root/drivers/net/virtio_net.c

/* [<][>][^][v][top][bottom][index][help] */

DEFINITIONS

This source file includes following definitions.
  1. is_xdp_frame
  2. xdp_to_ptr
  3. ptr_to_xdp
  4. vq2txq
  5. txq2vq
  6. vq2rxq
  7. rxq2vq
  8. skb_vnet_hdr
  9. give_pages
  10. get_a_page
  11. virtqueue_napi_schedule
  12. virtqueue_napi_complete
  13. skb_xmit_done
  14. mergeable_len_to_ctx
  15. mergeable_ctx_to_headroom
  16. mergeable_ctx_to_truesize
  17. page_to_skb
  18. __virtnet_xdp_xmit_one
  19. virtnet_xdp_sq
  20. virtnet_xdp_xmit
  21. virtnet_get_headroom
  22. xdp_linearize_page
  23. receive_small
  24. receive_big
  25. receive_mergeable
  26. receive_buf
  27. add_recvbuf_small
  28. add_recvbuf_big
  29. get_mergeable_buf_len
  30. add_recvbuf_mergeable
  31. try_fill_recv
  32. skb_recv_done
  33. virtnet_napi_enable
  34. virtnet_napi_tx_enable
  35. virtnet_napi_tx_disable
  36. refill_work
  37. virtnet_receive
  38. free_old_xmit_skbs
  39. is_xdp_raw_buffer_queue
  40. virtnet_poll_cleantx
  41. virtnet_poll
  42. virtnet_open
  43. virtnet_poll_tx
  44. xmit_skb
  45. start_xmit
  46. virtnet_send_command
  47. virtnet_set_mac_address
  48. virtnet_stats
  49. virtnet_ack_link_announce
  50. _virtnet_set_queues
  51. virtnet_set_queues
  52. virtnet_close
  53. virtnet_set_rx_mode
  54. virtnet_vlan_rx_add_vid
  55. virtnet_vlan_rx_kill_vid
  56. virtnet_clean_affinity
  57. virtnet_set_affinity
  58. virtnet_cpu_online
  59. virtnet_cpu_dead
  60. virtnet_cpu_down_prep
  61. virtnet_cpu_notif_add
  62. virtnet_cpu_notif_remove
  63. virtnet_get_ringparam
  64. virtnet_get_drvinfo
  65. virtnet_set_channels
  66. virtnet_get_strings
  67. virtnet_get_sset_count
  68. virtnet_get_ethtool_stats
  69. virtnet_get_channels
  70. virtnet_validate_ethtool_cmd
  71. virtnet_set_link_ksettings
  72. virtnet_get_link_ksettings
  73. virtnet_set_coalesce
  74. virtnet_get_coalesce
  75. virtnet_init_settings
  76. virtnet_update_settings
  77. virtnet_freeze_down
  78. virtnet_restore_up
  79. virtnet_set_guest_offloads
  80. virtnet_clear_guest_offloads
  81. virtnet_restore_guest_offloads
  82. virtnet_xdp_set
  83. virtnet_xdp_query
  84. virtnet_xdp
  85. virtnet_get_phys_port_name
  86. virtnet_set_features
  87. virtnet_config_changed_work
  88. virtnet_config_changed
  89. virtnet_free_queues
  90. _free_receive_bufs
  91. free_receive_bufs
  92. free_receive_page_frags
  93. free_unused_bufs
  94. virtnet_del_vqs
  95. mergeable_min_buf_len
  96. virtnet_find_vqs
  97. virtnet_alloc_queues
  98. init_vqs
  99. mergeable_rx_buffer_size_show
  100. virtnet_fail_on_feature
  101. virtnet_validate_features
  102. virtnet_validate
  103. virtnet_probe
  104. remove_vq_common
  105. virtnet_remove
  106. virtnet_freeze
  107. virtnet_restore
  108. virtio_net_driver_init
  109. virtio_net_driver_exit

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /* A network driver using virtio.
   3  *
   4  * Copyright 2007 Rusty Russell <rusty@rustcorp.com.au> IBM Corporation
   5  */
   6 //#define DEBUG
   7 #include <linux/netdevice.h>
   8 #include <linux/etherdevice.h>
   9 #include <linux/ethtool.h>
  10 #include <linux/module.h>
  11 #include <linux/virtio.h>
  12 #include <linux/virtio_net.h>
  13 #include <linux/bpf.h>
  14 #include <linux/bpf_trace.h>
  15 #include <linux/scatterlist.h>
  16 #include <linux/if_vlan.h>
  17 #include <linux/slab.h>
  18 #include <linux/cpu.h>
  19 #include <linux/average.h>
  20 #include <linux/filter.h>
  21 #include <linux/kernel.h>
  22 #include <net/route.h>
  23 #include <net/xdp.h>
  24 #include <net/net_failover.h>
  25 
  26 static int napi_weight = NAPI_POLL_WEIGHT;
  27 module_param(napi_weight, int, 0444);
  28 
  29 static bool csum = true, gso = true, napi_tx = true;
  30 module_param(csum, bool, 0444);
  31 module_param(gso, bool, 0444);
  32 module_param(napi_tx, bool, 0644);
  33 
  34 /* FIXME: MTU in config. */
  35 #define GOOD_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
  36 #define GOOD_COPY_LEN   128
  37 
  38 #define VIRTNET_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
  39 
  40 /* Amount of XDP headroom to prepend to packets for use by xdp_adjust_head */
  41 #define VIRTIO_XDP_HEADROOM 256
  42 
  43 /* Separating two types of XDP xmit */
  44 #define VIRTIO_XDP_TX           BIT(0)
  45 #define VIRTIO_XDP_REDIR        BIT(1)
  46 
  47 #define VIRTIO_XDP_FLAG BIT(0)
  48 
  49 /* RX packet size EWMA. The average packet size is used to determine the packet
  50  * buffer size when refilling RX rings. As the entire RX ring may be refilled
  51  * at once, the weight is chosen so that the EWMA will be insensitive to short-
  52  * term, transient changes in packet size.
  53  */
  54 DECLARE_EWMA(pkt_len, 0, 64)
  55 
  56 #define VIRTNET_DRIVER_VERSION "1.0.0"
  57 
  58 static const unsigned long guest_offloads[] = {
  59         VIRTIO_NET_F_GUEST_TSO4,
  60         VIRTIO_NET_F_GUEST_TSO6,
  61         VIRTIO_NET_F_GUEST_ECN,
  62         VIRTIO_NET_F_GUEST_UFO,
  63         VIRTIO_NET_F_GUEST_CSUM
  64 };
  65 
  66 struct virtnet_stat_desc {
  67         char desc[ETH_GSTRING_LEN];
  68         size_t offset;
  69 };
  70 
  71 struct virtnet_sq_stats {
  72         struct u64_stats_sync syncp;
  73         u64 packets;
  74         u64 bytes;
  75         u64 xdp_tx;
  76         u64 xdp_tx_drops;
  77         u64 kicks;
  78 };
  79 
  80 struct virtnet_rq_stats {
  81         struct u64_stats_sync syncp;
  82         u64 packets;
  83         u64 bytes;
  84         u64 drops;
  85         u64 xdp_packets;
  86         u64 xdp_tx;
  87         u64 xdp_redirects;
  88         u64 xdp_drops;
  89         u64 kicks;
  90 };
  91 
  92 #define VIRTNET_SQ_STAT(m)      offsetof(struct virtnet_sq_stats, m)
  93 #define VIRTNET_RQ_STAT(m)      offsetof(struct virtnet_rq_stats, m)
  94 
  95 static const struct virtnet_stat_desc virtnet_sq_stats_desc[] = {
  96         { "packets",            VIRTNET_SQ_STAT(packets) },
  97         { "bytes",              VIRTNET_SQ_STAT(bytes) },
  98         { "xdp_tx",             VIRTNET_SQ_STAT(xdp_tx) },
  99         { "xdp_tx_drops",       VIRTNET_SQ_STAT(xdp_tx_drops) },
 100         { "kicks",              VIRTNET_SQ_STAT(kicks) },
 101 };
 102 
 103 static const struct virtnet_stat_desc virtnet_rq_stats_desc[] = {
 104         { "packets",            VIRTNET_RQ_STAT(packets) },
 105         { "bytes",              VIRTNET_RQ_STAT(bytes) },
 106         { "drops",              VIRTNET_RQ_STAT(drops) },
 107         { "xdp_packets",        VIRTNET_RQ_STAT(xdp_packets) },
 108         { "xdp_tx",             VIRTNET_RQ_STAT(xdp_tx) },
 109         { "xdp_redirects",      VIRTNET_RQ_STAT(xdp_redirects) },
 110         { "xdp_drops",          VIRTNET_RQ_STAT(xdp_drops) },
 111         { "kicks",              VIRTNET_RQ_STAT(kicks) },
 112 };
 113 
 114 #define VIRTNET_SQ_STATS_LEN    ARRAY_SIZE(virtnet_sq_stats_desc)
 115 #define VIRTNET_RQ_STATS_LEN    ARRAY_SIZE(virtnet_rq_stats_desc)
 116 
 117 /* Internal representation of a send virtqueue */
 118 struct send_queue {
 119         /* Virtqueue associated with this send _queue */
 120         struct virtqueue *vq;
 121 
 122         /* TX: fragments + linear part + virtio header */
 123         struct scatterlist sg[MAX_SKB_FRAGS + 2];
 124 
 125         /* Name of the send queue: output.$index */
 126         char name[40];
 127 
 128         struct virtnet_sq_stats stats;
 129 
 130         struct napi_struct napi;
 131 };
 132 
 133 /* Internal representation of a receive virtqueue */
 134 struct receive_queue {
 135         /* Virtqueue associated with this receive_queue */
 136         struct virtqueue *vq;
 137 
 138         struct napi_struct napi;
 139 
 140         struct bpf_prog __rcu *xdp_prog;
 141 
 142         struct virtnet_rq_stats stats;
 143 
 144         /* Chain pages by the private ptr. */
 145         struct page *pages;
 146 
 147         /* Average packet length for mergeable receive buffers. */
 148         struct ewma_pkt_len mrg_avg_pkt_len;
 149 
 150         /* Page frag for packet buffer allocation. */
 151         struct page_frag alloc_frag;
 152 
 153         /* RX: fragments + linear part + virtio header */
 154         struct scatterlist sg[MAX_SKB_FRAGS + 2];
 155 
 156         /* Min single buffer size for mergeable buffers case. */
 157         unsigned int min_buf_len;
 158 
 159         /* Name of this receive queue: input.$index */
 160         char name[40];
 161 
 162         struct xdp_rxq_info xdp_rxq;
 163 };
 164 
 165 /* Control VQ buffers: protected by the rtnl lock */
 166 struct control_buf {
 167         struct virtio_net_ctrl_hdr hdr;
 168         virtio_net_ctrl_ack status;
 169         struct virtio_net_ctrl_mq mq;
 170         u8 promisc;
 171         u8 allmulti;
 172         __virtio16 vid;
 173         __virtio64 offloads;
 174 };
 175 
 176 struct virtnet_info {
 177         struct virtio_device *vdev;
 178         struct virtqueue *cvq;
 179         struct net_device *dev;
 180         struct send_queue *sq;
 181         struct receive_queue *rq;
 182         unsigned int status;
 183 
 184         /* Max # of queue pairs supported by the device */
 185         u16 max_queue_pairs;
 186 
 187         /* # of queue pairs currently used by the driver */
 188         u16 curr_queue_pairs;
 189 
 190         /* # of XDP queue pairs currently used by the driver */
 191         u16 xdp_queue_pairs;
 192 
 193         /* I like... big packets and I cannot lie! */
 194         bool big_packets;
 195 
 196         /* Host will merge rx buffers for big packets (shake it! shake it!) */
 197         bool mergeable_rx_bufs;
 198 
 199         /* Has control virtqueue */
 200         bool has_cvq;
 201 
 202         /* Host can handle any s/g split between our header and packet data */
 203         bool any_header_sg;
 204 
 205         /* Packet virtio header size */
 206         u8 hdr_len;
 207 
 208         /* Work struct for refilling if we run low on memory. */
 209         struct delayed_work refill;
 210 
 211         /* Work struct for config space updates */
 212         struct work_struct config_work;
 213 
 214         /* Does the affinity hint is set for virtqueues? */
 215         bool affinity_hint_set;
 216 
 217         /* CPU hotplug instances for online & dead */
 218         struct hlist_node node;
 219         struct hlist_node node_dead;
 220 
 221         struct control_buf *ctrl;
 222 
 223         /* Ethtool settings */
 224         u8 duplex;
 225         u32 speed;
 226 
 227         unsigned long guest_offloads;
 228         unsigned long guest_offloads_capable;
 229 
 230         /* failover when STANDBY feature enabled */
 231         struct failover *failover;
 232 };
 233 
 234 struct padded_vnet_hdr {
 235         struct virtio_net_hdr_mrg_rxbuf hdr;
 236         /*
 237          * hdr is in a separate sg buffer, and data sg buffer shares same page
 238          * with this header sg. This padding makes next sg 16 byte aligned
 239          * after the header.
 240          */
 241         char padding[4];
 242 };
 243 
 244 static bool is_xdp_frame(void *ptr)
 245 {
 246         return (unsigned long)ptr & VIRTIO_XDP_FLAG;
 247 }
 248 
 249 static void *xdp_to_ptr(struct xdp_frame *ptr)
 250 {
 251         return (void *)((unsigned long)ptr | VIRTIO_XDP_FLAG);
 252 }
 253 
 254 static struct xdp_frame *ptr_to_xdp(void *ptr)
 255 {
 256         return (struct xdp_frame *)((unsigned long)ptr & ~VIRTIO_XDP_FLAG);
 257 }
 258 
 259 /* Converting between virtqueue no. and kernel tx/rx queue no.
 260  * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq
 261  */
 262 static int vq2txq(struct virtqueue *vq)
 263 {
 264         return (vq->index - 1) / 2;
 265 }
 266 
 267 static int txq2vq(int txq)
 268 {
 269         return txq * 2 + 1;
 270 }
 271 
 272 static int vq2rxq(struct virtqueue *vq)
 273 {
 274         return vq->index / 2;
 275 }
 276 
 277 static int rxq2vq(int rxq)
 278 {
 279         return rxq * 2;
 280 }
 281 
 282 static inline struct virtio_net_hdr_mrg_rxbuf *skb_vnet_hdr(struct sk_buff *skb)
 283 {
 284         return (struct virtio_net_hdr_mrg_rxbuf *)skb->cb;
 285 }
 286 
 287 /*
 288  * private is used to chain pages for big packets, put the whole
 289  * most recent used list in the beginning for reuse
 290  */
 291 static void give_pages(struct receive_queue *rq, struct page *page)
 292 {
 293         struct page *end;
 294 
 295         /* Find end of list, sew whole thing into vi->rq.pages. */
 296         for (end = page; end->private; end = (struct page *)end->private);
 297         end->private = (unsigned long)rq->pages;
 298         rq->pages = page;
 299 }
 300 
 301 static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask)
 302 {
 303         struct page *p = rq->pages;
 304 
 305         if (p) {
 306                 rq->pages = (struct page *)p->private;
 307                 /* clear private here, it is used to chain pages */
 308                 p->private = 0;
 309         } else
 310                 p = alloc_page(gfp_mask);
 311         return p;
 312 }
 313 
 314 static void virtqueue_napi_schedule(struct napi_struct *napi,
 315                                     struct virtqueue *vq)
 316 {
 317         if (napi_schedule_prep(napi)) {
 318                 virtqueue_disable_cb(vq);
 319                 __napi_schedule(napi);
 320         }
 321 }
 322 
 323 static void virtqueue_napi_complete(struct napi_struct *napi,
 324                                     struct virtqueue *vq, int processed)
 325 {
 326         int opaque;
 327 
 328         opaque = virtqueue_enable_cb_prepare(vq);
 329         if (napi_complete_done(napi, processed)) {
 330                 if (unlikely(virtqueue_poll(vq, opaque)))
 331                         virtqueue_napi_schedule(napi, vq);
 332         } else {
 333                 virtqueue_disable_cb(vq);
 334         }
 335 }
 336 
 337 static void skb_xmit_done(struct virtqueue *vq)
 338 {
 339         struct virtnet_info *vi = vq->vdev->priv;
 340         struct napi_struct *napi = &vi->sq[vq2txq(vq)].napi;
 341 
 342         /* Suppress further interrupts. */
 343         virtqueue_disable_cb(vq);
 344 
 345         if (napi->weight)
 346                 virtqueue_napi_schedule(napi, vq);
 347         else
 348                 /* We were probably waiting for more output buffers. */
 349                 netif_wake_subqueue(vi->dev, vq2txq(vq));
 350 }
 351 
 352 #define MRG_CTX_HEADER_SHIFT 22
 353 static void *mergeable_len_to_ctx(unsigned int truesize,
 354                                   unsigned int headroom)
 355 {
 356         return (void *)(unsigned long)((headroom << MRG_CTX_HEADER_SHIFT) | truesize);
 357 }
 358 
 359 static unsigned int mergeable_ctx_to_headroom(void *mrg_ctx)
 360 {
 361         return (unsigned long)mrg_ctx >> MRG_CTX_HEADER_SHIFT;
 362 }
 363 
 364 static unsigned int mergeable_ctx_to_truesize(void *mrg_ctx)
 365 {
 366         return (unsigned long)mrg_ctx & ((1 << MRG_CTX_HEADER_SHIFT) - 1);
 367 }
 368 
 369 /* Called from bottom half context */
 370 static struct sk_buff *page_to_skb(struct virtnet_info *vi,
 371                                    struct receive_queue *rq,
 372                                    struct page *page, unsigned int offset,
 373                                    unsigned int len, unsigned int truesize,
 374                                    bool hdr_valid)
 375 {
 376         struct sk_buff *skb;
 377         struct virtio_net_hdr_mrg_rxbuf *hdr;
 378         unsigned int copy, hdr_len, hdr_padded_len;
 379         char *p;
 380 
 381         p = page_address(page) + offset;
 382 
 383         /* copy small packet so we can reuse these pages for small data */
 384         skb = napi_alloc_skb(&rq->napi, GOOD_COPY_LEN);
 385         if (unlikely(!skb))
 386                 return NULL;
 387 
 388         hdr = skb_vnet_hdr(skb);
 389 
 390         hdr_len = vi->hdr_len;
 391         if (vi->mergeable_rx_bufs)
 392                 hdr_padded_len = sizeof(*hdr);
 393         else
 394                 hdr_padded_len = sizeof(struct padded_vnet_hdr);
 395 
 396         if (hdr_valid)
 397                 memcpy(hdr, p, hdr_len);
 398 
 399         len -= hdr_len;
 400         offset += hdr_padded_len;
 401         p += hdr_padded_len;
 402 
 403         copy = len;
 404         if (copy > skb_tailroom(skb))
 405                 copy = skb_tailroom(skb);
 406         skb_put_data(skb, p, copy);
 407 
 408         len -= copy;
 409         offset += copy;
 410 
 411         if (vi->mergeable_rx_bufs) {
 412                 if (len)
 413                         skb_add_rx_frag(skb, 0, page, offset, len, truesize);
 414                 else
 415                         put_page(page);
 416                 return skb;
 417         }
 418 
 419         /*
 420          * Verify that we can indeed put this data into a skb.
 421          * This is here to handle cases when the device erroneously
 422          * tries to receive more than is possible. This is usually
 423          * the case of a broken device.
 424          */
 425         if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) {
 426                 net_dbg_ratelimited("%s: too much data\n", skb->dev->name);
 427                 dev_kfree_skb(skb);
 428                 return NULL;
 429         }
 430         BUG_ON(offset >= PAGE_SIZE);
 431         while (len) {
 432                 unsigned int frag_size = min((unsigned)PAGE_SIZE - offset, len);
 433                 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page, offset,
 434                                 frag_size, truesize);
 435                 len -= frag_size;
 436                 page = (struct page *)page->private;
 437                 offset = 0;
 438         }
 439 
 440         if (page)
 441                 give_pages(rq, page);
 442 
 443         return skb;
 444 }
 445 
 446 static int __virtnet_xdp_xmit_one(struct virtnet_info *vi,
 447                                    struct send_queue *sq,
 448                                    struct xdp_frame *xdpf)
 449 {
 450         struct virtio_net_hdr_mrg_rxbuf *hdr;
 451         int err;
 452 
 453         /* virtqueue want to use data area in-front of packet */
 454         if (unlikely(xdpf->metasize > 0))
 455                 return -EOPNOTSUPP;
 456 
 457         if (unlikely(xdpf->headroom < vi->hdr_len))
 458                 return -EOVERFLOW;
 459 
 460         /* Make room for virtqueue hdr (also change xdpf->headroom?) */
 461         xdpf->data -= vi->hdr_len;
 462         /* Zero header and leave csum up to XDP layers */
 463         hdr = xdpf->data;
 464         memset(hdr, 0, vi->hdr_len);
 465         xdpf->len   += vi->hdr_len;
 466 
 467         sg_init_one(sq->sg, xdpf->data, xdpf->len);
 468 
 469         err = virtqueue_add_outbuf(sq->vq, sq->sg, 1, xdp_to_ptr(xdpf),
 470                                    GFP_ATOMIC);
 471         if (unlikely(err))
 472                 return -ENOSPC; /* Caller handle free/refcnt */
 473 
 474         return 0;
 475 }
 476 
 477 static struct send_queue *virtnet_xdp_sq(struct virtnet_info *vi)
 478 {
 479         unsigned int qp;
 480 
 481         qp = vi->curr_queue_pairs - vi->xdp_queue_pairs + smp_processor_id();
 482         return &vi->sq[qp];
 483 }
 484 
 485 static int virtnet_xdp_xmit(struct net_device *dev,
 486                             int n, struct xdp_frame **frames, u32 flags)
 487 {
 488         struct virtnet_info *vi = netdev_priv(dev);
 489         struct receive_queue *rq = vi->rq;
 490         struct bpf_prog *xdp_prog;
 491         struct send_queue *sq;
 492         unsigned int len;
 493         int packets = 0;
 494         int bytes = 0;
 495         int drops = 0;
 496         int kicks = 0;
 497         int ret, err;
 498         void *ptr;
 499         int i;
 500 
 501         /* Only allow ndo_xdp_xmit if XDP is loaded on dev, as this
 502          * indicate XDP resources have been successfully allocated.
 503          */
 504         xdp_prog = rcu_dereference(rq->xdp_prog);
 505         if (!xdp_prog)
 506                 return -ENXIO;
 507 
 508         sq = virtnet_xdp_sq(vi);
 509 
 510         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK)) {
 511                 ret = -EINVAL;
 512                 drops = n;
 513                 goto out;
 514         }
 515 
 516         /* Free up any pending old buffers before queueing new ones. */
 517         while ((ptr = virtqueue_get_buf(sq->vq, &len)) != NULL) {
 518                 if (likely(is_xdp_frame(ptr))) {
 519                         struct xdp_frame *frame = ptr_to_xdp(ptr);
 520 
 521                         bytes += frame->len;
 522                         xdp_return_frame(frame);
 523                 } else {
 524                         struct sk_buff *skb = ptr;
 525 
 526                         bytes += skb->len;
 527                         napi_consume_skb(skb, false);
 528                 }
 529                 packets++;
 530         }
 531 
 532         for (i = 0; i < n; i++) {
 533                 struct xdp_frame *xdpf = frames[i];
 534 
 535                 err = __virtnet_xdp_xmit_one(vi, sq, xdpf);
 536                 if (err) {
 537                         xdp_return_frame_rx_napi(xdpf);
 538                         drops++;
 539                 }
 540         }
 541         ret = n - drops;
 542 
 543         if (flags & XDP_XMIT_FLUSH) {
 544                 if (virtqueue_kick_prepare(sq->vq) && virtqueue_notify(sq->vq))
 545                         kicks = 1;
 546         }
 547 out:
 548         u64_stats_update_begin(&sq->stats.syncp);
 549         sq->stats.bytes += bytes;
 550         sq->stats.packets += packets;
 551         sq->stats.xdp_tx += n;
 552         sq->stats.xdp_tx_drops += drops;
 553         sq->stats.kicks += kicks;
 554         u64_stats_update_end(&sq->stats.syncp);
 555 
 556         return ret;
 557 }
 558 
 559 static unsigned int virtnet_get_headroom(struct virtnet_info *vi)
 560 {
 561         return vi->xdp_queue_pairs ? VIRTIO_XDP_HEADROOM : 0;
 562 }
 563 
 564 /* We copy the packet for XDP in the following cases:
 565  *
 566  * 1) Packet is scattered across multiple rx buffers.
 567  * 2) Headroom space is insufficient.
 568  *
 569  * This is inefficient but it's a temporary condition that
 570  * we hit right after XDP is enabled and until queue is refilled
 571  * with large buffers with sufficient headroom - so it should affect
 572  * at most queue size packets.
 573  * Afterwards, the conditions to enable
 574  * XDP should preclude the underlying device from sending packets
 575  * across multiple buffers (num_buf > 1), and we make sure buffers
 576  * have enough headroom.
 577  */
 578 static struct page *xdp_linearize_page(struct receive_queue *rq,
 579                                        u16 *num_buf,
 580                                        struct page *p,
 581                                        int offset,
 582                                        int page_off,
 583                                        unsigned int *len)
 584 {
 585         struct page *page = alloc_page(GFP_ATOMIC);
 586 
 587         if (!page)
 588                 return NULL;
 589 
 590         memcpy(page_address(page) + page_off, page_address(p) + offset, *len);
 591         page_off += *len;
 592 
 593         while (--*num_buf) {
 594                 int tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
 595                 unsigned int buflen;
 596                 void *buf;
 597                 int off;
 598 
 599                 buf = virtqueue_get_buf(rq->vq, &buflen);
 600                 if (unlikely(!buf))
 601                         goto err_buf;
 602 
 603                 p = virt_to_head_page(buf);
 604                 off = buf - page_address(p);
 605 
 606                 /* guard against a misconfigured or uncooperative backend that
 607                  * is sending packet larger than the MTU.
 608                  */
 609                 if ((page_off + buflen + tailroom) > PAGE_SIZE) {
 610                         put_page(p);
 611                         goto err_buf;
 612                 }
 613 
 614                 memcpy(page_address(page) + page_off,
 615                        page_address(p) + off, buflen);
 616                 page_off += buflen;
 617                 put_page(p);
 618         }
 619 
 620         /* Headroom does not contribute to packet length */
 621         *len = page_off - VIRTIO_XDP_HEADROOM;
 622         return page;
 623 err_buf:
 624         __free_pages(page, 0);
 625         return NULL;
 626 }
 627 
 628 static struct sk_buff *receive_small(struct net_device *dev,
 629                                      struct virtnet_info *vi,
 630                                      struct receive_queue *rq,
 631                                      void *buf, void *ctx,
 632                                      unsigned int len,
 633                                      unsigned int *xdp_xmit,
 634                                      struct virtnet_rq_stats *stats)
 635 {
 636         struct sk_buff *skb;
 637         struct bpf_prog *xdp_prog;
 638         unsigned int xdp_headroom = (unsigned long)ctx;
 639         unsigned int header_offset = VIRTNET_RX_PAD + xdp_headroom;
 640         unsigned int headroom = vi->hdr_len + header_offset;
 641         unsigned int buflen = SKB_DATA_ALIGN(GOOD_PACKET_LEN + headroom) +
 642                               SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
 643         struct page *page = virt_to_head_page(buf);
 644         unsigned int delta = 0;
 645         struct page *xdp_page;
 646         int err;
 647 
 648         len -= vi->hdr_len;
 649         stats->bytes += len;
 650 
 651         rcu_read_lock();
 652         xdp_prog = rcu_dereference(rq->xdp_prog);
 653         if (xdp_prog) {
 654                 struct virtio_net_hdr_mrg_rxbuf *hdr = buf + header_offset;
 655                 struct xdp_frame *xdpf;
 656                 struct xdp_buff xdp;
 657                 void *orig_data;
 658                 u32 act;
 659 
 660                 if (unlikely(hdr->hdr.gso_type))
 661                         goto err_xdp;
 662 
 663                 if (unlikely(xdp_headroom < virtnet_get_headroom(vi))) {
 664                         int offset = buf - page_address(page) + header_offset;
 665                         unsigned int tlen = len + vi->hdr_len;
 666                         u16 num_buf = 1;
 667 
 668                         xdp_headroom = virtnet_get_headroom(vi);
 669                         header_offset = VIRTNET_RX_PAD + xdp_headroom;
 670                         headroom = vi->hdr_len + header_offset;
 671                         buflen = SKB_DATA_ALIGN(GOOD_PACKET_LEN + headroom) +
 672                                  SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
 673                         xdp_page = xdp_linearize_page(rq, &num_buf, page,
 674                                                       offset, header_offset,
 675                                                       &tlen);
 676                         if (!xdp_page)
 677                                 goto err_xdp;
 678 
 679                         buf = page_address(xdp_page);
 680                         put_page(page);
 681                         page = xdp_page;
 682                 }
 683 
 684                 xdp.data_hard_start = buf + VIRTNET_RX_PAD + vi->hdr_len;
 685                 xdp.data = xdp.data_hard_start + xdp_headroom;
 686                 xdp_set_data_meta_invalid(&xdp);
 687                 xdp.data_end = xdp.data + len;
 688                 xdp.rxq = &rq->xdp_rxq;
 689                 orig_data = xdp.data;
 690                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
 691                 stats->xdp_packets++;
 692 
 693                 switch (act) {
 694                 case XDP_PASS:
 695                         /* Recalculate length in case bpf program changed it */
 696                         delta = orig_data - xdp.data;
 697                         len = xdp.data_end - xdp.data;
 698                         break;
 699                 case XDP_TX:
 700                         stats->xdp_tx++;
 701                         xdpf = convert_to_xdp_frame(&xdp);
 702                         if (unlikely(!xdpf))
 703                                 goto err_xdp;
 704                         err = virtnet_xdp_xmit(dev, 1, &xdpf, 0);
 705                         if (unlikely(err < 0)) {
 706                                 trace_xdp_exception(vi->dev, xdp_prog, act);
 707                                 goto err_xdp;
 708                         }
 709                         *xdp_xmit |= VIRTIO_XDP_TX;
 710                         rcu_read_unlock();
 711                         goto xdp_xmit;
 712                 case XDP_REDIRECT:
 713                         stats->xdp_redirects++;
 714                         err = xdp_do_redirect(dev, &xdp, xdp_prog);
 715                         if (err)
 716                                 goto err_xdp;
 717                         *xdp_xmit |= VIRTIO_XDP_REDIR;
 718                         rcu_read_unlock();
 719                         goto xdp_xmit;
 720                 default:
 721                         bpf_warn_invalid_xdp_action(act);
 722                         /* fall through */
 723                 case XDP_ABORTED:
 724                         trace_xdp_exception(vi->dev, xdp_prog, act);
 725                 case XDP_DROP:
 726                         goto err_xdp;
 727                 }
 728         }
 729         rcu_read_unlock();
 730 
 731         skb = build_skb(buf, buflen);
 732         if (!skb) {
 733                 put_page(page);
 734                 goto err;
 735         }
 736         skb_reserve(skb, headroom - delta);
 737         skb_put(skb, len);
 738         if (!delta) {
 739                 buf += header_offset;
 740                 memcpy(skb_vnet_hdr(skb), buf, vi->hdr_len);
 741         } /* keep zeroed vnet hdr since packet was changed by bpf */
 742 
 743 err:
 744         return skb;
 745 
 746 err_xdp:
 747         rcu_read_unlock();
 748         stats->xdp_drops++;
 749         stats->drops++;
 750         put_page(page);
 751 xdp_xmit:
 752         return NULL;
 753 }
 754 
 755 static struct sk_buff *receive_big(struct net_device *dev,
 756                                    struct virtnet_info *vi,
 757                                    struct receive_queue *rq,
 758                                    void *buf,
 759                                    unsigned int len,
 760                                    struct virtnet_rq_stats *stats)
 761 {
 762         struct page *page = buf;
 763         struct sk_buff *skb = page_to_skb(vi, rq, page, 0, len,
 764                                           PAGE_SIZE, true);
 765 
 766         stats->bytes += len - vi->hdr_len;
 767         if (unlikely(!skb))
 768                 goto err;
 769 
 770         return skb;
 771 
 772 err:
 773         stats->drops++;
 774         give_pages(rq, page);
 775         return NULL;
 776 }
 777 
 778 static struct sk_buff *receive_mergeable(struct net_device *dev,
 779                                          struct virtnet_info *vi,
 780                                          struct receive_queue *rq,
 781                                          void *buf,
 782                                          void *ctx,
 783                                          unsigned int len,
 784                                          unsigned int *xdp_xmit,
 785                                          struct virtnet_rq_stats *stats)
 786 {
 787         struct virtio_net_hdr_mrg_rxbuf *hdr = buf;
 788         u16 num_buf = virtio16_to_cpu(vi->vdev, hdr->num_buffers);
 789         struct page *page = virt_to_head_page(buf);
 790         int offset = buf - page_address(page);
 791         struct sk_buff *head_skb, *curr_skb;
 792         struct bpf_prog *xdp_prog;
 793         unsigned int truesize;
 794         unsigned int headroom = mergeable_ctx_to_headroom(ctx);
 795         int err;
 796 
 797         head_skb = NULL;
 798         stats->bytes += len - vi->hdr_len;
 799 
 800         rcu_read_lock();
 801         xdp_prog = rcu_dereference(rq->xdp_prog);
 802         if (xdp_prog) {
 803                 struct xdp_frame *xdpf;
 804                 struct page *xdp_page;
 805                 struct xdp_buff xdp;
 806                 void *data;
 807                 u32 act;
 808 
 809                 /* Transient failure which in theory could occur if
 810                  * in-flight packets from before XDP was enabled reach
 811                  * the receive path after XDP is loaded.
 812                  */
 813                 if (unlikely(hdr->hdr.gso_type))
 814                         goto err_xdp;
 815 
 816                 /* This happens when rx buffer size is underestimated
 817                  * or headroom is not enough because of the buffer
 818                  * was refilled before XDP is set. This should only
 819                  * happen for the first several packets, so we don't
 820                  * care much about its performance.
 821                  */
 822                 if (unlikely(num_buf > 1 ||
 823                              headroom < virtnet_get_headroom(vi))) {
 824                         /* linearize data for XDP */
 825                         xdp_page = xdp_linearize_page(rq, &num_buf,
 826                                                       page, offset,
 827                                                       VIRTIO_XDP_HEADROOM,
 828                                                       &len);
 829                         if (!xdp_page)
 830                                 goto err_xdp;
 831                         offset = VIRTIO_XDP_HEADROOM;
 832                 } else {
 833                         xdp_page = page;
 834                 }
 835 
 836                 /* Allow consuming headroom but reserve enough space to push
 837                  * the descriptor on if we get an XDP_TX return code.
 838                  */
 839                 data = page_address(xdp_page) + offset;
 840                 xdp.data_hard_start = data - VIRTIO_XDP_HEADROOM + vi->hdr_len;
 841                 xdp.data = data + vi->hdr_len;
 842                 xdp_set_data_meta_invalid(&xdp);
 843                 xdp.data_end = xdp.data + (len - vi->hdr_len);
 844                 xdp.rxq = &rq->xdp_rxq;
 845 
 846                 act = bpf_prog_run_xdp(xdp_prog, &xdp);
 847                 stats->xdp_packets++;
 848 
 849                 switch (act) {
 850                 case XDP_PASS:
 851                         /* recalculate offset to account for any header
 852                          * adjustments. Note other cases do not build an
 853                          * skb and avoid using offset
 854                          */
 855                         offset = xdp.data -
 856                                         page_address(xdp_page) - vi->hdr_len;
 857 
 858                         /* recalculate len if xdp.data or xdp.data_end were
 859                          * adjusted
 860                          */
 861                         len = xdp.data_end - xdp.data + vi->hdr_len;
 862                         /* We can only create skb based on xdp_page. */
 863                         if (unlikely(xdp_page != page)) {
 864                                 rcu_read_unlock();
 865                                 put_page(page);
 866                                 head_skb = page_to_skb(vi, rq, xdp_page,
 867                                                        offset, len,
 868                                                        PAGE_SIZE, false);
 869                                 return head_skb;
 870                         }
 871                         break;
 872                 case XDP_TX:
 873                         stats->xdp_tx++;
 874                         xdpf = convert_to_xdp_frame(&xdp);
 875                         if (unlikely(!xdpf))
 876                                 goto err_xdp;
 877                         err = virtnet_xdp_xmit(dev, 1, &xdpf, 0);
 878                         if (unlikely(err < 0)) {
 879                                 trace_xdp_exception(vi->dev, xdp_prog, act);
 880                                 if (unlikely(xdp_page != page))
 881                                         put_page(xdp_page);
 882                                 goto err_xdp;
 883                         }
 884                         *xdp_xmit |= VIRTIO_XDP_TX;
 885                         if (unlikely(xdp_page != page))
 886                                 put_page(page);
 887                         rcu_read_unlock();
 888                         goto xdp_xmit;
 889                 case XDP_REDIRECT:
 890                         stats->xdp_redirects++;
 891                         err = xdp_do_redirect(dev, &xdp, xdp_prog);
 892                         if (err) {
 893                                 if (unlikely(xdp_page != page))
 894                                         put_page(xdp_page);
 895                                 goto err_xdp;
 896                         }
 897                         *xdp_xmit |= VIRTIO_XDP_REDIR;
 898                         if (unlikely(xdp_page != page))
 899                                 put_page(page);
 900                         rcu_read_unlock();
 901                         goto xdp_xmit;
 902                 default:
 903                         bpf_warn_invalid_xdp_action(act);
 904                         /* fall through */
 905                 case XDP_ABORTED:
 906                         trace_xdp_exception(vi->dev, xdp_prog, act);
 907                         /* fall through */
 908                 case XDP_DROP:
 909                         if (unlikely(xdp_page != page))
 910                                 __free_pages(xdp_page, 0);
 911                         goto err_xdp;
 912                 }
 913         }
 914         rcu_read_unlock();
 915 
 916         truesize = mergeable_ctx_to_truesize(ctx);
 917         if (unlikely(len > truesize)) {
 918                 pr_debug("%s: rx error: len %u exceeds truesize %lu\n",
 919                          dev->name, len, (unsigned long)ctx);
 920                 dev->stats.rx_length_errors++;
 921                 goto err_skb;
 922         }
 923 
 924         head_skb = page_to_skb(vi, rq, page, offset, len, truesize, !xdp_prog);
 925         curr_skb = head_skb;
 926 
 927         if (unlikely(!curr_skb))
 928                 goto err_skb;
 929         while (--num_buf) {
 930                 int num_skb_frags;
 931 
 932                 buf = virtqueue_get_buf_ctx(rq->vq, &len, &ctx);
 933                 if (unlikely(!buf)) {
 934                         pr_debug("%s: rx error: %d buffers out of %d missing\n",
 935                                  dev->name, num_buf,
 936                                  virtio16_to_cpu(vi->vdev,
 937                                                  hdr->num_buffers));
 938                         dev->stats.rx_length_errors++;
 939                         goto err_buf;
 940                 }
 941 
 942                 stats->bytes += len;
 943                 page = virt_to_head_page(buf);
 944 
 945                 truesize = mergeable_ctx_to_truesize(ctx);
 946                 if (unlikely(len > truesize)) {
 947                         pr_debug("%s: rx error: len %u exceeds truesize %lu\n",
 948                                  dev->name, len, (unsigned long)ctx);
 949                         dev->stats.rx_length_errors++;
 950                         goto err_skb;
 951                 }
 952 
 953                 num_skb_frags = skb_shinfo(curr_skb)->nr_frags;
 954                 if (unlikely(num_skb_frags == MAX_SKB_FRAGS)) {
 955                         struct sk_buff *nskb = alloc_skb(0, GFP_ATOMIC);
 956 
 957                         if (unlikely(!nskb))
 958                                 goto err_skb;
 959                         if (curr_skb == head_skb)
 960                                 skb_shinfo(curr_skb)->frag_list = nskb;
 961                         else
 962                                 curr_skb->next = nskb;
 963                         curr_skb = nskb;
 964                         head_skb->truesize += nskb->truesize;
 965                         num_skb_frags = 0;
 966                 }
 967                 if (curr_skb != head_skb) {
 968                         head_skb->data_len += len;
 969                         head_skb->len += len;
 970                         head_skb->truesize += truesize;
 971                 }
 972                 offset = buf - page_address(page);
 973                 if (skb_can_coalesce(curr_skb, num_skb_frags, page, offset)) {
 974                         put_page(page);
 975                         skb_coalesce_rx_frag(curr_skb, num_skb_frags - 1,
 976                                              len, truesize);
 977                 } else {
 978                         skb_add_rx_frag(curr_skb, num_skb_frags, page,
 979                                         offset, len, truesize);
 980                 }
 981         }
 982 
 983         ewma_pkt_len_add(&rq->mrg_avg_pkt_len, head_skb->len);
 984         return head_skb;
 985 
 986 err_xdp:
 987         rcu_read_unlock();
 988         stats->xdp_drops++;
 989 err_skb:
 990         put_page(page);
 991         while (num_buf-- > 1) {
 992                 buf = virtqueue_get_buf(rq->vq, &len);
 993                 if (unlikely(!buf)) {
 994                         pr_debug("%s: rx error: %d buffers missing\n",
 995                                  dev->name, num_buf);
 996                         dev->stats.rx_length_errors++;
 997                         break;
 998                 }
 999                 stats->bytes += len;
1000                 page = virt_to_head_page(buf);
1001                 put_page(page);
1002         }
1003 err_buf:
1004         stats->drops++;
1005         dev_kfree_skb(head_skb);
1006 xdp_xmit:
1007         return NULL;
1008 }
1009 
1010 static void receive_buf(struct virtnet_info *vi, struct receive_queue *rq,
1011                         void *buf, unsigned int len, void **ctx,
1012                         unsigned int *xdp_xmit,
1013                         struct virtnet_rq_stats *stats)
1014 {
1015         struct net_device *dev = vi->dev;
1016         struct sk_buff *skb;
1017         struct virtio_net_hdr_mrg_rxbuf *hdr;
1018 
1019         if (unlikely(len < vi->hdr_len + ETH_HLEN)) {
1020                 pr_debug("%s: short packet %i\n", dev->name, len);
1021                 dev->stats.rx_length_errors++;
1022                 if (vi->mergeable_rx_bufs) {
1023                         put_page(virt_to_head_page(buf));
1024                 } else if (vi->big_packets) {
1025                         give_pages(rq, buf);
1026                 } else {
1027                         put_page(virt_to_head_page(buf));
1028                 }
1029                 return;
1030         }
1031 
1032         if (vi->mergeable_rx_bufs)
1033                 skb = receive_mergeable(dev, vi, rq, buf, ctx, len, xdp_xmit,
1034                                         stats);
1035         else if (vi->big_packets)
1036                 skb = receive_big(dev, vi, rq, buf, len, stats);
1037         else
1038                 skb = receive_small(dev, vi, rq, buf, ctx, len, xdp_xmit, stats);
1039 
1040         if (unlikely(!skb))
1041                 return;
1042 
1043         hdr = skb_vnet_hdr(skb);
1044 
1045         if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID)
1046                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1047 
1048         if (virtio_net_hdr_to_skb(skb, &hdr->hdr,
1049                                   virtio_is_little_endian(vi->vdev))) {
1050                 net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n",
1051                                      dev->name, hdr->hdr.gso_type,
1052                                      hdr->hdr.gso_size);
1053                 goto frame_err;
1054         }
1055 
1056         skb_record_rx_queue(skb, vq2rxq(rq->vq));
1057         skb->protocol = eth_type_trans(skb, dev);
1058         pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
1059                  ntohs(skb->protocol), skb->len, skb->pkt_type);
1060 
1061         napi_gro_receive(&rq->napi, skb);
1062         return;
1063 
1064 frame_err:
1065         dev->stats.rx_frame_errors++;
1066         dev_kfree_skb(skb);
1067 }
1068 
1069 /* Unlike mergeable buffers, all buffers are allocated to the
1070  * same size, except for the headroom. For this reason we do
1071  * not need to use  mergeable_len_to_ctx here - it is enough
1072  * to store the headroom as the context ignoring the truesize.
1073  */
1074 static int add_recvbuf_small(struct virtnet_info *vi, struct receive_queue *rq,
1075                              gfp_t gfp)
1076 {
1077         struct page_frag *alloc_frag = &rq->alloc_frag;
1078         char *buf;
1079         unsigned int xdp_headroom = virtnet_get_headroom(vi);
1080         void *ctx = (void *)(unsigned long)xdp_headroom;
1081         int len = vi->hdr_len + VIRTNET_RX_PAD + GOOD_PACKET_LEN + xdp_headroom;
1082         int err;
1083 
1084         len = SKB_DATA_ALIGN(len) +
1085               SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1086         if (unlikely(!skb_page_frag_refill(len, alloc_frag, gfp)))
1087                 return -ENOMEM;
1088 
1089         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1090         get_page(alloc_frag->page);
1091         alloc_frag->offset += len;
1092         sg_init_one(rq->sg, buf + VIRTNET_RX_PAD + xdp_headroom,
1093                     vi->hdr_len + GOOD_PACKET_LEN);
1094         err = virtqueue_add_inbuf_ctx(rq->vq, rq->sg, 1, buf, ctx, gfp);
1095         if (err < 0)
1096                 put_page(virt_to_head_page(buf));
1097         return err;
1098 }
1099 
1100 static int add_recvbuf_big(struct virtnet_info *vi, struct receive_queue *rq,
1101                            gfp_t gfp)
1102 {
1103         struct page *first, *list = NULL;
1104         char *p;
1105         int i, err, offset;
1106 
1107         sg_init_table(rq->sg, MAX_SKB_FRAGS + 2);
1108 
1109         /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */
1110         for (i = MAX_SKB_FRAGS + 1; i > 1; --i) {
1111                 first = get_a_page(rq, gfp);
1112                 if (!first) {
1113                         if (list)
1114                                 give_pages(rq, list);
1115                         return -ENOMEM;
1116                 }
1117                 sg_set_buf(&rq->sg[i], page_address(first), PAGE_SIZE);
1118 
1119                 /* chain new page in list head to match sg */
1120                 first->private = (unsigned long)list;
1121                 list = first;
1122         }
1123 
1124         first = get_a_page(rq, gfp);
1125         if (!first) {
1126                 give_pages(rq, list);
1127                 return -ENOMEM;
1128         }
1129         p = page_address(first);
1130 
1131         /* rq->sg[0], rq->sg[1] share the same page */
1132         /* a separated rq->sg[0] for header - required in case !any_header_sg */
1133         sg_set_buf(&rq->sg[0], p, vi->hdr_len);
1134 
1135         /* rq->sg[1] for data packet, from offset */
1136         offset = sizeof(struct padded_vnet_hdr);
1137         sg_set_buf(&rq->sg[1], p + offset, PAGE_SIZE - offset);
1138 
1139         /* chain first in list head */
1140         first->private = (unsigned long)list;
1141         err = virtqueue_add_inbuf(rq->vq, rq->sg, MAX_SKB_FRAGS + 2,
1142                                   first, gfp);
1143         if (err < 0)
1144                 give_pages(rq, first);
1145 
1146         return err;
1147 }
1148 
1149 static unsigned int get_mergeable_buf_len(struct receive_queue *rq,
1150                                           struct ewma_pkt_len *avg_pkt_len,
1151                                           unsigned int room)
1152 {
1153         const size_t hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1154         unsigned int len;
1155 
1156         if (room)
1157                 return PAGE_SIZE - room;
1158 
1159         len = hdr_len + clamp_t(unsigned int, ewma_pkt_len_read(avg_pkt_len),
1160                                 rq->min_buf_len, PAGE_SIZE - hdr_len);
1161 
1162         return ALIGN(len, L1_CACHE_BYTES);
1163 }
1164 
1165 static int add_recvbuf_mergeable(struct virtnet_info *vi,
1166                                  struct receive_queue *rq, gfp_t gfp)
1167 {
1168         struct page_frag *alloc_frag = &rq->alloc_frag;
1169         unsigned int headroom = virtnet_get_headroom(vi);
1170         unsigned int tailroom = headroom ? sizeof(struct skb_shared_info) : 0;
1171         unsigned int room = SKB_DATA_ALIGN(headroom + tailroom);
1172         char *buf;
1173         void *ctx;
1174         int err;
1175         unsigned int len, hole;
1176 
1177         /* Extra tailroom is needed to satisfy XDP's assumption. This
1178          * means rx frags coalescing won't work, but consider we've
1179          * disabled GSO for XDP, it won't be a big issue.
1180          */
1181         len = get_mergeable_buf_len(rq, &rq->mrg_avg_pkt_len, room);
1182         if (unlikely(!skb_page_frag_refill(len + room, alloc_frag, gfp)))
1183                 return -ENOMEM;
1184 
1185         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1186         buf += headroom; /* advance address leaving hole at front of pkt */
1187         get_page(alloc_frag->page);
1188         alloc_frag->offset += len + room;
1189         hole = alloc_frag->size - alloc_frag->offset;
1190         if (hole < len + room) {
1191                 /* To avoid internal fragmentation, if there is very likely not
1192                  * enough space for another buffer, add the remaining space to
1193                  * the current buffer.
1194                  */
1195                 len += hole;
1196                 alloc_frag->offset += hole;
1197         }
1198 
1199         sg_init_one(rq->sg, buf, len);
1200         ctx = mergeable_len_to_ctx(len, headroom);
1201         err = virtqueue_add_inbuf_ctx(rq->vq, rq->sg, 1, buf, ctx, gfp);
1202         if (err < 0)
1203                 put_page(virt_to_head_page(buf));
1204 
1205         return err;
1206 }
1207 
1208 /*
1209  * Returns false if we couldn't fill entirely (OOM).
1210  *
1211  * Normally run in the receive path, but can also be run from ndo_open
1212  * before we're receiving packets, or from refill_work which is
1213  * careful to disable receiving (using napi_disable).
1214  */
1215 static bool try_fill_recv(struct virtnet_info *vi, struct receive_queue *rq,
1216                           gfp_t gfp)
1217 {
1218         int err;
1219         bool oom;
1220 
1221         do {
1222                 if (vi->mergeable_rx_bufs)
1223                         err = add_recvbuf_mergeable(vi, rq, gfp);
1224                 else if (vi->big_packets)
1225                         err = add_recvbuf_big(vi, rq, gfp);
1226                 else
1227                         err = add_recvbuf_small(vi, rq, gfp);
1228 
1229                 oom = err == -ENOMEM;
1230                 if (err)
1231                         break;
1232         } while (rq->vq->num_free);
1233         if (virtqueue_kick_prepare(rq->vq) && virtqueue_notify(rq->vq)) {
1234                 unsigned long flags;
1235 
1236                 flags = u64_stats_update_begin_irqsave(&rq->stats.syncp);
1237                 rq->stats.kicks++;
1238                 u64_stats_update_end_irqrestore(&rq->stats.syncp, flags);
1239         }
1240 
1241         return !oom;
1242 }
1243 
1244 static void skb_recv_done(struct virtqueue *rvq)
1245 {
1246         struct virtnet_info *vi = rvq->vdev->priv;
1247         struct receive_queue *rq = &vi->rq[vq2rxq(rvq)];
1248 
1249         virtqueue_napi_schedule(&rq->napi, rvq);
1250 }
1251 
1252 static void virtnet_napi_enable(struct virtqueue *vq, struct napi_struct *napi)
1253 {
1254         napi_enable(napi);
1255 
1256         /* If all buffers were filled by other side before we napi_enabled, we
1257          * won't get another interrupt, so process any outstanding packets now.
1258          * Call local_bh_enable after to trigger softIRQ processing.
1259          */
1260         local_bh_disable();
1261         virtqueue_napi_schedule(napi, vq);
1262         local_bh_enable();
1263 }
1264 
1265 static void virtnet_napi_tx_enable(struct virtnet_info *vi,
1266                                    struct virtqueue *vq,
1267                                    struct napi_struct *napi)
1268 {
1269         if (!napi->weight)
1270                 return;
1271 
1272         /* Tx napi touches cachelines on the cpu handling tx interrupts. Only
1273          * enable the feature if this is likely affine with the transmit path.
1274          */
1275         if (!vi->affinity_hint_set) {
1276                 napi->weight = 0;
1277                 return;
1278         }
1279 
1280         return virtnet_napi_enable(vq, napi);
1281 }
1282 
1283 static void virtnet_napi_tx_disable(struct napi_struct *napi)
1284 {
1285         if (napi->weight)
1286                 napi_disable(napi);
1287 }
1288 
1289 static void refill_work(struct work_struct *work)
1290 {
1291         struct virtnet_info *vi =
1292                 container_of(work, struct virtnet_info, refill.work);
1293         bool still_empty;
1294         int i;
1295 
1296         for (i = 0; i < vi->curr_queue_pairs; i++) {
1297                 struct receive_queue *rq = &vi->rq[i];
1298 
1299                 napi_disable(&rq->napi);
1300                 still_empty = !try_fill_recv(vi, rq, GFP_KERNEL);
1301                 virtnet_napi_enable(rq->vq, &rq->napi);
1302 
1303                 /* In theory, this can happen: if we don't get any buffers in
1304                  * we will *never* try to fill again.
1305                  */
1306                 if (still_empty)
1307                         schedule_delayed_work(&vi->refill, HZ/2);
1308         }
1309 }
1310 
1311 static int virtnet_receive(struct receive_queue *rq, int budget,
1312                            unsigned int *xdp_xmit)
1313 {
1314         struct virtnet_info *vi = rq->vq->vdev->priv;
1315         struct virtnet_rq_stats stats = {};
1316         unsigned int len;
1317         void *buf;
1318         int i;
1319 
1320         if (!vi->big_packets || vi->mergeable_rx_bufs) {
1321                 void *ctx;
1322 
1323                 while (stats.packets < budget &&
1324                        (buf = virtqueue_get_buf_ctx(rq->vq, &len, &ctx))) {
1325                         receive_buf(vi, rq, buf, len, ctx, xdp_xmit, &stats);
1326                         stats.packets++;
1327                 }
1328         } else {
1329                 while (stats.packets < budget &&
1330                        (buf = virtqueue_get_buf(rq->vq, &len)) != NULL) {
1331                         receive_buf(vi, rq, buf, len, NULL, xdp_xmit, &stats);
1332                         stats.packets++;
1333                 }
1334         }
1335 
1336         if (rq->vq->num_free > min((unsigned int)budget, virtqueue_get_vring_size(rq->vq)) / 2) {
1337                 if (!try_fill_recv(vi, rq, GFP_ATOMIC))
1338                         schedule_delayed_work(&vi->refill, 0);
1339         }
1340 
1341         u64_stats_update_begin(&rq->stats.syncp);
1342         for (i = 0; i < VIRTNET_RQ_STATS_LEN; i++) {
1343                 size_t offset = virtnet_rq_stats_desc[i].offset;
1344                 u64 *item;
1345 
1346                 item = (u64 *)((u8 *)&rq->stats + offset);
1347                 *item += *(u64 *)((u8 *)&stats + offset);
1348         }
1349         u64_stats_update_end(&rq->stats.syncp);
1350 
1351         return stats.packets;
1352 }
1353 
1354 static void free_old_xmit_skbs(struct send_queue *sq, bool in_napi)
1355 {
1356         unsigned int len;
1357         unsigned int packets = 0;
1358         unsigned int bytes = 0;
1359         void *ptr;
1360 
1361         while ((ptr = virtqueue_get_buf(sq->vq, &len)) != NULL) {
1362                 if (likely(!is_xdp_frame(ptr))) {
1363                         struct sk_buff *skb = ptr;
1364 
1365                         pr_debug("Sent skb %p\n", skb);
1366 
1367                         bytes += skb->len;
1368                         napi_consume_skb(skb, in_napi);
1369                 } else {
1370                         struct xdp_frame *frame = ptr_to_xdp(ptr);
1371 
1372                         bytes += frame->len;
1373                         xdp_return_frame(frame);
1374                 }
1375                 packets++;
1376         }
1377 
1378         /* Avoid overhead when no packets have been processed
1379          * happens when called speculatively from start_xmit.
1380          */
1381         if (!packets)
1382                 return;
1383 
1384         u64_stats_update_begin(&sq->stats.syncp);
1385         sq->stats.bytes += bytes;
1386         sq->stats.packets += packets;
1387         u64_stats_update_end(&sq->stats.syncp);
1388 }
1389 
1390 static bool is_xdp_raw_buffer_queue(struct virtnet_info *vi, int q)
1391 {
1392         if (q < (vi->curr_queue_pairs - vi->xdp_queue_pairs))
1393                 return false;
1394         else if (q < vi->curr_queue_pairs)
1395                 return true;
1396         else
1397                 return false;
1398 }
1399 
1400 static void virtnet_poll_cleantx(struct receive_queue *rq)
1401 {
1402         struct virtnet_info *vi = rq->vq->vdev->priv;
1403         unsigned int index = vq2rxq(rq->vq);
1404         struct send_queue *sq = &vi->sq[index];
1405         struct netdev_queue *txq = netdev_get_tx_queue(vi->dev, index);
1406 
1407         if (!sq->napi.weight || is_xdp_raw_buffer_queue(vi, index))
1408                 return;
1409 
1410         if (__netif_tx_trylock(txq)) {
1411                 free_old_xmit_skbs(sq, true);
1412                 __netif_tx_unlock(txq);
1413         }
1414 
1415         if (sq->vq->num_free >= 2 + MAX_SKB_FRAGS)
1416                 netif_tx_wake_queue(txq);
1417 }
1418 
1419 static int virtnet_poll(struct napi_struct *napi, int budget)
1420 {
1421         struct receive_queue *rq =
1422                 container_of(napi, struct receive_queue, napi);
1423         struct virtnet_info *vi = rq->vq->vdev->priv;
1424         struct send_queue *sq;
1425         unsigned int received;
1426         unsigned int xdp_xmit = 0;
1427 
1428         virtnet_poll_cleantx(rq);
1429 
1430         received = virtnet_receive(rq, budget, &xdp_xmit);
1431 
1432         /* Out of packets? */
1433         if (received < budget)
1434                 virtqueue_napi_complete(napi, rq->vq, received);
1435 
1436         if (xdp_xmit & VIRTIO_XDP_REDIR)
1437                 xdp_do_flush_map();
1438 
1439         if (xdp_xmit & VIRTIO_XDP_TX) {
1440                 sq = virtnet_xdp_sq(vi);
1441                 if (virtqueue_kick_prepare(sq->vq) && virtqueue_notify(sq->vq)) {
1442                         u64_stats_update_begin(&sq->stats.syncp);
1443                         sq->stats.kicks++;
1444                         u64_stats_update_end(&sq->stats.syncp);
1445                 }
1446         }
1447 
1448         return received;
1449 }
1450 
1451 static int virtnet_open(struct net_device *dev)
1452 {
1453         struct virtnet_info *vi = netdev_priv(dev);
1454         int i, err;
1455 
1456         for (i = 0; i < vi->max_queue_pairs; i++) {
1457                 if (i < vi->curr_queue_pairs)
1458                         /* Make sure we have some buffers: if oom use wq. */
1459                         if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL))
1460                                 schedule_delayed_work(&vi->refill, 0);
1461 
1462                 err = xdp_rxq_info_reg(&vi->rq[i].xdp_rxq, dev, i);
1463                 if (err < 0)
1464                         return err;
1465 
1466                 err = xdp_rxq_info_reg_mem_model(&vi->rq[i].xdp_rxq,
1467                                                  MEM_TYPE_PAGE_SHARED, NULL);
1468                 if (err < 0) {
1469                         xdp_rxq_info_unreg(&vi->rq[i].xdp_rxq);
1470                         return err;
1471                 }
1472 
1473                 virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi);
1474                 virtnet_napi_tx_enable(vi, vi->sq[i].vq, &vi->sq[i].napi);
1475         }
1476 
1477         return 0;
1478 }
1479 
1480 static int virtnet_poll_tx(struct napi_struct *napi, int budget)
1481 {
1482         struct send_queue *sq = container_of(napi, struct send_queue, napi);
1483         struct virtnet_info *vi = sq->vq->vdev->priv;
1484         unsigned int index = vq2txq(sq->vq);
1485         struct netdev_queue *txq;
1486 
1487         if (unlikely(is_xdp_raw_buffer_queue(vi, index))) {
1488                 /* We don't need to enable cb for XDP */
1489                 napi_complete_done(napi, 0);
1490                 return 0;
1491         }
1492 
1493         txq = netdev_get_tx_queue(vi->dev, index);
1494         __netif_tx_lock(txq, raw_smp_processor_id());
1495         free_old_xmit_skbs(sq, true);
1496         __netif_tx_unlock(txq);
1497 
1498         virtqueue_napi_complete(napi, sq->vq, 0);
1499 
1500         if (sq->vq->num_free >= 2 + MAX_SKB_FRAGS)
1501                 netif_tx_wake_queue(txq);
1502 
1503         return 0;
1504 }
1505 
1506 static int xmit_skb(struct send_queue *sq, struct sk_buff *skb)
1507 {
1508         struct virtio_net_hdr_mrg_rxbuf *hdr;
1509         const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
1510         struct virtnet_info *vi = sq->vq->vdev->priv;
1511         int num_sg;
1512         unsigned hdr_len = vi->hdr_len;
1513         bool can_push;
1514 
1515         pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
1516 
1517         can_push = vi->any_header_sg &&
1518                 !((unsigned long)skb->data & (__alignof__(*hdr) - 1)) &&
1519                 !skb_header_cloned(skb) && skb_headroom(skb) >= hdr_len;
1520         /* Even if we can, don't push here yet as this would skew
1521          * csum_start offset below. */
1522         if (can_push)
1523                 hdr = (struct virtio_net_hdr_mrg_rxbuf *)(skb->data - hdr_len);
1524         else
1525                 hdr = skb_vnet_hdr(skb);
1526 
1527         if (virtio_net_hdr_from_skb(skb, &hdr->hdr,
1528                                     virtio_is_little_endian(vi->vdev), false,
1529                                     0))
1530                 BUG();
1531 
1532         if (vi->mergeable_rx_bufs)
1533                 hdr->num_buffers = 0;
1534 
1535         sg_init_table(sq->sg, skb_shinfo(skb)->nr_frags + (can_push ? 1 : 2));
1536         if (can_push) {
1537                 __skb_push(skb, hdr_len);
1538                 num_sg = skb_to_sgvec(skb, sq->sg, 0, skb->len);
1539                 if (unlikely(num_sg < 0))
1540                         return num_sg;
1541                 /* Pull header back to avoid skew in tx bytes calculations. */
1542                 __skb_pull(skb, hdr_len);
1543         } else {
1544                 sg_set_buf(sq->sg, hdr, hdr_len);
1545                 num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len);
1546                 if (unlikely(num_sg < 0))
1547                         return num_sg;
1548                 num_sg++;
1549         }
1550         return virtqueue_add_outbuf(sq->vq, sq->sg, num_sg, skb, GFP_ATOMIC);
1551 }
1552 
1553 static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
1554 {
1555         struct virtnet_info *vi = netdev_priv(dev);
1556         int qnum = skb_get_queue_mapping(skb);
1557         struct send_queue *sq = &vi->sq[qnum];
1558         int err;
1559         struct netdev_queue *txq = netdev_get_tx_queue(dev, qnum);
1560         bool kick = !netdev_xmit_more();
1561         bool use_napi = sq->napi.weight;
1562 
1563         /* Free up any pending old buffers before queueing new ones. */
1564         free_old_xmit_skbs(sq, false);
1565 
1566         if (use_napi && kick)
1567                 virtqueue_enable_cb_delayed(sq->vq);
1568 
1569         /* timestamp packet in software */
1570         skb_tx_timestamp(skb);
1571 
1572         /* Try to transmit */
1573         err = xmit_skb(sq, skb);
1574 
1575         /* This should not happen! */
1576         if (unlikely(err)) {
1577                 dev->stats.tx_fifo_errors++;
1578                 if (net_ratelimit())
1579                         dev_warn(&dev->dev,
1580                                  "Unexpected TXQ (%d) queue failure: %d\n",
1581                                  qnum, err);
1582                 dev->stats.tx_dropped++;
1583                 dev_kfree_skb_any(skb);
1584                 return NETDEV_TX_OK;
1585         }
1586 
1587         /* Don't wait up for transmitted skbs to be freed. */
1588         if (!use_napi) {
1589                 skb_orphan(skb);
1590                 nf_reset_ct(skb);
1591         }
1592 
1593         /* If running out of space, stop queue to avoid getting packets that we
1594          * are then unable to transmit.
1595          * An alternative would be to force queuing layer to requeue the skb by
1596          * returning NETDEV_TX_BUSY. However, NETDEV_TX_BUSY should not be
1597          * returned in a normal path of operation: it means that driver is not
1598          * maintaining the TX queue stop/start state properly, and causes
1599          * the stack to do a non-trivial amount of useless work.
1600          * Since most packets only take 1 or 2 ring slots, stopping the queue
1601          * early means 16 slots are typically wasted.
1602          */
1603         if (sq->vq->num_free < 2+MAX_SKB_FRAGS) {
1604                 netif_stop_subqueue(dev, qnum);
1605                 if (!use_napi &&
1606                     unlikely(!virtqueue_enable_cb_delayed(sq->vq))) {
1607                         /* More just got used, free them then recheck. */
1608                         free_old_xmit_skbs(sq, false);
1609                         if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) {
1610                                 netif_start_subqueue(dev, qnum);
1611                                 virtqueue_disable_cb(sq->vq);
1612                         }
1613                 }
1614         }
1615 
1616         if (kick || netif_xmit_stopped(txq)) {
1617                 if (virtqueue_kick_prepare(sq->vq) && virtqueue_notify(sq->vq)) {
1618                         u64_stats_update_begin(&sq->stats.syncp);
1619                         sq->stats.kicks++;
1620                         u64_stats_update_end(&sq->stats.syncp);
1621                 }
1622         }
1623 
1624         return NETDEV_TX_OK;
1625 }
1626 
1627 /*
1628  * Send command via the control virtqueue and check status.  Commands
1629  * supported by the hypervisor, as indicated by feature bits, should
1630  * never fail unless improperly formatted.
1631  */
1632 static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
1633                                  struct scatterlist *out)
1634 {
1635         struct scatterlist *sgs[4], hdr, stat;
1636         unsigned out_num = 0, tmp;
1637 
1638         /* Caller should know better */
1639         BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ));
1640 
1641         vi->ctrl->status = ~0;
1642         vi->ctrl->hdr.class = class;
1643         vi->ctrl->hdr.cmd = cmd;
1644         /* Add header */
1645         sg_init_one(&hdr, &vi->ctrl->hdr, sizeof(vi->ctrl->hdr));
1646         sgs[out_num++] = &hdr;
1647 
1648         if (out)
1649                 sgs[out_num++] = out;
1650 
1651         /* Add return status. */
1652         sg_init_one(&stat, &vi->ctrl->status, sizeof(vi->ctrl->status));
1653         sgs[out_num] = &stat;
1654 
1655         BUG_ON(out_num + 1 > ARRAY_SIZE(sgs));
1656         virtqueue_add_sgs(vi->cvq, sgs, out_num, 1, vi, GFP_ATOMIC);
1657 
1658         if (unlikely(!virtqueue_kick(vi->cvq)))
1659                 return vi->ctrl->status == VIRTIO_NET_OK;
1660 
1661         /* Spin for a response, the kick causes an ioport write, trapping
1662          * into the hypervisor, so the request should be handled immediately.
1663          */
1664         while (!virtqueue_get_buf(vi->cvq, &tmp) &&
1665                !virtqueue_is_broken(vi->cvq))
1666                 cpu_relax();
1667 
1668         return vi->ctrl->status == VIRTIO_NET_OK;
1669 }
1670 
1671 static int virtnet_set_mac_address(struct net_device *dev, void *p)
1672 {
1673         struct virtnet_info *vi = netdev_priv(dev);
1674         struct virtio_device *vdev = vi->vdev;
1675         int ret;
1676         struct sockaddr *addr;
1677         struct scatterlist sg;
1678 
1679         if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STANDBY))
1680                 return -EOPNOTSUPP;
1681 
1682         addr = kmemdup(p, sizeof(*addr), GFP_KERNEL);
1683         if (!addr)
1684                 return -ENOMEM;
1685 
1686         ret = eth_prepare_mac_addr_change(dev, addr);
1687         if (ret)
1688                 goto out;
1689 
1690         if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1691                 sg_init_one(&sg, addr->sa_data, dev->addr_len);
1692                 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
1693                                           VIRTIO_NET_CTRL_MAC_ADDR_SET, &sg)) {
1694                         dev_warn(&vdev->dev,
1695                                  "Failed to set mac address by vq command.\n");
1696                         ret = -EINVAL;
1697                         goto out;
1698                 }
1699         } else if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC) &&
1700                    !virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) {
1701                 unsigned int i;
1702 
1703                 /* Naturally, this has an atomicity problem. */
1704                 for (i = 0; i < dev->addr_len; i++)
1705                         virtio_cwrite8(vdev,
1706                                        offsetof(struct virtio_net_config, mac) +
1707                                        i, addr->sa_data[i]);
1708         }
1709 
1710         eth_commit_mac_addr_change(dev, p);
1711         ret = 0;
1712 
1713 out:
1714         kfree(addr);
1715         return ret;
1716 }
1717 
1718 static void virtnet_stats(struct net_device *dev,
1719                           struct rtnl_link_stats64 *tot)
1720 {
1721         struct virtnet_info *vi = netdev_priv(dev);
1722         unsigned int start;
1723         int i;
1724 
1725         for (i = 0; i < vi->max_queue_pairs; i++) {
1726                 u64 tpackets, tbytes, rpackets, rbytes, rdrops;
1727                 struct receive_queue *rq = &vi->rq[i];
1728                 struct send_queue *sq = &vi->sq[i];
1729 
1730                 do {
1731                         start = u64_stats_fetch_begin_irq(&sq->stats.syncp);
1732                         tpackets = sq->stats.packets;
1733                         tbytes   = sq->stats.bytes;
1734                 } while (u64_stats_fetch_retry_irq(&sq->stats.syncp, start));
1735 
1736                 do {
1737                         start = u64_stats_fetch_begin_irq(&rq->stats.syncp);
1738                         rpackets = rq->stats.packets;
1739                         rbytes   = rq->stats.bytes;
1740                         rdrops   = rq->stats.drops;
1741                 } while (u64_stats_fetch_retry_irq(&rq->stats.syncp, start));
1742 
1743                 tot->rx_packets += rpackets;
1744                 tot->tx_packets += tpackets;
1745                 tot->rx_bytes   += rbytes;
1746                 tot->tx_bytes   += tbytes;
1747                 tot->rx_dropped += rdrops;
1748         }
1749 
1750         tot->tx_dropped = dev->stats.tx_dropped;
1751         tot->tx_fifo_errors = dev->stats.tx_fifo_errors;
1752         tot->rx_length_errors = dev->stats.rx_length_errors;
1753         tot->rx_frame_errors = dev->stats.rx_frame_errors;
1754 }
1755 
1756 static void virtnet_ack_link_announce(struct virtnet_info *vi)
1757 {
1758         rtnl_lock();
1759         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE,
1760                                   VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL))
1761                 dev_warn(&vi->dev->dev, "Failed to ack link announce.\n");
1762         rtnl_unlock();
1763 }
1764 
1765 static int _virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs)
1766 {
1767         struct scatterlist sg;
1768         struct net_device *dev = vi->dev;
1769 
1770         if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ))
1771                 return 0;
1772 
1773         vi->ctrl->mq.virtqueue_pairs = cpu_to_virtio16(vi->vdev, queue_pairs);
1774         sg_init_one(&sg, &vi->ctrl->mq, sizeof(vi->ctrl->mq));
1775 
1776         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ,
1777                                   VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg)) {
1778                 dev_warn(&dev->dev, "Fail to set num of queue pairs to %d\n",
1779                          queue_pairs);
1780                 return -EINVAL;
1781         } else {
1782                 vi->curr_queue_pairs = queue_pairs;
1783                 /* virtnet_open() will refill when device is going to up. */
1784                 if (dev->flags & IFF_UP)
1785                         schedule_delayed_work(&vi->refill, 0);
1786         }
1787 
1788         return 0;
1789 }
1790 
1791 static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs)
1792 {
1793         int err;
1794 
1795         rtnl_lock();
1796         err = _virtnet_set_queues(vi, queue_pairs);
1797         rtnl_unlock();
1798         return err;
1799 }
1800 
1801 static int virtnet_close(struct net_device *dev)
1802 {
1803         struct virtnet_info *vi = netdev_priv(dev);
1804         int i;
1805 
1806         /* Make sure refill_work doesn't re-enable napi! */
1807         cancel_delayed_work_sync(&vi->refill);
1808 
1809         for (i = 0; i < vi->max_queue_pairs; i++) {
1810                 xdp_rxq_info_unreg(&vi->rq[i].xdp_rxq);
1811                 napi_disable(&vi->rq[i].napi);
1812                 virtnet_napi_tx_disable(&vi->sq[i].napi);
1813         }
1814 
1815         return 0;
1816 }
1817 
1818 static void virtnet_set_rx_mode(struct net_device *dev)
1819 {
1820         struct virtnet_info *vi = netdev_priv(dev);
1821         struct scatterlist sg[2];
1822         struct virtio_net_ctrl_mac *mac_data;
1823         struct netdev_hw_addr *ha;
1824         int uc_count;
1825         int mc_count;
1826         void *buf;
1827         int i;
1828 
1829         /* We can't dynamically set ndo_set_rx_mode, so return gracefully */
1830         if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
1831                 return;
1832 
1833         vi->ctrl->promisc = ((dev->flags & IFF_PROMISC) != 0);
1834         vi->ctrl->allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
1835 
1836         sg_init_one(sg, &vi->ctrl->promisc, sizeof(vi->ctrl->promisc));
1837 
1838         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
1839                                   VIRTIO_NET_CTRL_RX_PROMISC, sg))
1840                 dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
1841                          vi->ctrl->promisc ? "en" : "dis");
1842 
1843         sg_init_one(sg, &vi->ctrl->allmulti, sizeof(vi->ctrl->allmulti));
1844 
1845         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
1846                                   VIRTIO_NET_CTRL_RX_ALLMULTI, sg))
1847                 dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
1848                          vi->ctrl->allmulti ? "en" : "dis");
1849 
1850         uc_count = netdev_uc_count(dev);
1851         mc_count = netdev_mc_count(dev);
1852         /* MAC filter - use one buffer for both lists */
1853         buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) +
1854                       (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
1855         mac_data = buf;
1856         if (!buf)
1857                 return;
1858 
1859         sg_init_table(sg, 2);
1860 
1861         /* Store the unicast list and count in the front of the buffer */
1862         mac_data->entries = cpu_to_virtio32(vi->vdev, uc_count);
1863         i = 0;
1864         netdev_for_each_uc_addr(ha, dev)
1865                 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
1866 
1867         sg_set_buf(&sg[0], mac_data,
1868                    sizeof(mac_data->entries) + (uc_count * ETH_ALEN));
1869 
1870         /* multicast list and count fill the end */
1871         mac_data = (void *)&mac_data->macs[uc_count][0];
1872 
1873         mac_data->entries = cpu_to_virtio32(vi->vdev, mc_count);
1874         i = 0;
1875         netdev_for_each_mc_addr(ha, dev)
1876                 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
1877 
1878         sg_set_buf(&sg[1], mac_data,
1879                    sizeof(mac_data->entries) + (mc_count * ETH_ALEN));
1880 
1881         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
1882                                   VIRTIO_NET_CTRL_MAC_TABLE_SET, sg))
1883                 dev_warn(&dev->dev, "Failed to set MAC filter table.\n");
1884 
1885         kfree(buf);
1886 }
1887 
1888 static int virtnet_vlan_rx_add_vid(struct net_device *dev,
1889                                    __be16 proto, u16 vid)
1890 {
1891         struct virtnet_info *vi = netdev_priv(dev);
1892         struct scatterlist sg;
1893 
1894         vi->ctrl->vid = cpu_to_virtio16(vi->vdev, vid);
1895         sg_init_one(&sg, &vi->ctrl->vid, sizeof(vi->ctrl->vid));
1896 
1897         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
1898                                   VIRTIO_NET_CTRL_VLAN_ADD, &sg))
1899                 dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
1900         return 0;
1901 }
1902 
1903 static int virtnet_vlan_rx_kill_vid(struct net_device *dev,
1904                                     __be16 proto, u16 vid)
1905 {
1906         struct virtnet_info *vi = netdev_priv(dev);
1907         struct scatterlist sg;
1908 
1909         vi->ctrl->vid = cpu_to_virtio16(vi->vdev, vid);
1910         sg_init_one(&sg, &vi->ctrl->vid, sizeof(vi->ctrl->vid));
1911 
1912         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
1913                                   VIRTIO_NET_CTRL_VLAN_DEL, &sg))
1914                 dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
1915         return 0;
1916 }
1917 
1918 static void virtnet_clean_affinity(struct virtnet_info *vi)
1919 {
1920         int i;
1921 
1922         if (vi->affinity_hint_set) {
1923                 for (i = 0; i < vi->max_queue_pairs; i++) {
1924                         virtqueue_set_affinity(vi->rq[i].vq, NULL);
1925                         virtqueue_set_affinity(vi->sq[i].vq, NULL);
1926                 }
1927 
1928                 vi->affinity_hint_set = false;
1929         }
1930 }
1931 
1932 static void virtnet_set_affinity(struct virtnet_info *vi)
1933 {
1934         cpumask_var_t mask;
1935         int stragglers;
1936         int group_size;
1937         int i, j, cpu;
1938         int num_cpu;
1939         int stride;
1940 
1941         if (!zalloc_cpumask_var(&mask, GFP_KERNEL)) {
1942                 virtnet_clean_affinity(vi);
1943                 return;
1944         }
1945 
1946         num_cpu = num_online_cpus();
1947         stride = max_t(int, num_cpu / vi->curr_queue_pairs, 1);
1948         stragglers = num_cpu >= vi->curr_queue_pairs ?
1949                         num_cpu % vi->curr_queue_pairs :
1950                         0;
1951         cpu = cpumask_next(-1, cpu_online_mask);
1952 
1953         for (i = 0; i < vi->curr_queue_pairs; i++) {
1954                 group_size = stride + (i < stragglers ? 1 : 0);
1955 
1956                 for (j = 0; j < group_size; j++) {
1957                         cpumask_set_cpu(cpu, mask);
1958                         cpu = cpumask_next_wrap(cpu, cpu_online_mask,
1959                                                 nr_cpu_ids, false);
1960                 }
1961                 virtqueue_set_affinity(vi->rq[i].vq, mask);
1962                 virtqueue_set_affinity(vi->sq[i].vq, mask);
1963                 __netif_set_xps_queue(vi->dev, cpumask_bits(mask), i, false);
1964                 cpumask_clear(mask);
1965         }
1966 
1967         vi->affinity_hint_set = true;
1968         free_cpumask_var(mask);
1969 }
1970 
1971 static int virtnet_cpu_online(unsigned int cpu, struct hlist_node *node)
1972 {
1973         struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info,
1974                                                    node);
1975         virtnet_set_affinity(vi);
1976         return 0;
1977 }
1978 
1979 static int virtnet_cpu_dead(unsigned int cpu, struct hlist_node *node)
1980 {
1981         struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info,
1982                                                    node_dead);
1983         virtnet_set_affinity(vi);
1984         return 0;
1985 }
1986 
1987 static int virtnet_cpu_down_prep(unsigned int cpu, struct hlist_node *node)
1988 {
1989         struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info,
1990                                                    node);
1991 
1992         virtnet_clean_affinity(vi);
1993         return 0;
1994 }
1995 
1996 static enum cpuhp_state virtionet_online;
1997 
1998 static int virtnet_cpu_notif_add(struct virtnet_info *vi)
1999 {
2000         int ret;
2001 
2002         ret = cpuhp_state_add_instance_nocalls(virtionet_online, &vi->node);
2003         if (ret)
2004                 return ret;
2005         ret = cpuhp_state_add_instance_nocalls(CPUHP_VIRT_NET_DEAD,
2006                                                &vi->node_dead);
2007         if (!ret)
2008                 return ret;
2009         cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node);
2010         return ret;
2011 }
2012 
2013 static void virtnet_cpu_notif_remove(struct virtnet_info *vi)
2014 {
2015         cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node);
2016         cpuhp_state_remove_instance_nocalls(CPUHP_VIRT_NET_DEAD,
2017                                             &vi->node_dead);
2018 }
2019 
2020 static void virtnet_get_ringparam(struct net_device *dev,
2021                                 struct ethtool_ringparam *ring)
2022 {
2023         struct virtnet_info *vi = netdev_priv(dev);
2024 
2025         ring->rx_max_pending = virtqueue_get_vring_size(vi->rq[0].vq);
2026         ring->tx_max_pending = virtqueue_get_vring_size(vi->sq[0].vq);
2027         ring->rx_pending = ring->rx_max_pending;
2028         ring->tx_pending = ring->tx_max_pending;
2029 }
2030 
2031 
2032 static void virtnet_get_drvinfo(struct net_device *dev,
2033                                 struct ethtool_drvinfo *info)
2034 {
2035         struct virtnet_info *vi = netdev_priv(dev);
2036         struct virtio_device *vdev = vi->vdev;
2037 
2038         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
2039         strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version));
2040         strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info));
2041 
2042 }
2043 
2044 /* TODO: Eliminate OOO packets during switching */
2045 static int virtnet_set_channels(struct net_device *dev,
2046                                 struct ethtool_channels *channels)
2047 {
2048         struct virtnet_info *vi = netdev_priv(dev);
2049         u16 queue_pairs = channels->combined_count;
2050         int err;
2051 
2052         /* We don't support separate rx/tx channels.
2053          * We don't allow setting 'other' channels.
2054          */
2055         if (channels->rx_count || channels->tx_count || channels->other_count)
2056                 return -EINVAL;
2057 
2058         if (queue_pairs > vi->max_queue_pairs || queue_pairs == 0)
2059                 return -EINVAL;
2060 
2061         /* For now we don't support modifying channels while XDP is loaded
2062          * also when XDP is loaded all RX queues have XDP programs so we only
2063          * need to check a single RX queue.
2064          */
2065         if (vi->rq[0].xdp_prog)
2066                 return -EINVAL;
2067 
2068         get_online_cpus();
2069         err = _virtnet_set_queues(vi, queue_pairs);
2070         if (!err) {
2071                 netif_set_real_num_tx_queues(dev, queue_pairs);
2072                 netif_set_real_num_rx_queues(dev, queue_pairs);
2073 
2074                 virtnet_set_affinity(vi);
2075         }
2076         put_online_cpus();
2077 
2078         return err;
2079 }
2080 
2081 static void virtnet_get_strings(struct net_device *dev, u32 stringset, u8 *data)
2082 {
2083         struct virtnet_info *vi = netdev_priv(dev);
2084         char *p = (char *)data;
2085         unsigned int i, j;
2086 
2087         switch (stringset) {
2088         case ETH_SS_STATS:
2089                 for (i = 0; i < vi->curr_queue_pairs; i++) {
2090                         for (j = 0; j < VIRTNET_RQ_STATS_LEN; j++) {
2091                                 snprintf(p, ETH_GSTRING_LEN, "rx_queue_%u_%s",
2092                                          i, virtnet_rq_stats_desc[j].desc);
2093                                 p += ETH_GSTRING_LEN;
2094                         }
2095                 }
2096 
2097                 for (i = 0; i < vi->curr_queue_pairs; i++) {
2098                         for (j = 0; j < VIRTNET_SQ_STATS_LEN; j++) {
2099                                 snprintf(p, ETH_GSTRING_LEN, "tx_queue_%u_%s",
2100                                          i, virtnet_sq_stats_desc[j].desc);
2101                                 p += ETH_GSTRING_LEN;
2102                         }
2103                 }
2104                 break;
2105         }
2106 }
2107 
2108 static int virtnet_get_sset_count(struct net_device *dev, int sset)
2109 {
2110         struct virtnet_info *vi = netdev_priv(dev);
2111 
2112         switch (sset) {
2113         case ETH_SS_STATS:
2114                 return vi->curr_queue_pairs * (VIRTNET_RQ_STATS_LEN +
2115                                                VIRTNET_SQ_STATS_LEN);
2116         default:
2117                 return -EOPNOTSUPP;
2118         }
2119 }
2120 
2121 static void virtnet_get_ethtool_stats(struct net_device *dev,
2122                                       struct ethtool_stats *stats, u64 *data)
2123 {
2124         struct virtnet_info *vi = netdev_priv(dev);
2125         unsigned int idx = 0, start, i, j;
2126         const u8 *stats_base;
2127         size_t offset;
2128 
2129         for (i = 0; i < vi->curr_queue_pairs; i++) {
2130                 struct receive_queue *rq = &vi->rq[i];
2131 
2132                 stats_base = (u8 *)&rq->stats;
2133                 do {
2134                         start = u64_stats_fetch_begin_irq(&rq->stats.syncp);
2135                         for (j = 0; j < VIRTNET_RQ_STATS_LEN; j++) {
2136                                 offset = virtnet_rq_stats_desc[j].offset;
2137                                 data[idx + j] = *(u64 *)(stats_base + offset);
2138                         }
2139                 } while (u64_stats_fetch_retry_irq(&rq->stats.syncp, start));
2140                 idx += VIRTNET_RQ_STATS_LEN;
2141         }
2142 
2143         for (i = 0; i < vi->curr_queue_pairs; i++) {
2144                 struct send_queue *sq = &vi->sq[i];
2145 
2146                 stats_base = (u8 *)&sq->stats;
2147                 do {
2148                         start = u64_stats_fetch_begin_irq(&sq->stats.syncp);
2149                         for (j = 0; j < VIRTNET_SQ_STATS_LEN; j++) {
2150                                 offset = virtnet_sq_stats_desc[j].offset;
2151                                 data[idx + j] = *(u64 *)(stats_base + offset);
2152                         }
2153                 } while (u64_stats_fetch_retry_irq(&sq->stats.syncp, start));
2154                 idx += VIRTNET_SQ_STATS_LEN;
2155         }
2156 }
2157 
2158 static void virtnet_get_channels(struct net_device *dev,
2159                                  struct ethtool_channels *channels)
2160 {
2161         struct virtnet_info *vi = netdev_priv(dev);
2162 
2163         channels->combined_count = vi->curr_queue_pairs;
2164         channels->max_combined = vi->max_queue_pairs;
2165         channels->max_other = 0;
2166         channels->rx_count = 0;
2167         channels->tx_count = 0;
2168         channels->other_count = 0;
2169 }
2170 
2171 /* Check if the user is trying to change anything besides speed/duplex */
2172 static bool
2173 virtnet_validate_ethtool_cmd(const struct ethtool_link_ksettings *cmd)
2174 {
2175         struct ethtool_link_ksettings diff1 = *cmd;
2176         struct ethtool_link_ksettings diff2 = {};
2177 
2178         /* cmd is always set so we need to clear it, validate the port type
2179          * and also without autonegotiation we can ignore advertising
2180          */
2181         diff1.base.speed = 0;
2182         diff2.base.port = PORT_OTHER;
2183         ethtool_link_ksettings_zero_link_mode(&diff1, advertising);
2184         diff1.base.duplex = 0;
2185         diff1.base.cmd = 0;
2186         diff1.base.link_mode_masks_nwords = 0;
2187 
2188         return !memcmp(&diff1.base, &diff2.base, sizeof(diff1.base)) &&
2189                 bitmap_empty(diff1.link_modes.supported,
2190                              __ETHTOOL_LINK_MODE_MASK_NBITS) &&
2191                 bitmap_empty(diff1.link_modes.advertising,
2192                              __ETHTOOL_LINK_MODE_MASK_NBITS) &&
2193                 bitmap_empty(diff1.link_modes.lp_advertising,
2194                              __ETHTOOL_LINK_MODE_MASK_NBITS);
2195 }
2196 
2197 static int virtnet_set_link_ksettings(struct net_device *dev,
2198                                       const struct ethtool_link_ksettings *cmd)
2199 {
2200         struct virtnet_info *vi = netdev_priv(dev);
2201         u32 speed;
2202 
2203         speed = cmd->base.speed;
2204         /* don't allow custom speed and duplex */
2205         if (!ethtool_validate_speed(speed) ||
2206             !ethtool_validate_duplex(cmd->base.duplex) ||
2207             !virtnet_validate_ethtool_cmd(cmd))
2208                 return -EINVAL;
2209         vi->speed = speed;
2210         vi->duplex = cmd->base.duplex;
2211 
2212         return 0;
2213 }
2214 
2215 static int virtnet_get_link_ksettings(struct net_device *dev,
2216                                       struct ethtool_link_ksettings *cmd)
2217 {
2218         struct virtnet_info *vi = netdev_priv(dev);
2219 
2220         cmd->base.speed = vi->speed;
2221         cmd->base.duplex = vi->duplex;
2222         cmd->base.port = PORT_OTHER;
2223 
2224         return 0;
2225 }
2226 
2227 static int virtnet_set_coalesce(struct net_device *dev,
2228                                 struct ethtool_coalesce *ec)
2229 {
2230         struct ethtool_coalesce ec_default = {
2231                 .cmd = ETHTOOL_SCOALESCE,
2232                 .rx_max_coalesced_frames = 1,
2233         };
2234         struct virtnet_info *vi = netdev_priv(dev);
2235         int i, napi_weight;
2236 
2237         if (ec->tx_max_coalesced_frames > 1)
2238                 return -EINVAL;
2239 
2240         ec_default.tx_max_coalesced_frames = ec->tx_max_coalesced_frames;
2241         napi_weight = ec->tx_max_coalesced_frames ? NAPI_POLL_WEIGHT : 0;
2242 
2243         /* disallow changes to fields not explicitly tested above */
2244         if (memcmp(ec, &ec_default, sizeof(ec_default)))
2245                 return -EINVAL;
2246 
2247         if (napi_weight ^ vi->sq[0].napi.weight) {
2248                 if (dev->flags & IFF_UP)
2249                         return -EBUSY;
2250                 for (i = 0; i < vi->max_queue_pairs; i++)
2251                         vi->sq[i].napi.weight = napi_weight;
2252         }
2253 
2254         return 0;
2255 }
2256 
2257 static int virtnet_get_coalesce(struct net_device *dev,
2258                                 struct ethtool_coalesce *ec)
2259 {
2260         struct ethtool_coalesce ec_default = {
2261                 .cmd = ETHTOOL_GCOALESCE,
2262                 .rx_max_coalesced_frames = 1,
2263         };
2264         struct virtnet_info *vi = netdev_priv(dev);
2265 
2266         memcpy(ec, &ec_default, sizeof(ec_default));
2267 
2268         if (vi->sq[0].napi.weight)
2269                 ec->tx_max_coalesced_frames = 1;
2270 
2271         return 0;
2272 }
2273 
2274 static void virtnet_init_settings(struct net_device *dev)
2275 {
2276         struct virtnet_info *vi = netdev_priv(dev);
2277 
2278         vi->speed = SPEED_UNKNOWN;
2279         vi->duplex = DUPLEX_UNKNOWN;
2280 }
2281 
2282 static void virtnet_update_settings(struct virtnet_info *vi)
2283 {
2284         u32 speed;
2285         u8 duplex;
2286 
2287         if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_SPEED_DUPLEX))
2288                 return;
2289 
2290         speed = virtio_cread32(vi->vdev, offsetof(struct virtio_net_config,
2291                                                   speed));
2292         if (ethtool_validate_speed(speed))
2293                 vi->speed = speed;
2294         duplex = virtio_cread8(vi->vdev, offsetof(struct virtio_net_config,
2295                                                   duplex));
2296         if (ethtool_validate_duplex(duplex))
2297                 vi->duplex = duplex;
2298 }
2299 
2300 static const struct ethtool_ops virtnet_ethtool_ops = {
2301         .get_drvinfo = virtnet_get_drvinfo,
2302         .get_link = ethtool_op_get_link,
2303         .get_ringparam = virtnet_get_ringparam,
2304         .get_strings = virtnet_get_strings,
2305         .get_sset_count = virtnet_get_sset_count,
2306         .get_ethtool_stats = virtnet_get_ethtool_stats,
2307         .set_channels = virtnet_set_channels,
2308         .get_channels = virtnet_get_channels,
2309         .get_ts_info = ethtool_op_get_ts_info,
2310         .get_link_ksettings = virtnet_get_link_ksettings,
2311         .set_link_ksettings = virtnet_set_link_ksettings,
2312         .set_coalesce = virtnet_set_coalesce,
2313         .get_coalesce = virtnet_get_coalesce,
2314 };
2315 
2316 static void virtnet_freeze_down(struct virtio_device *vdev)
2317 {
2318         struct virtnet_info *vi = vdev->priv;
2319         int i;
2320 
2321         /* Make sure no work handler is accessing the device */
2322         flush_work(&vi->config_work);
2323 
2324         netif_tx_lock_bh(vi->dev);
2325         netif_device_detach(vi->dev);
2326         netif_tx_unlock_bh(vi->dev);
2327         cancel_delayed_work_sync(&vi->refill);
2328 
2329         if (netif_running(vi->dev)) {
2330                 for (i = 0; i < vi->max_queue_pairs; i++) {
2331                         napi_disable(&vi->rq[i].napi);
2332                         virtnet_napi_tx_disable(&vi->sq[i].napi);
2333                 }
2334         }
2335 }
2336 
2337 static int init_vqs(struct virtnet_info *vi);
2338 
2339 static int virtnet_restore_up(struct virtio_device *vdev)
2340 {
2341         struct virtnet_info *vi = vdev->priv;
2342         int err, i;
2343 
2344         err = init_vqs(vi);
2345         if (err)
2346                 return err;
2347 
2348         virtio_device_ready(vdev);
2349 
2350         if (netif_running(vi->dev)) {
2351                 for (i = 0; i < vi->curr_queue_pairs; i++)
2352                         if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL))
2353                                 schedule_delayed_work(&vi->refill, 0);
2354 
2355                 for (i = 0; i < vi->max_queue_pairs; i++) {
2356                         virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi);
2357                         virtnet_napi_tx_enable(vi, vi->sq[i].vq,
2358                                                &vi->sq[i].napi);
2359                 }
2360         }
2361 
2362         netif_tx_lock_bh(vi->dev);
2363         netif_device_attach(vi->dev);
2364         netif_tx_unlock_bh(vi->dev);
2365         return err;
2366 }
2367 
2368 static int virtnet_set_guest_offloads(struct virtnet_info *vi, u64 offloads)
2369 {
2370         struct scatterlist sg;
2371         vi->ctrl->offloads = cpu_to_virtio64(vi->vdev, offloads);
2372 
2373         sg_init_one(&sg, &vi->ctrl->offloads, sizeof(vi->ctrl->offloads));
2374 
2375         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_GUEST_OFFLOADS,
2376                                   VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET, &sg)) {
2377                 dev_warn(&vi->dev->dev, "Fail to set guest offload.\n");
2378                 return -EINVAL;
2379         }
2380 
2381         return 0;
2382 }
2383 
2384 static int virtnet_clear_guest_offloads(struct virtnet_info *vi)
2385 {
2386         u64 offloads = 0;
2387 
2388         if (!vi->guest_offloads)
2389                 return 0;
2390 
2391         return virtnet_set_guest_offloads(vi, offloads);
2392 }
2393 
2394 static int virtnet_restore_guest_offloads(struct virtnet_info *vi)
2395 {
2396         u64 offloads = vi->guest_offloads;
2397 
2398         if (!vi->guest_offloads)
2399                 return 0;
2400 
2401         return virtnet_set_guest_offloads(vi, offloads);
2402 }
2403 
2404 static int virtnet_xdp_set(struct net_device *dev, struct bpf_prog *prog,
2405                            struct netlink_ext_ack *extack)
2406 {
2407         unsigned long int max_sz = PAGE_SIZE - sizeof(struct padded_vnet_hdr);
2408         struct virtnet_info *vi = netdev_priv(dev);
2409         struct bpf_prog *old_prog;
2410         u16 xdp_qp = 0, curr_qp;
2411         int i, err;
2412 
2413         if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)
2414             && (virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO4) ||
2415                 virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO6) ||
2416                 virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_ECN) ||
2417                 virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_UFO) ||
2418                 virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_CSUM))) {
2419                 NL_SET_ERR_MSG_MOD(extack, "Can't set XDP while host is implementing LRO/CSUM, disable LRO/CSUM first");
2420                 return -EOPNOTSUPP;
2421         }
2422 
2423         if (vi->mergeable_rx_bufs && !vi->any_header_sg) {
2424                 NL_SET_ERR_MSG_MOD(extack, "XDP expects header/data in single page, any_header_sg required");
2425                 return -EINVAL;
2426         }
2427 
2428         if (dev->mtu > max_sz) {
2429                 NL_SET_ERR_MSG_MOD(extack, "MTU too large to enable XDP");
2430                 netdev_warn(dev, "XDP requires MTU less than %lu\n", max_sz);
2431                 return -EINVAL;
2432         }
2433 
2434         curr_qp = vi->curr_queue_pairs - vi->xdp_queue_pairs;
2435         if (prog)
2436                 xdp_qp = nr_cpu_ids;
2437 
2438         /* XDP requires extra queues for XDP_TX */
2439         if (curr_qp + xdp_qp > vi->max_queue_pairs) {
2440                 NL_SET_ERR_MSG_MOD(extack, "Too few free TX rings available");
2441                 netdev_warn(dev, "request %i queues but max is %i\n",
2442                             curr_qp + xdp_qp, vi->max_queue_pairs);
2443                 return -ENOMEM;
2444         }
2445 
2446         old_prog = rtnl_dereference(vi->rq[0].xdp_prog);
2447         if (!prog && !old_prog)
2448                 return 0;
2449 
2450         if (prog) {
2451                 prog = bpf_prog_add(prog, vi->max_queue_pairs - 1);
2452                 if (IS_ERR(prog))
2453                         return PTR_ERR(prog);
2454         }
2455 
2456         /* Make sure NAPI is not using any XDP TX queues for RX. */
2457         if (netif_running(dev)) {
2458                 for (i = 0; i < vi->max_queue_pairs; i++) {
2459                         napi_disable(&vi->rq[i].napi);
2460                         virtnet_napi_tx_disable(&vi->sq[i].napi);
2461                 }
2462         }
2463 
2464         if (!prog) {
2465                 for (i = 0; i < vi->max_queue_pairs; i++) {
2466                         rcu_assign_pointer(vi->rq[i].xdp_prog, prog);
2467                         if (i == 0)
2468                                 virtnet_restore_guest_offloads(vi);
2469                 }
2470                 synchronize_net();
2471         }
2472 
2473         err = _virtnet_set_queues(vi, curr_qp + xdp_qp);
2474         if (err)
2475                 goto err;
2476         netif_set_real_num_rx_queues(dev, curr_qp + xdp_qp);
2477         vi->xdp_queue_pairs = xdp_qp;
2478 
2479         if (prog) {
2480                 for (i = 0; i < vi->max_queue_pairs; i++) {
2481                         rcu_assign_pointer(vi->rq[i].xdp_prog, prog);
2482                         if (i == 0 && !old_prog)
2483                                 virtnet_clear_guest_offloads(vi);
2484                 }
2485         }
2486 
2487         for (i = 0; i < vi->max_queue_pairs; i++) {
2488                 if (old_prog)
2489                         bpf_prog_put(old_prog);
2490                 if (netif_running(dev)) {
2491                         virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi);
2492                         virtnet_napi_tx_enable(vi, vi->sq[i].vq,
2493                                                &vi->sq[i].napi);
2494                 }
2495         }
2496 
2497         return 0;
2498 
2499 err:
2500         if (!prog) {
2501                 virtnet_clear_guest_offloads(vi);
2502                 for (i = 0; i < vi->max_queue_pairs; i++)
2503                         rcu_assign_pointer(vi->rq[i].xdp_prog, old_prog);
2504         }
2505 
2506         if (netif_running(dev)) {
2507                 for (i = 0; i < vi->max_queue_pairs; i++) {
2508                         virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi);
2509                         virtnet_napi_tx_enable(vi, vi->sq[i].vq,
2510                                                &vi->sq[i].napi);
2511                 }
2512         }
2513         if (prog)
2514                 bpf_prog_sub(prog, vi->max_queue_pairs - 1);
2515         return err;
2516 }
2517 
2518 static u32 virtnet_xdp_query(struct net_device *dev)
2519 {
2520         struct virtnet_info *vi = netdev_priv(dev);
2521         const struct bpf_prog *xdp_prog;
2522         int i;
2523 
2524         for (i = 0; i < vi->max_queue_pairs; i++) {
2525                 xdp_prog = rtnl_dereference(vi->rq[i].xdp_prog);
2526                 if (xdp_prog)
2527                         return xdp_prog->aux->id;
2528         }
2529         return 0;
2530 }
2531 
2532 static int virtnet_xdp(struct net_device *dev, struct netdev_bpf *xdp)
2533 {
2534         switch (xdp->command) {
2535         case XDP_SETUP_PROG:
2536                 return virtnet_xdp_set(dev, xdp->prog, xdp->extack);
2537         case XDP_QUERY_PROG:
2538                 xdp->prog_id = virtnet_xdp_query(dev);
2539                 return 0;
2540         default:
2541                 return -EINVAL;
2542         }
2543 }
2544 
2545 static int virtnet_get_phys_port_name(struct net_device *dev, char *buf,
2546                                       size_t len)
2547 {
2548         struct virtnet_info *vi = netdev_priv(dev);
2549         int ret;
2550 
2551         if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_STANDBY))
2552                 return -EOPNOTSUPP;
2553 
2554         ret = snprintf(buf, len, "sby");
2555         if (ret >= len)
2556                 return -EOPNOTSUPP;
2557 
2558         return 0;
2559 }
2560 
2561 static int virtnet_set_features(struct net_device *dev,
2562                                 netdev_features_t features)
2563 {
2564         struct virtnet_info *vi = netdev_priv(dev);
2565         u64 offloads;
2566         int err;
2567 
2568         if ((dev->features ^ features) & NETIF_F_LRO) {
2569                 if (vi->xdp_queue_pairs)
2570                         return -EBUSY;
2571 
2572                 if (features & NETIF_F_LRO)
2573                         offloads = vi->guest_offloads_capable;
2574                 else
2575                         offloads = 0;
2576 
2577                 err = virtnet_set_guest_offloads(vi, offloads);
2578                 if (err)
2579                         return err;
2580                 vi->guest_offloads = offloads;
2581         }
2582 
2583         return 0;
2584 }
2585 
2586 static const struct net_device_ops virtnet_netdev = {
2587         .ndo_open            = virtnet_open,
2588         .ndo_stop            = virtnet_close,
2589         .ndo_start_xmit      = start_xmit,
2590         .ndo_validate_addr   = eth_validate_addr,
2591         .ndo_set_mac_address = virtnet_set_mac_address,
2592         .ndo_set_rx_mode     = virtnet_set_rx_mode,
2593         .ndo_get_stats64     = virtnet_stats,
2594         .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
2595         .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
2596         .ndo_bpf                = virtnet_xdp,
2597         .ndo_xdp_xmit           = virtnet_xdp_xmit,
2598         .ndo_features_check     = passthru_features_check,
2599         .ndo_get_phys_port_name = virtnet_get_phys_port_name,
2600         .ndo_set_features       = virtnet_set_features,
2601 };
2602 
2603 static void virtnet_config_changed_work(struct work_struct *work)
2604 {
2605         struct virtnet_info *vi =
2606                 container_of(work, struct virtnet_info, config_work);
2607         u16 v;
2608 
2609         if (virtio_cread_feature(vi->vdev, VIRTIO_NET_F_STATUS,
2610                                  struct virtio_net_config, status, &v) < 0)
2611                 return;
2612 
2613         if (v & VIRTIO_NET_S_ANNOUNCE) {
2614                 netdev_notify_peers(vi->dev);
2615                 virtnet_ack_link_announce(vi);
2616         }
2617 
2618         /* Ignore unknown (future) status bits */
2619         v &= VIRTIO_NET_S_LINK_UP;
2620 
2621         if (vi->status == v)
2622                 return;
2623 
2624         vi->status = v;
2625 
2626         if (vi->status & VIRTIO_NET_S_LINK_UP) {
2627                 virtnet_update_settings(vi);
2628                 netif_carrier_on(vi->dev);
2629                 netif_tx_wake_all_queues(vi->dev);
2630         } else {
2631                 netif_carrier_off(vi->dev);
2632                 netif_tx_stop_all_queues(vi->dev);
2633         }
2634 }
2635 
2636 static void virtnet_config_changed(struct virtio_device *vdev)
2637 {
2638         struct virtnet_info *vi = vdev->priv;
2639 
2640         schedule_work(&vi->config_work);
2641 }
2642 
2643 static void virtnet_free_queues(struct virtnet_info *vi)
2644 {
2645         int i;
2646 
2647         for (i = 0; i < vi->max_queue_pairs; i++) {
2648                 napi_hash_del(&vi->rq[i].napi);
2649                 netif_napi_del(&vi->rq[i].napi);
2650                 netif_napi_del(&vi->sq[i].napi);
2651         }
2652 
2653         /* We called napi_hash_del() before netif_napi_del(),
2654          * we need to respect an RCU grace period before freeing vi->rq
2655          */
2656         synchronize_net();
2657 
2658         kfree(vi->rq);
2659         kfree(vi->sq);
2660         kfree(vi->ctrl);
2661 }
2662 
2663 static void _free_receive_bufs(struct virtnet_info *vi)
2664 {
2665         struct bpf_prog *old_prog;
2666         int i;
2667 
2668         for (i = 0; i < vi->max_queue_pairs; i++) {
2669                 while (vi->rq[i].pages)
2670                         __free_pages(get_a_page(&vi->rq[i], GFP_KERNEL), 0);
2671 
2672                 old_prog = rtnl_dereference(vi->rq[i].xdp_prog);
2673                 RCU_INIT_POINTER(vi->rq[i].xdp_prog, NULL);
2674                 if (old_prog)
2675                         bpf_prog_put(old_prog);
2676         }
2677 }
2678 
2679 static void free_receive_bufs(struct virtnet_info *vi)
2680 {
2681         rtnl_lock();
2682         _free_receive_bufs(vi);
2683         rtnl_unlock();
2684 }
2685 
2686 static void free_receive_page_frags(struct virtnet_info *vi)
2687 {
2688         int i;
2689         for (i = 0; i < vi->max_queue_pairs; i++)
2690                 if (vi->rq[i].alloc_frag.page)
2691                         put_page(vi->rq[i].alloc_frag.page);
2692 }
2693 
2694 static void free_unused_bufs(struct virtnet_info *vi)
2695 {
2696         void *buf;
2697         int i;
2698 
2699         for (i = 0; i < vi->max_queue_pairs; i++) {
2700                 struct virtqueue *vq = vi->sq[i].vq;
2701                 while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) {
2702                         if (!is_xdp_frame(buf))
2703                                 dev_kfree_skb(buf);
2704                         else
2705                                 xdp_return_frame(ptr_to_xdp(buf));
2706                 }
2707         }
2708 
2709         for (i = 0; i < vi->max_queue_pairs; i++) {
2710                 struct virtqueue *vq = vi->rq[i].vq;
2711 
2712                 while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) {
2713                         if (vi->mergeable_rx_bufs) {
2714                                 put_page(virt_to_head_page(buf));
2715                         } else if (vi->big_packets) {
2716                                 give_pages(&vi->rq[i], buf);
2717                         } else {
2718                                 put_page(virt_to_head_page(buf));
2719                         }
2720                 }
2721         }
2722 }
2723 
2724 static void virtnet_del_vqs(struct virtnet_info *vi)
2725 {
2726         struct virtio_device *vdev = vi->vdev;
2727 
2728         virtnet_clean_affinity(vi);
2729 
2730         vdev->config->del_vqs(vdev);
2731 
2732         virtnet_free_queues(vi);
2733 }
2734 
2735 /* How large should a single buffer be so a queue full of these can fit at
2736  * least one full packet?
2737  * Logic below assumes the mergeable buffer header is used.
2738  */
2739 static unsigned int mergeable_min_buf_len(struct virtnet_info *vi, struct virtqueue *vq)
2740 {
2741         const unsigned int hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
2742         unsigned int rq_size = virtqueue_get_vring_size(vq);
2743         unsigned int packet_len = vi->big_packets ? IP_MAX_MTU : vi->dev->max_mtu;
2744         unsigned int buf_len = hdr_len + ETH_HLEN + VLAN_HLEN + packet_len;
2745         unsigned int min_buf_len = DIV_ROUND_UP(buf_len, rq_size);
2746 
2747         return max(max(min_buf_len, hdr_len) - hdr_len,
2748                    (unsigned int)GOOD_PACKET_LEN);
2749 }
2750 
2751 static int virtnet_find_vqs(struct virtnet_info *vi)
2752 {
2753         vq_callback_t **callbacks;
2754         struct virtqueue **vqs;
2755         int ret = -ENOMEM;
2756         int i, total_vqs;
2757         const char **names;
2758         bool *ctx;
2759 
2760         /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by
2761          * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by
2762          * possible control vq.
2763          */
2764         total_vqs = vi->max_queue_pairs * 2 +
2765                     virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ);
2766 
2767         /* Allocate space for find_vqs parameters */
2768         vqs = kcalloc(total_vqs, sizeof(*vqs), GFP_KERNEL);
2769         if (!vqs)
2770                 goto err_vq;
2771         callbacks = kmalloc_array(total_vqs, sizeof(*callbacks), GFP_KERNEL);
2772         if (!callbacks)
2773                 goto err_callback;
2774         names = kmalloc_array(total_vqs, sizeof(*names), GFP_KERNEL);
2775         if (!names)
2776                 goto err_names;
2777         if (!vi->big_packets || vi->mergeable_rx_bufs) {
2778                 ctx = kcalloc(total_vqs, sizeof(*ctx), GFP_KERNEL);
2779                 if (!ctx)
2780                         goto err_ctx;
2781         } else {
2782                 ctx = NULL;
2783         }
2784 
2785         /* Parameters for control virtqueue, if any */
2786         if (vi->has_cvq) {
2787                 callbacks[total_vqs - 1] = NULL;
2788                 names[total_vqs - 1] = "control";
2789         }
2790 
2791         /* Allocate/initialize parameters for send/receive virtqueues */
2792         for (i = 0; i < vi->max_queue_pairs; i++) {
2793                 callbacks[rxq2vq(i)] = skb_recv_done;
2794                 callbacks[txq2vq(i)] = skb_xmit_done;
2795                 sprintf(vi->rq[i].name, "input.%d", i);
2796                 sprintf(vi->sq[i].name, "output.%d", i);
2797                 names[rxq2vq(i)] = vi->rq[i].name;
2798                 names[txq2vq(i)] = vi->sq[i].name;
2799                 if (ctx)
2800                         ctx[rxq2vq(i)] = true;
2801         }
2802 
2803         ret = vi->vdev->config->find_vqs(vi->vdev, total_vqs, vqs, callbacks,
2804                                          names, ctx, NULL);
2805         if (ret)
2806                 goto err_find;
2807 
2808         if (vi->has_cvq) {
2809                 vi->cvq = vqs[total_vqs - 1];
2810                 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
2811                         vi->dev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
2812         }
2813 
2814         for (i = 0; i < vi->max_queue_pairs; i++) {
2815                 vi->rq[i].vq = vqs[rxq2vq(i)];
2816                 vi->rq[i].min_buf_len = mergeable_min_buf_len(vi, vi->rq[i].vq);
2817                 vi->sq[i].vq = vqs[txq2vq(i)];
2818         }
2819 
2820         /* run here: ret == 0. */
2821 
2822 
2823 err_find:
2824         kfree(ctx);
2825 err_ctx:
2826         kfree(names);
2827 err_names:
2828         kfree(callbacks);
2829 err_callback:
2830         kfree(vqs);
2831 err_vq:
2832         return ret;
2833 }
2834 
2835 static int virtnet_alloc_queues(struct virtnet_info *vi)
2836 {
2837         int i;
2838 
2839         vi->ctrl = kzalloc(sizeof(*vi->ctrl), GFP_KERNEL);
2840         if (!vi->ctrl)
2841                 goto err_ctrl;
2842         vi->sq = kcalloc(vi->max_queue_pairs, sizeof(*vi->sq), GFP_KERNEL);
2843         if (!vi->sq)
2844                 goto err_sq;
2845         vi->rq = kcalloc(vi->max_queue_pairs, sizeof(*vi->rq), GFP_KERNEL);
2846         if (!vi->rq)
2847                 goto err_rq;
2848 
2849         INIT_DELAYED_WORK(&vi->refill, refill_work);
2850         for (i = 0; i < vi->max_queue_pairs; i++) {
2851                 vi->rq[i].pages = NULL;
2852                 netif_napi_add(vi->dev, &vi->rq[i].napi, virtnet_poll,
2853                                napi_weight);
2854                 netif_tx_napi_add(vi->dev, &vi->sq[i].napi, virtnet_poll_tx,
2855                                   napi_tx ? napi_weight : 0);
2856 
2857                 sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg));
2858                 ewma_pkt_len_init(&vi->rq[i].mrg_avg_pkt_len);
2859                 sg_init_table(vi->sq[i].sg, ARRAY_SIZE(vi->sq[i].sg));
2860 
2861                 u64_stats_init(&vi->rq[i].stats.syncp);
2862                 u64_stats_init(&vi->sq[i].stats.syncp);
2863         }
2864 
2865         return 0;
2866 
2867 err_rq:
2868         kfree(vi->sq);
2869 err_sq:
2870         kfree(vi->ctrl);
2871 err_ctrl:
2872         return -ENOMEM;
2873 }
2874 
2875 static int init_vqs(struct virtnet_info *vi)
2876 {
2877         int ret;
2878 
2879         /* Allocate send & receive queues */
2880         ret = virtnet_alloc_queues(vi);
2881         if (ret)
2882                 goto err;
2883 
2884         ret = virtnet_find_vqs(vi);
2885         if (ret)
2886                 goto err_free;
2887 
2888         get_online_cpus();
2889         virtnet_set_affinity(vi);
2890         put_online_cpus();
2891 
2892         return 0;
2893 
2894 err_free:
2895         virtnet_free_queues(vi);
2896 err:
2897         return ret;
2898 }
2899 
2900 #ifdef CONFIG_SYSFS
2901 static ssize_t mergeable_rx_buffer_size_show(struct netdev_rx_queue *queue,
2902                 char *buf)
2903 {
2904         struct virtnet_info *vi = netdev_priv(queue->dev);
2905         unsigned int queue_index = get_netdev_rx_queue_index(queue);
2906         unsigned int headroom = virtnet_get_headroom(vi);
2907         unsigned int tailroom = headroom ? sizeof(struct skb_shared_info) : 0;
2908         struct ewma_pkt_len *avg;
2909 
2910         BUG_ON(queue_index >= vi->max_queue_pairs);
2911         avg = &vi->rq[queue_index].mrg_avg_pkt_len;
2912         return sprintf(buf, "%u\n",
2913                        get_mergeable_buf_len(&vi->rq[queue_index], avg,
2914                                        SKB_DATA_ALIGN(headroom + tailroom)));
2915 }
2916 
2917 static struct rx_queue_attribute mergeable_rx_buffer_size_attribute =
2918         __ATTR_RO(mergeable_rx_buffer_size);
2919 
2920 static struct attribute *virtio_net_mrg_rx_attrs[] = {
2921         &mergeable_rx_buffer_size_attribute.attr,
2922         NULL
2923 };
2924 
2925 static const struct attribute_group virtio_net_mrg_rx_group = {
2926         .name = "virtio_net",
2927         .attrs = virtio_net_mrg_rx_attrs
2928 };
2929 #endif
2930 
2931 static bool virtnet_fail_on_feature(struct virtio_device *vdev,
2932                                     unsigned int fbit,
2933                                     const char *fname, const char *dname)
2934 {
2935         if (!virtio_has_feature(vdev, fbit))
2936                 return false;
2937 
2938         dev_err(&vdev->dev, "device advertises feature %s but not %s",
2939                 fname, dname);
2940 
2941         return true;
2942 }
2943 
2944 #define VIRTNET_FAIL_ON(vdev, fbit, dbit)                       \
2945         virtnet_fail_on_feature(vdev, fbit, #fbit, dbit)
2946 
2947 static bool virtnet_validate_features(struct virtio_device *vdev)
2948 {
2949         if (!virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ) &&
2950             (VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_RX,
2951                              "VIRTIO_NET_F_CTRL_VQ") ||
2952              VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_VLAN,
2953                              "VIRTIO_NET_F_CTRL_VQ") ||
2954              VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE,
2955                              "VIRTIO_NET_F_CTRL_VQ") ||
2956              VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_MQ, "VIRTIO_NET_F_CTRL_VQ") ||
2957              VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR,
2958                              "VIRTIO_NET_F_CTRL_VQ"))) {
2959                 return false;
2960         }
2961 
2962         return true;
2963 }
2964 
2965 #define MIN_MTU ETH_MIN_MTU
2966 #define MAX_MTU ETH_MAX_MTU
2967 
2968 static int virtnet_validate(struct virtio_device *vdev)
2969 {
2970         if (!vdev->config->get) {
2971                 dev_err(&vdev->dev, "%s failure: config access disabled\n",
2972                         __func__);
2973                 return -EINVAL;
2974         }
2975 
2976         if (!virtnet_validate_features(vdev))
2977                 return -EINVAL;
2978 
2979         if (virtio_has_feature(vdev, VIRTIO_NET_F_MTU)) {
2980                 int mtu = virtio_cread16(vdev,
2981                                          offsetof(struct virtio_net_config,
2982                                                   mtu));
2983                 if (mtu < MIN_MTU)
2984                         __virtio_clear_bit(vdev, VIRTIO_NET_F_MTU);
2985         }
2986 
2987         return 0;
2988 }
2989 
2990 static int virtnet_probe(struct virtio_device *vdev)
2991 {
2992         int i, err = -ENOMEM;
2993         struct net_device *dev;
2994         struct virtnet_info *vi;
2995         u16 max_queue_pairs;
2996         int mtu;
2997 
2998         /* Find if host supports multiqueue virtio_net device */
2999         err = virtio_cread_feature(vdev, VIRTIO_NET_F_MQ,
3000                                    struct virtio_net_config,
3001                                    max_virtqueue_pairs, &max_queue_pairs);
3002 
3003         /* We need at least 2 queue's */
3004         if (err || max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
3005             max_queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
3006             !virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
3007                 max_queue_pairs = 1;
3008 
3009         /* Allocate ourselves a network device with room for our info */
3010         dev = alloc_etherdev_mq(sizeof(struct virtnet_info), max_queue_pairs);
3011         if (!dev)
3012                 return -ENOMEM;
3013 
3014         /* Set up network device as normal. */
3015         dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE;
3016         dev->netdev_ops = &virtnet_netdev;
3017         dev->features = NETIF_F_HIGHDMA;
3018 
3019         dev->ethtool_ops = &virtnet_ethtool_ops;
3020         SET_NETDEV_DEV(dev, &vdev->dev);
3021 
3022         /* Do we support "hardware" checksums? */
3023         if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
3024                 /* This opens up the world of extra features. */
3025                 dev->hw_features |= NETIF_F_HW_CSUM | NETIF_F_SG;
3026                 if (csum)
3027                         dev->features |= NETIF_F_HW_CSUM | NETIF_F_SG;
3028 
3029                 if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
3030                         dev->hw_features |= NETIF_F_TSO
3031                                 | NETIF_F_TSO_ECN | NETIF_F_TSO6;
3032                 }
3033                 /* Individual feature bits: what can host handle? */
3034                 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
3035                         dev->hw_features |= NETIF_F_TSO;
3036                 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
3037                         dev->hw_features |= NETIF_F_TSO6;
3038                 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
3039                         dev->hw_features |= NETIF_F_TSO_ECN;
3040 
3041                 dev->features |= NETIF_F_GSO_ROBUST;
3042 
3043                 if (gso)
3044                         dev->features |= dev->hw_features & NETIF_F_ALL_TSO;
3045                 /* (!csum && gso) case will be fixed by register_netdev() */
3046         }
3047         if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_CSUM))
3048                 dev->features |= NETIF_F_RXCSUM;
3049         if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
3050             virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6))
3051                 dev->features |= NETIF_F_LRO;
3052         if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS))
3053                 dev->hw_features |= NETIF_F_LRO;
3054 
3055         dev->vlan_features = dev->features;
3056 
3057         /* MTU range: 68 - 65535 */
3058         dev->min_mtu = MIN_MTU;
3059         dev->max_mtu = MAX_MTU;
3060 
3061         /* Configuration may specify what MAC to use.  Otherwise random. */
3062         if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC))
3063                 virtio_cread_bytes(vdev,
3064                                    offsetof(struct virtio_net_config, mac),
3065                                    dev->dev_addr, dev->addr_len);
3066         else
3067                 eth_hw_addr_random(dev);
3068 
3069         /* Set up our device-specific information */
3070         vi = netdev_priv(dev);
3071         vi->dev = dev;
3072         vi->vdev = vdev;
3073         vdev->priv = vi;
3074 
3075         INIT_WORK(&vi->config_work, virtnet_config_changed_work);
3076 
3077         /* If we can receive ANY GSO packets, we must allocate large ones. */
3078         if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
3079             virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) ||
3080             virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN) ||
3081             virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_UFO))
3082                 vi->big_packets = true;
3083 
3084         if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
3085                 vi->mergeable_rx_bufs = true;
3086 
3087         if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF) ||
3088             virtio_has_feature(vdev, VIRTIO_F_VERSION_1))
3089                 vi->hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
3090         else
3091                 vi->hdr_len = sizeof(struct virtio_net_hdr);
3092 
3093         if (virtio_has_feature(vdev, VIRTIO_F_ANY_LAYOUT) ||
3094             virtio_has_feature(vdev, VIRTIO_F_VERSION_1))
3095                 vi->any_header_sg = true;
3096 
3097         if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
3098                 vi->has_cvq = true;
3099 
3100         if (virtio_has_feature(vdev, VIRTIO_NET_F_MTU)) {
3101                 mtu = virtio_cread16(vdev,
3102                                      offsetof(struct virtio_net_config,
3103                                               mtu));
3104                 if (mtu < dev->min_mtu) {
3105                         /* Should never trigger: MTU was previously validated
3106                          * in virtnet_validate.
3107                          */
3108                         dev_err(&vdev->dev,
3109                                 "device MTU appears to have changed it is now %d < %d",
3110                                 mtu, dev->min_mtu);
3111                         goto free;
3112                 }
3113 
3114                 dev->mtu = mtu;
3115                 dev->max_mtu = mtu;
3116 
3117                 /* TODO: size buffers correctly in this case. */
3118                 if (dev->mtu > ETH_DATA_LEN)
3119                         vi->big_packets = true;
3120         }
3121 
3122         if (vi->any_header_sg)
3123                 dev->needed_headroom = vi->hdr_len;
3124 
3125         /* Enable multiqueue by default */
3126         if (num_online_cpus() >= max_queue_pairs)
3127                 vi->curr_queue_pairs = max_queue_pairs;
3128         else
3129                 vi->curr_queue_pairs = num_online_cpus();
3130         vi->max_queue_pairs = max_queue_pairs;
3131 
3132         /* Allocate/initialize the rx/tx queues, and invoke find_vqs */
3133         err = init_vqs(vi);
3134         if (err)
3135                 goto free;
3136 
3137 #ifdef CONFIG_SYSFS
3138         if (vi->mergeable_rx_bufs)
3139                 dev->sysfs_rx_queue_group = &virtio_net_mrg_rx_group;
3140 #endif
3141         netif_set_real_num_tx_queues(dev, vi->curr_queue_pairs);
3142         netif_set_real_num_rx_queues(dev, vi->curr_queue_pairs);
3143 
3144         virtnet_init_settings(dev);
3145 
3146         if (virtio_has_feature(vdev, VIRTIO_NET_F_STANDBY)) {
3147                 vi->failover = net_failover_create(vi->dev);
3148                 if (IS_ERR(vi->failover)) {
3149                         err = PTR_ERR(vi->failover);
3150                         goto free_vqs;
3151                 }
3152         }
3153 
3154         err = register_netdev(dev);
3155         if (err) {
3156                 pr_debug("virtio_net: registering device failed\n");
3157                 goto free_failover;
3158         }
3159 
3160         virtio_device_ready(vdev);
3161 
3162         err = virtnet_cpu_notif_add(vi);
3163         if (err) {
3164                 pr_debug("virtio_net: registering cpu notifier failed\n");
3165                 goto free_unregister_netdev;
3166         }
3167 
3168         virtnet_set_queues(vi, vi->curr_queue_pairs);
3169 
3170         /* Assume link up if device can't report link status,
3171            otherwise get link status from config. */
3172         netif_carrier_off(dev);
3173         if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) {
3174                 schedule_work(&vi->config_work);
3175         } else {
3176                 vi->status = VIRTIO_NET_S_LINK_UP;
3177                 virtnet_update_settings(vi);
3178                 netif_carrier_on(dev);
3179         }
3180 
3181         for (i = 0; i < ARRAY_SIZE(guest_offloads); i++)
3182                 if (virtio_has_feature(vi->vdev, guest_offloads[i]))
3183                         set_bit(guest_offloads[i], &vi->guest_offloads);
3184         vi->guest_offloads_capable = vi->guest_offloads;
3185 
3186         pr_debug("virtnet: registered device %s with %d RX and TX vq's\n",
3187                  dev->name, max_queue_pairs);
3188 
3189         return 0;
3190 
3191 free_unregister_netdev:
3192         vi->vdev->config->reset(vdev);
3193 
3194         unregister_netdev(dev);
3195 free_failover:
3196         net_failover_destroy(vi->failover);
3197 free_vqs:
3198         cancel_delayed_work_sync(&vi->refill);
3199         free_receive_page_frags(vi);
3200         virtnet_del_vqs(vi);
3201 free:
3202         free_netdev(dev);
3203         return err;
3204 }
3205 
3206 static void remove_vq_common(struct virtnet_info *vi)
3207 {
3208         vi->vdev->config->reset(vi->vdev);
3209 
3210         /* Free unused buffers in both send and recv, if any. */
3211         free_unused_bufs(vi);
3212 
3213         free_receive_bufs(vi);
3214 
3215         free_receive_page_frags(vi);
3216 
3217         virtnet_del_vqs(vi);
3218 }
3219 
3220 static void virtnet_remove(struct virtio_device *vdev)
3221 {
3222         struct virtnet_info *vi = vdev->priv;
3223 
3224         virtnet_cpu_notif_remove(vi);
3225 
3226         /* Make sure no work handler is accessing the device. */
3227         flush_work(&vi->config_work);
3228 
3229         unregister_netdev(vi->dev);
3230 
3231         net_failover_destroy(vi->failover);
3232 
3233         remove_vq_common(vi);
3234 
3235         free_netdev(vi->dev);
3236 }
3237 
3238 static __maybe_unused int virtnet_freeze(struct virtio_device *vdev)
3239 {
3240         struct virtnet_info *vi = vdev->priv;
3241 
3242         virtnet_cpu_notif_remove(vi);
3243         virtnet_freeze_down(vdev);
3244         remove_vq_common(vi);
3245 
3246         return 0;
3247 }
3248 
3249 static __maybe_unused int virtnet_restore(struct virtio_device *vdev)
3250 {
3251         struct virtnet_info *vi = vdev->priv;
3252         int err;
3253 
3254         err = virtnet_restore_up(vdev);
3255         if (err)
3256                 return err;
3257         virtnet_set_queues(vi, vi->curr_queue_pairs);
3258 
3259         err = virtnet_cpu_notif_add(vi);
3260         if (err)
3261                 return err;
3262 
3263         return 0;
3264 }
3265 
3266 static struct virtio_device_id id_table[] = {
3267         { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
3268         { 0 },
3269 };
3270 
3271 #define VIRTNET_FEATURES \
3272         VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM, \
3273         VIRTIO_NET_F_MAC, \
3274         VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6, \
3275         VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6, \
3276         VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO, \
3277         VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ, \
3278         VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN, \
3279         VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ, \
3280         VIRTIO_NET_F_CTRL_MAC_ADDR, \
3281         VIRTIO_NET_F_MTU, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS, \
3282         VIRTIO_NET_F_SPEED_DUPLEX, VIRTIO_NET_F_STANDBY
3283 
3284 static unsigned int features[] = {
3285         VIRTNET_FEATURES,
3286 };
3287 
3288 static unsigned int features_legacy[] = {
3289         VIRTNET_FEATURES,
3290         VIRTIO_NET_F_GSO,
3291         VIRTIO_F_ANY_LAYOUT,
3292 };
3293 
3294 static struct virtio_driver virtio_net_driver = {
3295         .feature_table = features,
3296         .feature_table_size = ARRAY_SIZE(features),
3297         .feature_table_legacy = features_legacy,
3298         .feature_table_size_legacy = ARRAY_SIZE(features_legacy),
3299         .driver.name =  KBUILD_MODNAME,
3300         .driver.owner = THIS_MODULE,
3301         .id_table =     id_table,
3302         .validate =     virtnet_validate,
3303         .probe =        virtnet_probe,
3304         .remove =       virtnet_remove,
3305         .config_changed = virtnet_config_changed,
3306 #ifdef CONFIG_PM_SLEEP
3307         .freeze =       virtnet_freeze,
3308         .restore =      virtnet_restore,
3309 #endif
3310 };
3311 
3312 static __init int virtio_net_driver_init(void)
3313 {
3314         int ret;
3315 
3316         ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "virtio/net:online",
3317                                       virtnet_cpu_online,
3318                                       virtnet_cpu_down_prep);
3319         if (ret < 0)
3320                 goto out;
3321         virtionet_online = ret;
3322         ret = cpuhp_setup_state_multi(CPUHP_VIRT_NET_DEAD, "virtio/net:dead",
3323                                       NULL, virtnet_cpu_dead);
3324         if (ret)
3325                 goto err_dead;
3326 
3327         ret = register_virtio_driver(&virtio_net_driver);
3328         if (ret)
3329                 goto err_virtio;
3330         return 0;
3331 err_virtio:
3332         cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD);
3333 err_dead:
3334         cpuhp_remove_multi_state(virtionet_online);
3335 out:
3336         return ret;
3337 }
3338 module_init(virtio_net_driver_init);
3339 
3340 static __exit void virtio_net_driver_exit(void)
3341 {
3342         unregister_virtio_driver(&virtio_net_driver);
3343         cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD);
3344         cpuhp_remove_multi_state(virtionet_online);
3345 }
3346 module_exit(virtio_net_driver_exit);
3347 
3348 MODULE_DEVICE_TABLE(virtio, id_table);
3349 MODULE_DESCRIPTION("Virtio network driver");
3350 MODULE_LICENSE("GPL");

/* [<][>][^][v][top][bottom][index][help] */