drivers/net/virtio_net.c

   1 /* A simple network driver using virtio.
   2  *
   3  * Copyright 2007 Rusty Russell <rusty@rustcorp.com.au> IBM Corporation
   4  *
   5  * This program is free software; you can redistribute it and/or modify
   6  * it under the terms of the GNU General Public License as published by
   7  * the Free Software Foundation; either version 2 of the License, or
   8  * (at your option) any later version.
   9  *
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License for more details.
  14  *
  15  * You should have received a copy of the GNU General Public License
  16  * along with this program; if not, write to the Free Software
  17  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  18  */
  19 //#define DEBUG
  20 #include <linux/netdevice.h>
  21 #include <linux/etherdevice.h>
  22 #include <linux/ethtool.h>
  23 #include <linux/module.h>
  24 #include <linux/virtio.h>
  25 #include <linux/virtio_net.h>
  26 #include <linux/scatterlist.h>
  27
  28 static int napi_weight = 128;
  29 module_param(napi_weight, int, 0444);
  30
  31 static int csum = 1, gso = 1;
  32 module_param(csum, bool, 0444);
  33 module_param(gso, bool, 0444);
  34
  35 /* FIXME: MTU in config. */
  36 #define MAX_PACKET_LEN (ETH_HLEN+ETH_DATA_LEN)
  37 #define GOOD_COPY_LEN   128
  38
  39 struct virtnet_info
  40 {
  41         struct virtio_device *vdev;
  42         struct virtqueue *rvq, *svq;
  43         struct net_device *dev;
  44         struct napi_struct napi;
  45
  46         /* The skb we couldn't send because buffers were full. */
  47         struct sk_buff *last_xmit_skb;
  48
  49         /* If we need to free in a timer, this is it. */
  50         struct timer_list xmit_free_timer;
  51
  52         /* Number of input buffers, and max we've ever had. */
  53         unsigned int num, max;
  54
  55         /* For cleaning up after transmission. */
  56         struct tasklet_struct tasklet;
  57         bool free_in_tasklet;
  58
  59         /* I like... big packets and I cannot lie! */
  60         bool big_packets;
  61
  62         /* Host will merge rx buffers for big packets (shake it! shake it!) */
  63         bool mergeable_rx_bufs;
  64
  65         /* Receive & send queues. */
  66         struct sk_buff_head recv;
  67         struct sk_buff_head send;
  68
  69         /* Chain pages by the private ptr. */
  70         struct page *pages;
  71 };
  72
  73 static inline void *skb_vnet_hdr(struct sk_buff *skb)
  74 {
  75         return (struct virtio_net_hdr *)skb->cb;
  76 }
  77
  78 static void give_a_page(struct virtnet_info *vi, struct page *page)
  79 {
  80         page->private = (unsigned long)vi->pages;
  81         vi->pages = page;
  82 }
  83
  84 static void trim_pages(struct virtnet_info *vi, struct sk_buff *skb)
  85 {
  86         unsigned int i;
  87
  88         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
  89                 give_a_page(vi, skb_shinfo(skb)->frags[i].page);
  90         skb_shinfo(skb)->nr_frags = 0;
  91         skb->data_len = 0;
  92 }
  93
  94 static struct page *get_a_page(struct virtnet_info *vi, gfp_t gfp_mask)
  95 {
  96         struct page *p = vi->pages;
  97
  98         if (p)
  99                 vi->pages = (struct page *)p->private;
 100         else
 101                 p = alloc_page(gfp_mask);
 102         return p;
 103 }
 104
 105 static void skb_xmit_done(struct virtqueue *svq)
 106 {
 107         struct virtnet_info *vi = svq->vdev->priv;
 108
 109         /* Suppress further interrupts. */
 110         svq->vq_ops->disable_cb(svq);
 111
 112         /* We were probably waiting for more output buffers. */
 113         netif_wake_queue(vi->dev);
 114
 115         /* Make sure we re-xmit last_xmit_skb: if there are no more packets
 116          * queued, start_xmit won't be called. */
 117         tasklet_schedule(&vi->tasklet);
 118 }
 119
 120 static void receive_skb(struct net_device *dev, struct sk_buff *skb,
 121                         unsigned len)
 122 {
 123         struct virtnet_info *vi = netdev_priv(dev);
 124         struct virtio_net_hdr *hdr = skb_vnet_hdr(skb);
 125         int err;
 126         int i;
 127
 128         if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) {
 129                 pr_debug("%s: short packet %i\n", dev->name, len);
 130                 dev->stats.rx_length_errors++;
 131                 goto drop;
 132         }
 133
 134         if (vi->mergeable_rx_bufs) {
 135                 struct virtio_net_hdr_mrg_rxbuf *mhdr = skb_vnet_hdr(skb);
 136                 unsigned int copy;
 137                 char *p = page_address(skb_shinfo(skb)->frags[0].page);
 138
 139                 if (len > PAGE_SIZE)
 140                         len = PAGE_SIZE;
 141                 len -= sizeof(struct virtio_net_hdr_mrg_rxbuf);
 142
 143                 memcpy(hdr, p, sizeof(*mhdr));
 144                 p += sizeof(*mhdr);
 145
 146                 copy = len;
 147                 if (copy > skb_tailroom(skb))
 148                         copy = skb_tailroom(skb);
 149
 150                 memcpy(skb_put(skb, copy), p, copy);
 151
 152                 len -= copy;
 153
 154                 if (!len) {
 155                         give_a_page(vi, skb_shinfo(skb)->frags[0].page);
 156                         skb_shinfo(skb)->nr_frags--;
 157                 } else {
 158                         skb_shinfo(skb)->frags[0].page_offset +=
 159                                 sizeof(*mhdr) + copy;
 160                         skb_shinfo(skb)->frags[0].size = len;
 161                         skb->data_len += len;
 162                         skb->len += len;
 163                 }
 164
 165                 while (--mhdr->num_buffers) {
 166                         struct sk_buff *nskb;
 167
 168                         i = skb_shinfo(skb)->nr_frags;
 169                         if (i >= MAX_SKB_FRAGS) {
 170                                 pr_debug("%s: packet too long %d\n", dev->name,
 171                                          len);
 172                                 dev->stats.rx_length_errors++;
 173                                 goto drop;
 174                         }
 175
 176                         nskb = vi->rvq->vq_ops->get_buf(vi->rvq, &len);
 177                         if (!nskb) {
 178                                 pr_debug("%s: rx error: %d buffers missing\n",
 179                                          dev->name, mhdr->num_buffers);
 180                                 dev->stats.rx_length_errors++;
 181                                 goto drop;
 182                         }
 183
 184                         __skb_unlink(nskb, &vi->recv);
 185                         vi->num--;
 186
 187                         skb_shinfo(skb)->frags[i] = skb_shinfo(nskb)->frags[0];
 188                         skb_shinfo(nskb)->nr_frags = 0;
 189                         kfree_skb(nskb);
 190
 191                         if (len > PAGE_SIZE)
 192                                 len = PAGE_SIZE;
 193
 194                         skb_shinfo(skb)->frags[i].size = len;
 195                         skb_shinfo(skb)->nr_frags++;
 196                         skb->data_len += len;
 197                         skb->len += len;
 198                 }
 199         } else {
 200                 len -= sizeof(struct virtio_net_hdr);
 201
 202                 if (len <= MAX_PACKET_LEN)
 203                         trim_pages(vi, skb);
 204
 205                 err = pskb_trim(skb, len);
 206                 if (err) {
 207                         pr_debug("%s: pskb_trim failed %i %d\n", dev->name,
 208                                  len, err);
 209                         dev->stats.rx_dropped++;
 210                         goto drop;
 211                 }
 212         }
 213
 214         skb->truesize += skb->data_len;
 215         dev->stats.rx_bytes += skb->len;
 216         dev->stats.rx_packets++;
 217
 218         if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
 219                 pr_debug("Needs csum!\n");
 220                 if (!skb_partial_csum_set(skb,hdr->csum_start,hdr->csum_offset))
 221                         goto frame_err;
 222         }
 223
 224         skb->protocol = eth_type_trans(skb, dev);
 225         pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
 226                  ntohs(skb->protocol), skb->len, skb->pkt_type);
 227
 228         if (hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
 229                 pr_debug("GSO!\n");
 230                 switch (hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
 231                 case VIRTIO_NET_HDR_GSO_TCPV4:
 232                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
 233                         break;
 234                 case VIRTIO_NET_HDR_GSO_UDP:
 235                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
 236                         break;
 237                 case VIRTIO_NET_HDR_GSO_TCPV6:
 238                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
 239                         break;
 240                 default:
 241                         if (net_ratelimit())
 242                                 printk(KERN_WARNING "%s: bad gso type %u.\n",
 243                                        dev->name, hdr->gso_type);
 244                         goto frame_err;
 245                 }
 246
 247                 if (hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
 248                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
 249
 250                 skb_shinfo(skb)->gso_size = hdr->gso_size;
 251                 if (skb_shinfo(skb)->gso_size == 0) {
 252                         if (net_ratelimit())
 253                                 printk(KERN_WARNING "%s: zero gso size.\n",
 254                                        dev->name);
 255                         goto frame_err;
 256                 }
 257
 258                 /* Header must be checked, and gso_segs computed. */
 259                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
 260                 skb_shinfo(skb)->gso_segs = 0;
 261         }
 262
 263         netif_receive_skb(skb);
 264         return;
 265
 266 frame_err:
 267         dev->stats.rx_frame_errors++;
 268 drop:
 269         dev_kfree_skb(skb);
 270 }
 271
 272 static void try_fill_recv_maxbufs(struct virtnet_info *vi)
 273 {
 274         struct sk_buff *skb;
 275         struct scatterlist sg[2+MAX_SKB_FRAGS];
 276         int num, err, i;
 277
 278         sg_init_table(sg, 2+MAX_SKB_FRAGS);
 279         for (;;) {
 280                 struct virtio_net_hdr *hdr;
 281
 282                 skb = netdev_alloc_skb(vi->dev, MAX_PACKET_LEN);
 283                 if (unlikely(!skb))
 284                         break;
 285
 286                 skb_put(skb, MAX_PACKET_LEN);
 287
 288                 hdr = skb_vnet_hdr(skb);
 289                 sg_init_one(sg, hdr, sizeof(*hdr));
 290
 291                 if (vi->big_packets) {
 292                         for (i = 0; i < MAX_SKB_FRAGS; i++) {
 293                                 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
 294                                 f->page = get_a_page(vi, GFP_ATOMIC);
 295                                 if (!f->page)
 296                                         break;
 297
 298                                 f->page_offset = 0;
 299                                 f->size = PAGE_SIZE;
 300
 301                                 skb->data_len += PAGE_SIZE;
 302                                 skb->len += PAGE_SIZE;
 303
 304                                 skb_shinfo(skb)->nr_frags++;
 305                         }
 306                 }
 307
 308                 num = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
 309                 skb_queue_head(&vi->recv, skb);
 310
 311                 err = vi->rvq->vq_ops->add_buf(vi->rvq, sg, 0, num, skb);
 312                 if (err) {
 313                         skb_unlink(skb, &vi->recv);
 314                         trim_pages(vi, skb);
 315                         kfree_skb(skb);
 316                         break;
 317                 }
 318                 vi->num++;
 319         }
 320         if (unlikely(vi->num > vi->max))
 321                 vi->max = vi->num;
 322         vi->rvq->vq_ops->kick(vi->rvq);
 323 }
 324
 325 static void try_fill_recv(struct virtnet_info *vi)
 326 {
 327         struct sk_buff *skb;
 328         struct scatterlist sg[1];
 329         int err;
 330
 331         if (!vi->mergeable_rx_bufs) {
 332                 try_fill_recv_maxbufs(vi);
 333                 return;
 334         }
 335
 336         for (;;) {
 337                 skb_frag_t *f;
 338
 339                 skb = netdev_alloc_skb(vi->dev, GOOD_COPY_LEN + NET_IP_ALIGN);
 340                 if (unlikely(!skb))
 341                         break;
 342
 343                 skb_reserve(skb, NET_IP_ALIGN);
 344
 345                 f = &skb_shinfo(skb)->frags[0];
 346                 f->page = get_a_page(vi, GFP_ATOMIC);
 347                 if (!f->page) {
 348                         kfree_skb(skb);
 349                         break;
 350                 }
 351
 352                 f->page_offset = 0;
 353                 f->size = PAGE_SIZE;
 354
 355                 skb_shinfo(skb)->nr_frags++;
 356
 357                 sg_init_one(sg, page_address(f->page), PAGE_SIZE);
 358                 skb_queue_head(&vi->recv, skb);
 359
 360                 err = vi->rvq->vq_ops->add_buf(vi->rvq, sg, 0, 1, skb);
 361                 if (err) {
 362                         skb_unlink(skb, &vi->recv);
 363                         kfree_skb(skb);
 364                         break;
 365                 }
 366                 vi->num++;
 367         }
 368         if (unlikely(vi->num > vi->max))
 369                 vi->max = vi->num;
 370         vi->rvq->vq_ops->kick(vi->rvq);
 371 }
 372
 373 static void skb_recv_done(struct virtqueue *rvq)
 374 {
 375         struct virtnet_info *vi = rvq->vdev->priv;
 376         /* Schedule NAPI, Suppress further interrupts if successful. */
 377         if (netif_rx_schedule_prep(&vi->napi)) {
 378                 rvq->vq_ops->disable_cb(rvq);
 379                 __netif_rx_schedule(&vi->napi);
 380         }
 381 }
 382
 383 static int virtnet_poll(struct napi_struct *napi, int budget)
 384 {
 385         struct virtnet_info *vi = container_of(napi, struct virtnet_info, napi);
 386         struct sk_buff *skb = NULL;
 387         unsigned int len, received = 0;
 388
 389 again:
 390         while (received < budget &&
 391                (skb = vi->rvq->vq_ops->get_buf(vi->rvq, &len)) != NULL) {
 392                 __skb_unlink(skb, &vi->recv);
 393                 receive_skb(vi->dev, skb, len);
 394                 vi->num--;
 395                 received++;
 396         }
 397
 398         /* FIXME: If we oom and completely run out of inbufs, we need
 399          * to start a timer trying to fill more. */
 400         if (vi->num < vi->max / 2)
 401                 try_fill_recv(vi);
 402
 403         /* Out of packets? */
 404         if (received < budget) {
 405                 netif_rx_complete(napi);
 406                 if (unlikely(!vi->rvq->vq_ops->enable_cb(vi->rvq))
 407                     && napi_schedule_prep(napi)) {
 408                         vi->rvq->vq_ops->disable_cb(vi->rvq);
 409                         __netif_rx_schedule(napi);
 410                         goto again;
 411                 }
 412         }
 413
 414         return received;
 415 }
 416
 417 static void free_old_xmit_skbs(struct virtnet_info *vi)
 418 {
 419         struct sk_buff *skb;
 420         unsigned int len;
 421
 422         while ((skb = vi->svq->vq_ops->get_buf(vi->svq, &len)) != NULL) {
 423                 pr_debug("Sent skb %p\n", skb);
 424                 __skb_unlink(skb, &vi->send);
 425                 vi->dev->stats.tx_bytes += skb->len;
 426                 vi->dev->stats.tx_packets++;
 427                 kfree_skb(skb);
 428         }
 429 }
 430
 431 /* If the virtio transport doesn't always notify us when all in-flight packets
 432  * are consumed, we fall back to using this function on a timer to free them. */
 433 static void xmit_free(unsigned long data)
 434 {
 435         struct virtnet_info *vi = (void *)data;
 436
 437         netif_tx_lock(vi->dev);
 438
 439         free_old_xmit_skbs(vi);
 440
 441         if (!skb_queue_empty(&vi->send))
 442                 mod_timer(&vi->xmit_free_timer, jiffies + (HZ/10));
 443
 444         netif_tx_unlock(vi->dev);
 445 }
 446
 447 static int xmit_skb(struct virtnet_info *vi, struct sk_buff *skb)
 448 {
 449         int num, err;
 450         struct scatterlist sg[2+MAX_SKB_FRAGS];
 451         struct virtio_net_hdr_mrg_rxbuf *mhdr = skb_vnet_hdr(skb);
 452         struct virtio_net_hdr *hdr = skb_vnet_hdr(skb);
 453         const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
 454
 455         sg_init_table(sg, 2+MAX_SKB_FRAGS);
 456
 457         pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
 458
 459         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 460                 hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
 461                 hdr->csum_start = skb->csum_start - skb_headroom(skb);
 462                 hdr->csum_offset = skb->csum_offset;
 463         } else {
 464                 hdr->flags = 0;
 465                 hdr->csum_offset = hdr->csum_start = 0;
 466         }
 467
 468         if (skb_is_gso(skb)) {
 469                 hdr->hdr_len = skb_transport_header(skb) - skb->data;
 470                 hdr->gso_size = skb_shinfo(skb)->gso_size;
 471                 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
 472                         hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
 473                 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
 474                         hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
 475                 else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 476                         hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
 477                 else
 478                         BUG();
 479                 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
 480                         hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
 481         } else {
 482                 hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
 483                 hdr->gso_size = hdr->hdr_len = 0;
 484         }
 485
 486         mhdr->num_buffers = 0;
 487
 488         /* Encode metadata header at front. */
 489         if (vi->mergeable_rx_bufs)
 490                 sg_init_one(sg, mhdr, sizeof(*mhdr));
 491         else
 492                 sg_init_one(sg, hdr, sizeof(*hdr));
 493
 494         num = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
 495
 496         err = vi->svq->vq_ops->add_buf(vi->svq, sg, num, 0, skb);
 497         if (!err && !vi->free_in_tasklet)
 498                 mod_timer(&vi->xmit_free_timer, jiffies + (HZ/10));
 499
 500         return err;
 501 }
 502
 503 static void xmit_tasklet(unsigned long data)
 504 {
 505         struct virtnet_info *vi = (void *)data;
 506
 507         netif_tx_lock_bh(vi->dev);
 508         if (vi->last_xmit_skb && xmit_skb(vi, vi->last_xmit_skb) == 0) {
 509                 vi->svq->vq_ops->kick(vi->svq);
 510                 vi->last_xmit_skb = NULL;
 511         }
 512         if (vi->free_in_tasklet)
 513                 free_old_xmit_skbs(vi);
 514         netif_tx_unlock_bh(vi->dev);
 515 }
 516
 517 static int start_xmit(struct sk_buff *skb, struct net_device *dev)
 518 {
 519         struct virtnet_info *vi = netdev_priv(dev);
 520
 521 again:
 522         /* Free up any pending old buffers before queueing new ones. */
 523         free_old_xmit_skbs(vi);
 524
 525         /* If we has a buffer left over from last time, send it now. */
 526         if (unlikely(vi->last_xmit_skb) &&
 527             xmit_skb(vi, vi->last_xmit_skb) != 0)
 528                 goto stop_queue;
 529
 530         vi->last_xmit_skb = NULL;
 531
 532         /* Put new one in send queue and do transmit */
 533         if (likely(skb)) {
 534                 __skb_queue_head(&vi->send, skb);
 535                 if (xmit_skb(vi, skb) != 0) {
 536                         vi->last_xmit_skb = skb;
 537                         skb = NULL;
 538                         goto stop_queue;
 539                 }
 540         }
 541 done:
 542         vi->svq->vq_ops->kick(vi->svq);
 543         return NETDEV_TX_OK;
 544
 545 stop_queue:
 546         pr_debug("%s: virtio not prepared to send\n", dev->name);
 547         netif_stop_queue(dev);
 548
 549         /* Activate callback for using skbs: if this returns false it
 550          * means some were used in the meantime. */
 551         if (unlikely(!vi->svq->vq_ops->enable_cb(vi->svq))) {
 552                 vi->svq->vq_ops->disable_cb(vi->svq);
 553                 netif_start_queue(dev);
 554                 goto again;
 555         }
 556         if (skb) {
 557                 /* Drop this skb: we only queue one. */
 558                 vi->dev->stats.tx_dropped++;
 559                 kfree_skb(skb);
 560         }
 561         goto done;
 562 }
 563
 564 #ifdef CONFIG_NET_POLL_CONTROLLER
 565 static void virtnet_netpoll(struct net_device *dev)
 566 {
 567         struct virtnet_info *vi = netdev_priv(dev);
 568
 569         napi_schedule(&vi->napi);
 570 }
 571 #endif
 572
 573 static int virtnet_open(struct net_device *dev)
 574 {
 575         struct virtnet_info *vi = netdev_priv(dev);
 576
 577         napi_enable(&vi->napi);
 578
 579         /* If all buffers were filled by other side before we napi_enabled, we
 580          * won't get another interrupt, so process any outstanding packets
 581          * now.  virtnet_poll wants re-enable the queue, so we disable here.
 582          * We synchronize against interrupts via NAPI_STATE_SCHED */
 583         if (netif_rx_schedule_prep(&vi->napi)) {
 584                 vi->rvq->vq_ops->disable_cb(vi->rvq);
 585                 __netif_rx_schedule(&vi->napi);
 586         }
 587         return 0;
 588 }
 589
 590 static int virtnet_close(struct net_device *dev)
 591 {
 592         struct virtnet_info *vi = netdev_priv(dev);
 593
 594         napi_disable(&vi->napi);
 595
 596         return 0;
 597 }
 598
 599 static int virtnet_set_tx_csum(struct net_device *dev, u32 data)
 600 {
 601         struct virtnet_info *vi = netdev_priv(dev);
 602         struct virtio_device *vdev = vi->vdev;
 603
 604         if (data && !virtio_has_feature(vdev, VIRTIO_NET_F_CSUM))
 605                 return -ENOSYS;
 606
 607         return ethtool_op_set_tx_hw_csum(dev, data);
 608 }
 609
 610 static struct ethtool_ops virtnet_ethtool_ops = {
 611         .set_tx_csum = virtnet_set_tx_csum,
 612         .set_sg = ethtool_op_set_sg,
 613         .set_tso = ethtool_op_set_tso,
 614 };
 615
 616 #define MIN_MTU 68
 617 #define MAX_MTU 65535
 618
 619 static int virtnet_change_mtu(struct net_device *dev, int new_mtu)
 620 {
 621         if (new_mtu < MIN_MTU || new_mtu > MAX_MTU)
 622                 return -EINVAL;
 623         dev->mtu = new_mtu;
 624         return 0;
 625 }
 626
 627 static const struct net_device_ops virtnet_netdev = {
 628         .ndo_open            = virtnet_open,
 629         .ndo_stop            = virtnet_close,
 630         .ndo_start_xmit      = start_xmit,
 631         .ndo_validate_addr   = eth_validate_addr,
 632         .ndo_set_mac_address = eth_mac_addr,
 633         .ndo_change_mtu      = virtnet_change_mtu,
 634 #ifdef CONFIG_NET_POLL_CONTROLLER
 635         .ndo_poll_controller = virtnet_netpoll,
 636 #endif
 637 };
 638
 639 static int virtnet_probe(struct virtio_device *vdev)
 640 {
 641         int err;
 642         struct net_device *dev;
 643         struct virtnet_info *vi;
 644
 645         /* Allocate ourselves a network device with room for our info */
 646         dev = alloc_etherdev(sizeof(struct virtnet_info));
 647         if (!dev)
 648                 return -ENOMEM;
 649
 650         /* Set up network device as normal. */
 651         dev->netdev_ops = &virtnet_netdev;
 652         dev->features = NETIF_F_HIGHDMA;
 653         SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops);
 654         SET_NETDEV_DEV(dev, &vdev->dev);
 655
 656         /* Do we support "hardware" checksums? */
 657         if (csum && virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
 658                 /* This opens up the world of extra features. */
 659                 dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
 660                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
 661                         dev->features |= NETIF_F_TSO | NETIF_F_UFO
 662                                 | NETIF_F_TSO_ECN | NETIF_F_TSO6;
 663                 }
 664                 /* Individual feature bits: what can host handle? */
 665                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
 666                         dev->features |= NETIF_F_TSO;
 667                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
 668                         dev->features |= NETIF_F_TSO6;
 669                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
 670                         dev->features |= NETIF_F_TSO_ECN;
 671                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
 672                         dev->features |= NETIF_F_UFO;
 673         }
 674
 675         /* Configuration may specify what MAC to use.  Otherwise random. */
 676         if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) {
 677                 vdev->config->get(vdev,
 678                                   offsetof(struct virtio_net_config, mac),
 679                                   dev->dev_addr, dev->addr_len);
 680         } else
 681                 random_ether_addr(dev->dev_addr);
 682
 683         /* Set up our device-specific information */
 684         vi = netdev_priv(dev);
 685         netif_napi_add(dev, &vi->napi, virtnet_poll, napi_weight);
 686         vi->dev = dev;
 687         vi->vdev = vdev;
 688         vdev->priv = vi;
 689         vi->pages = NULL;
 690
 691         /* If they give us a callback when all buffers are done, we don't need
 692          * the timer. */
 693         vi->free_in_tasklet = virtio_has_feature(vdev,VIRTIO_F_NOTIFY_ON_EMPTY);
 694
 695         /* If we can receive ANY GSO packets, we must allocate large ones. */
 696         if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4)
 697             || virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6)
 698             || virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN))
 699                 vi->big_packets = true;
 700
 701         if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
 702                 vi->mergeable_rx_bufs = true;
 703
 704         /* We expect two virtqueues, receive then send. */
 705         vi->rvq = vdev->config->find_vq(vdev, 0, skb_recv_done);
 706         if (IS_ERR(vi->rvq)) {
 707                 err = PTR_ERR(vi->rvq);
 708                 goto free;
 709         }
 710
 711         vi->svq = vdev->config->find_vq(vdev, 1, skb_xmit_done);
 712         if (IS_ERR(vi->svq)) {
 713                 err = PTR_ERR(vi->svq);
 714                 goto free_recv;
 715         }
 716
 717         /* Initialize our empty receive and send queues. */
 718         skb_queue_head_init(&vi->recv);
 719         skb_queue_head_init(&vi->send);
 720
 721         tasklet_init(&vi->tasklet, xmit_tasklet, (unsigned long)vi);
 722
 723         if (!vi->free_in_tasklet)
 724                 setup_timer(&vi->xmit_free_timer, xmit_free, (unsigned long)vi);
 725
 726         err = register_netdev(dev);
 727         if (err) {
 728                 pr_debug("virtio_net: registering device failed\n");
 729                 goto free_send;
 730         }
 731
 732         /* Last of all, set up some receive buffers. */
 733         try_fill_recv(vi);
 734
 735         /* If we didn't even get one input buffer, we're useless. */
 736         if (vi->num == 0) {
 737                 err = -ENOMEM;
 738                 goto unregister;
 739         }
 740
 741         pr_debug("virtnet: registered device %s\n", dev->name);
 742         return 0;
 743
 744 unregister:
 745         unregister_netdev(dev);
 746 free_send:
 747         vdev->config->del_vq(vi->svq);
 748 free_recv:
 749         vdev->config->del_vq(vi->rvq);
 750 free:
 751         free_netdev(dev);
 752         return err;
 753 }
 754
 755 static void virtnet_remove(struct virtio_device *vdev)
 756 {
 757         struct virtnet_info *vi = vdev->priv;
 758         struct sk_buff *skb;
 759
 760         /* Stop all the virtqueues. */
 761         vdev->config->reset(vdev);
 762
 763         if (!vi->free_in_tasklet)
 764                 del_timer_sync(&vi->xmit_free_timer);
 765
 766         /* Free our skbs in send and recv queues, if any. */
 767         while ((skb = __skb_dequeue(&vi->recv)) != NULL) {
 768                 kfree_skb(skb);
 769                 vi->num--;
 770         }
 771         __skb_queue_purge(&vi->send);
 772
 773         BUG_ON(vi->num != 0);
 774
 775         vdev->config->del_vq(vi->svq);
 776         vdev->config->del_vq(vi->rvq);
 777         unregister_netdev(vi->dev);
 778
 779         while (vi->pages)
 780                 __free_pages(get_a_page(vi, GFP_KERNEL), 0);
 781
 782         free_netdev(vi->dev);
 783 }
 784
 785 static struct virtio_device_id id_table[] = {
 786         { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
 787         { 0 },
 788 };
 789
 790 static unsigned int features[] = {
 791         VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM,
 792         VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC,
 793         VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6,
 794         VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6,
 795         VIRTIO_NET_F_GUEST_ECN, /* We don't yet handle UFO input. */
 796         VIRTIO_NET_F_MRG_RXBUF,
 797         VIRTIO_F_NOTIFY_ON_EMPTY,
 798 };
 799
 800 static struct virtio_driver virtio_net = {
 801         .feature_table = features,
 802         .feature_table_size = ARRAY_SIZE(features),
 803         .driver.name =  KBUILD_MODNAME,
 804         .driver.owner = THIS_MODULE,
 805         .id_table =     id_table,
 806         .probe =        virtnet_probe,
 807         .remove =       __devexit_p(virtnet_remove),
 808 };
 809
 810 static int __init init(void)
 811 {
 812         return register_virtio_driver(&virtio_net);
 813 }
 814
 815 static void __exit fini(void)
 816 {
 817         unregister_virtio_driver(&virtio_net);
 818 }
 819 module_init(init);
 820 module_exit(fini);
 821
 822 MODULE_DEVICE_TABLE(virtio, id_table);
 823 MODULE_DESCRIPTION("Virtio network driver");
 824 MODULE_LICENSE("GPL");