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
  40 #include <rte_ethdev.h>
  41 #include <rte_memcpy.h>
  42 #include <rte_string_fns.h>
  43 #include <rte_memzone.h>
  44 #include <rte_malloc.h>
  45 #include <rte_atomic.h>
  46 #include <rte_branch_prediction.h>
  47 #include <rte_pci.h>
  48 #include <rte_ether.h>
  49 #include <rte_common.h>
  50 #include <rte_errno.h>
  51
  52 #include <rte_memory.h>
  53 #include <rte_eal.h>
  54 #include <rte_dev.h>
  55
  56 #include "virtio_ethdev.h"
  57 #include "virtio_pci.h"
  58 #include "virtio_logs.h"
  59 #include "virtqueue.h"
  60 #include "virtio_rxtx.h"
  61
  62 static int eth_virtio_dev_init(struct rte_eth_dev *eth_dev);
  63 static int eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev);
  64 static int  virtio_dev_configure(struct rte_eth_dev *dev);
  65 static int  virtio_dev_start(struct rte_eth_dev *dev);
  66 static void virtio_dev_stop(struct rte_eth_dev *dev);
  67 static void virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
  68 static void virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
  69 static void virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
  70 static void virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
  71 static void virtio_dev_info_get(struct rte_eth_dev *dev,
  72                                 struct rte_eth_dev_info *dev_info);
  73 static int virtio_dev_link_update(struct rte_eth_dev *dev,
  74         __rte_unused int wait_to_complete);
  75
  76 static void virtio_set_hwaddr(struct virtio_hw *hw);
  77 static void virtio_get_hwaddr(struct virtio_hw *hw);
  78
  79 static void virtio_dev_stats_get(struct rte_eth_dev *dev,
  80                                  struct rte_eth_stats *stats);
  81 static int virtio_dev_xstats_get(struct rte_eth_dev *dev,
  82                                  struct rte_eth_xstats *xstats, unsigned n);
  83 static void virtio_dev_stats_reset(struct rte_eth_dev *dev);
  84 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
  85 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
  86                                 uint16_t vlan_id, int on);
  87 static void virtio_mac_addr_add(struct rte_eth_dev *dev,
  88                                 struct ether_addr *mac_addr,
  89                                 uint32_t index, uint32_t vmdq __rte_unused);
  90 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
  91 static void virtio_mac_addr_set(struct rte_eth_dev *dev,
  92                                 struct ether_addr *mac_addr);
  93
  94 static int virtio_dev_queue_stats_mapping_set(
  95         __rte_unused struct rte_eth_dev *eth_dev,
  96         __rte_unused uint16_t queue_id,
  97         __rte_unused uint8_t stat_idx,
  98         __rte_unused uint8_t is_rx);
  99
 100 /*
 101  * The set of PCI devices this driver supports
 102  */
 103 static const struct rte_pci_id pci_id_virtio_map[] = {
 104
 105 #define RTE_PCI_DEV_ID_DECL_VIRTIO(vend, dev) {RTE_PCI_DEVICE(vend, dev)},
 106 #include "rte_pci_dev_ids.h"
 107
 108 { .vendor_id = 0, /* sentinel */ },
 109 };
 110
 111 struct rte_virtio_xstats_name_off {
 112         char name[RTE_ETH_XSTATS_NAME_SIZE];
 113         unsigned offset;
 114 };
 115
 116 /* [rt]x_qX_ is prepended to the name string here */
 117 static const struct rte_virtio_xstats_name_off rte_virtio_q_stat_strings[] = {
 118         {"good_packets",           offsetof(struct virtqueue, packets)},
 119         {"good_bytes",             offsetof(struct virtqueue, bytes)},
 120         {"errors",                 offsetof(struct virtqueue, errors)},
 121         {"multicast_packets",      offsetof(struct virtqueue, multicast)},
 122         {"broadcast_packets",      offsetof(struct virtqueue, broadcast)},
 123         {"undersize_packets",      offsetof(struct virtqueue, size_bins[0])},
 124         {"size_64_packets",        offsetof(struct virtqueue, size_bins[1])},
 125         {"size_65_127_packets",    offsetof(struct virtqueue, size_bins[2])},
 126         {"size_128_255_packets",   offsetof(struct virtqueue, size_bins[3])},
 127         {"size_256_511_packets",   offsetof(struct virtqueue, size_bins[4])},
 128         {"size_512_1023_packets",  offsetof(struct virtqueue, size_bins[5])},
 129         {"size_1024_1517_packets", offsetof(struct virtqueue, size_bins[6])},
 130         {"size_1518_max_packets",  offsetof(struct virtqueue, size_bins[7])},
 131 };
 132
 133 #define VIRTIO_NB_Q_XSTATS (sizeof(rte_virtio_q_stat_strings) / \
 134                             sizeof(rte_virtio_q_stat_strings[0]))
 135
 136 static int
 137 virtio_send_command(struct virtqueue *vq, struct virtio_pmd_ctrl *ctrl,
 138                 int *dlen, int pkt_num)
 139 {
 140         uint32_t head, i;
 141         int k, sum = 0;
 142         virtio_net_ctrl_ack status = ~0;
 143         struct virtio_pmd_ctrl result;
 144
 145         ctrl->status = status;
 146
 147         if (!(vq && vq->hw->cvq)) {
 148                 PMD_INIT_LOG(ERR, "Control queue is not supported.");
 149                 return -1;
 150         }
 151         head = vq->vq_desc_head_idx;
 152
 153         PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
 154                 "vq->hw->cvq = %p vq = %p",
 155                 vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
 156
 157         if ((vq->vq_free_cnt < ((uint32_t)pkt_num + 2)) || (pkt_num < 1))
 158                 return -1;
 159
 160         memcpy(vq->virtio_net_hdr_mz->addr, ctrl,
 161                 sizeof(struct virtio_pmd_ctrl));
 162
 163         /*
 164          * Format is enforced in qemu code:
 165          * One TX packet for header;
 166          * At least one TX packet per argument;
 167          * One RX packet for ACK.
 168          */
 169         vq->vq_ring.desc[head].flags = VRING_DESC_F_NEXT;
 170         vq->vq_ring.desc[head].addr = vq->virtio_net_hdr_mz->phys_addr;
 171         vq->vq_ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
 172         vq->vq_free_cnt--;
 173         i = vq->vq_ring.desc[head].next;
 174
 175         for (k = 0; k < pkt_num; k++) {
 176                 vq->vq_ring.desc[i].flags = VRING_DESC_F_NEXT;
 177                 vq->vq_ring.desc[i].addr = vq->virtio_net_hdr_mz->phys_addr
 178                         + sizeof(struct virtio_net_ctrl_hdr)
 179                         + sizeof(ctrl->status) + sizeof(uint8_t)*sum;
 180                 vq->vq_ring.desc[i].len = dlen[k];
 181                 sum += dlen[k];
 182                 vq->vq_free_cnt--;
 183                 i = vq->vq_ring.desc[i].next;
 184         }
 185
 186         vq->vq_ring.desc[i].flags = VRING_DESC_F_WRITE;
 187         vq->vq_ring.desc[i].addr = vq->virtio_net_hdr_mz->phys_addr
 188                         + sizeof(struct virtio_net_ctrl_hdr);
 189         vq->vq_ring.desc[i].len = sizeof(ctrl->status);
 190         vq->vq_free_cnt--;
 191
 192         vq->vq_desc_head_idx = vq->vq_ring.desc[i].next;
 193
 194         vq_update_avail_ring(vq, head);
 195         vq_update_avail_idx(vq);
 196
 197         PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
 198
 199         virtqueue_notify(vq);
 200
 201         rte_rmb();
 202         while (vq->vq_used_cons_idx == vq->vq_ring.used->idx) {
 203                 rte_rmb();
 204                 usleep(100);
 205         }
 206
 207         while (vq->vq_used_cons_idx != vq->vq_ring.used->idx) {
 208                 uint32_t idx, desc_idx, used_idx;
 209                 struct vring_used_elem *uep;
 210
 211                 used_idx = (uint32_t)(vq->vq_used_cons_idx
 212                                 & (vq->vq_nentries - 1));
 213                 uep = &vq->vq_ring.used->ring[used_idx];
 214                 idx = (uint32_t) uep->id;
 215                 desc_idx = idx;
 216
 217                 while (vq->vq_ring.desc[desc_idx].flags & VRING_DESC_F_NEXT) {
 218                         desc_idx = vq->vq_ring.desc[desc_idx].next;
 219                         vq->vq_free_cnt++;
 220                 }
 221
 222                 vq->vq_ring.desc[desc_idx].next = vq->vq_desc_head_idx;
 223                 vq->vq_desc_head_idx = idx;
 224
 225                 vq->vq_used_cons_idx++;
 226                 vq->vq_free_cnt++;
 227         }
 228
 229         PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
 230                         vq->vq_free_cnt, vq->vq_desc_head_idx);
 231
 232         memcpy(&result, vq->virtio_net_hdr_mz->addr,
 233                         sizeof(struct virtio_pmd_ctrl));
 234
 235         return result.status;
 236 }
 237
 238 static int
 239 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
 240 {
 241         struct virtio_hw *hw = dev->data->dev_private;
 242         struct virtio_pmd_ctrl ctrl;
 243         int dlen[1];
 244         int ret;
 245
 246         ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
 247         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
 248         memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
 249
 250         dlen[0] = sizeof(uint16_t);
 251
 252         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 253         if (ret) {
 254                 PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
 255                           "failed, this is too late now...");
 256                 return -EINVAL;
 257         }
 258
 259         return 0;
 260 }
 261
 262 void
 263 virtio_dev_queue_release(struct virtqueue *vq) {
 264         struct virtio_hw *hw;
 265
 266         if (vq) {
 267                 hw = vq->hw;
 268                 hw->vtpci_ops->del_queue(hw, vq);
 269
 270                 rte_free(vq->sw_ring);
 271                 rte_free(vq);
 272         }
 273 }
 274
 275 int virtio_dev_queue_setup(struct rte_eth_dev *dev,
 276                         int queue_type,
 277                         uint16_t queue_idx,
 278                         uint16_t vtpci_queue_idx,
 279                         uint16_t nb_desc,
 280                         unsigned int socket_id,
 281                         struct virtqueue **pvq)
 282 {
 283         char vq_name[VIRTQUEUE_MAX_NAME_SZ];
 284         const struct rte_memzone *mz;
 285         unsigned int vq_size, size;
 286         struct virtio_hw *hw = dev->data->dev_private;
 287         struct virtqueue *vq = NULL;
 288         const char *queue_names[] = {"rvq", "txq", "cvq"};
 289
 290         PMD_INIT_LOG(DEBUG, "setting up queue: %u", vtpci_queue_idx);
 291
 292         /*
 293          * Read the virtqueue size from the Queue Size field
 294          * Always power of 2 and if 0 virtqueue does not exist
 295          */
 296         vq_size = hw->vtpci_ops->get_queue_num(hw, vtpci_queue_idx);
 297         PMD_INIT_LOG(DEBUG, "vq_size: %u nb_desc:%u", vq_size, nb_desc);
 298         if (vq_size == 0) {
 299                 PMD_INIT_LOG(ERR, "virtqueue does not exist");
 300                 return -EINVAL;
 301         }
 302
 303         if (!rte_is_power_of_2(vq_size)) {
 304                 PMD_INIT_LOG(ERR, "virtqueue size is not powerof 2");
 305                 return -EINVAL;
 306         }
 307
 308         snprintf(vq_name, sizeof(vq_name), "port%d_%s%d",
 309                  dev->data->port_id, queue_names[queue_type], queue_idx);
 310         vq = rte_zmalloc(vq_name, sizeof(struct virtqueue) +
 311                          vq_size * sizeof(struct vq_desc_extra),
 312                          RTE_CACHE_LINE_SIZE);
 313         if (vq == NULL) {
 314                 PMD_INIT_LOG(ERR, "Can not allocate virtqueue");
 315                 return -ENOMEM;
 316         }
 317
 318         if (queue_type == VTNET_RQ) {
 319                 size_t sz_sw;
 320
 321                 sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
 322                         sizeof(vq->sw_ring[0]);
 323                 vq->sw_ring = rte_zmalloc_socket("rxq->sw_ring", sz_sw,
 324                                                  RTE_CACHE_LINE_SIZE,
 325                                                  socket_id);
 326                 if (!vq->sw_ring) {
 327                         PMD_INIT_LOG(ERR, "Can not allocate RX soft ring");
 328                         rte_free(vq);
 329                         return -ENOMEM;
 330                 }
 331         }
 332
 333         vq->hw = hw;
 334         vq->port_id = dev->data->port_id;
 335         vq->queue_id = queue_idx;
 336         vq->vq_queue_index = vtpci_queue_idx;
 337         vq->vq_nentries = vq_size;
 338
 339         if (nb_desc == 0 || nb_desc > vq_size)
 340                 nb_desc = vq_size;
 341         vq->vq_free_cnt = nb_desc;
 342
 343         /*
 344          * Reserve a memzone for vring elements
 345          */
 346         size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
 347         vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
 348         PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d", size, vq->vq_ring_size);
 349
 350         mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size,
 351                 socket_id, 0, VIRTIO_PCI_VRING_ALIGN);
 352         if (mz == NULL) {
 353                 if (rte_errno == EEXIST)
 354                         mz = rte_memzone_lookup(vq_name);
 355                 if (mz == NULL) {
 356                         rte_free(vq);
 357                         return -ENOMEM;
 358                 }
 359         }
 360
 361         /*
 362          * Virtio PCI device VIRTIO_PCI_QUEUE_PF register is 32bit,
 363          * and only accepts 32 bit page frame number.
 364          * Check if the allocated physical memory exceeds 16TB.
 365          */
 366         if ((mz->phys_addr + vq->vq_ring_size - 1) >> (VIRTIO_PCI_QUEUE_ADDR_SHIFT + 32)) {
 367                 PMD_INIT_LOG(ERR, "vring address shouldn't be above 16TB!");
 368                 rte_free(vq);
 369                 return -ENOMEM;
 370         }
 371
 372         memset(mz->addr, 0, sizeof(mz->len));
 373         vq->mz = mz;
 374         vq->vq_ring_mem = mz->phys_addr;
 375         vq->vq_ring_virt_mem = mz->addr;
 376         PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%"PRIx64, (uint64_t)mz->phys_addr);
 377         PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%"PRIx64, (uint64_t)(uintptr_t)mz->addr);
 378         vq->virtio_net_hdr_mz  = NULL;
 379         vq->virtio_net_hdr_mem = 0;
 380
 381         if (queue_type == VTNET_TQ) {
 382                 const struct rte_memzone *hdr_mz;
 383                 struct virtio_tx_region *txr;
 384                 unsigned int i;
 385
 386                 /*
 387                  * For each xmit packet, allocate a virtio_net_hdr
 388                  * and indirect ring elements
 389                  */
 390                 snprintf(vq_name, sizeof(vq_name), "port%d_tvq%d_hdrzone",
 391                          dev->data->port_id, queue_idx);
 392                 hdr_mz = rte_memzone_reserve_aligned(vq_name,
 393                                                      vq_size * sizeof(*txr),
 394                                                      socket_id, 0,
 395                                                      RTE_CACHE_LINE_SIZE);
 396                 if (hdr_mz == NULL) {
 397                         if (rte_errno == EEXIST)
 398                                 hdr_mz = rte_memzone_lookup(vq_name);
 399                         if (hdr_mz == NULL) {
 400                                 rte_free(vq);
 401                                 return -ENOMEM;
 402                         }
 403                 }
 404                 vq->virtio_net_hdr_mz = hdr_mz;
 405                 vq->virtio_net_hdr_mem = hdr_mz->phys_addr;
 406
 407                 txr = hdr_mz->addr;
 408                 memset(txr, 0, vq_size * sizeof(*txr));
 409                 for (i = 0; i < vq_size; i++) {
 410                         struct vring_desc *start_dp = txr[i].tx_indir;
 411
 412                         vring_desc_init(start_dp, RTE_DIM(txr[i].tx_indir));
 413
 414                         /* first indirect descriptor is always the tx header */
 415                         start_dp->addr = vq->virtio_net_hdr_mem
 416                                 + i * sizeof(*txr)
 417                                 + offsetof(struct virtio_tx_region, tx_hdr);
 418
 419                         start_dp->len = vq->hw->vtnet_hdr_size;
 420                         start_dp->flags = VRING_DESC_F_NEXT;
 421                 }
 422
 423         } else if (queue_type == VTNET_CQ) {
 424                 /* Allocate a page for control vq command, data and status */
 425                 snprintf(vq_name, sizeof(vq_name), "port%d_cvq_hdrzone",
 426                         dev->data->port_id);
 427                 vq->virtio_net_hdr_mz = rte_memzone_reserve_aligned(vq_name,
 428                         PAGE_SIZE, socket_id, 0, RTE_CACHE_LINE_SIZE);
 429                 if (vq->virtio_net_hdr_mz == NULL) {
 430                         if (rte_errno == EEXIST)
 431                                 vq->virtio_net_hdr_mz =
 432                                         rte_memzone_lookup(vq_name);
 433                         if (vq->virtio_net_hdr_mz == NULL) {
 434                                 rte_free(vq);
 435                                 return -ENOMEM;
 436                         }
 437                 }
 438                 vq->virtio_net_hdr_mem =
 439                         vq->virtio_net_hdr_mz->phys_addr;
 440                 memset(vq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
 441         }
 442
 443         hw->vtpci_ops->setup_queue(hw, vq);
 444
 445         *pvq = vq;
 446         return 0;
 447 }
 448
 449 static int
 450 virtio_dev_cq_queue_setup(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx,
 451                 uint32_t socket_id)
 452 {
 453         struct virtqueue *vq;
 454         int ret;
 455         struct virtio_hw *hw = dev->data->dev_private;
 456
 457         PMD_INIT_FUNC_TRACE();
 458         ret = virtio_dev_queue_setup(dev, VTNET_CQ, VTNET_SQ_CQ_QUEUE_IDX,
 459                         vtpci_queue_idx, 0, socket_id, &vq);
 460         if (ret < 0) {
 461                 PMD_INIT_LOG(ERR, "control vq initialization failed");
 462                 return ret;
 463         }
 464
 465         hw->cvq = vq;
 466         return 0;
 467 }
 468
 469 static void
 470 virtio_free_queues(struct rte_eth_dev *dev)
 471 {
 472         unsigned int i;
 473
 474         for (i = 0; i < dev->data->nb_rx_queues; i++)
 475                 virtio_dev_rx_queue_release(dev->data->rx_queues[i]);
 476
 477         dev->data->nb_rx_queues = 0;
 478
 479         for (i = 0; i < dev->data->nb_tx_queues; i++)
 480                 virtio_dev_tx_queue_release(dev->data->tx_queues[i]);
 481
 482         dev->data->nb_tx_queues = 0;
 483 }
 484
 485 static void
 486 virtio_dev_close(struct rte_eth_dev *dev)
 487 {
 488         struct virtio_hw *hw = dev->data->dev_private;
 489
 490         PMD_INIT_LOG(DEBUG, "virtio_dev_close");
 491
 492         if (hw->started == 1)
 493                 virtio_dev_stop(dev);
 494
 495         /* reset the NIC */
 496         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
 497                 vtpci_irq_config(hw, VIRTIO_MSI_NO_VECTOR);
 498         vtpci_reset(hw);
 499         virtio_dev_free_mbufs(dev);
 500         virtio_free_queues(dev);
 501 }
 502
 503 static void
 504 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
 505 {
 506         struct virtio_hw *hw = dev->data->dev_private;
 507         struct virtio_pmd_ctrl ctrl;
 508         int dlen[1];
 509         int ret;
 510
 511         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 512                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 513                 return;
 514         }
 515
 516         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 517         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 518         ctrl.data[0] = 1;
 519         dlen[0] = 1;
 520
 521         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 522         if (ret)
 523                 PMD_INIT_LOG(ERR, "Failed to enable promisc");
 524 }
 525
 526 static void
 527 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
 528 {
 529         struct virtio_hw *hw = dev->data->dev_private;
 530         struct virtio_pmd_ctrl ctrl;
 531         int dlen[1];
 532         int ret;
 533
 534         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 535                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 536                 return;
 537         }
 538
 539         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 540         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 541         ctrl.data[0] = 0;
 542         dlen[0] = 1;
 543
 544         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 545         if (ret)
 546                 PMD_INIT_LOG(ERR, "Failed to disable promisc");
 547 }
 548
 549 static void
 550 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
 551 {
 552         struct virtio_hw *hw = dev->data->dev_private;
 553         struct virtio_pmd_ctrl ctrl;
 554         int dlen[1];
 555         int ret;
 556
 557         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 558                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 559                 return;
 560         }
 561
 562         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 563         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 564         ctrl.data[0] = 1;
 565         dlen[0] = 1;
 566
 567         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 568         if (ret)
 569                 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
 570 }
 571
 572 static void
 573 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
 574 {
 575         struct virtio_hw *hw = dev->data->dev_private;
 576         struct virtio_pmd_ctrl ctrl;
 577         int dlen[1];
 578         int ret;
 579
 580         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 581                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 582                 return;
 583         }
 584
 585         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 586         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 587         ctrl.data[0] = 0;
 588         dlen[0] = 1;
 589
 590         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 591         if (ret)
 592                 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
 593 }
 594
 595 /*
 596  * dev_ops for virtio, bare necessities for basic operation
 597  */
 598 static const struct eth_dev_ops virtio_eth_dev_ops = {
 599         .dev_configure           = virtio_dev_configure,
 600         .dev_start               = virtio_dev_start,
 601         .dev_stop                = virtio_dev_stop,
 602         .dev_close               = virtio_dev_close,
 603         .promiscuous_enable      = virtio_dev_promiscuous_enable,
 604         .promiscuous_disable     = virtio_dev_promiscuous_disable,
 605         .allmulticast_enable     = virtio_dev_allmulticast_enable,
 606         .allmulticast_disable    = virtio_dev_allmulticast_disable,
 607
 608         .dev_infos_get           = virtio_dev_info_get,
 609         .stats_get               = virtio_dev_stats_get,
 610         .xstats_get              = virtio_dev_xstats_get,
 611         .stats_reset             = virtio_dev_stats_reset,
 612         .xstats_reset            = virtio_dev_stats_reset,
 613         .link_update             = virtio_dev_link_update,
 614         .rx_queue_setup          = virtio_dev_rx_queue_setup,
 615         .rx_queue_release        = virtio_dev_rx_queue_release,
 616         .tx_queue_setup          = virtio_dev_tx_queue_setup,
 617         .tx_queue_release        = virtio_dev_tx_queue_release,
 618         /* collect stats per queue */
 619         .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
 620         .vlan_filter_set         = virtio_vlan_filter_set,
 621         .mac_addr_add            = virtio_mac_addr_add,
 622         .mac_addr_remove         = virtio_mac_addr_remove,
 623         .mac_addr_set            = virtio_mac_addr_set,
 624 };
 625
 626 static inline int
 627 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
 628                                 struct rte_eth_link *link)
 629 {
 630         struct rte_eth_link *dst = link;
 631         struct rte_eth_link *src = &(dev->data->dev_link);
 632
 633         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
 634                         *(uint64_t *)src) == 0)
 635                 return -1;
 636
 637         return 0;
 638 }
 639
 640 /**
 641  * Atomically writes the link status information into global
 642  * structure rte_eth_dev.
 643  *
 644  * @param dev
 645  *   - Pointer to the structure rte_eth_dev to read from.
 646  *   - Pointer to the buffer to be saved with the link status.
 647  *
 648  * @return
 649  *   - On success, zero.
 650  *   - On failure, negative value.
 651  */
 652 static inline int
 653 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
 654                 struct rte_eth_link *link)
 655 {
 656         struct rte_eth_link *dst = &(dev->data->dev_link);
 657         struct rte_eth_link *src = link;
 658
 659         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
 660                                         *(uint64_t *)src) == 0)
 661                 return -1;
 662
 663         return 0;
 664 }
 665
 666 static void
 667 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 668 {
 669         unsigned i;
 670
 671         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 672                 const struct virtqueue *txvq = dev->data->tx_queues[i];
 673                 if (txvq == NULL)
 674                         continue;
 675
 676                 stats->opackets += txvq->packets;
 677                 stats->obytes += txvq->bytes;
 678                 stats->oerrors += txvq->errors;
 679
 680                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 681                         stats->q_opackets[i] = txvq->packets;
 682                         stats->q_obytes[i] = txvq->bytes;
 683                 }
 684         }
 685
 686         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 687                 const struct virtqueue *rxvq = dev->data->rx_queues[i];
 688                 if (rxvq == NULL)
 689                         continue;
 690
 691                 stats->ipackets += rxvq->packets;
 692                 stats->ibytes += rxvq->bytes;
 693                 stats->ierrors += rxvq->errors;
 694
 695                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 696                         stats->q_ipackets[i] = rxvq->packets;
 697                         stats->q_ibytes[i] = rxvq->bytes;
 698                 }
 699         }
 700
 701         stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
 702 }
 703
 704 static int
 705 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstats *xstats,
 706                       unsigned n)
 707 {
 708         unsigned i;
 709         unsigned count = 0;
 710
 711         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_Q_XSTATS +
 712                 dev->data->nb_rx_queues * VIRTIO_NB_Q_XSTATS;
 713
 714         if (n < nstats)
 715                 return nstats;
 716
 717         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 718                 struct virtqueue *rxvq = dev->data->rx_queues[i];
 719
 720                 if (rxvq == NULL)
 721                         continue;
 722
 723                 unsigned t;
 724
 725                 for (t = 0; t < VIRTIO_NB_Q_XSTATS; t++) {
 726                         snprintf(xstats[count].name, sizeof(xstats[count].name),
 727                                  "rx_q%u_%s", i,
 728                                  rte_virtio_q_stat_strings[t].name);
 729                         xstats[count].value = *(uint64_t *)(((char *)rxvq) +
 730                                 rte_virtio_q_stat_strings[t].offset);
 731                         count++;
 732                 }
 733         }
 734
 735         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 736                 struct virtqueue *txvq = dev->data->tx_queues[i];
 737
 738                 if (txvq == NULL)
 739                         continue;
 740
 741                 unsigned t;
 742
 743                 for (t = 0; t < VIRTIO_NB_Q_XSTATS; t++) {
 744                         snprintf(xstats[count].name, sizeof(xstats[count].name),
 745                                  "tx_q%u_%s", i,
 746                                  rte_virtio_q_stat_strings[t].name);
 747                         xstats[count].value = *(uint64_t *)(((char *)txvq) +
 748                                 rte_virtio_q_stat_strings[t].offset);
 749                         count++;
 750                 }
 751         }
 752
 753         return count;
 754 }
 755
 756 static void
 757 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 758 {
 759         virtio_update_stats(dev, stats);
 760 }
 761
 762 static void
 763 virtio_dev_stats_reset(struct rte_eth_dev *dev)
 764 {
 765         unsigned int i;
 766
 767         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 768                 struct virtqueue *txvq = dev->data->tx_queues[i];
 769                 if (txvq == NULL)
 770                         continue;
 771
 772                 txvq->packets = 0;
 773                 txvq->bytes = 0;
 774                 txvq->errors = 0;
 775                 txvq->multicast = 0;
 776                 txvq->broadcast = 0;
 777                 memset(txvq->size_bins, 0, sizeof(txvq->size_bins[0]) * 8);
 778         }
 779
 780         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 781                 struct virtqueue *rxvq = dev->data->rx_queues[i];
 782                 if (rxvq == NULL)
 783                         continue;
 784
 785                 rxvq->packets = 0;
 786                 rxvq->bytes = 0;
 787                 rxvq->errors = 0;
 788                 rxvq->multicast = 0;
 789                 rxvq->broadcast = 0;
 790                 memset(rxvq->size_bins, 0, sizeof(rxvq->size_bins[0]) * 8);
 791         }
 792 }
 793
 794 static void
 795 virtio_set_hwaddr(struct virtio_hw *hw)
 796 {
 797         vtpci_write_dev_config(hw,
 798                         offsetof(struct virtio_net_config, mac),
 799                         &hw->mac_addr, ETHER_ADDR_LEN);
 800 }
 801
 802 static void
 803 virtio_get_hwaddr(struct virtio_hw *hw)
 804 {
 805         if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
 806                 vtpci_read_dev_config(hw,
 807                         offsetof(struct virtio_net_config, mac),
 808                         &hw->mac_addr, ETHER_ADDR_LEN);
 809         } else {
 810                 eth_random_addr(&hw->mac_addr[0]);
 811                 virtio_set_hwaddr(hw);
 812         }
 813 }
 814
 815 static void
 816 virtio_mac_table_set(struct virtio_hw *hw,
 817                      const struct virtio_net_ctrl_mac *uc,
 818                      const struct virtio_net_ctrl_mac *mc)
 819 {
 820         struct virtio_pmd_ctrl ctrl;
 821         int err, len[2];
 822
 823         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
 824                 PMD_DRV_LOG(INFO, "host does not support mac table");
 825                 return;
 826         }
 827
 828         ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
 829         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
 830
 831         len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
 832         memcpy(ctrl.data, uc, len[0]);
 833
 834         len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
 835         memcpy(ctrl.data + len[0], mc, len[1]);
 836
 837         err = virtio_send_command(hw->cvq, &ctrl, len, 2);
 838         if (err != 0)
 839                 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
 840 }
 841
 842 static void
 843 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
 844                     uint32_t index, uint32_t vmdq __rte_unused)
 845 {
 846         struct virtio_hw *hw = dev->data->dev_private;
 847         const struct ether_addr *addrs = dev->data->mac_addrs;
 848         unsigned int i;
 849         struct virtio_net_ctrl_mac *uc, *mc;
 850
 851         if (index >= VIRTIO_MAX_MAC_ADDRS) {
 852                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
 853                 return;
 854         }
 855
 856         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
 857         uc->entries = 0;
 858         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
 859         mc->entries = 0;
 860
 861         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
 862                 const struct ether_addr *addr
 863                         = (i == index) ? mac_addr : addrs + i;
 864                 struct virtio_net_ctrl_mac *tbl
 865                         = is_multicast_ether_addr(addr) ? mc : uc;
 866
 867                 memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
 868         }
 869
 870         virtio_mac_table_set(hw, uc, mc);
 871 }
 872
 873 static void
 874 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
 875 {
 876         struct virtio_hw *hw = dev->data->dev_private;
 877         struct ether_addr *addrs = dev->data->mac_addrs;
 878         struct virtio_net_ctrl_mac *uc, *mc;
 879         unsigned int i;
 880
 881         if (index >= VIRTIO_MAX_MAC_ADDRS) {
 882                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
 883                 return;
 884         }
 885
 886         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
 887         uc->entries = 0;
 888         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
 889         mc->entries = 0;
 890
 891         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
 892                 struct virtio_net_ctrl_mac *tbl;
 893
 894                 if (i == index || is_zero_ether_addr(addrs + i))
 895                         continue;
 896
 897                 tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
 898                 memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
 899         }
 900
 901         virtio_mac_table_set(hw, uc, mc);
 902 }
 903
 904 static void
 905 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 906 {
 907         struct virtio_hw *hw = dev->data->dev_private;
 908
 909         memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
 910
 911         /* Use atomic update if available */
 912         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
 913                 struct virtio_pmd_ctrl ctrl;
 914                 int len = ETHER_ADDR_LEN;
 915
 916                 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
 917                 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
 918
 919                 memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
 920                 virtio_send_command(hw->cvq, &ctrl, &len, 1);
 921         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
 922                 virtio_set_hwaddr(hw);
 923 }
 924
 925 static int
 926 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
 927 {
 928         struct virtio_hw *hw = dev->data->dev_private;
 929         struct virtio_pmd_ctrl ctrl;
 930         int len;
 931
 932         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
 933                 return -ENOTSUP;
 934
 935         ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
 936         ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
 937         memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
 938         len = sizeof(vlan_id);
 939
 940         return virtio_send_command(hw->cvq, &ctrl, &len, 1);
 941 }
 942
 943 static int
 944 virtio_negotiate_features(struct virtio_hw *hw)
 945 {
 946         uint64_t host_features;
 947
 948         /* Prepare guest_features: feature that driver wants to support */
 949         hw->guest_features = VIRTIO_PMD_GUEST_FEATURES;
 950         PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
 951                 hw->guest_features);
 952
 953         /* Read device(host) feature bits */
 954         host_features = hw->vtpci_ops->get_features(hw);
 955         PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
 956                 host_features);
 957
 958         /*
 959          * Negotiate features: Subset of device feature bits are written back
 960          * guest feature bits.
 961          */
 962         hw->guest_features = vtpci_negotiate_features(hw, host_features);
 963         PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
 964                 hw->guest_features);
 965
 966         if (hw->modern) {
 967                 if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
 968                         PMD_INIT_LOG(ERR,
 969                                 "VIRTIO_F_VERSION_1 features is not enabled.");
 970                         return -1;
 971                 }
 972                 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
 973                 if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
 974                         PMD_INIT_LOG(ERR,
 975                                 "failed to set FEATURES_OK status!");
 976                         return -1;
 977                 }
 978         }
 979
 980         return 0;
 981 }
 982
 983 /*
 984  * Process Virtio Config changed interrupt and call the callback
 985  * if link state changed.
 986  */
 987 static void
 988 virtio_interrupt_handler(__rte_unused struct rte_intr_handle *handle,
 989                          void *param)
 990 {
 991         struct rte_eth_dev *dev = param;
 992         struct virtio_hw *hw = dev->data->dev_private;
 993         uint8_t isr;
 994
 995         /* Read interrupt status which clears interrupt */
 996         isr = vtpci_isr(hw);
 997         PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
 998
 999         if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0)
1000                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1001
1002         if (isr & VIRTIO_PCI_ISR_CONFIG) {
1003                 if (virtio_dev_link_update(dev, 0) == 0)
1004                         _rte_eth_dev_callback_process(dev,
1005                                                       RTE_ETH_EVENT_INTR_LSC);
1006         }
1007
1008 }
1009
1010 static void
1011 rx_func_get(struct rte_eth_dev *eth_dev)
1012 {
1013         struct virtio_hw *hw = eth_dev->data->dev_private;
1014         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF))
1015                 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1016         else
1017                 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1018 }
1019
1020 /*
1021  * This function is based on probe() function in virtio_pci.c
1022  * It returns 0 on success.
1023  */
1024 static int
1025 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1026 {
1027         struct virtio_hw *hw = eth_dev->data->dev_private;
1028         struct virtio_net_config *config;
1029         struct virtio_net_config local_config;
1030         struct rte_pci_device *pci_dev;
1031         uint32_t dev_flags = RTE_ETH_DEV_DETACHABLE;
1032         int ret;
1033
1034         RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr_mrg_rxbuf));
1035
1036         eth_dev->dev_ops = &virtio_eth_dev_ops;
1037         eth_dev->tx_pkt_burst = &virtio_xmit_pkts;
1038
1039         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1040                 rx_func_get(eth_dev);
1041                 return 0;
1042         }
1043
1044         /* Allocate memory for storing MAC addresses */
1045         eth_dev->data->mac_addrs = rte_zmalloc("virtio", VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN, 0);
1046         if (eth_dev->data->mac_addrs == NULL) {
1047                 PMD_INIT_LOG(ERR,
1048                         "Failed to allocate %d bytes needed to store MAC addresses",
1049                         VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN);
1050                 return -ENOMEM;
1051         }
1052
1053         pci_dev = eth_dev->pci_dev;
1054
1055         ret = vtpci_init(pci_dev, hw, &dev_flags);
1056         if (ret)
1057                 return ret;
1058
1059         /* Reset the device although not necessary at startup */
1060         vtpci_reset(hw);
1061
1062         /* Tell the host we've noticed this device. */
1063         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1064
1065         /* Tell the host we've known how to drive the device. */
1066         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1067         if (virtio_negotiate_features(hw) < 0)
1068                 return -1;
1069
1070         /* If host does not support status then disable LSC */
1071         if (!vtpci_with_feature(hw, VIRTIO_NET_F_STATUS))
1072                 dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1073
1074         rte_eth_copy_pci_info(eth_dev, pci_dev);
1075         eth_dev->data->dev_flags = dev_flags;
1076
1077         rx_func_get(eth_dev);
1078
1079         /* Setting up rx_header size for the device */
1080         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1081             vtpci_with_feature(hw, VIRTIO_F_VERSION_1))
1082                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1083         else
1084                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1085
1086         /* Copy the permanent MAC address to: virtio_hw */
1087         virtio_get_hwaddr(hw);
1088         ether_addr_copy((struct ether_addr *) hw->mac_addr,
1089                         &eth_dev->data->mac_addrs[0]);
1090         PMD_INIT_LOG(DEBUG,
1091                      "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1092                      hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1093                      hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1094
1095         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1096                 config = &local_config;
1097
1098                 vtpci_read_dev_config(hw,
1099                         offsetof(struct virtio_net_config, mac),
1100                         &config->mac, sizeof(config->mac));
1101
1102                 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1103                         vtpci_read_dev_config(hw,
1104                                 offsetof(struct virtio_net_config, status),
1105                                 &config->status, sizeof(config->status));
1106                 } else {
1107                         PMD_INIT_LOG(DEBUG,
1108                                      "VIRTIO_NET_F_STATUS is not supported");
1109                         config->status = 0;
1110                 }
1111
1112                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1113                         vtpci_read_dev_config(hw,
1114                                 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1115                                 &config->max_virtqueue_pairs,
1116                                 sizeof(config->max_virtqueue_pairs));
1117                 } else {
1118                         PMD_INIT_LOG(DEBUG,
1119                                      "VIRTIO_NET_F_MQ is not supported");
1120                         config->max_virtqueue_pairs = 1;
1121                 }
1122
1123                 hw->max_rx_queues =
1124                         (VIRTIO_MAX_RX_QUEUES < config->max_virtqueue_pairs) ?
1125                         VIRTIO_MAX_RX_QUEUES : config->max_virtqueue_pairs;
1126                 hw->max_tx_queues =
1127                         (VIRTIO_MAX_TX_QUEUES < config->max_virtqueue_pairs) ?
1128                         VIRTIO_MAX_TX_QUEUES : config->max_virtqueue_pairs;
1129
1130                 virtio_dev_cq_queue_setup(eth_dev,
1131                                         config->max_virtqueue_pairs * 2,
1132                                         SOCKET_ID_ANY);
1133
1134                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1135                                 config->max_virtqueue_pairs);
1136                 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1137                 PMD_INIT_LOG(DEBUG,
1138                                 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1139                                 config->mac[0], config->mac[1],
1140                                 config->mac[2], config->mac[3],
1141                                 config->mac[4], config->mac[5]);
1142         } else {
1143                 hw->max_rx_queues = 1;
1144                 hw->max_tx_queues = 1;
1145         }
1146
1147         PMD_INIT_LOG(DEBUG, "hw->max_rx_queues=%d   hw->max_tx_queues=%d",
1148                         hw->max_rx_queues, hw->max_tx_queues);
1149         PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1150                         eth_dev->data->port_id, pci_dev->id.vendor_id,
1151                         pci_dev->id.device_id);
1152
1153         /* Setup interrupt callback  */
1154         if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1155                 rte_intr_callback_register(&pci_dev->intr_handle,
1156                                    virtio_interrupt_handler, eth_dev);
1157
1158         virtio_dev_cq_start(eth_dev);
1159
1160         return 0;
1161 }
1162
1163 static int
1164 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1165 {
1166         struct rte_pci_device *pci_dev;
1167         struct virtio_hw *hw = eth_dev->data->dev_private;
1168
1169         PMD_INIT_FUNC_TRACE();
1170
1171         if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1172                 return -EPERM;
1173
1174         /* Close it anyway since there's no way to know if closed */
1175         virtio_dev_close(eth_dev);
1176
1177         pci_dev = eth_dev->pci_dev;
1178
1179         eth_dev->dev_ops = NULL;
1180         eth_dev->tx_pkt_burst = NULL;
1181         eth_dev->rx_pkt_burst = NULL;
1182
1183         virtio_dev_queue_release(hw->cvq);
1184
1185         rte_free(eth_dev->data->mac_addrs);
1186         eth_dev->data->mac_addrs = NULL;
1187
1188         /* reset interrupt callback  */
1189         if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1190                 rte_intr_callback_unregister(&pci_dev->intr_handle,
1191                                                 virtio_interrupt_handler,
1192                                                 eth_dev);
1193         rte_eal_pci_unmap_device(pci_dev);
1194
1195         PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1196
1197         return 0;
1198 }
1199
1200 static struct eth_driver rte_virtio_pmd = {
1201         .pci_drv = {
1202                 .name = "rte_virtio_pmd",
1203                 .id_table = pci_id_virtio_map,
1204                 .drv_flags = RTE_PCI_DRV_DETACHABLE,
1205         },
1206         .eth_dev_init = eth_virtio_dev_init,
1207         .eth_dev_uninit = eth_virtio_dev_uninit,
1208         .dev_private_size = sizeof(struct virtio_hw),
1209 };
1210
1211 /*
1212  * Driver initialization routine.
1213  * Invoked once at EAL init time.
1214  * Register itself as the [Poll Mode] Driver of PCI virtio devices.
1215  * Returns 0 on success.
1216  */
1217 static int
1218 rte_virtio_pmd_init(const char *name __rte_unused,
1219                     const char *param __rte_unused)
1220 {
1221         if (rte_eal_iopl_init() != 0) {
1222                 PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1223                 return -1;
1224         }
1225
1226         rte_eth_driver_register(&rte_virtio_pmd);
1227         return 0;
1228 }
1229
1230 /*
1231  * Configure virtio device
1232  * It returns 0 on success.
1233  */
1234 static int
1235 virtio_dev_configure(struct rte_eth_dev *dev)
1236 {
1237         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1238         struct virtio_hw *hw = dev->data->dev_private;
1239
1240         PMD_INIT_LOG(DEBUG, "configure");
1241
1242         if (rxmode->hw_ip_checksum) {
1243                 PMD_DRV_LOG(ERR, "HW IP checksum not supported");
1244                 return -EINVAL;
1245         }
1246
1247         hw->vlan_strip = rxmode->hw_vlan_strip;
1248
1249         if (rxmode->hw_vlan_filter
1250             && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1251                 PMD_DRV_LOG(NOTICE,
1252                             "vlan filtering not available on this host");
1253                 return -ENOTSUP;
1254         }
1255
1256         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1257                 if (vtpci_irq_config(hw, 0) == VIRTIO_MSI_NO_VECTOR) {
1258                         PMD_DRV_LOG(ERR, "failed to set config vector");
1259                         return -EBUSY;
1260                 }
1261
1262         return 0;
1263 }
1264
1265
1266 static int
1267 virtio_dev_start(struct rte_eth_dev *dev)
1268 {
1269         uint16_t nb_queues, i;
1270         struct virtio_hw *hw = dev->data->dev_private;
1271
1272         /* check if lsc interrupt feature is enabled */
1273         if (dev->data->dev_conf.intr_conf.lsc) {
1274                 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
1275                         PMD_DRV_LOG(ERR, "link status not supported by host");
1276                         return -ENOTSUP;
1277                 }
1278
1279                 if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0) {
1280                         PMD_DRV_LOG(ERR, "interrupt enable failed");
1281                         return -EIO;
1282                 }
1283         }
1284
1285         /* Initialize Link state */
1286         virtio_dev_link_update(dev, 0);
1287
1288         /* On restart after stop do not touch queues */
1289         if (hw->started)
1290                 return 0;
1291
1292         /* Do final configuration before rx/tx engine starts */
1293         virtio_dev_rxtx_start(dev);
1294         vtpci_reinit_complete(hw);
1295
1296         hw->started = 1;
1297
1298         /*Notify the backend
1299          *Otherwise the tap backend might already stop its queue due to fullness.
1300          *vhost backend will have no chance to be waked up
1301          */
1302         nb_queues = dev->data->nb_rx_queues;
1303         if (nb_queues > 1) {
1304                 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1305                         return -EINVAL;
1306         }
1307
1308         PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1309
1310         for (i = 0; i < nb_queues; i++)
1311                 virtqueue_notify(dev->data->rx_queues[i]);
1312
1313         PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1314
1315         for (i = 0; i < dev->data->nb_rx_queues; i++)
1316                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1317
1318         for (i = 0; i < dev->data->nb_tx_queues; i++)
1319                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1320
1321         return 0;
1322 }
1323
1324 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1325 {
1326         struct rte_mbuf *buf;
1327         int i, mbuf_num = 0;
1328
1329         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1330                 PMD_INIT_LOG(DEBUG,
1331                              "Before freeing rxq[%d] used and unused buf", i);
1332                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1333
1334                 PMD_INIT_LOG(DEBUG, "rx_queues[%d]=%p",
1335                                 i, dev->data->rx_queues[i]);
1336                 while ((buf = (struct rte_mbuf *)virtqueue_detatch_unused(
1337                                         dev->data->rx_queues[i])) != NULL) {
1338                         rte_pktmbuf_free(buf);
1339                         mbuf_num++;
1340                 }
1341
1342                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1343                 PMD_INIT_LOG(DEBUG,
1344                              "After freeing rxq[%d] used and unused buf", i);
1345                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->rx_queues[i]);
1346         }
1347
1348         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1349                 PMD_INIT_LOG(DEBUG,
1350                              "Before freeing txq[%d] used and unused bufs",
1351                              i);
1352                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1353
1354                 mbuf_num = 0;
1355                 while ((buf = (struct rte_mbuf *)virtqueue_detatch_unused(
1356                                         dev->data->tx_queues[i])) != NULL) {
1357                         rte_pktmbuf_free(buf);
1358
1359                         mbuf_num++;
1360                 }
1361
1362                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1363                 PMD_INIT_LOG(DEBUG,
1364                              "After freeing txq[%d] used and unused buf", i);
1365                 VIRTQUEUE_DUMP((struct virtqueue *)dev->data->tx_queues[i]);
1366         }
1367 }
1368
1369 /*
1370  * Stop device: disable interrupt and mark link down
1371  */
1372 static void
1373 virtio_dev_stop(struct rte_eth_dev *dev)
1374 {
1375         struct rte_eth_link link;
1376         struct virtio_hw *hw = dev->data->dev_private;
1377
1378         PMD_INIT_LOG(DEBUG, "stop");
1379
1380         hw->started = 0;
1381
1382         if (dev->data->dev_conf.intr_conf.lsc)
1383                 rte_intr_disable(&dev->pci_dev->intr_handle);
1384
1385         memset(&link, 0, sizeof(link));
1386         virtio_dev_atomic_write_link_status(dev, &link);
1387 }
1388
1389 static int
1390 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
1391 {
1392         struct rte_eth_link link, old;
1393         uint16_t status;
1394         struct virtio_hw *hw = dev->data->dev_private;
1395         memset(&link, 0, sizeof(link));
1396         virtio_dev_atomic_read_link_status(dev, &link);
1397         old = link;
1398         link.link_duplex = ETH_LINK_FULL_DUPLEX;
1399         link.link_speed  = SPEED_10G;
1400
1401         if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1402                 PMD_INIT_LOG(DEBUG, "Get link status from hw");
1403                 vtpci_read_dev_config(hw,
1404                                 offsetof(struct virtio_net_config, status),
1405                                 &status, sizeof(status));
1406                 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
1407                         link.link_status = ETH_LINK_DOWN;
1408                         PMD_INIT_LOG(DEBUG, "Port %d is down",
1409                                      dev->data->port_id);
1410                 } else {
1411                         link.link_status = ETH_LINK_UP;
1412                         PMD_INIT_LOG(DEBUG, "Port %d is up",
1413                                      dev->data->port_id);
1414                 }
1415         } else {
1416                 link.link_status = ETH_LINK_UP;
1417         }
1418         virtio_dev_atomic_write_link_status(dev, &link);
1419
1420         return (old.link_status == link.link_status) ? -1 : 0;
1421 }
1422
1423 static void
1424 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1425 {
1426         struct virtio_hw *hw = dev->data->dev_private;
1427
1428         dev_info->driver_name = dev->driver->pci_drv.name;
1429         dev_info->max_rx_queues = (uint16_t)hw->max_rx_queues;
1430         dev_info->max_tx_queues = (uint16_t)hw->max_tx_queues;
1431         dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
1432         dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
1433         dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
1434         dev_info->default_txconf = (struct rte_eth_txconf) {
1435                 .txq_flags = ETH_TXQ_FLAGS_NOOFFLOADS
1436         };
1437 }
1438
1439 /*
1440  * It enables testpmd to collect per queue stats.
1441  */
1442 static int
1443 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
1444 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
1445 __rte_unused uint8_t is_rx)
1446 {
1447         return 0;
1448 }
1449
1450 static struct rte_driver rte_virtio_driver = {
1451         .type = PMD_PDEV,
1452         .init = rte_virtio_pmd_init,
1453 };
1454
1455 PMD_REGISTER_DRIVER(rte_virtio_driver);