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);
 287                 if (unlikely(!skb))
 288                         break;
 289
 290                 skb_put(skb, MAX_PACKET_LEN);
 291
 292                 hdr = skb_vnet_hdr(skb);
 293                 sg_set_buf(sg, hdr, sizeof(*hdr));
 294
 295                 if (vi->big_packets) {
 296                         for (i = 0; i < MAX_SKB_FRAGS; i++) {
 297                                 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
 298                                 f->page = get_a_page(vi, GFP_ATOMIC);
 299                                 if (!f->page)
 300                                         break;
 301
 302                                 f->page_offset = 0;
 303                                 f->size = PAGE_SIZE;
 304
 305                                 skb->data_len += PAGE_SIZE;
 306                                 skb->len += PAGE_SIZE;
 307
 308                                 skb_shinfo(skb)->nr_frags++;
 309                         }
 310                 }
 311
 312                 num = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
 313                 skb_queue_head(&vi->recv, skb);
 314
 315                 err = vi->rvq->vq_ops->add_buf(vi->rvq, sg, 0, num, skb);
 316                 if (err) {
 317                         skb_unlink(skb, &vi->recv);
 318                         trim_pages(vi, skb);
 319                         kfree_skb(skb);
 320                         break;
 321                 }
 322                 vi->num++;
 323         }
 324         if (unlikely(vi->num > vi->max))
 325                 vi->max = vi->num;
 326         vi->rvq->vq_ops->kick(vi->rvq);
 327 }
 328
 329 static void try_fill_recv(struct virtnet_info *vi)
 330 {
 331         struct sk_buff *skb;
 332         struct scatterlist sg[1];
 333         int err;
 334
 335         if (!vi->mergeable_rx_bufs) {
 336                 try_fill_recv_maxbufs(vi);
 337                 return;
 338         }
 339
 340         for (;;) {
 341                 skb_frag_t *f;
 342
 343                 skb = netdev_alloc_skb(vi->dev, GOOD_COPY_LEN + NET_IP_ALIGN);
 344                 if (unlikely(!skb))
 345                         break;
 346
 347                 skb_reserve(skb, NET_IP_ALIGN);
 348
 349                 f = &skb_shinfo(skb)->frags[0];
 350                 f->page = get_a_page(vi, GFP_ATOMIC);
 351                 if (!f->page) {
 352                         kfree_skb(skb);
 353                         break;
 354                 }
 355
 356                 f->page_offset = 0;
 357                 f->size = PAGE_SIZE;
 358
 359                 skb_shinfo(skb)->nr_frags++;
 360
 361                 sg_init_one(sg, page_address(f->page), PAGE_SIZE);
 362                 skb_queue_head(&vi->recv, skb);
 363
 364                 err = vi->rvq->vq_ops->add_buf(vi->rvq, sg, 0, 1, skb);
 365                 if (err) {
 366                         skb_unlink(skb, &vi->recv);
 367                         kfree_skb(skb);
 368                         break;
 369                 }
 370                 vi->num++;
 371         }
 372         if (unlikely(vi->num > vi->max))
 373                 vi->max = vi->num;
 374         vi->rvq->vq_ops->kick(vi->rvq);
 375 }
 376
 377 static void skb_recv_done(struct virtqueue *rvq)
 378 {
 379         struct virtnet_info *vi = rvq->vdev->priv;
 380         /* Schedule NAPI, Suppress further interrupts if successful. */
 381         if (napi_schedule_prep(&vi->napi)) {
 382                 rvq->vq_ops->disable_cb(rvq);
 383                 __napi_schedule(&vi->napi);
 384         }
 385 }
 386
 387 static int virtnet_poll(struct napi_struct *napi, int budget)
 388 {
 389         struct virtnet_info *vi = container_of(napi, struct virtnet_info, napi);
 390         struct sk_buff *skb = NULL;
 391         unsigned int len, received = 0;
 392
 393 again:
 394         while (received < budget &&
 395                (skb = vi->rvq->vq_ops->get_buf(vi->rvq, &len)) != NULL) {
 396                 __skb_unlink(skb, &vi->recv);
 397                 receive_skb(vi->dev, skb, len);
 398                 vi->num--;
 399                 received++;
 400         }
 401
 402         /* FIXME: If we oom and completely run out of inbufs, we need
 403          * to start a timer trying to fill more. */
 404         if (vi->num < vi->max / 2)
 405                 try_fill_recv(vi);
 406
 407         /* Out of packets? */
 408         if (received < budget) {
 409                 napi_complete(napi);
 410                 if (unlikely(!vi->rvq->vq_ops->enable_cb(vi->rvq))
 411                     && napi_schedule_prep(napi)) {
 412                         vi->rvq->vq_ops->disable_cb(vi->rvq);
 413                         __napi_schedule(napi);
 414                         goto again;
 415                 }
 416         }
 417
 418         return received;
 419 }
 420
 421 static void free_old_xmit_skbs(struct virtnet_info *vi)
 422 {
 423         struct sk_buff *skb;
 424         unsigned int len;
 425
 426         while ((skb = vi->svq->vq_ops->get_buf(vi->svq, &len)) != NULL) {
 427                 pr_debug("Sent skb %p\n", skb);
 428                 __skb_unlink(skb, &vi->send);
 429                 vi->dev->stats.tx_bytes += skb->len;
 430                 vi->dev->stats.tx_packets++;
 431                 kfree_skb(skb);
 432         }
 433 }
 434
 435 /* If the virtio transport doesn't always notify us when all in-flight packets
 436  * are consumed, we fall back to using this function on a timer to free them. */
 437 static void xmit_free(unsigned long data)
 438 {
 439         struct virtnet_info *vi = (void *)data;
 440
 441         netif_tx_lock(vi->dev);
 442
 443         free_old_xmit_skbs(vi);
 444
 445         if (!skb_queue_empty(&vi->send))
 446                 mod_timer(&vi->xmit_free_timer, jiffies + (HZ/10));
 447
 448         netif_tx_unlock(vi->dev);
 449 }
 450
 451 static int xmit_skb(struct virtnet_info *vi, struct sk_buff *skb)
 452 {
 453         int num, err;
 454         struct scatterlist sg[2+MAX_SKB_FRAGS];
 455         struct virtio_net_hdr_mrg_rxbuf *mhdr = skb_vnet_hdr(skb);
 456         struct virtio_net_hdr *hdr = skb_vnet_hdr(skb);
 457         const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
 458
 459         sg_init_table(sg, 2+MAX_SKB_FRAGS);
 460
 461         pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
 462
 463         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 464                 hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
 465                 hdr->csum_start = skb->csum_start - skb_headroom(skb);
 466                 hdr->csum_offset = skb->csum_offset;
 467         } else {
 468                 hdr->flags = 0;
 469                 hdr->csum_offset = hdr->csum_start = 0;
 470         }
 471
 472         if (skb_is_gso(skb)) {
 473                 hdr->hdr_len = skb_transport_header(skb) - skb->data;
 474                 hdr->gso_size = skb_shinfo(skb)->gso_size;
 475                 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
 476                         hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
 477                 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
 478                         hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
 479                 else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 480                         hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
 481                 else
 482                         BUG();
 483                 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
 484                         hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
 485         } else {
 486                 hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
 487                 hdr->gso_size = hdr->hdr_len = 0;
 488         }
 489
 490         mhdr->num_buffers = 0;
 491
 492         /* Encode metadata header at front. */
 493         if (vi->mergeable_rx_bufs)
 494                 sg_set_buf(sg, mhdr, sizeof(*mhdr));
 495         else
 496                 sg_set_buf(sg, hdr, sizeof(*hdr));
 497
 498         num = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
 499
 500         err = vi->svq->vq_ops->add_buf(vi->svq, sg, num, 0, skb);
 501         if (!err && !vi->free_in_tasklet)
 502                 mod_timer(&vi->xmit_free_timer, jiffies + (HZ/10));
 503
 504         return err;
 505 }
 506
 507 static void xmit_tasklet(unsigned long data)
 508 {
 509         struct virtnet_info *vi = (void *)data;
 510
 511         netif_tx_lock_bh(vi->dev);
 512         if (vi->last_xmit_skb && xmit_skb(vi, vi->last_xmit_skb) == 0) {
 513                 vi->svq->vq_ops->kick(vi->svq);
 514                 vi->last_xmit_skb = NULL;
 515         }
 516         if (vi->free_in_tasklet)
 517                 free_old_xmit_skbs(vi);
 518         netif_tx_unlock_bh(vi->dev);
 519 }
 520
 521 static int start_xmit(struct sk_buff *skb, struct net_device *dev)
 522 {
 523         struct virtnet_info *vi = netdev_priv(dev);
 524
 525 again:
 526         /* Free up any pending old buffers before queueing new ones. */
 527         free_old_xmit_skbs(vi);
 528
 529         /* If we has a buffer left over from last time, send it now. */
 530         if (unlikely(vi->last_xmit_skb) &&
 531             xmit_skb(vi, vi->last_xmit_skb) != 0)
 532                 goto stop_queue;
 533
 534         vi->last_xmit_skb = NULL;
 535
 536         /* Put new one in send queue and do transmit */
 537         if (likely(skb)) {
 538                 __skb_queue_head(&vi->send, skb);
 539                 if (xmit_skb(vi, skb) != 0) {
 540                         vi->last_xmit_skb = skb;
 541                         skb = NULL;
 542                         goto stop_queue;
 543                 }
 544         }
 545 done:
 546         vi->svq->vq_ops->kick(vi->svq);
 547         return NETDEV_TX_OK;
 548
 549 stop_queue:
 550         pr_debug("%s: virtio not prepared to send\n", dev->name);
 551         netif_stop_queue(dev);
 552
 553         /* Activate callback for using skbs: if this returns false it
 554          * means some were used in the meantime. */
 555         if (unlikely(!vi->svq->vq_ops->enable_cb(vi->svq))) {
 556                 vi->svq->vq_ops->disable_cb(vi->svq);
 557                 netif_start_queue(dev);
 558                 goto again;
 559         }
 560         if (skb) {
 561                 /* Drop this skb: we only queue one. */
 562                 vi->dev->stats.tx_dropped++;
 563                 kfree_skb(skb);
 564         }
 565         goto done;
 566 }
 567
 568 static int virtnet_set_mac_address(struct net_device *dev, void *p)
 569 {
 570         struct virtnet_info *vi = netdev_priv(dev);
 571         struct virtio_device *vdev = vi->vdev;
 572         int ret;
 573
 574         ret = eth_mac_addr(dev, p);
 575         if (ret)
 576                 return ret;
 577
 578         if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC))
 579                 vdev->config->set(vdev, offsetof(struct virtio_net_config, mac),
 580                                   dev->dev_addr, dev->addr_len);
 581
 582         return 0;
 583 }
 584
 585 #ifdef CONFIG_NET_POLL_CONTROLLER
 586 static void virtnet_netpoll(struct net_device *dev)
 587 {
 588         struct virtnet_info *vi = netdev_priv(dev);
 589
 590         napi_schedule(&vi->napi);
 591 }
 592 #endif
 593
 594 static int virtnet_open(struct net_device *dev)
 595 {
 596         struct virtnet_info *vi = netdev_priv(dev);
 597
 598         napi_enable(&vi->napi);
 599
 600         /* If all buffers were filled by other side before we napi_enabled, we
 601          * won't get another interrupt, so process any outstanding packets
 602          * now.  virtnet_poll wants re-enable the queue, so we disable here.
 603          * We synchronize against interrupts via NAPI_STATE_SCHED */
 604         if (napi_schedule_prep(&vi->napi)) {
 605                 vi->rvq->vq_ops->disable_cb(vi->rvq);
 606                 __napi_schedule(&vi->napi);
 607         }
 608         return 0;
 609 }
 610
 611 /*
 612  * Send command via the control virtqueue and check status.  Commands
 613  * supported by the hypervisor, as indicated by feature bits, should
 614  * never fail unless improperly formated.
 615  */
 616 static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
 617                                  struct scatterlist *data, int out, int in)
 618 {
 619         struct scatterlist *s, sg[VIRTNET_SEND_COMMAND_SG_MAX + 2];
 620         struct virtio_net_ctrl_hdr ctrl;
 621         virtio_net_ctrl_ack status = ~0;
 622         unsigned int tmp;
 623         int i;
 624
 625         /* Caller should know better */
 626         BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ||
 627                 (out + in > VIRTNET_SEND_COMMAND_SG_MAX));
 628
 629         out++; /* Add header */
 630         in++; /* Add return status */
 631
 632         ctrl.class = class;
 633         ctrl.cmd = cmd;
 634
 635         sg_init_table(sg, out + in);
 636
 637         sg_set_buf(&sg[0], &ctrl, sizeof(ctrl));
 638         for_each_sg(data, s, out + in - 2, i)
 639                 sg_set_buf(&sg[i + 1], sg_virt(s), s->length);
 640         sg_set_buf(&sg[out + in - 1], &status, sizeof(status));
 641
 642         BUG_ON(vi->cvq->vq_ops->add_buf(vi->cvq, sg, out, in, vi));
 643
 644         vi->cvq->vq_ops->kick(vi->cvq);
 645
 646         /*
 647          * Spin for a response, the kick causes an ioport write, trapping
 648          * into the hypervisor, so the request should be handled immediately.
 649          */
 650         while (!vi->cvq->vq_ops->get_buf(vi->cvq, &tmp))
 651                 cpu_relax();
 652
 653         return status == VIRTIO_NET_OK;
 654 }
 655
 656 static int virtnet_close(struct net_device *dev)
 657 {
 658         struct virtnet_info *vi = netdev_priv(dev);
 659
 660         napi_disable(&vi->napi);
 661
 662         return 0;
 663 }
 664
 665 static int virtnet_set_tx_csum(struct net_device *dev, u32 data)
 666 {
 667         struct virtnet_info *vi = netdev_priv(dev);
 668         struct virtio_device *vdev = vi->vdev;
 669
 670         if (data && !virtio_has_feature(vdev, VIRTIO_NET_F_CSUM))
 671                 return -ENOSYS;
 672
 673         return ethtool_op_set_tx_hw_csum(dev, data);
 674 }
 675
 676 static void virtnet_set_rx_mode(struct net_device *dev)
 677 {
 678         struct virtnet_info *vi = netdev_priv(dev);
 679         struct scatterlist sg[2];
 680         u8 promisc, allmulti;
 681         struct virtio_net_ctrl_mac *mac_data;
 682         struct dev_addr_list *addr;
 683         struct netdev_hw_addr *ha;
 684         void *buf;
 685         int i;
 686
 687         /* We can't dynamicaly set ndo_set_rx_mode, so return gracefully */
 688         if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
 689                 return;
 690
 691         promisc = ((dev->flags & IFF_PROMISC) != 0);
 692         allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
 693
 694         sg_init_one(sg, &promisc, sizeof(promisc));
 695
 696         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
 697                                   VIRTIO_NET_CTRL_RX_PROMISC,
 698                                   sg, 1, 0))
 699                 dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
 700                          promisc ? "en" : "dis");
 701
 702         sg_init_one(sg, &allmulti, sizeof(allmulti));
 703
 704         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
 705                                   VIRTIO_NET_CTRL_RX_ALLMULTI,
 706                                   sg, 1, 0))
 707                 dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
 708                          allmulti ? "en" : "dis");
 709
 710         /* MAC filter - use one buffer for both lists */
 711         mac_data = buf = kzalloc(((dev->uc_count + dev->mc_count) * ETH_ALEN) +
 712                                  (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
 713         if (!buf) {
 714                 dev_warn(&dev->dev, "No memory for MAC address buffer\n");
 715                 return;
 716         }
 717
 718         sg_init_table(sg, 2);
 719
 720         /* Store the unicast list and count in the front of the buffer */
 721         mac_data->entries = dev->uc_count;
 722         i = 0;
 723         list_for_each_entry(ha, &dev->uc_list, list)
 724                 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
 725
 726         sg_set_buf(&sg[0], mac_data,
 727                    sizeof(mac_data->entries) + (dev->uc_count * ETH_ALEN));
 728
 729         /* multicast list and count fill the end */
 730         mac_data = (void *)&mac_data->macs[dev->uc_count][0];
 731
 732         mac_data->entries = dev->mc_count;
 733         addr = dev->mc_list;
 734         for (i = 0; i < dev->mc_count; i++, addr = addr->next)
 735                 memcpy(&mac_data->macs[i][0], addr->da_addr, ETH_ALEN);
 736
 737         sg_set_buf(&sg[1], mac_data,
 738                    sizeof(mac_data->entries) + (dev->mc_count * ETH_ALEN));
 739
 740         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
 741                                   VIRTIO_NET_CTRL_MAC_TABLE_SET,
 742                                   sg, 2, 0))
 743                 dev_warn(&dev->dev, "Failed to set MAC fitler table.\n");
 744
 745         kfree(buf);
 746 }
 747
 748 static void virtnet_vlan_rx_add_vid(struct net_device *dev, u16 vid)
 749 {
 750         struct virtnet_info *vi = netdev_priv(dev);
 751         struct scatterlist sg;
 752
 753         sg_init_one(&sg, &vid, sizeof(vid));
 754
 755         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
 756                                   VIRTIO_NET_CTRL_VLAN_ADD, &sg, 1, 0))
 757                 dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
 758 }
 759
 760 static void virtnet_vlan_rx_kill_vid(struct net_device *dev, u16 vid)
 761 {
 762         struct virtnet_info *vi = netdev_priv(dev);
 763         struct scatterlist sg;
 764
 765         sg_init_one(&sg, &vid, sizeof(vid));
 766
 767         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
 768                                   VIRTIO_NET_CTRL_VLAN_DEL, &sg, 1, 0))
 769                 dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
 770 }
 771
 772 static struct ethtool_ops virtnet_ethtool_ops = {
 773         .set_tx_csum = virtnet_set_tx_csum,
 774         .set_sg = ethtool_op_set_sg,
 775         .set_tso = ethtool_op_set_tso,
 776         .get_link = ethtool_op_get_link,
 777 };
 778
 779 #define MIN_MTU 68
 780 #define MAX_MTU 65535
 781
 782 static int virtnet_change_mtu(struct net_device *dev, int new_mtu)
 783 {
 784         if (new_mtu < MIN_MTU || new_mtu > MAX_MTU)
 785                 return -EINVAL;
 786         dev->mtu = new_mtu;
 787         return 0;
 788 }
 789
 790 static const struct net_device_ops virtnet_netdev = {
 791         .ndo_open            = virtnet_open,
 792         .ndo_stop            = virtnet_close,
 793         .ndo_start_xmit      = start_xmit,
 794         .ndo_validate_addr   = eth_validate_addr,
 795         .ndo_set_mac_address = virtnet_set_mac_address,
 796         .ndo_set_rx_mode     = virtnet_set_rx_mode,
 797         .ndo_change_mtu      = virtnet_change_mtu,
 798         .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
 799         .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
 800 #ifdef CONFIG_NET_POLL_CONTROLLER
 801         .ndo_poll_controller = virtnet_netpoll,
 802 #endif
 803 };
 804
 805 static void virtnet_update_status(struct virtnet_info *vi)
 806 {
 807         u16 v;
 808
 809         if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS))
 810                 return;
 811
 812         vi->vdev->config->get(vi->vdev,
 813                               offsetof(struct virtio_net_config, status),
 814                               &v, sizeof(v));
 815
 816         /* Ignore unknown (future) status bits */
 817         v &= VIRTIO_NET_S_LINK_UP;
 818
 819         if (vi->status == v)
 820                 return;
 821
 822         vi->status = v;
 823
 824         if (vi->status & VIRTIO_NET_S_LINK_UP) {
 825                 netif_carrier_on(vi->dev);
 826                 netif_wake_queue(vi->dev);
 827         } else {
 828                 netif_carrier_off(vi->dev);
 829                 netif_stop_queue(vi->dev);
 830         }
 831 }
 832
 833 static void virtnet_config_changed(struct virtio_device *vdev)
 834 {
 835         struct virtnet_info *vi = vdev->priv;
 836
 837         virtnet_update_status(vi);
 838 }
 839
 840 static int virtnet_probe(struct virtio_device *vdev)
 841 {
 842         int err;
 843         struct net_device *dev;
 844         struct virtnet_info *vi;
 845
 846         /* Allocate ourselves a network device with room for our info */
 847         dev = alloc_etherdev(sizeof(struct virtnet_info));
 848         if (!dev)
 849                 return -ENOMEM;
 850
 851         /* Set up network device as normal. */
 852         dev->netdev_ops = &virtnet_netdev;
 853         dev->features = NETIF_F_HIGHDMA;
 854         SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops);
 855         SET_NETDEV_DEV(dev, &vdev->dev);
 856
 857         /* Do we support "hardware" checksums? */
 858         if (csum && virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
 859                 /* This opens up the world of extra features. */
 860                 dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
 861                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
 862                         dev->features |= NETIF_F_TSO | NETIF_F_UFO
 863                                 | NETIF_F_TSO_ECN | NETIF_F_TSO6;
 864                 }
 865                 /* Individual feature bits: what can host handle? */
 866                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
 867                         dev->features |= NETIF_F_TSO;
 868                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
 869                         dev->features |= NETIF_F_TSO6;
 870                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
 871                         dev->features |= NETIF_F_TSO_ECN;
 872                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
 873                         dev->features |= NETIF_F_UFO;
 874         }
 875
 876         /* Configuration may specify what MAC to use.  Otherwise random. */
 877         if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) {
 878                 vdev->config->get(vdev,
 879                                   offsetof(struct virtio_net_config, mac),
 880                                   dev->dev_addr, dev->addr_len);
 881         } else
 882                 random_ether_addr(dev->dev_addr);
 883
 884         /* Set up our device-specific information */
 885         vi = netdev_priv(dev);
 886         netif_napi_add(dev, &vi->napi, virtnet_poll, napi_weight);
 887         vi->dev = dev;
 888         vi->vdev = vdev;
 889         vdev->priv = vi;
 890         vi->pages = NULL;
 891
 892         /* If they give us a callback when all buffers are done, we don't need
 893          * the timer. */
 894         vi->free_in_tasklet = virtio_has_feature(vdev,VIRTIO_F_NOTIFY_ON_EMPTY);
 895
 896         /* If we can receive ANY GSO packets, we must allocate large ones. */
 897         if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4)
 898             || virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6)
 899             || virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN))
 900                 vi->big_packets = true;
 901
 902         if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
 903                 vi->mergeable_rx_bufs = true;
 904
 905         /* We expect two virtqueues, receive then send. */
 906         vi->rvq = vdev->config->find_vq(vdev, 0, skb_recv_done);
 907         if (IS_ERR(vi->rvq)) {
 908                 err = PTR_ERR(vi->rvq);
 909                 goto free;
 910         }
 911
 912         vi->svq = vdev->config->find_vq(vdev, 1, skb_xmit_done);
 913         if (IS_ERR(vi->svq)) {
 914                 err = PTR_ERR(vi->svq);
 915                 goto free_recv;
 916         }
 917
 918         if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)) {
 919                 vi->cvq = vdev->config->find_vq(vdev, 2, NULL);
 920                 if (IS_ERR(vi->cvq)) {
 921                         err = PTR_ERR(vi->svq);
 922                         goto free_send;
 923                 }
 924
 925                 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
 926                         dev->features |= NETIF_F_HW_VLAN_FILTER;
 927         }
 928
 929         /* Initialize our empty receive and send queues. */
 930         skb_queue_head_init(&vi->recv);
 931         skb_queue_head_init(&vi->send);
 932
 933         tasklet_init(&vi->tasklet, xmit_tasklet, (unsigned long)vi);
 934
 935         if (!vi->free_in_tasklet)
 936                 setup_timer(&vi->xmit_free_timer, xmit_free, (unsigned long)vi);
 937
 938         err = register_netdev(dev);
 939         if (err) {
 940                 pr_debug("virtio_net: registering device failed\n");
 941                 goto free_ctrl;
 942         }
 943
 944         /* Last of all, set up some receive buffers. */
 945         try_fill_recv(vi);
 946
 947         /* If we didn't even get one input buffer, we're useless. */
 948         if (vi->num == 0) {
 949                 err = -ENOMEM;
 950                 goto unregister;
 951         }
 952
 953         vi->status = VIRTIO_NET_S_LINK_UP;
 954         virtnet_update_status(vi);
 955         netif_carrier_on(dev);
 956
 957         pr_debug("virtnet: registered device %s\n", dev->name);
 958         return 0;
 959
 960 unregister:
 961         unregister_netdev(dev);
 962 free_ctrl:
 963         if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ))
 964                 vdev->config->del_vq(vi->cvq);
 965 free_send:
 966         vdev->config->del_vq(vi->svq);
 967 free_recv:
 968         vdev->config->del_vq(vi->rvq);
 969 free:
 970         free_netdev(dev);
 971         return err;
 972 }
 973
 974 static void virtnet_remove(struct virtio_device *vdev)
 975 {
 976         struct virtnet_info *vi = vdev->priv;
 977         struct sk_buff *skb;
 978
 979         /* Stop all the virtqueues. */
 980         vdev->config->reset(vdev);
 981
 982         if (!vi->free_in_tasklet)
 983                 del_timer_sync(&vi->xmit_free_timer);
 984
 985         /* Free our skbs in send and recv queues, if any. */
 986         while ((skb = __skb_dequeue(&vi->recv)) != NULL) {
 987                 kfree_skb(skb);
 988                 vi->num--;
 989         }
 990         __skb_queue_purge(&vi->send);
 991
 992         BUG_ON(vi->num != 0);
 993
 994         vdev->config->del_vq(vi->svq);
 995         vdev->config->del_vq(vi->rvq);
 996         if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ))
 997                 vdev->config->del_vq(vi->cvq);
 998         unregister_netdev(vi->dev);
 999
1000         while (vi->pages)
1001                 __free_pages(get_a_page(vi, GFP_KERNEL), 0);
1002
1003         free_netdev(vi->dev);
1004 }
1005
1006 static struct virtio_device_id id_table[] = {
1007         { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
1008         { 0 },
1009 };
1010
1011 static unsigned int features[] = {
1012         VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM,
1013         VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC,
1014         VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6,
1015         VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6,
1016         VIRTIO_NET_F_GUEST_ECN, /* We don't yet handle UFO input. */
1017         VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ,
1018         VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN,
1019         VIRTIO_F_NOTIFY_ON_EMPTY,
1020 };
1021
1022 static struct virtio_driver virtio_net = {
1023         .feature_table = features,
1024         .feature_table_size = ARRAY_SIZE(features),
1025         .driver.name =  KBUILD_MODNAME,
1026         .driver.owner = THIS_MODULE,
1027         .id_table =     id_table,
1028         .probe =        virtnet_probe,
1029         .remove =       __devexit_p(virtnet_remove),
1030         .config_changed = virtnet_config_changed,
1031 };
1032
1033 static int __init init(void)
1034 {
1035         return register_virtio_driver(&virtio_net);
1036 }
1037
1038 static void __exit fini(void)
1039 {
1040         unregister_virtio_driver(&virtio_net);
1041 }
1042 module_init(init);
1043 module_exit(fini);
1044
1045 MODULE_DEVICE_TABLE(virtio, id_table);
1046 MODULE_DESCRIPTION("Virtio network driver");
1047 MODULE_LICENSE("GPL");