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 #define VIRTNET_SEND_COMMAND_SG_MAX    2
  41
  42 struct virtnet_info
  43 {
  44         struct virtio_device *vdev;
  45         struct virtqueue *rvq, *svq, *cvq;
  46         struct net_device *dev;
  47         struct napi_struct napi;
  48         unsigned int status;
  49
  50         /* The skb we couldn't send because buffers were full. */
  51         struct sk_buff *last_xmit_skb;
  52
  53         /* If we need to free in a timer, this is it. */
  54         struct timer_list xmit_free_timer;
  55
  56         /* Number of input buffers, and max we've ever had. */
  57         unsigned int num, max;
  58
  59         /* For cleaning up after transmission. */
  60         struct tasklet_struct tasklet;
  61         bool free_in_tasklet;
  62
  63         /* I like... big packets and I cannot lie! */
  64         bool big_packets;
  65
  66         /* Host will merge rx buffers for big packets (shake it! shake it!) */
  67         bool mergeable_rx_bufs;
  68
  69         /* Receive & send queues. */
  70         struct sk_buff_head recv;
  71         struct sk_buff_head send;
  72
  73         /* Chain pages by the private ptr. */
  74         struct page *pages;
  75 };
  76
  77 static inline void *skb_vnet_hdr(struct sk_buff *skb)
  78 {
  79         return (struct virtio_net_hdr *)skb->cb;
  80 }
  81
  82 static void give_a_page(struct virtnet_info *vi, struct page *page)
  83 {
  84         page->private = (unsigned long)vi->pages;
  85         vi->pages = page;
  86 }
  87
  88 static void trim_pages(struct virtnet_info *vi, struct sk_buff *skb)
  89 {
  90         unsigned int i;
  91
  92         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
  93                 give_a_page(vi, skb_shinfo(skb)->frags[i].page);
  94         skb_shinfo(skb)->nr_frags = 0;
  95         skb->data_len = 0;
  96 }
  97
  98 static struct page *get_a_page(struct virtnet_info *vi, gfp_t gfp_mask)
  99 {
 100         struct page *p = vi->pages;
 101
 102         if (p)
 103                 vi->pages = (struct page *)p->private;
 104         else
 105                 p = alloc_page(gfp_mask);
 106         return p;
 107 }
 108
 109 static void skb_xmit_done(struct virtqueue *svq)
 110 {
 111         struct virtnet_info *vi = svq->vdev->priv;
 112
 113         /* Suppress further interrupts. */
 114         svq->vq_ops->disable_cb(svq);
 115
 116         /* We were probably waiting for more output buffers. */
 117         netif_wake_queue(vi->dev);
 118
 119         /* Make sure we re-xmit last_xmit_skb: if there are no more packets
 120          * queued, start_xmit won't be called. */
 121         tasklet_schedule(&vi->tasklet);
 122 }
 123
 124 static void receive_skb(struct net_device *dev, struct sk_buff *skb,
 125                         unsigned len)
 126 {
 127         struct virtnet_info *vi = netdev_priv(dev);
 128         struct virtio_net_hdr *hdr = skb_vnet_hdr(skb);
 129         int err;
 130         int i;
 131
 132         if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) {
 133                 pr_debug("%s: short packet %i\n", dev->name, len);
 134                 dev->stats.rx_length_errors++;
 135                 goto drop;
 136         }
 137
 138         if (vi->mergeable_rx_bufs) {
 139                 struct virtio_net_hdr_mrg_rxbuf *mhdr = skb_vnet_hdr(skb);
 140                 unsigned int copy;
 141                 char *p = page_address(skb_shinfo(skb)->frags[0].page);
 142
 143                 if (len > PAGE_SIZE)
 144                         len = PAGE_SIZE;
 145                 len -= sizeof(struct virtio_net_hdr_mrg_rxbuf);
 146
 147                 memcpy(hdr, p, sizeof(*mhdr));
 148                 p += sizeof(*mhdr);
 149
 150                 copy = len;
 151                 if (copy > skb_tailroom(skb))
 152                         copy = skb_tailroom(skb);
 153
 154                 memcpy(skb_put(skb, copy), p, copy);
 155
 156                 len -= copy;
 157
 158                 if (!len) {
 159                         give_a_page(vi, skb_shinfo(skb)->frags[0].page);
 160                         skb_shinfo(skb)->nr_frags--;
 161                 } else {
 162                         skb_shinfo(skb)->frags[0].page_offset +=
 163                                 sizeof(*mhdr) + copy;
 164                         skb_shinfo(skb)->frags[0].size = len;
 165                         skb->data_len += len;
 166                         skb->len += len;
 167                 }
 168
 169                 while (--mhdr->num_buffers) {
 170                         struct sk_buff *nskb;
 171
 172                         i = skb_shinfo(skb)->nr_frags;
 173                         if (i >= MAX_SKB_FRAGS) {
 174                                 pr_debug("%s: packet too long %d\n", dev->name,
 175                                          len);
 176                                 dev->stats.rx_length_errors++;
 177                                 goto drop;
 178                         }
 179
 180                         nskb = vi->rvq->vq_ops->get_buf(vi->rvq, &len);
 181                         if (!nskb) {
 182                                 pr_debug("%s: rx error: %d buffers missing\n",
 183                                          dev->name, mhdr->num_buffers);
 184                                 dev->stats.rx_length_errors++;
 185                                 goto drop;
 186                         }
 187
 188                         __skb_unlink(nskb, &vi->recv);
 189                         vi->num--;
 190
 191                         skb_shinfo(skb)->frags[i] = skb_shinfo(nskb)->frags[0];
 192                         skb_shinfo(nskb)->nr_frags = 0;
 193                         kfree_skb(nskb);
 194
 195                         if (len > PAGE_SIZE)
 196                                 len = PAGE_SIZE;
 197
 198                         skb_shinfo(skb)->frags[i].size = len;
 199                         skb_shinfo(skb)->nr_frags++;
 200                         skb->data_len += len;
 201                         skb->len += len;
 202                 }
 203         } else {
 204                 len -= sizeof(struct virtio_net_hdr);
 205
 206                 if (len <= MAX_PACKET_LEN)
 207                         trim_pages(vi, skb);
 208
 209                 err = pskb_trim(skb, len);
 210                 if (err) {
 211                         pr_debug("%s: pskb_trim failed %i %d\n", dev->name,
 212                                  len, err);
 213                         dev->stats.rx_dropped++;
 214                         goto drop;
 215                 }
 216         }
 217
 218         skb->truesize += skb->data_len;
 219         dev->stats.rx_bytes += skb->len;
 220         dev->stats.rx_packets++;
 221
 222         if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
 223                 pr_debug("Needs csum!\n");
 224                 if (!skb_partial_csum_set(skb,hdr->csum_start,hdr->csum_offset))
 225                         goto frame_err;
 226         }
 227
 228         skb->protocol = eth_type_trans(skb, dev);
 229         pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
 230                  ntohs(skb->protocol), skb->len, skb->pkt_type);
 231
 232         if (hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
 233                 pr_debug("GSO!\n");
 234                 switch (hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
 235                 case VIRTIO_NET_HDR_GSO_TCPV4:
 236                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
 237                         break;
 238                 case VIRTIO_NET_HDR_GSO_UDP:
 239                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
 240                         break;
 241                 case VIRTIO_NET_HDR_GSO_TCPV6:
 242                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
 243                         break;
 244                 default:
 245                         if (net_ratelimit())
 246                                 printk(KERN_WARNING "%s: bad gso type %u.\n",
 247                                        dev->name, hdr->gso_type);
 248                         goto frame_err;
 249                 }
 250
 251                 if (hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
 252                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
 253
 254                 skb_shinfo(skb)->gso_size = hdr->gso_size;
 255                 if (skb_shinfo(skb)->gso_size == 0) {
 256                         if (net_ratelimit())
 257                                 printk(KERN_WARNING "%s: zero gso size.\n",
 258                                        dev->name);
 259                         goto frame_err;
 260                 }
 261
 262                 /* Header must be checked, and gso_segs computed. */
 263                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
 264                 skb_shinfo(skb)->gso_segs = 0;
 265         }
 266
 267         netif_receive_skb(skb);
 268         return;
 269
 270 frame_err:
 271         dev->stats.rx_frame_errors++;
 272 drop:
 273         dev_kfree_skb(skb);
 274 }
 275
 276 static void try_fill_recv_maxbufs(struct virtnet_info *vi)
 277 {
 278         struct sk_buff *skb;
 279         struct scatterlist sg[2+MAX_SKB_FRAGS];
 280         int num, err, i;
 281
 282         sg_init_table(sg, 2+MAX_SKB_FRAGS);
 283         for (;;) {
 284                 struct virtio_net_hdr *hdr;
 285
 286                 skb = netdev_alloc_skb(vi->dev, MAX_PACKET_LEN + NET_IP_ALIGN);
 287                 if (unlikely(!skb))
 288                         break;
 289
 290                 skb_reserve(skb, NET_IP_ALIGN);
 291                 skb_put(skb, MAX_PACKET_LEN);
 292
 293                 hdr = skb_vnet_hdr(skb);
 294                 sg_set_buf(sg, hdr, sizeof(*hdr));
 295
 296                 if (vi->big_packets) {
 297                         for (i = 0; i < MAX_SKB_FRAGS; i++) {
 298                                 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
 299                                 f->page = get_a_page(vi, GFP_ATOMIC);
 300                                 if (!f->page)
 301                                         break;
 302
 303                                 f->page_offset = 0;
 304                                 f->size = PAGE_SIZE;
 305
 306                                 skb->data_len += PAGE_SIZE;
 307                                 skb->len += PAGE_SIZE;
 308
 309                                 skb_shinfo(skb)->nr_frags++;
 310                         }
 311                 }
 312
 313                 num = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
 314                 skb_queue_head(&vi->recv, skb);
 315
 316                 err = vi->rvq->vq_ops->add_buf(vi->rvq, sg, 0, num, skb);
 317                 if (err) {
 318                         skb_unlink(skb, &vi->recv);
 319                         trim_pages(vi, skb);
 320                         kfree_skb(skb);
 321                         break;
 322                 }
 323                 vi->num++;
 324         }
 325         if (unlikely(vi->num > vi->max))
 326                 vi->max = vi->num;
 327         vi->rvq->vq_ops->kick(vi->rvq);
 328 }
 329
 330 static void try_fill_recv(struct virtnet_info *vi)
 331 {
 332         struct sk_buff *skb;
 333         struct scatterlist sg[1];
 334         int err;
 335
 336         if (!vi->mergeable_rx_bufs) {
 337                 try_fill_recv_maxbufs(vi);
 338                 return;
 339         }
 340
 341         for (;;) {
 342                 skb_frag_t *f;
 343
 344                 skb = netdev_alloc_skb(vi->dev, GOOD_COPY_LEN + NET_IP_ALIGN);
 345                 if (unlikely(!skb))
 346                         break;
 347
 348                 skb_reserve(skb, NET_IP_ALIGN);
 349
 350                 f = &skb_shinfo(skb)->frags[0];
 351                 f->page = get_a_page(vi, GFP_ATOMIC);
 352                 if (!f->page) {
 353                         kfree_skb(skb);
 354                         break;
 355                 }
 356
 357                 f->page_offset = 0;
 358                 f->size = PAGE_SIZE;
 359
 360                 skb_shinfo(skb)->nr_frags++;
 361
 362                 sg_init_one(sg, page_address(f->page), PAGE_SIZE);
 363                 skb_queue_head(&vi->recv, skb);
 364
 365                 err = vi->rvq->vq_ops->add_buf(vi->rvq, sg, 0, 1, skb);
 366                 if (err) {
 367                         skb_unlink(skb, &vi->recv);
 368                         kfree_skb(skb);
 369                         break;
 370                 }
 371                 vi->num++;
 372         }
 373         if (unlikely(vi->num > vi->max))
 374                 vi->max = vi->num;
 375         vi->rvq->vq_ops->kick(vi->rvq);
 376 }
 377
 378 static void skb_recv_done(struct virtqueue *rvq)
 379 {
 380         struct virtnet_info *vi = rvq->vdev->priv;
 381         /* Schedule NAPI, Suppress further interrupts if successful. */
 382         if (napi_schedule_prep(&vi->napi)) {
 383                 rvq->vq_ops->disable_cb(rvq);
 384                 __napi_schedule(&vi->napi);
 385         }
 386 }
 387
 388 static int virtnet_poll(struct napi_struct *napi, int budget)
 389 {
 390         struct virtnet_info *vi = container_of(napi, struct virtnet_info, napi);
 391         struct sk_buff *skb = NULL;
 392         unsigned int len, received = 0;
 393
 394 again:
 395         while (received < budget &&
 396                (skb = vi->rvq->vq_ops->get_buf(vi->rvq, &len)) != NULL) {
 397                 __skb_unlink(skb, &vi->recv);
 398                 receive_skb(vi->dev, skb, len);
 399                 vi->num--;
 400                 received++;
 401         }
 402
 403         /* FIXME: If we oom and completely run out of inbufs, we need
 404          * to start a timer trying to fill more. */
 405         if (vi->num < vi->max / 2)
 406                 try_fill_recv(vi);
 407
 408         /* Out of packets? */
 409         if (received < budget) {
 410                 napi_complete(napi);
 411                 if (unlikely(!vi->rvq->vq_ops->enable_cb(vi->rvq))
 412                     && napi_schedule_prep(napi)) {
 413                         vi->rvq->vq_ops->disable_cb(vi->rvq);
 414                         __napi_schedule(napi);
 415                         goto again;
 416                 }
 417         }
 418
 419         return received;
 420 }
 421
 422 static void free_old_xmit_skbs(struct virtnet_info *vi)
 423 {
 424         struct sk_buff *skb;
 425         unsigned int len;
 426
 427         while ((skb = vi->svq->vq_ops->get_buf(vi->svq, &len)) != NULL) {
 428                 pr_debug("Sent skb %p\n", skb);
 429                 __skb_unlink(skb, &vi->send);
 430                 vi->dev->stats.tx_bytes += skb->len;
 431                 vi->dev->stats.tx_packets++;
 432                 kfree_skb(skb);
 433         }
 434 }
 435
 436 /* If the virtio transport doesn't always notify us when all in-flight packets
 437  * are consumed, we fall back to using this function on a timer to free them. */
 438 static void xmit_free(unsigned long data)
 439 {
 440         struct virtnet_info *vi = (void *)data;
 441
 442         netif_tx_lock(vi->dev);
 443
 444         free_old_xmit_skbs(vi);
 445
 446         if (!skb_queue_empty(&vi->send))
 447                 mod_timer(&vi->xmit_free_timer, jiffies + (HZ/10));
 448
 449         netif_tx_unlock(vi->dev);
 450 }
 451
 452 static int xmit_skb(struct virtnet_info *vi, struct sk_buff *skb)
 453 {
 454         int num, err;
 455         struct scatterlist sg[2+MAX_SKB_FRAGS];
 456         struct virtio_net_hdr_mrg_rxbuf *mhdr = skb_vnet_hdr(skb);
 457         struct virtio_net_hdr *hdr = skb_vnet_hdr(skb);
 458         const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
 459
 460         sg_init_table(sg, 2+MAX_SKB_FRAGS);
 461
 462         pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
 463
 464         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 465                 hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
 466                 hdr->csum_start = skb->csum_start - skb_headroom(skb);
 467                 hdr->csum_offset = skb->csum_offset;
 468         } else {
 469                 hdr->flags = 0;
 470                 hdr->csum_offset = hdr->csum_start = 0;
 471         }
 472
 473         if (skb_is_gso(skb)) {
 474                 hdr->hdr_len = skb_headlen(skb);
 475                 hdr->gso_size = skb_shinfo(skb)->gso_size;
 476                 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
 477                         hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
 478                 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
 479                         hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
 480                 else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 481                         hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
 482                 else
 483                         BUG();
 484                 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
 485                         hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
 486         } else {
 487                 hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
 488                 hdr->gso_size = hdr->hdr_len = 0;
 489         }
 490
 491         mhdr->num_buffers = 0;
 492
 493         /* Encode metadata header at front. */
 494         if (vi->mergeable_rx_bufs)
 495                 sg_set_buf(sg, mhdr, sizeof(*mhdr));
 496         else
 497                 sg_set_buf(sg, hdr, sizeof(*hdr));
 498
 499         num = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
 500
 501         err = vi->svq->vq_ops->add_buf(vi->svq, sg, num, 0, skb);
 502         if (!err && !vi->free_in_tasklet)
 503                 mod_timer(&vi->xmit_free_timer, jiffies + (HZ/10));
 504
 505         return err;
 506 }
 507
 508 static void xmit_tasklet(unsigned long data)
 509 {
 510         struct virtnet_info *vi = (void *)data;
 511
 512         netif_tx_lock_bh(vi->dev);
 513         if (vi->last_xmit_skb && xmit_skb(vi, vi->last_xmit_skb) == 0) {
 514                 vi->svq->vq_ops->kick(vi->svq);
 515                 vi->last_xmit_skb = NULL;
 516         }
 517         if (vi->free_in_tasklet)
 518                 free_old_xmit_skbs(vi);
 519         netif_tx_unlock_bh(vi->dev);
 520 }
 521
 522 static int start_xmit(struct sk_buff *skb, struct net_device *dev)
 523 {
 524         struct virtnet_info *vi = netdev_priv(dev);
 525
 526 again:
 527         /* Free up any pending old buffers before queueing new ones. */
 528         free_old_xmit_skbs(vi);
 529
 530         /* If we has a buffer left over from last time, send it now. */
 531         if (unlikely(vi->last_xmit_skb) &&
 532             xmit_skb(vi, vi->last_xmit_skb) != 0)
 533                 goto stop_queue;
 534
 535         vi->last_xmit_skb = NULL;
 536
 537         /* Put new one in send queue and do transmit */
 538         if (likely(skb)) {
 539                 __skb_queue_head(&vi->send, skb);
 540                 if (xmit_skb(vi, skb) != 0) {
 541                         vi->last_xmit_skb = skb;
 542                         skb = NULL;
 543                         goto stop_queue;
 544                 }
 545         }
 546 done:
 547         vi->svq->vq_ops->kick(vi->svq);
 548         return NETDEV_TX_OK;
 549
 550 stop_queue:
 551         pr_debug("%s: virtio not prepared to send\n", dev->name);
 552         netif_stop_queue(dev);
 553
 554         /* Activate callback for using skbs: if this returns false it
 555          * means some were used in the meantime. */
 556         if (unlikely(!vi->svq->vq_ops->enable_cb(vi->svq))) {
 557                 vi->svq->vq_ops->disable_cb(vi->svq);
 558                 netif_start_queue(dev);
 559                 goto again;
 560         }
 561         if (skb) {
 562                 /* Drop this skb: we only queue one. */
 563                 vi->dev->stats.tx_dropped++;
 564                 kfree_skb(skb);
 565         }
 566         goto done;
 567 }
 568
 569 static int virtnet_set_mac_address(struct net_device *dev, void *p)
 570 {
 571         struct virtnet_info *vi = netdev_priv(dev);
 572         struct virtio_device *vdev = vi->vdev;
 573         int ret;
 574
 575         ret = eth_mac_addr(dev, p);
 576         if (ret)
 577                 return ret;
 578
 579         if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC))
 580                 vdev->config->set(vdev, offsetof(struct virtio_net_config, mac),
 581                                   dev->dev_addr, dev->addr_len);
 582
 583         return 0;
 584 }
 585
 586 #ifdef CONFIG_NET_POLL_CONTROLLER
 587 static void virtnet_netpoll(struct net_device *dev)
 588 {
 589         struct virtnet_info *vi = netdev_priv(dev);
 590
 591         napi_schedule(&vi->napi);
 592 }
 593 #endif
 594
 595 static int virtnet_open(struct net_device *dev)
 596 {
 597         struct virtnet_info *vi = netdev_priv(dev);
 598
 599         napi_enable(&vi->napi);
 600
 601         /* If all buffers were filled by other side before we napi_enabled, we
 602          * won't get another interrupt, so process any outstanding packets
 603          * now.  virtnet_poll wants re-enable the queue, so we disable here.
 604          * We synchronize against interrupts via NAPI_STATE_SCHED */
 605         if (napi_schedule_prep(&vi->napi)) {
 606                 vi->rvq->vq_ops->disable_cb(vi->rvq);
 607                 __napi_schedule(&vi->napi);
 608         }
 609         return 0;
 610 }
 611
 612 /*
 613  * Send command via the control virtqueue and check status.  Commands
 614  * supported by the hypervisor, as indicated by feature bits, should
 615  * never fail unless improperly formated.
 616  */
 617 static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
 618                                  struct scatterlist *data, int out, int in)
 619 {
 620         struct scatterlist *s, sg[VIRTNET_SEND_COMMAND_SG_MAX + 2];
 621         struct virtio_net_ctrl_hdr ctrl;
 622         virtio_net_ctrl_ack status = ~0;
 623         unsigned int tmp;
 624         int i;
 625
 626         /* Caller should know better */
 627         BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ||
 628                 (out + in > VIRTNET_SEND_COMMAND_SG_MAX));
 629
 630         out++; /* Add header */
 631         in++; /* Add return status */
 632
 633         ctrl.class = class;
 634         ctrl.cmd = cmd;
 635
 636         sg_init_table(sg, out + in);
 637
 638         sg_set_buf(&sg[0], &ctrl, sizeof(ctrl));
 639         for_each_sg(data, s, out + in - 2, i)
 640                 sg_set_buf(&sg[i + 1], sg_virt(s), s->length);
 641         sg_set_buf(&sg[out + in - 1], &status, sizeof(status));
 642
 643         BUG_ON(vi->cvq->vq_ops->add_buf(vi->cvq, sg, out, in, vi));
 644
 645         vi->cvq->vq_ops->kick(vi->cvq);
 646
 647         /*
 648          * Spin for a response, the kick causes an ioport write, trapping
 649          * into the hypervisor, so the request should be handled immediately.
 650          */
 651         while (!vi->cvq->vq_ops->get_buf(vi->cvq, &tmp))
 652                 cpu_relax();
 653
 654         return status == VIRTIO_NET_OK;
 655 }
 656
 657 static int virtnet_close(struct net_device *dev)
 658 {
 659         struct virtnet_info *vi = netdev_priv(dev);
 660
 661         napi_disable(&vi->napi);
 662
 663         return 0;
 664 }
 665
 666 static int virtnet_set_tx_csum(struct net_device *dev, u32 data)
 667 {
 668         struct virtnet_info *vi = netdev_priv(dev);
 669         struct virtio_device *vdev = vi->vdev;
 670
 671         if (data && !virtio_has_feature(vdev, VIRTIO_NET_F_CSUM))
 672                 return -ENOSYS;
 673
 674         return ethtool_op_set_tx_hw_csum(dev, data);
 675 }
 676
 677 static void virtnet_set_rx_mode(struct net_device *dev)
 678 {
 679         struct virtnet_info *vi = netdev_priv(dev);
 680         struct scatterlist sg[2];
 681         u8 promisc, allmulti;
 682         struct virtio_net_ctrl_mac *mac_data;
 683         struct dev_addr_list *addr;
 684         struct netdev_hw_addr *ha;
 685         void *buf;
 686         int i;
 687
 688         /* We can't dynamicaly set ndo_set_rx_mode, so return gracefully */
 689         if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
 690                 return;
 691
 692         promisc = ((dev->flags & IFF_PROMISC) != 0);
 693         allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
 694
 695         sg_init_one(sg, &promisc, sizeof(promisc));
 696
 697         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
 698                                   VIRTIO_NET_CTRL_RX_PROMISC,
 699                                   sg, 1, 0))
 700                 dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
 701                          promisc ? "en" : "dis");
 702
 703         sg_init_one(sg, &allmulti, sizeof(allmulti));
 704
 705         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
 706                                   VIRTIO_NET_CTRL_RX_ALLMULTI,
 707                                   sg, 1, 0))
 708                 dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
 709                          allmulti ? "en" : "dis");
 710
 711         /* MAC filter - use one buffer for both lists */
 712         mac_data = buf = kzalloc(((dev->uc_count + dev->mc_count) * ETH_ALEN) +
 713                                  (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
 714         if (!buf) {
 715                 dev_warn(&dev->dev, "No memory for MAC address buffer\n");
 716                 return;
 717         }
 718
 719         sg_init_table(sg, 2);
 720
 721         /* Store the unicast list and count in the front of the buffer */
 722         mac_data->entries = dev->uc_count;
 723         i = 0;
 724         list_for_each_entry(ha, &dev->uc_list, list)
 725                 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
 726
 727         sg_set_buf(&sg[0], mac_data,
 728                    sizeof(mac_data->entries) + (dev->uc_count * ETH_ALEN));
 729
 730         /* multicast list and count fill the end */
 731         mac_data = (void *)&mac_data->macs[dev->uc_count][0];
 732
 733         mac_data->entries = dev->mc_count;
 734         addr = dev->mc_list;
 735         for (i = 0; i < dev->mc_count; i++, addr = addr->next)
 736                 memcpy(&mac_data->macs[i][0], addr->da_addr, ETH_ALEN);
 737
 738         sg_set_buf(&sg[1], mac_data,
 739                    sizeof(mac_data->entries) + (dev->mc_count * ETH_ALEN));
 740
 741         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
 742                                   VIRTIO_NET_CTRL_MAC_TABLE_SET,
 743                                   sg, 2, 0))
 744                 dev_warn(&dev->dev, "Failed to set MAC fitler table.\n");
 745
 746         kfree(buf);
 747 }
 748
 749 static void virtnet_vlan_rx_add_vid(struct net_device *dev, u16 vid)
 750 {
 751         struct virtnet_info *vi = netdev_priv(dev);
 752         struct scatterlist sg;
 753
 754         sg_init_one(&sg, &vid, sizeof(vid));
 755
 756         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
 757                                   VIRTIO_NET_CTRL_VLAN_ADD, &sg, 1, 0))
 758                 dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
 759 }
 760
 761 static void virtnet_vlan_rx_kill_vid(struct net_device *dev, u16 vid)
 762 {
 763         struct virtnet_info *vi = netdev_priv(dev);
 764         struct scatterlist sg;
 765
 766         sg_init_one(&sg, &vid, sizeof(vid));
 767
 768         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
 769                                   VIRTIO_NET_CTRL_VLAN_DEL, &sg, 1, 0))
 770                 dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
 771 }
 772
 773 static struct ethtool_ops virtnet_ethtool_ops = {
 774         .set_tx_csum = virtnet_set_tx_csum,
 775         .set_sg = ethtool_op_set_sg,
 776         .set_tso = ethtool_op_set_tso,
 777         .get_link = ethtool_op_get_link,
 778 };
 779
 780 #define MIN_MTU 68
 781 #define MAX_MTU 65535
 782
 783 static int virtnet_change_mtu(struct net_device *dev, int new_mtu)
 784 {
 785         if (new_mtu < MIN_MTU || new_mtu > MAX_MTU)
 786                 return -EINVAL;
 787         dev->mtu = new_mtu;
 788         return 0;
 789 }
 790
 791 static const struct net_device_ops virtnet_netdev = {
 792         .ndo_open            = virtnet_open,
 793         .ndo_stop            = virtnet_close,
 794         .ndo_start_xmit      = start_xmit,
 795         .ndo_validate_addr   = eth_validate_addr,
 796         .ndo_set_mac_address = virtnet_set_mac_address,
 797         .ndo_set_rx_mode     = virtnet_set_rx_mode,
 798         .ndo_change_mtu      = virtnet_change_mtu,
 799         .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
 800         .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
 801 #ifdef CONFIG_NET_POLL_CONTROLLER
 802         .ndo_poll_controller = virtnet_netpoll,
 803 #endif
 804 };
 805
 806 static void virtnet_update_status(struct virtnet_info *vi)
 807 {
 808         u16 v;
 809
 810         if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS))
 811                 return;
 812
 813         vi->vdev->config->get(vi->vdev,
 814                               offsetof(struct virtio_net_config, status),
 815                               &v, sizeof(v));
 816
 817         /* Ignore unknown (future) status bits */
 818         v &= VIRTIO_NET_S_LINK_UP;
 819
 820         if (vi->status == v)
 821                 return;
 822
 823         vi->status = v;
 824
 825         if (vi->status & VIRTIO_NET_S_LINK_UP) {
 826                 netif_carrier_on(vi->dev);
 827                 netif_wake_queue(vi->dev);
 828         } else {
 829                 netif_carrier_off(vi->dev);
 830                 netif_stop_queue(vi->dev);
 831         }
 832 }
 833
 834 static void virtnet_config_changed(struct virtio_device *vdev)
 835 {
 836         struct virtnet_info *vi = vdev->priv;
 837
 838         virtnet_update_status(vi);
 839 }
 840
 841 static int virtnet_probe(struct virtio_device *vdev)
 842 {
 843         int err;
 844         struct net_device *dev;
 845         struct virtnet_info *vi;
 846         struct virtqueue *vqs[3];
 847         vq_callback_t *callbacks[] = { skb_recv_done, skb_xmit_done, NULL};
 848         const char *names[] = { "input", "output", "control" };
 849         int nvqs;
 850
 851         /* Allocate ourselves a network device with room for our info */
 852         dev = alloc_etherdev(sizeof(struct virtnet_info));
 853         if (!dev)
 854                 return -ENOMEM;
 855
 856         /* Set up network device as normal. */
 857         dev->netdev_ops = &virtnet_netdev;
 858         dev->features = NETIF_F_HIGHDMA;
 859         SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops);
 860         SET_NETDEV_DEV(dev, &vdev->dev);
 861
 862         /* Do we support "hardware" checksums? */
 863         if (csum && virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
 864                 /* This opens up the world of extra features. */
 865                 dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
 866                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
 867                         dev->features |= NETIF_F_TSO | NETIF_F_UFO
 868                                 | NETIF_F_TSO_ECN | NETIF_F_TSO6;
 869                 }
 870                 /* Individual feature bits: what can host handle? */
 871                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
 872                         dev->features |= NETIF_F_TSO;
 873                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
 874                         dev->features |= NETIF_F_TSO6;
 875                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
 876                         dev->features |= NETIF_F_TSO_ECN;
 877                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
 878                         dev->features |= NETIF_F_UFO;
 879         }
 880
 881         /* Configuration may specify what MAC to use.  Otherwise random. */
 882         if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) {
 883                 vdev->config->get(vdev,
 884                                   offsetof(struct virtio_net_config, mac),
 885                                   dev->dev_addr, dev->addr_len);
 886         } else
 887                 random_ether_addr(dev->dev_addr);
 888
 889         /* Set up our device-specific information */
 890         vi = netdev_priv(dev);
 891         netif_napi_add(dev, &vi->napi, virtnet_poll, napi_weight);
 892         vi->dev = dev;
 893         vi->vdev = vdev;
 894         vdev->priv = vi;
 895         vi->pages = NULL;
 896
 897         /* If they give us a callback when all buffers are done, we don't need
 898          * the timer. */
 899         vi->free_in_tasklet = virtio_has_feature(vdev,VIRTIO_F_NOTIFY_ON_EMPTY);
 900
 901         /* If we can receive ANY GSO packets, we must allocate large ones. */
 902         if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4)
 903             || virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6)
 904             || virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN))
 905                 vi->big_packets = true;
 906
 907         if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
 908                 vi->mergeable_rx_bufs = true;
 909
 910         /* We expect two virtqueues, receive then send,
 911          * and optionally control. */
 912         nvqs = virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ? 3 : 2;
 913
 914         err = vdev->config->find_vqs(vdev, nvqs, vqs, callbacks, names);
 915         if (err)
 916                 goto free;
 917
 918         vi->rvq = vqs[0];
 919         vi->svq = vqs[1];
 920
 921         if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)) {
 922                 vi->cvq = vqs[2];
 923
 924                 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
 925                         dev->features |= NETIF_F_HW_VLAN_FILTER;
 926         }
 927
 928         /* Initialize our empty receive and send queues. */
 929         skb_queue_head_init(&vi->recv);
 930         skb_queue_head_init(&vi->send);
 931
 932         tasklet_init(&vi->tasklet, xmit_tasklet, (unsigned long)vi);
 933
 934         if (!vi->free_in_tasklet)
 935                 setup_timer(&vi->xmit_free_timer, xmit_free, (unsigned long)vi);
 936
 937         err = register_netdev(dev);
 938         if (err) {
 939                 pr_debug("virtio_net: registering device failed\n");
 940                 goto free_vqs;
 941         }
 942
 943         /* Last of all, set up some receive buffers. */
 944         try_fill_recv(vi);
 945
 946         /* If we didn't even get one input buffer, we're useless. */
 947         if (vi->num == 0) {
 948                 err = -ENOMEM;
 949                 goto unregister;
 950         }
 951
 952         vi->status = VIRTIO_NET_S_LINK_UP;
 953         virtnet_update_status(vi);
 954         netif_carrier_on(dev);
 955
 956         pr_debug("virtnet: registered device %s\n", dev->name);
 957         return 0;
 958
 959 unregister:
 960         unregister_netdev(dev);
 961 free_vqs:
 962         vdev->config->del_vqs(vdev);
 963 free:
 964         free_netdev(dev);
 965         return err;
 966 }
 967
 968 static void virtnet_remove(struct virtio_device *vdev)
 969 {
 970         struct virtnet_info *vi = vdev->priv;
 971         struct sk_buff *skb;
 972
 973         /* Stop all the virtqueues. */
 974         vdev->config->reset(vdev);
 975
 976         if (!vi->free_in_tasklet)
 977                 del_timer_sync(&vi->xmit_free_timer);
 978
 979         /* Free our skbs in send and recv queues, if any. */
 980         while ((skb = __skb_dequeue(&vi->recv)) != NULL) {
 981                 kfree_skb(skb);
 982                 vi->num--;
 983         }
 984         __skb_queue_purge(&vi->send);
 985
 986         BUG_ON(vi->num != 0);
 987
 988         unregister_netdev(vi->dev);
 989
 990         vdev->config->del_vqs(vi->vdev);
 991
 992         while (vi->pages)
 993                 __free_pages(get_a_page(vi, GFP_KERNEL), 0);
 994
 995         free_netdev(vi->dev);
 996 }
 997
 998 static struct virtio_device_id id_table[] = {
 999         { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
1000         { 0 },
1001 };
1002
1003 static unsigned int features[] = {
1004         VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM,
1005         VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC,
1006         VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6,
1007         VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6,
1008         VIRTIO_NET_F_GUEST_ECN, /* We don't yet handle UFO input. */
1009         VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ,
1010         VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN,
1011         VIRTIO_F_NOTIFY_ON_EMPTY,
1012 };
1013
1014 static struct virtio_driver virtio_net = {
1015         .feature_table = features,
1016         .feature_table_size = ARRAY_SIZE(features),
1017         .driver.name =  KBUILD_MODNAME,
1018         .driver.owner = THIS_MODULE,
1019         .id_table =     id_table,
1020         .probe =        virtnet_probe,
1021         .remove =       __devexit_p(virtnet_remove),
1022         .config_changed = virtnet_config_changed,
1023 };
1024
1025 static int __init init(void)
1026 {
1027         return register_virtio_driver(&virtio_net);
1028 }
1029
1030 static void __exit fini(void)
1031 {
1032         unregister_virtio_driver(&virtio_net);
1033 }
1034 module_init(init);
1035 module_exit(fini);
1036
1037 MODULE_DEVICE_TABLE(virtio, id_table);
1038 MODULE_DESCRIPTION("Virtio network driver");
1039 MODULE_LICENSE("GPL");