno == 0 && endptr != (e->d_name + longprefix_len)) {
 946                         snprintf(buf, buflen, "%s/uio:uio%u", dirname,
 947                                      *uio_num);
 948                         break;
 949                 }
 950         }
 951         closedir(dir);
 952
 953         /* No uio resource found */
 954         if (e == NULL) {
 955                 PMD_INIT_LOG(ERR, "Could not find uio resource");
 956                 return -1;
 957         }
 958
 959         return 0;
 960 }
 961
 962 static int
 963 virtio_has_msix(const struct rte_pci_addr *loc)
 964 {
 965         DIR *d;
 966         char dirname[PATH_MAX];
 967
 968         snprintf(dirname, sizeof(dirname),
 969                      SYSFS_PCI_DEVICES "/" PCI_PRI_FMT "/msi_irqs",
 970                      loc->domain, loc->bus, loc->devid, loc->function);
 971
 972         d = opendir(dirname);
 973         if (d)
 974                 closedir(d);
 975
 976         return (d != NULL);
 977 }
 978
 979 /* Extract I/O port numbers from sysfs */
 980 static int virtio_resource_init_by_uio(struct rte_pci_device *pci_dev)
 981 {
 982         char dirname[PATH_MAX];
 983         char filename[PATH_MAX];
 984         unsigned long start, size;
 985         unsigned int uio_num;
 986
 987         if (get_uio_dev(&pci_dev->addr, dirname, sizeof(dirname), &uio_num) < 0)
 988                 return -1;
 989
 990         /* get portio size */
 991         snprintf(filename, sizeof(filename),
 992                      "%s/portio/port0/size", dirname);
 993         if (parse_sysfs_value(filename, &size) < 0) {
 994                 PMD_INIT_LOG(ERR, "%s(): cannot parse size",
 995                              __func__);
 996                 return -1;
 997         }
 998
 999         /* get portio start */
1000         snprintf(filename, sizeof(filename),
1001                  "%s/portio/port0/start", dirname);
1002         if (parse_sysfs_value(filename, &start) < 0) {
1003                 PMD_INIT_LOG(ERR, "%s(): cannot parse portio start",
1004                              __func__);
1005                 return -1;
1006         }
1007         pci_dev->mem_resource[0].addr = (void *)(uintptr_t)start;
1008         pci_dev->mem_resource[0].len =  (uint64_t)size;
1009         PMD_INIT_LOG(DEBUG,
1010                      "PCI Port IO found start=0x%lx with size=0x%lx",
1011                      start, size);
1012
1013         /* save fd */
1014         memset(dirname, 0, sizeof(dirname));
1015         snprintf(dirname, sizeof(dirname), "/dev/uio%u", uio_num);
1016         pci_dev->intr_handle.fd = open(dirname, O_RDWR);
1017         if (pci_dev->intr_handle.fd < 0) {
1018                 PMD_INIT_LOG(ERR, "Cannot open %s: %s\n",
1019                         dirname, strerror(errno));
1020                 return -1;
1021         }
1022
1023         pci_dev->intr_handle.type = RTE_INTR_HANDLE_UIO;
1024         pci_dev->driver->drv_flags |= RTE_PCI_DRV_INTR_LSC;
1025
1026         return 0;
1027 }
1028
1029 /* Extract port I/O numbers from proc/ioports */
1030 static int virtio_resource_init_by_ioports(struct rte_pci_device *pci_dev)
1031 {
1032         uint16_t start, end;
1033         int size;
1034         FILE *fp;
1035         char *line = NULL;
1036         char pci_id[16];
1037         int found = 0;
1038         size_t linesz;
1039
1040         snprintf(pci_id, sizeof(pci_id), PCI_PRI_FMT,
1041                  pci_dev->addr.domain,
1042                  pci_dev->addr.bus,
1043                  pci_dev->addr.devid,
1044                  pci_dev->addr.function);
1045
1046         fp = fopen("/proc/ioports", "r");
1047         if (fp == NULL) {
1048                 PMD_INIT_LOG(ERR, "%s(): can't open ioports", __func__);
1049                 return -1;
1050         }
1051
1052         while (getdelim(&line, &linesz, '\n', fp) > 0) {
1053                 char *ptr = line;
1054                 char *left;
1055                 int n;
1056
1057                 n = strcspn(ptr, ":");
1058                 ptr[n] = 0;
1059                 left = &ptr[n+1];
1060
1061                 while (*left && isspace(*left))
1062                         left++;
1063
1064                 if (!strncmp(left, pci_id, strlen(pci_id))) {
1065                         found = 1;
1066
1067                         while (*ptr && isspace(*ptr))
1068                                 ptr++;
1069
1070                         sscanf(ptr, "%04hx-%04hx", &start, &end);
1071                         size = end - start + 1;
1072
1073                         break;
1074                 }
1075         }
1076
1077         free(line);
1078         fclose(fp);
1079
1080         if (!found)
1081                 return -1;
1082
1083         pci_dev->mem_resource[0].addr = (void *)(uintptr_t)(uint32_t)start;
1084         pci_dev->mem_resource[0].len =  (uint64_t)size;
1085         PMD_INIT_LOG(DEBUG,
1086                 "PCI Port IO found start=0x%x with size=0x%x",
1087                 start, size);
1088
1089         /* can't support lsc interrupt without uio */
1090         pci_dev->driver->drv_flags &= ~RTE_PCI_DRV_INTR_LSC;
1091
1092         return 0;
1093 }
1094
1095 /* Extract I/O port numbers from sysfs */
1096 static int virtio_resource_init(struct rte_pci_device *pci_dev)
1097 {
1098         if (virtio_resource_init_by_uio(pci_dev) == 0)
1099                 return 0;
1100         else
1101                 return virtio_resource_init_by_ioports(pci_dev);
1102 }
1103
1104 #else
1105 static int
1106 virtio_has_msix(const struct rte_pci_addr *loc __rte_unused)
1107 {
1108         /* nic_uio does not enable interrupts, return 0 (false). */
1109         return 0;
1110 }
1111
1112 static int virtio_resource_init(struct rte_pci_device *pci_dev __rte_unused)
1113 {
1114         /* no setup required */
1115         return 0;
1116 }
1117 #endif
1118
1119 /*
1120  * Process Virtio Config changed interrupt and call the callback
1121  * if link state changed.
1122  */
1123 static void
1124 virtio_interrupt_handler(__rte_unused struct rte_intr_handle *handle,
1125                          void *param)
1126 {
1127         struct rte_eth_dev *dev = param;
1128         struct virtio_hw *hw = dev->data->dev_private;
1129         uint8_t isr;
1130
1131         /* Read interrupt status which clears interrupt */
1132         isr = vtpci_isr(hw);
1133         PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1134
1135         if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0)
1136                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1137
1138         if (isr & VIRTIO_PCI_ISR_CONFIG) {
1139                 if (virtio_dev_link_update(dev, 0) == 0)
1140                         _rte_eth_dev_callback_process(dev,
1141                                                       RTE_ETH_EVENT_INTR_LSC);
1142         }
1143
1144 }
1145
1146 static void
1147 rx_func_get(struct rte_eth_dev *eth_dev)
1148 {
1149         struct virtio_hw *hw = eth_dev->data->dev_private;
1150         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF))
1151                 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1152         else
1153                 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1154 }
1155
1156 /*
1157  * This function is based on probe() function in virtio_pci.c
1158  * It returns 0 on success.
1159  */
1160 static int
1161 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1162 {
1163         struct virtio_hw *hw = eth_dev->data->dev_private;
1164         struct virtio_net_config *config;
1165         struct virtio_net_config local_config;
1166         uint32_t offset_conf = sizeof(config->mac);
1167         struct rte_pci_device *pci_dev;
1168
1169         RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr));
1170
1171         eth_dev->dev_ops = &virtio_eth_dev_ops;
1172         eth_dev->tx_pkt_burst = &virtio_xmit_pkts;
1173
1174         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1175                 rx_func_get(eth_dev);
1176                 return 0;
1177         }
1178
1179         /* Allocate memory for storing MAC addresses */
1180         eth_dev->data->mac_addrs = rte_zmalloc("virtio", ETHER_ADDR_LEN, 0);
1181         if (eth_dev->data->mac_addrs == NULL) {
1182                 PMD_INIT_LOG(ERR,
1183                         "Failed to allocate %d bytes needed to store MAC addresses",
1184                         ETHER_ADDR_LEN);
1185                 return -ENOMEM;
1186         }
1187
1188         pci_dev = eth_dev->pci_dev;
1189         if (virtio_resource_init(pci_dev) < 0)
1190                 return -1;
1191
1192         hw->use_msix = virtio_has_msix(&pci_dev->addr);
1193         hw->io_base = (uint32_t)(uintptr_t)pci_dev->mem_resource[0].addr;
1194
1195         /* Reset the device although not necessary at startup */
1196         vtpci_reset(hw);
1197
1198         /* Tell the host we've noticed this device. */
1199         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1200
1201         /* Tell the host we've known how to drive the device. */
1202         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1203         virtio_negotiate_features(hw);
1204
1205         /* If host does not support status then disable LSC */
1206         if (!vtpci_with_feature(hw, VIRTIO_NET_F_STATUS))
1207                 pci_dev->driver->drv_flags &= ~RTE_PCI_DRV_INTR_LSC;
1208
1209         rx_func_get(eth_dev);
1210
1211         /* Setting up rx_header size for the device */
1212         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF))
1213                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1214         else
1215                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1216
1217         /* Copy the permanent MAC address to: virtio_hw */
1218         virtio_get_hwaddr(hw);
1219         ether_addr_copy((struct ether_addr *) hw->mac_addr,
1220                         &eth_dev->data->mac_addrs[0]);
1221         PMD_INIT_LOG(DEBUG,
1222                      "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1223                      hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1224                      hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1225
1226         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1227                 config = &local_config;
1228
1229                 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1230                         offset_conf += sizeof(config->status);
1231                 } else {
1232                         PMD_INIT_LOG(DEBUG,
1233                                      "VIRTIO_NET_F_STATUS is not supported");
1234                         config->status = 0;
1235                 }
1236
1237                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1238                         offset_conf += sizeof(config->max_virtqueue_pairs);
1239                 } else {
1240                         PMD_INIT_LOG(DEBUG,
1241                                      "VIRTIO_NET_F_MQ is not supported");
1242                         config->max_virtqueue_pairs = 1;
1243                 }
1244
1245                 vtpci_read_dev_config(hw, 0, (uint8_t *)config, offset_conf);
1246
1247                 hw->max_rx_queues =
1248                         (VIRTIO_MAX_RX_QUEUES < config->max_virtqueue_pairs) ?
1249                         VIRTIO_MAX_RX_QUEUES : config->max_virtqueue_pairs;
1250                 hw->max_tx_queues =
1251                         (VIRTIO_MAX_TX_QUEUES < config->max_virtqueue_pairs) ?
1252                         VIRTIO_MAX_TX_QUEUES : config->max_virtqueue_pairs;
1253
1254                 virtio_dev_cq_queue_setup(eth_dev,
1255                                         config->max_virtqueue_pairs * 2,
1256                                         SOCKET_ID_ANY);
1257
1258                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1259                                 config->max_virtqueue_pairs);
1260                 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1261                 PMD_INIT_LOG(DEBUG,
1262                                 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1263                                 config->mac[0], config->mac[1],
1264                                 config->mac[2], config->mac[3],
1265                                 config->mac[4], config->mac[5]);
1266         } else {
1267                 hw->max_rx_queues = 1;
1268                 hw->max_tx_queues = 1;
1269         }
1270
1271         eth_dev->data->nb_rx_queues = hw->max_rx_queues;
1272         eth_dev->data->nb_tx_queues = hw->max_tx_queues;
1273
1274         PMD_INIT_LOG(DEBUG, "hw->max_rx_queues=%d   hw->max_tx_queues=%d",
1275                         hw->max_rx_queues, hw->max_tx_queues);
1276         PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1277                         eth_dev->data->port_id, pci_dev->id.vendor_id,
1278                         pci_dev->id.device_id);
1279
1280         /* Setup interrupt callback  */
1281         if (pci_dev->driver->drv_flags & RTE_PCI_DRV_INTR_LSC)
1282                 rte_intr_callback_register(&pci_dev->intr_handle,
1283                                    virtio_interrupt_handler, eth_dev);
1284
1285         virtio_dev_cq_start(eth_dev);
1286
1287         return 0;
1288 }
1289
1290 static int
1291 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1292 {
1293         struct rte_pci_device *pci_dev;
1294         struct virtio_hw *hw = eth_dev->data->dev_private;
1295
1296         PMD_INIT_FUNC_TRACE();
1297
1298         if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1299                 return -EPERM;
1300
1301         if (hw->started == 1) {
1302                 virtio_dev_stop(eth_dev);
1303                 virtio_dev_close(eth_dev);
1304         }
1305         pci_dev = eth_dev->pci_dev;
1306
1307         eth_dev->dev_ops = NULL;
1308         eth_dev->tx_pkt_burst = NULL;
1309         eth_dev->rx_pkt_burst = NULL;
1310
1311         virtio_dev_queue_release(hw->cvq);
1312
1313         rte_free(eth_dev->data->mac_addrs);
1314         eth_dev->data->mac_addrs = NULL;
1315
1316         /* reset interrupt callback  */
1317         if (pci_dev->driver->drv_flags & RTE_PCI_DRV_INTR_LSC)
1318                 rte_intr_callback_unregister(&pci_dev->intr_handle,
1319                                                 virtio_interrupt_handler,
1320                                                 eth_dev);
1321
1322         PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1323
1324         return 0;
1325 }
1326
1327 static struct eth_driver rte_virtio_pmd = {
1328         .pci_drv = {
1329                 .name = "rte_virtio_pmd",
1330                 .id_table = pci_id_virtio_map,
1331                 .drv_flags = RTE_PCI_DRV_DETACHABLE,
1332         },
1333         .eth_dev_init = eth_virtio_dev_init,
1334         .eth_dev_uninit = eth_virtio_dev_uninit,
1335         .dev_private_size = sizeof(struct virtio_hw),
1336 };
1337
1338 /*
1339  * Driver initialization routine.
1340  * Invoked once at EAL init time.
1341  * Register itself as the [Poll Mode] Driver of PCI virtio devices.
1342  * Returns 0 on success.
1343  */
1344 static int
1345 rte_virtio_pmd_init(const char *name __rte_unused,
1346                     const char *param __rte_unused)
1347 {
1348         if (rte_eal_iopl_init() != 0) {
1349                 PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1350                 return -1;
1351         }
1352
1353         rte_eth_driver_register(&rte_virtio_pmd);
1354         return 0;
1355 }
1356
1357 /*
1358  * Configure virtio device
1359  * It returns 0 on success.
1360  */
1361 static int
1362 virtio_dev_configure(struct rte_eth_dev *dev)
1363 {
1364         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1365         struct virtio_hw *hw = dev->data->dev_private;
1366         struct rte_pci_device *pci_dev = dev->pci_dev;
1367
1368         PMD_INIT_LOG(DEBUG, "configure");
1369
1370         if (rxmode->hw_ip_checksum) {
1371                 PMD_DRV_LOG(ERR, "HW IP checksum not supported");
1372                 return (-EINVAL);
1373         }
1374
1375         hw->vlan_strip = rxmode->hw_vlan_strip;
1376
1377         if (rxmode->hw_vlan_filter
1378             && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1379                 PMD_DRV_LOG(NOTICE,
1380                             "vlan filtering not available on this host");
1381                 return -ENOTSUP;
1382         }
1383
1384         if (pci_dev->driver->drv_flags & RTE_PCI_DRV_INTR_LSC)
1385                 if (vtpci_irq_config(hw, 0) == VIRTIO_MSI_NO_VECTOR) {
1386                         PMD_DRV_LOG(ERR, "failed to set config vector");
1387                         return -EBUSY;
1388                 }
1389
1390         return 0;
1391 }
1392
1393
1394 static int
1395 virtio_dev_start(struct rte_eth_dev *dev)
1396 {
1397         uint16_t nb_queues, i;
1398         struct virtio_hw *hw = dev->data->dev_private;
1399         struct rte_pci_device *pci_dev = dev->pci_dev;
1400
1401         /* check if lsc interrupt feature is enabled */
1402         if (dev->data->dev_conf.intr_conf.lsc) {
1403                 if (!(pci_dev->driver->drv_flags & RTE_PCI_DRV_INTR_LSC)) {
1404                         PMD_DRV_LOG(ERR, "link status not supported by host");
1405                         return -ENOTSUP;
1406                 }
1407
1408                 if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0) {
1409                         PMD_DRV_LOG(ERR, "interrupt enable failed");
1410                         return -EIO;
1411                 }
1412         }
1413
1414         /* Initialize Link state */
1415         virtio_dev_link_update(dev, 0);
1416
1417         /* On restart after stop do not touch queues */
1418         if (hw->started)
1419                 return 0;
1420
1421         /* Do final configuration before rx/tx engine starts */
1422         virtio_dev_rxtx_start(dev);
1423         vtpci_reinit_complete(hw);
1424
1425         hw->started = 1;
1426
1427         /*Notify the backend
1428          *Otherwise the tap backend might already stop its queue due to fullness.
1429          *vhost backend will have no chance to be waked up
1430          */
1431         nb_queues = dev->data->nb_rx_queues;
1432         if (nb_queues > 1) {
1433                 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1434                         return -EINVAL;
1435         }
1436
1437         PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1438
1439         for (i = 0; i < nb_queues; i++)
1440                 virtqueue_notify(dev->data->rx_queues[i]);
1441
1442         PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1443
1444         for (i = 0; i < dev->data->nb_rx_queues; i++)
1445                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1446
1447         for (i = 0; i < dev->data->nb_tx_queues; i++)
1448                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1449
1450         return 0;
1451 }
1452
1453 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1454 {
1455         struct rte_mbuf *buf;
1456         int i, mbuf_num = 0;
1457
1458         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1459                 PMD_INIT_LOG(DEBUG,
1460                              "Before freeing rxq[%d] used and unused buf", i);
1461                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1462
1463                 PMD_INIT_LOG(DEBUG, "rx_queues[%d]=%p",
1464                                 i, dev->data->rx_queues[i]);
1465                 while ((buf = (struct rte_mbuf *)virtqueue_detatch_unused(
1466                                         dev->data->rx_queues[i])) != NULL) {
1467                         rte_pktmbuf_free(buf);
1468                         mbuf_num++;
1469                 }
1470
1471                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1472                 PMD_INIT_LOG(DEBUG,
1473                              "After freeing rxq[%d] used and unused buf", i);
1474                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1475         }
1476
1477         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1478                 PMD_INIT_LOG(DEBUG,
1479                              "Before freeing txq[%d] used and unused bufs",
1480                              i);
1481                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1482
1483                 mbuf_num = 0;
1484                 while ((buf = (struct rte_mbuf *)virtqueue_detatch_unused(
1485                                         dev->data->tx_queues[i])) != NULL) {
1486                         rte_pktmbuf_free(buf);
1487
1488                         mbuf_num++;
1489                 }
1490
1491                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1492                 PMD_INIT_LOG(DEBUG,
1493                              "After freeing txq[%d] used and unused buf", i);
1494                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1495         }
1496 }
1497
1498 /*
1499  * Stop device: disable interrupt and mark link down
1500  */
1501 static void
1502 virtio_dev_stop(struct rte_eth_dev *dev)
1503 {
1504         struct rte_eth_link link;
1505
1506         PMD_INIT_LOG(DEBUG, "stop");
1507
1508         if (dev->data->dev_conf.intr_conf.lsc)
1509                 rte_intr_disable(&dev->pci_dev->intr_handle);
1510
1511         memset(&link, 0, sizeof(link));
1512         virtio_dev_atomic_write_link_status(dev, &link);
1513 }
1514
1515 static int
1516 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
1517 {
1518         struct rte_eth_link link, old;
1519         uint16_t status;
1520         struct virtio_hw *hw = dev->data->dev_private;
1521         memset(&link, 0, sizeof(link));
1522         virtio_dev_atomic_read_link_status(dev, &link);
1523         old = link;
1524         link.link_duplex = FULL_DUPLEX;
1525         link.link_speed  = SPEED_10G;
1526
1527         if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1528                 PMD_INIT_LOG(DEBUG, "Get link status from hw");
1529                 vtpci_read_dev_config(hw,
1530                                 offsetof(struct virtio_net_config, status),
1531                                 &status, sizeof(status));
1532                 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
1533                         link.link_status = 0;
1534                         PMD_INIT_LOG(DEBUG, "Port %d is down",
1535                                      dev->data->port_id);
1536                 } else {
1537                         link.link_status = 1;
1538                         PMD_INIT_LOG(DEBUG, "Port %d is up",
1539                                      dev->data->port_id);
1540                 }
1541         } else {
1542                 link.link_status = 1;   /* Link up */
1543         }
1544         virtio_dev_atomic_write_link_status(dev, &link);
1545
1546         return (old.link_status == link.link_status) ? -1 : 0;
1547 }
1548
1549 static void
1550 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1551 {
1552         struct virtio_hw *hw = dev->data->dev_private;
1553
1554         dev_info->driver_name = dev->driver->pci_drv.name;
1555         dev_info->max_rx_queues = (uint16_t)hw->max_rx_queues;
1556         dev_info->max_tx_queues = (uint16_t)hw->max_tx_queues;
1557         dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
1558         dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
1559         dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
1560         dev_info->default_txconf = (struct rte_eth_txconf) {
1561                 .txq_flags = ETH_TXQ_FLAGS_NOOFFLOADS
1562         };
1563 }
1564
1565 /*
1566  * It enables testpmd to collect per queue stats.
1567  */
1568 static int
1569 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
1570 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
1571 __rte_unused uint8_t is_rx)
1572 {
1573         return 0;
1574 }
1575
1576 static struct rte_driver rte_virtio_driver = {
1577         .type = PMD_PDEV,
1578         .init = rte_virtio_pmd_init,
1579 };
1580
1581 PMD_REGISTER_DRIVER(rte_virtio_driver);