git.oblomov.eu Git - linux-2.6/blob - 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 #include <linux/if_vlan.h>
  28
  29 static int napi_weight = 128;
  30 module_param(napi_weight, int, 0444);
  31
  32 static int csum = 1, gso = 1;
  33 module_param(csum, bool, 0444);
  34 module_param(gso, bool, 0444);
  35
  36 /* FIXME: MTU in config. */
  37 #define MAX_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
  38 #define GOOD_COPY_LEN   128
  39
  40 struct virtnet_info
  41 {
  42         struct virtio_device *vdev;
  43         struct virtqueue *rvq, *svq;
  44         struct net_device *dev;
  45         struct napi_struct napi;
  46         unsigned int status;
  47
  48         /* The skb we couldn't send because buffers were full. */
  49         struct sk_buff *last_xmit_skb;
  50
  51         /* If we need to free in a timer, this is it. */
  52         struct timer_list xmit_free_timer;
  53
  54         /* Number of input buffers, and max we've ever had. */
  55         unsigned int num, max;
  56
  57         /* For cleaning up after transmission. */
  58         struct tasklet_struct tasklet;
  59         bool free_in_tasklet;
  60
  61         /* I like... big packets and I cannot lie! */
  62         bool big_packets;
  63
  64         /* Host will merge rx buffers for big packets (shake it! shake it!) */
  65         bool mergeable_rx_bufs;
  66
  67         /* Receive & send queues. */
  68         struct sk_buff_head recv;
  69         struct sk_buff_head send;
  70
  71         /* Chain pages by the private ptr. */
  72         struct page *pages;
  73 };
  74
  75 static inline void *skb_vnet_hdr(struct sk_buff *skb)
  76 {
  77         return (struct virtio_net_hdr *)skb->cb;
  78 }
  79
  80 static void give_a_page(struct virtnet_info *vi, struct page *page)
  81 {
  82         page->private = (unsigned long)vi->pages;
  83         vi->pages = page;
  84 }
  85
  86 static void trim_pages(struct virtnet_info *vi, struct sk_buff *skb)
  87 {
  88         unsigned int i;
  89
  90         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
  91                 give_a_page(vi, skb_shinfo(skb)->frags[i].page);
  92         skb_shinfo(skb)->nr_frags = 0;
  93         skb->data_len = 0;
  94 }
  95
  96 static struct page *get_a_page(struct virtnet_info *vi, gfp_t gfp_mask)
  97 {
  98         struct page *p = vi->pages;
  99
 100         if (p)
 101                 vi->pages = (struct page *)p->private;
 102         else
 103                 p = alloc_page(gfp_mask);
 104         return p;
 105 }
 106
 107 static void skb_xmit_done(struct virtqueue *svq)
 108 {
 109         struct virtnet_info *vi = svq->vdev->priv;
 110
 111         /* Suppress further interrupts. */
 112         svq->vq_ops->disable_cb(svq);
 113
 114         /* We were probably waiting for more output buffers. */
 115         netif_wake_queue(vi->dev);
 116
 117         /* Make sure we re-xmit last_xmit_skb: if there are no more packets
 118          * queued, start_xmit won't be called. */
 119         tasklet_schedule(&vi->tasklet);
 120 }
 121
 122 static void receive_skb(struct net_device *dev, struct sk_buff *skb,
 123                         unsigned len)
 124 {
 125         struct virtnet_info *vi = netdev_priv(dev);
 126         struct virtio_net_hdr *hdr = skb_vnet_hdr(skb);
 127         int err;
 128         int i;
 129
 130         if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) {
 131                 pr_debug("%s: short packet %i\n", dev->name, len);
 132                 dev->stats.rx_length_errors++;
 133                 goto drop;
 134         }
 135
 136         if (vi->mergeable_rx_bufs) {
 137                 struct virtio_net_hdr_mrg_rxbuf *mhdr = skb_vnet_hdr(skb);
 138                 unsigned int copy;
 139                 char *p = page_address(skb_shinfo(skb)->frags[0].page);
 140
 141                 if (len > PAGE_SIZE)
 142                         len = PAGE_SIZE;
 143                 len -= sizeof(struct virtio_net_hdr_mrg_rxbuf);
 144
 145                 memcpy(hdr, p, sizeof(*mhdr));
 146                 p += sizeof(*mhdr);
 147
 148                 copy = len;
 149                 if (copy > skb_tailroom(skb))
 150                         copy = skb_tailroom(skb);
 151
 152                 memcpy(skb_put(skb, copy), p, copy);
 153
 154                 len -= copy;
 155
 156                 if (!len) {
 157                         give_a_page(vi, skb_shinfo(skb)->frags[0].page);
 158                         skb_shinfo(skb)->nr_frags--;
 159                 } else {
 160                         skb_shinfo(skb)->frags[0].page_offset +=
 161                                 sizeof(*mhdr) + copy;
 162                         skb_shinfo(skb)->frags[0].size = len;
 163                         skb->data_len += len;
 164                         skb->len += len;
 165                 }
 166
 167                 while (--mhdr->num_buffers) {
 168                         struct sk_buff *nskb;
 169
 170                         i = skb_shinfo(skb)->nr_frags;
 171                         if (i >= MAX_SKB_FRAGS) {
 172                                 pr_debug("%s: packet too long %d\n", dev->name,
 173                                          len);
 174                                 dev->stats.rx_length_errors++;
 175                                 goto drop;
 176                         }
 177
 178                         nskb = vi->rvq->vq_ops->get_buf(vi->rvq, &len);
 179                         if (!nskb) {
 180                                 pr_debug("%s: rx error: %d buffers missing\n",
 181                                          dev->name, mhdr->num_buffers);
 182                                 dev->stats.rx_length_errors++;
 183                                 goto drop;
 184                         }
 185
 186                         __skb_unlink(nskb, &vi->recv);
 187                         vi->num--;
 188
 189                         skb_shinfo(skb)->frags[i] = skb_shinfo(nskb)->frags[0];
 190                         skb_shinfo(nskb)->nr_frags = 0;
 191                         kfree_skb(nskb);
 192
 193                         if (len > PAGE_SIZE)
 194                                 len = PAGE_SIZE;
 195
 196                         skb_shinfo(skb)->frags[i].size = len;
 197                         skb_shinfo(skb)->nr_frags++;
 198                         skb->data_len += len;
 199                         skb->len += len;
 200                 }
 201         } else {
 202                 len -= sizeof(struct virtio_net_hdr);
 203
 204                 if (len <= MAX_PACKET_LEN)
 205                         trim_pages(vi, skb);
 206
 207                 err = pskb_trim(skb, len);
 208                 if (err) {
 209                         pr_debug("%s: pskb_trim failed %i %d\n", dev->name,
 210                                  len, err);
 211                         dev->stats.rx_dropped++;
 212                         goto drop;
 213                 }
 214         }
 215
 216         skb->truesize += skb->data_len;
 217         dev->stats.rx_bytes += skb->len;
 218         dev->stats.rx_packets++;
 219
 220         if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
 221                 pr_debug("Needs csum!\n");
 222                 if (!skb_partial_csum_set(skb,hdr->csum_start,hdr->csum_offset))
 223                         goto frame_err;
 224         }
 225
 226         skb->protocol = eth_type_trans(skb, dev);
 227         pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
 228                  ntohs(skb->protocol), skb->len, skb->pkt_type);
 229
 230         if (hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
 231                 pr_debug("GSO!\n");
 232                 switch (hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
 233                 case VIRTIO_NET_HDR_GSO_TCPV4:
 234                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
 235                         break;
 236                 case VIRTIO_NET_HDR_GSO_UDP:
 237                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
 238                         break;
 239                 case VIRTIO_NET_HDR_GSO_TCPV6:
 240                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
 241                         break;
 242                 default:
 243                         if (net_ratelimit())
 244                                 printk(KERN_WARNING "%s: bad gso type %u.\n",
 245                                        dev->name, hdr->gso_type);
 246                         goto frame_err;
 247                 }
 248
 249                 if (hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
 250                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
 251
 252                 skb_shinfo(skb)->gso_size = hdr->gso_size;
 253                 if (skb_shinfo(skb)->gso_size == 0) {
 254                         if (net_ratelimit())
 255                                 printk(KERN_WARNING "%s: zero gso size.\n",
 256                                        dev->name);
 257                         goto frame_err;
 258                 }
 259
 260                 /* Header must be checked, and gso_segs computed. */
 261                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
 262                 skb_shinfo(skb)->gso_segs = 0;
 263         }
 264
 265         netif_receive_skb(skb);
 266         return;
 267
 268 frame_err:
 269         dev->stats.rx_frame_errors++;
 270 drop:
 271         dev_kfree_skb(skb);
 272 }
 273
 274 static void try_fill_recv_maxbufs(struct virtnet_info *vi)
 275 {
 276         struct sk_buff *skb;
 277         struct scatterlist sg[2+MAX_SKB_FRAGS];
 278         int num, err, i;
 279
 280         sg_init_table(sg, 2+MAX_SKB_FRAGS);
 281         for (;;) {
 282                 struct virtio_net_hdr *hdr;
 283
 284                 skb = netdev_alloc_skb(vi->dev, MAX_PACKET_LEN);
 285                 if (unlikely(!skb))
 286                         break;
 287
 288                 skb_put(skb, MAX_PACKET_LEN);
 289
 290                 hdr = skb_vnet_hdr(skb);
 291                 sg_init_one(sg, hdr, sizeof(*hdr));
 292
 293                 if (vi->big_packets) {
 294                         for (i = 0; i < MAX_SKB_FRAGS; i++) {
 295                                 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
 296                                 f->page = get_a_page(vi, GFP_ATOMIC);
 297                                 if (!f->page)
 298                                         break;
 299
 300                                 f->page_offset = 0;
 301                                 f->size = PAGE_SIZE;
 302
 303                                 skb->data_len += PAGE_SIZE;
 304                                 skb->len += PAGE_SIZE;
 305
 306                                 skb_shinfo(skb)->nr_frags++;
 307                         }
 308                 }
 309
 310                 num = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
 311                 skb_queue_head(&vi->recv, skb);
 312
 313                 err = vi->rvq->vq_ops->add_buf(vi->rvq, sg, 0, num, skb);
 314                 if (err) {
 315                         skb_unlink(skb, &vi->recv);
 316                         trim_pages(vi, skb);
 317                         kfree_skb(skb);
 318                         break;
 319                 }
 320                 vi->num++;
 321         }
 322         if (unlikely(vi->num > vi->max))
 323                 vi->max = vi->num;
 324         vi->rvq->vq_ops->kick(vi->rvq);
 325 }
 326
 327 static void try_fill_recv(struct virtnet_info *vi)
 328 {
 329         struct sk_buff *skb;
 330         struct scatterlist sg[1];
 331         int err;
 332
 333         if (!vi->mergeable_rx_bufs) {
 334                 try_fill_recv_maxbufs(vi);
 335                 return;
 336         }
 337
 338         for (;;) {
 339                 skb_frag_t *f;
 340
 341                 skb = netdev_alloc_skb(vi->dev, GOOD_COPY_LEN + NET_IP_ALIGN);
 342                 if (unlikely(!skb))
 343                         break;
 344
 345                 skb_reserve(skb, NET_IP_ALIGN);
 346
 347                 f = &skb_shinfo(skb)->frags[0];
 348                 f->page = get_a_page(vi, GFP_ATOMIC);
 349                 if (!f->page) {
 350                         kfree_skb(skb);
 351                         break;
 352                 }
 353
 354                 f->page_offset = 0;
 355                 f->size = PAGE_SIZE;
 356
 357                 skb_shinfo(skb)->nr_frags++;
 358
 359                 sg_init_one(sg, page_address(f->page), PAGE_SIZE);
 360                 skb_queue_head(&vi->recv, skb);
 361
 362                 err = vi->rvq->vq_ops->add_buf(vi->rvq, sg, 0, 1, skb);
 363                 if (err) {
 364                         skb_unlink(skb, &vi->recv);
 365                         kfree_skb(skb);
 366                         break;
 367                 }
 368                 vi->num++;
 369         }
 370         if (unlikely(vi->num > vi->max))
 371                 vi->max = vi->num;
 372         vi->rvq->vq_ops->kick(vi->rvq);
 373 }
 374
 375 static void skb_recv_done(struct virtqueue *rvq)
 376 {
 377         struct virtnet_info *vi = rvq->vdev->priv;
 378         /* Schedule NAPI, Suppress further interrupts if successful. */
 379         if (napi_schedule_prep(&vi->napi)) {
 380                 rvq->vq_ops->disable_cb(rvq);
 381                 __napi_schedule(&vi->napi);
 382         }
 383 }
 384
 385 static int virtnet_poll(struct napi_struct *napi, int budget)
 386 {
 387         struct virtnet_info *vi = container_of(napi, struct virtnet_info, napi);
 388         struct sk_buff *skb = NULL;
 389         unsigned int len, received = 0;
 390
 391 again:
 392         while (received < budget &&
 393                (skb = vi->rvq->vq_ops->get_buf(vi->rvq, &len)) != NULL) {
 394                 __skb_unlink(skb, &vi->recv);
 395                 receive_skb(vi->dev, skb, len);
 396                 vi->num--;
 397                 received++;
 398         }
 399
 400         /* FIXME: If we oom and completely run out of inbufs, we need
 401          * to start a timer trying to fill more. */
 402         if (vi->num < vi->max / 2)
 403                 try_fill_recv(vi);
 404
 405         /* Out of packets? */
 406         if (received < budget) {
 407                 napi_complete(napi);
 408                 if (unlikely(!vi->rvq->vq_ops->enable_cb(vi->rvq))
 409                     && napi_schedule_prep(napi)) {
 410                         vi->rvq->vq_ops->disable_cb(vi->rvq);
 411                         __napi_schedule(napi);
 412                         goto again;
 413                 }
 414         }
 415
 416         return received;
 417 }
 418
 419 static void free_old_xmit_skbs(struct virtnet_info *vi)
 420 {
 421         struct sk_buff *skb;
 422         unsigned int len;
 423
 424         while ((skb = vi->svq->vq_ops->get_buf(vi->svq, &len)) != NULL) {
 425                 pr_debug("Sent skb %p\n", skb);
 426                 __skb_unlink(skb, &vi->send);
 427                 vi->dev->stats.tx_bytes += skb->len;
 428                 vi->dev->stats.tx_packets++;
 429                 kfree_skb(skb);
 430         }
 431 }
 432
 433 /* If the virtio transport doesn't always notify us when all in-flight packets
 434  * are consumed, we fall back to using this function on a timer to free them. */
 435 static void xmit_free(unsigned long data)
 436 {
 437         struct virtnet_info *vi = (void *)data;
 438
 439         netif_tx_lock(vi->dev);
 440
 441         free_old_xmit_skbs(vi);
 442
 443         if (!skb_queue_empty(&vi->send))
 444                 mod_timer(&vi->xmit_free_timer, jiffies + (HZ/10));
 445
 446         netif_tx_unlock(vi->dev);
 447 }
 448
 449 static int xmit_skb(struct virtnet_info *vi, struct sk_buff *skb)
 450 {
 451         int num, err;
 452         struct scatterlist sg[2+MAX_SKB_FRAGS];
 453         struct virtio_net_hdr_mrg_rxbuf *mhdr = skb_vnet_hdr(skb);
 454         struct virtio_net_hdr *hdr = skb_vnet_hdr(skb);
 455         const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
 456
 457         sg_init_table(sg, 2+MAX_SKB_FRAGS);
 458
 459         pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
 460
 461         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 462                 hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
 463                 hdr->csum_start = skb->csum_start - skb_headroom(skb);
 464                 hdr->csum_offset = skb->csum_offset;
 465         } else {
 466                 hdr->flags = 0;
 467                 hdr->csum_offset = hdr->csum_start = 0;
 468         }
 469
 470         if (skb_is_gso(skb)) {
 471                 hdr->hdr_len = skb_transport_header(skb) - skb->data;
 472                 hdr->gso_size = skb_shinfo(skb)->gso_size;
 473                 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
 474                         hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
 475                 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
 476                         hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
 477                 else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 478                         hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
 479                 else
 480                         BUG();
 481                 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
 482                         hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
 483         } else {
 484                 hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
 485                 hdr->gso_size = hdr->hdr_len = 0;
 486         }
 487
 488         mhdr->num_buffers = 0;
 489
 490         /* Encode metadata header at front. */
 491         if (vi->mergeable_rx_bufs)
 492                 sg_init_one(sg, mhdr, sizeof(*mhdr));
 493         else
 494                 sg_init_one(sg, hdr, sizeof(*hdr));
 495
 496         num = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
 497
 498         err = vi->svq->vq_ops->add_buf(vi->svq, sg, num, 0, skb);
 499         if (!err && !vi->free_in_tasklet)
 500                 mod_timer(&vi->xmit_free_timer, jiffies + (HZ/10));
 501
 502         return err;
 503 }
 504
 505 static void xmit_tasklet(unsigned long data)
 506 {
 507         struct virtnet_info *vi = (void *)data;
 508
 509         netif_tx_lock_bh(vi->dev);
 510         if (vi->last_xmit_skb && xmit_skb(vi, vi->last_xmit_skb) == 0) {
 511                 vi->svq->vq_ops->kick(vi->svq);
 512                 vi->last_xmit_skb = NULL;
 513         }
 514         if (vi->free_in_tasklet)
 515                 free_old_xmit_skbs(vi);
 516         netif_tx_unlock_bh(vi->dev);
 517 }
 518
 519 static int start_xmit(struct sk_buff *skb, struct net_device *dev)
 520 {
 521         struct virtnet_info *vi = netdev_priv(dev);
 522
 523 again:
 524         /* Free up any pending old buffers before queueing new ones. */
 525         free_old_xmit_skbs(vi);
 526
 527         /* If we has a buffer left over from last time, send it now. */
 528         if (unlikely(vi->last_xmit_skb) &&
 529             xmit_skb(vi, vi->last_xmit_skb) != 0)
 530                 goto stop_queue;
 531
 532         vi->last_xmit_skb = NULL;
 533
 534         /* Put new one in send queue and do transmit */
 535         if (likely(skb)) {
 536                 __skb_queue_head(&vi->send, skb);
 537                 if (xmit_skb(vi, skb) != 0) {
 538                         vi->last_xmit_skb = skb;
 539                         skb = NULL;
 540                         goto stop_queue;
 541                 }
 542         }
 543 done:
 544         vi->svq->vq_ops->kick(vi->svq);
 545         return NETDEV_TX_OK;
 546
 547 stop_queue:
 548         pr_debug("%s: virtio not prepared to send\n", dev->name);
 549         netif_stop_queue(dev);
 550
 551         /* Activate callback for using skbs: if this returns false it
 552          * means some were used in the meantime. */
 553         if (unlikely(!vi->svq->vq_ops->enable_cb(vi->svq))) {
 554                 vi->svq->vq_ops->disable_cb(vi->svq);
 555                 netif_start_queue(dev);
 556                 goto again;
 557         }
 558         if (skb) {
 559                 /* Drop this skb: we only queue one. */
 560                 vi->dev->stats.tx_dropped++;
 561                 kfree_skb(skb);
 562         }
 563         goto done;
 564 }
 565
 566 #ifdef CONFIG_NET_POLL_CONTROLLER
 567 static void virtnet_netpoll(struct net_device *dev)
 568 {
 569         struct virtnet_info *vi = netdev_priv(dev);
 570
 571         napi_schedule(&vi->napi);
 572 }
 573 #endif
 574
 575 static int virtnet_open(struct net_device *dev)
 576 {
 577         struct virtnet_info *vi = netdev_priv(dev);
 578
 579         napi_enable(&vi->napi);
 580
 581         /* If all buffers were filled by other side before we napi_enabled, we
 582          * won't get another interrupt, so process any outstanding packets
 583          * now.  virtnet_poll wants re-enable the queue, so we disable here.
 584          * We synchronize against interrupts via NAPI_STATE_SCHED */
 585         if (napi_schedule_prep(&vi->napi)) {
 586                 vi->rvq->vq_ops->disable_cb(vi->rvq);
 587                 __napi_schedule(&vi->napi);
 588         }
 589         return 0;
 590 }
 591
 592 static int virtnet_close(struct net_device *dev)
 593 {
 594         struct virtnet_info *vi = netdev_priv(dev);
 595
 596         napi_disable(&vi->napi);
 597
 598         return 0;
 599 }
 600
 601 static int virtnet_set_tx_csum(struct net_device *dev, u32 data)
 602 {
 603         struct virtnet_info *vi = netdev_priv(dev);
 604         struct virtio_device *vdev = vi->vdev;
 605
 606         if (data && !virtio_has_feature(vdev, VIRTIO_NET_F_CSUM))
 607                 return -ENOSYS;
 608
 609         return ethtool_op_set_tx_hw_csum(dev, data);
 610 }
 611
 612 static struct ethtool_ops virtnet_ethtool_ops = {
 613         .set_tx_csum = virtnet_set_tx_csum,
 614         .set_sg = ethtool_op_set_sg,
 615         .set_tso = ethtool_op_set_tso,
 616         .get_link = ethtool_op_get_link,
 617 };
 618
 619 #define MIN_MTU 68
 620 #define MAX_MTU 65535
 621
 622 static int virtnet_change_mtu(struct net_device *dev, int new_mtu)
 623 {
 624         if (new_mtu < MIN_MTU || new_mtu > MAX_MTU)
 625                 return -EINVAL;
 626         dev->mtu = new_mtu;
 627         return 0;
 628 }
 629
 630 static const struct net_device_ops virtnet_netdev = {
 631         .ndo_open            = virtnet_open,
 632         .ndo_stop            = virtnet_close,
 633         .ndo_start_xmit      = start_xmit,
 634         .ndo_validate_addr   = eth_validate_addr,
 635         .ndo_set_mac_address = eth_mac_addr,
 636         .ndo_change_mtu      = virtnet_change_mtu,
 637 #ifdef CONFIG_NET_POLL_CONTROLLER
 638         .ndo_poll_controller = virtnet_netpoll,
 639 #endif
 640 };
 641
 642 static void virtnet_update_status(struct virtnet_info *vi)
 643 {
 644         u16 v;
 645
 646         if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS))
 647                 return;
 648
 649         vi->vdev->config->get(vi->vdev,
 650                               offsetof(struct virtio_net_config, status),
 651                               &v, sizeof(v));
 652
 653         /* Ignore unknown (future) status bits */
 654         v &= VIRTIO_NET_S_LINK_UP;
 655
 656         if (vi->status == v)
 657                 return;
 658
 659         vi->status = v;
 660
 661         if (vi->status & VIRTIO_NET_S_LINK_UP) {
 662                 netif_carrier_on(vi->dev);
 663                 netif_wake_queue(vi->dev);
 664         } else {
 665                 netif_carrier_off(vi->dev);
 666                 netif_stop_queue(vi->dev);
 667         }
 668 }
 669
 670 static void virtnet_config_changed(struct virtio_device *vdev)
 671 {
 672         struct virtnet_info *vi = vdev->priv;
 673
 674         virtnet_update_status(vi);
 675 }
 676
 677 static int virtnet_probe(struct virtio_device *vdev)
 678 {
 679         int err;
 680         struct net_device *dev;
 681         struct virtnet_info *vi;
 682
 683         /* Allocate ourselves a network device with room for our info */
 684         dev = alloc_etherdev(sizeof(struct virtnet_info));
 685         if (!dev)
 686                 return -ENOMEM;
 687
 688         /* Set up network device as normal. */
 689         dev->netdev_ops = &virtnet_netdev;
 690         dev->features = NETIF_F_HIGHDMA;
 691         SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops);
 692         SET_NETDEV_DEV(dev, &vdev->dev);
 693
 694         /* Do we support "hardware" checksums? */
 695         if (csum && virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
 696                 /* This opens up the world of extra features. */
 697                 dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
 698                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
 699                         dev->features |= NETIF_F_TSO | NETIF_F_UFO
 700                                 | NETIF_F_TSO_ECN | NETIF_F_TSO6;
 701                 }
 702                 /* Individual feature bits: what can host handle? */
 703                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
 704                         dev->features |= NETIF_F_TSO;
 705                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
 706                         dev->features |= NETIF_F_TSO6;
 707                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
 708                         dev->features |= NETIF_F_TSO_ECN;
 709                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
 710                         dev->features |= NETIF_F_UFO;
 711         }
 712
 713         /* Configuration may specify what MAC to use.  Otherwise random. */
 714         if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) {
 715                 vdev->config->get(vdev,
 716                                   offsetof(struct virtio_net_config, mac),
 717                                   dev->dev_addr, dev->addr_len);
 718         } else
 719                 random_ether_addr(dev->dev_addr);
 720
 721         /* Set up our device-specific information */
 722         vi = netdev_priv(dev);
 723         netif_napi_add(dev, &vi->napi, virtnet_poll, napi_weight);
 724         vi->dev = dev;
 725         vi->vdev = vdev;
 726         vdev->priv = vi;
 727         vi->pages = NULL;
 728
 729         /* If they give us a callback when all buffers are done, we don't need
 730          * the timer. */
 731         vi->free_in_tasklet = virtio_has_feature(vdev,VIRTIO_F_NOTIFY_ON_EMPTY);
 732
 733         /* If we can receive ANY GSO packets, we must allocate large ones. */
 734         if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4)
 735             || virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6)
 736             || virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN))
 737                 vi->big_packets = true;
 738
 739         if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
 740                 vi->mergeable_rx_bufs = true;
 741
 742         /* We expect two virtqueues, receive then send. */
 743         vi->rvq = vdev->config->find_vq(vdev, 0, skb_recv_done);
 744         if (IS_ERR(vi->rvq)) {
 745                 err = PTR_ERR(vi->rvq);
 746                 goto free;
 747         }
 748
 749         vi->svq = vdev->config->find_vq(vdev, 1, skb_xmit_done);
 750         if (IS_ERR(vi->svq)) {
 751                 err = PTR_ERR(vi->svq);
 752                 goto free_recv;
 753         }
 754
 755         /* Initialize our empty receive and send queues. */
 756         skb_queue_head_init(&vi->recv);
 757         skb_queue_head_init(&vi->send);
 758
 759         tasklet_init(&vi->tasklet, xmit_tasklet, (unsigned long)vi);
 760
 761         if (!vi->free_in_tasklet)
 762                 setup_timer(&vi->xmit_free_timer, xmit_free, (unsigned long)vi);
 763
 764         err = register_netdev(dev);
 765         if (err) {
 766                 pr_debug("virtio_net: registering device failed\n");
 767                 goto free_send;
 768         }
 769
 770         /* Last of all, set up some receive buffers. */
 771         try_fill_recv(vi);
 772
 773         /* If we didn't even get one input buffer, we're useless. */
 774         if (vi->num == 0) {
 775                 err = -ENOMEM;
 776                 goto unregister;
 777         }
 778
 779         vi->status = VIRTIO_NET_S_LINK_UP;
 780         virtnet_update_status(vi);
 781
 782         pr_debug("virtnet: registered device %s\n", dev->name);
 783         return 0;
 784
 785 unregister:
 786         unregister_netdev(dev);
 787 free_send:
 788         vdev->config->del_vq(vi->svq);
 789 free_recv:
 790         vdev->config->del_vq(vi->rvq);
 791 free:
 792         free_netdev(dev);
 793         return err;
 794 }
 795
 796 static void virtnet_remove(struct virtio_device *vdev)
 797 {
 798         struct virtnet_info *vi = vdev->priv;
 799         struct sk_buff *skb;
 800
 801         /* Stop all the virtqueues. */
 802         vdev->config->reset(vdev);
 803
 804         if (!vi->free_in_tasklet)
 805                 del_timer_sync(&vi->xmit_free_timer);
 806
 807         /* Free our skbs in send and recv queues, if any. */
 808         while ((skb = __skb_dequeue(&vi->recv)) != NULL) {
 809                 kfree_skb(skb);
 810                 vi->num--;
 811         }
 812         __skb_queue_purge(&vi->send);
 813
 814         BUG_ON(vi->num != 0);
 815
 816         vdev->config->del_vq(vi->svq);
 817         vdev->config->del_vq(vi->rvq);
 818         unregister_netdev(vi->dev);
 819
 820         while (vi->pages)
 821                 __free_pages(get_a_page(vi, GFP_KERNEL), 0);
 822
 823         free_netdev(vi->dev);
 824 }
 825
 826 static struct virtio_device_id id_table[] = {
 827         { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
 828         { 0 },
 829 };
 830
 831 static unsigned int features[] = {
 832         VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM,
 833         VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC,
 834         VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6,
 835         VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6,
 836         VIRTIO_NET_F_GUEST_ECN, /* We don't yet handle UFO input. */
 837         VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS,
 838         VIRTIO_F_NOTIFY_ON_EMPTY,
 839 };
 840
 841 static struct virtio_driver virtio_net = {
 842         .feature_table = features,
 843         .feature_table_size = ARRAY_SIZE(features),
 844         .driver.name =  KBUILD_MODNAME,
 845         .driver.owner = THIS_MODULE,
 846         .id_table =     id_table,
 847         .probe =        virtnet_probe,
 848         .remove =       __devexit_p(virtnet_remove),
 849         .config_changed = virtnet_config_changed,
 850 };
 851
 852 static int __init init(void)
 853 {
 854         return register_virtio_driver(&virtio_net);
 855 }
 856
 857 static void __exit fini(void)
 858 {
 859         unregister_virtio_driver(&virtio_net);
 860 }
 861 module_init(init);
 862 module_exit(fini);
 863
 864 MODULE_DEVICE_TABLE(virtio, id_table);
 865 MODULE_DESCRIPTION("Virtio network driver");
 866 MODULE_LICENSE("GPL");