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, see <http://www.gnu.org/licenses/>.
  17  */
  18 //#define DEBUG
  19 #include <linux/netdevice.h>
  20 #include <linux/etherdevice.h>
  21 #include <linux/ethtool.h>
  22 #include <linux/module.h>
  23 #include <linux/virtio.h>
  24 #include <linux/virtio_net.h>
  25 #include <linux/bpf.h>
  26 #include <linux/scatterlist.h>
  27 #include <linux/if_vlan.h>
  28 #include <linux/slab.h>
  29 #include <linux/cpu.h>
  30 #include <linux/average.h>
  31 #include <net/busy_poll.h>
  32
  33 static int napi_weight = NAPI_POLL_WEIGHT;
  34 module_param(napi_weight, int, 0444);
  35
  36 static bool csum = true, gso = true;
  37 module_param(csum, bool, 0444);
  38 module_param(gso, bool, 0444);
  39
  40 /* FIXME: MTU in config. */
  41 #define GOOD_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
  42 #define GOOD_COPY_LEN   128
  43
  44 /* RX packet size EWMA. The average packet size is used to determine the packet
  45  * buffer size when refilling RX rings. As the entire RX ring may be refilled
  46  * at once, the weight is chosen so that the EWMA will be insensitive to short-
  47  * term, transient changes in packet size.
  48  */
  49 DECLARE_EWMA(pkt_len, 1, 64)
  50
  51 /* Minimum alignment for mergeable packet buffers. */
  52 #define MERGEABLE_BUFFER_ALIGN max(L1_CACHE_BYTES, 256)
  53
  54 #define VIRTNET_DRIVER_VERSION "1.0.0"
  55
  56 struct virtnet_stats {
  57         struct u64_stats_sync tx_syncp;
  58         struct u64_stats_sync rx_syncp;
  59         u64 tx_bytes;
  60         u64 tx_packets;
  61
  62         u64 rx_bytes;
  63         u64 rx_packets;
  64 };
  65
  66 /* Internal representation of a send virtqueue */
  67 struct send_queue {
  68         /* Virtqueue associated with this send _queue */
  69         struct virtqueue *vq;
  70
  71         /* TX: fragments + linear part + virtio header */
  72         struct scatterlist sg[MAX_SKB_FRAGS + 2];
  73
  74         /* Name of the send queue: output.$index */
  75         char name[40];
  76 };
  77
  78 /* Internal representation of a receive virtqueue */
  79 struct receive_queue {
  80         /* Virtqueue associated with this receive_queue */
  81         struct virtqueue *vq;
  82
  83         struct napi_struct napi;
  84
  85         struct bpf_prog __rcu *xdp_prog;
  86
  87         /* Chain pages by the private ptr. */
  88         struct page *pages;
  89
  90         /* Average packet length for mergeable receive buffers. */
  91         struct ewma_pkt_len mrg_avg_pkt_len;
  92
  93         /* Page frag for packet buffer allocation. */
  94         struct page_frag alloc_frag;
  95
  96         /* RX: fragments + linear part + virtio header */
  97         struct scatterlist sg[MAX_SKB_FRAGS + 2];
  98
  99         /* Name of this receive queue: input.$index */
 100         char name[40];
 101 };
 102
 103 struct virtnet_info {
 104         struct virtio_device *vdev;
 105         struct virtqueue *cvq;
 106         struct net_device *dev;
 107         struct send_queue *sq;
 108         struct receive_queue *rq;
 109         unsigned int status;
 110
 111         /* Max # of queue pairs supported by the device */
 112         u16 max_queue_pairs;
 113
 114         /* # of queue pairs currently used by the driver */
 115         u16 curr_queue_pairs;
 116
 117         /* # of XDP queue pairs currently used by the driver */
 118         u16 xdp_queue_pairs;
 119
 120         /* I like... big packets and I cannot lie! */
 121         bool big_packets;
 122
 123         /* Host will merge rx buffers for big packets (shake it! shake it!) */
 124         bool mergeable_rx_bufs;
 125
 126         /* Has control virtqueue */
 127         bool has_cvq;
 128
 129         /* Host can handle any s/g split between our header and packet data */
 130         bool any_header_sg;
 131
 132         /* Packet virtio header size */
 133         u8 hdr_len;
 134
 135         /* Active statistics */
 136         struct virtnet_stats __percpu *stats;
 137
 138         /* Work struct for refilling if we run low on memory. */
 139         struct delayed_work refill;
 140
 141         /* Work struct for config space updates */
 142         struct work_struct config_work;
 143
 144         /* Does the affinity hint is set for virtqueues? */
 145         bool affinity_hint_set;
 146
 147         /* CPU hotplug instances for online & dead */
 148         struct hlist_node node;
 149         struct hlist_node node_dead;
 150
 151         /* Control VQ buffers: protected by the rtnl lock */
 152         struct virtio_net_ctrl_hdr ctrl_hdr;
 153         virtio_net_ctrl_ack ctrl_status;
 154         struct virtio_net_ctrl_mq ctrl_mq;
 155         u8 ctrl_promisc;
 156         u8 ctrl_allmulti;
 157         u16 ctrl_vid;
 158
 159         /* Ethtool settings */
 160         u8 duplex;
 161         u32 speed;
 162 };
 163
 164 struct padded_vnet_hdr {
 165         struct virtio_net_hdr_mrg_rxbuf hdr;
 166         /*
 167          * hdr is in a separate sg buffer, and data sg buffer shares same page
 168          * with this header sg. This padding makes next sg 16 byte aligned
 169          * after the header.
 170          */
 171         char padding[4];
 172 };
 173
 174 /* Converting between virtqueue no. and kernel tx/rx queue no.
 175  * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq
 176  */
 177 static int vq2txq(struct virtqueue *vq)
 178 {
 179         return (vq->index - 1) / 2;
 180 }
 181
 182 static int txq2vq(int txq)
 183 {
 184         return txq * 2 + 1;
 185 }
 186
 187 static int vq2rxq(struct virtqueue *vq)
 188 {
 189         return vq->index / 2;
 190 }
 191
 192 static int rxq2vq(int rxq)
 193 {
 194         return rxq * 2;
 195 }
 196
 197 static inline struct virtio_net_hdr_mrg_rxbuf *skb_vnet_hdr(struct sk_buff *skb)
 198 {
 199         return (struct virtio_net_hdr_mrg_rxbuf *)skb->cb;
 200 }
 201
 202 /*
 203  * private is used to chain pages for big packets, put the whole
 204  * most recent used list in the beginning for reuse
 205  */
 206 static void give_pages(struct receive_queue *rq, struct page *page)
 207 {
 208         struct page *end;
 209
 210         /* Find end of list, sew whole thing into vi->rq.pages. */
 211         for (end = page; end->private; end = (struct page *)end->private);
 212         end->private = (unsigned long)rq->pages;
 213         rq->pages = page;
 214 }
 215
 216 static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask)
 217 {
 218         struct page *p = rq->pages;
 219
 220         if (p) {
 221                 rq->pages = (struct page *)p->private;
 222                 /* clear private here, it is used to chain pages */
 223                 p->private = 0;
 224         } else
 225                 p = alloc_page(gfp_mask);
 226         return p;
 227 }
 228
 229 static void skb_xmit_done(struct virtqueue *vq)
 230 {
 231         struct virtnet_info *vi = vq->vdev->priv;
 232
 233         /* Suppress further interrupts. */
 234         virtqueue_disable_cb(vq);
 235
 236         /* We were probably waiting for more output buffers. */
 237         netif_wake_subqueue(vi->dev, vq2txq(vq));
 238 }
 239
 240 static unsigned int mergeable_ctx_to_buf_truesize(unsigned long mrg_ctx)
 241 {
 242         unsigned int truesize = mrg_ctx & (MERGEABLE_BUFFER_ALIGN - 1);
 243         return (truesize + 1) * MERGEABLE_BUFFER_ALIGN;
 244 }
 245
 246 static void *mergeable_ctx_to_buf_address(unsigned long mrg_ctx)
 247 {
 248         return (void *)(mrg_ctx & -MERGEABLE_BUFFER_ALIGN);
 249
 250 }
 251
 252 static unsigned long mergeable_buf_to_ctx(void *buf, unsigned int truesize)
 253 {
 254         unsigned int size = truesize / MERGEABLE_BUFFER_ALIGN;
 255         return (unsigned long)buf | (size - 1);
 256 }
 257
 258 /* Called from bottom half context */
 259 static struct sk_buff *page_to_skb(struct virtnet_info *vi,
 260                                    struct receive_queue *rq,
 261                                    struct page *page, unsigned int offset,
 262                                    unsigned int len, unsigned int truesize)
 263 {
 264         struct sk_buff *skb;
 265         struct virtio_net_hdr_mrg_rxbuf *hdr;
 266         unsigned int copy, hdr_len, hdr_padded_len;
 267         char *p;
 268
 269         p = page_address(page) + offset;
 270
 271         /* copy small packet so we can reuse these pages for small data */
 272         skb = napi_alloc_skb(&rq->napi, GOOD_COPY_LEN);
 273         if (unlikely(!skb))
 274                 return NULL;
 275
 276         hdr = skb_vnet_hdr(skb);
 277
 278         hdr_len = vi->hdr_len;
 279         if (vi->mergeable_rx_bufs)
 280                 hdr_padded_len = sizeof *hdr;
 281         else
 282                 hdr_padded_len = sizeof(struct padded_vnet_hdr);
 283
 284         memcpy(hdr, p, hdr_len);
 285
 286         len -= hdr_len;
 287         offset += hdr_padded_len;
 288         p += hdr_padded_len;
 289
 290         copy = len;
 291         if (copy > skb_tailroom(skb))
 292                 copy = skb_tailroom(skb);
 293         memcpy(skb_put(skb, copy), p, copy);
 294
 295         len -= copy;
 296         offset += copy;
 297
 298         if (vi->mergeable_rx_bufs) {
 299                 if (len)
 300                         skb_add_rx_frag(skb, 0, page, offset, len, truesize);
 301                 else
 302                         put_page(page);
 303                 return skb;
 304         }
 305
 306         /*
 307          * Verify that we can indeed put this data into a skb.
 308          * This is here to handle cases when the device erroneously
 309          * tries to receive more than is possible. This is usually
 310          * the case of a broken device.
 311          */
 312         if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) {
 313                 net_dbg_ratelimited("%s: too much data\n", skb->dev->name);
 314                 dev_kfree_skb(skb);
 315                 return NULL;
 316         }
 317         BUG_ON(offset >= PAGE_SIZE);
 318         while (len) {
 319                 unsigned int frag_size = min((unsigned)PAGE_SIZE - offset, len);
 320                 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page, offset,
 321                                 frag_size, truesize);
 322                 len -= frag_size;
 323                 page = (struct page *)page->private;
 324                 offset = 0;
 325         }
 326
 327         if (page)
 328                 give_pages(rq, page);
 329
 330         return skb;
 331 }
 332
 333 static void virtnet_xdp_xmit(struct virtnet_info *vi,
 334                              struct receive_queue *rq,
 335                              struct send_queue *sq,
 336                              struct xdp_buff *xdp)
 337 {
 338         struct page *page = virt_to_head_page(xdp->data);
 339         struct virtio_net_hdr_mrg_rxbuf *hdr;
 340         unsigned int num_sg, len;
 341         void *xdp_sent;
 342         int err;
 343
 344         /* Free up any pending old buffers before queueing new ones. */
 345         while ((xdp_sent = virtqueue_get_buf(sq->vq, &len)) != NULL) {
 346                 struct page *sent_page = virt_to_head_page(xdp_sent);
 347
 348                 if (vi->mergeable_rx_bufs)
 349                         put_page(sent_page);
 350                 else
 351                         give_pages(rq, sent_page);
 352         }
 353
 354         /* Zero header and leave csum up to XDP layers */
 355         hdr = xdp->data;
 356         memset(hdr, 0, vi->hdr_len);
 357
 358         num_sg = 1;
 359         sg_init_one(sq->sg, xdp->data, xdp->data_end - xdp->data);
 360         err = virtqueue_add_outbuf(sq->vq, sq->sg, num_sg,
 361                                    xdp->data, GFP_ATOMIC);
 362         if (unlikely(err)) {
 363                 if (vi->mergeable_rx_bufs)
 364                         put_page(page);
 365                 else
 366                         give_pages(rq, page);
 367                 return; // On error abort to avoid unnecessary kick
 368         } else if (!vi->mergeable_rx_bufs) {
 369                 /* If not mergeable bufs must be big packets so cleanup pages */
 370                 give_pages(rq, (struct page *)page->private);
 371                 page->private = 0;
 372         }
 373
 374         virtqueue_kick(sq->vq);
 375 }
 376
 377 static u32 do_xdp_prog(struct virtnet_info *vi,
 378                        struct receive_queue *rq,
 379                        struct bpf_prog *xdp_prog,
 380                        struct page *page, int offset, int len)
 381 {
 382         int hdr_padded_len;
 383         struct xdp_buff xdp;
 384         unsigned int qp;
 385         u32 act;
 386         u8 *buf;
 387
 388         buf = page_address(page) + offset;
 389
 390         if (vi->mergeable_rx_bufs)
 391                 hdr_padded_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
 392         else
 393                 hdr_padded_len = sizeof(struct padded_vnet_hdr);
 394
 395         xdp.data = buf + hdr_padded_len;
 396         xdp.data_end = xdp.data + (len - vi->hdr_len);
 397
 398         act = bpf_prog_run_xdp(xdp_prog, &xdp);
 399         switch (act) {
 400         case XDP_PASS:
 401                 return XDP_PASS;
 402         case XDP_TX:
 403                 qp = vi->curr_queue_pairs -
 404                         vi->xdp_queue_pairs +
 405                         smp_processor_id();
 406                 xdp.data = buf + (vi->mergeable_rx_bufs ? 0 : 4);
 407                 virtnet_xdp_xmit(vi, rq, &vi->sq[qp], &xdp);
 408                 return XDP_TX;
 409         default:
 410                 bpf_warn_invalid_xdp_action(act);
 411         case XDP_ABORTED:
 412         case XDP_DROP:
 413                 return XDP_DROP;
 414         }
 415 }
 416
 417 static struct sk_buff *receive_small(struct virtnet_info *vi, void *buf, unsigned int len)
 418 {
 419         struct sk_buff * skb = buf;
 420
 421         len -= vi->hdr_len;
 422         skb_trim(skb, len);
 423
 424         return skb;
 425 }
 426
 427 static struct sk_buff *receive_big(struct net_device *dev,
 428                                    struct virtnet_info *vi,
 429                                    struct receive_queue *rq,
 430                                    void *buf,
 431                                    unsigned int len)
 432 {
 433         struct bpf_prog *xdp_prog;
 434         struct page *page = buf;
 435         struct sk_buff *skb;
 436
 437         rcu_read_lock();
 438         xdp_prog = rcu_dereference(rq->xdp_prog);
 439         if (xdp_prog) {
 440                 struct virtio_net_hdr_mrg_rxbuf *hdr = buf;
 441                 u32 act;
 442
 443                 if (unlikely(hdr->hdr.gso_type || hdr->hdr.flags))
 444                         goto err_xdp;
 445                 act = do_xdp_prog(vi, rq, xdp_prog, page, 0, len);
 446                 switch (act) {
 447                 case XDP_PASS:
 448                         break;
 449                 case XDP_TX:
 450                         rcu_read_unlock();
 451                         goto xdp_xmit;
 452                 case XDP_DROP:
 453                 default:
 454                         goto err_xdp;
 455                 }
 456         }
 457         rcu_read_unlock();
 458
 459         skb = page_to_skb(vi, rq, page, 0, len, PAGE_SIZE);
 460         if (unlikely(!skb))
 461                 goto err;
 462
 463         return skb;
 464
 465 err_xdp:
 466         rcu_read_unlock();
 467 err:
 468         dev->stats.rx_dropped++;
 469         give_pages(rq, page);
 470 xdp_xmit:
 471         return NULL;
 472 }
 473
 474 /* The conditions to enable XDP should preclude the underlying device from
 475  * sending packets across multiple buffers (num_buf > 1). However per spec
 476  * it does not appear to be illegal to do so but rather just against convention.
 477  * So in order to avoid making a system unresponsive the packets are pushed
 478  * into a page and the XDP program is run. This will be extremely slow and we
 479  * push a warning to the user to fix this as soon as possible. Fixing this may
 480  * require resolving the underlying hardware to determine why multiple buffers
 481  * are being received or simply loading the XDP program in the ingress stack
 482  * after the skb is built because there is no advantage to running it here
 483  * anymore.
 484  */
 485 static struct page *xdp_linearize_page(struct receive_queue *rq,
 486                                        u16 num_buf,
 487                                        struct page *p,
 488                                        int offset,
 489                                        unsigned int *len)
 490 {
 491         struct page *page = alloc_page(GFP_ATOMIC);
 492         unsigned int page_off = 0;
 493
 494         if (!page)
 495                 return NULL;
 496
 497         memcpy(page_address(page) + page_off, page_address(p) + offset, *len);
 498         page_off += *len;
 499
 500         while (--num_buf) {
 501                 unsigned int buflen;
 502                 unsigned long ctx;
 503                 void *buf;
 504                 int off;
 505
 506                 ctx = (unsigned long)virtqueue_get_buf(rq->vq, &buflen);
 507                 if (unlikely(!ctx))
 508                         goto err_buf;
 509
 510                 /* guard against a misconfigured or uncooperative backend that
 511                  * is sending packet larger than the MTU.
 512                  */
 513                 if ((page_off + buflen) > PAGE_SIZE)
 514                         goto err_buf;
 515
 516                 buf = mergeable_ctx_to_buf_address(ctx);
 517                 p = virt_to_head_page(buf);
 518                 off = buf - page_address(p);
 519
 520                 memcpy(page_address(page) + page_off,
 521                        page_address(p) + off, buflen);
 522                 page_off += buflen;
 523         }
 524
 525         *len = page_off;
 526         return page;
 527 err_buf:
 528         __free_pages(page, 0);
 529         return NULL;
 530 }
 531
 532 static struct sk_buff *receive_mergeable(struct net_device *dev,
 533                                          struct virtnet_info *vi,
 534                                          struct receive_queue *rq,
 535                                          unsigned long ctx,
 536                                          unsigned int len)
 537 {
 538         void *buf = mergeable_ctx_to_buf_address(ctx);
 539         struct virtio_net_hdr_mrg_rxbuf *hdr = buf;
 540         u16 num_buf = virtio16_to_cpu(vi->vdev, hdr->num_buffers);
 541         struct page *page = virt_to_head_page(buf);
 542         int offset = buf - page_address(page);
 543         struct sk_buff *head_skb, *curr_skb;
 544         struct bpf_prog *xdp_prog;
 545         unsigned int truesize;
 546
 547         head_skb = NULL;
 548
 549         rcu_read_lock();
 550         xdp_prog = rcu_dereference(rq->xdp_prog);
 551         if (xdp_prog) {
 552                 struct page *xdp_page;
 553                 u32 act;
 554
 555                 /* No known backend devices should send packets with
 556                  * more than a single buffer when XDP conditions are
 557                  * met. However it is not strictly illegal so the case
 558                  * is handled as an exception and a warning is thrown.
 559                  */
 560                 if (unlikely(num_buf > 1)) {
 561                         bpf_warn_invalid_xdp_buffer();
 562
 563                         /* linearize data for XDP */
 564                         xdp_page = xdp_linearize_page(rq, num_buf,
 565                                                       page, offset, &len);
 566                         if (!xdp_page)
 567                                 goto err_xdp;
 568                         offset = 0;
 569                 } else {
 570                         xdp_page = page;
 571                 }
 572
 573                 /* Transient failure which in theory could occur if
 574                  * in-flight packets from before XDP was enabled reach
 575                  * the receive path after XDP is loaded. In practice I
 576                  * was not able to create this condition.
 577                  */
 578                 if (unlikely(hdr->hdr.gso_type || hdr->hdr.flags))
 579                         goto err_xdp;
 580
 581                 act = do_xdp_prog(vi, rq, xdp_prog, page, offset, len);
 582                 switch (act) {
 583                 case XDP_PASS:
 584                         if (unlikely(xdp_page != page))
 585                                 __free_pages(xdp_page, 0);
 586                         break;
 587                 case XDP_TX:
 588                         if (unlikely(xdp_page != page))
 589                                 goto err_xdp;
 590                         rcu_read_unlock();
 591                         goto xdp_xmit;
 592                 case XDP_DROP:
 593                 default:
 594                         if (unlikely(xdp_page != page))
 595                                 __free_pages(xdp_page, 0);
 596                         goto err_xdp;
 597                 }
 598         }
 599         rcu_read_unlock();
 600
 601         truesize = max(len, mergeable_ctx_to_buf_truesize(ctx));
 602         head_skb = page_to_skb(vi, rq, page, offset, len, truesize);
 603         curr_skb = head_skb;
 604
 605         if (unlikely(!curr_skb))
 606                 goto err_skb;
 607         while (--num_buf) {
 608                 int num_skb_frags;
 609
 610                 ctx = (unsigned long)virtqueue_get_buf(rq->vq, &len);
 611                 if (unlikely(!ctx)) {
 612                         pr_debug("%s: rx error: %d buffers out of %d missing\n",
 613                                  dev->name, num_buf,
 614                                  virtio16_to_cpu(vi->vdev,
 615                                                  hdr->num_buffers));
 616                         dev->stats.rx_length_errors++;
 617                         goto err_buf;
 618                 }
 619
 620                 buf = mergeable_ctx_to_buf_address(ctx);
 621                 page = virt_to_head_page(buf);
 622
 623                 num_skb_frags = skb_shinfo(curr_skb)->nr_frags;
 624                 if (unlikely(num_skb_frags == MAX_SKB_FRAGS)) {
 625                         struct sk_buff *nskb = alloc_skb(0, GFP_ATOMIC);
 626
 627                         if (unlikely(!nskb))
 628                                 goto err_skb;
 629                         if (curr_skb == head_skb)
 630                                 skb_shinfo(curr_skb)->frag_list = nskb;
 631                         else
 632                                 curr_skb->next = nskb;
 633                         curr_skb = nskb;
 634                         head_skb->truesize += nskb->truesize;
 635                         num_skb_frags = 0;
 636                 }
 637                 truesize = max(len, mergeable_ctx_to_buf_truesize(ctx));
 638                 if (curr_skb != head_skb) {
 639                         head_skb->data_len += len;
 640                         head_skb->len += len;
 641                         head_skb->truesize += truesize;
 642                 }
 643                 offset = buf - page_address(page);
 644                 if (skb_can_coalesce(curr_skb, num_skb_frags, page, offset)) {
 645                         put_page(page);
 646                         skb_coalesce_rx_frag(curr_skb, num_skb_frags - 1,
 647                                              len, truesize);
 648                 } else {
 649                         skb_add_rx_frag(curr_skb, num_skb_frags, page,
 650                                         offset, len, truesize);
 651                 }
 652         }
 653
 654         ewma_pkt_len_add(&rq->mrg_avg_pkt_len, head_skb->len);
 655         return head_skb;
 656
 657 err_xdp:
 658         rcu_read_unlock();
 659 err_skb:
 660         put_page(page);
 661         while (--num_buf) {
 662                 ctx = (unsigned long)virtqueue_get_buf(rq->vq, &len);
 663                 if (unlikely(!ctx)) {
 664                         pr_debug("%s: rx error: %d buffers missing\n",
 665                                  dev->name, num_buf);
 666                         dev->stats.rx_length_errors++;
 667                         break;
 668                 }
 669                 page = virt_to_head_page(mergeable_ctx_to_buf_address(ctx));
 670                 put_page(page);
 671         }
 672 err_buf:
 673         dev->stats.rx_dropped++;
 674         dev_kfree_skb(head_skb);
 675 xdp_xmit:
 676         return NULL;
 677 }
 678
 679 static void receive_buf(struct virtnet_info *vi, struct receive_queue *rq,
 680                         void *buf, unsigned int len)
 681 {
 682         struct net_device *dev = vi->dev;
 683         struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
 684         struct sk_buff *skb;
 685         struct virtio_net_hdr_mrg_rxbuf *hdr;
 686
 687         if (unlikely(len < vi->hdr_len + ETH_HLEN)) {
 688                 pr_debug("%s: short packet %i\n", dev->name, len);
 689                 dev->stats.rx_length_errors++;
 690                 if (vi->mergeable_rx_bufs) {
 691                         unsigned long ctx = (unsigned long)buf;
 692                         void *base = mergeable_ctx_to_buf_address(ctx);
 693                         put_page(virt_to_head_page(base));
 694                 } else if (vi->big_packets) {
 695                         give_pages(rq, buf);
 696                 } else {
 697                         dev_kfree_skb(buf);
 698                 }
 699                 return;
 700         }
 701
 702         if (vi->mergeable_rx_bufs)
 703                 skb = receive_mergeable(dev, vi, rq, (unsigned long)buf, len);
 704         else if (vi->big_packets)
 705                 skb = receive_big(dev, vi, rq, buf, len);
 706         else
 707                 skb = receive_small(vi, buf, len);
 708
 709         if (unlikely(!skb))
 710                 return;
 711
 712         hdr = skb_vnet_hdr(skb);
 713
 714         u64_stats_update_begin(&stats->rx_syncp);
 715         stats->rx_bytes += skb->len;
 716         stats->rx_packets++;
 717         u64_stats_update_end(&stats->rx_syncp);
 718
 719         if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID)
 720                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 721
 722         if (virtio_net_hdr_to_skb(skb, &hdr->hdr,
 723                                   virtio_is_little_endian(vi->vdev))) {
 724                 net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n",
 725                                      dev->name, hdr->hdr.gso_type,
 726                                      hdr->hdr.gso_size);
 727                 goto frame_err;
 728         }
 729
 730         skb->protocol = eth_type_trans(skb, dev);
 731         pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
 732                  ntohs(skb->protocol), skb->len, skb->pkt_type);
 733
 734         napi_gro_receive(&rq->napi, skb);
 735         return;
 736
 737 frame_err:
 738         dev->stats.rx_frame_errors++;
 739         dev_kfree_skb(skb);
 740 }
 741
 742 static int add_recvbuf_small(struct virtnet_info *vi, struct receive_queue *rq,
 743                              gfp_t gfp)
 744 {
 745         struct sk_buff *skb;
 746         struct virtio_net_hdr_mrg_rxbuf *hdr;
 747         int err;
 748
 749         skb = __netdev_alloc_skb_ip_align(vi->dev, GOOD_PACKET_LEN, gfp);
 750         if (unlikely(!skb))
 751                 return -ENOMEM;
 752
 753         skb_put(skb, GOOD_PACKET_LEN);
 754
 755         hdr = skb_vnet_hdr(skb);
 756         sg_init_table(rq->sg, 2);
 757         sg_set_buf(rq->sg, hdr, vi->hdr_len);
 758         skb_to_sgvec(skb, rq->sg + 1, 0, skb->len);
 759
 760         err = virtqueue_add_inbuf(rq->vq, rq->sg, 2, skb, gfp);
 761         if (err < 0)
 762                 dev_kfree_skb(skb);
 763
 764         return err;
 765 }
 766
 767 static int add_recvbuf_big(struct virtnet_info *vi, struct receive_queue *rq,
 768                            gfp_t gfp)
 769 {
 770         struct page *first, *list = NULL;
 771         char *p;
 772         int i, err, offset;
 773
 774         sg_init_table(rq->sg, MAX_SKB_FRAGS + 2);
 775
 776         /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */
 777         for (i = MAX_SKB_FRAGS + 1; i > 1; --i) {
 778                 first = get_a_page(rq, gfp);
 779                 if (!first) {
 780                         if (list)
 781                                 give_pages(rq, list);
 782                         return -ENOMEM;
 783                 }
 784                 sg_set_buf(&rq->sg[i], page_address(first), PAGE_SIZE);
 785
 786                 /* chain new page in list head to match sg */
 787                 first->private = (unsigned long)list;
 788                 list = first;
 789         }
 790
 791         first = get_a_page(rq, gfp);
 792         if (!first) {
 793                 give_pages(rq, list);
 794                 return -ENOMEM;
 795         }
 796         p = page_address(first);
 797
 798         /* rq->sg[0], rq->sg[1] share the same page */
 799         /* a separated rq->sg[0] for header - required in case !any_header_sg */
 800         sg_set_buf(&rq->sg[0], p, vi->hdr_len);
 801
 802         /* rq->sg[1] for data packet, from offset */
 803         offset = sizeof(struct padded_vnet_hdr);
 804         sg_set_buf(&rq->sg[1], p + offset, PAGE_SIZE - offset);
 805
 806         /* chain first in list head */
 807         first->private = (unsigned long)list;
 808         err = virtqueue_add_inbuf(rq->vq, rq->sg, MAX_SKB_FRAGS + 2,
 809                                   first, gfp);
 810         if (err < 0)
 811                 give_pages(rq, first);
 812
 813         return err;
 814 }
 815
 816 static unsigned int get_mergeable_buf_len(struct ewma_pkt_len *avg_pkt_len)
 817 {
 818         const size_t hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
 819         unsigned int len;
 820
 821         len = hdr_len + clamp_t(unsigned int, ewma_pkt_len_read(avg_pkt_len),
 822                         GOOD_PACKET_LEN, PAGE_SIZE - hdr_len);
 823         return ALIGN(len, MERGEABLE_BUFFER_ALIGN);
 824 }
 825
 826 static int add_recvbuf_mergeable(struct receive_queue *rq, gfp_t gfp)
 827 {
 828         struct page_frag *alloc_frag = &rq->alloc_frag;
 829         char *buf;
 830         unsigned long ctx;
 831         int err;
 832         unsigned int len, hole;
 833
 834         len = get_mergeable_buf_len(&rq->mrg_avg_pkt_len);
 835         if (unlikely(!skb_page_frag_refill(len, alloc_frag, gfp)))
 836                 return -ENOMEM;
 837
 838         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
 839         ctx = mergeable_buf_to_ctx(buf, len);
 840         get_page(alloc_frag->page);
 841         alloc_frag->offset += len;
 842         hole = alloc_frag->size - alloc_frag->offset;
 843         if (hole < len) {
 844                 /* To avoid internal fragmentation, if there is very likely not
 845                  * enough space for another buffer, add the remaining space to
 846                  * the current buffer. This extra space is not included in
 847                  * the truesize stored in ctx.
 848                  */
 849                 len += hole;
 850                 alloc_frag->offset += hole;
 851         }
 852
 853         sg_init_one(rq->sg, buf, len);
 854         err = virtqueue_add_inbuf(rq->vq, rq->sg, 1, (void *)ctx, gfp);
 855         if (err < 0)
 856                 put_page(virt_to_head_page(buf));
 857
 858         return err;
 859 }
 860
 861 /*
 862  * Returns false if we couldn't fill entirely (OOM).
 863  *
 864  * Normally run in the receive path, but can also be run from ndo_open
 865  * before we're receiving packets, or from refill_work which is
 866  * careful to disable receiving (using napi_disable).
 867  */
 868 static bool try_fill_recv(struct virtnet_info *vi, struct receive_queue *rq,
 869                           gfp_t gfp)
 870 {
 871         int err;
 872         bool oom;
 873
 874         gfp |= __GFP_COLD;
 875         do {
 876                 if (vi->mergeable_rx_bufs)
 877                         err = add_recvbuf_mergeable(rq, gfp);
 878                 else if (vi->big_packets)
 879                         err = add_recvbuf_big(vi, rq, gfp);
 880                 else
 881                         err = add_recvbuf_small(vi, rq, gfp);
 882
 883                 oom = err == -ENOMEM;
 884                 if (err)
 885                         break;
 886         } while (rq->vq->num_free);
 887         virtqueue_kick(rq->vq);
 888         return !oom;
 889 }
 890
 891 static void skb_recv_done(struct virtqueue *rvq)
 892 {
 893         struct virtnet_info *vi = rvq->vdev->priv;
 894         struct receive_queue *rq = &vi->rq[vq2rxq(rvq)];
 895
 896         /* Schedule NAPI, Suppress further interrupts if successful. */
 897         if (napi_schedule_prep(&rq->napi)) {
 898                 virtqueue_disable_cb(rvq);
 899                 __napi_schedule(&rq->napi);
 900         }
 901 }
 902
 903 static void virtnet_napi_enable(struct receive_queue *rq)
 904 {
 905         napi_enable(&rq->napi);
 906
 907         /* If all buffers were filled by other side before we napi_enabled, we
 908          * won't get another interrupt, so process any outstanding packets
 909          * now.  virtnet_poll wants re-enable the queue, so we disable here.
 910          * We synchronize against interrupts via NAPI_STATE_SCHED */
 911         if (napi_schedule_prep(&rq->napi)) {
 912                 virtqueue_disable_cb(rq->vq);
 913                 local_bh_disable();
 914                 __napi_schedule(&rq->napi);
 915                 local_bh_enable();
 916         }
 917 }
 918
 919 static void refill_work(struct work_struct *work)
 920 {
 921         struct virtnet_info *vi =
 922                 container_of(work, struct virtnet_info, refill.work);
 923         bool still_empty;
 924         int i;
 925
 926         for (i = 0; i < vi->curr_queue_pairs; i++) {
 927                 struct receive_queue *rq = &vi->rq[i];
 928
 929                 napi_disable(&rq->napi);
 930                 still_empty = !try_fill_recv(vi, rq, GFP_KERNEL);
 931                 virtnet_napi_enable(rq);
 932
 933                 /* In theory, this can happen: if we don't get any buffers in
 934                  * we will *never* try to fill again.
 935                  */
 936                 if (still_empty)
 937                         schedule_delayed_work(&vi->refill, HZ/2);
 938         }
 939 }
 940
 941 static int virtnet_receive(struct receive_queue *rq, int budget)
 942 {
 943         struct virtnet_info *vi = rq->vq->vdev->priv;
 944         unsigned int len, received = 0;
 945         void *buf;
 946
 947         while (received < budget &&
 948                (buf = virtqueue_get_buf(rq->vq, &len)) != NULL) {
 949                 receive_buf(vi, rq, buf, len);
 950                 received++;
 951         }
 952
 953         if (rq->vq->num_free > virtqueue_get_vring_size(rq->vq) / 2) {
 954                 if (!try_fill_recv(vi, rq, GFP_ATOMIC))
 955                         schedule_delayed_work(&vi->refill, 0);
 956         }
 957
 958         return received;
 959 }
 960
 961 static int virtnet_poll(struct napi_struct *napi, int budget)
 962 {
 963         struct receive_queue *rq =
 964                 container_of(napi, struct receive_queue, napi);
 965         unsigned int r, received;
 966
 967         received = virtnet_receive(rq, budget);
 968
 969         /* Out of packets? */
 970         if (received < budget) {
 971                 r = virtqueue_enable_cb_prepare(rq->vq);
 972                 napi_complete_done(napi, received);
 973                 if (unlikely(virtqueue_poll(rq->vq, r)) &&
 974                     napi_schedule_prep(napi)) {
 975                         virtqueue_disable_cb(rq->vq);
 976                         __napi_schedule(napi);
 977                 }
 978         }
 979
 980         return received;
 981 }
 982
 983 #ifdef CONFIG_NET_RX_BUSY_POLL
 984 /* must be called with local_bh_disable()d */
 985 static int virtnet_busy_poll(struct napi_struct *napi)
 986 {
 987         struct receive_queue *rq =
 988                 container_of(napi, struct receive_queue, napi);
 989         struct virtnet_info *vi = rq->vq->vdev->priv;
 990         int r, received = 0, budget = 4;
 991
 992         if (!(vi->status & VIRTIO_NET_S_LINK_UP))
 993                 return LL_FLUSH_FAILED;
 994
 995         if (!napi_schedule_prep(napi))
 996                 return LL_FLUSH_BUSY;
 997
 998         virtqueue_disable_cb(rq->vq);
 999
1000 again:
1001         received += virtnet_receive(rq, budget);
1002
1003         r = virtqueue_enable_cb_prepare(rq->vq);
1004         clear_bit(NAPI_STATE_SCHED, &napi->state);
1005         if (unlikely(virtqueue_poll(rq->vq, r)) &&
1006             napi_schedule_prep(napi)) {
1007                 virtqueue_disable_cb(rq->vq);
1008                 if (received < budget) {
1009                         budget -= received;
1010                         goto again;
1011                 } else {
1012                         __napi_schedule(napi);
1013                 }
1014         }
1015
1016         return received;
1017 }
1018 #endif  /* CONFIG_NET_RX_BUSY_POLL */
1019
1020 static int virtnet_open(struct net_device *dev)
1021 {
1022         struct virtnet_info *vi = netdev_priv(dev);
1023         int i;
1024
1025         for (i = 0; i < vi->max_queue_pairs; i++) {
1026                 if (i < vi->curr_queue_pairs)
1027                         /* Make sure we have some buffers: if oom use wq. */
1028                         if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL))
1029                                 schedule_delayed_work(&vi->refill, 0);
1030                 virtnet_napi_enable(&vi->rq[i]);
1031         }
1032
1033         return 0;
1034 }
1035
1036 static void free_old_xmit_skbs(struct send_queue *sq)
1037 {
1038         struct sk_buff *skb;
1039         unsigned int len;
1040         struct virtnet_info *vi = sq->vq->vdev->priv;
1041         struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
1042
1043         while ((skb = virtqueue_get_buf(sq->vq, &len)) != NULL) {
1044                 pr_debug("Sent skb %p\n", skb);
1045
1046                 u64_stats_update_begin(&stats->tx_syncp);
1047                 stats->tx_bytes += skb->len;
1048                 stats->tx_packets++;
1049                 u64_stats_update_end(&stats->tx_syncp);
1050
1051                 dev_kfree_skb_any(skb);
1052         }
1053 }
1054
1055 static int xmit_skb(struct send_queue *sq, struct sk_buff *skb)
1056 {
1057         struct virtio_net_hdr_mrg_rxbuf *hdr;
1058         const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
1059         struct virtnet_info *vi = sq->vq->vdev->priv;
1060         unsigned num_sg;
1061         unsigned hdr_len = vi->hdr_len;
1062         bool can_push;
1063
1064         pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
1065
1066         can_push = vi->any_header_sg &&
1067                 !((unsigned long)skb->data & (__alignof__(*hdr) - 1)) &&
1068                 !skb_header_cloned(skb) && skb_headroom(skb) >= hdr_len;
1069         /* Even if we can, don't push here yet as this would skew
1070          * csum_start offset below. */
1071         if (can_push)
1072                 hdr = (struct virtio_net_hdr_mrg_rxbuf *)(skb->data - hdr_len);
1073         else
1074                 hdr = skb_vnet_hdr(skb);
1075
1076         if (virtio_net_hdr_from_skb(skb, &hdr->hdr,
1077                                     virtio_is_little_endian(vi->vdev)))
1078                 BUG();
1079
1080         if (vi->mergeable_rx_bufs)
1081                 hdr->num_buffers = 0;
1082
1083         sg_init_table(sq->sg, skb_shinfo(skb)->nr_frags + (can_push ? 1 : 2));
1084         if (can_push) {
1085                 __skb_push(skb, hdr_len);
1086                 num_sg = skb_to_sgvec(skb, sq->sg, 0, skb->len);
1087                 /* Pull header back to avoid skew in tx bytes calculations. */
1088                 __skb_pull(skb, hdr_len);
1089         } else {
1090                 sg_set_buf(sq->sg, hdr, hdr_len);
1091                 num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len) + 1;
1092         }
1093         return virtqueue_add_outbuf(sq->vq, sq->sg, num_sg, skb, GFP_ATOMIC);
1094 }
1095
1096 static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
1097 {
1098         struct virtnet_info *vi = netdev_priv(dev);
1099         int qnum = skb_get_queue_mapping(skb);
1100         struct send_queue *sq = &vi->sq[qnum];
1101         int err;
1102         struct netdev_queue *txq = netdev_get_tx_queue(dev, qnum);
1103         bool kick = !skb->xmit_more;
1104
1105         /* Free up any pending old buffers before queueing new ones. */
1106         free_old_xmit_skbs(sq);
1107
1108         /* timestamp packet in software */
1109         skb_tx_timestamp(skb);
1110
1111         /* Try to transmit */
1112         err = xmit_skb(sq, skb);
1113
1114         /* This should not happen! */
1115         if (unlikely(err)) {
1116                 dev->stats.tx_fifo_errors++;
1117                 if (net_ratelimit())
1118                         dev_warn(&dev->dev,
1119                                  "Unexpected TXQ (%d) queue failure: %d\n", qnum, err);
1120                 dev->stats.tx_dropped++;
1121                 dev_kfree_skb_any(skb);
1122                 return NETDEV_TX_OK;
1123         }
1124
1125         /* Don't wait up for transmitted skbs to be freed. */
1126         skb_orphan(skb);
1127         nf_reset(skb);
1128
1129         /* If running out of space, stop queue to avoid getting packets that we
1130          * are then unable to transmit.
1131          * An alternative would be to force queuing layer to requeue the skb by
1132          * returning NETDEV_TX_BUSY. However, NETDEV_TX_BUSY should not be
1133          * returned in a normal path of operation: it means that driver is not
1134          * maintaining the TX queue stop/start state properly, and causes
1135          * the stack to do a non-trivial amount of useless work.
1136          * Since most packets only take 1 or 2 ring slots, stopping the queue
1137          * early means 16 slots are typically wasted.
1138          */
1139         if (sq->vq->num_free < 2+MAX_SKB_FRAGS) {
1140                 netif_stop_subqueue(dev, qnum);
1141                 if (unlikely(!virtqueue_enable_cb_delayed(sq->vq))) {
1142                         /* More just got used, free them then recheck. */
1143                         free_old_xmit_skbs(sq);
1144                         if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) {
1145                                 netif_start_subqueue(dev, qnum);
1146                                 virtqueue_disable_cb(sq->vq);
1147                         }
1148                 }
1149         }
1150
1151         if (kick || netif_xmit_stopped(txq))
1152                 virtqueue_kick(sq->vq);
1153
1154         return NETDEV_TX_OK;
1155 }
1156
1157 /*
1158  * Send command via the control virtqueue and check status.  Commands
1159  * supported by the hypervisor, as indicated by feature bits, should
1160  * never fail unless improperly formatted.
1161  */
1162 static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
1163                                  struct scatterlist *out)
1164 {
1165         struct scatterlist *sgs[4], hdr, stat;
1166         unsigned out_num = 0, tmp;
1167
1168         /* Caller should know better */
1169         BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ));
1170
1171         vi->ctrl_status = ~0;
1172         vi->ctrl_hdr.class = class;
1173         vi->ctrl_hdr.cmd = cmd;
1174         /* Add header */
1175         sg_init_one(&hdr, &vi->ctrl_hdr, sizeof(vi->ctrl_hdr));
1176         sgs[out_num++] = &hdr;
1177
1178         if (out)
1179                 sgs[out_num++] = out;
1180
1181         /* Add return status. */
1182         sg_init_one(&stat, &vi->ctrl_status, sizeof(vi->ctrl_status));
1183         sgs[out_num] = &stat;
1184
1185         BUG_ON(out_num + 1 > ARRAY_SIZE(sgs));
1186         virtqueue_add_sgs(vi->cvq, sgs, out_num, 1, vi, GFP_ATOMIC);
1187
1188         if (unlikely(!virtqueue_kick(vi->cvq)))
1189                 return vi->ctrl_status == VIRTIO_NET_OK;
1190
1191         /* Spin for a response, the kick causes an ioport write, trapping
1192          * into the hypervisor, so the request should be handled immediately.
1193          */
1194         while (!virtqueue_get_buf(vi->cvq, &tmp) &&
1195                !virtqueue_is_broken(vi->cvq))
1196                 cpu_relax();
1197
1198         return vi->ctrl_status == VIRTIO_NET_OK;
1199 }
1200
1201 static int virtnet_set_mac_address(struct net_device *dev, void *p)
1202 {
1203         struct virtnet_info *vi = netdev_priv(dev);
1204         struct virtio_device *vdev = vi->vdev;
1205         int ret;
1206         struct sockaddr *addr;
1207         struct scatterlist sg;
1208
1209         addr = kmalloc(sizeof(*addr), GFP_KERNEL);
1210         if (!addr)
1211                 return -ENOMEM;
1212         memcpy(addr, p, sizeof(*addr));
1213
1214         ret = eth_prepare_mac_addr_change(dev, addr);
1215         if (ret)
1216                 goto out;
1217
1218         if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1219                 sg_init_one(&sg, addr->sa_data, dev->addr_len);
1220                 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
1221                                           VIRTIO_NET_CTRL_MAC_ADDR_SET, &sg)) {
1222                         dev_warn(&vdev->dev,
1223                                  "Failed to set mac address by vq command.\n");
1224                         ret = -EINVAL;
1225                         goto out;
1226                 }
1227         } else if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC) &&
1228                    !virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) {
1229                 unsigned int i;
1230
1231                 /* Naturally, this has an atomicity problem. */
1232                 for (i = 0; i < dev->addr_len; i++)
1233                         virtio_cwrite8(vdev,
1234                                        offsetof(struct virtio_net_config, mac) +
1235                                        i, addr->sa_data[i]);
1236         }
1237
1238         eth_commit_mac_addr_change(dev, p);
1239         ret = 0;
1240
1241 out:
1242         kfree(addr);
1243         return ret;
1244 }
1245
1246 static struct rtnl_link_stats64 *virtnet_stats(struct net_device *dev,
1247                                                struct rtnl_link_stats64 *tot)
1248 {
1249         struct virtnet_info *vi = netdev_priv(dev);
1250         int cpu;
1251         unsigned int start;
1252
1253         for_each_possible_cpu(cpu) {
1254                 struct virtnet_stats *stats = per_cpu_ptr(vi->stats, cpu);
1255                 u64 tpackets, tbytes, rpackets, rbytes;
1256
1257                 do {
1258                         start = u64_stats_fetch_begin_irq(&stats->tx_syncp);
1259                         tpackets = stats->tx_packets;
1260                         tbytes   = stats->tx_bytes;
1261                 } while (u64_stats_fetch_retry_irq(&stats->tx_syncp, start));
1262
1263                 do {
1264                         start = u64_stats_fetch_begin_irq(&stats->rx_syncp);
1265                         rpackets = stats->rx_packets;
1266                         rbytes   = stats->rx_bytes;
1267                 } while (u64_stats_fetch_retry_irq(&stats->rx_syncp, start));
1268
1269                 tot->rx_packets += rpackets;
1270                 tot->tx_packets += tpackets;
1271                 tot->rx_bytes   += rbytes;
1272                 tot->tx_bytes   += tbytes;
1273         }
1274
1275         tot->tx_dropped = dev->stats.tx_dropped;
1276         tot->tx_fifo_errors = dev->stats.tx_fifo_errors;
1277         tot->rx_dropped = dev->stats.rx_dropped;
1278         tot->rx_length_errors = dev->stats.rx_length_errors;
1279         tot->rx_frame_errors = dev->stats.rx_frame_errors;
1280
1281         return tot;
1282 }
1283
1284 #ifdef CONFIG_NET_POLL_CONTROLLER
1285 static void virtnet_netpoll(struct net_device *dev)
1286 {
1287         struct virtnet_info *vi = netdev_priv(dev);
1288         int i;
1289
1290         for (i = 0; i < vi->curr_queue_pairs; i++)
1291                 napi_schedule(&vi->rq[i].napi);
1292 }
1293 #endif
1294
1295 static void virtnet_ack_link_announce(struct virtnet_info *vi)
1296 {
1297         rtnl_lock();
1298         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE,
1299                                   VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL))
1300                 dev_warn(&vi->dev->dev, "Failed to ack link announce.\n");
1301         rtnl_unlock();
1302 }
1303
1304 static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs)
1305 {
1306         struct scatterlist sg;
1307         struct net_device *dev = vi->dev;
1308
1309         if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ))
1310                 return 0;
1311
1312         vi->ctrl_mq.virtqueue_pairs = cpu_to_virtio16(vi->vdev, queue_pairs);
1313         sg_init_one(&sg, &vi->ctrl_mq, sizeof(vi->ctrl_mq));
1314
1315         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ,
1316                                   VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg)) {
1317                 dev_warn(&dev->dev, "Fail to set num of queue pairs to %d\n",
1318                          queue_pairs);
1319                 return -EINVAL;
1320         } else {
1321                 vi->curr_queue_pairs = queue_pairs;
1322                 /* virtnet_open() will refill when device is going to up. */
1323                 if (dev->flags & IFF_UP)
1324                         schedule_delayed_work(&vi->refill, 0);
1325         }
1326
1327         return 0;
1328 }
1329
1330 static int virtnet_close(struct net_device *dev)
1331 {
1332         struct virtnet_info *vi = netdev_priv(dev);
1333         int i;
1334
1335         /* Make sure refill_work doesn't re-enable napi! */
1336         cancel_delayed_work_sync(&vi->refill);
1337
1338         for (i = 0; i < vi->max_queue_pairs; i++)
1339                 napi_disable(&vi->rq[i].napi);
1340
1341         return 0;
1342 }
1343
1344 static void virtnet_set_rx_mode(struct net_device *dev)
1345 {
1346         struct virtnet_info *vi = netdev_priv(dev);
1347         struct scatterlist sg[2];
1348         struct virtio_net_ctrl_mac *mac_data;
1349         struct netdev_hw_addr *ha;
1350         int uc_count;
1351         int mc_count;
1352         void *buf;
1353         int i;
1354
1355         /* We can't dynamically set ndo_set_rx_mode, so return gracefully */
1356         if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
1357                 return;
1358
1359         vi->ctrl_promisc = ((dev->flags & IFF_PROMISC) != 0);
1360         vi->ctrl_allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
1361
1362         sg_init_one(sg, &vi->ctrl_promisc, sizeof(vi->ctrl_promisc));
1363
1364         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
1365                                   VIRTIO_NET_CTRL_RX_PROMISC, sg))
1366                 dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
1367                          vi->ctrl_promisc ? "en" : "dis");
1368
1369         sg_init_one(sg, &vi->ctrl_allmulti, sizeof(vi->ctrl_allmulti));
1370
1371         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
1372                                   VIRTIO_NET_CTRL_RX_ALLMULTI, sg))
1373                 dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
1374                          vi->ctrl_allmulti ? "en" : "dis");
1375
1376         uc_count = netdev_uc_count(dev);
1377         mc_count = netdev_mc_count(dev);
1378         /* MAC filter - use one buffer for both lists */
1379         buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) +
1380                       (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
1381         mac_data = buf;
1382         if (!buf)
1383                 return;
1384
1385         sg_init_table(sg, 2);
1386
1387         /* Store the unicast list and count in the front of the buffer */
1388         mac_data->entries = cpu_to_virtio32(vi->vdev, uc_count);
1389         i = 0;
1390         netdev_for_each_uc_addr(ha, dev)
1391                 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
1392
1393         sg_set_buf(&sg[0], mac_data,
1394                    sizeof(mac_data->entries) + (uc_count * ETH_ALEN));
1395
1396         /* multicast list and count fill the end */
1397         mac_data = (void *)&mac_data->macs[uc_count][0];
1398
1399         mac_data->entries = cpu_to_virtio32(vi->vdev, mc_count);
1400         i = 0;
1401         netdev_for_each_mc_addr(ha, dev)
1402                 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
1403
1404         sg_set_buf(&sg[1], mac_data,
1405                    sizeof(mac_data->entries) + (mc_count * ETH_ALEN));
1406
1407         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
1408                                   VIRTIO_NET_CTRL_MAC_TABLE_SET, sg))
1409                 dev_warn(&dev->dev, "Failed to set MAC filter table.\n");
1410
1411         kfree(buf);
1412 }
1413
1414 static int virtnet_vlan_rx_add_vid(struct net_device *dev,
1415                                    __be16 proto, u16 vid)
1416 {
1417         struct virtnet_info *vi = netdev_priv(dev);
1418         struct scatterlist sg;
1419
1420         vi->ctrl_vid = vid;
1421         sg_init_one(&sg, &vi->ctrl_vid, sizeof(vi->ctrl_vid));
1422
1423         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
1424                                   VIRTIO_NET_CTRL_VLAN_ADD, &sg))
1425                 dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
1426         return 0;
1427 }
1428
1429 static int virtnet_vlan_rx_kill_vid(struct net_device *dev,
1430                                     __be16 proto, u16 vid)
1431 {
1432         struct virtnet_info *vi = netdev_priv(dev);
1433         struct scatterlist sg;
1434
1435         vi->ctrl_vid = vid;
1436         sg_init_one(&sg, &vi->ctrl_vid, sizeof(vi->ctrl_vid));
1437
1438         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
1439                                   VIRTIO_NET_CTRL_VLAN_DEL, &sg))
1440                 dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
1441         return 0;
1442 }
1443
1444 static void virtnet_clean_affinity(struct virtnet_info *vi, long hcpu)
1445 {
1446         int i;
1447
1448         if (vi->affinity_hint_set) {
1449                 for (i = 0; i < vi->max_queue_pairs; i++) {
1450                         virtqueue_set_affinity(vi->rq[i].vq, -1);
1451                         virtqueue_set_affinity(vi->sq[i].vq, -1);
1452                 }
1453
1454                 vi->affinity_hint_set = false;
1455         }
1456 }
1457
1458 static void virtnet_set_affinity(struct virtnet_info *vi)
1459 {
1460         int i;
1461         int cpu;
1462
1463         /* In multiqueue mode, when the number of cpu is equal to the number of
1464          * queue pairs, we let the queue pairs to be private to one cpu by
1465          * setting the affinity hint to eliminate the contention.
1466          */
1467         if (vi->curr_queue_pairs == 1 ||
1468             vi->max_queue_pairs != num_online_cpus()) {
1469                 virtnet_clean_affinity(vi, -1);
1470                 return;
1471         }
1472
1473         i = 0;
1474         for_each_online_cpu(cpu) {
1475                 virtqueue_set_affinity(vi->rq[i].vq, cpu);
1476                 virtqueue_set_affinity(vi->sq[i].vq, cpu);
1477                 netif_set_xps_queue(vi->dev, cpumask_of(cpu), i);
1478                 i++;
1479         }
1480
1481         vi->affinity_hint_set = true;
1482 }
1483
1484 static int virtnet_cpu_online(unsigned int cpu, struct hlist_node *node)
1485 {
1486         struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info,
1487                                                    node);
1488         virtnet_set_affinity(vi);
1489         return 0;
1490 }
1491
1492 static int virtnet_cpu_dead(unsigned int cpu, struct hlist_node *node)
1493 {
1494         struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info,
1495                                                    node_dead);
1496         virtnet_set_affinity(vi);
1497         return 0;
1498 }
1499
1500 static int virtnet_cpu_down_prep(unsigned int cpu, struct hlist_node *node)
1501 {
1502         struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info,
1503                                                    node);
1504
1505         virtnet_clean_affinity(vi, cpu);
1506         return 0;
1507 }
1508
1509 static enum cpuhp_state virtionet_online;
1510
1511 static int virtnet_cpu_notif_add(struct virtnet_info *vi)
1512 {
1513         int ret;
1514
1515         ret = cpuhp_state_add_instance_nocalls(virtionet_online, &vi->node);
1516         if (ret)
1517                 return ret;
1518         ret = cpuhp_state_add_instance_nocalls(CPUHP_VIRT_NET_DEAD,
1519                                                &vi->node_dead);
1520         if (!ret)
1521                 return ret;
1522         cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node);
1523         return ret;
1524 }
1525
1526 static void virtnet_cpu_notif_remove(struct virtnet_info *vi)
1527 {
1528         cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node);
1529         cpuhp_state_remove_instance_nocalls(CPUHP_VIRT_NET_DEAD,
1530                                             &vi->node_dead);
1531 }
1532
1533 static void virtnet_get_ringparam(struct net_device *dev,
1534                                 struct ethtool_ringparam *ring)
1535 {
1536         struct virtnet_info *vi = netdev_priv(dev);
1537
1538         ring->rx_max_pending = virtqueue_get_vring_size(vi->rq[0].vq);
1539         ring->tx_max_pending = virtqueue_get_vring_size(vi->sq[0].vq);
1540         ring->rx_pending = ring->rx_max_pending;
1541         ring->tx_pending = ring->tx_max_pending;
1542 }
1543
1544
1545 static void virtnet_get_drvinfo(struct net_device *dev,
1546                                 struct ethtool_drvinfo *info)
1547 {
1548         struct virtnet_info *vi = netdev_priv(dev);
1549         struct virtio_device *vdev = vi->vdev;
1550
1551         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
1552         strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version));
1553         strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info));
1554
1555 }
1556
1557 /* TODO: Eliminate OOO packets during switching */
1558 static int virtnet_set_channels(struct net_device *dev,
1559                                 struct ethtool_channels *channels)
1560 {
1561         struct virtnet_info *vi = netdev_priv(dev);
1562         u16 queue_pairs = channels->combined_count;
1563         int err;
1564
1565         /* We don't support separate rx/tx channels.
1566          * We don't allow setting 'other' channels.
1567          */
1568         if (channels->rx_count || channels->tx_count || channels->other_count)
1569                 return -EINVAL;
1570
1571         if (queue_pairs > vi->max_queue_pairs || queue_pairs == 0)
1572                 return -EINVAL;
1573
1574         /* For now we don't support modifying channels while XDP is loaded
1575          * also when XDP is loaded all RX queues have XDP programs so we only
1576          * need to check a single RX queue.
1577          */
1578         if (vi->rq[0].xdp_prog)
1579                 return -EINVAL;
1580
1581         get_online_cpus();
1582         err = virtnet_set_queues(vi, queue_pairs);
1583         if (!err) {
1584                 netif_set_real_num_tx_queues(dev, queue_pairs);
1585                 netif_set_real_num_rx_queues(dev, queue_pairs);
1586
1587                 virtnet_set_affinity(vi);
1588         }
1589         put_online_cpus();
1590
1591         return err;
1592 }
1593
1594 static void virtnet_get_channels(struct net_device *dev,
1595                                  struct ethtool_channels *channels)
1596 {
1597         struct virtnet_info *vi = netdev_priv(dev);
1598
1599         channels->combined_count = vi->curr_queue_pairs;
1600         channels->max_combined = vi->max_queue_pairs;
1601         channels->max_other = 0;
1602         channels->rx_count = 0;
1603         channels->tx_count = 0;
1604         channels->other_count = 0;
1605 }
1606
1607 /* Check if the user is trying to change anything besides speed/duplex */
1608 static bool virtnet_validate_ethtool_cmd(const struct ethtool_cmd *cmd)
1609 {
1610         struct ethtool_cmd diff1 = *cmd;
1611         struct ethtool_cmd diff2 = {};
1612
1613         /* cmd is always set so we need to clear it, validate the port type
1614          * and also without autonegotiation we can ignore advertising
1615          */
1616         ethtool_cmd_speed_set(&diff1, 0);
1617         diff2.port = PORT_OTHER;
1618         diff1.advertising = 0;
1619         diff1.duplex = 0;
1620         diff1.cmd = 0;
1621
1622         return !memcmp(&diff1, &diff2, sizeof(diff1));
1623 }
1624
1625 static int virtnet_set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1626 {
1627         struct virtnet_info *vi = netdev_priv(dev);
1628         u32 speed;
1629
1630         speed = ethtool_cmd_speed(cmd);
1631         /* don't allow custom speed and duplex */
1632         if (!ethtool_validate_speed(speed) ||
1633             !ethtool_validate_duplex(cmd->duplex) ||
1634             !virtnet_validate_ethtool_cmd(cmd))
1635                 return -EINVAL;
1636         vi->speed = speed;
1637         vi->duplex = cmd->duplex;
1638
1639         return 0;
1640 }
1641
1642 static int virtnet_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1643 {
1644         struct virtnet_info *vi = netdev_priv(dev);
1645
1646         ethtool_cmd_speed_set(cmd, vi->speed);
1647         cmd->duplex = vi->duplex;
1648         cmd->port = PORT_OTHER;
1649
1650         return 0;
1651 }
1652
1653 static void virtnet_init_settings(struct net_device *dev)
1654 {
1655         struct virtnet_info *vi = netdev_priv(dev);
1656
1657         vi->speed = SPEED_UNKNOWN;
1658         vi->duplex = DUPLEX_UNKNOWN;
1659 }
1660
1661 static const struct ethtool_ops virtnet_ethtool_ops = {
1662         .get_drvinfo = virtnet_get_drvinfo,
1663         .get_link = ethtool_op_get_link,
1664         .get_ringparam = virtnet_get_ringparam,
1665         .set_channels = virtnet_set_channels,
1666         .get_channels = virtnet_get_channels,
1667         .get_ts_info = ethtool_op_get_ts_info,
1668         .get_settings = virtnet_get_settings,
1669         .set_settings = virtnet_set_settings,
1670 };
1671
1672 static int virtnet_xdp_set(struct net_device *dev, struct bpf_prog *prog)
1673 {
1674         unsigned long int max_sz = PAGE_SIZE - sizeof(struct padded_vnet_hdr);
1675         struct virtnet_info *vi = netdev_priv(dev);
1676         struct bpf_prog *old_prog;
1677         u16 xdp_qp = 0, curr_qp;
1678         int i, err;
1679
1680         if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO4) ||
1681             virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO6)) {
1682                 netdev_warn(dev, "can't set XDP while host is implementing LRO, disable LRO first\n");
1683                 return -EOPNOTSUPP;
1684         }
1685
1686         if (vi->mergeable_rx_bufs && !vi->any_header_sg) {
1687                 netdev_warn(dev, "XDP expects header/data in single page, any_header_sg required\n");
1688                 return -EINVAL;
1689         }
1690
1691         if (dev->mtu > max_sz) {
1692                 netdev_warn(dev, "XDP requires MTU less than %lu\n", max_sz);
1693                 return -EINVAL;
1694         }
1695
1696         curr_qp = vi->curr_queue_pairs - vi->xdp_queue_pairs;
1697         if (prog)
1698                 xdp_qp = nr_cpu_ids;
1699
1700         /* XDP requires extra queues for XDP_TX */
1701         if (curr_qp + xdp_qp > vi->max_queue_pairs) {
1702                 netdev_warn(dev, "request %i queues but max is %i\n",
1703                             curr_qp + xdp_qp, vi->max_queue_pairs);
1704                 return -ENOMEM;
1705         }
1706
1707         err = virtnet_set_queues(vi, curr_qp + xdp_qp);
1708         if (err) {
1709                 dev_warn(&dev->dev, "XDP Device queue allocation failure.\n");
1710                 return err;
1711         }
1712
1713         if (prog) {
1714                 prog = bpf_prog_add(prog, vi->max_queue_pairs - 1);
1715                 if (IS_ERR(prog)) {
1716                         virtnet_set_queues(vi, curr_qp);
1717                         return PTR_ERR(prog);
1718                 }
1719         }
1720
1721         vi->xdp_queue_pairs = xdp_qp;
1722         netif_set_real_num_rx_queues(dev, curr_qp + xdp_qp);
1723
1724         for (i = 0; i < vi->max_queue_pairs; i++) {
1725                 old_prog = rtnl_dereference(vi->rq[i].xdp_prog);
1726                 rcu_assign_pointer(vi->rq[i].xdp_prog, prog);
1727                 if (old_prog)
1728                         bpf_prog_put(old_prog);
1729         }
1730
1731         return 0;
1732 }
1733
1734 static bool virtnet_xdp_query(struct net_device *dev)
1735 {
1736         struct virtnet_info *vi = netdev_priv(dev);
1737         int i;
1738
1739         for (i = 0; i < vi->max_queue_pairs; i++) {
1740                 if (vi->rq[i].xdp_prog)
1741                         return true;
1742         }
1743         return false;
1744 }
1745
1746 static int virtnet_xdp(struct net_device *dev, struct netdev_xdp *xdp)
1747 {
1748         switch (xdp->command) {
1749         case XDP_SETUP_PROG:
1750                 return virtnet_xdp_set(dev, xdp->prog);
1751         case XDP_QUERY_PROG:
1752                 xdp->prog_attached = virtnet_xdp_query(dev);
1753                 return 0;
1754         default:
1755                 return -EINVAL;
1756         }
1757 }
1758
1759 static const struct net_device_ops virtnet_netdev = {
1760         .ndo_open            = virtnet_open,
1761         .ndo_stop            = virtnet_close,
1762         .ndo_start_xmit      = start_xmit,
1763         .ndo_validate_addr   = eth_validate_addr,
1764         .ndo_set_mac_address = virtnet_set_mac_address,
1765         .ndo_set_rx_mode     = virtnet_set_rx_mode,
1766         .ndo_get_stats64     = virtnet_stats,
1767         .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
1768         .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
1769 #ifdef CONFIG_NET_POLL_CONTROLLER
1770         .ndo_poll_controller = virtnet_netpoll,
1771 #endif
1772 #ifdef CONFIG_NET_RX_BUSY_POLL
1773         .ndo_busy_poll          = virtnet_busy_poll,
1774 #endif
1775         .ndo_xdp                = virtnet_xdp,
1776 };
1777
1778 static void virtnet_config_changed_work(struct work_struct *work)
1779 {
1780         struct virtnet_info *vi =
1781                 container_of(work, struct virtnet_info, config_work);
1782         u16 v;
1783
1784         if (virtio_cread_feature(vi->vdev, VIRTIO_NET_F_STATUS,
1785                                  struct virtio_net_config, status, &v) < 0)
1786                 return;
1787
1788         if (v & VIRTIO_NET_S_ANNOUNCE) {
1789                 netdev_notify_peers(vi->dev);
1790                 virtnet_ack_link_announce(vi);
1791         }
1792
1793         /* Ignore unknown (future) status bits */
1794         v &= VIRTIO_NET_S_LINK_UP;
1795
1796         if (vi->status == v)
1797                 return;
1798
1799         vi->status = v;
1800
1801         if (vi->status & VIRTIO_NET_S_LINK_UP) {
1802                 netif_carrier_on(vi->dev);
1803                 netif_tx_wake_all_queues(vi->dev);
1804         } else {
1805                 netif_carrier_off(vi->dev);
1806                 netif_tx_stop_all_queues(vi->dev);
1807         }
1808 }
1809
1810 static void virtnet_config_changed(struct virtio_device *vdev)
1811 {
1812         struct virtnet_info *vi = vdev->priv;
1813
1814         schedule_work(&vi->config_work);
1815 }
1816
1817 static void virtnet_free_queues(struct virtnet_info *vi)
1818 {
1819         int i;
1820
1821         for (i = 0; i < vi->max_queue_pairs; i++) {
1822                 napi_hash_del(&vi->rq[i].napi);
1823                 netif_napi_del(&vi->rq[i].napi);
1824         }
1825
1826         /* We called napi_hash_del() before netif_napi_del(),
1827          * we need to respect an RCU grace period before freeing vi->rq
1828          */
1829         synchronize_net();
1830
1831         kfree(vi->rq);
1832         kfree(vi->sq);
1833 }
1834
1835 static void free_receive_bufs(struct virtnet_info *vi)
1836 {
1837         struct bpf_prog *old_prog;
1838         int i;
1839
1840         rtnl_lock();
1841         for (i = 0; i < vi->max_queue_pairs; i++) {
1842                 while (vi->rq[i].pages)
1843                         __free_pages(get_a_page(&vi->rq[i], GFP_KERNEL), 0);
1844
1845                 old_prog = rtnl_dereference(vi->rq[i].xdp_prog);
1846                 RCU_INIT_POINTER(vi->rq[i].xdp_prog, NULL);
1847                 if (old_prog)
1848                         bpf_prog_put(old_prog);
1849         }
1850         rtnl_unlock();
1851 }
1852
1853 static void free_receive_page_frags(struct virtnet_info *vi)
1854 {
1855         int i;
1856         for (i = 0; i < vi->max_queue_pairs; i++)
1857                 if (vi->rq[i].alloc_frag.page)
1858                         put_page(vi->rq[i].alloc_frag.page);
1859 }
1860
1861 static bool is_xdp_queue(struct virtnet_info *vi, int q)
1862 {
1863         if (q < (vi->curr_queue_pairs - vi->xdp_queue_pairs))
1864                 return false;
1865         else if (q < vi->curr_queue_pairs)
1866                 return true;
1867         else
1868                 return false;
1869 }
1870
1871 static void free_unused_bufs(struct virtnet_info *vi)
1872 {
1873         void *buf;
1874         int i;
1875
1876         for (i = 0; i < vi->max_queue_pairs; i++) {
1877                 struct virtqueue *vq = vi->sq[i].vq;
1878                 while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) {
1879                         if (!is_xdp_queue(vi, i))
1880                                 dev_kfree_skb(buf);
1881                         else
1882                                 put_page(virt_to_head_page(buf));
1883                 }
1884         }
1885
1886         for (i = 0; i < vi->max_queue_pairs; i++) {
1887                 struct virtqueue *vq = vi->rq[i].vq;
1888
1889                 while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) {
1890                         if (vi->mergeable_rx_bufs) {
1891                                 unsigned long ctx = (unsigned long)buf;
1892                                 void *base = mergeable_ctx_to_buf_address(ctx);
1893                                 put_page(virt_to_head_page(base));
1894                         } else if (vi->big_packets) {
1895                                 give_pages(&vi->rq[i], buf);
1896                         } else {
1897                                 dev_kfree_skb(buf);
1898                         }
1899                 }
1900         }
1901 }
1902
1903 static void virtnet_del_vqs(struct virtnet_info *vi)
1904 {
1905         struct virtio_device *vdev = vi->vdev;
1906
1907         virtnet_clean_affinity(vi, -1);
1908
1909         vdev->config->del_vqs(vdev);
1910
1911         virtnet_free_queues(vi);
1912 }
1913
1914 static int virtnet_find_vqs(struct virtnet_info *vi)
1915 {
1916         vq_callback_t **callbacks;
1917         struct virtqueue **vqs;
1918         int ret = -ENOMEM;
1919         int i, total_vqs;
1920         const char **names;
1921
1922         /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by
1923          * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by
1924          * possible control vq.
1925          */
1926         total_vqs = vi->max_queue_pairs * 2 +
1927                     virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ);
1928
1929         /* Allocate space for find_vqs parameters */
1930         vqs = kzalloc(total_vqs * sizeof(*vqs), GFP_KERNEL);
1931         if (!vqs)
1932                 goto err_vq;
1933         callbacks = kmalloc(total_vqs * sizeof(*callbacks), GFP_KERNEL);
1934         if (!callbacks)
1935                 goto err_callback;
1936         names = kmalloc(total_vqs * sizeof(*names), GFP_KERNEL);
1937         if (!names)
1938                 goto err_names;
1939
1940         /* Parameters for control virtqueue, if any */
1941         if (vi->has_cvq) {
1942                 callbacks[total_vqs - 1] = NULL;
1943                 names[total_vqs - 1] = "control";
1944         }
1945
1946         /* Allocate/initialize parameters for send/receive virtqueues */
1947         for (i = 0; i < vi->max_queue_pairs; i++) {
1948                 callbacks[rxq2vq(i)] = skb_recv_done;
1949                 callbacks[txq2vq(i)] = skb_xmit_done;
1950                 sprintf(vi->rq[i].name, "input.%d", i);
1951                 sprintf(vi->sq[i].name, "output.%d", i);
1952                 names[rxq2vq(i)] = vi->rq[i].name;
1953                 names[txq2vq(i)] = vi->sq[i].name;
1954         }
1955
1956         ret = vi->vdev->config->find_vqs(vi->vdev, total_vqs, vqs, callbacks,
1957                                          names);
1958         if (ret)
1959                 goto err_find;
1960
1961         if (vi->has_cvq) {
1962                 vi->cvq = vqs[total_vqs - 1];
1963                 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
1964                         vi->dev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
1965         }
1966
1967         for (i = 0; i < vi->max_queue_pairs; i++) {
1968                 vi->rq[i].vq = vqs[rxq2vq(i)];
1969                 vi->sq[i].vq = vqs[txq2vq(i)];
1970         }
1971
1972         kfree(names);
1973         kfree(callbacks);
1974         kfree(vqs);
1975
1976         return 0;
1977
1978 err_find:
1979         kfree(names);
1980 err_names:
1981         kfree(callbacks);
1982 err_callback:
1983         kfree(vqs);
1984 err_vq:
1985         return ret;
1986 }
1987
1988 static int virtnet_alloc_queues(struct virtnet_info *vi)
1989 {
1990         int i;
1991
1992         vi->sq = kzalloc(sizeof(*vi->sq) * vi->max_queue_pairs, GFP_KERNEL);
1993         if (!vi->sq)
1994                 goto err_sq;
1995         vi->rq = kzalloc(sizeof(*vi->rq) * vi->max_queue_pairs, GFP_KERNEL);
1996         if (!vi->rq)
1997                 goto err_rq;
1998
1999         INIT_DELAYED_WORK(&vi->refill, refill_work);
2000         for (i = 0; i < vi->max_queue_pairs; i++) {
2001                 vi->rq[i].pages = NULL;
2002                 netif_napi_add(vi->dev, &vi->rq[i].napi, virtnet_poll,
2003                                napi_weight);
2004
2005                 sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg));
2006                 ewma_pkt_len_init(&vi->rq[i].mrg_avg_pkt_len);
2007                 sg_init_table(vi->sq[i].sg, ARRAY_SIZE(vi->sq[i].sg));
2008         }
2009
2010         return 0;
2011
2012 err_rq:
2013         kfree(vi->sq);
2014 err_sq:
2015         return -ENOMEM;
2016 }
2017
2018 static int init_vqs(struct virtnet_info *vi)
2019 {
2020         int ret;
2021
2022         /* Allocate send & receive queues */
2023         ret = virtnet_alloc_queues(vi);
2024         if (ret)
2025                 goto err;
2026
2027         ret = virtnet_find_vqs(vi);
2028         if (ret)
2029                 goto err_free;
2030
2031         get_online_cpus();
2032         virtnet_set_affinity(vi);
2033         put_online_cpus();
2034
2035         return 0;
2036
2037 err_free:
2038         virtnet_free_queues(vi);
2039 err:
2040         return ret;
2041 }
2042
2043 #ifdef CONFIG_SYSFS
2044 static ssize_t mergeable_rx_buffer_size_show(struct netdev_rx_queue *queue,
2045                 struct rx_queue_attribute *attribute, char *buf)
2046 {
2047         struct virtnet_info *vi = netdev_priv(queue->dev);
2048         unsigned int queue_index = get_netdev_rx_queue_index(queue);
2049         struct ewma_pkt_len *avg;
2050
2051         BUG_ON(queue_index >= vi->max_queue_pairs);
2052         avg = &vi->rq[queue_index].mrg_avg_pkt_len;
2053         return sprintf(buf, "%u\n", get_mergeable_buf_len(avg));
2054 }
2055
2056 static struct rx_queue_attribute mergeable_rx_buffer_size_attribute =
2057         __ATTR_RO(mergeable_rx_buffer_size);
2058
2059 static struct attribute *virtio_net_mrg_rx_attrs[] = {
2060         &mergeable_rx_buffer_size_attribute.attr,
2061         NULL
2062 };
2063
2064 static const struct attribute_group virtio_net_mrg_rx_group = {
2065         .name = "virtio_net",
2066         .attrs = virtio_net_mrg_rx_attrs
2067 };
2068 #endif
2069
2070 static bool virtnet_fail_on_feature(struct virtio_device *vdev,
2071                                     unsigned int fbit,
2072                                     const char *fname, const char *dname)
2073 {
2074         if (!virtio_has_feature(vdev, fbit))
2075                 return false;
2076
2077         dev_err(&vdev->dev, "device advertises feature %s but not %s",
2078                 fname, dname);
2079
2080         return true;
2081 }
2082
2083 #define VIRTNET_FAIL_ON(vdev, fbit, dbit)                       \
2084         virtnet_fail_on_feature(vdev, fbit, #fbit, dbit)
2085
2086 static bool virtnet_validate_features(struct virtio_device *vdev)
2087 {
2088         if (!virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ) &&
2089             (VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_RX,
2090                              "VIRTIO_NET_F_CTRL_VQ") ||
2091              VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_VLAN,
2092                              "VIRTIO_NET_F_CTRL_VQ") ||
2093              VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE,
2094                              "VIRTIO_NET_F_CTRL_VQ") ||
2095              VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_MQ, "VIRTIO_NET_F_CTRL_VQ") ||
2096              VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR,
2097                              "VIRTIO_NET_F_CTRL_VQ"))) {
2098                 return false;
2099         }
2100
2101         return true;
2102 }
2103
2104 #define MIN_MTU ETH_MIN_MTU
2105 #define MAX_MTU ETH_MAX_MTU
2106
2107 static int virtnet_probe(struct virtio_device *vdev)
2108 {
2109         int i, err;
2110         struct net_device *dev;
2111         struct virtnet_info *vi;
2112         u16 max_queue_pairs;
2113         int mtu;
2114
2115         if (!vdev->config->get) {
2116                 dev_err(&vdev->dev, "%s failure: config access disabled\n",
2117                         __func__);
2118                 return -EINVAL;
2119         }
2120
2121         if (!virtnet_validate_features(vdev))
2122                 return -EINVAL;
2123
2124         /* Find if host supports multiqueue virtio_net device */
2125         err = virtio_cread_feature(vdev, VIRTIO_NET_F_MQ,
2126                                    struct virtio_net_config,
2127                                    max_virtqueue_pairs, &max_queue_pairs);
2128
2129         /* We need at least 2 queue's */
2130         if (err || max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
2131             max_queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
2132             !virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
2133                 max_queue_pairs = 1;
2134
2135         /* Allocate ourselves a network device with room for our info */
2136         dev = alloc_etherdev_mq(sizeof(struct virtnet_info), max_queue_pairs);
2137         if (!dev)
2138                 return -ENOMEM;
2139
2140         /* Set up network device as normal. */
2141         dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE;
2142         dev->netdev_ops = &virtnet_netdev;
2143         dev->features = NETIF_F_HIGHDMA;
2144
2145         dev->ethtool_ops = &virtnet_ethtool_ops;
2146         SET_NETDEV_DEV(dev, &vdev->dev);
2147
2148         /* Do we support "hardware" checksums? */
2149         if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
2150                 /* This opens up the world of extra features. */
2151                 dev->hw_features |= NETIF_F_HW_CSUM | NETIF_F_SG;
2152                 if (csum)
2153                         dev->features |= NETIF_F_HW_CSUM | NETIF_F_SG;
2154
2155                 if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
2156                         dev->hw_features |= NETIF_F_TSO | NETIF_F_UFO
2157                                 | NETIF_F_TSO_ECN | NETIF_F_TSO6;
2158                 }
2159                 /* Individual feature bits: what can host handle? */
2160                 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
2161                         dev->hw_features |= NETIF_F_TSO;
2162                 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
2163                         dev->hw_features |= NETIF_F_TSO6;
2164                 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
2165                         dev->hw_features |= NETIF_F_TSO_ECN;
2166                 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
2167                         dev->hw_features |= NETIF_F_UFO;
2168
2169                 dev->features |= NETIF_F_GSO_ROBUST;
2170
2171                 if (gso)
2172                         dev->features |= dev->hw_features & (NETIF_F_ALL_TSO|NETIF_F_UFO);
2173                 /* (!csum && gso) case will be fixed by register_netdev() */
2174         }
2175         if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_CSUM))
2176                 dev->features |= NETIF_F_RXCSUM;
2177
2178         dev->vlan_features = dev->features;
2179
2180         /* MTU range: 68 - 65535 */
2181         dev->min_mtu = MIN_MTU;
2182         dev->max_mtu = MAX_MTU;
2183
2184         /* Configuration may specify what MAC to use.  Otherwise random. */
2185         if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC))
2186                 virtio_cread_bytes(vdev,
2187                                    offsetof(struct virtio_net_config, mac),
2188                                    dev->dev_addr, dev->addr_len);
2189         else
2190                 eth_hw_addr_random(dev);
2191
2192         /* Set up our device-specific information */
2193         vi = netdev_priv(dev);
2194         vi->dev = dev;
2195         vi->vdev = vdev;
2196         vdev->priv = vi;
2197         vi->stats = alloc_percpu(struct virtnet_stats);
2198         err = -ENOMEM;
2199         if (vi->stats == NULL)
2200                 goto free;
2201
2202         for_each_possible_cpu(i) {
2203                 struct virtnet_stats *virtnet_stats;
2204                 virtnet_stats = per_cpu_ptr(vi->stats, i);
2205                 u64_stats_init(&virtnet_stats->tx_syncp);
2206                 u64_stats_init(&virtnet_stats->rx_syncp);
2207         }
2208
2209         INIT_WORK(&vi->config_work, virtnet_config_changed_work);
2210
2211         /* If we can receive ANY GSO packets, we must allocate large ones. */
2212         if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
2213             virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) ||
2214             virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN) ||
2215             virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_UFO))
2216                 vi->big_packets = true;
2217
2218         if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
2219                 vi->mergeable_rx_bufs = true;
2220
2221         if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF) ||
2222             virtio_has_feature(vdev, VIRTIO_F_VERSION_1))
2223                 vi->hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
2224         else
2225                 vi->hdr_len = sizeof(struct virtio_net_hdr);
2226
2227         if (virtio_has_feature(vdev, VIRTIO_F_ANY_LAYOUT) ||
2228             virtio_has_feature(vdev, VIRTIO_F_VERSION_1))
2229                 vi->any_header_sg = true;
2230
2231         if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
2232                 vi->has_cvq = true;
2233
2234         if (virtio_has_feature(vdev, VIRTIO_NET_F_MTU)) {
2235                 mtu = virtio_cread16(vdev,
2236                                      offsetof(struct virtio_net_config,
2237                                               mtu));
2238                 if (mtu < dev->min_mtu) {
2239                         __virtio_clear_bit(vdev, VIRTIO_NET_F_MTU);
2240                 } else {
2241                         dev->mtu = mtu;
2242                         dev->max_mtu = mtu;
2243                 }
2244         }
2245
2246         if (vi->any_header_sg)
2247                 dev->needed_headroom = vi->hdr_len;
2248
2249         /* Enable multiqueue by default */
2250         if (num_online_cpus() >= max_queue_pairs)
2251                 vi->curr_queue_pairs = max_queue_pairs;
2252         else
2253                 vi->curr_queue_pairs = num_online_cpus();
2254         vi->max_queue_pairs = max_queue_pairs;
2255
2256         /* Allocate/initialize the rx/tx queues, and invoke find_vqs */
2257         err = init_vqs(vi);
2258         if (err)
2259                 goto free_stats;
2260
2261 #ifdef CONFIG_SYSFS
2262         if (vi->mergeable_rx_bufs)
2263                 dev->sysfs_rx_queue_group = &virtio_net_mrg_rx_group;
2264 #endif
2265         netif_set_real_num_tx_queues(dev, vi->curr_queue_pairs);
2266         netif_set_real_num_rx_queues(dev, vi->curr_queue_pairs);
2267
2268         virtnet_init_settings(dev);
2269
2270         err = register_netdev(dev);
2271         if (err) {
2272                 pr_debug("virtio_net: registering device failed\n");
2273                 goto free_vqs;
2274         }
2275
2276         virtio_device_ready(vdev);
2277
2278         err = virtnet_cpu_notif_add(vi);
2279         if (err) {
2280                 pr_debug("virtio_net: registering cpu notifier failed\n");
2281                 goto free_unregister_netdev;
2282         }
2283
2284         rtnl_lock();
2285         virtnet_set_queues(vi, vi->curr_queue_pairs);
2286         rtnl_unlock();
2287
2288         /* Assume link up if device can't report link status,
2289            otherwise get link status from config. */
2290         if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) {
2291                 netif_carrier_off(dev);
2292                 schedule_work(&vi->config_work);
2293         } else {
2294                 vi->status = VIRTIO_NET_S_LINK_UP;
2295                 netif_carrier_on(dev);
2296         }
2297
2298         pr_debug("virtnet: registered device %s with %d RX and TX vq's\n",
2299                  dev->name, max_queue_pairs);
2300
2301         return 0;
2302
2303 free_unregister_netdev:
2304         vi->vdev->config->reset(vdev);
2305
2306         unregister_netdev(dev);
2307 free_vqs:
2308         cancel_delayed_work_sync(&vi->refill);
2309         free_receive_page_frags(vi);
2310         virtnet_del_vqs(vi);
2311 free_stats:
2312         free_percpu(vi->stats);
2313 free:
2314         free_netdev(dev);
2315         return err;
2316 }
2317
2318 static void remove_vq_common(struct virtnet_info *vi)
2319 {
2320         vi->vdev->config->reset(vi->vdev);
2321
2322         /* Free unused buffers in both send and recv, if any. */
2323         free_unused_bufs(vi);
2324
2325         free_receive_bufs(vi);
2326
2327         free_receive_page_frags(vi);
2328
2329         virtnet_del_vqs(vi);
2330 }
2331
2332 static void virtnet_remove(struct virtio_device *vdev)
2333 {
2334         struct virtnet_info *vi = vdev->priv;
2335
2336         virtnet_cpu_notif_remove(vi);
2337
2338         /* Make sure no work handler is accessing the device. */
2339         flush_work(&vi->config_work);
2340
2341         unregister_netdev(vi->dev);
2342
2343         remove_vq_common(vi);
2344
2345         free_percpu(vi->stats);
2346         free_netdev(vi->dev);
2347 }
2348
2349 #ifdef CONFIG_PM_SLEEP
2350 static int virtnet_freeze(struct virtio_device *vdev)
2351 {
2352         struct virtnet_info *vi = vdev->priv;
2353         int i;
2354
2355         virtnet_cpu_notif_remove(vi);
2356
2357         /* Make sure no work handler is accessing the device */
2358         flush_work(&vi->config_work);
2359
2360         netif_device_detach(vi->dev);
2361         cancel_delayed_work_sync(&vi->refill);
2362
2363         if (netif_running(vi->dev)) {
2364                 for (i = 0; i < vi->max_queue_pairs; i++)
2365                         napi_disable(&vi->rq[i].napi);
2366         }
2367
2368         remove_vq_common(vi);
2369
2370         return 0;
2371 }
2372
2373 static int virtnet_restore(struct virtio_device *vdev)
2374 {
2375         struct virtnet_info *vi = vdev->priv;
2376         int err, i;
2377
2378         err = init_vqs(vi);
2379         if (err)
2380                 return err;
2381
2382         virtio_device_ready(vdev);
2383
2384         if (netif_running(vi->dev)) {
2385                 for (i = 0; i < vi->curr_queue_pairs; i++)
2386                         if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL))
2387                                 schedule_delayed_work(&vi->refill, 0);
2388
2389                 for (i = 0; i < vi->max_queue_pairs; i++)
2390                         virtnet_napi_enable(&vi->rq[i]);
2391         }
2392
2393         netif_device_attach(vi->dev);
2394
2395         rtnl_lock();
2396         virtnet_set_queues(vi, vi->curr_queue_pairs);
2397         rtnl_unlock();
2398
2399         err = virtnet_cpu_notif_add(vi);
2400         if (err)
2401                 return err;
2402
2403         return 0;
2404 }
2405 #endif
2406
2407 static struct virtio_device_id id_table[] = {
2408         { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
2409         { 0 },
2410 };
2411
2412 #define VIRTNET_FEATURES \
2413         VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM, \
2414         VIRTIO_NET_F_MAC, \
2415         VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6, \
2416         VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6, \
2417         VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO, \
2418         VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ, \
2419         VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN, \
2420         VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ, \
2421         VIRTIO_NET_F_CTRL_MAC_ADDR, \
2422         VIRTIO_NET_F_MTU
2423
2424 static unsigned int features[] = {
2425         VIRTNET_FEATURES,
2426 };
2427
2428 static unsigned int features_legacy[] = {
2429         VIRTNET_FEATURES,
2430         VIRTIO_NET_F_GSO,
2431         VIRTIO_F_ANY_LAYOUT,
2432 };
2433
2434 static struct virtio_driver virtio_net_driver = {
2435         .feature_table = features,
2436         .feature_table_size = ARRAY_SIZE(features),
2437         .feature_table_legacy = features_legacy,
2438         .feature_table_size_legacy = ARRAY_SIZE(features_legacy),
2439         .driver.name =  KBUILD_MODNAME,
2440         .driver.owner = THIS_MODULE,
2441         .id_table =     id_table,
2442         .probe =        virtnet_probe,
2443         .remove =       virtnet_remove,
2444         .config_changed = virtnet_config_changed,
2445 #ifdef CONFIG_PM_SLEEP
2446         .freeze =       virtnet_freeze,
2447         .restore =      virtnet_restore,
2448 #endif
2449 };
2450
2451 static __init int virtio_net_driver_init(void)
2452 {
2453         int ret;
2454
2455         ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "AP_VIRT_NET_ONLINE",
2456                                       virtnet_cpu_online,
2457                                       virtnet_cpu_down_prep);
2458         if (ret < 0)
2459                 goto out;
2460         virtionet_online = ret;
2461         ret = cpuhp_setup_state_multi(CPUHP_VIRT_NET_DEAD, "VIRT_NET_DEAD",
2462                                       NULL, virtnet_cpu_dead);
2463         if (ret)
2464                 goto err_dead;
2465
2466         ret = register_virtio_driver(&virtio_net_driver);
2467         if (ret)
2468                 goto err_virtio;
2469         return 0;
2470 err_virtio:
2471         cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD);
2472 err_dead:
2473         cpuhp_remove_multi_state(virtionet_online);
2474 out:
2475         return ret;
2476 }
2477 module_init(virtio_net_driver_init);
2478
2479 static __exit void virtio_net_driver_exit(void)
2480 {
2481         cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD);
2482         cpuhp_remove_multi_state(virtionet_online);
2483         unregister_virtio_driver(&virtio_net_driver);
2484 }
2485 module_exit(virtio_net_driver_exit);
2486
2487 MODULE_DEVICE_TABLE(virtio, id_table);
2488 MODULE_DESCRIPTION("Virtio network driver");
2489 MODULE_LICENSE("GPL");