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