linux/drivers/net/virtio_net.c
<<
>>
Prefs
   1/* A network driver using virtio.
   2 *
   3 * Copyright 2007 Rusty Russell <rusty@rustcorp.com.au> IBM Corporation
   4 *
   5 * This program is free software; you can redistribute it and/or modify
   6 * it under the terms of the GNU General Public License as published by
   7 * the Free Software Foundation; either version 2 of the License, or
   8 * (at your option) any later version.
   9 *
  10 * This program is distributed in the hope that it will be useful,
  11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13 * GNU General Public License for more details.
  14 *
  15 * You should have received a copy of the GNU General Public License
  16 * along with this program; if not, write to the Free Software
  17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  18 */
  19//#define DEBUG
  20#include <linux/netdevice.h>
  21#include <linux/etherdevice.h>
  22#include <linux/ethtool.h>
  23#include <linux/module.h>
  24#include <linux/virtio.h>
  25#include <linux/virtio_net.h>
  26#include <linux/scatterlist.h>
  27#include <linux/if_vlan.h>
  28#include <linux/slab.h>
  29#include <linux/cpu.h>
  30
  31static int napi_weight = NAPI_POLL_WEIGHT;
  32module_param(napi_weight, int, 0444);
  33
  34static bool csum = true, gso = true;
  35module_param(csum, bool, 0444);
  36module_param(gso, bool, 0444);
  37
  38/* FIXME: MTU in config. */
  39#define MAX_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
  40#define GOOD_COPY_LEN   128
  41
  42#define VIRTNET_DRIVER_VERSION "1.0.0"
  43
  44struct virtnet_stats {
  45        struct u64_stats_sync tx_syncp;
  46        struct u64_stats_sync rx_syncp;
  47        u64 tx_bytes;
  48        u64 tx_packets;
  49
  50        u64 rx_bytes;
  51        u64 rx_packets;
  52};
  53
  54/* Internal representation of a send virtqueue */
  55struct send_queue {
  56        /* Virtqueue associated with this send _queue */
  57        struct virtqueue *vq;
  58
  59        /* TX: fragments + linear part + virtio header */
  60        struct scatterlist sg[MAX_SKB_FRAGS + 2];
  61
  62        /* Name of the send queue: output.$index */
  63        char name[40];
  64};
  65
  66/* Internal representation of a receive virtqueue */
  67struct receive_queue {
  68        /* Virtqueue associated with this receive_queue */
  69        struct virtqueue *vq;
  70
  71        struct napi_struct napi;
  72
  73        /* Number of input buffers, and max we've ever had. */
  74        unsigned int num, max;
  75
  76        /* Chain pages by the private ptr. */
  77        struct page *pages;
  78
  79        /* RX: fragments + linear part + virtio header */
  80        struct scatterlist sg[MAX_SKB_FRAGS + 2];
  81
  82        /* Name of this receive queue: input.$index */
  83        char name[40];
  84};
  85
  86struct virtnet_info {
  87        struct virtio_device *vdev;
  88        struct virtqueue *cvq;
  89        struct net_device *dev;
  90        struct send_queue *sq;
  91        struct receive_queue *rq;
  92        unsigned int status;
  93
  94        /* Max # of queue pairs supported by the device */
  95        u16 max_queue_pairs;
  96
  97        /* # of queue pairs currently used by the driver */
  98        u16 curr_queue_pairs;
  99
 100        /* I like... big packets and I cannot lie! */
 101        bool big_packets;
 102
 103        /* Host will merge rx buffers for big packets (shake it! shake it!) */
 104        bool mergeable_rx_bufs;
 105
 106        /* Has control virtqueue */
 107        bool has_cvq;
 108
 109        /* enable config space updates */
 110        bool config_enable;
 111
 112        /* Active statistics */
 113        struct virtnet_stats __percpu *stats;
 114
 115        /* Work struct for refilling if we run low on memory. */
 116        struct delayed_work refill;
 117
 118        /* Work struct for config space updates */
 119        struct work_struct config_work;
 120
 121        /* Lock for config space updates */
 122        struct mutex config_lock;
 123
 124        /* Does the affinity hint is set for virtqueues? */
 125        bool affinity_hint_set;
 126
 127        /* Per-cpu variable to show the mapping from CPU to virtqueue */
 128        int __percpu *vq_index;
 129
 130        /* CPU hot plug notifier */
 131        struct notifier_block nb;
 132};
 133
 134struct skb_vnet_hdr {
 135        union {
 136                struct virtio_net_hdr hdr;
 137                struct virtio_net_hdr_mrg_rxbuf mhdr;
 138        };
 139};
 140
 141struct padded_vnet_hdr {
 142        struct virtio_net_hdr hdr;
 143        /*
 144         * virtio_net_hdr should be in a separated sg buffer because of a
 145         * QEMU bug, and data sg buffer shares same page with this header sg.
 146         * This padding makes next sg 16 byte aligned after virtio_net_hdr.
 147         */
 148        char padding[6];
 149};
 150
 151/* Converting between virtqueue no. and kernel tx/rx queue no.
 152 * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq
 153 */
 154static int vq2txq(struct virtqueue *vq)
 155{
 156        return (vq->index - 1) / 2;
 157}
 158
 159static int txq2vq(int txq)
 160{
 161        return txq * 2 + 1;
 162}
 163
 164static int vq2rxq(struct virtqueue *vq)
 165{
 166        return vq->index / 2;
 167}
 168
 169static int rxq2vq(int rxq)
 170{
 171        return rxq * 2;
 172}
 173
 174static inline struct skb_vnet_hdr *skb_vnet_hdr(struct sk_buff *skb)
 175{
 176        return (struct skb_vnet_hdr *)skb->cb;
 177}
 178
 179/*
 180 * private is used to chain pages for big packets, put the whole
 181 * most recent used list in the beginning for reuse
 182 */
 183static void give_pages(struct receive_queue *rq, struct page *page)
 184{
 185        struct page *end;
 186
 187        /* Find end of list, sew whole thing into vi->rq.pages. */
 188        for (end = page; end->private; end = (struct page *)end->private);
 189        end->private = (unsigned long)rq->pages;
 190        rq->pages = page;
 191}
 192
 193static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask)
 194{
 195        struct page *p = rq->pages;
 196
 197        if (p) {
 198                rq->pages = (struct page *)p->private;
 199                /* clear private here, it is used to chain pages */
 200                p->private = 0;
 201        } else
 202                p = alloc_page(gfp_mask);
 203        return p;
 204}
 205
 206static void skb_xmit_done(struct virtqueue *vq)
 207{
 208        struct virtnet_info *vi = vq->vdev->priv;
 209
 210        /* Suppress further interrupts. */
 211        virtqueue_disable_cb(vq);
 212
 213        /* We were probably waiting for more output buffers. */
 214        netif_wake_subqueue(vi->dev, vq2txq(vq));
 215}
 216
 217static void set_skb_frag(struct sk_buff *skb, struct page *page,
 218                         unsigned int offset, unsigned int *len)
 219{
 220        int size = min((unsigned)PAGE_SIZE - offset, *len);
 221        int i = skb_shinfo(skb)->nr_frags;
 222
 223        __skb_fill_page_desc(skb, i, page, offset, size);
 224
 225        skb->data_len += size;
 226        skb->len += size;
 227        skb->truesize += PAGE_SIZE;
 228        skb_shinfo(skb)->nr_frags++;
 229        skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
 230        *len -= size;
 231}
 232
 233/* Called from bottom half context */
 234static struct sk_buff *page_to_skb(struct receive_queue *rq,
 235                                   struct page *page, unsigned int len)
 236{
 237        struct virtnet_info *vi = rq->vq->vdev->priv;
 238        struct sk_buff *skb;
 239        struct skb_vnet_hdr *hdr;
 240        unsigned int copy, hdr_len, offset;
 241        char *p;
 242
 243        p = page_address(page);
 244
 245        /* copy small packet so we can reuse these pages for small data */
 246        skb = netdev_alloc_skb_ip_align(vi->dev, GOOD_COPY_LEN);
 247        if (unlikely(!skb))
 248                return NULL;
 249
 250        hdr = skb_vnet_hdr(skb);
 251
 252        if (vi->mergeable_rx_bufs) {
 253                hdr_len = sizeof hdr->mhdr;
 254                offset = hdr_len;
 255        } else {
 256                hdr_len = sizeof hdr->hdr;
 257                offset = sizeof(struct padded_vnet_hdr);
 258        }
 259
 260        memcpy(hdr, p, hdr_len);
 261
 262        len -= hdr_len;
 263        p += offset;
 264
 265        copy = len;
 266        if (copy > skb_tailroom(skb))
 267                copy = skb_tailroom(skb);
 268        memcpy(skb_put(skb, copy), p, copy);
 269
 270        len -= copy;
 271        offset += copy;
 272
 273        /*
 274         * Verify that we can indeed put this data into a skb.
 275         * This is here to handle cases when the device erroneously
 276         * tries to receive more than is possible. This is usually
 277         * the case of a broken device.
 278         */
 279        if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) {
 280                net_dbg_ratelimited("%s: too much data\n", skb->dev->name);
 281                dev_kfree_skb(skb);
 282                return NULL;
 283        }
 284
 285        while (len) {
 286                set_skb_frag(skb, page, offset, &len);
 287                page = (struct page *)page->private;
 288                offset = 0;
 289        }
 290
 291        if (page)
 292                give_pages(rq, page);
 293
 294        return skb;
 295}
 296
 297static int receive_mergeable(struct receive_queue *rq, struct sk_buff *skb)
 298{
 299        struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
 300        struct page *page;
 301        int num_buf, i, len;
 302
 303        num_buf = hdr->mhdr.num_buffers;
 304        while (--num_buf) {
 305                i = skb_shinfo(skb)->nr_frags;
 306                if (i >= MAX_SKB_FRAGS) {
 307                        pr_debug("%s: packet too long\n", skb->dev->name);
 308                        skb->dev->stats.rx_length_errors++;
 309                        return -EINVAL;
 310                }
 311                page = virtqueue_get_buf(rq->vq, &len);
 312                if (!page) {
 313                        pr_debug("%s: rx error: %d buffers missing\n",
 314                                 skb->dev->name, hdr->mhdr.num_buffers);
 315                        skb->dev->stats.rx_length_errors++;
 316                        return -EINVAL;
 317                }
 318
 319                if (len > PAGE_SIZE)
 320                        len = PAGE_SIZE;
 321
 322                set_skb_frag(skb, page, 0, &len);
 323
 324                --rq->num;
 325        }
 326        return 0;
 327}
 328
 329static void receive_buf(struct receive_queue *rq, void *buf, unsigned int len)
 330{
 331        struct virtnet_info *vi = rq->vq->vdev->priv;
 332        struct net_device *dev = vi->dev;
 333        struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
 334        struct sk_buff *skb;
 335        struct page *page;
 336        struct skb_vnet_hdr *hdr;
 337
 338        if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) {
 339                pr_debug("%s: short packet %i\n", dev->name, len);
 340                dev->stats.rx_length_errors++;
 341                if (vi->mergeable_rx_bufs || vi->big_packets)
 342                        give_pages(rq, buf);
 343                else
 344                        dev_kfree_skb(buf);
 345                return;
 346        }
 347
 348        if (!vi->mergeable_rx_bufs && !vi->big_packets) {
 349                skb = buf;
 350                len -= sizeof(struct virtio_net_hdr);
 351                skb_trim(skb, len);
 352        } else {
 353                page = buf;
 354                skb = page_to_skb(rq, page, len);
 355                if (unlikely(!skb)) {
 356                        dev->stats.rx_dropped++;
 357                        give_pages(rq, page);
 358                        return;
 359                }
 360                if (vi->mergeable_rx_bufs)
 361                        if (receive_mergeable(rq, skb)) {
 362                                dev_kfree_skb(skb);
 363                                return;
 364                        }
 365        }
 366
 367        hdr = skb_vnet_hdr(skb);
 368
 369        u64_stats_update_begin(&stats->rx_syncp);
 370        stats->rx_bytes += skb->len;
 371        stats->rx_packets++;
 372        u64_stats_update_end(&stats->rx_syncp);
 373
 374        if (hdr->hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
 375                pr_debug("Needs csum!\n");
 376                if (!skb_partial_csum_set(skb,
 377                                          hdr->hdr.csum_start,
 378                                          hdr->hdr.csum_offset))
 379                        goto frame_err;
 380        } else if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID) {
 381                skb->ip_summed = CHECKSUM_UNNECESSARY;
 382        }
 383
 384        skb->protocol = eth_type_trans(skb, dev);
 385        pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
 386                 ntohs(skb->protocol), skb->len, skb->pkt_type);
 387
 388        if (hdr->hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
 389                pr_debug("GSO!\n");
 390                switch (hdr->hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
 391                case VIRTIO_NET_HDR_GSO_TCPV4:
 392                        skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
 393                        break;
 394                case VIRTIO_NET_HDR_GSO_UDP:
 395                        skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
 396                        break;
 397                case VIRTIO_NET_HDR_GSO_TCPV6:
 398                        skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
 399                        break;
 400                default:
 401                        net_warn_ratelimited("%s: bad gso type %u.\n",
 402                                             dev->name, hdr->hdr.gso_type);
 403                        goto frame_err;
 404                }
 405
 406                if (hdr->hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
 407                        skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
 408
 409                skb_shinfo(skb)->gso_size = hdr->hdr.gso_size;
 410                if (skb_shinfo(skb)->gso_size == 0) {
 411                        net_warn_ratelimited("%s: zero gso size.\n", dev->name);
 412                        goto frame_err;
 413                }
 414
 415                /* Header must be checked, and gso_segs computed. */
 416                skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
 417                skb_shinfo(skb)->gso_segs = 0;
 418        }
 419
 420        netif_receive_skb(skb);
 421        return;
 422
 423frame_err:
 424        dev->stats.rx_frame_errors++;
 425        dev_kfree_skb(skb);
 426}
 427
 428static int add_recvbuf_small(struct receive_queue *rq, gfp_t gfp)
 429{
 430        struct virtnet_info *vi = rq->vq->vdev->priv;
 431        struct sk_buff *skb;
 432        struct skb_vnet_hdr *hdr;
 433        int err;
 434
 435        skb = __netdev_alloc_skb_ip_align(vi->dev, MAX_PACKET_LEN, gfp);
 436        if (unlikely(!skb))
 437                return -ENOMEM;
 438
 439        skb_put(skb, MAX_PACKET_LEN);
 440
 441        hdr = skb_vnet_hdr(skb);
 442        sg_set_buf(rq->sg, &hdr->hdr, sizeof hdr->hdr);
 443
 444        skb_to_sgvec(skb, rq->sg + 1, 0, skb->len);
 445
 446        err = virtqueue_add_inbuf(rq->vq, rq->sg, 2, skb, gfp);
 447        if (err < 0)
 448                dev_kfree_skb(skb);
 449
 450        return err;
 451}
 452
 453static int add_recvbuf_big(struct receive_queue *rq, gfp_t gfp)
 454{
 455        struct page *first, *list = NULL;
 456        char *p;
 457        int i, err, offset;
 458
 459        /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */
 460        for (i = MAX_SKB_FRAGS + 1; i > 1; --i) {
 461                first = get_a_page(rq, gfp);
 462                if (!first) {
 463                        if (list)
 464                                give_pages(rq, list);
 465                        return -ENOMEM;
 466                }
 467                sg_set_buf(&rq->sg[i], page_address(first), PAGE_SIZE);
 468
 469                /* chain new page in list head to match sg */
 470                first->private = (unsigned long)list;
 471                list = first;
 472        }
 473
 474        first = get_a_page(rq, gfp);
 475        if (!first) {
 476                give_pages(rq, list);
 477                return -ENOMEM;
 478        }
 479        p = page_address(first);
 480
 481        /* rq->sg[0], rq->sg[1] share the same page */
 482        /* a separated rq->sg[0] for virtio_net_hdr only due to QEMU bug */
 483        sg_set_buf(&rq->sg[0], p, sizeof(struct virtio_net_hdr));
 484
 485        /* rq->sg[1] for data packet, from offset */
 486        offset = sizeof(struct padded_vnet_hdr);
 487        sg_set_buf(&rq->sg[1], p + offset, PAGE_SIZE - offset);
 488
 489        /* chain first in list head */
 490        first->private = (unsigned long)list;
 491        err = virtqueue_add_inbuf(rq->vq, rq->sg, MAX_SKB_FRAGS + 2,
 492                                  first, gfp);
 493        if (err < 0)
 494                give_pages(rq, first);
 495
 496        return err;
 497}
 498
 499static int add_recvbuf_mergeable(struct receive_queue *rq, gfp_t gfp)
 500{
 501        struct page *page;
 502        int err;
 503
 504        page = get_a_page(rq, gfp);
 505        if (!page)
 506                return -ENOMEM;
 507
 508        sg_init_one(rq->sg, page_address(page), PAGE_SIZE);
 509
 510        err = virtqueue_add_inbuf(rq->vq, rq->sg, 1, page, gfp);
 511        if (err < 0)
 512                give_pages(rq, page);
 513
 514        return err;
 515}
 516
 517/*
 518 * Returns false if we couldn't fill entirely (OOM).
 519 *
 520 * Normally run in the receive path, but can also be run from ndo_open
 521 * before we're receiving packets, or from refill_work which is
 522 * careful to disable receiving (using napi_disable).
 523 */
 524static bool try_fill_recv(struct receive_queue *rq, gfp_t gfp)
 525{
 526        struct virtnet_info *vi = rq->vq->vdev->priv;
 527        int err;
 528        bool oom;
 529
 530        do {
 531                if (vi->mergeable_rx_bufs)
 532                        err = add_recvbuf_mergeable(rq, gfp);
 533                else if (vi->big_packets)
 534                        err = add_recvbuf_big(rq, gfp);
 535                else
 536                        err = add_recvbuf_small(rq, gfp);
 537
 538                oom = err == -ENOMEM;
 539                if (err)
 540                        break;
 541                ++rq->num;
 542        } while (rq->vq->num_free);
 543        if (unlikely(rq->num > rq->max))
 544                rq->max = rq->num;
 545        virtqueue_kick(rq->vq);
 546        return !oom;
 547}
 548
 549static void skb_recv_done(struct virtqueue *rvq)
 550{
 551        struct virtnet_info *vi = rvq->vdev->priv;
 552        struct receive_queue *rq = &vi->rq[vq2rxq(rvq)];
 553
 554        /* Schedule NAPI, Suppress further interrupts if successful. */
 555        if (napi_schedule_prep(&rq->napi)) {
 556                virtqueue_disable_cb(rvq);
 557                __napi_schedule(&rq->napi);
 558        }
 559}
 560
 561static void virtnet_napi_enable(struct receive_queue *rq)
 562{
 563        napi_enable(&rq->napi);
 564
 565        /* If all buffers were filled by other side before we napi_enabled, we
 566         * won't get another interrupt, so process any outstanding packets
 567         * now.  virtnet_poll wants re-enable the queue, so we disable here.
 568         * We synchronize against interrupts via NAPI_STATE_SCHED */
 569        if (napi_schedule_prep(&rq->napi)) {
 570                virtqueue_disable_cb(rq->vq);
 571                local_bh_disable();
 572                __napi_schedule(&rq->napi);
 573                local_bh_enable();
 574        }
 575}
 576
 577static void refill_work(struct work_struct *work)
 578{
 579        struct virtnet_info *vi =
 580                container_of(work, struct virtnet_info, refill.work);
 581        bool still_empty;
 582        int i;
 583
 584        for (i = 0; i < vi->curr_queue_pairs; i++) {
 585                struct receive_queue *rq = &vi->rq[i];
 586
 587                napi_disable(&rq->napi);
 588                still_empty = !try_fill_recv(rq, GFP_KERNEL);
 589                virtnet_napi_enable(rq);
 590
 591                /* In theory, this can happen: if we don't get any buffers in
 592                 * we will *never* try to fill again.
 593                 */
 594                if (still_empty)
 595                        schedule_delayed_work(&vi->refill, HZ/2);
 596        }
 597}
 598
 599static int virtnet_poll(struct napi_struct *napi, int budget)
 600{
 601        struct receive_queue *rq =
 602                container_of(napi, struct receive_queue, napi);
 603        struct virtnet_info *vi = rq->vq->vdev->priv;
 604        void *buf;
 605        unsigned int r, len, received = 0;
 606
 607again:
 608        while (received < budget &&
 609               (buf = virtqueue_get_buf(rq->vq, &len)) != NULL) {
 610                receive_buf(rq, buf, len);
 611                --rq->num;
 612                received++;
 613        }
 614
 615        if (rq->num < rq->max / 2) {
 616                if (!try_fill_recv(rq, GFP_ATOMIC))
 617                        schedule_delayed_work(&vi->refill, 0);
 618        }
 619
 620        /* Out of packets? */
 621        if (received < budget) {
 622                r = virtqueue_enable_cb_prepare(rq->vq);
 623                napi_complete(napi);
 624                if (unlikely(virtqueue_poll(rq->vq, r)) &&
 625                    napi_schedule_prep(napi)) {
 626                        virtqueue_disable_cb(rq->vq);
 627                        __napi_schedule(napi);
 628                        goto again;
 629                }
 630        }
 631
 632        return received;
 633}
 634
 635static int virtnet_open(struct net_device *dev)
 636{
 637        struct virtnet_info *vi = netdev_priv(dev);
 638        int i;
 639
 640        for (i = 0; i < vi->max_queue_pairs; i++) {
 641                if (i < vi->curr_queue_pairs)
 642                        /* Make sure we have some buffers: if oom use wq. */
 643                        if (!try_fill_recv(&vi->rq[i], GFP_KERNEL))
 644                                schedule_delayed_work(&vi->refill, 0);
 645                virtnet_napi_enable(&vi->rq[i]);
 646        }
 647
 648        return 0;
 649}
 650
 651static void free_old_xmit_skbs(struct send_queue *sq)
 652{
 653        struct sk_buff *skb;
 654        unsigned int len;
 655        struct virtnet_info *vi = sq->vq->vdev->priv;
 656        struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
 657
 658        while ((skb = virtqueue_get_buf(sq->vq, &len)) != NULL) {
 659                pr_debug("Sent skb %p\n", skb);
 660
 661                u64_stats_update_begin(&stats->tx_syncp);
 662                stats->tx_bytes += skb->len;
 663                stats->tx_packets++;
 664                u64_stats_update_end(&stats->tx_syncp);
 665
 666                dev_kfree_skb_any(skb);
 667        }
 668}
 669
 670static int xmit_skb(struct send_queue *sq, struct sk_buff *skb)
 671{
 672        struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
 673        const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
 674        struct virtnet_info *vi = sq->vq->vdev->priv;
 675        unsigned num_sg;
 676
 677        pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
 678
 679        if (skb->ip_summed == CHECKSUM_PARTIAL) {
 680                hdr->hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
 681                hdr->hdr.csum_start = skb_checksum_start_offset(skb);
 682                hdr->hdr.csum_offset = skb->csum_offset;
 683        } else {
 684                hdr->hdr.flags = 0;
 685                hdr->hdr.csum_offset = hdr->hdr.csum_start = 0;
 686        }
 687
 688        if (skb_is_gso(skb)) {
 689                hdr->hdr.hdr_len = skb_headlen(skb);
 690                hdr->hdr.gso_size = skb_shinfo(skb)->gso_size;
 691                if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
 692                        hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
 693                else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
 694                        hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
 695                else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
 696                        hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
 697                else
 698                        BUG();
 699                if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
 700                        hdr->hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
 701        } else {
 702                hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
 703                hdr->hdr.gso_size = hdr->hdr.hdr_len = 0;
 704        }
 705
 706        hdr->mhdr.num_buffers = 0;
 707
 708        /* Encode metadata header at front. */
 709        if (vi->mergeable_rx_bufs)
 710                sg_set_buf(sq->sg, &hdr->mhdr, sizeof hdr->mhdr);
 711        else
 712                sg_set_buf(sq->sg, &hdr->hdr, sizeof hdr->hdr);
 713
 714        num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len) + 1;
 715        return virtqueue_add_outbuf(sq->vq, sq->sg, num_sg, skb, GFP_ATOMIC);
 716}
 717
 718static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
 719{
 720        struct virtnet_info *vi = netdev_priv(dev);
 721        int qnum = skb_get_queue_mapping(skb);
 722        struct send_queue *sq = &vi->sq[qnum];
 723        int err;
 724
 725        /* Free up any pending old buffers before queueing new ones. */
 726        free_old_xmit_skbs(sq);
 727
 728        /* Try to transmit */
 729        err = xmit_skb(sq, skb);
 730
 731        /* This should not happen! */
 732        if (unlikely(err)) {
 733                dev->stats.tx_fifo_errors++;
 734                if (net_ratelimit())
 735                        dev_warn(&dev->dev,
 736                                 "Unexpected TXQ (%d) queue failure: %d\n", qnum, err);
 737                dev->stats.tx_dropped++;
 738                kfree_skb(skb);
 739                return NETDEV_TX_OK;
 740        }
 741        virtqueue_kick(sq->vq);
 742
 743        /* Don't wait up for transmitted skbs to be freed. */
 744        skb_orphan(skb);
 745        nf_reset(skb);
 746
 747        /* Apparently nice girls don't return TX_BUSY; stop the queue
 748         * before it gets out of hand.  Naturally, this wastes entries. */
 749        if (sq->vq->num_free < 2+MAX_SKB_FRAGS) {
 750                netif_stop_subqueue(dev, qnum);
 751                if (unlikely(!virtqueue_enable_cb_delayed(sq->vq))) {
 752                        /* More just got used, free them then recheck. */
 753                        free_old_xmit_skbs(sq);
 754                        if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) {
 755                                netif_start_subqueue(dev, qnum);
 756                                virtqueue_disable_cb(sq->vq);
 757                        }
 758                }
 759        }
 760
 761        return NETDEV_TX_OK;
 762}
 763
 764/*
 765 * Send command via the control virtqueue and check status.  Commands
 766 * supported by the hypervisor, as indicated by feature bits, should
 767 * never fail unless improperly formated.
 768 */
 769static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
 770                                 struct scatterlist *out,
 771                                 struct scatterlist *in)
 772{
 773        struct scatterlist *sgs[4], hdr, stat;
 774        struct virtio_net_ctrl_hdr ctrl;
 775        virtio_net_ctrl_ack status = ~0;
 776        unsigned out_num = 0, in_num = 0, tmp;
 777
 778        /* Caller should know better */
 779        BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ));
 780
 781        ctrl.class = class;
 782        ctrl.cmd = cmd;
 783        /* Add header */
 784        sg_init_one(&hdr, &ctrl, sizeof(ctrl));
 785        sgs[out_num++] = &hdr;
 786
 787        if (out)
 788                sgs[out_num++] = out;
 789        if (in)
 790                sgs[out_num + in_num++] = in;
 791
 792        /* Add return status. */
 793        sg_init_one(&stat, &status, sizeof(status));
 794        sgs[out_num + in_num++] = &stat;
 795
 796        BUG_ON(out_num + in_num > ARRAY_SIZE(sgs));
 797        BUG_ON(virtqueue_add_sgs(vi->cvq, sgs, out_num, in_num, vi, GFP_ATOMIC)
 798               < 0);
 799
 800        virtqueue_kick(vi->cvq);
 801
 802        /* Spin for a response, the kick causes an ioport write, trapping
 803         * into the hypervisor, so the request should be handled immediately.
 804         */
 805        while (!virtqueue_get_buf(vi->cvq, &tmp))
 806                cpu_relax();
 807
 808        return status == VIRTIO_NET_OK;
 809}
 810
 811static int virtnet_set_mac_address(struct net_device *dev, void *p)
 812{
 813        struct virtnet_info *vi = netdev_priv(dev);
 814        struct virtio_device *vdev = vi->vdev;
 815        int ret;
 816        struct sockaddr *addr = p;
 817        struct scatterlist sg;
 818
 819        ret = eth_prepare_mac_addr_change(dev, p);
 820        if (ret)
 821                return ret;
 822
 823        if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
 824                sg_init_one(&sg, addr->sa_data, dev->addr_len);
 825                if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
 826                                          VIRTIO_NET_CTRL_MAC_ADDR_SET,
 827                                          &sg, NULL)) {
 828                        dev_warn(&vdev->dev,
 829                                 "Failed to set mac address by vq command.\n");
 830                        return -EINVAL;
 831                }
 832        } else if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) {
 833                vdev->config->set(vdev, offsetof(struct virtio_net_config, mac),
 834                                  addr->sa_data, dev->addr_len);
 835        }
 836
 837        eth_commit_mac_addr_change(dev, p);
 838
 839        return 0;
 840}
 841
 842static struct rtnl_link_stats64 *virtnet_stats(struct net_device *dev,
 843                                               struct rtnl_link_stats64 *tot)
 844{
 845        struct virtnet_info *vi = netdev_priv(dev);
 846        int cpu;
 847        unsigned int start;
 848
 849        for_each_possible_cpu(cpu) {
 850                struct virtnet_stats *stats = per_cpu_ptr(vi->stats, cpu);
 851                u64 tpackets, tbytes, rpackets, rbytes;
 852
 853                do {
 854                        start = u64_stats_fetch_begin_bh(&stats->tx_syncp);
 855                        tpackets = stats->tx_packets;
 856                        tbytes   = stats->tx_bytes;
 857                } while (u64_stats_fetch_retry_bh(&stats->tx_syncp, start));
 858
 859                do {
 860                        start = u64_stats_fetch_begin_bh(&stats->rx_syncp);
 861                        rpackets = stats->rx_packets;
 862                        rbytes   = stats->rx_bytes;
 863                } while (u64_stats_fetch_retry_bh(&stats->rx_syncp, start));
 864
 865                tot->rx_packets += rpackets;
 866                tot->tx_packets += tpackets;
 867                tot->rx_bytes   += rbytes;
 868                tot->tx_bytes   += tbytes;
 869        }
 870
 871        tot->tx_dropped = dev->stats.tx_dropped;
 872        tot->tx_fifo_errors = dev->stats.tx_fifo_errors;
 873        tot->rx_dropped = dev->stats.rx_dropped;
 874        tot->rx_length_errors = dev->stats.rx_length_errors;
 875        tot->rx_frame_errors = dev->stats.rx_frame_errors;
 876
 877        return tot;
 878}
 879
 880#ifdef CONFIG_NET_POLL_CONTROLLER
 881static void virtnet_netpoll(struct net_device *dev)
 882{
 883        struct virtnet_info *vi = netdev_priv(dev);
 884        int i;
 885
 886        for (i = 0; i < vi->curr_queue_pairs; i++)
 887                napi_schedule(&vi->rq[i].napi);
 888}
 889#endif
 890
 891static void virtnet_ack_link_announce(struct virtnet_info *vi)
 892{
 893        rtnl_lock();
 894        if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE,
 895                                  VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL, NULL))
 896                dev_warn(&vi->dev->dev, "Failed to ack link announce.\n");
 897        rtnl_unlock();
 898}
 899
 900static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs)
 901{
 902        struct scatterlist sg;
 903        struct virtio_net_ctrl_mq s;
 904        struct net_device *dev = vi->dev;
 905
 906        if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ))
 907                return 0;
 908
 909        s.virtqueue_pairs = queue_pairs;
 910        sg_init_one(&sg, &s, sizeof(s));
 911
 912        if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ,
 913                                  VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg, NULL)) {
 914                dev_warn(&dev->dev, "Fail to set num of queue pairs to %d\n",
 915                         queue_pairs);
 916                return -EINVAL;
 917        } else {
 918                vi->curr_queue_pairs = queue_pairs;
 919                schedule_delayed_work(&vi->refill, 0);
 920        }
 921
 922        return 0;
 923}
 924
 925static int virtnet_close(struct net_device *dev)
 926{
 927        struct virtnet_info *vi = netdev_priv(dev);
 928        int i;
 929
 930        /* Make sure refill_work doesn't re-enable napi! */
 931        cancel_delayed_work_sync(&vi->refill);
 932
 933        for (i = 0; i < vi->max_queue_pairs; i++)
 934                napi_disable(&vi->rq[i].napi);
 935
 936        return 0;
 937}
 938
 939static void virtnet_set_rx_mode(struct net_device *dev)
 940{
 941        struct virtnet_info *vi = netdev_priv(dev);
 942        struct scatterlist sg[2];
 943        u8 promisc, allmulti;
 944        struct virtio_net_ctrl_mac *mac_data;
 945        struct netdev_hw_addr *ha;
 946        int uc_count;
 947        int mc_count;
 948        void *buf;
 949        int i;
 950
 951        /* We can't dynamicaly set ndo_set_rx_mode, so return gracefully */
 952        if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
 953                return;
 954
 955        promisc = ((dev->flags & IFF_PROMISC) != 0);
 956        allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
 957
 958        sg_init_one(sg, &promisc, sizeof(promisc));
 959
 960        if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
 961                                  VIRTIO_NET_CTRL_RX_PROMISC,
 962                                  sg, NULL))
 963                dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
 964                         promisc ? "en" : "dis");
 965
 966        sg_init_one(sg, &allmulti, sizeof(allmulti));
 967
 968        if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
 969                                  VIRTIO_NET_CTRL_RX_ALLMULTI,
 970                                  sg, NULL))
 971                dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
 972                         allmulti ? "en" : "dis");
 973
 974        uc_count = netdev_uc_count(dev);
 975        mc_count = netdev_mc_count(dev);
 976        /* MAC filter - use one buffer for both lists */
 977        buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) +
 978                      (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
 979        mac_data = buf;
 980        if (!buf)
 981                return;
 982
 983        sg_init_table(sg, 2);
 984
 985        /* Store the unicast list and count in the front of the buffer */
 986        mac_data->entries = uc_count;
 987        i = 0;
 988        netdev_for_each_uc_addr(ha, dev)
 989                memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
 990
 991        sg_set_buf(&sg[0], mac_data,
 992                   sizeof(mac_data->entries) + (uc_count * ETH_ALEN));
 993
 994        /* multicast list and count fill the end */
 995        mac_data = (void *)&mac_data->macs[uc_count][0];
 996
 997        mac_data->entries = mc_count;
 998        i = 0;
 999        netdev_for_each_mc_addr(ha, dev)
1000                memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
1001
1002        sg_set_buf(&sg[1], mac_data,
1003                   sizeof(mac_data->entries) + (mc_count * ETH_ALEN));
1004
1005        if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
1006                                  VIRTIO_NET_CTRL_MAC_TABLE_SET,
1007                                  sg, NULL))
1008                dev_warn(&dev->dev, "Failed to set MAC fitler table.\n");
1009
1010        kfree(buf);
1011}
1012
1013static int virtnet_vlan_rx_add_vid(struct net_device *dev,
1014                                   __be16 proto, u16 vid)
1015{
1016        struct virtnet_info *vi = netdev_priv(dev);
1017        struct scatterlist sg;
1018
1019        sg_init_one(&sg, &vid, sizeof(vid));
1020
1021        if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
1022                                  VIRTIO_NET_CTRL_VLAN_ADD, &sg, NULL))
1023                dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
1024        return 0;
1025}
1026
1027static int virtnet_vlan_rx_kill_vid(struct net_device *dev,
1028                                    __be16 proto, u16 vid)
1029{
1030        struct virtnet_info *vi = netdev_priv(dev);
1031        struct scatterlist sg;
1032
1033        sg_init_one(&sg, &vid, sizeof(vid));
1034
1035        if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
1036                                  VIRTIO_NET_CTRL_VLAN_DEL, &sg, NULL))
1037                dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
1038        return 0;
1039}
1040
1041static void virtnet_clean_affinity(struct virtnet_info *vi, long hcpu)
1042{
1043        int i;
1044        int cpu;
1045
1046        if (vi->affinity_hint_set) {
1047                for (i = 0; i < vi->max_queue_pairs; i++) {
1048                        virtqueue_set_affinity(vi->rq[i].vq, -1);
1049                        virtqueue_set_affinity(vi->sq[i].vq, -1);
1050                }
1051
1052                vi->affinity_hint_set = false;
1053        }
1054
1055        i = 0;
1056        for_each_online_cpu(cpu) {
1057                if (cpu == hcpu) {
1058                        *per_cpu_ptr(vi->vq_index, cpu) = -1;
1059                } else {
1060                        *per_cpu_ptr(vi->vq_index, cpu) =
1061                                ++i % vi->curr_queue_pairs;
1062                }
1063        }
1064}
1065
1066static void virtnet_set_affinity(struct virtnet_info *vi)
1067{
1068        int i;
1069        int cpu;
1070
1071        /* In multiqueue mode, when the number of cpu is equal to the number of
1072         * queue pairs, we let the queue pairs to be private to one cpu by
1073         * setting the affinity hint to eliminate the contention.
1074         */
1075        if (vi->curr_queue_pairs == 1 ||
1076            vi->max_queue_pairs != num_online_cpus()) {
1077                virtnet_clean_affinity(vi, -1);
1078                return;
1079        }
1080
1081        i = 0;
1082        for_each_online_cpu(cpu) {
1083                virtqueue_set_affinity(vi->rq[i].vq, cpu);
1084                virtqueue_set_affinity(vi->sq[i].vq, cpu);
1085                *per_cpu_ptr(vi->vq_index, cpu) = i;
1086                i++;
1087        }
1088
1089        vi->affinity_hint_set = true;
1090}
1091
1092static int virtnet_cpu_callback(struct notifier_block *nfb,
1093                                unsigned long action, void *hcpu)
1094{
1095        struct virtnet_info *vi = container_of(nfb, struct virtnet_info, nb);
1096
1097        switch(action & ~CPU_TASKS_FROZEN) {
1098        case CPU_ONLINE:
1099        case CPU_DOWN_FAILED:
1100        case CPU_DEAD:
1101                virtnet_set_affinity(vi);
1102                break;
1103        case CPU_DOWN_PREPARE:
1104                virtnet_clean_affinity(vi, (long)hcpu);
1105                break;
1106        default:
1107                break;
1108        }
1109        return NOTIFY_OK;
1110}
1111
1112static void virtnet_get_ringparam(struct net_device *dev,
1113                                struct ethtool_ringparam *ring)
1114{
1115        struct virtnet_info *vi = netdev_priv(dev);
1116
1117        ring->rx_max_pending = virtqueue_get_vring_size(vi->rq[0].vq);
1118        ring->tx_max_pending = virtqueue_get_vring_size(vi->sq[0].vq);
1119        ring->rx_pending = ring->rx_max_pending;
1120        ring->tx_pending = ring->tx_max_pending;
1121}
1122
1123
1124static void virtnet_get_drvinfo(struct net_device *dev,
1125                                struct ethtool_drvinfo *info)
1126{
1127        struct virtnet_info *vi = netdev_priv(dev);
1128        struct virtio_device *vdev = vi->vdev;
1129
1130        strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
1131        strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version));
1132        strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info));
1133
1134}
1135
1136/* TODO: Eliminate OOO packets during switching */
1137static int virtnet_set_channels(struct net_device *dev,
1138                                struct ethtool_channels *channels)
1139{
1140        struct virtnet_info *vi = netdev_priv(dev);
1141        u16 queue_pairs = channels->combined_count;
1142        int err;
1143
1144        /* We don't support separate rx/tx channels.
1145         * We don't allow setting 'other' channels.
1146         */
1147        if (channels->rx_count || channels->tx_count || channels->other_count)
1148                return -EINVAL;
1149
1150        if (queue_pairs > vi->max_queue_pairs)
1151                return -EINVAL;
1152
1153        get_online_cpus();
1154        err = virtnet_set_queues(vi, queue_pairs);
1155        if (!err) {
1156                netif_set_real_num_tx_queues(dev, queue_pairs);
1157                netif_set_real_num_rx_queues(dev, queue_pairs);
1158
1159                virtnet_set_affinity(vi);
1160        }
1161        put_online_cpus();
1162
1163        return err;
1164}
1165
1166static void virtnet_get_channels(struct net_device *dev,
1167                                 struct ethtool_channels *channels)
1168{
1169        struct virtnet_info *vi = netdev_priv(dev);
1170
1171        channels->combined_count = vi->curr_queue_pairs;
1172        channels->max_combined = vi->max_queue_pairs;
1173        channels->max_other = 0;
1174        channels->rx_count = 0;
1175        channels->tx_count = 0;
1176        channels->other_count = 0;
1177}
1178
1179static const struct ethtool_ops virtnet_ethtool_ops = {
1180        .get_drvinfo = virtnet_get_drvinfo,
1181        .get_link = ethtool_op_get_link,
1182        .get_ringparam = virtnet_get_ringparam,
1183        .set_channels = virtnet_set_channels,
1184        .get_channels = virtnet_get_channels,
1185};
1186
1187#define MIN_MTU 68
1188#define MAX_MTU 65535
1189
1190static int virtnet_change_mtu(struct net_device *dev, int new_mtu)
1191{
1192        if (new_mtu < MIN_MTU || new_mtu > MAX_MTU)
1193                return -EINVAL;
1194        dev->mtu = new_mtu;
1195        return 0;
1196}
1197
1198/* To avoid contending a lock hold by a vcpu who would exit to host, select the
1199 * txq based on the processor id.
1200 */
1201static u16 virtnet_select_queue(struct net_device *dev, struct sk_buff *skb)
1202{
1203        int txq;
1204        struct virtnet_info *vi = netdev_priv(dev);
1205
1206        if (skb_rx_queue_recorded(skb)) {
1207                txq = skb_get_rx_queue(skb);
1208        } else {
1209                txq = *__this_cpu_ptr(vi->vq_index);
1210                if (txq == -1)
1211                        txq = 0;
1212        }
1213
1214        while (unlikely(txq >= dev->real_num_tx_queues))
1215                txq -= dev->real_num_tx_queues;
1216
1217        return txq;
1218}
1219
1220static const struct net_device_ops virtnet_netdev = {
1221        .ndo_open            = virtnet_open,
1222        .ndo_stop            = virtnet_close,
1223        .ndo_start_xmit      = start_xmit,
1224        .ndo_validate_addr   = eth_validate_addr,
1225        .ndo_set_mac_address = virtnet_set_mac_address,
1226        .ndo_set_rx_mode     = virtnet_set_rx_mode,
1227        .ndo_change_mtu      = virtnet_change_mtu,
1228        .ndo_get_stats64     = virtnet_stats,
1229        .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
1230        .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
1231        .ndo_select_queue     = virtnet_select_queue,
1232#ifdef CONFIG_NET_POLL_CONTROLLER
1233        .ndo_poll_controller = virtnet_netpoll,
1234#endif
1235};
1236
1237static void virtnet_config_changed_work(struct work_struct *work)
1238{
1239        struct virtnet_info *vi =
1240                container_of(work, struct virtnet_info, config_work);
1241        u16 v;
1242
1243        mutex_lock(&vi->config_lock);
1244        if (!vi->config_enable)
1245                goto done;
1246
1247        if (virtio_config_val(vi->vdev, VIRTIO_NET_F_STATUS,
1248                              offsetof(struct virtio_net_config, status),
1249                              &v) < 0)
1250                goto done;
1251
1252        if (v & VIRTIO_NET_S_ANNOUNCE) {
1253                netdev_notify_peers(vi->dev);
1254                virtnet_ack_link_announce(vi);
1255        }
1256
1257        /* Ignore unknown (future) status bits */
1258        v &= VIRTIO_NET_S_LINK_UP;
1259
1260        if (vi->status == v)
1261                goto done;
1262
1263        vi->status = v;
1264
1265        if (vi->status & VIRTIO_NET_S_LINK_UP) {
1266                netif_carrier_on(vi->dev);
1267                netif_tx_wake_all_queues(vi->dev);
1268        } else {
1269                netif_carrier_off(vi->dev);
1270                netif_tx_stop_all_queues(vi->dev);
1271        }
1272done:
1273        mutex_unlock(&vi->config_lock);
1274}
1275
1276static void virtnet_config_changed(struct virtio_device *vdev)
1277{
1278        struct virtnet_info *vi = vdev->priv;
1279
1280        schedule_work(&vi->config_work);
1281}
1282
1283static void virtnet_free_queues(struct virtnet_info *vi)
1284{
1285        kfree(vi->rq);
1286        kfree(vi->sq);
1287}
1288
1289static void free_receive_bufs(struct virtnet_info *vi)
1290{
1291        int i;
1292
1293        for (i = 0; i < vi->max_queue_pairs; i++) {
1294                while (vi->rq[i].pages)
1295                        __free_pages(get_a_page(&vi->rq[i], GFP_KERNEL), 0);
1296        }
1297}
1298
1299static void free_unused_bufs(struct virtnet_info *vi)
1300{
1301        void *buf;
1302        int i;
1303
1304        for (i = 0; i < vi->max_queue_pairs; i++) {
1305                struct virtqueue *vq = vi->sq[i].vq;
1306                while ((buf = virtqueue_detach_unused_buf(vq)) != NULL)
1307                        dev_kfree_skb(buf);
1308        }
1309
1310        for (i = 0; i < vi->max_queue_pairs; i++) {
1311                struct virtqueue *vq = vi->rq[i].vq;
1312
1313                while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) {
1314                        if (vi->mergeable_rx_bufs || vi->big_packets)
1315                                give_pages(&vi->rq[i], buf);
1316                        else
1317                                dev_kfree_skb(buf);
1318                        --vi->rq[i].num;
1319                }
1320                BUG_ON(vi->rq[i].num != 0);
1321        }
1322}
1323
1324static void virtnet_del_vqs(struct virtnet_info *vi)
1325{
1326        struct virtio_device *vdev = vi->vdev;
1327
1328        virtnet_clean_affinity(vi, -1);
1329
1330        vdev->config->del_vqs(vdev);
1331
1332        virtnet_free_queues(vi);
1333}
1334
1335static int virtnet_find_vqs(struct virtnet_info *vi)
1336{
1337        vq_callback_t **callbacks;
1338        struct virtqueue **vqs;
1339        int ret = -ENOMEM;
1340        int i, total_vqs;
1341        const char **names;
1342
1343        /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by
1344         * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by
1345         * possible control vq.
1346         */
1347        total_vqs = vi->max_queue_pairs * 2 +
1348                    virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ);
1349
1350        /* Allocate space for find_vqs parameters */
1351        vqs = kzalloc(total_vqs * sizeof(*vqs), GFP_KERNEL);
1352        if (!vqs)
1353                goto err_vq;
1354        callbacks = kmalloc(total_vqs * sizeof(*callbacks), GFP_KERNEL);
1355        if (!callbacks)
1356                goto err_callback;
1357        names = kmalloc(total_vqs * sizeof(*names), GFP_KERNEL);
1358        if (!names)
1359                goto err_names;
1360
1361        /* Parameters for control virtqueue, if any */
1362        if (vi->has_cvq) {
1363                callbacks[total_vqs - 1] = NULL;
1364                names[total_vqs - 1] = "control";
1365        }
1366
1367        /* Allocate/initialize parameters for send/receive virtqueues */
1368        for (i = 0; i < vi->max_queue_pairs; i++) {
1369                callbacks[rxq2vq(i)] = skb_recv_done;
1370                callbacks[txq2vq(i)] = skb_xmit_done;
1371                sprintf(vi->rq[i].name, "input.%d", i);
1372                sprintf(vi->sq[i].name, "output.%d", i);
1373                names[rxq2vq(i)] = vi->rq[i].name;
1374                names[txq2vq(i)] = vi->sq[i].name;
1375        }
1376
1377        ret = vi->vdev->config->find_vqs(vi->vdev, total_vqs, vqs, callbacks,
1378                                         names);
1379        if (ret)
1380                goto err_find;
1381
1382        if (vi->has_cvq) {
1383                vi->cvq = vqs[total_vqs - 1];
1384                if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
1385                        vi->dev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
1386        }
1387
1388        for (i = 0; i < vi->max_queue_pairs; i++) {
1389                vi->rq[i].vq = vqs[rxq2vq(i)];
1390                vi->sq[i].vq = vqs[txq2vq(i)];
1391        }
1392
1393        kfree(names);
1394        kfree(callbacks);
1395        kfree(vqs);
1396
1397        return 0;
1398
1399err_find:
1400        kfree(names);
1401err_names:
1402        kfree(callbacks);
1403err_callback:
1404        kfree(vqs);
1405err_vq:
1406        return ret;
1407}
1408
1409static int virtnet_alloc_queues(struct virtnet_info *vi)
1410{
1411        int i;
1412
1413        vi->sq = kzalloc(sizeof(*vi->sq) * vi->max_queue_pairs, GFP_KERNEL);
1414        if (!vi->sq)
1415                goto err_sq;
1416        vi->rq = kzalloc(sizeof(*vi->rq) * vi->max_queue_pairs, GFP_KERNEL);
1417        if (!vi->rq)
1418                goto err_rq;
1419
1420        INIT_DELAYED_WORK(&vi->refill, refill_work);
1421        for (i = 0; i < vi->max_queue_pairs; i++) {
1422                vi->rq[i].pages = NULL;
1423                netif_napi_add(vi->dev, &vi->rq[i].napi, virtnet_poll,
1424                               napi_weight);
1425
1426                sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg));
1427                sg_init_table(vi->sq[i].sg, ARRAY_SIZE(vi->sq[i].sg));
1428        }
1429
1430        return 0;
1431
1432err_rq:
1433        kfree(vi->sq);
1434err_sq:
1435        return -ENOMEM;
1436}
1437
1438static int init_vqs(struct virtnet_info *vi)
1439{
1440        int ret;
1441
1442        /* Allocate send & receive queues */
1443        ret = virtnet_alloc_queues(vi);
1444        if (ret)
1445                goto err;
1446
1447        ret = virtnet_find_vqs(vi);
1448        if (ret)
1449                goto err_free;
1450
1451        get_online_cpus();
1452        virtnet_set_affinity(vi);
1453        put_online_cpus();
1454
1455        return 0;
1456
1457err_free:
1458        virtnet_free_queues(vi);
1459err:
1460        return ret;
1461}
1462
1463static int virtnet_probe(struct virtio_device *vdev)
1464{
1465        int i, err;
1466        struct net_device *dev;
1467        struct virtnet_info *vi;
1468        u16 max_queue_pairs;
1469
1470        /* Find if host supports multiqueue virtio_net device */
1471        err = virtio_config_val(vdev, VIRTIO_NET_F_MQ,
1472                                offsetof(struct virtio_net_config,
1473                                max_virtqueue_pairs), &max_queue_pairs);
1474
1475        /* We need at least 2 queue's */
1476        if (err || max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
1477            max_queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
1478            !virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
1479                max_queue_pairs = 1;
1480
1481        /* Allocate ourselves a network device with room for our info */
1482        dev = alloc_etherdev_mq(sizeof(struct virtnet_info), max_queue_pairs);
1483        if (!dev)
1484                return -ENOMEM;
1485
1486        /* Set up network device as normal. */
1487        dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE;
1488        dev->netdev_ops = &virtnet_netdev;
1489        dev->features = NETIF_F_HIGHDMA;
1490
1491        SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops);
1492        SET_NETDEV_DEV(dev, &vdev->dev);
1493
1494        /* Do we support "hardware" checksums? */
1495        if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
1496                /* This opens up the world of extra features. */
1497                dev->hw_features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
1498                if (csum)
1499                        dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
1500
1501                if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
1502                        dev->hw_features |= NETIF_F_TSO | NETIF_F_UFO
1503                                | NETIF_F_TSO_ECN | NETIF_F_TSO6;
1504                }
1505                /* Individual feature bits: what can host handle? */
1506                if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
1507                        dev->hw_features |= NETIF_F_TSO;
1508                if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
1509                        dev->hw_features |= NETIF_F_TSO6;
1510                if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
1511                        dev->hw_features |= NETIF_F_TSO_ECN;
1512                if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
1513                        dev->hw_features |= NETIF_F_UFO;
1514
1515                if (gso)
1516                        dev->features |= dev->hw_features & (NETIF_F_ALL_TSO|NETIF_F_UFO);
1517                /* (!csum && gso) case will be fixed by register_netdev() */
1518        }
1519
1520        dev->vlan_features = dev->features;
1521
1522        /* Configuration may specify what MAC to use.  Otherwise random. */
1523        if (virtio_config_val_len(vdev, VIRTIO_NET_F_MAC,
1524                                  offsetof(struct virtio_net_config, mac),
1525                                  dev->dev_addr, dev->addr_len) < 0)
1526                eth_hw_addr_random(dev);
1527
1528        /* Set up our device-specific information */
1529        vi = netdev_priv(dev);
1530        vi->dev = dev;
1531        vi->vdev = vdev;
1532        vdev->priv = vi;
1533        vi->stats = alloc_percpu(struct virtnet_stats);
1534        err = -ENOMEM;
1535        if (vi->stats == NULL)
1536                goto free;
1537
1538        vi->vq_index = alloc_percpu(int);
1539        if (vi->vq_index == NULL)
1540                goto free_stats;
1541
1542        mutex_init(&vi->config_lock);
1543        vi->config_enable = true;
1544        INIT_WORK(&vi->config_work, virtnet_config_changed_work);
1545
1546        /* If we can receive ANY GSO packets, we must allocate large ones. */
1547        if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
1548            virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) ||
1549            virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN))
1550                vi->big_packets = true;
1551
1552        if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
1553                vi->mergeable_rx_bufs = true;
1554
1555        if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
1556                vi->has_cvq = true;
1557
1558        /* Use single tx/rx queue pair as default */
1559        vi->curr_queue_pairs = 1;
1560        vi->max_queue_pairs = max_queue_pairs;
1561
1562        /* Allocate/initialize the rx/tx queues, and invoke find_vqs */
1563        err = init_vqs(vi);
1564        if (err)
1565                goto free_index;
1566
1567        netif_set_real_num_tx_queues(dev, 1);
1568        netif_set_real_num_rx_queues(dev, 1);
1569
1570        err = register_netdev(dev);
1571        if (err) {
1572                pr_debug("virtio_net: registering device failed\n");
1573                goto free_vqs;
1574        }
1575
1576        /* Last of all, set up some receive buffers. */
1577        for (i = 0; i < vi->curr_queue_pairs; i++) {
1578                try_fill_recv(&vi->rq[i], GFP_KERNEL);
1579
1580                /* If we didn't even get one input buffer, we're useless. */
1581                if (vi->rq[i].num == 0) {
1582                        free_unused_bufs(vi);
1583                        err = -ENOMEM;
1584                        goto free_recv_bufs;
1585                }
1586        }
1587
1588        vi->nb.notifier_call = &virtnet_cpu_callback;
1589        err = register_hotcpu_notifier(&vi->nb);
1590        if (err) {
1591                pr_debug("virtio_net: registering cpu notifier failed\n");
1592                goto free_recv_bufs;
1593        }
1594
1595        /* Assume link up if device can't report link status,
1596           otherwise get link status from config. */
1597        if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) {
1598                netif_carrier_off(dev);
1599                schedule_work(&vi->config_work);
1600        } else {
1601                vi->status = VIRTIO_NET_S_LINK_UP;
1602                netif_carrier_on(dev);
1603        }
1604
1605        pr_debug("virtnet: registered device %s with %d RX and TX vq's\n",
1606                 dev->name, max_queue_pairs);
1607
1608        return 0;
1609
1610free_recv_bufs:
1611        free_receive_bufs(vi);
1612        unregister_netdev(dev);
1613free_vqs:
1614        cancel_delayed_work_sync(&vi->refill);
1615        virtnet_del_vqs(vi);
1616free_index:
1617        free_percpu(vi->vq_index);
1618free_stats:
1619        free_percpu(vi->stats);
1620free:
1621        free_netdev(dev);
1622        return err;
1623}
1624
1625static void remove_vq_common(struct virtnet_info *vi)
1626{
1627        vi->vdev->config->reset(vi->vdev);
1628
1629        /* Free unused buffers in both send and recv, if any. */
1630        free_unused_bufs(vi);
1631
1632        free_receive_bufs(vi);
1633
1634        virtnet_del_vqs(vi);
1635}
1636
1637static void virtnet_remove(struct virtio_device *vdev)
1638{
1639        struct virtnet_info *vi = vdev->priv;
1640
1641        unregister_hotcpu_notifier(&vi->nb);
1642
1643        /* Prevent config work handler from accessing the device. */
1644        mutex_lock(&vi->config_lock);
1645        vi->config_enable = false;
1646        mutex_unlock(&vi->config_lock);
1647
1648        unregister_netdev(vi->dev);
1649
1650        remove_vq_common(vi);
1651
1652        flush_work(&vi->config_work);
1653
1654        free_percpu(vi->vq_index);
1655        free_percpu(vi->stats);
1656        free_netdev(vi->dev);
1657}
1658
1659#ifdef CONFIG_PM
1660static int virtnet_freeze(struct virtio_device *vdev)
1661{
1662        struct virtnet_info *vi = vdev->priv;
1663        int i;
1664
1665        /* Prevent config work handler from accessing the device */
1666        mutex_lock(&vi->config_lock);
1667        vi->config_enable = false;
1668        mutex_unlock(&vi->config_lock);
1669
1670        netif_device_detach(vi->dev);
1671        cancel_delayed_work_sync(&vi->refill);
1672
1673        if (netif_running(vi->dev))
1674                for (i = 0; i < vi->max_queue_pairs; i++) {
1675                        napi_disable(&vi->rq[i].napi);
1676                        netif_napi_del(&vi->rq[i].napi);
1677                }
1678
1679        remove_vq_common(vi);
1680
1681        flush_work(&vi->config_work);
1682
1683        return 0;
1684}
1685
1686static int virtnet_restore(struct virtio_device *vdev)
1687{
1688        struct virtnet_info *vi = vdev->priv;
1689        int err, i;
1690
1691        err = init_vqs(vi);
1692        if (err)
1693                return err;
1694
1695        if (netif_running(vi->dev))
1696                for (i = 0; i < vi->max_queue_pairs; i++)
1697                        virtnet_napi_enable(&vi->rq[i]);
1698
1699        netif_device_attach(vi->dev);
1700
1701        for (i = 0; i < vi->curr_queue_pairs; i++)
1702                if (!try_fill_recv(&vi->rq[i], GFP_KERNEL))
1703                        schedule_delayed_work(&vi->refill, 0);
1704
1705        mutex_lock(&vi->config_lock);
1706        vi->config_enable = true;
1707        mutex_unlock(&vi->config_lock);
1708
1709        virtnet_set_queues(vi, vi->curr_queue_pairs);
1710
1711        return 0;
1712}
1713#endif
1714
1715static struct virtio_device_id id_table[] = {
1716        { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
1717        { 0 },
1718};
1719
1720static unsigned int features[] = {
1721        VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM,
1722        VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC,
1723        VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6,
1724        VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6,
1725        VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO,
1726        VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ,
1727        VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN,
1728        VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ,
1729        VIRTIO_NET_F_CTRL_MAC_ADDR,
1730};
1731
1732static struct virtio_driver virtio_net_driver = {
1733        .feature_table = features,
1734        .feature_table_size = ARRAY_SIZE(features),
1735        .driver.name =  KBUILD_MODNAME,
1736        .driver.owner = THIS_MODULE,
1737        .id_table =     id_table,
1738        .probe =        virtnet_probe,
1739        .remove =       virtnet_remove,
1740        .config_changed = virtnet_config_changed,
1741#ifdef CONFIG_PM
1742        .freeze =       virtnet_freeze,
1743        .restore =      virtnet_restore,
1744#endif
1745};
1746
1747module_virtio_driver(virtio_net_driver);
1748
1749MODULE_DEVICE_TABLE(virtio, id_table);
1750MODULE_DESCRIPTION("Virtio network driver");
1751MODULE_LICENSE("GPL");
1752
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.