Pileus Git - ~andy/linux/blob - drivers/net/virtio_net.c

   1 /* A 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 #include <linux/slab.h>
  29
  30 static int napi_weight = 128;
  31 module_param(napi_weight, int, 0444);
  32
  33 static bool csum = true, gso = true;
  34 module_param(csum, bool, 0444);
  35 module_param(gso, bool, 0444);
  36
  37 /* FIXME: MTU in config. */
  38 #define MAX_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
  39 #define GOOD_COPY_LEN   128
  40
  41 #define VIRTNET_SEND_COMMAND_SG_MAX    2
  42 #define VIRTNET_DRIVER_VERSION "1.0.0"
  43
  44 struct virtnet_stats {
  45         struct u64_stats_sync tx_syncp;
  46         struct u64_stats_sync rx_syncp;
  47         u64 tx_bytes;
  48         u64 tx_packets;
  49
  50         u64 rx_bytes;
  51         u64 rx_packets;
  52 };
  53
  54 struct virtnet_info {
  55         struct virtio_device *vdev;
  56         struct virtqueue *rvq, *svq, *cvq;
  57         struct net_device *dev;
  58         struct napi_struct napi;
  59         unsigned int status;
  60
  61         /* Number of input buffers, and max we've ever had. */
  62         unsigned int num, max;
  63
  64         /* I like... big packets and I cannot lie! */
  65         bool big_packets;
  66
  67         /* Host will merge rx buffers for big packets (shake it! shake it!) */
  68         bool mergeable_rx_bufs;
  69
  70         /* enable config space updates */
  71         bool config_enable;
  72
  73         /* Active statistics */
  74         struct virtnet_stats __percpu *stats;
  75
  76         /* Work struct for refilling if we run low on memory. */
  77         struct delayed_work refill;
  78
  79         /* Work struct for config space updates */
  80         struct work_struct config_work;
  81
  82         /* Lock for config space updates */
  83         struct mutex config_lock;
  84
  85         /* Chain pages by the private ptr. */
  86         struct page *pages;
  87
  88         /* fragments + linear part + virtio header */
  89         struct scatterlist rx_sg[MAX_SKB_FRAGS + 2];
  90         struct scatterlist tx_sg[MAX_SKB_FRAGS + 2];
  91 };
  92
  93 struct skb_vnet_hdr {
  94         union {
  95                 struct virtio_net_hdr hdr;
  96                 struct virtio_net_hdr_mrg_rxbuf mhdr;
  97         };
  98         unsigned int num_sg;
  99 };
 100
 101 struct padded_vnet_hdr {
 102         struct virtio_net_hdr hdr;
 103         /*
 104          * virtio_net_hdr should be in a separated sg buffer because of a
 105          * QEMU bug, and data sg buffer shares same page with this header sg.
 106          * This padding makes next sg 16 byte aligned after virtio_net_hdr.
 107          */
 108         char padding[6];
 109 };
 110
 111 static inline struct skb_vnet_hdr *skb_vnet_hdr(struct sk_buff *skb)
 112 {
 113         return (struct skb_vnet_hdr *)skb->cb;
 114 }
 115
 116 /*
 117  * private is used to chain pages for big packets, put the whole
 118  * most recent used list in the beginning for reuse
 119  */
 120 static void give_pages(struct virtnet_info *vi, struct page *page)
 121 {
 122         struct page *end;
 123
 124         /* Find end of list, sew whole thing into vi->pages. */
 125         for (end = page; end->private; end = (struct page *)end->private);
 126         end->private = (unsigned long)vi->pages;
 127         vi->pages = page;
 128 }
 129
 130 static struct page *get_a_page(struct virtnet_info *vi, gfp_t gfp_mask)
 131 {
 132         struct page *p = vi->pages;
 133
 134         if (p) {
 135                 vi->pages = (struct page *)p->private;
 136                 /* clear private here, it is used to chain pages */
 137                 p->private = 0;
 138         } else
 139                 p = alloc_page(gfp_mask);
 140         return p;
 141 }
 142
 143 static void skb_xmit_done(struct virtqueue *svq)
 144 {
 145         struct virtnet_info *vi = svq->vdev->priv;
 146
 147         /* Suppress further interrupts. */
 148         virtqueue_disable_cb(svq);
 149
 150         /* We were probably waiting for more output buffers. */
 151         netif_wake_queue(vi->dev);
 152 }
 153
 154 static void set_skb_frag(struct sk_buff *skb, struct page *page,
 155                          unsigned int offset, unsigned int *len)
 156 {
 157         int size = min((unsigned)PAGE_SIZE - offset, *len);
 158         int i = skb_shinfo(skb)->nr_frags;
 159
 160         __skb_fill_page_desc(skb, i, page, offset, size);
 161
 162         skb->data_len += size;
 163         skb->len += size;
 164         skb->truesize += PAGE_SIZE;
 165         skb_shinfo(skb)->nr_frags++;
 166         *len -= size;
 167 }
 168
 169 /* Called from bottom half context */
 170 static struct sk_buff *page_to_skb(struct virtnet_info *vi,
 171                                    struct page *page, unsigned int len)
 172 {
 173         struct sk_buff *skb;
 174         struct skb_vnet_hdr *hdr;
 175         unsigned int copy, hdr_len, offset;
 176         char *p;
 177
 178         p = page_address(page);
 179
 180         /* copy small packet so we can reuse these pages for small data */
 181         skb = netdev_alloc_skb_ip_align(vi->dev, GOOD_COPY_LEN);
 182         if (unlikely(!skb))
 183                 return NULL;
 184
 185         hdr = skb_vnet_hdr(skb);
 186
 187         if (vi->mergeable_rx_bufs) {
 188                 hdr_len = sizeof hdr->mhdr;
 189                 offset = hdr_len;
 190         } else {
 191                 hdr_len = sizeof hdr->hdr;
 192                 offset = sizeof(struct padded_vnet_hdr);
 193         }
 194
 195         memcpy(hdr, p, hdr_len);
 196
 197         len -= hdr_len;
 198         p += offset;
 199
 200         copy = len;
 201         if (copy > skb_tailroom(skb))
 202                 copy = skb_tailroom(skb);
 203         memcpy(skb_put(skb, copy), p, copy);
 204
 205         len -= copy;
 206         offset += copy;
 207
 208         /*
 209          * Verify that we can indeed put this data into a skb.
 210          * This is here to handle cases when the device erroneously
 211          * tries to receive more than is possible. This is usually
 212          * the case of a broken device.
 213          */
 214         if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) {
 215                 if (net_ratelimit())
 216                         pr_debug("%s: too much data\n", skb->dev->name);
 217                 dev_kfree_skb(skb);
 218                 return NULL;
 219         }
 220
 221         while (len) {
 222                 set_skb_frag(skb, page, offset, &len);
 223                 page = (struct page *)page->private;
 224                 offset = 0;
 225         }
 226
 227         if (page)
 228                 give_pages(vi, page);
 229
 230         return skb;
 231 }
 232
 233 static int receive_mergeable(struct virtnet_info *vi, struct sk_buff *skb)
 234 {
 235         struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
 236         struct page *page;
 237         int num_buf, i, len;
 238
 239         num_buf = hdr->mhdr.num_buffers;
 240         while (--num_buf) {
 241                 i = skb_shinfo(skb)->nr_frags;
 242                 if (i >= MAX_SKB_FRAGS) {
 243                         pr_debug("%s: packet too long\n", skb->dev->name);
 244                         skb->dev->stats.rx_length_errors++;
 245                         return -EINVAL;
 246                 }
 247                 page = virtqueue_get_buf(vi->rvq, &len);
 248                 if (!page) {
 249                         pr_debug("%s: rx error: %d buffers missing\n",
 250                                  skb->dev->name, hdr->mhdr.num_buffers);
 251                         skb->dev->stats.rx_length_errors++;
 252                         return -EINVAL;
 253                 }
 254
 255                 if (len > PAGE_SIZE)
 256                         len = PAGE_SIZE;
 257
 258                 set_skb_frag(skb, page, 0, &len);
 259
 260                 --vi->num;
 261         }
 262         return 0;
 263 }
 264
 265 static void receive_buf(struct net_device *dev, void *buf, unsigned int len)
 266 {
 267         struct virtnet_info *vi = netdev_priv(dev);
 268         struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
 269         struct sk_buff *skb;
 270         struct page *page;
 271         struct skb_vnet_hdr *hdr;
 272
 273         if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) {
 274                 pr_debug("%s: short packet %i\n", dev->name, len);
 275                 dev->stats.rx_length_errors++;
 276                 if (vi->mergeable_rx_bufs || vi->big_packets)
 277                         give_pages(vi, buf);
 278                 else
 279                         dev_kfree_skb(buf);
 280                 return;
 281         }
 282
 283         if (!vi->mergeable_rx_bufs && !vi->big_packets) {
 284                 skb = buf;
 285                 len -= sizeof(struct virtio_net_hdr);
 286                 skb_trim(skb, len);
 287         } else {
 288                 page = buf;
 289                 skb = page_to_skb(vi, page, len);
 290                 if (unlikely(!skb)) {
 291                         dev->stats.rx_dropped++;
 292                         give_pages(vi, page);
 293                         return;
 294                 }
 295                 if (vi->mergeable_rx_bufs)
 296                         if (receive_mergeable(vi, skb)) {
 297                                 dev_kfree_skb(skb);
 298                                 return;
 299                         }
 300         }
 301
 302         hdr = skb_vnet_hdr(skb);
 303
 304         u64_stats_update_begin(&stats->rx_syncp);
 305         stats->rx_bytes += skb->len;
 306         stats->rx_packets++;
 307         u64_stats_update_end(&stats->rx_syncp);
 308
 309         if (hdr->hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
 310                 pr_debug("Needs csum!\n");
 311                 if (!skb_partial_csum_set(skb,
 312                                           hdr->hdr.csum_start,
 313                                           hdr->hdr.csum_offset))
 314                         goto frame_err;
 315         } else if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID) {
 316                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 317         }
 318
 319         skb->protocol = eth_type_trans(skb, dev);
 320         pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
 321                  ntohs(skb->protocol), skb->len, skb->pkt_type);
 322
 323         if (hdr->hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
 324                 pr_debug("GSO!\n");
 325                 switch (hdr->hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
 326                 case VIRTIO_NET_HDR_GSO_TCPV4:
 327                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
 328                         break;
 329                 case VIRTIO_NET_HDR_GSO_UDP:
 330                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
 331                         break;
 332                 case VIRTIO_NET_HDR_GSO_TCPV6:
 333                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
 334                         break;
 335                 default:
 336                         if (net_ratelimit())
 337                                 printk(KERN_WARNING "%s: bad gso type %u.\n",
 338                                        dev->name, hdr->hdr.gso_type);
 339                         goto frame_err;
 340                 }
 341
 342                 if (hdr->hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
 343                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
 344
 345                 skb_shinfo(skb)->gso_size = hdr->hdr.gso_size;
 346                 if (skb_shinfo(skb)->gso_size == 0) {
 347                         if (net_ratelimit())
 348                                 printk(KERN_WARNING "%s: zero gso size.\n",
 349                                        dev->name);
 350                         goto frame_err;
 351                 }
 352
 353                 /* Header must be checked, and gso_segs computed. */
 354                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
 355                 skb_shinfo(skb)->gso_segs = 0;
 356         }
 357
 358         netif_receive_skb(skb);
 359         return;
 360
 361 frame_err:
 362         dev->stats.rx_frame_errors++;
 363         dev_kfree_skb(skb);
 364 }
 365
 366 static int add_recvbuf_small(struct virtnet_info *vi, gfp_t gfp)
 367 {
 368         struct sk_buff *skb;
 369         struct skb_vnet_hdr *hdr;
 370         int err;
 371
 372         skb = __netdev_alloc_skb_ip_align(vi->dev, MAX_PACKET_LEN, gfp);
 373         if (unlikely(!skb))
 374                 return -ENOMEM;
 375
 376         skb_put(skb, MAX_PACKET_LEN);
 377
 378         hdr = skb_vnet_hdr(skb);
 379         sg_set_buf(vi->rx_sg, &hdr->hdr, sizeof hdr->hdr);
 380
 381         skb_to_sgvec(skb, vi->rx_sg + 1, 0, skb->len);
 382
 383         err = virtqueue_add_buf(vi->rvq, vi->rx_sg, 0, 2, skb, gfp);
 384         if (err < 0)
 385                 dev_kfree_skb(skb);
 386
 387         return err;
 388 }
 389
 390 static int add_recvbuf_big(struct virtnet_info *vi, gfp_t gfp)
 391 {
 392         struct page *first, *list = NULL;
 393         char *p;
 394         int i, err, offset;
 395
 396         /* page in vi->rx_sg[MAX_SKB_FRAGS + 1] is list tail */
 397         for (i = MAX_SKB_FRAGS + 1; i > 1; --i) {
 398                 first = get_a_page(vi, gfp);
 399                 if (!first) {
 400                         if (list)
 401                                 give_pages(vi, list);
 402                         return -ENOMEM;
 403                 }
 404                 sg_set_buf(&vi->rx_sg[i], page_address(first), PAGE_SIZE);
 405
 406                 /* chain new page in list head to match sg */
 407                 first->private = (unsigned long)list;
 408                 list = first;
 409         }
 410
 411         first = get_a_page(vi, gfp);
 412         if (!first) {
 413                 give_pages(vi, list);
 414                 return -ENOMEM;
 415         }
 416         p = page_address(first);
 417
 418         /* vi->rx_sg[0], vi->rx_sg[1] share the same page */
 419         /* a separated vi->rx_sg[0] for virtio_net_hdr only due to QEMU bug */
 420         sg_set_buf(&vi->rx_sg[0], p, sizeof(struct virtio_net_hdr));
 421
 422         /* vi->rx_sg[1] for data packet, from offset */
 423         offset = sizeof(struct padded_vnet_hdr);
 424         sg_set_buf(&vi->rx_sg[1], p + offset, PAGE_SIZE - offset);
 425
 426         /* chain first in list head */
 427         first->private = (unsigned long)list;
 428         err = virtqueue_add_buf(vi->rvq, vi->rx_sg, 0, MAX_SKB_FRAGS + 2,
 429                                 first, gfp);
 430         if (err < 0)
 431                 give_pages(vi, first);
 432
 433         return err;
 434 }
 435
 436 static int add_recvbuf_mergeable(struct virtnet_info *vi, gfp_t gfp)
 437 {
 438         struct page *page;
 439         int err;
 440
 441         page = get_a_page(vi, gfp);
 442         if (!page)
 443                 return -ENOMEM;
 444
 445         sg_init_one(vi->rx_sg, page_address(page), PAGE_SIZE);
 446
 447         err = virtqueue_add_buf(vi->rvq, vi->rx_sg, 0, 1, page, gfp);
 448         if (err < 0)
 449                 give_pages(vi, page);
 450
 451         return err;
 452 }
 453
 454 /*
 455  * Returns false if we couldn't fill entirely (OOM).
 456  *
 457  * Normally run in the receive path, but can also be run from ndo_open
 458  * before we're receiving packets, or from refill_work which is
 459  * careful to disable receiving (using napi_disable).
 460  */
 461 static bool try_fill_recv(struct virtnet_info *vi, gfp_t gfp)
 462 {
 463         int err;
 464         bool oom;
 465
 466         do {
 467                 if (vi->mergeable_rx_bufs)
 468                         err = add_recvbuf_mergeable(vi, gfp);
 469                 else if (vi->big_packets)
 470                         err = add_recvbuf_big(vi, gfp);
 471                 else
 472                         err = add_recvbuf_small(vi, gfp);
 473
 474                 oom = err == -ENOMEM;
 475                 if (err < 0)
 476                         break;
 477                 ++vi->num;
 478         } while (err > 0);
 479         if (unlikely(vi->num > vi->max))
 480                 vi->max = vi->num;
 481         virtqueue_kick(vi->rvq);
 482         return !oom;
 483 }
 484
 485 static void skb_recv_done(struct virtqueue *rvq)
 486 {
 487         struct virtnet_info *vi = rvq->vdev->priv;
 488         /* Schedule NAPI, Suppress further interrupts if successful. */
 489         if (napi_schedule_prep(&vi->napi)) {
 490                 virtqueue_disable_cb(rvq);
 491                 __napi_schedule(&vi->napi);
 492         }
 493 }
 494
 495 static void virtnet_napi_enable(struct virtnet_info *vi)
 496 {
 497         napi_enable(&vi->napi);
 498
 499         /* If all buffers were filled by other side before we napi_enabled, we
 500          * won't get another interrupt, so process any outstanding packets
 501          * now.  virtnet_poll wants re-enable the queue, so we disable here.
 502          * We synchronize against interrupts via NAPI_STATE_SCHED */
 503         if (napi_schedule_prep(&vi->napi)) {
 504                 virtqueue_disable_cb(vi->rvq);
 505                 local_bh_disable();
 506                 __napi_schedule(&vi->napi);
 507                 local_bh_enable();
 508         }
 509 }
 510
 511 static void refill_work(struct work_struct *work)
 512 {
 513         struct virtnet_info *vi;
 514         bool still_empty;
 515
 516         vi = container_of(work, struct virtnet_info, refill.work);
 517         napi_disable(&vi->napi);
 518         still_empty = !try_fill_recv(vi, GFP_KERNEL);
 519         virtnet_napi_enable(vi);
 520
 521         /* In theory, this can happen: if we don't get any buffers in
 522          * we will *never* try to fill again. */
 523         if (still_empty)
 524                 schedule_delayed_work(&vi->refill, HZ/2);
 525 }
 526
 527 static int virtnet_poll(struct napi_struct *napi, int budget)
 528 {
 529         struct virtnet_info *vi = container_of(napi, struct virtnet_info, napi);
 530         void *buf;
 531         unsigned int len, received = 0;
 532
 533 again:
 534         while (received < budget &&
 535                (buf = virtqueue_get_buf(vi->rvq, &len)) != NULL) {
 536                 receive_buf(vi->dev, buf, len);
 537                 --vi->num;
 538                 received++;
 539         }
 540
 541         if (vi->num < vi->max / 2) {
 542                 if (!try_fill_recv(vi, GFP_ATOMIC))
 543                         schedule_delayed_work(&vi->refill, 0);
 544         }
 545
 546         /* Out of packets? */
 547         if (received < budget) {
 548                 napi_complete(napi);
 549                 if (unlikely(!virtqueue_enable_cb(vi->rvq)) &&
 550                     napi_schedule_prep(napi)) {
 551                         virtqueue_disable_cb(vi->rvq);
 552                         __napi_schedule(napi);
 553                         goto again;
 554                 }
 555         }
 556
 557         return received;
 558 }
 559
 560 static void free_old_xmit_skbs(struct virtnet_info *vi)
 561 {
 562         struct sk_buff *skb;
 563         unsigned int len;
 564         struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
 565
 566         while ((skb = virtqueue_get_buf(vi->svq, &len)) != NULL) {
 567                 pr_debug("Sent skb %p\n", skb);
 568
 569                 u64_stats_update_begin(&stats->tx_syncp);
 570                 stats->tx_bytes += skb->len;
 571                 stats->tx_packets++;
 572                 u64_stats_update_end(&stats->tx_syncp);
 573
 574                 dev_kfree_skb_any(skb);
 575         }
 576 }
 577
 578 static int xmit_skb(struct virtnet_info *vi, struct sk_buff *skb)
 579 {
 580         struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
 581         const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
 582
 583         pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
 584
 585         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 586                 hdr->hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
 587                 hdr->hdr.csum_start = skb_checksum_start_offset(skb);
 588                 hdr->hdr.csum_offset = skb->csum_offset;
 589         } else {
 590                 hdr->hdr.flags = 0;
 591                 hdr->hdr.csum_offset = hdr->hdr.csum_start = 0;
 592         }
 593
 594         if (skb_is_gso(skb)) {
 595                 hdr->hdr.hdr_len = skb_headlen(skb);
 596                 hdr->hdr.gso_size = skb_shinfo(skb)->gso_size;
 597                 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
 598                         hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
 599                 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
 600                         hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
 601                 else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 602                         hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
 603                 else
 604                         BUG();
 605                 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
 606                         hdr->hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
 607         } else {
 608                 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
 609                 hdr->hdr.gso_size = hdr->hdr.hdr_len = 0;
 610         }
 611
 612         hdr->mhdr.num_buffers = 0;
 613
 614         /* Encode metadata header at front. */
 615         if (vi->mergeable_rx_bufs)
 616                 sg_set_buf(vi->tx_sg, &hdr->mhdr, sizeof hdr->mhdr);
 617         else
 618                 sg_set_buf(vi->tx_sg, &hdr->hdr, sizeof hdr->hdr);
 619
 620         hdr->num_sg = skb_to_sgvec(skb, vi->tx_sg + 1, 0, skb->len) + 1;
 621         return virtqueue_add_buf(vi->svq, vi->tx_sg, hdr->num_sg,
 622                                  0, skb, GFP_ATOMIC);
 623 }
 624
 625 static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
 626 {
 627         struct virtnet_info *vi = netdev_priv(dev);
 628         int capacity;
 629
 630         /* Free up any pending old buffers before queueing new ones. */
 631         free_old_xmit_skbs(vi);
 632
 633         /* Try to transmit */
 634         capacity = xmit_skb(vi, skb);
 635
 636         /* This can happen with OOM and indirect buffers. */
 637         if (unlikely(capacity < 0)) {
 638                 if (likely(capacity == -ENOMEM)) {
 639                         if (net_ratelimit())
 640                                 dev_warn(&dev->dev,
 641                                          "TX queue failure: out of memory\n");
 642                 } else {
 643                         dev->stats.tx_fifo_errors++;
 644                         if (net_ratelimit())
 645                                 dev_warn(&dev->dev,
 646                                          "Unexpected TX queue failure: %d\n",
 647                                          capacity);
 648                 }
 649                 dev->stats.tx_dropped++;
 650                 kfree_skb(skb);
 651                 return NETDEV_TX_OK;
 652         }
 653         virtqueue_kick(vi->svq);
 654
 655         /* Don't wait up for transmitted skbs to be freed. */
 656         skb_orphan(skb);
 657         nf_reset(skb);
 658
 659         /* Apparently nice girls don't return TX_BUSY; stop the queue
 660          * before it gets out of hand.  Naturally, this wastes entries. */
 661         if (capacity < 2+MAX_SKB_FRAGS) {
 662                 netif_stop_queue(dev);
 663                 if (unlikely(!virtqueue_enable_cb_delayed(vi->svq))) {
 664                         /* More just got used, free them then recheck. */
 665                         free_old_xmit_skbs(vi);
 666                         capacity = vi->svq->num_free;
 667                         if (capacity >= 2+MAX_SKB_FRAGS) {
 668                                 netif_start_queue(dev);
 669                                 virtqueue_disable_cb(vi->svq);
 670                         }
 671                 }
 672         }
 673
 674         return NETDEV_TX_OK;
 675 }
 676
 677 static int virtnet_set_mac_address(struct net_device *dev, void *p)
 678 {
 679         struct virtnet_info *vi = netdev_priv(dev);
 680         struct virtio_device *vdev = vi->vdev;
 681         int ret;
 682
 683         ret = eth_mac_addr(dev, p);
 684         if (ret)
 685                 return ret;
 686
 687         if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC))
 688                 vdev->config->set(vdev, offsetof(struct virtio_net_config, mac),
 689                                   dev->dev_addr, dev->addr_len);
 690
 691         return 0;
 692 }
 693
 694 static struct rtnl_link_stats64 *virtnet_stats(struct net_device *dev,
 695                                                struct rtnl_link_stats64 *tot)
 696 {
 697         struct virtnet_info *vi = netdev_priv(dev);
 698         int cpu;
 699         unsigned int start;
 700
 701         for_each_possible_cpu(cpu) {
 702                 struct virtnet_stats *stats = per_cpu_ptr(vi->stats, cpu);
 703                 u64 tpackets, tbytes, rpackets, rbytes;
 704
 705                 do {
 706                         start = u64_stats_fetch_begin_bh(&stats->tx_syncp);
 707                         tpackets = stats->tx_packets;
 708                         tbytes   = stats->tx_bytes;
 709                 } while (u64_stats_fetch_retry_bh(&stats->tx_syncp, start));
 710
 711                 do {
 712                         start = u64_stats_fetch_begin_bh(&stats->rx_syncp);
 713                         rpackets = stats->rx_packets;
 714                         rbytes   = stats->rx_bytes;
 715                 } while (u64_stats_fetch_retry_bh(&stats->rx_syncp, start));
 716
 717                 tot->rx_packets += rpackets;
 718                 tot->tx_packets += tpackets;
 719                 tot->rx_bytes   += rbytes;
 720                 tot->tx_bytes   += tbytes;
 721         }
 722
 723         tot->tx_dropped = dev->stats.tx_dropped;
 724         tot->tx_fifo_errors = dev->stats.tx_fifo_errors;
 725         tot->rx_dropped = dev->stats.rx_dropped;
 726         tot->rx_length_errors = dev->stats.rx_length_errors;
 727         tot->rx_frame_errors = dev->stats.rx_frame_errors;
 728
 729         return tot;
 730 }
 731
 732 #ifdef CONFIG_NET_POLL_CONTROLLER
 733 static void virtnet_netpoll(struct net_device *dev)
 734 {
 735         struct virtnet_info *vi = netdev_priv(dev);
 736
 737         napi_schedule(&vi->napi);
 738 }
 739 #endif
 740
 741 static int virtnet_open(struct net_device *dev)
 742 {
 743         struct virtnet_info *vi = netdev_priv(dev);
 744
 745         /* Make sure we have some buffers: if oom use wq. */
 746         if (!try_fill_recv(vi, GFP_KERNEL))
 747                 schedule_delayed_work(&vi->refill, 0);
 748
 749         virtnet_napi_enable(vi);
 750         return 0;
 751 }
 752
 753 /*
 754  * Send command via the control virtqueue and check status.  Commands
 755  * supported by the hypervisor, as indicated by feature bits, should
 756  * never fail unless improperly formated.
 757  */
 758 static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
 759                                  struct scatterlist *data, int out, int in)
 760 {
 761         struct scatterlist *s, sg[VIRTNET_SEND_COMMAND_SG_MAX + 2];
 762         struct virtio_net_ctrl_hdr ctrl;
 763         virtio_net_ctrl_ack status = ~0;
 764         unsigned int tmp;
 765         int i;
 766
 767         /* Caller should know better */
 768         BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ||
 769                 (out + in > VIRTNET_SEND_COMMAND_SG_MAX));
 770
 771         out++; /* Add header */
 772         in++; /* Add return status */
 773
 774         ctrl.class = class;
 775         ctrl.cmd = cmd;
 776
 777         sg_init_table(sg, out + in);
 778
 779         sg_set_buf(&sg[0], &ctrl, sizeof(ctrl));
 780         for_each_sg(data, s, out + in - 2, i)
 781                 sg_set_buf(&sg[i + 1], sg_virt(s), s->length);
 782         sg_set_buf(&sg[out + in - 1], &status, sizeof(status));
 783
 784         BUG_ON(virtqueue_add_buf(vi->cvq, sg, out, in, vi, GFP_ATOMIC) < 0);
 785
 786         virtqueue_kick(vi->cvq);
 787
 788         /*
 789          * Spin for a response, the kick causes an ioport write, trapping
 790          * into the hypervisor, so the request should be handled immediately.
 791          */
 792         while (!virtqueue_get_buf(vi->cvq, &tmp))
 793                 cpu_relax();
 794
 795         return status == VIRTIO_NET_OK;
 796 }
 797
 798 static void virtnet_ack_link_announce(struct virtnet_info *vi)
 799 {
 800         rtnl_lock();
 801         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE,
 802                                   VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL,
 803                                   0, 0))
 804                 dev_warn(&vi->dev->dev, "Failed to ack link announce.\n");
 805         rtnl_unlock();
 806 }
 807
 808 static int virtnet_close(struct net_device *dev)
 809 {
 810         struct virtnet_info *vi = netdev_priv(dev);
 811
 812         /* Make sure refill_work doesn't re-enable napi! */
 813         cancel_delayed_work_sync(&vi->refill);
 814         napi_disable(&vi->napi);
 815
 816         return 0;
 817 }
 818
 819 static void virtnet_set_rx_mode(struct net_device *dev)
 820 {
 821         struct virtnet_info *vi = netdev_priv(dev);
 822         struct scatterlist sg[2];
 823         u8 promisc, allmulti;
 824         struct virtio_net_ctrl_mac *mac_data;
 825         struct netdev_hw_addr *ha;
 826         int uc_count;
 827         int mc_count;
 828         void *buf;
 829         int i;
 830
 831         /* We can't dynamicaly set ndo_set_rx_mode, so return gracefully */
 832         if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
 833                 return;
 834
 835         promisc = ((dev->flags & IFF_PROMISC) != 0);
 836         allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
 837
 838         sg_init_one(sg, &promisc, sizeof(promisc));
 839
 840         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
 841                                   VIRTIO_NET_CTRL_RX_PROMISC,
 842                                   sg, 1, 0))
 843                 dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
 844                          promisc ? "en" : "dis");
 845
 846         sg_init_one(sg, &allmulti, sizeof(allmulti));
 847
 848         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
 849                                   VIRTIO_NET_CTRL_RX_ALLMULTI,
 850                                   sg, 1, 0))
 851                 dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
 852                          allmulti ? "en" : "dis");
 853
 854         uc_count = netdev_uc_count(dev);
 855         mc_count = netdev_mc_count(dev);
 856         /* MAC filter - use one buffer for both lists */
 857         buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) +
 858                       (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
 859         mac_data = buf;
 860         if (!buf) {
 861                 dev_warn(&dev->dev, "No memory for MAC address buffer\n");
 862                 return;
 863         }
 864
 865         sg_init_table(sg, 2);
 866
 867         /* Store the unicast list and count in the front of the buffer */
 868         mac_data->entries = uc_count;
 869         i = 0;
 870         netdev_for_each_uc_addr(ha, dev)
 871                 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
 872
 873         sg_set_buf(&sg[0], mac_data,
 874                    sizeof(mac_data->entries) + (uc_count * ETH_ALEN));
 875
 876         /* multicast list and count fill the end */
 877         mac_data = (void *)&mac_data->macs[uc_count][0];
 878
 879         mac_data->entries = mc_count;
 880         i = 0;
 881         netdev_for_each_mc_addr(ha, dev)
 882                 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
 883
 884         sg_set_buf(&sg[1], mac_data,
 885                    sizeof(mac_data->entries) + (mc_count * ETH_ALEN));
 886
 887         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
 888                                   VIRTIO_NET_CTRL_MAC_TABLE_SET,
 889                                   sg, 2, 0))
 890                 dev_warn(&dev->dev, "Failed to set MAC fitler table.\n");
 891
 892         kfree(buf);
 893 }
 894
 895 static int virtnet_vlan_rx_add_vid(struct net_device *dev, u16 vid)
 896 {
 897         struct virtnet_info *vi = netdev_priv(dev);
 898         struct scatterlist sg;
 899
 900         sg_init_one(&sg, &vid, sizeof(vid));
 901
 902         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
 903                                   VIRTIO_NET_CTRL_VLAN_ADD, &sg, 1, 0))
 904                 dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
 905         return 0;
 906 }
 907
 908 static int virtnet_vlan_rx_kill_vid(struct net_device *dev, u16 vid)
 909 {
 910         struct virtnet_info *vi = netdev_priv(dev);
 911         struct scatterlist sg;
 912
 913         sg_init_one(&sg, &vid, sizeof(vid));
 914
 915         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
 916                                   VIRTIO_NET_CTRL_VLAN_DEL, &sg, 1, 0))
 917                 dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
 918         return 0;
 919 }
 920
 921 static void virtnet_get_ringparam(struct net_device *dev,
 922                                 struct ethtool_ringparam *ring)
 923 {
 924         struct virtnet_info *vi = netdev_priv(dev);
 925
 926         ring->rx_max_pending = virtqueue_get_vring_size(vi->rvq);
 927         ring->tx_max_pending = virtqueue_get_vring_size(vi->svq);
 928         ring->rx_pending = ring->rx_max_pending;
 929         ring->tx_pending = ring->tx_max_pending;
 930
 931 }
 932
 933
 934 static void virtnet_get_drvinfo(struct net_device *dev,
 935                                 struct ethtool_drvinfo *info)
 936 {
 937         struct virtnet_info *vi = netdev_priv(dev);
 938         struct virtio_device *vdev = vi->vdev;
 939
 940         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
 941         strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version));
 942         strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info));
 943
 944 }
 945
 946 static const struct ethtool_ops virtnet_ethtool_ops = {
 947         .get_drvinfo = virtnet_get_drvinfo,
 948         .get_link = ethtool_op_get_link,
 949         .get_ringparam = virtnet_get_ringparam,
 950 };
 951
 952 #define MIN_MTU 68
 953 #define MAX_MTU 65535
 954
 955 static int virtnet_change_mtu(struct net_device *dev, int new_mtu)
 956 {
 957         if (new_mtu < MIN_MTU || new_mtu > MAX_MTU)
 958                 return -EINVAL;
 959         dev->mtu = new_mtu;
 960         return 0;
 961 }
 962
 963 static const struct net_device_ops virtnet_netdev = {
 964         .ndo_open            = virtnet_open,
 965         .ndo_stop            = virtnet_close,
 966         .ndo_start_xmit      = start_xmit,
 967         .ndo_validate_addr   = eth_validate_addr,
 968         .ndo_set_mac_address = virtnet_set_mac_address,
 969         .ndo_set_rx_mode     = virtnet_set_rx_mode,
 970         .ndo_change_mtu      = virtnet_change_mtu,
 971         .ndo_get_stats64     = virtnet_stats,
 972         .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
 973         .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
 974 #ifdef CONFIG_NET_POLL_CONTROLLER
 975         .ndo_poll_controller = virtnet_netpoll,
 976 #endif
 977 };
 978
 979 static void virtnet_config_changed_work(struct work_struct *work)
 980 {
 981         struct virtnet_info *vi =
 982                 container_of(work, struct virtnet_info, config_work);
 983         u16 v;
 984
 985         mutex_lock(&vi->config_lock);
 986         if (!vi->config_enable)
 987                 goto done;
 988
 989         if (virtio_config_val(vi->vdev, VIRTIO_NET_F_STATUS,
 990                               offsetof(struct virtio_net_config, status),
 991                               &v) < 0)
 992                 goto done;
 993
 994         if (v & VIRTIO_NET_S_ANNOUNCE) {
 995                 netdev_notify_peers(vi->dev);
 996                 virtnet_ack_link_announce(vi);
 997         }
 998
 999         /* Ignore unknown (future) status bits */
1000         v &= VIRTIO_NET_S_LINK_UP;
1001
1002         if (vi->status == v)
1003                 goto done;
1004
1005         vi->status = v;
1006
1007         if (vi->status & VIRTIO_NET_S_LINK_UP) {
1008                 netif_carrier_on(vi->dev);
1009                 netif_wake_queue(vi->dev);
1010         } else {
1011                 netif_carrier_off(vi->dev);
1012                 netif_stop_queue(vi->dev);
1013         }
1014 done:
1015         mutex_unlock(&vi->config_lock);
1016 }
1017
1018 static void virtnet_config_changed(struct virtio_device *vdev)
1019 {
1020         struct virtnet_info *vi = vdev->priv;
1021
1022         schedule_work(&vi->config_work);
1023 }
1024
1025 static int init_vqs(struct virtnet_info *vi)
1026 {
1027         struct virtqueue *vqs[3];
1028         vq_callback_t *callbacks[] = { skb_recv_done, skb_xmit_done, NULL};
1029         const char *names[] = { "input", "output", "control" };
1030         int nvqs, err;
1031
1032         /* We expect two virtqueues, receive then send,
1033          * and optionally control. */
1034         nvqs = virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ? 3 : 2;
1035
1036         err = vi->vdev->config->find_vqs(vi->vdev, nvqs, vqs, callbacks, names);
1037         if (err)
1038                 return err;
1039
1040         vi->rvq = vqs[0];
1041         vi->svq = vqs[1];
1042
1043         if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)) {
1044                 vi->cvq = vqs[2];
1045
1046                 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
1047                         vi->dev->features |= NETIF_F_HW_VLAN_FILTER;
1048         }
1049         return 0;
1050 }
1051
1052 static int virtnet_probe(struct virtio_device *vdev)
1053 {
1054         int err;
1055         struct net_device *dev;
1056         struct virtnet_info *vi;
1057
1058         /* Allocate ourselves a network device with room for our info */
1059         dev = alloc_etherdev(sizeof(struct virtnet_info));
1060         if (!dev)
1061                 return -ENOMEM;
1062
1063         /* Set up network device as normal. */
1064         dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE;
1065         dev->netdev_ops = &virtnet_netdev;
1066         dev->features = NETIF_F_HIGHDMA;
1067
1068         SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops);
1069         SET_NETDEV_DEV(dev, &vdev->dev);
1070
1071         /* Do we support "hardware" checksums? */
1072         if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
1073                 /* This opens up the world of extra features. */
1074                 dev->hw_features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
1075                 if (csum)
1076                         dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
1077
1078                 if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
1079                         dev->hw_features |= NETIF_F_TSO | NETIF_F_UFO
1080                                 | NETIF_F_TSO_ECN | NETIF_F_TSO6;
1081                 }
1082                 /* Individual feature bits: what can host handle? */
1083                 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
1084                         dev->hw_features |= NETIF_F_TSO;
1085                 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
1086                         dev->hw_features |= NETIF_F_TSO6;
1087                 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
1088                         dev->hw_features |= NETIF_F_TSO_ECN;
1089                 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
1090                         dev->hw_features |= NETIF_F_UFO;
1091
1092                 if (gso)
1093                         dev->features |= dev->hw_features & (NETIF_F_ALL_TSO|NETIF_F_UFO);
1094                 /* (!csum && gso) case will be fixed by register_netdev() */
1095         }
1096
1097         /* Configuration may specify what MAC to use.  Otherwise random. */
1098         if (virtio_config_val_len(vdev, VIRTIO_NET_F_MAC,
1099                                   offsetof(struct virtio_net_config, mac),
1100                                   dev->dev_addr, dev->addr_len) < 0)
1101                 eth_hw_addr_random(dev);
1102
1103         /* Set up our device-specific information */
1104         vi = netdev_priv(dev);
1105         netif_napi_add(dev, &vi->napi, virtnet_poll, napi_weight);
1106         vi->dev = dev;
1107         vi->vdev = vdev;
1108         vdev->priv = vi;
1109         vi->pages = NULL;
1110         vi->stats = alloc_percpu(struct virtnet_stats);
1111         err = -ENOMEM;
1112         if (vi->stats == NULL)
1113                 goto free;
1114
1115         INIT_DELAYED_WORK(&vi->refill, refill_work);
1116         mutex_init(&vi->config_lock);
1117         vi->config_enable = true;
1118         INIT_WORK(&vi->config_work, virtnet_config_changed_work);
1119         sg_init_table(vi->rx_sg, ARRAY_SIZE(vi->rx_sg));
1120         sg_init_table(vi->tx_sg, ARRAY_SIZE(vi->tx_sg));
1121
1122         /* If we can receive ANY GSO packets, we must allocate large ones. */
1123         if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
1124             virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) ||
1125             virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN))
1126                 vi->big_packets = true;
1127
1128         if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
1129                 vi->mergeable_rx_bufs = true;
1130
1131         err = init_vqs(vi);
1132         if (err)
1133                 goto free_stats;
1134
1135         err = register_netdev(dev);
1136         if (err) {
1137                 pr_debug("virtio_net: registering device failed\n");
1138                 goto free_vqs;
1139         }
1140
1141         /* Last of all, set up some receive buffers. */
1142         try_fill_recv(vi, GFP_KERNEL);
1143
1144         /* If we didn't even get one input buffer, we're useless. */
1145         if (vi->num == 0) {
1146                 err = -ENOMEM;
1147                 goto unregister;
1148         }
1149
1150         /* Assume link up if device can't report link status,
1151            otherwise get link status from config. */
1152         if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) {
1153                 netif_carrier_off(dev);
1154                 schedule_work(&vi->config_work);
1155         } else {
1156                 vi->status = VIRTIO_NET_S_LINK_UP;
1157                 netif_carrier_on(dev);
1158         }
1159
1160         pr_debug("virtnet: registered device %s\n", dev->name);
1161         return 0;
1162
1163 unregister:
1164         unregister_netdev(dev);
1165 free_vqs:
1166         vdev->config->del_vqs(vdev);
1167 free_stats:
1168         free_percpu(vi->stats);
1169 free:
1170         free_netdev(dev);
1171         return err;
1172 }
1173
1174 static void free_unused_bufs(struct virtnet_info *vi)
1175 {
1176         void *buf;
1177         while (1) {
1178                 buf = virtqueue_detach_unused_buf(vi->svq);
1179                 if (!buf)
1180                         break;
1181                 dev_kfree_skb(buf);
1182         }
1183         while (1) {
1184                 buf = virtqueue_detach_unused_buf(vi->rvq);
1185                 if (!buf)
1186                         break;
1187                 if (vi->mergeable_rx_bufs || vi->big_packets)
1188                         give_pages(vi, buf);
1189                 else
1190                         dev_kfree_skb(buf);
1191                 --vi->num;
1192         }
1193         BUG_ON(vi->num != 0);
1194 }
1195
1196 static void remove_vq_common(struct virtnet_info *vi)
1197 {
1198         vi->vdev->config->reset(vi->vdev);
1199
1200         /* Free unused buffers in both send and recv, if any. */
1201         free_unused_bufs(vi);
1202
1203         vi->vdev->config->del_vqs(vi->vdev);
1204
1205         while (vi->pages)
1206                 __free_pages(get_a_page(vi, GFP_KERNEL), 0);
1207 }
1208
1209 static void __devexit virtnet_remove(struct virtio_device *vdev)
1210 {
1211         struct virtnet_info *vi = vdev->priv;
1212
1213         /* Prevent config work handler from accessing the device. */
1214         mutex_lock(&vi->config_lock);
1215         vi->config_enable = false;
1216         mutex_unlock(&vi->config_lock);
1217
1218         unregister_netdev(vi->dev);
1219
1220         remove_vq_common(vi);
1221
1222         flush_work(&vi->config_work);
1223
1224         free_percpu(vi->stats);
1225         free_netdev(vi->dev);
1226 }
1227
1228 #ifdef CONFIG_PM
1229 static int virtnet_freeze(struct virtio_device *vdev)
1230 {
1231         struct virtnet_info *vi = vdev->priv;
1232
1233         /* Prevent config work handler from accessing the device */
1234         mutex_lock(&vi->config_lock);
1235         vi->config_enable = false;
1236         mutex_unlock(&vi->config_lock);
1237
1238         netif_device_detach(vi->dev);
1239         cancel_delayed_work_sync(&vi->refill);
1240
1241         if (netif_running(vi->dev))
1242                 napi_disable(&vi->napi);
1243
1244         remove_vq_common(vi);
1245
1246         flush_work(&vi->config_work);
1247
1248         return 0;
1249 }
1250
1251 static int virtnet_restore(struct virtio_device *vdev)
1252 {
1253         struct virtnet_info *vi = vdev->priv;
1254         int err;
1255
1256         err = init_vqs(vi);
1257         if (err)
1258                 return err;
1259
1260         if (netif_running(vi->dev))
1261                 virtnet_napi_enable(vi);
1262
1263         netif_device_attach(vi->dev);
1264
1265         if (!try_fill_recv(vi, GFP_KERNEL))
1266                 schedule_delayed_work(&vi->refill, 0);
1267
1268         mutex_lock(&vi->config_lock);
1269         vi->config_enable = true;
1270         mutex_unlock(&vi->config_lock);
1271
1272         return 0;
1273 }
1274 #endif
1275
1276 static struct virtio_device_id id_table[] = {
1277         { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
1278         { 0 },
1279 };
1280
1281 static unsigned int features[] = {
1282         VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM,
1283         VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC,
1284         VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6,
1285         VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6,
1286         VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO,
1287         VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ,
1288         VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN,
1289         VIRTIO_NET_F_GUEST_ANNOUNCE,
1290 };
1291
1292 static struct virtio_driver virtio_net_driver = {
1293         .feature_table = features,
1294         .feature_table_size = ARRAY_SIZE(features),
1295         .driver.name =  KBUILD_MODNAME,
1296         .driver.owner = THIS_MODULE,
1297         .id_table =     id_table,
1298         .probe =        virtnet_probe,
1299         .remove =       __devexit_p(virtnet_remove),
1300         .config_changed = virtnet_config_changed,
1301 #ifdef CONFIG_PM
1302         .freeze =       virtnet_freeze,
1303         .restore =      virtnet_restore,
1304 #endif
1305 };
1306
1307 static int __init init(void)
1308 {
1309         return register_virtio_driver(&virtio_net_driver);
1310 }
1311
1312 static void __exit fini(void)
1313 {
1314         unregister_virtio_driver(&virtio_net_driver);
1315 }
1316 module_init(init);
1317 module_exit(fini);
1318
1319 MODULE_DEVICE_TABLE(virtio, id_table);
1320 MODULE_DESCRIPTION("Virtio network driver");
1321 MODULE_LICENSE("GPL");