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_uninit(struct rte_eth_dev *eth_dev);
  63 static int  virtio_dev_configure(struct rte_eth_dev *dev);
  64 static int  virtio_dev_start(struct rte_eth_dev *dev);
  65 static void virtio_dev_stop(struct rte_eth_dev *dev);
  66 static void virtio_dev_promiscuous_enable(struct rte_eth_dev *dev);
  67 static void virtio_dev_promiscuous_disable(struct rte_eth_dev *dev);
  68 static void virtio_dev_allmulticast_enable(struct rte_eth_dev *dev);
  69 static void virtio_dev_allmulticast_disable(struct rte_eth_dev *dev);
  70 static void virtio_dev_info_get(struct rte_eth_dev *dev,
  71                                 struct rte_eth_dev_info *dev_info);
  72 static int virtio_dev_link_update(struct rte_eth_dev *dev,
  73         __rte_unused int wait_to_complete);
  74
  75 static void virtio_set_hwaddr(struct virtio_hw *hw);
  76 static void virtio_get_hwaddr(struct virtio_hw *hw);
  77
  78 static void virtio_dev_stats_get(struct rte_eth_dev *dev,
  79                                  struct rte_eth_stats *stats);
  80 static int virtio_dev_xstats_get(struct rte_eth_dev *dev,
  81                                  struct rte_eth_xstat *xstats, unsigned n);
  82 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
  83                                        struct rte_eth_xstat_name *xstats_names,
  84                                        unsigned limit);
  85 static void virtio_dev_stats_reset(struct rte_eth_dev *dev);
  86 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev);
  87 static int virtio_vlan_filter_set(struct rte_eth_dev *dev,
  88                                 uint16_t vlan_id, int on);
  89 static void virtio_mac_addr_add(struct rte_eth_dev *dev,
  90                                 struct ether_addr *mac_addr,
  91                                 uint32_t index, uint32_t vmdq __rte_unused);
  92 static void virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index);
  93 static void virtio_mac_addr_set(struct rte_eth_dev *dev,
  94                                 struct ether_addr *mac_addr);
  95
  96 static int virtio_dev_queue_stats_mapping_set(
  97         __rte_unused struct rte_eth_dev *eth_dev,
  98         __rte_unused uint16_t queue_id,
  99         __rte_unused uint8_t stat_idx,
 100         __rte_unused uint8_t is_rx);
 101
 102 /*
 103  * The set of PCI devices this driver supports
 104  */
 105 static const struct rte_pci_id pci_id_virtio_map[] = {
 106         { RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_LEGACY_DEVICEID_NET) },
 107         { RTE_PCI_DEVICE(VIRTIO_PCI_VENDORID, VIRTIO_PCI_MODERN_DEVICEID_NET) },
 108         { .vendor_id = 0, /* sentinel */ },
 109 };
 110
 111 struct rte_virtio_xstats_name_off {
 112         char name[RTE_ETH_XSTATS_NAME_SIZE];
 113         unsigned offset;
 114 };
 115
 116 /* [rt]x_qX_ is prepended to the name string here */
 117 static const struct rte_virtio_xstats_name_off rte_virtio_rxq_stat_strings[] = {
 118         {"good_packets",           offsetof(struct virtnet_rx, stats.packets)},
 119         {"good_bytes",             offsetof(struct virtnet_rx, stats.bytes)},
 120         {"errors",                 offsetof(struct virtnet_rx, stats.errors)},
 121         {"multicast_packets",      offsetof(struct virtnet_rx, stats.multicast)},
 122         {"broadcast_packets",      offsetof(struct virtnet_rx, stats.broadcast)},
 123         {"undersize_packets",      offsetof(struct virtnet_rx, stats.size_bins[0])},
 124         {"size_64_packets",        offsetof(struct virtnet_rx, stats.size_bins[1])},
 125         {"size_65_127_packets",    offsetof(struct virtnet_rx, stats.size_bins[2])},
 126         {"size_128_255_packets",   offsetof(struct virtnet_rx, stats.size_bins[3])},
 127         {"size_256_511_packets",   offsetof(struct virtnet_rx, stats.size_bins[4])},
 128         {"size_512_1023_packets",  offsetof(struct virtnet_rx, stats.size_bins[5])},
 129         {"size_1024_1518_packets", offsetof(struct virtnet_rx, stats.size_bins[6])},
 130         {"size_1519_max_packets",  offsetof(struct virtnet_rx, stats.size_bins[7])},
 131 };
 132
 133 /* [rt]x_qX_ is prepended to the name string here */
 134 static const struct rte_virtio_xstats_name_off rte_virtio_txq_stat_strings[] = {
 135         {"good_packets",           offsetof(struct virtnet_tx, stats.packets)},
 136         {"good_bytes",             offsetof(struct virtnet_tx, stats.bytes)},
 137         {"errors",                 offsetof(struct virtnet_tx, stats.errors)},
 138         {"multicast_packets",      offsetof(struct virtnet_tx, stats.multicast)},
 139         {"broadcast_packets",      offsetof(struct virtnet_tx, stats.broadcast)},
 140         {"undersize_packets",      offsetof(struct virtnet_tx, stats.size_bins[0])},
 141         {"size_64_packets",        offsetof(struct virtnet_tx, stats.size_bins[1])},
 142         {"size_65_127_packets",    offsetof(struct virtnet_tx, stats.size_bins[2])},
 143         {"size_128_255_packets",   offsetof(struct virtnet_tx, stats.size_bins[3])},
 144         {"size_256_511_packets",   offsetof(struct virtnet_tx, stats.size_bins[4])},
 145         {"size_512_1023_packets",  offsetof(struct virtnet_tx, stats.size_bins[5])},
 146         {"size_1024_1518_packets", offsetof(struct virtnet_tx, stats.size_bins[6])},
 147         {"size_1519_max_packets",  offsetof(struct virtnet_tx, stats.size_bins[7])},
 148 };
 149
 150 #define VIRTIO_NB_RXQ_XSTATS (sizeof(rte_virtio_rxq_stat_strings) / \
 151                             sizeof(rte_virtio_rxq_stat_strings[0]))
 152 #define VIRTIO_NB_TXQ_XSTATS (sizeof(rte_virtio_txq_stat_strings) / \
 153                             sizeof(rte_virtio_txq_stat_strings[0]))
 154
 155 static int
 156 virtio_send_command(struct virtnet_ctl *cvq, struct virtio_pmd_ctrl *ctrl,
 157                 int *dlen, int pkt_num)
 158 {
 159         uint32_t head, i;
 160         int k, sum = 0;
 161         virtio_net_ctrl_ack status = ~0;
 162         struct virtio_pmd_ctrl result;
 163         struct virtqueue *vq;
 164
 165         ctrl->status = status;
 166
 167         if (!cvq || !cvq->vq) {
 168                 PMD_INIT_LOG(ERR, "Control queue is not supported.");
 169                 return -1;
 170         }
 171         vq = cvq->vq;
 172         head = vq->vq_desc_head_idx;
 173
 174         PMD_INIT_LOG(DEBUG, "vq->vq_desc_head_idx = %d, status = %d, "
 175                 "vq->hw->cvq = %p vq = %p",
 176                 vq->vq_desc_head_idx, status, vq->hw->cvq, vq);
 177
 178         if ((vq->vq_free_cnt < ((uint32_t)pkt_num + 2)) || (pkt_num < 1))
 179                 return -1;
 180
 181         memcpy(cvq->virtio_net_hdr_mz->addr, ctrl,
 182                 sizeof(struct virtio_pmd_ctrl));
 183
 184         /*
 185          * Format is enforced in qemu code:
 186          * One TX packet for header;
 187          * At least one TX packet per argument;
 188          * One RX packet for ACK.
 189          */
 190         vq->vq_ring.desc[head].flags = VRING_DESC_F_NEXT;
 191         vq->vq_ring.desc[head].addr = cvq->virtio_net_hdr_mem;
 192         vq->vq_ring.desc[head].len = sizeof(struct virtio_net_ctrl_hdr);
 193         vq->vq_free_cnt--;
 194         i = vq->vq_ring.desc[head].next;
 195
 196         for (k = 0; k < pkt_num; k++) {
 197                 vq->vq_ring.desc[i].flags = VRING_DESC_F_NEXT;
 198                 vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
 199                         + sizeof(struct virtio_net_ctrl_hdr)
 200                         + sizeof(ctrl->status) + sizeof(uint8_t)*sum;
 201                 vq->vq_ring.desc[i].len = dlen[k];
 202                 sum += dlen[k];
 203                 vq->vq_free_cnt--;
 204                 i = vq->vq_ring.desc[i].next;
 205         }
 206
 207         vq->vq_ring.desc[i].flags = VRING_DESC_F_WRITE;
 208         vq->vq_ring.desc[i].addr = cvq->virtio_net_hdr_mem
 209                         + sizeof(struct virtio_net_ctrl_hdr);
 210         vq->vq_ring.desc[i].len = sizeof(ctrl->status);
 211         vq->vq_free_cnt--;
 212
 213         vq->vq_desc_head_idx = vq->vq_ring.desc[i].next;
 214
 215         vq_update_avail_ring(vq, head);
 216         vq_update_avail_idx(vq);
 217
 218         PMD_INIT_LOG(DEBUG, "vq->vq_queue_index = %d", vq->vq_queue_index);
 219
 220         virtqueue_notify(vq);
 221
 222         rte_rmb();
 223         while (VIRTQUEUE_NUSED(vq) == 0) {
 224                 rte_rmb();
 225                 usleep(100);
 226         }
 227
 228         while (VIRTQUEUE_NUSED(vq)) {
 229                 uint32_t idx, desc_idx, used_idx;
 230                 struct vring_used_elem *uep;
 231
 232                 used_idx = (uint32_t)(vq->vq_used_cons_idx
 233                                 & (vq->vq_nentries - 1));
 234                 uep = &vq->vq_ring.used->ring[used_idx];
 235                 idx = (uint32_t) uep->id;
 236                 desc_idx = idx;
 237
 238                 while (vq->vq_ring.desc[desc_idx].flags & VRING_DESC_F_NEXT) {
 239                         desc_idx = vq->vq_ring.desc[desc_idx].next;
 240                         vq->vq_free_cnt++;
 241                 }
 242
 243                 vq->vq_ring.desc[desc_idx].next = vq->vq_desc_head_idx;
 244                 vq->vq_desc_head_idx = idx;
 245
 246                 vq->vq_used_cons_idx++;
 247                 vq->vq_free_cnt++;
 248         }
 249
 250         PMD_INIT_LOG(DEBUG, "vq->vq_free_cnt=%d\nvq->vq_desc_head_idx=%d",
 251                         vq->vq_free_cnt, vq->vq_desc_head_idx);
 252
 253         memcpy(&result, cvq->virtio_net_hdr_mz->addr,
 254                         sizeof(struct virtio_pmd_ctrl));
 255
 256         return result.status;
 257 }
 258
 259 static int
 260 virtio_set_multiple_queues(struct rte_eth_dev *dev, uint16_t nb_queues)
 261 {
 262         struct virtio_hw *hw = dev->data->dev_private;
 263         struct virtio_pmd_ctrl ctrl;
 264         int dlen[1];
 265         int ret;
 266
 267         ctrl.hdr.class = VIRTIO_NET_CTRL_MQ;
 268         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET;
 269         memcpy(ctrl.data, &nb_queues, sizeof(uint16_t));
 270
 271         dlen[0] = sizeof(uint16_t);
 272
 273         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 274         if (ret) {
 275                 PMD_INIT_LOG(ERR, "Multiqueue configured but send command "
 276                           "failed, this is too late now...");
 277                 return -EINVAL;
 278         }
 279
 280         return 0;
 281 }
 282
 283 void
 284 virtio_dev_queue_release(struct virtqueue *vq)
 285 {
 286         struct virtio_hw *hw;
 287
 288         if (vq) {
 289                 hw = vq->hw;
 290                 if (vq->configured)
 291                         hw->vtpci_ops->del_queue(hw, vq);
 292
 293                 rte_free(vq->sw_ring);
 294                 rte_free(vq);
 295         }
 296 }
 297
 298 int virtio_dev_queue_setup(struct rte_eth_dev *dev,
 299                         int queue_type,
 300                         uint16_t queue_idx,
 301                         uint16_t vtpci_queue_idx,
 302                         uint16_t nb_desc,
 303                         unsigned int socket_id,
 304                         void **pvq)
 305 {
 306         char vq_name[VIRTQUEUE_MAX_NAME_SZ];
 307         char vq_hdr_name[VIRTQUEUE_MAX_NAME_SZ];
 308         const struct rte_memzone *mz = NULL, *hdr_mz = NULL;
 309         unsigned int vq_size, size;
 310         struct virtio_hw *hw = dev->data->dev_private;
 311         struct virtnet_rx *rxvq = NULL;
 312         struct virtnet_tx *txvq = NULL;
 313         struct virtnet_ctl *cvq = NULL;
 314         struct virtqueue *vq;
 315         const char *queue_names[] = {"rvq", "txq", "cvq"};
 316         size_t sz_vq, sz_q = 0, sz_hdr_mz = 0;
 317         void *sw_ring = NULL;
 318         int ret;
 319
 320         PMD_INIT_LOG(DEBUG, "setting up queue: %u", vtpci_queue_idx);
 321
 322         /*
 323          * Read the virtqueue size from the Queue Size field
 324          * Always power of 2 and if 0 virtqueue does not exist
 325          */
 326         vq_size = hw->vtpci_ops->get_queue_num(hw, vtpci_queue_idx);
 327         PMD_INIT_LOG(DEBUG, "vq_size: %u nb_desc:%u", vq_size, nb_desc);
 328         if (vq_size == 0) {
 329                 PMD_INIT_LOG(ERR, "virtqueue does not exist");
 330                 return -EINVAL;
 331         }
 332
 333         if (!rte_is_power_of_2(vq_size)) {
 334                 PMD_INIT_LOG(ERR, "virtqueue size is not powerof 2");
 335                 return -EINVAL;
 336         }
 337
 338         snprintf(vq_name, sizeof(vq_name), "port%d_%s%d",
 339                  dev->data->port_id, queue_names[queue_type], queue_idx);
 340
 341         sz_vq = RTE_ALIGN_CEIL(sizeof(*vq) +
 342                                 vq_size * sizeof(struct vq_desc_extra),
 343                                 RTE_CACHE_LINE_SIZE);
 344         if (queue_type == VTNET_RQ) {
 345                 sz_q = sz_vq + sizeof(*rxvq);
 346         } else if (queue_type == VTNET_TQ) {
 347                 sz_q = sz_vq + sizeof(*txvq);
 348                 /*
 349                  * For each xmit packet, allocate a virtio_net_hdr
 350                  * and indirect ring elements
 351                  */
 352                 sz_hdr_mz = vq_size * sizeof(struct virtio_tx_region);
 353         } else if (queue_type == VTNET_CQ) {
 354                 sz_q = sz_vq + sizeof(*cvq);
 355                 /* Allocate a page for control vq command, data and status */
 356                 sz_hdr_mz = PAGE_SIZE;
 357         }
 358
 359         vq = rte_zmalloc_socket(vq_name, sz_q, RTE_CACHE_LINE_SIZE, socket_id);
 360         if (vq == NULL) {
 361                 PMD_INIT_LOG(ERR, "can not allocate vq");
 362                 return -ENOMEM;
 363         }
 364         vq->hw = hw;
 365         vq->vq_queue_index = vtpci_queue_idx;
 366         vq->vq_nentries = vq_size;
 367
 368         if (nb_desc == 0 || nb_desc > vq_size)
 369                 nb_desc = vq_size;
 370         vq->vq_free_cnt = nb_desc;
 371
 372         /*
 373          * Reserve a memzone for vring elements
 374          */
 375         size = vring_size(vq_size, VIRTIO_PCI_VRING_ALIGN);
 376         vq->vq_ring_size = RTE_ALIGN_CEIL(size, VIRTIO_PCI_VRING_ALIGN);
 377         PMD_INIT_LOG(DEBUG, "vring_size: %d, rounded_vring_size: %d",
 378                      size, vq->vq_ring_size);
 379
 380         mz = rte_memzone_reserve_aligned(vq_name, vq->vq_ring_size, socket_id,
 381                                          0, VIRTIO_PCI_VRING_ALIGN);
 382         if (mz == NULL) {
 383                 if (rte_errno == EEXIST)
 384                         mz = rte_memzone_lookup(vq_name);
 385                 if (mz == NULL) {
 386                         ret = -ENOMEM;
 387                         goto fail_q_alloc;
 388                 }
 389         }
 390
 391         memset(mz->addr, 0, sizeof(mz->len));
 392
 393         vq->vq_ring_mem = mz->phys_addr;
 394         vq->vq_ring_virt_mem = mz->addr;
 395         PMD_INIT_LOG(DEBUG, "vq->vq_ring_mem:      0x%" PRIx64,
 396                      (uint64_t)mz->phys_addr);
 397         PMD_INIT_LOG(DEBUG, "vq->vq_ring_virt_mem: 0x%" PRIx64,
 398                      (uint64_t)(uintptr_t)mz->addr);
 399
 400         if (sz_hdr_mz) {
 401                 snprintf(vq_hdr_name, sizeof(vq_hdr_name), "port%d_%s%d_hdr",
 402                          dev->data->port_id, queue_names[queue_type],
 403                          queue_idx);
 404                 hdr_mz = rte_memzone_reserve_aligned(vq_hdr_name, sz_hdr_mz,
 405                                                      socket_id, 0,
 406                                                      RTE_CACHE_LINE_SIZE);
 407                 if (hdr_mz == NULL) {
 408                         if (rte_errno == EEXIST)
 409                                 hdr_mz = rte_memzone_lookup(vq_hdr_name);
 410                         if (hdr_mz == NULL) {
 411                                 ret = -ENOMEM;
 412                                 goto fail_q_alloc;
 413                         }
 414                 }
 415         }
 416
 417         if (queue_type == VTNET_RQ) {
 418                 size_t sz_sw = (RTE_PMD_VIRTIO_RX_MAX_BURST + vq_size) *
 419                                sizeof(vq->sw_ring[0]);
 420
 421                 sw_ring = rte_zmalloc_socket("sw_ring", sz_sw,
 422                                              RTE_CACHE_LINE_SIZE, socket_id);
 423                 if (!sw_ring) {
 424                         PMD_INIT_LOG(ERR, "can not allocate RX soft ring");
 425                         ret = -ENOMEM;
 426                         goto fail_q_alloc;
 427                 }
 428
 429                 vq->sw_ring = sw_ring;
 430                 rxvq = (struct virtnet_rx *)RTE_PTR_ADD(vq, sz_vq);
 431                 rxvq->vq = vq;
 432                 rxvq->port_id = dev->data->port_id;
 433                 rxvq->queue_id = queue_idx;
 434                 rxvq->mz = mz;
 435                 *pvq = rxvq;
 436         } else if (queue_type == VTNET_TQ) {
 437                 txvq = (struct virtnet_tx *)RTE_PTR_ADD(vq, sz_vq);
 438                 txvq->vq = vq;
 439                 txvq->port_id = dev->data->port_id;
 440                 txvq->queue_id = queue_idx;
 441                 txvq->mz = mz;
 442                 txvq->virtio_net_hdr_mz = hdr_mz;
 443                 txvq->virtio_net_hdr_mem = hdr_mz->phys_addr;
 444
 445                 *pvq = txvq;
 446         } else if (queue_type == VTNET_CQ) {
 447                 cvq = (struct virtnet_ctl *)RTE_PTR_ADD(vq, sz_vq);
 448                 cvq->vq = vq;
 449                 cvq->mz = mz;
 450                 cvq->virtio_net_hdr_mz = hdr_mz;
 451                 cvq->virtio_net_hdr_mem = hdr_mz->phys_addr;
 452                 memset(cvq->virtio_net_hdr_mz->addr, 0, PAGE_SIZE);
 453                 *pvq = cvq;
 454         }
 455
 456         /* For virtio_user case (that is when dev->pci_dev is NULL), we use
 457          * virtual address. And we need properly set _offset_, please see
 458          * VIRTIO_MBUF_DATA_DMA_ADDR in virtqueue.h for more information.
 459          */
 460         if (dev->pci_dev)
 461                 vq->offset = offsetof(struct rte_mbuf, buf_physaddr);
 462         else {
 463                 vq->vq_ring_mem = (uintptr_t)mz->addr;
 464                 vq->offset = offsetof(struct rte_mbuf, buf_addr);
 465                 if (queue_type == VTNET_TQ)
 466                         txvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
 467                 else if (queue_type == VTNET_CQ)
 468                         cvq->virtio_net_hdr_mem = (uintptr_t)hdr_mz->addr;
 469         }
 470
 471         if (queue_type == VTNET_TQ) {
 472                 struct virtio_tx_region *txr;
 473                 unsigned int i;
 474
 475                 txr = hdr_mz->addr;
 476                 memset(txr, 0, vq_size * sizeof(*txr));
 477                 for (i = 0; i < vq_size; i++) {
 478                         struct vring_desc *start_dp = txr[i].tx_indir;
 479
 480                         vring_desc_init(start_dp, RTE_DIM(txr[i].tx_indir));
 481
 482                         /* first indirect descriptor is always the tx header */
 483                         start_dp->addr = txvq->virtio_net_hdr_mem
 484                                 + i * sizeof(*txr)
 485                                 + offsetof(struct virtio_tx_region, tx_hdr);
 486
 487                         start_dp->len = hw->vtnet_hdr_size;
 488                         start_dp->flags = VRING_DESC_F_NEXT;
 489                 }
 490         }
 491
 492         if (hw->vtpci_ops->setup_queue(hw, vq) < 0) {
 493                 PMD_INIT_LOG(ERR, "setup_queue failed");
 494                 virtio_dev_queue_release(vq);
 495                 return -EINVAL;
 496         }
 497
 498         vq->configured = 1;
 499         return 0;
 500
 501 fail_q_alloc:
 502         rte_free(sw_ring);
 503         rte_memzone_free(hdr_mz);
 504         rte_memzone_free(mz);
 505         rte_free(vq);
 506
 507         return ret;
 508 }
 509
 510 static int
 511 virtio_dev_cq_queue_setup(struct rte_eth_dev *dev, uint16_t vtpci_queue_idx,
 512                 uint32_t socket_id)
 513 {
 514         struct virtnet_ctl *cvq;
 515         int ret;
 516         struct virtio_hw *hw = dev->data->dev_private;
 517
 518         PMD_INIT_FUNC_TRACE();
 519         ret = virtio_dev_queue_setup(dev, VTNET_CQ, VTNET_SQ_CQ_QUEUE_IDX,
 520                         vtpci_queue_idx, 0, socket_id, (void **)&cvq);
 521         if (ret < 0) {
 522                 PMD_INIT_LOG(ERR, "control vq initialization failed");
 523                 return ret;
 524         }
 525
 526         hw->cvq = cvq;
 527         return 0;
 528 }
 529
 530 static void
 531 virtio_free_queues(struct rte_eth_dev *dev)
 532 {
 533         unsigned int i;
 534
 535         for (i = 0; i < dev->data->nb_rx_queues; i++)
 536                 virtio_dev_rx_queue_release(dev->data->rx_queues[i]);
 537
 538         dev->data->nb_rx_queues = 0;
 539
 540         for (i = 0; i < dev->data->nb_tx_queues; i++)
 541                 virtio_dev_tx_queue_release(dev->data->tx_queues[i]);
 542
 543         dev->data->nb_tx_queues = 0;
 544 }
 545
 546 static void
 547 virtio_dev_close(struct rte_eth_dev *dev)
 548 {
 549         struct virtio_hw *hw = dev->data->dev_private;
 550
 551         PMD_INIT_LOG(DEBUG, "virtio_dev_close");
 552
 553         if (hw->started == 1)
 554                 virtio_dev_stop(dev);
 555
 556         /* reset the NIC */
 557         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
 558                 vtpci_irq_config(hw, VIRTIO_MSI_NO_VECTOR);
 559         vtpci_reset(hw);
 560         virtio_dev_free_mbufs(dev);
 561         virtio_free_queues(dev);
 562 }
 563
 564 static void
 565 virtio_dev_promiscuous_enable(struct rte_eth_dev *dev)
 566 {
 567         struct virtio_hw *hw = dev->data->dev_private;
 568         struct virtio_pmd_ctrl ctrl;
 569         int dlen[1];
 570         int ret;
 571
 572         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 573                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 574                 return;
 575         }
 576
 577         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 578         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 579         ctrl.data[0] = 1;
 580         dlen[0] = 1;
 581
 582         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 583         if (ret)
 584                 PMD_INIT_LOG(ERR, "Failed to enable promisc");
 585 }
 586
 587 static void
 588 virtio_dev_promiscuous_disable(struct rte_eth_dev *dev)
 589 {
 590         struct virtio_hw *hw = dev->data->dev_private;
 591         struct virtio_pmd_ctrl ctrl;
 592         int dlen[1];
 593         int ret;
 594
 595         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 596                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 597                 return;
 598         }
 599
 600         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 601         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_PROMISC;
 602         ctrl.data[0] = 0;
 603         dlen[0] = 1;
 604
 605         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 606         if (ret)
 607                 PMD_INIT_LOG(ERR, "Failed to disable promisc");
 608 }
 609
 610 static void
 611 virtio_dev_allmulticast_enable(struct rte_eth_dev *dev)
 612 {
 613         struct virtio_hw *hw = dev->data->dev_private;
 614         struct virtio_pmd_ctrl ctrl;
 615         int dlen[1];
 616         int ret;
 617
 618         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 619                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 620                 return;
 621         }
 622
 623         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 624         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 625         ctrl.data[0] = 1;
 626         dlen[0] = 1;
 627
 628         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 629         if (ret)
 630                 PMD_INIT_LOG(ERR, "Failed to enable allmulticast");
 631 }
 632
 633 static void
 634 virtio_dev_allmulticast_disable(struct rte_eth_dev *dev)
 635 {
 636         struct virtio_hw *hw = dev->data->dev_private;
 637         struct virtio_pmd_ctrl ctrl;
 638         int dlen[1];
 639         int ret;
 640
 641         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_RX)) {
 642                 PMD_INIT_LOG(INFO, "host does not support rx control\n");
 643                 return;
 644         }
 645
 646         ctrl.hdr.class = VIRTIO_NET_CTRL_RX;
 647         ctrl.hdr.cmd = VIRTIO_NET_CTRL_RX_ALLMULTI;
 648         ctrl.data[0] = 0;
 649         dlen[0] = 1;
 650
 651         ret = virtio_send_command(hw->cvq, &ctrl, dlen, 1);
 652         if (ret)
 653                 PMD_INIT_LOG(ERR, "Failed to disable allmulticast");
 654 }
 655
 656 #define VLAN_TAG_LEN           4    /* 802.3ac tag (not DMA'd) */
 657 static int
 658 virtio_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 659 {
 660         struct virtio_hw *hw = dev->data->dev_private;
 661         uint32_t ether_hdr_len = ETHER_HDR_LEN + VLAN_TAG_LEN +
 662                                  hw->vtnet_hdr_size;
 663         uint32_t frame_size = mtu + ether_hdr_len;
 664
 665         if (mtu < ETHER_MIN_MTU || frame_size > VIRTIO_MAX_RX_PKTLEN) {
 666                 PMD_INIT_LOG(ERR, "MTU should be between %d and %d\n",
 667                         ETHER_MIN_MTU, VIRTIO_MAX_RX_PKTLEN - ether_hdr_len);
 668                 return -EINVAL;
 669         }
 670         return 0;
 671 }
 672
 673 /*
 674  * dev_ops for virtio, bare necessities for basic operation
 675  */
 676 static const struct eth_dev_ops virtio_eth_dev_ops = {
 677         .dev_configure           = virtio_dev_configure,
 678         .dev_start               = virtio_dev_start,
 679         .dev_stop                = virtio_dev_stop,
 680         .dev_close               = virtio_dev_close,
 681         .promiscuous_enable      = virtio_dev_promiscuous_enable,
 682         .promiscuous_disable     = virtio_dev_promiscuous_disable,
 683         .allmulticast_enable     = virtio_dev_allmulticast_enable,
 684         .allmulticast_disable    = virtio_dev_allmulticast_disable,
 685         .mtu_set                 = virtio_mtu_set,
 686         .dev_infos_get           = virtio_dev_info_get,
 687         .stats_get               = virtio_dev_stats_get,
 688         .xstats_get              = virtio_dev_xstats_get,
 689         .xstats_get_names        = virtio_dev_xstats_get_names,
 690         .stats_reset             = virtio_dev_stats_reset,
 691         .xstats_reset            = virtio_dev_stats_reset,
 692         .link_update             = virtio_dev_link_update,
 693         .rx_queue_setup          = virtio_dev_rx_queue_setup,
 694         .rx_queue_release        = virtio_dev_rx_queue_release,
 695         .tx_queue_setup          = virtio_dev_tx_queue_setup,
 696         .tx_queue_release        = virtio_dev_tx_queue_release,
 697         /* collect stats per queue */
 698         .queue_stats_mapping_set = virtio_dev_queue_stats_mapping_set,
 699         .vlan_filter_set         = virtio_vlan_filter_set,
 700         .mac_addr_add            = virtio_mac_addr_add,
 701         .mac_addr_remove         = virtio_mac_addr_remove,
 702         .mac_addr_set            = virtio_mac_addr_set,
 703 };
 704
 705 static inline int
 706 virtio_dev_atomic_read_link_status(struct rte_eth_dev *dev,
 707                                 struct rte_eth_link *link)
 708 {
 709         struct rte_eth_link *dst = link;
 710         struct rte_eth_link *src = &(dev->data->dev_link);
 711
 712         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
 713                         *(uint64_t *)src) == 0)
 714                 return -1;
 715
 716         return 0;
 717 }
 718
 719 /**
 720  * Atomically writes the link status information into global
 721  * structure rte_eth_dev.
 722  *
 723  * @param dev
 724  *   - Pointer to the structure rte_eth_dev to read from.
 725  *   - Pointer to the buffer to be saved with the link status.
 726  *
 727  * @return
 728  *   - On success, zero.
 729  *   - On failure, negative value.
 730  */
 731 static inline int
 732 virtio_dev_atomic_write_link_status(struct rte_eth_dev *dev,
 733                 struct rte_eth_link *link)
 734 {
 735         struct rte_eth_link *dst = &(dev->data->dev_link);
 736         struct rte_eth_link *src = link;
 737
 738         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
 739                                         *(uint64_t *)src) == 0)
 740                 return -1;
 741
 742         return 0;
 743 }
 744
 745 static void
 746 virtio_update_stats(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 747 {
 748         unsigned i;
 749
 750         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 751                 const struct virtnet_tx *txvq = dev->data->tx_queues[i];
 752                 if (txvq == NULL)
 753                         continue;
 754
 755                 stats->opackets += txvq->stats.packets;
 756                 stats->obytes += txvq->stats.bytes;
 757                 stats->oerrors += txvq->stats.errors;
 758
 759                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 760                         stats->q_opackets[i] = txvq->stats.packets;
 761                         stats->q_obytes[i] = txvq->stats.bytes;
 762                 }
 763         }
 764
 765         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 766                 const struct virtnet_rx *rxvq = dev->data->rx_queues[i];
 767                 if (rxvq == NULL)
 768                         continue;
 769
 770                 stats->ipackets += rxvq->stats.packets;
 771                 stats->ibytes += rxvq->stats.bytes;
 772                 stats->ierrors += rxvq->stats.errors;
 773
 774                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 775                         stats->q_ipackets[i] = rxvq->stats.packets;
 776                         stats->q_ibytes[i] = rxvq->stats.bytes;
 777                 }
 778         }
 779
 780         stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
 781 }
 782
 783 static int virtio_dev_xstats_get_names(struct rte_eth_dev *dev,
 784                                        struct rte_eth_xstat_name *xstats_names,
 785                                        __rte_unused unsigned limit)
 786 {
 787         unsigned i;
 788         unsigned count = 0;
 789         unsigned t;
 790
 791         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
 792                 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
 793
 794         if (xstats_names != NULL) {
 795                 /* Note: limit checked in rte_eth_xstats_names() */
 796
 797                 for (i = 0; i < dev->data->nb_rx_queues; i++) {
 798                         struct virtqueue *rxvq = dev->data->rx_queues[i];
 799                         if (rxvq == NULL)
 800                                 continue;
 801                         for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
 802                                 snprintf(xstats_names[count].name,
 803                                         sizeof(xstats_names[count].name),
 804                                         "rx_q%u_%s", i,
 805                                         rte_virtio_rxq_stat_strings[t].name);
 806                                 count++;
 807                         }
 808                 }
 809
 810                 for (i = 0; i < dev->data->nb_tx_queues; i++) {
 811                         struct virtqueue *txvq = dev->data->tx_queues[i];
 812                         if (txvq == NULL)
 813                                 continue;
 814                         for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
 815                                 snprintf(xstats_names[count].name,
 816                                         sizeof(xstats_names[count].name),
 817                                         "tx_q%u_%s", i,
 818                                         rte_virtio_txq_stat_strings[t].name);
 819                                 count++;
 820                         }
 821                 }
 822                 return count;
 823         }
 824         return nstats;
 825 }
 826
 827 static int
 828 virtio_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 829                       unsigned n)
 830 {
 831         unsigned i;
 832         unsigned count = 0;
 833
 834         unsigned nstats = dev->data->nb_tx_queues * VIRTIO_NB_TXQ_XSTATS +
 835                 dev->data->nb_rx_queues * VIRTIO_NB_RXQ_XSTATS;
 836
 837         if (n < nstats)
 838                 return nstats;
 839
 840         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 841                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
 842
 843                 if (rxvq == NULL)
 844                         continue;
 845
 846                 unsigned t;
 847
 848                 for (t = 0; t < VIRTIO_NB_RXQ_XSTATS; t++) {
 849                         xstats[count].value = *(uint64_t *)(((char *)rxvq) +
 850                                 rte_virtio_rxq_stat_strings[t].offset);
 851                         count++;
 852                 }
 853         }
 854
 855         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 856                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
 857
 858                 if (txvq == NULL)
 859                         continue;
 860
 861                 unsigned t;
 862
 863                 for (t = 0; t < VIRTIO_NB_TXQ_XSTATS; t++) {
 864                         xstats[count].value = *(uint64_t *)(((char *)txvq) +
 865                                 rte_virtio_txq_stat_strings[t].offset);
 866                         count++;
 867                 }
 868         }
 869
 870         return count;
 871 }
 872
 873 static void
 874 virtio_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 875 {
 876         virtio_update_stats(dev, stats);
 877 }
 878
 879 static void
 880 virtio_dev_stats_reset(struct rte_eth_dev *dev)
 881 {
 882         unsigned int i;
 883
 884         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 885                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
 886                 if (txvq == NULL)
 887                         continue;
 888
 889                 txvq->stats.packets = 0;
 890                 txvq->stats.bytes = 0;
 891                 txvq->stats.errors = 0;
 892                 txvq->stats.multicast = 0;
 893                 txvq->stats.broadcast = 0;
 894                 memset(txvq->stats.size_bins, 0,
 895                        sizeof(txvq->stats.size_bins[0]) * 8);
 896         }
 897
 898         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 899                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
 900                 if (rxvq == NULL)
 901                         continue;
 902
 903                 rxvq->stats.packets = 0;
 904                 rxvq->stats.bytes = 0;
 905                 rxvq->stats.errors = 0;
 906                 rxvq->stats.multicast = 0;
 907                 rxvq->stats.broadcast = 0;
 908                 memset(rxvq->stats.size_bins, 0,
 909                        sizeof(rxvq->stats.size_bins[0]) * 8);
 910         }
 911 }
 912
 913 static void
 914 virtio_set_hwaddr(struct virtio_hw *hw)
 915 {
 916         vtpci_write_dev_config(hw,
 917                         offsetof(struct virtio_net_config, mac),
 918                         &hw->mac_addr, ETHER_ADDR_LEN);
 919 }
 920
 921 static void
 922 virtio_get_hwaddr(struct virtio_hw *hw)
 923 {
 924         if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC)) {
 925                 vtpci_read_dev_config(hw,
 926                         offsetof(struct virtio_net_config, mac),
 927                         &hw->mac_addr, ETHER_ADDR_LEN);
 928         } else {
 929                 eth_random_addr(&hw->mac_addr[0]);
 930                 virtio_set_hwaddr(hw);
 931         }
 932 }
 933
 934 static void
 935 virtio_mac_table_set(struct virtio_hw *hw,
 936                      const struct virtio_net_ctrl_mac *uc,
 937                      const struct virtio_net_ctrl_mac *mc)
 938 {
 939         struct virtio_pmd_ctrl ctrl;
 940         int err, len[2];
 941
 942         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
 943                 PMD_DRV_LOG(INFO, "host does not support mac table");
 944                 return;
 945         }
 946
 947         ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
 948         ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
 949
 950         len[0] = uc->entries * ETHER_ADDR_LEN + sizeof(uc->entries);
 951         memcpy(ctrl.data, uc, len[0]);
 952
 953         len[1] = mc->entries * ETHER_ADDR_LEN + sizeof(mc->entries);
 954         memcpy(ctrl.data + len[0], mc, len[1]);
 955
 956         err = virtio_send_command(hw->cvq, &ctrl, len, 2);
 957         if (err != 0)
 958                 PMD_DRV_LOG(NOTICE, "mac table set failed: %d", err);
 959 }
 960
 961 static void
 962 virtio_mac_addr_add(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
 963                     uint32_t index, uint32_t vmdq __rte_unused)
 964 {
 965         struct virtio_hw *hw = dev->data->dev_private;
 966         const struct ether_addr *addrs = dev->data->mac_addrs;
 967         unsigned int i;
 968         struct virtio_net_ctrl_mac *uc, *mc;
 969
 970         if (index >= VIRTIO_MAX_MAC_ADDRS) {
 971                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
 972                 return;
 973         }
 974
 975         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
 976         uc->entries = 0;
 977         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
 978         mc->entries = 0;
 979
 980         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
 981                 const struct ether_addr *addr
 982                         = (i == index) ? mac_addr : addrs + i;
 983                 struct virtio_net_ctrl_mac *tbl
 984                         = is_multicast_ether_addr(addr) ? mc : uc;
 985
 986                 memcpy(&tbl->macs[tbl->entries++], addr, ETHER_ADDR_LEN);
 987         }
 988
 989         virtio_mac_table_set(hw, uc, mc);
 990 }
 991
 992 static void
 993 virtio_mac_addr_remove(struct rte_eth_dev *dev, uint32_t index)
 994 {
 995         struct virtio_hw *hw = dev->data->dev_private;
 996         struct ether_addr *addrs = dev->data->mac_addrs;
 997         struct virtio_net_ctrl_mac *uc, *mc;
 998         unsigned int i;
 999
1000         if (index >= VIRTIO_MAX_MAC_ADDRS) {
1001                 PMD_DRV_LOG(ERR, "mac address index %u out of range", index);
1002                 return;
1003         }
1004
1005         uc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(uc->entries));
1006         uc->entries = 0;
1007         mc = alloca(VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN + sizeof(mc->entries));
1008         mc->entries = 0;
1009
1010         for (i = 0; i < VIRTIO_MAX_MAC_ADDRS; i++) {
1011                 struct virtio_net_ctrl_mac *tbl;
1012
1013                 if (i == index || is_zero_ether_addr(addrs + i))
1014                         continue;
1015
1016                 tbl = is_multicast_ether_addr(addrs + i) ? mc : uc;
1017                 memcpy(&tbl->macs[tbl->entries++], addrs + i, ETHER_ADDR_LEN);
1018         }
1019
1020         virtio_mac_table_set(hw, uc, mc);
1021 }
1022
1023 static void
1024 virtio_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1025 {
1026         struct virtio_hw *hw = dev->data->dev_private;
1027
1028         memcpy(hw->mac_addr, mac_addr, ETHER_ADDR_LEN);
1029
1030         /* Use atomic update if available */
1031         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1032                 struct virtio_pmd_ctrl ctrl;
1033                 int len = ETHER_ADDR_LEN;
1034
1035                 ctrl.hdr.class = VIRTIO_NET_CTRL_MAC;
1036                 ctrl.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
1037
1038                 memcpy(ctrl.data, mac_addr, ETHER_ADDR_LEN);
1039                 virtio_send_command(hw->cvq, &ctrl, &len, 1);
1040         } else if (vtpci_with_feature(hw, VIRTIO_NET_F_MAC))
1041                 virtio_set_hwaddr(hw);
1042 }
1043
1044 static int
1045 virtio_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
1046 {
1047         struct virtio_hw *hw = dev->data->dev_private;
1048         struct virtio_pmd_ctrl ctrl;
1049         int len;
1050
1051         if (!vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN))
1052                 return -ENOTSUP;
1053
1054         ctrl.hdr.class = VIRTIO_NET_CTRL_VLAN;
1055         ctrl.hdr.cmd = on ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
1056         memcpy(ctrl.data, &vlan_id, sizeof(vlan_id));
1057         len = sizeof(vlan_id);
1058
1059         return virtio_send_command(hw->cvq, &ctrl, &len, 1);
1060 }
1061
1062 static int
1063 virtio_negotiate_features(struct virtio_hw *hw)
1064 {
1065         uint64_t host_features;
1066
1067         /* Prepare guest_features: feature that driver wants to support */
1068         hw->guest_features = VIRTIO_PMD_GUEST_FEATURES;
1069         PMD_INIT_LOG(DEBUG, "guest_features before negotiate = %" PRIx64,
1070                 hw->guest_features);
1071
1072         /* Read device(host) feature bits */
1073         host_features = hw->vtpci_ops->get_features(hw);
1074         PMD_INIT_LOG(DEBUG, "host_features before negotiate = %" PRIx64,
1075                 host_features);
1076
1077         /*
1078          * Negotiate features: Subset of device feature bits are written back
1079          * guest feature bits.
1080          */
1081         hw->guest_features = vtpci_negotiate_features(hw, host_features);
1082         PMD_INIT_LOG(DEBUG, "features after negotiate = %" PRIx64,
1083                 hw->guest_features);
1084
1085         if (hw->modern) {
1086                 if (!vtpci_with_feature(hw, VIRTIO_F_VERSION_1)) {
1087                         PMD_INIT_LOG(ERR,
1088                                 "VIRTIO_F_VERSION_1 features is not enabled.");
1089                         return -1;
1090                 }
1091                 vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_FEATURES_OK);
1092                 if (!(vtpci_get_status(hw) & VIRTIO_CONFIG_STATUS_FEATURES_OK)) {
1093                         PMD_INIT_LOG(ERR,
1094                                 "failed to set FEATURES_OK status!");
1095                         return -1;
1096                 }
1097         }
1098
1099         return 0;
1100 }
1101
1102 /*
1103  * Process Virtio Config changed interrupt and call the callback
1104  * if link state changed.
1105  */
1106 static void
1107 virtio_interrupt_handler(__rte_unused struct rte_intr_handle *handle,
1108                          void *param)
1109 {
1110         struct rte_eth_dev *dev = param;
1111         struct virtio_hw *hw = dev->data->dev_private;
1112         uint8_t isr;
1113
1114         /* Read interrupt status which clears interrupt */
1115         isr = vtpci_isr(hw);
1116         PMD_DRV_LOG(INFO, "interrupt status = %#x", isr);
1117
1118         if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0)
1119                 PMD_DRV_LOG(ERR, "interrupt enable failed");
1120
1121         if (isr & VIRTIO_PCI_ISR_CONFIG) {
1122                 if (virtio_dev_link_update(dev, 0) == 0)
1123                         _rte_eth_dev_callback_process(dev,
1124                                                       RTE_ETH_EVENT_INTR_LSC);
1125         }
1126
1127 }
1128
1129 static void
1130 rx_func_get(struct rte_eth_dev *eth_dev)
1131 {
1132         struct virtio_hw *hw = eth_dev->data->dev_private;
1133         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF))
1134                 eth_dev->rx_pkt_burst = &virtio_recv_mergeable_pkts;
1135         else
1136                 eth_dev->rx_pkt_burst = &virtio_recv_pkts;
1137 }
1138
1139 /*
1140  * This function is based on probe() function in virtio_pci.c
1141  * It returns 0 on success.
1142  */
1143 int
1144 eth_virtio_dev_init(struct rte_eth_dev *eth_dev)
1145 {
1146         struct virtio_hw *hw = eth_dev->data->dev_private;
1147         struct virtio_net_config *config;
1148         struct virtio_net_config local_config;
1149         struct rte_pci_device *pci_dev;
1150         uint32_t dev_flags = RTE_ETH_DEV_DETACHABLE;
1151         int ret;
1152
1153         RTE_BUILD_BUG_ON(RTE_PKTMBUF_HEADROOM < sizeof(struct virtio_net_hdr_mrg_rxbuf));
1154
1155         eth_dev->dev_ops = &virtio_eth_dev_ops;
1156         eth_dev->tx_pkt_burst = &virtio_xmit_pkts;
1157
1158         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
1159                 rx_func_get(eth_dev);
1160                 return 0;
1161         }
1162
1163         /* Allocate memory for storing MAC addresses */
1164         eth_dev->data->mac_addrs = rte_zmalloc("virtio", VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN, 0);
1165         if (eth_dev->data->mac_addrs == NULL) {
1166                 PMD_INIT_LOG(ERR,
1167                         "Failed to allocate %d bytes needed to store MAC addresses",
1168                         VIRTIO_MAX_MAC_ADDRS * ETHER_ADDR_LEN);
1169                 return -ENOMEM;
1170         }
1171
1172         pci_dev = eth_dev->pci_dev;
1173
1174         if (pci_dev) {
1175                 ret = vtpci_init(pci_dev, hw, &dev_flags);
1176                 if (ret)
1177                         return ret;
1178         }
1179
1180         /* Reset the device although not necessary at startup */
1181         vtpci_reset(hw);
1182
1183         /* Tell the host we've noticed this device. */
1184         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_ACK);
1185
1186         /* Tell the host we've known how to drive the device. */
1187         vtpci_set_status(hw, VIRTIO_CONFIG_STATUS_DRIVER);
1188         if (virtio_negotiate_features(hw) < 0)
1189                 return -1;
1190
1191         /* If host does not support status then disable LSC */
1192         if (!vtpci_with_feature(hw, VIRTIO_NET_F_STATUS))
1193                 dev_flags &= ~RTE_ETH_DEV_INTR_LSC;
1194
1195         rte_eth_copy_pci_info(eth_dev, pci_dev);
1196         eth_dev->data->dev_flags = dev_flags;
1197
1198         rx_func_get(eth_dev);
1199
1200         /* Setting up rx_header size for the device */
1201         if (vtpci_with_feature(hw, VIRTIO_NET_F_MRG_RXBUF) ||
1202             vtpci_with_feature(hw, VIRTIO_F_VERSION_1))
1203                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1204         else
1205                 hw->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
1206
1207         /* Copy the permanent MAC address to: virtio_hw */
1208         virtio_get_hwaddr(hw);
1209         ether_addr_copy((struct ether_addr *) hw->mac_addr,
1210                         &eth_dev->data->mac_addrs[0]);
1211         PMD_INIT_LOG(DEBUG,
1212                      "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1213                      hw->mac_addr[0], hw->mac_addr[1], hw->mac_addr[2],
1214                      hw->mac_addr[3], hw->mac_addr[4], hw->mac_addr[5]);
1215
1216         if (vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VQ)) {
1217                 config = &local_config;
1218
1219                 vtpci_read_dev_config(hw,
1220                         offsetof(struct virtio_net_config, mac),
1221                         &config->mac, sizeof(config->mac));
1222
1223                 if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1224                         vtpci_read_dev_config(hw,
1225                                 offsetof(struct virtio_net_config, status),
1226                                 &config->status, sizeof(config->status));
1227                 } else {
1228                         PMD_INIT_LOG(DEBUG,
1229                                      "VIRTIO_NET_F_STATUS is not supported");
1230                         config->status = 0;
1231                 }
1232
1233                 if (vtpci_with_feature(hw, VIRTIO_NET_F_MQ)) {
1234                         vtpci_read_dev_config(hw,
1235                                 offsetof(struct virtio_net_config, max_virtqueue_pairs),
1236                                 &config->max_virtqueue_pairs,
1237                                 sizeof(config->max_virtqueue_pairs));
1238                 } else {
1239                         PMD_INIT_LOG(DEBUG,
1240                                      "VIRTIO_NET_F_MQ is not supported");
1241                         config->max_virtqueue_pairs = 1;
1242                 }
1243
1244                 hw->max_rx_queues =
1245                         (VIRTIO_MAX_RX_QUEUES < config->max_virtqueue_pairs) ?
1246                         VIRTIO_MAX_RX_QUEUES : config->max_virtqueue_pairs;
1247                 hw->max_tx_queues =
1248                         (VIRTIO_MAX_TX_QUEUES < config->max_virtqueue_pairs) ?
1249                         VIRTIO_MAX_TX_QUEUES : config->max_virtqueue_pairs;
1250
1251                 virtio_dev_cq_queue_setup(eth_dev,
1252                                         config->max_virtqueue_pairs * 2,
1253                                         SOCKET_ID_ANY);
1254
1255                 PMD_INIT_LOG(DEBUG, "config->max_virtqueue_pairs=%d",
1256                                 config->max_virtqueue_pairs);
1257                 PMD_INIT_LOG(DEBUG, "config->status=%d", config->status);
1258                 PMD_INIT_LOG(DEBUG,
1259                                 "PORT MAC: %02X:%02X:%02X:%02X:%02X:%02X",
1260                                 config->mac[0], config->mac[1],
1261                                 config->mac[2], config->mac[3],
1262                                 config->mac[4], config->mac[5]);
1263         } else {
1264                 hw->max_rx_queues = 1;
1265                 hw->max_tx_queues = 1;
1266         }
1267
1268         PMD_INIT_LOG(DEBUG, "hw->max_rx_queues=%d   hw->max_tx_queues=%d",
1269                         hw->max_rx_queues, hw->max_tx_queues);
1270         if (pci_dev)
1271                 PMD_INIT_LOG(DEBUG, "port %d vendorID=0x%x deviceID=0x%x",
1272                         eth_dev->data->port_id, pci_dev->id.vendor_id,
1273                         pci_dev->id.device_id);
1274
1275         /* Setup interrupt callback  */
1276         if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1277                 rte_intr_callback_register(&pci_dev->intr_handle,
1278                                    virtio_interrupt_handler, eth_dev);
1279
1280         virtio_dev_cq_start(eth_dev);
1281
1282         return 0;
1283 }
1284
1285 static int
1286 eth_virtio_dev_uninit(struct rte_eth_dev *eth_dev)
1287 {
1288         struct rte_pci_device *pci_dev;
1289         struct virtio_hw *hw = eth_dev->data->dev_private;
1290
1291         PMD_INIT_FUNC_TRACE();
1292
1293         if (rte_eal_process_type() == RTE_PROC_SECONDARY)
1294                 return -EPERM;
1295
1296         /* Close it anyway since there's no way to know if closed */
1297         virtio_dev_close(eth_dev);
1298
1299         pci_dev = eth_dev->pci_dev;
1300
1301         eth_dev->dev_ops = NULL;
1302         eth_dev->tx_pkt_burst = NULL;
1303         eth_dev->rx_pkt_burst = NULL;
1304
1305         if (hw->cvq)
1306                 virtio_dev_queue_release(hw->cvq->vq);
1307
1308         rte_free(eth_dev->data->mac_addrs);
1309         eth_dev->data->mac_addrs = NULL;
1310
1311         /* reset interrupt callback  */
1312         if (eth_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1313                 rte_intr_callback_unregister(&pci_dev->intr_handle,
1314                                                 virtio_interrupt_handler,
1315                                                 eth_dev);
1316         rte_eal_pci_unmap_device(pci_dev);
1317
1318         PMD_INIT_LOG(DEBUG, "dev_uninit completed");
1319
1320         return 0;
1321 }
1322
1323 static struct eth_driver rte_virtio_pmd = {
1324         .pci_drv = {
1325                 .driver = {
1326                         .name = "net_virtio",
1327                 },
1328                 .id_table = pci_id_virtio_map,
1329                 .drv_flags = RTE_PCI_DRV_DETACHABLE,
1330                 .probe = rte_eth_dev_pci_probe,
1331                 .remove = rte_eth_dev_pci_remove,
1332         },
1333         .eth_dev_init = eth_virtio_dev_init,
1334         .eth_dev_uninit = eth_virtio_dev_uninit,
1335         .dev_private_size = sizeof(struct virtio_hw),
1336 };
1337
1338 RTE_INIT(rte_virtio_pmd_init);
1339 static void
1340 rte_virtio_pmd_init(void)
1341 {
1342         if (rte_eal_iopl_init() != 0) {
1343                 PMD_INIT_LOG(ERR, "IOPL call failed - cannot use virtio PMD");
1344                 return;
1345         }
1346
1347         rte_eal_pci_register(&rte_virtio_pmd.pci_drv);
1348 }
1349
1350 /*
1351  * Configure virtio device
1352  * It returns 0 on success.
1353  */
1354 static int
1355 virtio_dev_configure(struct rte_eth_dev *dev)
1356 {
1357         const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1358         struct virtio_hw *hw = dev->data->dev_private;
1359
1360         PMD_INIT_LOG(DEBUG, "configure");
1361
1362         if (rxmode->hw_ip_checksum) {
1363                 PMD_DRV_LOG(ERR, "HW IP checksum not supported");
1364                 return -EINVAL;
1365         }
1366
1367         hw->vlan_strip = rxmode->hw_vlan_strip;
1368
1369         if (rxmode->hw_vlan_filter
1370             && !vtpci_with_feature(hw, VIRTIO_NET_F_CTRL_VLAN)) {
1371                 PMD_DRV_LOG(NOTICE,
1372                             "vlan filtering not available on this host");
1373                 return -ENOTSUP;
1374         }
1375
1376         if (dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)
1377                 if (vtpci_irq_config(hw, 0) == VIRTIO_MSI_NO_VECTOR) {
1378                         PMD_DRV_LOG(ERR, "failed to set config vector");
1379                         return -EBUSY;
1380                 }
1381
1382         return 0;
1383 }
1384
1385
1386 static int
1387 virtio_dev_start(struct rte_eth_dev *dev)
1388 {
1389         uint16_t nb_queues, i;
1390         struct virtio_hw *hw = dev->data->dev_private;
1391         struct virtnet_rx *rxvq;
1392         struct virtnet_tx *txvq __rte_unused;
1393
1394         /* check if lsc interrupt feature is enabled */
1395         if (dev->data->dev_conf.intr_conf.lsc) {
1396                 if (!(dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
1397                         PMD_DRV_LOG(ERR, "link status not supported by host");
1398                         return -ENOTSUP;
1399                 }
1400
1401                 if (rte_intr_enable(&dev->pci_dev->intr_handle) < 0) {
1402                         PMD_DRV_LOG(ERR, "interrupt enable failed");
1403                         return -EIO;
1404                 }
1405         }
1406
1407         /* Initialize Link state */
1408         virtio_dev_link_update(dev, 0);
1409
1410         /* On restart after stop do not touch queues */
1411         if (hw->started)
1412                 return 0;
1413
1414         /* Do final configuration before rx/tx engine starts */
1415         virtio_dev_rxtx_start(dev);
1416         vtpci_reinit_complete(hw);
1417
1418         hw->started = 1;
1419
1420         /*Notify the backend
1421          *Otherwise the tap backend might already stop its queue due to fullness.
1422          *vhost backend will have no chance to be waked up
1423          */
1424         nb_queues = dev->data->nb_rx_queues;
1425         if (nb_queues > 1) {
1426                 if (virtio_set_multiple_queues(dev, nb_queues) != 0)
1427                         return -EINVAL;
1428         }
1429
1430         PMD_INIT_LOG(DEBUG, "nb_queues=%d", nb_queues);
1431
1432         for (i = 0; i < nb_queues; i++) {
1433                 rxvq = dev->data->rx_queues[i];
1434                 virtqueue_notify(rxvq->vq);
1435         }
1436
1437         PMD_INIT_LOG(DEBUG, "Notified backend at initialization");
1438
1439         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1440                 rxvq = dev->data->rx_queues[i];
1441                 VIRTQUEUE_DUMP(rxvq->vq);
1442         }
1443
1444         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1445                 txvq = dev->data->tx_queues[i];
1446                 VIRTQUEUE_DUMP(txvq->vq);
1447         }
1448
1449         return 0;
1450 }
1451
1452 static void virtio_dev_free_mbufs(struct rte_eth_dev *dev)
1453 {
1454         struct rte_mbuf *buf;
1455         int i, mbuf_num = 0;
1456
1457         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1458                 struct virtnet_rx *rxvq = dev->data->rx_queues[i];
1459
1460                 PMD_INIT_LOG(DEBUG,
1461                              "Before freeing rxq[%d] used and unused buf", i);
1462                 VIRTQUEUE_DUMP(rxvq->vq);
1463
1464                 PMD_INIT_LOG(DEBUG, "rx_queues[%d]=%p", i, rxvq);
1465                 while ((buf = virtqueue_detatch_unused(rxvq->vq)) != NULL) {
1466                         rte_pktmbuf_free(buf);
1467                         mbuf_num++;
1468                 }
1469
1470                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1471                 PMD_INIT_LOG(DEBUG,
1472                              "After freeing rxq[%d] used and unused buf", i);
1473                 VIRTQUEUE_DUMP(rxvq->vq);
1474         }
1475
1476         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1477                 struct virtnet_tx *txvq = dev->data->tx_queues[i];
1478
1479                 PMD_INIT_LOG(DEBUG,
1480                              "Before freeing txq[%d] used and unused bufs",
1481                              i);
1482                 VIRTQUEUE_DUMP(txvq->vq);
1483
1484                 mbuf_num = 0;
1485                 while ((buf = virtqueue_detatch_unused(txvq->vq)) != NULL) {
1486                         rte_pktmbuf_free(buf);
1487                         mbuf_num++;
1488                 }
1489
1490                 PMD_INIT_LOG(DEBUG, "free %d mbufs", mbuf_num);
1491                 PMD_INIT_LOG(DEBUG,
1492                              "After freeing txq[%d] used and unused buf", i);
1493                 VIRTQUEUE_DUMP(txvq->vq);
1494         }
1495 }
1496
1497 /*
1498  * Stop device: disable interrupt and mark link down
1499  */
1500 static void
1501 virtio_dev_stop(struct rte_eth_dev *dev)
1502 {
1503         struct rte_eth_link link;
1504         struct virtio_hw *hw = dev->data->dev_private;
1505
1506         PMD_INIT_LOG(DEBUG, "stop");
1507
1508         hw->started = 0;
1509
1510         if (dev->data->dev_conf.intr_conf.lsc)
1511                 rte_intr_disable(&dev->pci_dev->intr_handle);
1512
1513         memset(&link, 0, sizeof(link));
1514         virtio_dev_atomic_write_link_status(dev, &link);
1515 }
1516
1517 static int
1518 virtio_dev_link_update(struct rte_eth_dev *dev, __rte_unused int wait_to_complete)
1519 {
1520         struct rte_eth_link link, old;
1521         uint16_t status;
1522         struct virtio_hw *hw = dev->data->dev_private;
1523         memset(&link, 0, sizeof(link));
1524         virtio_dev_atomic_read_link_status(dev, &link);
1525         old = link;
1526         link.link_duplex = ETH_LINK_FULL_DUPLEX;
1527         link.link_speed  = SPEED_10G;
1528
1529         if (vtpci_with_feature(hw, VIRTIO_NET_F_STATUS)) {
1530                 PMD_INIT_LOG(DEBUG, "Get link status from hw");
1531                 vtpci_read_dev_config(hw,
1532                                 offsetof(struct virtio_net_config, status),
1533                                 &status, sizeof(status));
1534                 if ((status & VIRTIO_NET_S_LINK_UP) == 0) {
1535                         link.link_status = ETH_LINK_DOWN;
1536                         PMD_INIT_LOG(DEBUG, "Port %d is down",
1537                                      dev->data->port_id);
1538                 } else {
1539                         link.link_status = ETH_LINK_UP;
1540                         PMD_INIT_LOG(DEBUG, "Port %d is up",
1541                                      dev->data->port_id);
1542                 }
1543         } else {
1544                 link.link_status = ETH_LINK_UP;
1545         }
1546         virtio_dev_atomic_write_link_status(dev, &link);
1547
1548         return (old.link_status == link.link_status) ? -1 : 0;
1549 }
1550
1551 static void
1552 virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1553 {
1554         struct virtio_hw *hw = dev->data->dev_private;
1555
1556         if (dev->pci_dev)
1557                 dev_info->driver_name = dev->driver->pci_drv.driver.name;
1558         else
1559                 dev_info->driver_name = "virtio_user PMD";
1560         dev_info->max_rx_queues = (uint16_t)hw->max_rx_queues;
1561         dev_info->max_tx_queues = (uint16_t)hw->max_tx_queues;
1562         dev_info->min_rx_bufsize = VIRTIO_MIN_RX_BUFSIZE;
1563         dev_info->max_rx_pktlen = VIRTIO_MAX_RX_PKTLEN;
1564         dev_info->max_mac_addrs = VIRTIO_MAX_MAC_ADDRS;
1565         dev_info->default_txconf = (struct rte_eth_txconf) {
1566                 .txq_flags = ETH_TXQ_FLAGS_NOOFFLOADS
1567         };
1568 }
1569
1570 /*
1571  * It enables testpmd to collect per queue stats.
1572  */
1573 static int
1574 virtio_dev_queue_stats_mapping_set(__rte_unused struct rte_eth_dev *eth_dev,
1575 __rte_unused uint16_t queue_id, __rte_unused uint8_t stat_idx,
1576 __rte_unused uint8_t is_rx)
1577 {
1578         return 0;
1579 }
1580
1581 DRIVER_EXPORT_NAME(net_virtio, __COUNTER__);
1582 DRIVER_REGISTER_PCI_TABLE(net_virtio, pci_id_virtio_map);