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_mz->phys_addr;
 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_mz->phys_addr
 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_mz->phys_addr
 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;
 315         struct virtnet_tx *txvq;
 316         struct virtnet_ctl *cvq;
 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                 struct virtio_tx_region *txr;
 441                 unsigned int i;
 442
 443                 txvq = (struct virtnet_tx *)RTE_PTR_ADD(vq, sz_vq);
 444                 txvq->vq = vq;
 445                 txvq->port_id = dev->data->port_id;
 446                 txvq->queue_id = queue_idx;
 447                 txvq->mz = mz;
 448                 txvq->virtio_net_hdr_mz = hdr_mz;
 449                 txvq->virtio_net_hdr_mem = hdr_mz->phys_addr;
 450
 451                 txr = hdr_mz->addr;
 452                 memset(txr, 0, vq_size * sizeof(*txr));
 453                 for (i = 0; i < vq_size; i++) {
 454                         struct vring_desc *start_dp = txr[i].tx_indir;
 455
 456                         vring_desc_init(start_dp, RTE_DIM(txr[i].tx_indir));
 457
 458                         /* first indirect descriptor is always the tx header */
 459                         start_dp->addr = txvq->virtio_net_hdr_mem
 460                                 + i * sizeof(*txr)
 461                                 + offsetof(struct virtio_tx_region, tx_hdr);
 462
 463                         start_dp->len = hw->vtnet_hdr_size;
 464                         start_dp->flags = VRING_DESC_F_NEXT;
 465                 }
 466
 467                 *pvq = txvq;
 468         } else if (queue_type == VTNET_CQ) {
 469                 cvq = (struct virtnet_ctl *)RTE_PTR_ADD(vq, sz_vq);
 470                 cvq->vq = vq;
 471                 cvq->mz = mz;
 472                 cvq->virtio_net_hdr_mz = hdr_mz;
 473                 cvq->virtio_net_hdr_mem = hdr_mz->phys_addr;
 474                 memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
 475                 *pvq = cvq;
 476         }
 477
 478         if (hw->vtpci_ops->setup_queue(hw, vq) < 0) {
 479                 PMD_INIT_LOG(ERR, "setup_queue failed");
 480                 virtio_dev_queue_release(vq);
 481                 return -EINVAL;
 482         }
 483
 484         vq->configured = 1;
 485         return 0;
 486
 487 fail_q_alloc:
 488         rte_free(sw_ring);
 489         rte_memzone_free(hdr_mz);
 490         rte_memzone_free(mz);
 491         rte_free(vq);
 492
 493         return ret;
 494 }
 495
 496 static int
 497 virtio_dev_cq_queue_setup(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx,
 498                 uint32_t socket_id)
 499 {
 500         struct virtnet_ctl *cvq;
 501         int ret;
 502         struct virtio_hw *hw = dev->data->dev_private;
 503
 504         PMD_INIT_FUNC_TRACE();
 505         ret = virtio_dev_queue_setup(dev, VTNET_CQ, VTNET_SQ_CQ_QUEUE_IDX,
 506                         vtpci_queue_idx, 0, socket_id, (void **)&cvq);
 507         if (ret < 0) {
 508                 PMD_INIT_LOG(ERR, "control vq initialization failed");
 509                 return ret;
 510         }
 511
 512         hw->cvq = cvq;
 513         return 0;
 514 }
 515
 516 static void
 517 virtio_free_queues(struct rte_eth_dev *dev)
 518 {
 519         unsigned int i;
 520
 521         for (i = 0; i < dev->data->nb_rx_queues; i++)
 522                 virtio_dev_rx_queue_release(dev->data->rx_queues[i]);
 523
 524         dev->data->nb_rx_queues = 0;
 525
 526         for (i = 0; i < dev->data->nb_tx_queues; i++)
 527                 virtio_dev_tx_queue_release(dev->data->tx_queues[i]);
 528
 529         dev->data->nb_tx_queues = 0;
 530 }
 531
 532 static void
 533 virtio_dev_close(struct rte_eth_dev *dev)
 534 {
 535         struct virtio_hw *hw = dev->data->dev_private;
 536
 537         PMD_INIT_LOG(DEBUG, "virtio_dev_close");
 538
 539         if (hw->started == 1)
 540                 virtio_dev_stop(dev);
 541
 542         /* reset the NIC */
 543         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
 544                 vtpci_irq_config(hw, VIRTIO_MSI_NO_VECTOR);
 545         vtpci_reset(hw);
 546         virtio_dev_free_mbufs(dev);
 547         virtio_free_queues(dev);
 548 }
 549
 550 static void
 551 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
 552 {
 553         struct virtio_hw *hw = dev->data->dev_private;
 554         struct virtio_pmd_ctrl ctrl;
 555         int dlen[1];
 556         int ret;
 557
 558         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 559                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 560                 return;
 561         }
 562
 563         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 564         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 565         ctrl.data[0] = 1;
 566         dlen[0] = 1;
 567
 568         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 569         if (ret)
 570                 PMD_INIT_LOG(ERR, "Failed to enable promisc");
 571 }
 572
 573 static void
 574 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
 575 {
 576         struct virtio_hw *hw = dev->data->dev_private;
 577         struct virtio_pmd_ctrl ctrl;
 578         int dlen[1];
 579         int ret;
 580
 581         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 582                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 583                 return;
 584         }
 585
 586         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 587         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 588         ctrl.data[0] = 0;
 589         dlen[0] = 1;
 590
 591         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 592         if (ret)
 593                 PMD_INIT_LOG(ERR, "Failed to disable promisc");
 594 }
 595
 596 static void
 597 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
 598 {
 599         struct virtio_hw *hw = dev->data->dev_private;
 600         struct virtio_pmd_ctrl ctrl;
 601         int dlen[1];
 602         int ret;
 603
 604         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 605                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 606                 return;
 607         }
 608
 609         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 610         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 611         ctrl.data[0] = 1;
 612         dlen[0] = 1;
 613
 614         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 615         if (ret)
 616                 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
 617 }
 618
 619 static void
 620 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
 621 {
 622         struct virtio_hw *hw = dev->data->dev_private;
 623         struct virtio_pmd_ctrl ctrl;
 624         int dlen[1];
 625         int ret;
 626
 627         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 628                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 629                 return;
 630         }
 631
 632         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 633         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 634         ctrl.data[0] = 0;
 635         dlen[0] = 1;
 636
 637         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 638         if (ret)
 639                 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
 640 }
 641
 642 /*
 643  * dev_ops for virtio, bare necessities for basic operation
 644  */
 645 static const struct eth_dev_ops virtio_eth_dev_ops = {
 646         .dev_configure           = virtio_dev_configure,
 647         .dev_start               = virtio_dev_start,
 648         .dev_stop                = virtio_dev_stop,
 649         .dev_close               = virtio_dev_close,
 650         .promiscuous_enable      = virtio_dev_promiscuous_enable,
 651         .promiscuous_disable     = virtio_dev_promiscuous_disable,
 652         .allmulticast_enable     = virtio_dev_allmulticast_enable,
 653         .allmulticast_disable    = virtio_dev_allmulticast_disable,
 654
 655         .dev_infos_get           = virtio_dev_info_get,
 656         .stats_get               = virtio_dev_stats_get,
 657         .xstats_get              = virtio_dev_xstats_get,
 658         .xstats_get_names        = virtio_dev_xstats_get_names,
 659         .stats_reset             = virtio_dev_stats_reset,
 660         .xstats_reset            = virtio_dev_stats_reset,
 661         .link_update             = virtio_dev_link_update,
 662         .rx_queue_setup          = virtio_dev_rx_queue_setup,
 663         .rx_queue_release        = virtio_dev_rx_queue_release,
 664         .tx_queue_setup          = virtio_dev_tx_queue_setup,
 665         .tx_queue_release        = virtio_dev_tx_queue_release,
 666         /* collect stats per queue */
 667         .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
 668         .vlan_filter_set         = virtio_vlan_filter_set,
 669         .mac_addr_add            = virtio_mac_addr_add,
 670         .mac_addr_remove         = virtio_mac_addr_remove,
 671         .mac_addr_set            = virtio_mac_addr_set,
 672 };
 673
 674 static inline int
 675 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
 676                                 struct rte_eth_link *link)
 677 {
 678         struct rte_eth_link *dst = link;
 679         struct rte_eth_link *src = &(dev->data->dev_link);
 680
 681         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
 682                         *(uint64_t *)src) == 0)
 683                 return -1;
 684
 685         return 0;
 686 }
 687
 688 /**
 689  * Atomically writes the link status information into global
 690  * structure rte_eth_dev.
 691  *
 692  * @param dev
 693  *   - Pointer to the structure rte_eth_dev to read from.
 694  *   - Pointer to the buffer to be saved with the link status.
 695  *
 696  * @return
 697  *   - On success, zero.
 698  *   - On failure, negative value.
 699  */
 700 static inline int
 701 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
 702                 struct rte_eth_link *link)
 703 {
 704         struct rte_eth_link *dst = &(dev->data->dev_link);
 705         struct rte_eth_link *src = link;
 706
 707         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
 708                                         *(uint64_t *)src) == 0)
 709                 return -1;
 710
 711         return 0;
 712 }
 713
 714 static void
 715 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 716 {
 717         unsigned i;
 718
 719         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 720                 const struct virtnet_tx *txvq = dev->data->tx_queues[i];
 721                 if (txvq == NULL)
 722                         continue;
 723
 724                 stats->opackets += txvq->stats.packets;
 725                 stats->obytes += txvq->stats.bytes;
 726                 stats->oerrors += txvq->stats.errors;
 727
 728                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 729                         stats->q_opackets[i] = txvq->stats.packets;
 730                         stats->q_obytes[i] = txvq->stats.bytes;
 731                 }
 732         }
 733
 734         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 735                 const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
 736                 if (rxvq == NULL)
 737                         continue;
 738
 739                 stats->ipackets += rxvq->stats.packets;
 740                 stats->ibytes += rxvq->stats.bytes;
 741                 stats->ierrors += rxvq->stats.errors;
 742
 743                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 744                         stats->q_ipackets[i] = rxvq->stats.packets;
 745                         stats->q_ibytes[i] = rxvq->stats.bytes;
 746                 }
 747         }
 748
 749         stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
 750 }
 751
 752 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
 753                                        struct rte_eth_xstat_name *xstats_names,
 754                                        __rte_unused unsigned limit)
 755 {
 756         unsigned i;
 757         unsigned count = 0;
 758         unsigned t;
 759
 760         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
 761                 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
 762
 763         if (xstats_names == NULL) {
 764                 /* Note: limit checked in rte_eth_xstats_names() */
 765
 766                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
 767                         struct virtqueue *rxvq = dev->data->rx_queues[i];
 768                         if (rxvq == NULL)
 769                                 continue;
 770                         for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
 771                                 snprintf(xstats_names[count].name,
 772                                         sizeof(xstats_names[count].name),
 773                                         "rx_q%u_%s", i,
 774                                         rte_virtio_rxq_stat_strings[t].name);
 775                                 xstats_names[count].id = count;
 776                                 count++;
 777                         }
 778                 }
 779
 780                 for (i = 0; i < dev->data->nb_tx_queues; i++) {
 781                         struct virtqueue *txvq = dev->data->tx_queues[i];
 782                         if (txvq == NULL)
 783                                 continue;
 784                         for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
 785                                 snprintf(xstats_names[count].name,
 786                                         sizeof(xstats_names[count].name),
 787                                         "tx_q%u_%s", i,
 788                                         rte_virtio_txq_stat_strings[t].name);
 789                                 xstats_names[count].id = count;
 790                                 count++;
 791                         }
 792                 }
 793                 return count;
 794         }
 795         return nstats;
 796 }
 797
 798 static int
 799 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 800                       unsigned n)
 801 {
 802         unsigned i;
 803         unsigned count = 0;
 804
 805         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
 806                 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
 807
 808         if (n < nstats)
 809                 return nstats;
 810
 811         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 812                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
 813
 814                 if (rxvq == NULL)
 815                         continue;
 816
 817                 unsigned t;
 818
 819                 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
 820                         xstats[count].id = count;
 821                         xstats[count].value = *(uint64_t *)(((char *)rxvq) +
 822                                 rte_virtio_rxq_stat_strings[t].offset);
 823                         count++;
 824                 }
 825         }
 826
 827         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 828                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
 829
 830                 if (txvq == NULL)
 831                         continue;
 832
 833                 unsigned t;
 834
 835                 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
 836                         xstats[count].id = count;
 837                         xstats[count].value = *(uint64_t *)(((char *)txvq) +
 838                                 rte_virtio_txq_stat_strings[t].offset);
 839                         count++;
 840                 }
 841         }
 842
 843         return count;
 844 }
 845
 846 static void
 847 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 848 {
 849         virtio_update_stats(dev, stats);
 850 }
 851
 852 static void
 853 virtio_dev_stats_reset(struct rte_eth_dev *dev)
 854 {
 855         unsigned int i;
 856
 857         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 858                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
 859                 if (txvq == NULL)
 860                         continue;
 861
 862                 txvq->stats.packets = 0;
 863                 txvq->stats.bytes = 0;
 864                 txvq->stats.errors = 0;
 865                 txvq->stats.multicast = 0;
 866                 txvq->stats.broadcast = 0;
 867                 memset(txvq->stats.size_bins, 0,
 868                        sizeof(txvq->stats.size_bins[0]) * 8);
 869         }
 870
 871         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 872                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
 873                 if (rxvq == NULL)
 874                         continue;
 875
 876                 rxvq->stats.packets = 0;
 877                 rxvq->stats.bytes = 0;
 878                 rxvq->stats.errors = 0;
 879                 rxvq->stats.multicast = 0;
 880                 rxvq->stats.broadcast = 0;
 881                 memset(rxvq->stats.size_bins, 0,
 882                        sizeof(rxvq->stats.size_bins[0]) * 8);
 883         }
 884 }
 885
 886 static void
 887 virtio_set_hwaddr(struct virtio_hw *hw)
 888 {
 889         vtpci_write_dev_config(hw,
 890                         offsetof(struct virtio_net_config, mac),
 891                         &hw->mac_addr, ETHER_ADDR_LEN);
 892 }
 893
 894 static void
 895 virtio_get_hwaddr(struct virtio_hw *hw)
 896 {
 897         if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
 898                 vtpci_read_dev_config(hw,
 899                         offsetof(struct virtio_net_config, mac),
 900                         &hw->mac_addr, ETHER_ADDR_LEN);
 901         } else {
 902                 eth_random_addr(&hw->mac_addr[0]);
 903                 virtio_set_hwaddr(hw);
 904         }
 905 }
 906
 907 static void
 908 virtio_mac_table_set(struct virtio_hw *hw,
 909                      const struct virtio_net_ctrl_mac *uc,
 910                      const struct virtio_net_ctrl_mac *mc)
 911 {
 912         struct virtio_pmd_ctrl ctrl;
 913         int err, len[2];
 914
 915         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
 916                 PMD_DRV_LOG(INFO, "host does not support mac table");
 917                 return;
 918         }
 919
 920         ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
 921         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
 922
 923         len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
 924         memcpy(ctrl.data, uc, len[0]);
 925
 926         len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
 927         memcpy(ctrl.data + len[0], mc, len[1]);
 928
 929         err = virtio_send_command(hw->cvq, &ctrl, len, 2);
 930         if (err != 0)
 931                 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
 932 }
 933
 934 static void
 935 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
 936                     uint32_t index, uint32_t vmdq __rte_unused)
 937 {
 938         struct virtio_hw *hw = dev->data->dev_private;
 939         const struct ether_addr *addrs = dev->data->mac_addrs;
 940         unsigned int i;
 941         struct virtio_net_ctrl_mac *uc, *mc;
 942
 943         if (index >= VIRTIO_MAX_MAC_ADDRS) {
 944                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
 945                 return;
 946         }
 947
 948         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
 949         uc->entries = 0;
 950         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
 951         mc->entries = 0;
 952
 953         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
 954                 const struct ether_addr *addr
 955                         = (i == index) ? mac_addr : addrs + i;
 956                 struct virtio_net_ctrl_mac *tbl
 957                         = is_multicast_ether_addr(addr) ? mc : uc;
 958
 959                 memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
 960         }
 961
 962         virtio_mac_table_set(hw, uc, mc);
 963 }
 964
 965 static void
 966 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
 967 {
 968         struct virtio_hw *hw = dev->data->dev_private;
 969         struct ether_addr *addrs = dev->data->mac_addrs;
 970         struct virtio_net_ctrl_mac *uc, *mc;
 971         unsigned int i;
 972
 973         if (index >= VIRTIO_MAX_MAC_ADDRS) {
 974                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
 975                 return;
 976         }
 977
 978         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
 979         uc->entries = 0;
 980         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
 981         mc->entries = 0;
 982
 983         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
 984                 struct virtio_net_ctrl_mac *tbl;
 985
 986                 if (i == index || is_zero_ether_addr(addrs + i))
 987                         continue;
 988
 989                 tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
 990                 memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
 991         }
 992
 993         virtio_mac_table_set(hw, uc, mc);
 994 }
 995
 996 static void
 997 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 998 {
 999         struct virtio_hw *hw = dev->data->dev_private;
1000
1001         memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
1002
1003         /* Use atomic update if available */
1004         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1005                 struct virtio_pmd_ctrl ctrl;
1006                 int len = ETHER_ADDR_LEN;
1007
1008                 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1009                 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1010
1011                 memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
1012                 virtio_send_command(hw->cvq, &ctrl, &len, 1);
1013         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1014                 virtio_set_hwaddr(hw);
1015 }
1016
1017 static int
1018 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1019 {
1020         struct virtio_hw *hw = dev->data->dev_private;
1021         struct virtio_pmd_ctrl ctrl;
1022         int len;
1023
1024         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1025                 return -ENOTSUP;
1026
1027         ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1028         ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1029         memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1030         len = sizeof(vlan_id);
1031
1032         return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1033 }
1034
1035 static int
1036 virtio_negotiate_features(struct virtio_hw *hw)
1037 {
1038         uint64_t host_features;
1039
1040         /* Prepare guest_features: feature that driver wants to support */
1041         hw->guest_features = VIRTIO_PMD_GUEST_FEATURES;
1042         PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1043                 hw->guest_features);
1044
1045         /* Read device(host) feature bits */
1046         host_features = hw->vtpci_ops->get_features(hw);
1047         PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1048                 host_features);
1049
1050         /*
1051          * Negotiate features: Subset of device feature bits are written back
1052          * guest feature bits.
1053          */
1054         hw->guest_features = vtpci_negotiate_features(hw, host_features);
1055         PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1056                 hw->guest_features);
1057
1058         if (hw->modern) {
1059                 if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1060                         PMD_INIT_LOG(ERR,
1061                                 "VIRTIO_F_VERSION_1 features is not enabled.");
1062                         return -1;
1063                 }
1064                 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1065                 if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1066                         PMD_INIT_LOG(ERR,
1067                                 "failed to set FEATURES_OK status!");
1068                         return -1;
1069                 }
1070         }
1071
1072         return 0;
1073 }
1074
1075 /*
1076  * Process Virtio Config changed interrupt and call the callback
1077  * if link state changed.
1078  */
1079 static void
1080 virtio_interrupt_handler(__rte_unused struct rte_intr_handle *handle,
1081                          void *param)
1082 {
1083         struct rte_eth_dev *dev = param;
1084         struct virtio_hw *hw = dev->data->dev_private;
1085         uint8_t isr;
1086
1087         /* Read interrupt status which clears interrupt */
1088         isr = vtpci_isr(hw);
1089         PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1090
1091         if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0)
1092                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1093
1094         if (isr & VIRTIO_PCI_ISR_CONFIG) {
1095                 if (virtio_dev_link_update(dev, 0) == 0)
1096                         _rte_eth_dev_callback_process(dev,
1097                                                       RTE_ETH_EVENT_INTR_LSC);
1098         }
1099
1100 }
1101
1102 static void
1103 rx_func_get(struct rte_eth_dev *eth_dev)
1104 {
1105         struct virtio_hw *hw = eth_dev->data->dev_private;
1106         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF))
1107                 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1108         else
1109                 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1110 }
1111
1112 /*
1113  * This function is based on probe() function in virtio_pci.c
1114  * It returns 0 on success.
1115  */
1116 static int
1117 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1118 {
1119         struct virtio_hw *hw = eth_dev->data->dev_private;
1120         struct virtio_net_config *config;
1121         struct virtio_net_config local_config;
1122         struct rte_pci_device *pci_dev;
1123         uint32_t dev_flags = RTE_ETH_DEV_DETACHABLE;
1124         int ret;
1125
1126         RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr_mrg_rxbuf));
1127
1128         eth_dev->dev_ops = &virtio_eth_dev_ops;
1129         eth_dev->tx_pkt_burst = &virtio_xmit_pkts;
1130
1131         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1132                 rx_func_get(eth_dev);
1133                 return 0;
1134         }
1135
1136         /* Allocate memory for storing MAC addresses */
1137         eth_dev->data->mac_addrs = rte_zmalloc("virtio", VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN, 0);
1138         if (eth_dev->data->mac_addrs == NULL) {
1139                 PMD_INIT_LOG(ERR,
1140                         "Failed to allocate %d bytes needed to store MAC addresses",
1141                         VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN);
1142                 return -ENOMEM;
1143         }
1144
1145         pci_dev = eth_dev->pci_dev;
1146
1147         ret = vtpci_init(pci_dev, hw, &dev_flags);
1148         if (ret)
1149                 return ret;
1150
1151         /* Reset the device although not necessary at startup */
1152         vtpci_reset(hw);
1153
1154         /* Tell the host we've noticed this device. */
1155         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1156
1157         /* Tell the host we've known how to drive the device. */
1158         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1159         if (virtio_negotiate_features(hw) < 0)
1160                 return -1;
1161
1162         /* If host does not support status then disable LSC */
1163         if (!vtpci_with_feature(hw, VIRTIO_NET_F_STATUS))
1164                 dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1165
1166         rte_eth_copy_pci_info(eth_dev, pci_dev);
1167         eth_dev->data->dev_flags = dev_flags;
1168
1169         rx_func_get(eth_dev);
1170
1171         /* Setting up rx_header size for the device */
1172         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1173             vtpci_with_feature(hw, VIRTIO_F_VERSION_1))
1174                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1175         else
1176                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1177
1178         /* Copy the permanent MAC address to: virtio_hw */
1179         virtio_get_hwaddr(hw);
1180         ether_addr_copy((struct ether_addr *) hw->mac_addr,
1181                         &eth_dev->data->mac_addrs[0]);
1182         PMD_INIT_LOG(DEBUG,
1183                      "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1184                      hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1185                      hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1186
1187         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1188                 config = &local_config;
1189
1190                 vtpci_read_dev_config(hw,
1191                         offsetof(struct virtio_net_config, mac),
1192                         &config->mac, sizeof(config->mac));
1193
1194                 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1195                         vtpci_read_dev_config(hw,
1196                                 offsetof(struct virtio_net_config, status),
1197                                 &config->status, sizeof(config->status));
1198                 } else {
1199                         PMD_INIT_LOG(DEBUG,
1200                                      "VIRTIO_NET_F_STATUS is not supported");
1201                         config->status = 0;
1202                 }
1203
1204                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1205                         vtpci_read_dev_config(hw,
1206                                 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1207                                 &config->max_virtqueue_pairs,
1208                                 sizeof(config->max_virtqueue_pairs));
1209                 } else {
1210                         PMD_INIT_LOG(DEBUG,
1211                                      "VIRTIO_NET_F_MQ is not supported");
1212                         config->max_virtqueue_pairs = 1;
1213                 }
1214
1215                 hw->max_rx_queues =
1216                         (VIRTIO_MAX_RX_QUEUES < config->max_virtqueue_pairs) ?
1217                         VIRTIO_MAX_RX_QUEUES : config->max_virtqueue_pairs;
1218                 hw->max_tx_queues =
1219                         (VIRTIO_MAX_TX_QUEUES < config->max_virtqueue_pairs) ?
1220                         VIRTIO_MAX_TX_QUEUES : config->max_virtqueue_pairs;
1221
1222                 virtio_dev_cq_queue_setup(eth_dev,
1223                                         config->max_virtqueue_pairs * 2,
1224                                         SOCKET_ID_ANY);
1225
1226                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1227                                 config->max_virtqueue_pairs);
1228                 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1229                 PMD_INIT_LOG(DEBUG,
1230                                 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1231                                 config->mac[0], config->mac[1],
1232                                 config->mac[2], config->mac[3],
1233                                 config->mac[4], config->mac[5]);
1234         } else {
1235                 hw->max_rx_queues = 1;
1236                 hw->max_tx_queues = 1;
1237         }
1238
1239         PMD_INIT_LOG(DEBUG, "hw->max_rx_queues=%d   hw->max_tx_queues=%d",
1240                         hw->max_rx_queues, hw->max_tx_queues);
1241         PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1242                         eth_dev->data->port_id, pci_dev->id.vendor_id,
1243                         pci_dev->id.device_id);
1244
1245         /* Setup interrupt callback  */
1246         if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1247                 rte_intr_callback_register(&pci_dev->intr_handle,
1248                                    virtio_interrupt_handler, eth_dev);
1249
1250         virtio_dev_cq_start(eth_dev);
1251
1252         return 0;
1253 }
1254
1255 static int
1256 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1257 {
1258         struct rte_pci_device *pci_dev;
1259         struct virtio_hw *hw = eth_dev->data->dev_private;
1260
1261         PMD_INIT_FUNC_TRACE();
1262
1263         if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1264                 return -EPERM;
1265
1266         /* Close it anyway since there's no way to know if closed */
1267         virtio_dev_close(eth_dev);
1268
1269         pci_dev = eth_dev->pci_dev;
1270
1271         eth_dev->dev_ops = NULL;
1272         eth_dev->tx_pkt_burst = NULL;
1273         eth_dev->rx_pkt_burst = NULL;
1274
1275         if (hw->cvq)
1276                 virtio_dev_queue_release(hw->cvq->vq);
1277
1278         rte_free(eth_dev->data->mac_addrs);
1279         eth_dev->data->mac_addrs = NULL;
1280
1281         /* reset interrupt callback  */
1282         if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1283                 rte_intr_callback_unregister(&pci_dev->intr_handle,
1284                                                 virtio_interrupt_handler,
1285                                                 eth_dev);
1286         rte_eal_pci_unmap_device(pci_dev);
1287
1288         PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1289
1290         return 0;
1291 }
1292
1293 static struct eth_driver rte_virtio_pmd = {
1294         .pci_drv = {
1295                 .name = "rte_virtio_pmd",
1296                 .id_table = pci_id_virtio_map,
1297                 .drv_flags = RTE_PCI_DRV_DETACHABLE,
1298         },
1299         .eth_dev_init = eth_virtio_dev_init,
1300         .eth_dev_uninit = eth_virtio_dev_uninit,
1301         .dev_private_size = sizeof(struct virtio_hw),
1302 };
1303
1304 /*
1305  * Driver initialization routine.
1306  * Invoked once at EAL init time.
1307  * Register itself as the [Poll Mode] Driver of PCI virtio devices.
1308  * Returns 0 on success.
1309  */
1310 static int
1311 rte_virtio_pmd_init(const char *name __rte_unused,
1312                     const char *param __rte_unused)
1313 {
1314         if (rte_eal_iopl_init() != 0) {
1315                 PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1316                 return -1;
1317         }
1318
1319         rte_eth_driver_register(&rte_virtio_pmd);
1320         return 0;
1321 }
1322
1323 /*
1324  * Configure virtio device
1325  * It returns 0 on success.
1326  */
1327 static int
1328 virtio_dev_configure(struct rte_eth_dev *dev)
1329 {
1330         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1331         struct virtio_hw *hw = dev->data->dev_private;
1332
1333         PMD_INIT_LOG(DEBUG, "configure");
1334
1335         if (rxmode->hw_ip_checksum) {
1336                 PMD_DRV_LOG(ERR, "HW IP checksum not supported");
1337                 return -EINVAL;
1338         }
1339
1340         hw->vlan_strip = rxmode->hw_vlan_strip;
1341
1342         if (rxmode->hw_vlan_filter
1343             && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1344                 PMD_DRV_LOG(NOTICE,
1345                             "vlan filtering not available on this host");
1346                 return -ENOTSUP;
1347         }
1348
1349         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1350                 if (vtpci_irq_config(hw, 0) == VIRTIO_MSI_NO_VECTOR) {
1351                         PMD_DRV_LOG(ERR, "failed to set config vector");
1352                         return -EBUSY;
1353                 }
1354
1355         return 0;
1356 }
1357
1358
1359 static int
1360 virtio_dev_start(struct rte_eth_dev *dev)
1361 {
1362         uint16_t nb_queues, i;
1363         struct virtio_hw *hw = dev->data->dev_private;
1364         struct virtnet_rx *rxvq;
1365         struct virtnet_tx *txvq __rte_unused;
1366
1367         /* check if lsc interrupt feature is enabled */
1368         if (dev->data->dev_conf.intr_conf.lsc) {
1369                 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
1370                         PMD_DRV_LOG(ERR, "link status not supported by host");
1371                         return -ENOTSUP;
1372                 }
1373
1374                 if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0) {
1375                         PMD_DRV_LOG(ERR, "interrupt enable failed");
1376                         return -EIO;
1377                 }
1378         }
1379
1380         /* Initialize Link state */
1381         virtio_dev_link_update(dev, 0);
1382
1383         /* On restart after stop do not touch queues */
1384         if (hw->started)
1385                 return 0;
1386
1387         /* Do final configuration before rx/tx engine starts */
1388         virtio_dev_rxtx_start(dev);
1389         vtpci_reinit_complete(hw);
1390
1391         hw->started = 1;
1392
1393         /*Notify the backend
1394          *Otherwise the tap backend might already stop its queue due to fullness.
1395          *vhost backend will have no chance to be waked up
1396          */
1397         nb_queues = dev->data->nb_rx_queues;
1398         if (nb_queues > 1) {
1399                 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1400                         return -EINVAL;
1401         }
1402
1403         PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1404
1405         for (i = 0; i < nb_queues; i++) {
1406                 rxvq = dev->data->rx_queues[i];
1407                 virtqueue_notify(rxvq->vq);
1408         }
1409
1410         PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1411
1412         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1413                 rxvq = dev->data->rx_queues[i];
1414                 VIRTQUEUE_DUMP(rxvq->vq);
1415         }
1416
1417         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1418                 txvq = dev->data->tx_queues[i];
1419                 VIRTQUEUE_DUMP(txvq->vq);
1420         }
1421
1422         return 0;
1423 }
1424
1425 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1426 {
1427         struct rte_mbuf *buf;
1428         int i, mbuf_num = 0;
1429
1430         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1431                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1432
1433                 PMD_INIT_LOG(DEBUG,
1434                              "Before freeing rxq[%d] used and unused buf", i);
1435                 VIRTQUEUE_DUMP(rxvq->vq);
1436
1437                 PMD_INIT_LOG(DEBUG, "rx_queues[%d]=%p", i, rxvq);
1438                 while ((buf = virtqueue_detatch_unused(rxvq->vq)) != NULL) {
1439                         rte_pktmbuf_free(buf);
1440                         mbuf_num++;
1441                 }
1442
1443                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1444                 PMD_INIT_LOG(DEBUG,
1445                              "After freeing rxq[%d] used and unused buf", i);
1446                 VIRTQUEUE_DUMP(rxvq->vq);
1447         }
1448
1449         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1450                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1451
1452                 PMD_INIT_LOG(DEBUG,
1453                              "Before freeing txq[%d] used and unused bufs",
1454                              i);
1455                 VIRTQUEUE_DUMP(txvq->vq);
1456
1457                 mbuf_num = 0;
1458                 while ((buf = virtqueue_detatch_unused(txvq->vq)) != NULL) {
1459                         rte_pktmbuf_free(buf);
1460                         mbuf_num++;
1461                 }
1462
1463                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1464                 PMD_INIT_LOG(DEBUG,
1465                              "After freeing txq[%d] used and unused buf", i);
1466                 VIRTQUEUE_DUMP(txvq->vq);
1467         }
1468 }
1469
1470 /*
1471  * Stop device: disable interrupt and mark link down
1472  */
1473 static void
1474 virtio_dev_stop(struct rte_eth_dev *dev)
1475 {
1476         struct rte_eth_link link;
1477         struct virtio_hw *hw = dev->data->dev_private;
1478
1479         PMD_INIT_LOG(DEBUG, "stop");
1480
1481         hw->started = 0;
1482
1483         if (dev->data->dev_conf.intr_conf.lsc)
1484                 rte_intr_disable(&dev->pci_dev->intr_handle);
1485
1486         memset(&link, 0, sizeof(link));
1487         virtio_dev_atomic_write_link_status(dev, &link);
1488 }
1489
1490 static int
1491 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
1492 {
1493         struct rte_eth_link link, old;
1494         uint16_t status;
1495         struct virtio_hw *hw = dev->data->dev_private;
1496         memset(&link, 0, sizeof(link));
1497         virtio_dev_atomic_read_link_status(dev, &link);
1498         old = link;
1499         link.link_duplex = ETH_LINK_FULL_DUPLEX;
1500         link.link_speed  = SPEED_10G;
1501
1502         if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1503                 PMD_INIT_LOG(DEBUG, "Get link status from hw");
1504                 vtpci_read_dev_config(hw,
1505                                 offsetof(struct virtio_net_config, status),
1506                                 &status, sizeof(status));
1507                 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
1508                         link.link_status = ETH_LINK_DOWN;
1509                         PMD_INIT_LOG(DEBUG, "Port %d is down",
1510                                      dev->data->port_id);
1511                 } else {
1512                         link.link_status = ETH_LINK_UP;
1513                         PMD_INIT_LOG(DEBUG, "Port %d is up",
1514                                      dev->data->port_id);
1515                 }
1516         } else {
1517                 link.link_status = ETH_LINK_UP;
1518         }
1519         virtio_dev_atomic_write_link_status(dev, &link);
1520
1521         return (old.link_status == link.link_status) ? -1 : 0;
1522 }
1523
1524 static void
1525 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1526 {
1527         struct virtio_hw *hw = dev->data->dev_private;
1528
1529         dev_info->driver_name = dev->driver->pci_drv.name;
1530         dev_info->max_rx_queues = (uint16_t)hw->max_rx_queues;
1531         dev_info->max_tx_queues = (uint16_t)hw->max_tx_queues;
1532         dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
1533         dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
1534         dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
1535         dev_info->default_txconf = (struct rte_eth_txconf) {
1536                 .txq_flags = ETH_TXQ_FLAGS_NOOFFLOADS
1537         };
1538 }
1539
1540 /*
1541  * It enables testpmd to collect per queue stats.
1542  */
1543 static int
1544 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
1545 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
1546 __rte_unused uint8_t is_rx)
1547 {
1548         return 0;
1549 }
1550
1551 static struct rte_driver rte_virtio_driver = {
1552         .type = PMD_PDEV,
1553         .init = rte_virtio_pmd_init,
1554 };
1555
1556 PMD_REGISTER_DRIVER(rte_virtio_driver);