drivers/net/virtio/virtio_ethdev.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2016 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_xstat *xstats, unsigned n);
  83 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
  84                                        struct rte_eth_xstat_name *xstats_names,
  85                                        unsigned limit);
  86 static void virtio_dev_stats_reset(struct rte_eth_dev *dev);
  87 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
  88 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
  89                                 uint16_t vlan_id, int on);
  90 static void virtio_mac_addr_add(struct rte_eth_dev *dev,
  91                                 struct ether_addr *mac_addr,
  92                                 uint32_t index, uint32_t vmdq __rte_unused);
  93 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
  94 static void virtio_mac_addr_set(struct rte_eth_dev *dev,
  95                                 struct ether_addr *mac_addr);
  96
  97 static int virtio_dev_queue_stats_mapping_set(
  98         __rte_unused struct rte_eth_dev *eth_dev,
  99         __rte_unused uint16_t queue_id,
 100         __rte_unused uint8_t stat_idx,
 101         __rte_unused uint8_t is_rx);
 102
 103 /*
 104  * The set of PCI devices this driver supports
 105  */
 106 static const struct rte_pci_id pci_id_virtio_map[] = {
 107
 108 #define RTE_PCI_DEV_ID_DECL_VIRTIO(vend, dev) {RTE_PCI_DEVICE(vend, dev)},
 109 #include "rte_pci_dev_ids.h"
 110
 111 { .vendor_id = 0, /* sentinel */ },
 112 };
 113
 114 struct rte_virtio_xstats_name_off {
 115         char name[RTE_ETH_XSTATS_NAME_SIZE];
 116         unsigned offset;
 117 };
 118
 119 /* [rt]x_qX_ is prepended to the name string here */
 120 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
 121         {"good_packets",           offsetof(struct virtnet_rx, stats.packets)},
 122         {"good_bytes",             offsetof(struct virtnet_rx, stats.bytes)},
 123         {"errors",                 offsetof(struct virtnet_rx, stats.errors)},
 124         {"multicast_packets",      offsetof(struct virtnet_rx, stats.multicast)},
 125         {"broadcast_packets",      offsetof(struct virtnet_rx, stats.broadcast)},
 126         {"undersize_packets",      offsetof(struct virtnet_rx, stats.size_bins[0])},
 127         {"size_64_packets",        offsetof(struct virtnet_rx, stats.size_bins[1])},
 128         {"size_65_127_packets",    offsetof(struct virtnet_rx, stats.size_bins[2])},
 129         {"size_128_255_packets",   offsetof(struct virtnet_rx, stats.size_bins[3])},
 130         {"size_256_511_packets",   offsetof(struct virtnet_rx, stats.size_bins[4])},
 131         {"size_512_1023_packets",  offsetof(struct virtnet_rx, stats.size_bins[5])},
 132         {"size_1024_1517_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
 133         {"size_1518_max_packets",  offsetof(struct virtnet_rx, stats.size_bins[7])},
 134 };
 135
 136 /* [rt]x_qX_ is prepended to the name string here */
 137 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
 138         {"good_packets",           offsetof(struct virtnet_tx, stats.packets)},
 139         {"good_bytes",             offsetof(struct virtnet_tx, stats.bytes)},
 140         {"errors",                 offsetof(struct virtnet_tx, stats.errors)},
 141         {"multicast_packets",      offsetof(struct virtnet_tx, stats.multicast)},
 142         {"broadcast_packets",      offsetof(struct virtnet_tx, stats.broadcast)},
 143         {"undersize_packets",      offsetof(struct virtnet_tx, stats.size_bins[0])},
 144         {"size_64_packets",        offsetof(struct virtnet_tx, stats.size_bins[1])},
 145         {"size_65_127_packets",    offsetof(struct virtnet_tx, stats.size_bins[2])},
 146         {"size_128_255_packets",   offsetof(struct virtnet_tx, stats.size_bins[3])},
 147         {"size_256_511_packets",   offsetof(struct virtnet_tx, stats.size_bins[4])},
 148         {"size_512_1023_packets",  offsetof(struct virtnet_tx, stats.size_bins[5])},
 149         {"size_1024_1517_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
 150         {"size_1518_max_packets",  offsetof(struct virtnet_tx, stats.size_bins[7])},
 151 };
 152
 153 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
 154                             sizeof(rte_virtio_rxq_stat_strings[0]))
 155 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
 156                             sizeof(rte_virtio_txq_stat_strings[0]))
 157
 158 static int
 159 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
 160                 int *dlen, int pkt_num)
 161 {
 162         uint32_t head, i;
 163         int k, sum = 0;
 164         virtio_net_ctrl_ack status = ~0;
 165         struct virtio_pmd_ctrl result;
 166         struct virtqueue *vq;
 167
 168         ctrl->status = status;
 169
 170         if (!cvq && !cvq->vq) {
 171                 PMD_INIT_LOG(ERR, "Control queue is not supported.");
 172                 return -1;
 173         }
 174         vq = cvq->vq;
 175         head = vq->vq_desc_head_idx;
 176
 177         PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
 178                 "vq->hw->cvq = %p vq = %p",
 179                 vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
 180
 181         if ((vq->vq_free_cnt < ((uint32_t)pkt_num + 2)) || (pkt_num < 1))
 182                 return -1;
 183
 184         memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
 185                 sizeof(struct virtio_pmd_ctrl));
 186
 187         /*
 188          * Format is enforced in qemu code:
 189          * One TX packet for header;
 190          * At least one TX packet per argument;
 191          * One RX packet for ACK.
 192          */
 193         vq->vq_ring.desc[head].flags = VRING_DESC_F_NEXT;
 194         vq->vq_ring.desc[head].addr = cvq->virtio_net_hdr_mem;
 195         vq->vq_ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
 196         vq->vq_free_cnt--;
 197         i = vq->vq_ring.desc[head].next;
 198
 199         for (k = 0; k < pkt_num; k++) {
 200                 vq->vq_ring.desc[i].flags = VRING_DESC_F_NEXT;
 201                 vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
 202                         + sizeof(struct virtio_net_ctrl_hdr)
 203                         + sizeof(ctrl->status) + sizeof(uint8_t)*sum;
 204                 vq->vq_ring.desc[i].len = dlen[k];
 205                 sum += dlen[k];
 206                 vq->vq_free_cnt--;
 207                 i = vq->vq_ring.desc[i].next;
 208         }
 209
 210         vq->vq_ring.desc[i].flags = VRING_DESC_F_WRITE;
 211         vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
 212                         + sizeof(struct virtio_net_ctrl_hdr);
 213         vq->vq_ring.desc[i].len = sizeof(ctrl->status);
 214         vq->vq_free_cnt--;
 215
 216         vq->vq_desc_head_idx = vq->vq_ring.desc[i].next;
 217
 218         vq_update_avail_ring(vq, head);
 219         vq_update_avail_idx(vq);
 220
 221         PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
 222
 223         virtqueue_notify(vq);
 224
 225         rte_rmb();
 226         while (vq->vq_used_cons_idx == vq->vq_ring.used->idx) {
 227                 rte_rmb();
 228                 usleep(100);
 229         }
 230
 231         while (vq->vq_used_cons_idx != vq->vq_ring.used->idx) {
 232                 uint32_t idx, desc_idx, used_idx;
 233                 struct vring_used_elem *uep;
 234
 235                 used_idx = (uint32_t)(vq->vq_used_cons_idx
 236                                 & (vq->vq_nentries - 1));
 237                 uep = &vq->vq_ring.used->ring[used_idx];
 238                 idx = (uint32_t) uep->id;
 239                 desc_idx = idx;
 240
 241                 while (vq->vq_ring.desc[desc_idx].flags & VRING_DESC_F_NEXT) {
 242                         desc_idx = vq->vq_ring.desc[desc_idx].next;
 243                         vq->vq_free_cnt++;
 244                 }
 245
 246                 vq->vq_ring.desc[desc_idx].next = vq->vq_desc_head_idx;
 247                 vq->vq_desc_head_idx = idx;
 248
 249                 vq->vq_used_cons_idx++;
 250                 vq->vq_free_cnt++;
 251         }
 252
 253         PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
 254                         vq->vq_free_cnt, vq->vq_desc_head_idx);
 255
 256         memcpy(&result, cvq->virtio_net_hdr_mz->addr,
 257                         sizeof(struct virtio_pmd_ctrl));
 258
 259         return result.status;
 260 }
 261
 262 static int
 263 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
 264 {
 265         struct virtio_hw *hw = dev->data->dev_private;
 266         struct virtio_pmd_ctrl ctrl;
 267         int dlen[1];
 268         int ret;
 269
 270         ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
 271         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
 272         memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
 273
 274         dlen[0] = sizeof(uint16_t);
 275
 276         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 277         if (ret) {
 278                 PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
 279                           "failed, this is too late now...");
 280                 return -EINVAL;
 281         }
 282
 283         return 0;
 284 }
 285
 286 void
 287 virtio_dev_queue_release(struct virtqueue *vq)
 288 {
 289         struct virtio_hw *hw;
 290
 291         if (vq) {
 292                 hw = vq->hw;
 293                 if (vq->configured)
 294                         hw->vtpci_ops->del_queue(hw, vq);
 295
 296                 rte_free(vq->sw_ring);
 297                 rte_free(vq);
 298         }
 299 }
 300
 301 int virtio_dev_queue_setup(struct rte_eth_dev *dev,
 302                         int queue_type,
 303                         uint16_t queue_idx,
 304                         uint16_t vtpci_queue_idx,
 305                         uint16_t nb_desc,
 306                         unsigned int socket_id,
 307                         void **pvq)
 308 {
 309         char vq_name[VIRTQUEUE_MAX_NAME_SZ];
 310         char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
 311         const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
 312         unsigned int vq_size, size;
 313         struct virtio_hw *hw = dev->data->dev_private;
 314         struct virtnet_rx *rxvq = NULL;
 315         struct virtnet_tx *txvq = NULL;
 316         struct virtnet_ctl *cvq = NULL;
 317         struct virtqueue *vq;
 318         const char *queue_names[] = {"rvq", "txq", "cvq"};
 319         size_t sz_vq, sz_q = 0, sz_hdr_mz = 0;
 320         void *sw_ring = NULL;
 321         int ret;
 322
 323         PMD_INIT_LOG(DEBUG, "setting up queue: %u", vtpci_queue_idx);
 324
 325         /*
 326          * Read the virtqueue size from the Queue Size field
 327          * Always power of 2 and if 0 virtqueue does not exist
 328          */
 329         vq_size = hw->vtpci_ops->get_queue_num(hw, vtpci_queue_idx);
 330         PMD_INIT_LOG(DEBUG, "vq_size: %u nb_desc:%u", vq_size, nb_desc);
 331         if (vq_size == 0) {
 332                 PMD_INIT_LOG(ERR, "virtqueue does not exist");
 333                 return -EINVAL;
 334         }
 335
 336         if (!rte_is_power_of_2(vq_size)) {
 337                 PMD_INIT_LOG(ERR, "virtqueue size is not powerof 2");
 338                 return -EINVAL;
 339         }
 340
 341         snprintf(vq_name, sizeof(vq_name), "port%d_%s%d",
 342                  dev->data->port_id, queue_names[queue_type], queue_idx);
 343
 344         sz_vq = RTE_ALIGN_CEIL(sizeof(*vq) +
 345                                 vq_size * sizeof(struct vq_desc_extra),
 346                                 RTE_CACHE_LINE_SIZE);
 347         if (queue_type == VTNET_RQ) {
 348                 sz_q = sz_vq + sizeof(*rxvq);
 349         } else if (queue_type == VTNET_TQ) {
 350                 sz_q = sz_vq + sizeof(*txvq);
 351                 /*
 352                  * For each xmit packet, allocate a virtio_net_hdr
 353                  * and indirect ring elements
 354                  */
 355                 sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
 356         } else if (queue_type == VTNET_CQ) {
 357                 sz_q = sz_vq + sizeof(*cvq);
 358                 /* Allocate a page for control vq command, data and status */
 359                 sz_hdr_mz = PAGE_SIZE;
 360         }
 361
 362         vq = rte_zmalloc_socket(vq_name, sz_q, RTE_CACHE_LINE_SIZE, socket_id);
 363         if (vq == NULL) {
 364                 PMD_INIT_LOG(ERR, "can not allocate vq");
 365                 return -ENOMEM;
 366         }
 367         vq->hw = hw;
 368         vq->vq_queue_index = vtpci_queue_idx;
 369         vq->vq_nentries = vq_size;
 370
 371         if (nb_desc == 0 || nb_desc > vq_size)
 372                 nb_desc = vq_size;
 373         vq->vq_free_cnt = nb_desc;
 374
 375         /*
 376          * Reserve a memzone for vring elements
 377          */
 378         size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
 379         vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
 380         PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
 381                      size, vq->vq_ring_size);
 382
 383         mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size, socket_id,
 384                                          0, VIRTIO_PCI_VRING_ALIGN);
 385         if (mz == NULL) {
 386                 if (rte_errno == EEXIST)
 387                         mz = rte_memzone_lookup(vq_name);
 388                 if (mz == NULL) {
 389                         ret = -ENOMEM;
 390                         goto fail_q_alloc;
 391                 }
 392         }
 393
 394         memset(mz->addr, 0, sizeof(mz->len));
 395
 396         vq->vq_ring_mem = mz->phys_addr;
 397         vq->vq_ring_virt_mem = mz->addr;
 398         PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%" PRIx64,
 399                      (uint64_t)mz->phys_addr);
 400         PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
 401                      (uint64_t)(uintptr_t)mz->addr);
 402
 403         if (sz_hdr_mz) {
 404                 snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_%s%d_hdr",
 405                          dev->data->port_id, queue_names[queue_type],
 406                          queue_idx);
 407                 hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
 408                                                      socket_id, 0,
 409                                                      RTE_CACHE_LINE_SIZE);
 410                 if (hdr_mz == NULL) {
 411                         if (rte_errno == EEXIST)
 412                                 hdr_mz = rte_memzone_lookup(vq_hdr_name);
 413                         if (hdr_mz == NULL) {
 414                                 ret = -ENOMEM;
 415                                 goto fail_q_alloc;
 416                         }
 417                 }
 418         }
 419
 420         if (queue_type == VTNET_RQ) {
 421                 size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
 422                                sizeof(vq->sw_ring[0]);
 423
 424                 sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
 425                                              RTE_CACHE_LINE_SIZE, socket_id);
 426                 if (!sw_ring) {
 427                         PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
 428                         ret = -ENOMEM;
 429                         goto fail_q_alloc;
 430                 }
 431
 432                 vq->sw_ring = sw_ring;
 433                 rxvq = (struct virtnet_rx *)RTE_PTR_ADD(vq, sz_vq);
 434                 rxvq->vq = vq;
 435                 rxvq->port_id = dev->data->port_id;
 436                 rxvq->queue_id = queue_idx;
 437                 rxvq->mz = mz;
 438                 *pvq = rxvq;
 439         } else if (queue_type == VTNET_TQ) {
 440                 txvq = (struct virtnet_tx *)RTE_PTR_ADD(vq, sz_vq);
 441                 txvq->vq = vq;
 442                 txvq->port_id = dev->data->port_id;
 443                 txvq->queue_id = queue_idx;
 444                 txvq->mz = mz;
 445                 txvq->virtio_net_hdr_mz = hdr_mz;
 446                 txvq->virtio_net_hdr_mem = hdr_mz->phys_addr;
 447
 448                 *pvq = txvq;
 449         } else if (queue_type == VTNET_CQ) {
 450                 cvq = (struct virtnet_ctl *)RTE_PTR_ADD(vq, sz_vq);
 451                 cvq->vq = vq;
 452                 cvq->mz = mz;
 453                 cvq->virtio_net_hdr_mz = hdr_mz;
 454                 cvq->virtio_net_hdr_mem = hdr_mz->phys_addr;
 455                 memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
 456                 *pvq = cvq;
 457         }
 458
 459         /* For virtio-user case (that is when dev->pci_dev is NULL), we use
 460          * virtual address. And we need properly set _offset_, please see
 461          * MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
 462          */
 463         if (dev->pci_dev)
 464                 vq->offset = offsetof(struct rte_mbuf, buf_physaddr);
 465         else {
 466                 vq->vq_ring_mem = (uintptr_t)mz->addr;
 467                 vq->offset = offsetof(struct rte_mbuf, buf_addr);
 468                 if (queue_type == VTNET_TQ)
 469                         txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
 470                 else if (queue_type == VTNET_CQ)
 471                         cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
 472         }
 473
 474         if (queue_type == VTNET_TQ) {
 475                 struct virtio_tx_region *txr;
 476                 unsigned int i;
 477
 478                 txr = hdr_mz->addr;
 479                 memset(txr, 0, vq_size * sizeof(*txr));
 480                 for (i = 0; i < vq_size; i++) {
 481                         struct vring_desc *start_dp = txr[i].tx_indir;
 482
 483                         vring_desc_init(start_dp, RTE_DIM(txr[i].tx_indir));
 484
 485                         /* first indirect descriptor is always the tx header */
 486                         start_dp->addr = txvq->virtio_net_hdr_mem
 487                                 + i * sizeof(*txr)
 488                                 + offsetof(struct virtio_tx_region, tx_hdr);
 489
 490                         start_dp->len = hw->vtnet_hdr_size;
 491                         start_dp->flags = VRING_DESC_F_NEXT;
 492                 }
 493         }
 494
 495         if (hw->vtpci_ops->setup_queue(hw, vq) < 0) {
 496                 PMD_INIT_LOG(ERR, "setup_queue failed");
 497                 virtio_dev_queue_release(vq);
 498                 return -EINVAL;
 499         }
 500
 501         vq->configured = 1;
 502         return 0;
 503
 504 fail_q_alloc:
 505         rte_free(sw_ring);
 506         rte_memzone_free(hdr_mz);
 507         rte_memzone_free(mz);
 508         rte_free(vq);
 509
 510         return ret;
 511 }
 512
 513 static int
 514 virtio_dev_cq_queue_setup(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx,
 515                 uint32_t socket_id)
 516 {
 517         struct virtnet_ctl *cvq;
 518         int ret;
 519         struct virtio_hw *hw = dev->data->dev_private;
 520
 521         PMD_INIT_FUNC_TRACE();
 522         ret = virtio_dev_queue_setup(dev, VTNET_CQ, VTNET_SQ_CQ_QUEUE_IDX,
 523                         vtpci_queue_idx, 0, socket_id, (void **)&cvq);
 524         if (ret < 0) {
 525                 PMD_INIT_LOG(ERR, "control vq initialization failed");
 526                 return ret;
 527         }
 528
 529         hw->cvq = cvq;
 530         return 0;
 531 }
 532
 533 static void
 534 virtio_free_queues(struct rte_eth_dev *dev)
 535 {
 536         unsigned int i;
 537
 538         for (i = 0; i < dev->data->nb_rx_queues; i++)
 539                 virtio_dev_rx_queue_release(dev->data->rx_queues[i]);
 540
 541         dev->data->nb_rx_queues = 0;
 542
 543         for (i = 0; i < dev->data->nb_tx_queues; i++)
 544                 virtio_dev_tx_queue_release(dev->data->tx_queues[i]);
 545
 546         dev->data->nb_tx_queues = 0;
 547 }
 548
 549 static void
 550 virtio_dev_close(struct rte_eth_dev *dev)
 551 {
 552         struct virtio_hw *hw = dev->data->dev_private;
 553
 554         PMD_INIT_LOG(DEBUG, "virtio_dev_close");
 555
 556         if (hw->started == 1)
 557                 virtio_dev_stop(dev);
 558
 559         /* reset the NIC */
 560         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
 561                 vtpci_irq_config(hw, VIRTIO_MSI_NO_VECTOR);
 562         vtpci_reset(hw);
 563         virtio_dev_free_mbufs(dev);
 564         virtio_free_queues(dev);
 565 }
 566
 567 static void
 568 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
 569 {
 570         struct virtio_hw *hw = dev->data->dev_private;
 571         struct virtio_pmd_ctrl ctrl;
 572         int dlen[1];
 573         int ret;
 574
 575         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 576                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 577                 return;
 578         }
 579
 580         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 581         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 582         ctrl.data[0] = 1;
 583         dlen[0] = 1;
 584
 585         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 586         if (ret)
 587                 PMD_INIT_LOG(ERR, "Failed to enable promisc");
 588 }
 589
 590 static void
 591 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
 592 {
 593         struct virtio_hw *hw = dev->data->dev_private;
 594         struct virtio_pmd_ctrl ctrl;
 595         int dlen[1];
 596         int ret;
 597
 598         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 599                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 600                 return;
 601         }
 602
 603         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 604         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 605         ctrl.data[0] = 0;
 606         dlen[0] = 1;
 607
 608         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 609         if (ret)
 610                 PMD_INIT_LOG(ERR, "Failed to disable promisc");
 611 }
 612
 613 static void
 614 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
 615 {
 616         struct virtio_hw *hw = dev->data->dev_private;
 617         struct virtio_pmd_ctrl ctrl;
 618         int dlen[1];
 619         int ret;
 620
 621         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 622                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 623                 return;
 624         }
 625
 626         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 627         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 628         ctrl.data[0] = 1;
 629         dlen[0] = 1;
 630
 631         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 632         if (ret)
 633                 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
 634 }
 635
 636 static void
 637 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
 638 {
 639         struct virtio_hw *hw = dev->data->dev_private;
 640         struct virtio_pmd_ctrl ctrl;
 641         int dlen[1];
 642         int ret;
 643
 644         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 645                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 646                 return;
 647         }
 648
 649         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 650         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 651         ctrl.data[0] = 0;
 652         dlen[0] = 1;
 653
 654         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 655         if (ret)
 656                 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
 657 }
 658
 659 /*
 660  * dev_ops for virtio, bare necessities for basic operation
 661  */
 662 static const struct eth_dev_ops virtio_eth_dev_ops = {
 663         .dev_configure           = virtio_dev_configure,
 664         .dev_start               = virtio_dev_start,
 665         .dev_stop                = virtio_dev_stop,
 666         .dev_close               = virtio_dev_close,
 667         .promiscuous_enable      = virtio_dev_promiscuous_enable,
 668         .promiscuous_disable     = virtio_dev_promiscuous_disable,
 669         .allmulticast_enable     = virtio_dev_allmulticast_enable,
 670         .allmulticast_disable    = virtio_dev_allmulticast_disable,
 671
 672         .dev_infos_get           = virtio_dev_info_get,
 673         .stats_get               = virtio_dev_stats_get,
 674         .xstats_get              = virtio_dev_xstats_get,
 675         .xstats_get_names        = virtio_dev_xstats_get_names,
 676         .stats_reset             = virtio_dev_stats_reset,
 677         .xstats_reset            = virtio_dev_stats_reset,
 678         .link_update             = virtio_dev_link_update,
 679         .rx_queue_setup          = virtio_dev_rx_queue_setup,
 680         .rx_queue_release        = virtio_dev_rx_queue_release,
 681         .tx_queue_setup          = virtio_dev_tx_queue_setup,
 682         .tx_queue_release        = virtio_dev_tx_queue_release,
 683         /* collect stats per queue */
 684         .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
 685         .vlan_filter_set         = virtio_vlan_filter_set,
 686         .mac_addr_add            = virtio_mac_addr_add,
 687         .mac_addr_remove         = virtio_mac_addr_remove,
 688         .mac_addr_set            = virtio_mac_addr_set,
 689 };
 690
 691 static inline int
 692 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
 693                                 struct rte_eth_link *link)
 694 {
 695         struct rte_eth_link *dst = link;
 696         struct rte_eth_link *src = &(dev->data->dev_link);
 697
 698         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
 699                         *(uint64_t *)src) == 0)
 700                 return -1;
 701
 702         return 0;
 703 }
 704
 705 /**
 706  * Atomically writes the link status information into global
 707  * structure rte_eth_dev.
 708  *
 709  * @param dev
 710  *   - Pointer to the structure rte_eth_dev to read from.
 711  *   - Pointer to the buffer to be saved with the link status.
 712  *
 713  * @return
 714  *   - On success, zero.
 715  *   - On failure, negative value.
 716  */
 717 static inline int
 718 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
 719                 struct rte_eth_link *link)
 720 {
 721         struct rte_eth_link *dst = &(dev->data->dev_link);
 722         struct rte_eth_link *src = link;
 723
 724         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
 725                                         *(uint64_t *)src) == 0)
 726                 return -1;
 727
 728         return 0;
 729 }
 730
 731 static void
 732 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 733 {
 734         unsigned i;
 735
 736         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 737                 const struct virtnet_tx *txvq = dev->data->tx_queues[i];
 738                 if (txvq == NULL)
 739                         continue;
 740
 741                 stats->opackets += txvq->stats.packets;
 742                 stats->obytes += txvq->stats.bytes;
 743                 stats->oerrors += txvq->stats.errors;
 744
 745                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 746                         stats->q_opackets[i] = txvq->stats.packets;
 747                         stats->q_obytes[i] = txvq->stats.bytes;
 748                 }
 749         }
 750
 751         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 752                 const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
 753                 if (rxvq == NULL)
 754                         continue;
 755
 756                 stats->ipackets += rxvq->stats.packets;
 757                 stats->ibytes += rxvq->stats.bytes;
 758                 stats->ierrors += rxvq->stats.errors;
 759
 760                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 761                         stats->q_ipackets[i] = rxvq->stats.packets;
 762                         stats->q_ibytes[i] = rxvq->stats.bytes;
 763                 }
 764         }
 765
 766         stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
 767 }
 768
 769 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
 770                                        struct rte_eth_xstat_name *xstats_names,
 771                                        __rte_unused unsigned limit)
 772 {
 773         unsigned i;
 774         unsigned count = 0;
 775         unsigned t;
 776
 777         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
 778                 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
 779
 780         if (xstats_names == NULL) {
 781                 /* Note: limit checked in rte_eth_xstats_names() */
 782
 783                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
 784                         struct virtqueue *rxvq = dev->data->rx_queues[i];
 785                         if (rxvq == NULL)
 786                                 continue;
 787                         for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
 788                                 snprintf(xstats_names[count].name,
 789                                         sizeof(xstats_names[count].name),
 790                                         "rx_q%u_%s", i,
 791                                         rte_virtio_rxq_stat_strings[t].name);
 792                                 xstats_names[count].id = count;
 793                                 count++;
 794                         }
 795                 }
 796
 797                 for (i = 0; i < dev->data->nb_tx_queues; i++) {
 798                         struct virtqueue *txvq = dev->data->tx_queues[i];
 799                         if (txvq == NULL)
 800                                 continue;
 801                         for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
 802                                 snprintf(xstats_names[count].name,
 803                                         sizeof(xstats_names[count].name),
 804                                         "tx_q%u_%s", i,
 805                                         rte_virtio_txq_stat_strings[t].name);
 806                                 xstats_names[count].id = count;
 807                                 count++;
 808                         }
 809                 }
 810                 return count;
 811         }
 812         return nstats;
 813 }
 814
 815 static int
 816 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 817                       unsigned n)
 818 {
 819         unsigned i;
 820         unsigned count = 0;
 821
 822         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
 823                 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
 824
 825         if (n < nstats)
 826                 return nstats;
 827
 828         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 829                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
 830
 831                 if (rxvq == NULL)
 832                         continue;
 833
 834                 unsigned t;
 835
 836                 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
 837                         xstats[count].id = count;
 838                         xstats[count].value = *(uint64_t *)(((char *)rxvq) +
 839                                 rte_virtio_rxq_stat_strings[t].offset);
 840                         count++;
 841                 }
 842         }
 843
 844         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 845                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
 846
 847                 if (txvq == NULL)
 848                         continue;
 849
 850                 unsigned t;
 851
 852                 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
 853                         xstats[count].id = count;
 854                         xstats[count].value = *(uint64_t *)(((char *)txvq) +
 855                                 rte_virtio_txq_stat_strings[t].offset);
 856                         count++;
 857                 }
 858         }
 859
 860         return count;
 861 }
 862
 863 static void
 864 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 865 {
 866         virtio_update_stats(dev, stats);
 867 }
 868
 869 static void
 870 virtio_dev_stats_reset(struct rte_eth_dev *dev)
 871 {
 872         unsigned int i;
 873
 874         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 875                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
 876                 if (txvq == NULL)
 877                         continue;
 878
 879                 txvq->stats.packets = 0;
 880                 txvq->stats.bytes = 0;
 881                 txvq->stats.errors = 0;
 882                 txvq->stats.multicast = 0;
 883                 txvq->stats.broadcast = 0;
 884                 memset(txvq->stats.size_bins, 0,
 885                        sizeof(txvq->stats.size_bins[0]) * 8);
 886         }
 887
 888         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 889                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
 890                 if (rxvq == NULL)
 891                         continue;
 892
 893                 rxvq->stats.packets = 0;
 894                 rxvq->stats.bytes = 0;
 895                 rxvq->stats.errors = 0;
 896                 rxvq->stats.multicast = 0;
 897                 rxvq->stats.broadcast = 0;
 898                 memset(rxvq->stats.size_bins, 0,
 899                        sizeof(rxvq->stats.size_bins[0]) * 8);
 900         }
 901 }
 902
 903 static void
 904 virtio_set_hwaddr(struct virtio_hw *hw)
 905 {
 906         vtpci_write_dev_config(hw,
 907                         offsetof(struct virtio_net_config, mac),
 908                         &hw->mac_addr, ETHER_ADDR_LEN);
 909 }
 910
 911 static void
 912 virtio_get_hwaddr(struct virtio_hw *hw)
 913 {
 914         if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
 915                 vtpci_read_dev_config(hw,
 916                         offsetof(struct virtio_net_config, mac),
 917                         &hw->mac_addr, ETHER_ADDR_LEN);
 918         } else {
 919                 eth_random_addr(&hw->mac_addr[0]);
 920                 virtio_set_hwaddr(hw);
 921         }
 922 }
 923
 924 static void
 925 virtio_mac_table_set(struct virtio_hw *hw,
 926                      const struct virtio_net_ctrl_mac *uc,
 927                      const struct virtio_net_ctrl_mac *mc)
 928 {
 929         struct virtio_pmd_ctrl ctrl;
 930         int err, len[2];
 931
 932         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
 933                 PMD_DRV_LOG(INFO, "host does not support mac table");
 934                 return;
 935         }
 936
 937         ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
 938         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
 939
 940         len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
 941         memcpy(ctrl.data, uc, len[0]);
 942
 943         len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
 944         memcpy(ctrl.data + len[0], mc, len[1]);
 945
 946         err = virtio_send_command(hw->cvq, &ctrl, len, 2);
 947         if (err != 0)
 948                 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
 949 }
 950
 951 static void
 952 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
 953                     uint32_t index, uint32_t vmdq __rte_unused)
 954 {
 955         struct virtio_hw *hw = dev->data->dev_private;
 956         const struct ether_addr *addrs = dev->data->mac_addrs;
 957         unsigned int i;
 958         struct virtio_net_ctrl_mac *uc, *mc;
 959
 960         if (index >= VIRTIO_MAX_MAC_ADDRS) {
 961                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
 962                 return;
 963         }
 964
 965         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
 966         uc->entries = 0;
 967         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
 968         mc->entries = 0;
 969
 970         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
 971                 const struct ether_addr *addr
 972                         = (i == index) ? mac_addr : addrs + i;
 973                 struct virtio_net_ctrl_mac *tbl
 974                         = is_multicast_ether_addr(addr) ? mc : uc;
 975
 976                 memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
 977         }
 978
 979         virtio_mac_table_set(hw, uc, mc);
 980 }
 981
 982 static void
 983 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
 984 {
 985         struct virtio_hw *hw = dev->data->dev_private;
 986         struct ether_addr *addrs = dev->data->mac_addrs;
 987         struct virtio_net_ctrl_mac *uc, *mc;
 988         unsigned int i;
 989
 990         if (index >= VIRTIO_MAX_MAC_ADDRS) {
 991                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
 992                 return;
 993         }
 994
 995         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
 996         uc->entries = 0;
 997         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
 998         mc->entries = 0;
 999
1000         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1001                 struct virtio_net_ctrl_mac *tbl;
1002
1003                 if (i == index || is_zero_ether_addr(addrs + i))
1004                         continue;
1005
1006                 tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
1007                 memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
1008         }
1009
1010         virtio_mac_table_set(hw, uc, mc);
1011 }
1012
1013 static void
1014 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1015 {
1016         struct virtio_hw *hw = dev->data->dev_private;
1017
1018         memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
1019
1020         /* Use atomic update if available */
1021         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1022                 struct virtio_pmd_ctrl ctrl;
1023                 int len = ETHER_ADDR_LEN;
1024
1025                 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1026                 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1027
1028                 memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
1029                 virtio_send_command(hw->cvq, &ctrl, &len, 1);
1030         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1031                 virtio_set_hwaddr(hw);
1032 }
1033
1034 static int
1035 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1036 {
1037         struct virtio_hw *hw = dev->data->dev_private;
1038         struct virtio_pmd_ctrl ctrl;
1039         int len;
1040
1041         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1042                 return -ENOTSUP;
1043
1044         ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1045         ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1046         memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1047         len = sizeof(vlan_id);
1048
1049         return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1050 }
1051
1052 static int
1053 virtio_negotiate_features(struct virtio_hw *hw)
1054 {
1055         uint64_t host_features;
1056
1057         /* Prepare guest_features: feature that driver wants to support */
1058         hw->guest_features = VIRTIO_PMD_GUEST_FEATURES;
1059         PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1060                 hw->guest_features);
1061
1062         /* Read device(host) feature bits */
1063         host_features = hw->vtpci_ops->get_features(hw);
1064         PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1065                 host_features);
1066
1067         /*
1068          * Negotiate features: Subset of device feature bits are written back
1069          * guest feature bits.
1070          */
1071         hw->guest_features = vtpci_negotiate_features(hw, host_features);
1072         PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1073                 hw->guest_features);
1074
1075         if (hw->modern) {
1076                 if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1077                         PMD_INIT_LOG(ERR,
1078                                 "VIRTIO_F_VERSION_1 features is not enabled.");
1079                         return -1;
1080                 }
1081                 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1082                 if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1083                         PMD_INIT_LOG(ERR,
1084                                 "failed to set FEATURES_OK status!");
1085                         return -1;
1086                 }
1087         }
1088
1089         return 0;
1090 }
1091
1092 /*
1093  * Process Virtio Config changed interrupt and call the callback
1094  * if link state changed.
1095  */
1096 static void
1097 virtio_interrupt_handler(__rte_unused struct rte_intr_handle *handle,
1098                          void *param)
1099 {
1100         struct rte_eth_dev *dev = param;
1101         struct virtio_hw *hw = dev->data->dev_private;
1102         uint8_t isr;
1103
1104         /* Read interrupt status which clears interrupt */
1105         isr = vtpci_isr(hw);
1106         PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1107
1108         if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0)
1109                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1110
1111         if (isr & VIRTIO_PCI_ISR_CONFIG) {
1112                 if (virtio_dev_link_update(dev, 0) == 0)
1113                         _rte_eth_dev_callback_process(dev,
1114                                                       RTE_ETH_EVENT_INTR_LSC);
1115         }
1116
1117 }
1118
1119 static void
1120 rx_func_get(struct rte_eth_dev *eth_dev)
1121 {
1122         struct virtio_hw *hw = eth_dev->data->dev_private;
1123         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF))
1124                 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1125         else
1126                 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1127 }
1128
1129 /*
1130  * This function is based on probe() function in virtio_pci.c
1131  * It returns 0 on success.
1132  */
1133 static int
1134 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1135 {
1136         struct virtio_hw *hw = eth_dev->data->dev_private;
1137         struct virtio_net_config *config;
1138         struct virtio_net_config local_config;
1139         struct rte_pci_device *pci_dev;
1140         uint32_t dev_flags = RTE_ETH_DEV_DETACHABLE;
1141         int ret;
1142
1143         RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr_mrg_rxbuf));
1144
1145         eth_dev->dev_ops = &virtio_eth_dev_ops;
1146         eth_dev->tx_pkt_burst = &virtio_xmit_pkts;
1147
1148         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1149                 rx_func_get(eth_dev);
1150                 return 0;
1151         }
1152
1153         /* Allocate memory for storing MAC addresses */
1154         eth_dev->data->mac_addrs = rte_zmalloc("virtio", VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN, 0);
1155         if (eth_dev->data->mac_addrs == NULL) {
1156                 PMD_INIT_LOG(ERR,
1157                         "Failed to allocate %d bytes needed to store MAC addresses",
1158                         VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN);
1159                 return -ENOMEM;
1160         }
1161
1162         pci_dev = eth_dev->pci_dev;
1163
1164         ret = vtpci_init(pci_dev, hw, &dev_flags);
1165         if (ret)
1166                 return ret;
1167
1168         /* Reset the device although not necessary at startup */
1169         vtpci_reset(hw);
1170
1171         /* Tell the host we've noticed this device. */
1172         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1173
1174         /* Tell the host we've known how to drive the device. */
1175         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1176         if (virtio_negotiate_features(hw) < 0)
1177                 return -1;
1178
1179         /* If host does not support status then disable LSC */
1180         if (!vtpci_with_feature(hw, VIRTIO_NET_F_STATUS))
1181                 dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1182
1183         rte_eth_copy_pci_info(eth_dev, pci_dev);
1184         eth_dev->data->dev_flags = dev_flags;
1185
1186         rx_func_get(eth_dev);
1187
1188         /* Setting up rx_header size for the device */
1189         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1190             vtpci_with_feature(hw, VIRTIO_F_VERSION_1))
1191                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1192         else
1193                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1194
1195         /* Copy the permanent MAC address to: virtio_hw */
1196         virtio_get_hwaddr(hw);
1197         ether_addr_copy((struct ether_addr *) hw->mac_addr,
1198                         &eth_dev->data->mac_addrs[0]);
1199         PMD_INIT_LOG(DEBUG,
1200                      "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1201                      hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1202                      hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1203
1204         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1205                 config = &local_config;
1206
1207                 vtpci_read_dev_config(hw,
1208                         offsetof(struct virtio_net_config, mac),
1209                         &config->mac, sizeof(config->mac));
1210
1211                 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1212                         vtpci_read_dev_config(hw,
1213                                 offsetof(struct virtio_net_config, status),
1214                                 &config->status, sizeof(config->status));
1215                 } else {
1216                         PMD_INIT_LOG(DEBUG,
1217                                      "VIRTIO_NET_F_STATUS is not supported");
1218                         config->status = 0;
1219                 }
1220
1221                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1222                         vtpci_read_dev_config(hw,
1223                                 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1224                                 &config->max_virtqueue_pairs,
1225                                 sizeof(config->max_virtqueue_pairs));
1226                 } else {
1227                         PMD_INIT_LOG(DEBUG,
1228                                      "VIRTIO_NET_F_MQ is not supported");
1229                         config->max_virtqueue_pairs = 1;
1230                 }
1231
1232                 hw->max_rx_queues =
1233                         (VIRTIO_MAX_RX_QUEUES < config->max_virtqueue_pairs) ?
1234                         VIRTIO_MAX_RX_QUEUES : config->max_virtqueue_pairs;
1235                 hw->max_tx_queues =
1236                         (VIRTIO_MAX_TX_QUEUES < config->max_virtqueue_pairs) ?
1237                         VIRTIO_MAX_TX_QUEUES : config->max_virtqueue_pairs;
1238
1239                 virtio_dev_cq_queue_setup(eth_dev,
1240                                         config->max_virtqueue_pairs * 2,
1241                                         SOCKET_ID_ANY);
1242
1243                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1244                                 config->max_virtqueue_pairs);
1245                 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1246                 PMD_INIT_LOG(DEBUG,
1247                                 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1248                                 config->mac[0], config->mac[1],
1249                                 config->mac[2], config->mac[3],
1250                                 config->mac[4], config->mac[5]);
1251         } else {
1252                 hw->max_rx_queues = 1;
1253                 hw->max_tx_queues = 1;
1254         }
1255
1256         PMD_INIT_LOG(DEBUG, "hw->max_rx_queues=%d   hw->max_tx_queues=%d",
1257                         hw->max_rx_queues, hw->max_tx_queues);
1258         PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1259                         eth_dev->data->port_id, pci_dev->id.vendor_id,
1260                         pci_dev->id.device_id);
1261
1262         /* Setup interrupt callback  */
1263         if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1264                 rte_intr_callback_register(&pci_dev->intr_handle,
1265                                    virtio_interrupt_handler, eth_dev);
1266
1267         virtio_dev_cq_start(eth_dev);
1268
1269         return 0;
1270 }
1271
1272 static int
1273 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1274 {
1275         struct rte_pci_device *pci_dev;
1276         struct virtio_hw *hw = eth_dev->data->dev_private;
1277
1278         PMD_INIT_FUNC_TRACE();
1279
1280         if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1281                 return -EPERM;
1282
1283         /* Close it anyway since there's no way to know if closed */
1284         virtio_dev_close(eth_dev);
1285
1286         pci_dev = eth_dev->pci_dev;
1287
1288         eth_dev->dev_ops = NULL;
1289         eth_dev->tx_pkt_burst = NULL;
1290         eth_dev->rx_pkt_burst = NULL;
1291
1292         if (hw->cvq)
1293                 virtio_dev_queue_release(hw->cvq->vq);
1294
1295         rte_free(eth_dev->data->mac_addrs);
1296         eth_dev->data->mac_addrs = NULL;
1297
1298         /* reset interrupt callback  */
1299         if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1300                 rte_intr_callback_unregister(&pci_dev->intr_handle,
1301                                                 virtio_interrupt_handler,
1302                                                 eth_dev);
1303         rte_eal_pci_unmap_device(pci_dev);
1304
1305         PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1306
1307         return 0;
1308 }
1309
1310 static struct eth_driver rte_virtio_pmd = {
1311         .pci_drv = {
1312                 .name = "rte_virtio_pmd",
1313                 .id_table = pci_id_virtio_map,
1314                 .drv_flags = RTE_PCI_DRV_DETACHABLE,
1315         },
1316         .eth_dev_init = eth_virtio_dev_init,
1317         .eth_dev_uninit = eth_virtio_dev_uninit,
1318         .dev_private_size = sizeof(struct virtio_hw),
1319 };
1320
1321 /*
1322  * Driver initialization routine.
1323  * Invoked once at EAL init time.
1324  * Register itself as the [Poll Mode] Driver of PCI virtio devices.
1325  * Returns 0 on success.
1326  */
1327 static int
1328 rte_virtio_pmd_init(const char *name __rte_unused,
1329                     const char *param __rte_unused)
1330 {
1331         if (rte_eal_iopl_init() != 0) {
1332                 PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1333                 return -1;
1334         }
1335
1336         rte_eth_driver_register(&rte_virtio_pmd);
1337         return 0;
1338 }
1339
1340 /*
1341  * Configure virtio device
1342  * It returns 0 on success.
1343  */
1344 static int
1345 virtio_dev_configure(struct rte_eth_dev *dev)
1346 {
1347         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1348         struct virtio_hw *hw = dev->data->dev_private;
1349
1350         PMD_INIT_LOG(DEBUG, "configure");
1351
1352         if (rxmode->hw_ip_checksum) {
1353                 PMD_DRV_LOG(ERR, "HW IP checksum not supported");
1354                 return -EINVAL;
1355         }
1356
1357         hw->vlan_strip = rxmode->hw_vlan_strip;
1358
1359         if (rxmode->hw_vlan_filter
1360             && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1361                 PMD_DRV_LOG(NOTICE,
1362                             "vlan filtering not available on this host");
1363                 return -ENOTSUP;
1364         }
1365
1366         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1367                 if (vtpci_irq_config(hw, 0) == VIRTIO_MSI_NO_VECTOR) {
1368                         PMD_DRV_LOG(ERR, "failed to set config vector");
1369                         return -EBUSY;
1370                 }
1371
1372         return 0;
1373 }
1374
1375
1376 static int
1377 virtio_dev_start(struct rte_eth_dev *dev)
1378 {
1379         uint16_t nb_queues, i;
1380         struct virtio_hw *hw = dev->data->dev_private;
1381         struct virtnet_rx *rxvq;
1382         struct virtnet_tx *txvq __rte_unused;
1383
1384         /* check if lsc interrupt feature is enabled */
1385         if (dev->data->dev_conf.intr_conf.lsc) {
1386                 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
1387                         PMD_DRV_LOG(ERR, "link status not supported by host");
1388                         return -ENOTSUP;
1389                 }
1390
1391                 if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0) {
1392                         PMD_DRV_LOG(ERR, "interrupt enable failed");
1393                         return -EIO;
1394                 }
1395         }
1396
1397         /* Initialize Link state */
1398         virtio_dev_link_update(dev, 0);
1399
1400         /* On restart after stop do not touch queues */
1401         if (hw->started)
1402                 return 0;
1403
1404         /* Do final configuration before rx/tx engine starts */
1405         virtio_dev_rxtx_start(dev);
1406         vtpci_reinit_complete(hw);
1407
1408         hw->started = 1;
1409
1410         /*Notify the backend
1411          *Otherwise the tap backend might already stop its queue due to fullness.
1412          *vhost backend will have no chance to be waked up
1413          */
1414         nb_queues = dev->data->nb_rx_queues;
1415         if (nb_queues > 1) {
1416                 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1417                         return -EINVAL;
1418         }
1419
1420         PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1421
1422         for (i = 0; i < nb_queues; i++) {
1423                 rxvq = dev->data->rx_queues[i];
1424                 virtqueue_notify(rxvq->vq);
1425         }
1426
1427         PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1428
1429         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1430                 rxvq = dev->data->rx_queues[i];
1431                 VIRTQUEUE_DUMP(rxvq->vq);
1432         }
1433
1434         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1435                 txvq = dev->data->tx_queues[i];
1436                 VIRTQUEUE_DUMP(txvq->vq);
1437         }
1438
1439         return 0;
1440 }
1441
1442 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1443 {
1444         struct rte_mbuf *buf;
1445         int i, mbuf_num = 0;
1446
1447         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1448                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1449
1450                 PMD_INIT_LOG(DEBUG,
1451                              "Before freeing rxq[%d] used and unused buf", i);
1452                 VIRTQUEUE_DUMP(rxvq->vq);
1453
1454                 PMD_INIT_LOG(DEBUG, "rx_queues[%d]=%p", i, rxvq);
1455                 while ((buf = virtqueue_detatch_unused(rxvq->vq)) != NULL) {
1456                         rte_pktmbuf_free(buf);
1457                         mbuf_num++;
1458                 }
1459
1460                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1461                 PMD_INIT_LOG(DEBUG,
1462                              "After freeing rxq[%d] used and unused buf", i);
1463                 VIRTQUEUE_DUMP(rxvq->vq);
1464         }
1465
1466         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1467                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1468
1469                 PMD_INIT_LOG(DEBUG,
1470                              "Before freeing txq[%d] used and unused bufs",
1471                              i);
1472                 VIRTQUEUE_DUMP(txvq->vq);
1473
1474                 mbuf_num = 0;
1475                 while ((buf = virtqueue_detatch_unused(txvq->vq)) != NULL) {
1476                         rte_pktmbuf_free(buf);
1477                         mbuf_num++;
1478                 }
1479
1480                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1481                 PMD_INIT_LOG(DEBUG,
1482                              "After freeing txq[%d] used and unused buf", i);
1483                 VIRTQUEUE_DUMP(txvq->vq);
1484         }
1485 }
1486
1487 /*
1488  * Stop device: disable interrupt and mark link down
1489  */
1490 static void
1491 virtio_dev_stop(struct rte_eth_dev *dev)
1492 {
1493         struct rte_eth_link link;
1494         struct virtio_hw *hw = dev->data->dev_private;
1495
1496         PMD_INIT_LOG(DEBUG, "stop");
1497
1498         hw->started = 0;
1499
1500         if (dev->data->dev_conf.intr_conf.lsc)
1501                 rte_intr_disable(&dev->pci_dev->intr_handle);
1502
1503         memset(&link, 0, sizeof(link));
1504         virtio_dev_atomic_write_link_status(dev, &link);
1505 }
1506
1507 static int
1508 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
1509 {
1510         struct rte_eth_link link, old;
1511         uint16_t status;
1512         struct virtio_hw *hw = dev->data->dev_private;
1513         memset(&link, 0, sizeof(link));
1514         virtio_dev_atomic_read_link_status(dev, &link);
1515         old = link;
1516         link.link_duplex = ETH_LINK_FULL_DUPLEX;
1517         link.link_speed  = SPEED_10G;
1518
1519         if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1520                 PMD_INIT_LOG(DEBUG, "Get link status from hw");
1521                 vtpci_read_dev_config(hw,
1522                                 offsetof(struct virtio_net_config, status),
1523                                 &status, sizeof(status));
1524                 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
1525                         link.link_status = ETH_LINK_DOWN;
1526                         PMD_INIT_LOG(DEBUG, "Port %d is down",
1527                                      dev->data->port_id);
1528                 } else {
1529                         link.link_status = ETH_LINK_UP;
1530                         PMD_INIT_LOG(DEBUG, "Port %d is up",
1531                                      dev->data->port_id);
1532                 }
1533         } else {
1534                 link.link_status = ETH_LINK_UP;
1535         }
1536         virtio_dev_atomic_write_link_status(dev, &link);
1537
1538         return (old.link_status == link.link_status) ? -1 : 0;
1539 }
1540
1541 static void
1542 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1543 {
1544         struct virtio_hw *hw = dev->data->dev_private;
1545
1546         dev_info->driver_name = dev->driver->pci_drv.name;
1547         dev_info->max_rx_queues = (uint16_t)hw->max_rx_queues;
1548         dev_info->max_tx_queues = (uint16_t)hw->max_tx_queues;
1549         dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
1550         dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
1551         dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
1552         dev_info->default_txconf = (struct rte_eth_txconf) {
1553                 .txq_flags = ETH_TXQ_FLAGS_NOOFFLOADS
1554         };
1555 }
1556
1557 /*
1558  * It enables testpmd to collect per queue stats.
1559  */
1560 static int
1561 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
1562 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
1563 __rte_unused uint8_t is_rx)
1564 {
1565         return 0;
1566 }
1567
1568 static struct rte_driver rte_virtio_driver = {
1569         .type = PMD_PDEV,
1570         .init = rte_virtio_pmd_init,
1571 };
1572
1573 PMD_REGISTER_DRIVER(rte_virtio_driver);