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 syncp;
  46         u64 tx_bytes;
  47         u64 tx_packets;
  48
  49         u64 rx_bytes;
  50         u64 rx_packets;
  51 };
  52
  53 struct virtnet_info {
  54         struct virtio_device *vdev;
  55         struct virtqueue *rvq, *svq, *cvq;
  56         struct net_device *dev;
  57         struct napi_struct napi;
  58         unsigned int status;
  59
  60         /* Number of input buffers, and max we've ever had. */
  61         unsigned int num, max;
  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         /* enable config space updates */
  70         bool config_enable;
  71
  72         /* Active statistics */
  73         struct virtnet_stats __percpu *stats;
  74
  75         /* Work struct for refilling if we run low on memory. */
  76         struct delayed_work refill;
  77
  78         /* Work struct for config space updates */
  79         struct work_struct config_work;
  80
  81         /* Lock for config space updates */
  82         struct mutex config_lock;
  83
  84         /* Chain pages by the private ptr. */
  85         struct page *pages;
  86
  87         /* fragments + linear part + virtio header */
  88         struct scatterlist rx_sg[MAX_SKB_FRAGS + 2];
  89         struct scatterlist tx_sg[MAX_SKB_FRAGS + 2];
  90 };
  91
  92 struct skb_vnet_hdr {
  93         union {
  94                 struct virtio_net_hdr hdr;
  95                 struct virtio_net_hdr_mrg_rxbuf mhdr;
  96         };
  97         unsigned int num_sg;
  98 };
  99
 100 struct padded_vnet_hdr {
 101         struct virtio_net_hdr hdr;
 102         /*
 103          * virtio_net_hdr should be in a separated sg buffer because of a
 104          * QEMU bug, and data sg buffer shares same page with this header sg.
 105          * This padding makes next sg 16 byte aligned after virtio_net_hdr.
 106          */
 107         char padding[6];
 108 };
 109
 110 static inline struct skb_vnet_hdr *skb_vnet_hdr(struct sk_buff *skb)
 111 {
 112         return (struct skb_vnet_hdr *)skb->cb;
 113 }
 114
 115 /*
 116  * private is used to chain pages for big packets, put the whole
 117  * most recent used list in the beginning for reuse
 118  */
 119 static void give_pages(struct virtnet_info *vi, struct page *page)
 120 {
 121         struct page *end;
 122
 123         /* Find end of list, sew whole thing into vi->pages. */
 124         for (end = page; end->private; end = (struct page *)end->private);
 125         end->private = (unsigned long)vi->pages;
 126         vi->pages = page;
 127 }
 128
 129 static struct page *get_a_page(struct virtnet_info *vi, gfp_t gfp_mask)
 130 {
 131         struct page *p = vi->pages;
 132
 133         if (p) {
 134                 vi->pages = (struct page *)p->private;
 135                 /* clear private here, it is used to chain pages */
 136                 p->private = 0;
 137         } else
 138                 p = alloc_page(gfp_mask);
 139         return p;
 140 }
 141
 142 static void skb_xmit_done(struct virtqueue *svq)
 143 {
 144         struct virtnet_info *vi = svq->vdev->priv;
 145
 146         /* Suppress further interrupts. */
 147         virtqueue_disable_cb(svq);
 148
 149         /* We were probably waiting for more output buffers. */
 150         netif_wake_queue(vi->dev);
 151 }
 152
 153 static void set_skb_frag(struct sk_buff *skb, struct page *page,
 154                          unsigned int offset, unsigned int *len)
 155 {
 156         int size = min((unsigned)PAGE_SIZE - offset, *len);
 157         int i = skb_shinfo(skb)->nr_frags;
 158
 159         __skb_fill_page_desc(skb, i, page, offset, size);
 160
 161         skb->data_len += size;
 162         skb->len += size;
 163         skb->truesize += PAGE_SIZE;
 164         skb_shinfo(skb)->nr_frags++;
 165         *len -= size;
 166 }
 167
 168 /* Called from bottom half context */
 169 static struct sk_buff *page_to_skb(struct virtnet_info *vi,
 170                                    struct page *page, unsigned int len)
 171 {
 172         struct sk_buff *skb;
 173         struct skb_vnet_hdr *hdr;
 174         unsigned int copy, hdr_len, offset;
 175         char *p;
 176
 177         p = page_address(page);
 178
 179         /* copy small packet so we can reuse these pages for small data */
 180         skb = netdev_alloc_skb_ip_align(vi->dev, GOOD_COPY_LEN);
 181         if (unlikely(!skb))
 182                 return NULL;
 183
 184         hdr = skb_vnet_hdr(skb);
 185
 186         if (vi->mergeable_rx_bufs) {
 187                 hdr_len = sizeof hdr->mhdr;
 188                 offset = hdr_len;
 189         } else {
 190                 hdr_len = sizeof hdr->hdr;
 191                 offset = sizeof(struct padded_vnet_hdr);
 192         }
 193
 194         memcpy(hdr, p, hdr_len);
 195
 196         len -= hdr_len;
 197         p += offset;
 198
 199         copy = len;
 200         if (copy > skb_tailroom(skb))
 201                 copy = skb_tailroom(skb);
 202         memcpy(skb_put(skb, copy), p, copy);
 203
 204         len -= copy;
 205         offset += copy;
 206
 207         /*
 208          * Verify that we can indeed put this data into a skb.
 209          * This is here to handle cases when the device erroneously
 210          * tries to receive more than is possible. This is usually
 211          * the case of a broken device.
 212          */
 213         if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) {
 214                 if (net_ratelimit())
 215                         pr_debug("%s: too much data\n", skb->dev->name);
 216                 dev_kfree_skb(skb);
 217                 return NULL;
 218         }
 219
 220         while (len) {
 221                 set_skb_frag(skb, page, offset, &len);
 222                 page = (struct page *)page->private;
 223                 offset = 0;
 224         }
 225
 226         if (page)
 227                 give_pages(vi, page);
 228
 229         return skb;
 230 }
 231
 232 static int receive_mergeable(struct virtnet_info *vi, struct sk_buff *skb)
 233 {
 234         struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
 235         struct page *page;
 236         int num_buf, i, len;
 237
 238         num_buf = hdr->mhdr.num_buffers;
 239         while (--num_buf) {
 240                 i = skb_shinfo(skb)->nr_frags;
 241                 if (i >= MAX_SKB_FRAGS) {
 242                         pr_debug("%s: packet too long\n", skb->dev->name);
 243                         skb->dev->stats.rx_length_errors++;
 244                         return -EINVAL;
 245                 }
 246                 page = virtqueue_get_buf(vi->rvq, &len);
 247                 if (!page) {
 248                         pr_debug("%s: rx error: %d buffers missing\n",
 249                                  skb->dev->name, hdr->mhdr.num_buffers);
 250                         skb->dev->stats.rx_length_errors++;
 251                         return -EINVAL;
 252                 }
 253
 254                 if (len > PAGE_SIZE)
 255                         len = PAGE_SIZE;
 256
 257                 set_skb_frag(skb, page, 0, &len);
 258
 259                 --vi->num;
 260         }
 261         return 0;
 262 }
 263
 264 static void receive_buf(struct net_device *dev, void *buf, unsigned int len)
 265 {
 266         struct virtnet_info *vi = netdev_priv(dev);
 267         struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
 268         struct sk_buff *skb;
 269         struct page *page;
 270         struct skb_vnet_hdr *hdr;
 271
 272         if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) {
 273                 pr_debug("%s: short packet %i\n", dev->name, len);
 274                 dev->stats.rx_length_errors++;
 275                 if (vi->mergeable_rx_bufs || vi->big_packets)
 276                         give_pages(vi, buf);
 277                 else
 278                         dev_kfree_skb(buf);
 279                 return;
 280         }
 281
 282         if (!vi->mergeable_rx_bufs && !vi->big_packets) {
 283                 skb = buf;
 284                 len -= sizeof(struct virtio_net_hdr);
 285                 skb_trim(skb, len);
 286         } else {
 287                 page = buf;
 288                 skb = page_to_skb(vi, page, len);
 289                 if (unlikely(!skb)) {
 290                         dev->stats.rx_dropped++;
 291                         give_pages(vi, page);
 292                         return;
 293                 }
 294                 if (vi->mergeable_rx_bufs)
 295                         if (receive_mergeable(vi, skb)) {
 296                                 dev_kfree_skb(skb);
 297                                 return;
 298                         }
 299         }
 300
 301         hdr = skb_vnet_hdr(skb);
 302
 303         u64_stats_update_begin(&stats->syncp);
 304         stats->rx_bytes += skb->len;
 305         stats->rx_packets++;
 306         u64_stats_update_end(&stats->syncp);
 307
 308         if (hdr->hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
 309                 pr_debug("Needs csum!\n");
 310                 if (!skb_partial_csum_set(skb,
 311                                           hdr->hdr.csum_start,
 312                                           hdr->hdr.csum_offset))
 313                         goto frame_err;
 314         } else if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID) {
 315                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 316         }
 317
 318         skb->protocol = eth_type_trans(skb, dev);
 319         pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
 320                  ntohs(skb->protocol), skb->len, skb->pkt_type);
 321
 322         if (hdr->hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
 323                 pr_debug("GSO!\n");
 324                 switch (hdr->hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
 325                 case VIRTIO_NET_HDR_GSO_TCPV4:
 326                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
 327                         break;
 328                 case VIRTIO_NET_HDR_GSO_UDP:
 329                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
 330                         break;
 331                 case VIRTIO_NET_HDR_GSO_TCPV6:
 332                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
 333                         break;
 334                 default:
 335                         if (net_ratelimit())
 336                                 printk(KERN_WARNING "%s: bad gso type %u.\n",
 337                                        dev->name, hdr->hdr.gso_type);
 338                         goto frame_err;
 339                 }
 340
 341                 if (hdr->hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
 342                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
 343
 344                 skb_shinfo(skb)->gso_size = hdr->hdr.gso_size;
 345                 if (skb_shinfo(skb)->gso_size == 0) {
 346                         if (net_ratelimit())
 347                                 printk(KERN_WARNING "%s: zero gso size.\n",
 348                                        dev->name);
 349                         goto frame_err;
 350                 }
 351
 352                 /* Header must be checked, and gso_segs computed. */
 353                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
 354                 skb_shinfo(skb)->gso_segs = 0;
 355         }
 356
 357         netif_receive_skb(skb);
 358         return;
 359
 360 frame_err:
 361         dev->stats.rx_frame_errors++;
 362         dev_kfree_skb(skb);
 363 }
 364
 365 static int add_recvbuf_small(struct virtnet_info *vi, gfp_t gfp)
 366 {
 367         struct sk_buff *skb;
 368         struct skb_vnet_hdr *hdr;
 369         int err;
 370
 371         skb = __netdev_alloc_skb_ip_align(vi->dev, MAX_PACKET_LEN, gfp);
 372         if (unlikely(!skb))
 373                 return -ENOMEM;
 374
 375         skb_put(skb, MAX_PACKET_LEN);
 376
 377         hdr = skb_vnet_hdr(skb);
 378         sg_set_buf(vi->rx_sg, &hdr->hdr, sizeof hdr->hdr);
 379
 380         skb_to_sgvec(skb, vi->rx_sg + 1, 0, skb->len);
 381
 382         err = virtqueue_add_buf(vi->rvq, vi->rx_sg, 0, 2, skb, gfp);
 383         if (err < 0)
 384                 dev_kfree_skb(skb);
 385
 386         return err;
 387 }
 388
 389 static int add_recvbuf_big(struct virtnet_info *vi, gfp_t gfp)
 390 {
 391         struct page *first, *list = NULL;
 392         char *p;
 393         int i, err, offset;
 394
 395         /* page in vi->rx_sg[MAX_SKB_FRAGS + 1] is list tail */
 396         for (i = MAX_SKB_FRAGS + 1; i > 1; --i) {
 397                 first = get_a_page(vi, gfp);
 398                 if (!first) {
 399                         if (list)
 400                                 give_pages(vi, list);
 401                         return -ENOMEM;
 402                 }
 403                 sg_set_buf(&vi->rx_sg[i], page_address(first), PAGE_SIZE);
 404
 405                 /* chain new page in list head to match sg */
 406                 first->private = (unsigned long)list;
 407                 list = first;
 408         }
 409
 410         first = get_a_page(vi, gfp);
 411         if (!first) {
 412                 give_pages(vi, list);
 413                 return -ENOMEM;
 414         }
 415         p = page_address(first);
 416
 417         /* vi->rx_sg[0], vi->rx_sg[1] share the same page */
 418         /* a separated vi->rx_sg[0] for virtio_net_hdr only due to QEMU bug */
 419         sg_set_buf(&vi->rx_sg[0], p, sizeof(struct virtio_net_hdr));
 420
 421         /* vi->rx_sg[1] for data packet, from offset */
 422         offset = sizeof(struct padded_vnet_hdr);
 423         sg_set_buf(&vi->rx_sg[1], p + offset, PAGE_SIZE - offset);
 424
 425         /* chain first in list head */
 426         first->private = (unsigned long)list;
 427         err = virtqueue_add_buf(vi->rvq, vi->rx_sg, 0, MAX_SKB_FRAGS + 2,
 428                                 first, gfp);
 429         if (err < 0)
 430                 give_pages(vi, first);
 431
 432         return err;
 433 }
 434
 435 static int add_recvbuf_mergeable(struct virtnet_info *vi, gfp_t gfp)
 436 {
 437         struct page *page;
 438         int err;
 439
 440         page = get_a_page(vi, gfp);
 441         if (!page)
 442                 return -ENOMEM;
 443
 444         sg_init_one(vi->rx_sg, page_address(page), PAGE_SIZE);
 445
 446         err = virtqueue_add_buf(vi->rvq, vi->rx_sg, 0, 1, page, gfp);
 447         if (err < 0)
 448                 give_pages(vi, page);
 449
 450         return err;
 451 }
 452
 453 /*
 454  * Returns false if we couldn't fill entirely (OOM).
 455  *
 456  * Normally run in the receive path, but can also be run from ndo_open
 457  * before we're receiving packets, or from refill_work which is
 458  * careful to disable receiving (using napi_disable).
 459  */
 460 static bool try_fill_recv(struct virtnet_info *vi, gfp_t gfp)
 461 {
 462         int err;
 463         bool oom;
 464
 465         do {
 466                 if (vi->mergeable_rx_bufs)
 467                         err = add_recvbuf_mergeable(vi, gfp);
 468                 else if (vi->big_packets)
 469                         err = add_recvbuf_big(vi, gfp);
 470                 else
 471                         err = add_recvbuf_small(vi, gfp);
 472
 473                 oom = err == -ENOMEM;
 474                 if (err < 0)
 475                         break;
 476                 ++vi->num;
 477         } while (err > 0);
 478         if (unlikely(vi->num > vi->max))
 479                 vi->max = vi->num;
 480         virtqueue_kick(vi->rvq);
 481         return !oom;
 482 }
 483
 484 static void skb_recv_done(struct virtqueue *rvq)
 485 {
 486         struct virtnet_info *vi = rvq->vdev->priv;
 487         /* Schedule NAPI, Suppress further interrupts if successful. */
 488         if (napi_schedule_prep(&vi->napi)) {
 489                 virtqueue_disable_cb(rvq);
 490                 __napi_schedule(&vi->napi);
 491         }
 492 }
 493
 494 static void virtnet_napi_enable(struct virtnet_info *vi)
 495 {
 496         napi_enable(&vi->napi);
 497
 498         /* If all buffers were filled by other side before we napi_enabled, we
 499          * won't get another interrupt, so process any outstanding packets
 500          * now.  virtnet_poll wants re-enable the queue, so we disable here.
 501          * We synchronize against interrupts via NAPI_STATE_SCHED */
 502         if (napi_schedule_prep(&vi->napi)) {
 503                 virtqueue_disable_cb(vi->rvq);
 504                 local_bh_disable();
 505                 __napi_schedule(&vi->napi);
 506                 local_bh_enable();
 507         }
 508 }
 509
 510 static void refill_work(struct work_struct *work)
 511 {
 512         struct virtnet_info *vi;
 513         bool still_empty;
 514
 515         vi = container_of(work, struct virtnet_info, refill.work);
 516         napi_disable(&vi->napi);
 517         still_empty = !try_fill_recv(vi, GFP_KERNEL);
 518         virtnet_napi_enable(vi);
 519
 520         /* In theory, this can happen: if we don't get any buffers in
 521          * we will *never* try to fill again. */
 522         if (still_empty)
 523                 queue_delayed_work(system_nrt_wq, &vi->refill, HZ/2);
 524 }
 525
 526 static int virtnet_poll(struct napi_struct *napi, int budget)
 527 {
 528         struct virtnet_info *vi = container_of(napi, struct virtnet_info, napi);
 529         void *buf;
 530         unsigned int len, received = 0;
 531
 532 again:
 533         while (received < budget &&
 534                (buf = virtqueue_get_buf(vi->rvq, &len)) != NULL) {
 535                 receive_buf(vi->dev, buf, len);
 536                 --vi->num;
 537                 received++;
 538         }
 539
 540         if (vi->num < vi->max / 2) {
 541                 if (!try_fill_recv(vi, GFP_ATOMIC))
 542                         queue_delayed_work(system_nrt_wq, &vi->refill, 0);
 543         }
 544
 545         /* Out of packets? */
 546         if (received < budget) {
 547                 napi_complete(napi);
 548                 if (unlikely(!virtqueue_enable_cb(vi->rvq)) &&
 549                     napi_schedule_prep(napi)) {
 550                         virtqueue_disable_cb(vi->rvq);
 551                         __napi_schedule(napi);
 552                         goto again;
 553                 }
 554         }
 555
 556         return received;
 557 }
 558
 559 static unsigned int free_old_xmit_skbs(struct virtnet_info *vi)
 560 {
 561         struct sk_buff *skb;
 562         unsigned int len, tot_sgs = 0;
 563         struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
 564
 565         while ((skb = virtqueue_get_buf(vi->svq, &len)) != NULL) {
 566                 pr_debug("Sent skb %p\n", skb);
 567
 568                 u64_stats_update_begin(&stats->syncp);
 569                 stats->tx_bytes += skb->len;
 570                 stats->tx_packets++;
 571                 u64_stats_update_end(&stats->syncp);
 572
 573                 tot_sgs += skb_vnet_hdr(skb)->num_sg;
 574                 dev_kfree_skb_any(skb);
 575         }
 576         return tot_sgs;
 577 }
 578
 579 static int xmit_skb(struct virtnet_info *vi, struct sk_buff *skb)
 580 {
 581         struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
 582         const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
 583
 584         pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
 585
 586         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 587                 hdr->hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
 588                 hdr->hdr.csum_start = skb_checksum_start_offset(skb);
 589                 hdr->hdr.csum_offset = skb->csum_offset;
 590         } else {
 591                 hdr->hdr.flags = 0;
 592                 hdr->hdr.csum_offset = hdr->hdr.csum_start = 0;
 593         }
 594
 595         if (skb_is_gso(skb)) {
 596                 hdr->hdr.hdr_len = skb_headlen(skb);
 597                 hdr->hdr.gso_size = skb_shinfo(skb)->gso_size;
 598                 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
 599                         hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
 600                 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
 601                         hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
 602                 else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 603                         hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
 604                 else
 605                         BUG();
 606                 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
 607                         hdr->hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
 608         } else {
 609                 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
 610                 hdr->hdr.gso_size = hdr->hdr.hdr_len = 0;
 611         }
 612
 613         hdr->mhdr.num_buffers = 0;
 614
 615         /* Encode metadata header at front. */
 616         if (vi->mergeable_rx_bufs)
 617                 sg_set_buf(vi->tx_sg, &hdr->mhdr, sizeof hdr->mhdr);
 618         else
 619                 sg_set_buf(vi->tx_sg, &hdr->hdr, sizeof hdr->hdr);
 620
 621         hdr->num_sg = skb_to_sgvec(skb, vi->tx_sg + 1, 0, skb->len) + 1;
 622         return virtqueue_add_buf(vi->svq, vi->tx_sg, hdr->num_sg,
 623                                  0, skb, GFP_ATOMIC);
 624 }
 625
 626 static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
 627 {
 628         struct virtnet_info *vi = netdev_priv(dev);
 629         int capacity;
 630
 631         /* Free up any pending old buffers before queueing new ones. */
 632         free_old_xmit_skbs(vi);
 633
 634         /* Try to transmit */
 635         capacity = xmit_skb(vi, skb);
 636
 637         /* This can happen with OOM and indirect buffers. */
 638         if (unlikely(capacity < 0)) {
 639                 if (likely(capacity == -ENOMEM)) {
 640                         if (net_ratelimit())
 641                                 dev_warn(&dev->dev,
 642                                          "TX queue failure: out of memory\n");
 643                 } else {
 644                         dev->stats.tx_fifo_errors++;
 645                         if (net_ratelimit())
 646                                 dev_warn(&dev->dev,
 647                                          "Unexpected TX queue failure: %d\n",
 648                                          capacity);
 649                 }
 650                 dev->stats.tx_dropped++;
 651                 kfree_skb(skb);
 652                 return NETDEV_TX_OK;
 653         }
 654         virtqueue_kick(vi->svq);
 655
 656         /* Don't wait up for transmitted skbs to be freed. */
 657         skb_orphan(skb);
 658         nf_reset(skb);
 659
 660         /* Apparently nice girls don't return TX_BUSY; stop the queue
 661          * before it gets out of hand.  Naturally, this wastes entries. */
 662         if (capacity < 2+MAX_SKB_FRAGS) {
 663                 netif_stop_queue(dev);
 664                 if (unlikely(!virtqueue_enable_cb_delayed(vi->svq))) {
 665                         /* More just got used, free them then recheck. */
 666                         capacity += free_old_xmit_skbs(vi);
 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(&stats->syncp);
 707                         tpackets = stats->tx_packets;
 708                         tbytes   = stats->tx_bytes;
 709                         rpackets = stats->rx_packets;
 710                         rbytes   = stats->rx_bytes;
 711                 } while (u64_stats_fetch_retry(&stats->syncp, start));
 712
 713                 tot->rx_packets += rpackets;
 714                 tot->tx_packets += tpackets;
 715                 tot->rx_bytes   += rbytes;
 716                 tot->tx_bytes   += tbytes;
 717         }
 718
 719         tot->tx_dropped = dev->stats.tx_dropped;
 720         tot->tx_fifo_errors = dev->stats.tx_fifo_errors;
 721         tot->rx_dropped = dev->stats.rx_dropped;
 722         tot->rx_length_errors = dev->stats.rx_length_errors;
 723         tot->rx_frame_errors = dev->stats.rx_frame_errors;
 724
 725         return tot;
 726 }
 727
 728 #ifdef CONFIG_NET_POLL_CONTROLLER
 729 static void virtnet_netpoll(struct net_device *dev)
 730 {
 731         struct virtnet_info *vi = netdev_priv(dev);
 732
 733         napi_schedule(&vi->napi);
 734 }
 735 #endif
 736
 737 static int virtnet_open(struct net_device *dev)
 738 {
 739         struct virtnet_info *vi = netdev_priv(dev);
 740
 741         /* Make sure we have some buffers: if oom use wq. */
 742         if (!try_fill_recv(vi, GFP_KERNEL))
 743                 queue_delayed_work(system_nrt_wq, &vi->refill, 0);
 744
 745         virtnet_napi_enable(vi);
 746         return 0;
 747 }
 748
 749 /*
 750  * Send command via the control virtqueue and check status.  Commands
 751  * supported by the hypervisor, as indicated by feature bits, should
 752  * never fail unless improperly formated.
 753  */
 754 static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
 755                                  struct scatterlist *data, int out, int in)
 756 {
 757         struct scatterlist *s, sg[VIRTNET_SEND_COMMAND_SG_MAX + 2];
 758         struct virtio_net_ctrl_hdr ctrl;
 759         virtio_net_ctrl_ack status = ~0;
 760         unsigned int tmp;
 761         int i;
 762
 763         /* Caller should know better */
 764         BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ||
 765                 (out + in > VIRTNET_SEND_COMMAND_SG_MAX));
 766
 767         out++; /* Add header */
 768         in++; /* Add return status */
 769
 770         ctrl.class = class;
 771         ctrl.cmd = cmd;
 772
 773         sg_init_table(sg, out + in);
 774
 775         sg_set_buf(&sg[0], &ctrl, sizeof(ctrl));
 776         for_each_sg(data, s, out + in - 2, i)
 777                 sg_set_buf(&sg[i + 1], sg_virt(s), s->length);
 778         sg_set_buf(&sg[out + in - 1], &status, sizeof(status));
 779
 780         BUG_ON(virtqueue_add_buf(vi->cvq, sg, out, in, vi, GFP_ATOMIC) < 0);
 781
 782         virtqueue_kick(vi->cvq);
 783
 784         /*
 785          * Spin for a response, the kick causes an ioport write, trapping
 786          * into the hypervisor, so the request should be handled immediately.
 787          */
 788         while (!virtqueue_get_buf(vi->cvq, &tmp))
 789                 cpu_relax();
 790
 791         return status == VIRTIO_NET_OK;
 792 }
 793
 794 static void virtnet_ack_link_announce(struct virtnet_info *vi)
 795 {
 796         rtnl_lock();
 797         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE,
 798                                   VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL,
 799                                   0, 0))
 800                 dev_warn(&vi->dev->dev, "Failed to ack link announce.\n");
 801         rtnl_unlock();
 802 }
 803
 804 static int virtnet_close(struct net_device *dev)
 805 {
 806         struct virtnet_info *vi = netdev_priv(dev);
 807
 808         /* Make sure refill_work doesn't re-enable napi! */
 809         cancel_delayed_work_sync(&vi->refill);
 810         napi_disable(&vi->napi);
 811
 812         return 0;
 813 }
 814
 815 static void virtnet_set_rx_mode(struct net_device *dev)
 816 {
 817         struct virtnet_info *vi = netdev_priv(dev);
 818         struct scatterlist sg[2];
 819         u8 promisc, allmulti;
 820         struct virtio_net_ctrl_mac *mac_data;
 821         struct netdev_hw_addr *ha;
 822         int uc_count;
 823         int mc_count;
 824         void *buf;
 825         int i;
 826
 827         /* We can't dynamicaly set ndo_set_rx_mode, so return gracefully */
 828         if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
 829                 return;
 830
 831         promisc = ((dev->flags & IFF_PROMISC) != 0);
 832         allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
 833
 834         sg_init_one(sg, &promisc, sizeof(promisc));
 835
 836         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
 837                                   VIRTIO_NET_CTRL_RX_PROMISC,
 838                                   sg, 1, 0))
 839                 dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
 840                          promisc ? "en" : "dis");
 841
 842         sg_init_one(sg, &allmulti, sizeof(allmulti));
 843
 844         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
 845                                   VIRTIO_NET_CTRL_RX_ALLMULTI,
 846                                   sg, 1, 0))
 847                 dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
 848                          allmulti ? "en" : "dis");
 849
 850         uc_count = netdev_uc_count(dev);
 851         mc_count = netdev_mc_count(dev);
 852         /* MAC filter - use one buffer for both lists */
 853         buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) +
 854                       (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
 855         mac_data = buf;
 856         if (!buf) {
 857                 dev_warn(&dev->dev, "No memory for MAC address buffer\n");
 858                 return;
 859         }
 860
 861         sg_init_table(sg, 2);
 862
 863         /* Store the unicast list and count in the front of the buffer */
 864         mac_data->entries = uc_count;
 865         i = 0;
 866         netdev_for_each_uc_addr(ha, dev)
 867                 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
 868
 869         sg_set_buf(&sg[0], mac_data,
 870                    sizeof(mac_data->entries) + (uc_count * ETH_ALEN));
 871
 872         /* multicast list and count fill the end */
 873         mac_data = (void *)&mac_data->macs[uc_count][0];
 874
 875         mac_data->entries = mc_count;
 876         i = 0;
 877         netdev_for_each_mc_addr(ha, dev)
 878                 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
 879
 880         sg_set_buf(&sg[1], mac_data,
 881                    sizeof(mac_data->entries) + (mc_count * ETH_ALEN));
 882
 883         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
 884                                   VIRTIO_NET_CTRL_MAC_TABLE_SET,
 885                                   sg, 2, 0))
 886                 dev_warn(&dev->dev, "Failed to set MAC fitler table.\n");
 887
 888         kfree(buf);
 889 }
 890
 891 static int virtnet_vlan_rx_add_vid(struct net_device *dev, u16 vid)
 892 {
 893         struct virtnet_info *vi = netdev_priv(dev);
 894         struct scatterlist sg;
 895
 896         sg_init_one(&sg, &vid, sizeof(vid));
 897
 898         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
 899                                   VIRTIO_NET_CTRL_VLAN_ADD, &sg, 1, 0))
 900                 dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
 901         return 0;
 902 }
 903
 904 static int virtnet_vlan_rx_kill_vid(struct net_device *dev, u16 vid)
 905 {
 906         struct virtnet_info *vi = netdev_priv(dev);
 907         struct scatterlist sg;
 908
 909         sg_init_one(&sg, &vid, sizeof(vid));
 910
 911         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
 912                                   VIRTIO_NET_CTRL_VLAN_DEL, &sg, 1, 0))
 913                 dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
 914         return 0;
 915 }
 916
 917 static void virtnet_get_ringparam(struct net_device *dev,
 918                                 struct ethtool_ringparam *ring)
 919 {
 920         struct virtnet_info *vi = netdev_priv(dev);
 921
 922         ring->rx_max_pending = virtqueue_get_vring_size(vi->rvq);
 923         ring->tx_max_pending = virtqueue_get_vring_size(vi->svq);
 924         ring->rx_pending = ring->rx_max_pending;
 925         ring->tx_pending = ring->tx_max_pending;
 926
 927 }
 928
 929
 930 static void virtnet_get_drvinfo(struct net_device *dev,
 931                                 struct ethtool_drvinfo *info)
 932 {
 933         struct virtnet_info *vi = netdev_priv(dev);
 934         struct virtio_device *vdev = vi->vdev;
 935
 936         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
 937         strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version));
 938         strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info));
 939
 940 }
 941
 942 static const struct ethtool_ops virtnet_ethtool_ops = {
 943         .get_drvinfo = virtnet_get_drvinfo,
 944         .get_link = ethtool_op_get_link,
 945         .get_ringparam = virtnet_get_ringparam,
 946 };
 947
 948 #define MIN_MTU 68
 949 #define MAX_MTU 65535
 950
 951 static int virtnet_change_mtu(struct net_device *dev, int new_mtu)
 952 {
 953         if (new_mtu < MIN_MTU || new_mtu > MAX_MTU)
 954                 return -EINVAL;
 955         dev->mtu = new_mtu;
 956         return 0;
 957 }
 958
 959 static const struct net_device_ops virtnet_netdev = {
 960         .ndo_open            = virtnet_open,
 961         .ndo_stop            = virtnet_close,
 962         .ndo_start_xmit      = start_xmit,
 963         .ndo_validate_addr   = eth_validate_addr,
 964         .ndo_set_mac_address = virtnet_set_mac_address,
 965         .ndo_set_rx_mode     = virtnet_set_rx_mode,
 966         .ndo_change_mtu      = virtnet_change_mtu,
 967         .ndo_get_stats64     = virtnet_stats,
 968         .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
 969         .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
 970 #ifdef CONFIG_NET_POLL_CONTROLLER
 971         .ndo_poll_controller = virtnet_netpoll,
 972 #endif
 973 };
 974
 975 static void virtnet_config_changed_work(struct work_struct *work)
 976 {
 977         struct virtnet_info *vi =
 978                 container_of(work, struct virtnet_info, config_work);
 979         u16 v;
 980
 981         mutex_lock(&vi->config_lock);
 982         if (!vi->config_enable)
 983                 goto done;
 984
 985         if (virtio_config_val(vi->vdev, VIRTIO_NET_F_STATUS,
 986                               offsetof(struct virtio_net_config, status),
 987                               &v) < 0)
 988                 goto done;
 989
 990         if (v & VIRTIO_NET_S_ANNOUNCE) {
 991                 netif_notify_peers(vi->dev);
 992                 virtnet_ack_link_announce(vi);
 993         }
 994
 995         /* Ignore unknown (future) status bits */
 996         v &= VIRTIO_NET_S_LINK_UP;
 997
 998         if (vi->status == v)
 999                 goto done;
1000
1001         vi->status = v;
1002
1003         if (vi->status & VIRTIO_NET_S_LINK_UP) {
1004                 netif_carrier_on(vi->dev);
1005                 netif_wake_queue(vi->dev);
1006         } else {
1007                 netif_carrier_off(vi->dev);
1008                 netif_stop_queue(vi->dev);
1009         }
1010 done:
1011         mutex_unlock(&vi->config_lock);
1012 }
1013
1014 static void virtnet_config_changed(struct virtio_device *vdev)
1015 {
1016         struct virtnet_info *vi = vdev->priv;
1017
1018         queue_work(system_nrt_wq, &vi->config_work);
1019 }
1020
1021 static int init_vqs(struct virtnet_info *vi)
1022 {
1023         struct virtqueue *vqs[3];
1024         vq_callback_t *callbacks[] = { skb_recv_done, skb_xmit_done, NULL};
1025         const char *names[] = { "input", "output", "control" };
1026         int nvqs, err;
1027
1028         /* We expect two virtqueues, receive then send,
1029          * and optionally control. */
1030         nvqs = virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ? 3 : 2;
1031
1032         err = vi->vdev->config->find_vqs(vi->vdev, nvqs, vqs, callbacks, names);
1033         if (err)
1034                 return err;
1035
1036         vi->rvq = vqs[0];
1037         vi->svq = vqs[1];
1038
1039         if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)) {
1040                 vi->cvq = vqs[2];
1041
1042                 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
1043                         vi->dev->features |= NETIF_F_HW_VLAN_FILTER;
1044         }
1045         return 0;
1046 }
1047
1048 static int virtnet_probe(struct virtio_device *vdev)
1049 {
1050         int err;
1051         struct net_device *dev;
1052         struct virtnet_info *vi;
1053
1054         /* Allocate ourselves a network device with room for our info */
1055         dev = alloc_etherdev(sizeof(struct virtnet_info));
1056         if (!dev)
1057                 return -ENOMEM;
1058
1059         /* Set up network device as normal. */
1060         dev->priv_flags |= IFF_UNICAST_FLT;
1061         dev->netdev_ops = &virtnet_netdev;
1062         dev->features = NETIF_F_HIGHDMA;
1063
1064         SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops);
1065         SET_NETDEV_DEV(dev, &vdev->dev);
1066
1067         /* Do we support "hardware" checksums? */
1068         if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
1069                 /* This opens up the world of extra features. */
1070                 dev->hw_features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
1071                 if (csum)
1072                         dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
1073
1074                 if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
1075                         dev->hw_features |= NETIF_F_TSO | NETIF_F_UFO
1076                                 | NETIF_F_TSO_ECN | NETIF_F_TSO6;
1077                 }
1078                 /* Individual feature bits: what can host handle? */
1079                 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
1080                         dev->hw_features |= NETIF_F_TSO;
1081                 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
1082                         dev->hw_features |= NETIF_F_TSO6;
1083                 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
1084                         dev->hw_features |= NETIF_F_TSO_ECN;
1085                 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
1086                         dev->hw_features |= NETIF_F_UFO;
1087
1088                 if (gso)
1089                         dev->features |= dev->hw_features & (NETIF_F_ALL_TSO|NETIF_F_UFO);
1090                 /* (!csum && gso) case will be fixed by register_netdev() */
1091         }
1092
1093         /* Configuration may specify what MAC to use.  Otherwise random. */
1094         if (virtio_config_val_len(vdev, VIRTIO_NET_F_MAC,
1095                                   offsetof(struct virtio_net_config, mac),
1096                                   dev->dev_addr, dev->addr_len) < 0)
1097                 eth_hw_addr_random(dev);
1098
1099         /* Set up our device-specific information */
1100         vi = netdev_priv(dev);
1101         netif_napi_add(dev, &vi->napi, virtnet_poll, napi_weight);
1102         vi->dev = dev;
1103         vi->vdev = vdev;
1104         vdev->priv = vi;
1105         vi->pages = NULL;
1106         vi->stats = alloc_percpu(struct virtnet_stats);
1107         err = -ENOMEM;
1108         if (vi->stats == NULL)
1109                 goto free;
1110
1111         INIT_DELAYED_WORK(&vi->refill, refill_work);
1112         mutex_init(&vi->config_lock);
1113         vi->config_enable = true;
1114         INIT_WORK(&vi->config_work, virtnet_config_changed_work);
1115         sg_init_table(vi->rx_sg, ARRAY_SIZE(vi->rx_sg));
1116         sg_init_table(vi->tx_sg, ARRAY_SIZE(vi->tx_sg));
1117
1118         /* If we can receive ANY GSO packets, we must allocate large ones. */
1119         if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
1120             virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) ||
1121             virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN))
1122                 vi->big_packets = true;
1123
1124         if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
1125                 vi->mergeable_rx_bufs = true;
1126
1127         err = init_vqs(vi);
1128         if (err)
1129                 goto free_stats;
1130
1131         err = register_netdev(dev);
1132         if (err) {
1133                 pr_debug("virtio_net: registering device failed\n");
1134                 goto free_vqs;
1135         }
1136
1137         /* Last of all, set up some receive buffers. */
1138         try_fill_recv(vi, GFP_KERNEL);
1139
1140         /* If we didn't even get one input buffer, we're useless. */
1141         if (vi->num == 0) {
1142                 err = -ENOMEM;
1143                 goto unregister;
1144         }
1145
1146         /* Assume link up if device can't report link status,
1147            otherwise get link status from config. */
1148         if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) {
1149                 netif_carrier_off(dev);
1150                 queue_work(system_nrt_wq, &vi->config_work);
1151         } else {
1152                 vi->status = VIRTIO_NET_S_LINK_UP;
1153                 netif_carrier_on(dev);
1154         }
1155
1156         pr_debug("virtnet: registered device %s\n", dev->name);
1157         return 0;
1158
1159 unregister:
1160         unregister_netdev(dev);
1161 free_vqs:
1162         vdev->config->del_vqs(vdev);
1163 free_stats:
1164         free_percpu(vi->stats);
1165 free:
1166         free_netdev(dev);
1167         return err;
1168 }
1169
1170 static void free_unused_bufs(struct virtnet_info *vi)
1171 {
1172         void *buf;
1173         while (1) {
1174                 buf = virtqueue_detach_unused_buf(vi->svq);
1175                 if (!buf)
1176                         break;
1177                 dev_kfree_skb(buf);
1178         }
1179         while (1) {
1180                 buf = virtqueue_detach_unused_buf(vi->rvq);
1181                 if (!buf)
1182                         break;
1183                 if (vi->mergeable_rx_bufs || vi->big_packets)
1184                         give_pages(vi, buf);
1185                 else
1186                         dev_kfree_skb(buf);
1187                 --vi->num;
1188         }
1189         BUG_ON(vi->num != 0);
1190 }
1191
1192 static void remove_vq_common(struct virtnet_info *vi)
1193 {
1194         vi->vdev->config->reset(vi->vdev);
1195
1196         /* Free unused buffers in both send and recv, if any. */
1197         free_unused_bufs(vi);
1198
1199         vi->vdev->config->del_vqs(vi->vdev);
1200
1201         while (vi->pages)
1202                 __free_pages(get_a_page(vi, GFP_KERNEL), 0);
1203 }
1204
1205 static void __devexit virtnet_remove(struct virtio_device *vdev)
1206 {
1207         struct virtnet_info *vi = vdev->priv;
1208
1209         /* Prevent config work handler from accessing the device. */
1210         mutex_lock(&vi->config_lock);
1211         vi->config_enable = false;
1212         mutex_unlock(&vi->config_lock);
1213
1214         unregister_netdev(vi->dev);
1215
1216         remove_vq_common(vi);
1217
1218         flush_work(&vi->config_work);
1219
1220         free_percpu(vi->stats);
1221         free_netdev(vi->dev);
1222 }
1223
1224 #ifdef CONFIG_PM
1225 static int virtnet_freeze(struct virtio_device *vdev)
1226 {
1227         struct virtnet_info *vi = vdev->priv;
1228
1229         /* Prevent config work handler from accessing the device */
1230         mutex_lock(&vi->config_lock);
1231         vi->config_enable = false;
1232         mutex_unlock(&vi->config_lock);
1233
1234         virtqueue_disable_cb(vi->rvq);
1235         virtqueue_disable_cb(vi->svq);
1236         if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ))
1237                 virtqueue_disable_cb(vi->cvq);
1238
1239         netif_device_detach(vi->dev);
1240         cancel_delayed_work_sync(&vi->refill);
1241
1242         if (netif_running(vi->dev))
1243                 napi_disable(&vi->napi);
1244
1245         remove_vq_common(vi);
1246
1247         flush_work(&vi->config_work);
1248
1249         return 0;
1250 }
1251
1252 static int virtnet_restore(struct virtio_device *vdev)
1253 {
1254         struct virtnet_info *vi = vdev->priv;
1255         int err;
1256
1257         err = init_vqs(vi);
1258         if (err)
1259                 return err;
1260
1261         if (netif_running(vi->dev))
1262                 virtnet_napi_enable(vi);
1263
1264         netif_device_attach(vi->dev);
1265
1266         if (!try_fill_recv(vi, GFP_KERNEL))
1267                 queue_delayed_work(system_nrt_wq, &vi->refill, 0);
1268
1269         mutex_lock(&vi->config_lock);
1270         vi->config_enable = true;
1271         mutex_unlock(&vi->config_lock);
1272
1273         return 0;
1274 }
1275 #endif
1276
1277 static struct virtio_device_id id_table[] = {
1278         { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
1279         { 0 },
1280 };
1281
1282 static unsigned int features[] = {
1283         VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM,
1284         VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC,
1285         VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6,
1286         VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6,
1287         VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO,
1288         VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ,
1289         VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN,
1290         VIRTIO_NET_F_GUEST_ANNOUNCE,
1291 };
1292
1293 static struct virtio_driver virtio_net_driver = {
1294         .feature_table = features,
1295         .feature_table_size = ARRAY_SIZE(features),
1296         .driver.name =  KBUILD_MODNAME,
1297         .driver.owner = THIS_MODULE,
1298         .id_table =     id_table,
1299         .probe =        virtnet_probe,
1300         .remove =       __devexit_p(virtnet_remove),
1301         .config_changed = virtnet_config_changed,
1302 #ifdef CONFIG_PM
1303         .freeze =       virtnet_freeze,
1304         .restore =      virtnet_restore,
1305 #endif
1306 };
1307
1308 static int __init init(void)
1309 {
1310         return register_virtio_driver(&virtio_net_driver);
1311 }
1312
1313 static void __exit fini(void)
1314 {
1315         unregister_virtio_driver(&virtio_net_driver);
1316 }
1317 module_init(init);
1318 module_exit(fini);
1319
1320 MODULE_DEVICE_TABLE(virtio, id_table);
1321 MODULE_DESCRIPTION("Virtio network driver");
1322 MODULE_LICENSE("GPL");