drivers/net/xen-netfront.c

   1 /*
   2  * Virtual network driver for conversing with remote driver backends.
   3  *
   4  * Copyright (c) 2002-2005, K A Fraser
   5  * Copyright (c) 2005, XenSource Ltd
   6  *
   7  * This program is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU General Public License version 2
   9  * as published by the Free Software Foundation; or, when distributed
  10  * separately from the Linux kernel or incorporated into other
  11  * software packages, subject to the following license:
  12  *
  13  * Permission is hereby granted, free of charge, to any person obtaining a copy
  14  * of this source file (the "Software"), to deal in the Software without
  15  * restriction, including without limitation the rights to use, copy, modify,
  16  * merge, publish, distribute, sublicense, and/or sell copies of the Software,
  17  * and to permit persons to whom the Software is furnished to do so, subject to
  18  * the following conditions:
  19  *
  20  * The above copyright notice and this permission notice shall be included in
  21  * all copies or substantial portions of the Software.
  22  *
  23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  24  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  25  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  26  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  27  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  28  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  29  * IN THE SOFTWARE.
  30  */
  31
  32 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  33
  34 #include <linux/module.h>
  35 #include <linux/kernel.h>
  36 #include <linux/netdevice.h>
  37 #include <linux/etherdevice.h>
  38 #include <linux/skbuff.h>
  39 #include <linux/ethtool.h>
  40 #include <linux/if_ether.h>
  41 #include <net/tcp.h>
  42 #include <linux/udp.h>
  43 #include <linux/moduleparam.h>
  44 #include <linux/mm.h>
  45 #include <linux/slab.h>
  46 #include <net/ip.h>
  47 #include <linux/bpf.h>
  48 #include <net/page_pool.h>
  49 #include <linux/bpf_trace.h>
  50
  51 #include <xen/xen.h>
  52 #include <xen/xenbus.h>
  53 #include <xen/events.h>
  54 #include <xen/page.h>
  55 #include <xen/platform_pci.h>
  56 #include <xen/grant_table.h>
  57
  58 #include <xen/interface/io/netif.h>
  59 #include <xen/interface/memory.h>
  60 #include <xen/interface/grant_table.h>
  61
  62 /* Module parameters */
  63 #define MAX_QUEUES_DEFAULT 8
  64 static unsigned int xennet_max_queues;
  65 module_param_named(max_queues, xennet_max_queues, uint, 0644);
  66 MODULE_PARM_DESC(max_queues,
  67                  "Maximum number of queues per virtual interface");
  68
  69 #define XENNET_TIMEOUT  (5 * HZ)
  70
  71 static const struct ethtool_ops xennet_ethtool_ops;
  72
  73 struct netfront_cb {
  74         int pull_to;
  75 };
  76
  77 #define NETFRONT_SKB_CB(skb)    ((struct netfront_cb *)((skb)->cb))
  78
  79 #define RX_COPY_THRESHOLD 256
  80
  81 #define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, XEN_PAGE_SIZE)
  82 #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, XEN_PAGE_SIZE)
  83
  84 /* Minimum number of Rx slots (includes slot for GSO metadata). */
  85 #define NET_RX_SLOTS_MIN (XEN_NETIF_NR_SLOTS_MIN + 1)
  86
  87 /* Queue name is interface name with "-qNNN" appended */
  88 #define QUEUE_NAME_SIZE (IFNAMSIZ + 6)
  89
  90 /* IRQ name is queue name with "-tx" or "-rx" appended */
  91 #define IRQ_NAME_SIZE (QUEUE_NAME_SIZE + 3)
  92
  93 static DECLARE_WAIT_QUEUE_HEAD(module_wq);
  94
  95 struct netfront_stats {
  96         u64                     packets;
  97         u64                     bytes;
  98         struct u64_stats_sync   syncp;
  99 };
 100
 101 struct netfront_info;
 102
 103 struct netfront_queue {
 104         unsigned int id; /* Queue ID, 0-based */
 105         char name[QUEUE_NAME_SIZE]; /* DEVNAME-qN */
 106         struct netfront_info *info;
 107
 108         struct bpf_prog __rcu *xdp_prog;
 109
 110         struct napi_struct napi;
 111
 112         /* Split event channels support, tx_* == rx_* when using
 113          * single event channel.
 114          */
 115         unsigned int tx_evtchn, rx_evtchn;
 116         unsigned int tx_irq, rx_irq;
 117         /* Only used when split event channels support is enabled */
 118         char tx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-tx */
 119         char rx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-rx */
 120
 121         spinlock_t   tx_lock;
 122         struct xen_netif_tx_front_ring tx;
 123         int tx_ring_ref;
 124
 125         /*
 126          * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
 127          * are linked from tx_skb_freelist through tx_link.
 128          */
 129         struct sk_buff *tx_skbs[NET_TX_RING_SIZE];
 130         unsigned short tx_link[NET_TX_RING_SIZE];
 131 #define TX_LINK_NONE 0xffff
 132 #define TX_PENDING   0xfffe
 133         grant_ref_t gref_tx_head;
 134         grant_ref_t grant_tx_ref[NET_TX_RING_SIZE];
 135         struct page *grant_tx_page[NET_TX_RING_SIZE];
 136         unsigned tx_skb_freelist;
 137         unsigned int tx_pend_queue;
 138
 139         spinlock_t   rx_lock ____cacheline_aligned_in_smp;
 140         struct xen_netif_rx_front_ring rx;
 141         int rx_ring_ref;
 142
 143         struct timer_list rx_refill_timer;
 144
 145         struct sk_buff *rx_skbs[NET_RX_RING_SIZE];
 146         grant_ref_t gref_rx_head;
 147         grant_ref_t grant_rx_ref[NET_RX_RING_SIZE];
 148
 149         unsigned int rx_rsp_unconsumed;
 150         spinlock_t rx_cons_lock;
 151
 152         struct page_pool *page_pool;
 153         struct xdp_rxq_info xdp_rxq;
 154 };
 155
 156 struct netfront_info {
 157         struct list_head list;
 158         struct net_device *netdev;
 159
 160         struct xenbus_device *xbdev;
 161
 162         /* Multi-queue support */
 163         struct netfront_queue *queues;
 164
 165         /* Statistics */
 166         struct netfront_stats __percpu *rx_stats;
 167         struct netfront_stats __percpu *tx_stats;
 168
 169         /* XDP state */
 170         bool netback_has_xdp_headroom;
 171         bool netfront_xdp_enabled;
 172
 173         /* Is device behaving sane? */
 174         bool broken;
 175
 176         atomic_t rx_gso_checksum_fixup;
 177 };
 178
 179 struct netfront_rx_info {
 180         struct xen_netif_rx_response rx;
 181         struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
 182 };
 183
 184 /*
 185  * Access macros for acquiring freeing slots in tx_skbs[].
 186  */
 187
 188 static void add_id_to_list(unsigned *head, unsigned short *list,
 189                            unsigned short id)
 190 {
 191         list[id] = *head;
 192         *head = id;
 193 }
 194
 195 static unsigned short get_id_from_list(unsigned *head, unsigned short *list)
 196 {
 197         unsigned int id = *head;
 198
 199         if (id != TX_LINK_NONE) {
 200                 *head = list[id];
 201                 list[id] = TX_LINK_NONE;
 202         }
 203         return id;
 204 }
 205
 206 static int xennet_rxidx(RING_IDX idx)
 207 {
 208         return idx & (NET_RX_RING_SIZE - 1);
 209 }
 210
 211 static struct sk_buff *xennet_get_rx_skb(struct netfront_queue *queue,
 212                                          RING_IDX ri)
 213 {
 214         int i = xennet_rxidx(ri);
 215         struct sk_buff *skb = queue->rx_skbs[i];
 216         queue->rx_skbs[i] = NULL;
 217         return skb;
 218 }
 219
 220 static grant_ref_t xennet_get_rx_ref(struct netfront_queue *queue,
 221                                             RING_IDX ri)
 222 {
 223         int i = xennet_rxidx(ri);
 224         grant_ref_t ref = queue->grant_rx_ref[i];
 225         queue->grant_rx_ref[i] = INVALID_GRANT_REF;
 226         return ref;
 227 }
 228
 229 #ifdef CONFIG_SYSFS
 230 static const struct attribute_group xennet_dev_group;
 231 #endif
 232
 233 static bool xennet_can_sg(struct net_device *dev)
 234 {
 235         return dev->features & NETIF_F_SG;
 236 }
 237
 238
 239 static void rx_refill_timeout(struct timer_list *t)
 240 {
 241         struct netfront_queue *queue = from_timer(queue, t, rx_refill_timer);
 242         napi_schedule(&queue->napi);
 243 }
 244
 245 static int netfront_tx_slot_available(struct netfront_queue *queue)
 246 {
 247         return (queue->tx.req_prod_pvt - queue->tx.rsp_cons) <
 248                 (NET_TX_RING_SIZE - XEN_NETIF_NR_SLOTS_MIN - 1);
 249 }
 250
 251 static void xennet_maybe_wake_tx(struct netfront_queue *queue)
 252 {
 253         struct net_device *dev = queue->info->netdev;
 254         struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, queue->id);
 255
 256         if (unlikely(netif_tx_queue_stopped(dev_queue)) &&
 257             netfront_tx_slot_available(queue) &&
 258             likely(netif_running(dev)))
 259                 netif_tx_wake_queue(netdev_get_tx_queue(dev, queue->id));
 260 }
 261
 262
 263 static struct sk_buff *xennet_alloc_one_rx_buffer(struct netfront_queue *queue)
 264 {
 265         struct sk_buff *skb;
 266         struct page *page;
 267
 268         skb = __netdev_alloc_skb(queue->info->netdev,
 269                                  RX_COPY_THRESHOLD + NET_IP_ALIGN,
 270                                  GFP_ATOMIC | __GFP_NOWARN);
 271         if (unlikely(!skb))
 272                 return NULL;
 273
 274         page = page_pool_dev_alloc_pages(queue->page_pool);
 275         if (unlikely(!page)) {
 276                 kfree_skb(skb);
 277                 return NULL;
 278         }
 279         skb_add_rx_frag(skb, 0, page, 0, 0, PAGE_SIZE);
 280
 281         /* Align ip header to a 16 bytes boundary */
 282         skb_reserve(skb, NET_IP_ALIGN);
 283         skb->dev = queue->info->netdev;
 284
 285         return skb;
 286 }
 287
 288
 289 static void xennet_alloc_rx_buffers(struct netfront_queue *queue)
 290 {
 291         RING_IDX req_prod = queue->rx.req_prod_pvt;
 292         int notify;
 293         int err = 0;
 294
 295         if (unlikely(!netif_carrier_ok(queue->info->netdev)))
 296                 return;
 297
 298         for (req_prod = queue->rx.req_prod_pvt;
 299              req_prod - queue->rx.rsp_cons < NET_RX_RING_SIZE;
 300              req_prod++) {
 301                 struct sk_buff *skb;
 302                 unsigned short id;
 303                 grant_ref_t ref;
 304                 struct page *page;
 305                 struct xen_netif_rx_request *req;
 306
 307                 skb = xennet_alloc_one_rx_buffer(queue);
 308                 if (!skb) {
 309                         err = -ENOMEM;
 310                         break;
 311                 }
 312
 313                 id = xennet_rxidx(req_prod);
 314
 315                 BUG_ON(queue->rx_skbs[id]);
 316                 queue->rx_skbs[id] = skb;
 317
 318                 ref = gnttab_claim_grant_reference(&queue->gref_rx_head);
 319                 WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
 320                 queue->grant_rx_ref[id] = ref;
 321
 322                 page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
 323
 324                 req = RING_GET_REQUEST(&queue->rx, req_prod);
 325                 gnttab_page_grant_foreign_access_ref_one(ref,
 326                                                          queue->info->xbdev->otherend_id,
 327                                                          page,
 328                                                          0);
 329                 req->id = id;
 330                 req->gref = ref;
 331         }
 332
 333         queue->rx.req_prod_pvt = req_prod;
 334
 335         /* Try again later if there are not enough requests or skb allocation
 336          * failed.
 337          * Enough requests is quantified as the sum of newly created slots and
 338          * the unconsumed slots at the backend.
 339          */
 340         if (req_prod - queue->rx.rsp_cons < NET_RX_SLOTS_MIN ||
 341             unlikely(err)) {
 342                 mod_timer(&queue->rx_refill_timer, jiffies + (HZ/10));
 343                 return;
 344         }
 345
 346         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->rx, notify);
 347         if (notify)
 348                 notify_remote_via_irq(queue->rx_irq);
 349 }
 350
 351 static int xennet_open(struct net_device *dev)
 352 {
 353         struct netfront_info *np = netdev_priv(dev);
 354         unsigned int num_queues = dev->real_num_tx_queues;
 355         unsigned int i = 0;
 356         struct netfront_queue *queue = NULL;
 357
 358         if (!np->queues || np->broken)
 359                 return -ENODEV;
 360
 361         for (i = 0; i < num_queues; ++i) {
 362                 queue = &np->queues[i];
 363                 napi_enable(&queue->napi);
 364
 365                 spin_lock_bh(&queue->rx_lock);
 366                 if (netif_carrier_ok(dev)) {
 367                         xennet_alloc_rx_buffers(queue);
 368                         queue->rx.sring->rsp_event = queue->rx.rsp_cons + 1;
 369                         if (RING_HAS_UNCONSUMED_RESPONSES(&queue->rx))
 370                                 napi_schedule(&queue->napi);
 371                 }
 372                 spin_unlock_bh(&queue->rx_lock);
 373         }
 374
 375         netif_tx_start_all_queues(dev);
 376
 377         return 0;
 378 }
 379
 380 static bool xennet_tx_buf_gc(struct netfront_queue *queue)
 381 {
 382         RING_IDX cons, prod;
 383         unsigned short id;
 384         struct sk_buff *skb;
 385         bool more_to_do;
 386         bool work_done = false;
 387         const struct device *dev = &queue->info->netdev->dev;
 388
 389         BUG_ON(!netif_carrier_ok(queue->info->netdev));
 390
 391         do {
 392                 prod = queue->tx.sring->rsp_prod;
 393                 if (RING_RESPONSE_PROD_OVERFLOW(&queue->tx, prod)) {
 394                         dev_alert(dev, "Illegal number of responses %u\n",
 395                                   prod - queue->tx.rsp_cons);
 396                         goto err;
 397                 }
 398                 rmb(); /* Ensure we see responses up to 'rp'. */
 399
 400                 for (cons = queue->tx.rsp_cons; cons != prod; cons++) {
 401                         struct xen_netif_tx_response txrsp;
 402
 403                         work_done = true;
 404
 405                         RING_COPY_RESPONSE(&queue->tx, cons, &txrsp);
 406                         if (txrsp.status == XEN_NETIF_RSP_NULL)
 407                                 continue;
 408
 409                         id = txrsp.id;
 410                         if (id >= RING_SIZE(&queue->tx)) {
 411                                 dev_alert(dev,
 412                                           "Response has incorrect id (%u)\n",
 413                                           id);
 414                                 goto err;
 415                         }
 416                         if (queue->tx_link[id] != TX_PENDING) {
 417                                 dev_alert(dev,
 418                                           "Response for inactive request\n");
 419                                 goto err;
 420                         }
 421
 422                         queue->tx_link[id] = TX_LINK_NONE;
 423                         skb = queue->tx_skbs[id];
 424                         queue->tx_skbs[id] = NULL;
 425                         if (unlikely(!gnttab_end_foreign_access_ref(
 426                                 queue->grant_tx_ref[id]))) {
 427                                 dev_alert(dev,
 428                                           "Grant still in use by backend domain\n");
 429                                 goto err;
 430                         }
 431                         gnttab_release_grant_reference(
 432                                 &queue->gref_tx_head, queue->grant_tx_ref[id]);
 433                         queue->grant_tx_ref[id] = INVALID_GRANT_REF;
 434                         queue->grant_tx_page[id] = NULL;
 435                         add_id_to_list(&queue->tx_skb_freelist, queue->tx_link, id);
 436                         dev_kfree_skb_irq(skb);
 437                 }
 438
 439                 queue->tx.rsp_cons = prod;
 440
 441                 RING_FINAL_CHECK_FOR_RESPONSES(&queue->tx, more_to_do);
 442         } while (more_to_do);
 443
 444         xennet_maybe_wake_tx(queue);
 445
 446         return work_done;
 447
 448  err:
 449         queue->info->broken = true;
 450         dev_alert(dev, "Disabled for further use\n");
 451
 452         return work_done;
 453 }
 454
 455 struct xennet_gnttab_make_txreq {
 456         struct netfront_queue *queue;
 457         struct sk_buff *skb;
 458         struct page *page;
 459         struct xen_netif_tx_request *tx;      /* Last request on ring page */
 460         struct xen_netif_tx_request tx_local; /* Last request local copy*/
 461         unsigned int size;
 462 };
 463
 464 static void xennet_tx_setup_grant(unsigned long gfn, unsigned int offset,
 465                                   unsigned int len, void *data)
 466 {
 467         struct xennet_gnttab_make_txreq *info = data;
 468         unsigned int id;
 469         struct xen_netif_tx_request *tx;
 470         grant_ref_t ref;
 471         /* convenient aliases */
 472         struct page *page = info->page;
 473         struct netfront_queue *queue = info->queue;
 474         struct sk_buff *skb = info->skb;
 475
 476         id = get_id_from_list(&queue->tx_skb_freelist, queue->tx_link);
 477         tx = RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
 478         ref = gnttab_claim_grant_reference(&queue->gref_tx_head);
 479         WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
 480
 481         gnttab_grant_foreign_access_ref(ref, queue->info->xbdev->otherend_id,
 482                                         gfn, GNTMAP_readonly);
 483
 484         queue->tx_skbs[id] = skb;
 485         queue->grant_tx_page[id] = page;
 486         queue->grant_tx_ref[id] = ref;
 487
 488         info->tx_local.id = id;
 489         info->tx_local.gref = ref;
 490         info->tx_local.offset = offset;
 491         info->tx_local.size = len;
 492         info->tx_local.flags = 0;
 493
 494         *tx = info->tx_local;
 495
 496         /*
 497          * Put the request in the pending queue, it will be set to be pending
 498          * when the producer index is about to be raised.
 499          */
 500         add_id_to_list(&queue->tx_pend_queue, queue->tx_link, id);
 501
 502         info->tx = tx;
 503         info->size += info->tx_local.size;
 504 }
 505
 506 static struct xen_netif_tx_request *xennet_make_first_txreq(
 507         struct xennet_gnttab_make_txreq *info,
 508         unsigned int offset, unsigned int len)
 509 {
 510         info->size = 0;
 511
 512         gnttab_for_one_grant(info->page, offset, len, xennet_tx_setup_grant, info);
 513
 514         return info->tx;
 515 }
 516
 517 static void xennet_make_one_txreq(unsigned long gfn, unsigned int offset,
 518                                   unsigned int len, void *data)
 519 {
 520         struct xennet_gnttab_make_txreq *info = data;
 521
 522         info->tx->flags |= XEN_NETTXF_more_data;
 523         skb_get(info->skb);
 524         xennet_tx_setup_grant(gfn, offset, len, data);
 525 }
 526
 527 static void xennet_make_txreqs(
 528         struct xennet_gnttab_make_txreq *info,
 529         struct page *page,
 530         unsigned int offset, unsigned int len)
 531 {
 532         /* Skip unused frames from start of page */
 533         page += offset >> PAGE_SHIFT;
 534         offset &= ~PAGE_MASK;
 535
 536         while (len) {
 537                 info->page = page;
 538                 info->size = 0;
 539
 540                 gnttab_foreach_grant_in_range(page, offset, len,
 541                                               xennet_make_one_txreq,
 542                                               info);
 543
 544                 page++;
 545                 offset = 0;
 546                 len -= info->size;
 547         }
 548 }
 549
 550 /*
 551  * Count how many ring slots are required to send this skb. Each frag
 552  * might be a compound page.
 553  */
 554 static int xennet_count_skb_slots(struct sk_buff *skb)
 555 {
 556         int i, frags = skb_shinfo(skb)->nr_frags;
 557         int slots;
 558
 559         slots = gnttab_count_grant(offset_in_page(skb->data),
 560                                    skb_headlen(skb));
 561
 562         for (i = 0; i < frags; i++) {
 563                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 564                 unsigned long size = skb_frag_size(frag);
 565                 unsigned long offset = skb_frag_off(frag);
 566
 567                 /* Skip unused frames from start of page */
 568                 offset &= ~PAGE_MASK;
 569
 570                 slots += gnttab_count_grant(offset, size);
 571         }
 572
 573         return slots;
 574 }
 575
 576 static u16 xennet_select_queue(struct net_device *dev, struct sk_buff *skb,
 577                                struct net_device *sb_dev)
 578 {
 579         unsigned int num_queues = dev->real_num_tx_queues;
 580         u32 hash;
 581         u16 queue_idx;
 582
 583         /* First, check if there is only one queue */
 584         if (num_queues == 1) {
 585                 queue_idx = 0;
 586         } else {
 587                 hash = skb_get_hash(skb);
 588                 queue_idx = hash % num_queues;
 589         }
 590
 591         return queue_idx;
 592 }
 593
 594 static void xennet_mark_tx_pending(struct netfront_queue *queue)
 595 {
 596         unsigned int i;
 597
 598         while ((i = get_id_from_list(&queue->tx_pend_queue, queue->tx_link)) !=
 599                TX_LINK_NONE)
 600                 queue->tx_link[i] = TX_PENDING;
 601 }
 602
 603 static int xennet_xdp_xmit_one(struct net_device *dev,
 604                                struct netfront_queue *queue,
 605                                struct xdp_frame *xdpf)
 606 {
 607         struct netfront_info *np = netdev_priv(dev);
 608         struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats);
 609         struct xennet_gnttab_make_txreq info = {
 610                 .queue = queue,
 611                 .skb = NULL,
 612                 .page = virt_to_page(xdpf->data),
 613         };
 614         int notify;
 615
 616         xennet_make_first_txreq(&info,
 617                                 offset_in_page(xdpf->data),
 618                                 xdpf->len);
 619
 620         xennet_mark_tx_pending(queue);
 621
 622         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
 623         if (notify)
 624                 notify_remote_via_irq(queue->tx_irq);
 625
 626         u64_stats_update_begin(&tx_stats->syncp);
 627         tx_stats->bytes += xdpf->len;
 628         tx_stats->packets++;
 629         u64_stats_update_end(&tx_stats->syncp);
 630
 631         xennet_tx_buf_gc(queue);
 632
 633         return 0;
 634 }
 635
 636 static int xennet_xdp_xmit(struct net_device *dev, int n,
 637                            struct xdp_frame **frames, u32 flags)
 638 {
 639         unsigned int num_queues = dev->real_num_tx_queues;
 640         struct netfront_info *np = netdev_priv(dev);
 641         struct netfront_queue *queue = NULL;
 642         unsigned long irq_flags;
 643         int nxmit = 0;
 644         int i;
 645
 646         if (unlikely(np->broken))
 647                 return -ENODEV;
 648         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
 649                 return -EINVAL;
 650
 651         queue = &np->queues[smp_processor_id() % num_queues];
 652
 653         spin_lock_irqsave(&queue->tx_lock, irq_flags);
 654         for (i = 0; i < n; i++) {
 655                 struct xdp_frame *xdpf = frames[i];
 656
 657                 if (!xdpf)
 658                         continue;
 659                 if (xennet_xdp_xmit_one(dev, queue, xdpf))
 660                         break;
 661                 nxmit++;
 662         }
 663         spin_unlock_irqrestore(&queue->tx_lock, irq_flags);
 664
 665         return nxmit;
 666 }
 667
 668
 669 #define MAX_XEN_SKB_FRAGS (65536 / XEN_PAGE_SIZE + 1)
 670
 671 static netdev_tx_t xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
 672 {
 673         struct netfront_info *np = netdev_priv(dev);
 674         struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats);
 675         struct xen_netif_tx_request *first_tx;
 676         unsigned int i;
 677         int notify;
 678         int slots;
 679         struct page *page;
 680         unsigned int offset;
 681         unsigned int len;
 682         unsigned long flags;
 683         struct netfront_queue *queue = NULL;
 684         struct xennet_gnttab_make_txreq info = { };
 685         unsigned int num_queues = dev->real_num_tx_queues;
 686         u16 queue_index;
 687         struct sk_buff *nskb;
 688
 689         /* Drop the packet if no queues are set up */
 690         if (num_queues < 1)
 691                 goto drop;
 692         if (unlikely(np->broken))
 693                 goto drop;
 694         /* Determine which queue to transmit this SKB on */
 695         queue_index = skb_get_queue_mapping(skb);
 696         queue = &np->queues[queue_index];
 697
 698         /* If skb->len is too big for wire format, drop skb and alert
 699          * user about misconfiguration.
 700          */
 701         if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) {
 702                 net_alert_ratelimited(
 703                         "xennet: skb->len = %u, too big for wire format\n",
 704                         skb->len);
 705                 goto drop;
 706         }
 707
 708         slots = xennet_count_skb_slots(skb);
 709         if (unlikely(slots > MAX_XEN_SKB_FRAGS + 1)) {
 710                 net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n",
 711                                     slots, skb->len);
 712                 if (skb_linearize(skb))
 713                         goto drop;
 714         }
 715
 716         page = virt_to_page(skb->data);
 717         offset = offset_in_page(skb->data);
 718
 719         /* The first req should be at least ETH_HLEN size or the packet will be
 720          * dropped by netback.
 721          */
 722         if (unlikely(PAGE_SIZE - offset < ETH_HLEN)) {
 723                 nskb = skb_copy(skb, GFP_ATOMIC);
 724                 if (!nskb)
 725                         goto drop;
 726                 dev_consume_skb_any(skb);
 727                 skb = nskb;
 728                 page = virt_to_page(skb->data);
 729                 offset = offset_in_page(skb->data);
 730         }
 731
 732         len = skb_headlen(skb);
 733
 734         spin_lock_irqsave(&queue->tx_lock, flags);
 735
 736         if (unlikely(!netif_carrier_ok(dev) ||
 737                      (slots > 1 && !xennet_can_sg(dev)) ||
 738                      netif_needs_gso(skb, netif_skb_features(skb)))) {
 739                 spin_unlock_irqrestore(&queue->tx_lock, flags);
 740                 goto drop;
 741         }
 742
 743         /* First request for the linear area. */
 744         info.queue = queue;
 745         info.skb = skb;
 746         info.page = page;
 747         first_tx = xennet_make_first_txreq(&info, offset, len);
 748         offset += info.tx_local.size;
 749         if (offset == PAGE_SIZE) {
 750                 page++;
 751                 offset = 0;
 752         }
 753         len -= info.tx_local.size;
 754
 755         if (skb->ip_summed == CHECKSUM_PARTIAL)
 756                 /* local packet? */
 757                 first_tx->flags |= XEN_NETTXF_csum_blank |
 758                                    XEN_NETTXF_data_validated;
 759         else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
 760                 /* remote but checksummed. */
 761                 first_tx->flags |= XEN_NETTXF_data_validated;
 762
 763         /* Optional extra info after the first request. */
 764         if (skb_shinfo(skb)->gso_size) {
 765                 struct xen_netif_extra_info *gso;
 766
 767                 gso = (struct xen_netif_extra_info *)
 768                         RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
 769
 770                 first_tx->flags |= XEN_NETTXF_extra_info;
 771
 772                 gso->u.gso.size = skb_shinfo(skb)->gso_size;
 773                 gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ?
 774                         XEN_NETIF_GSO_TYPE_TCPV6 :
 775                         XEN_NETIF_GSO_TYPE_TCPV4;
 776                 gso->u.gso.pad = 0;
 777                 gso->u.gso.features = 0;
 778
 779                 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
 780                 gso->flags = 0;
 781         }
 782
 783         /* Requests for the rest of the linear area. */
 784         xennet_make_txreqs(&info, page, offset, len);
 785
 786         /* Requests for all the frags. */
 787         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 788                 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
 789                 xennet_make_txreqs(&info, skb_frag_page(frag),
 790                                         skb_frag_off(frag),
 791                                         skb_frag_size(frag));
 792         }
 793
 794         /* First request has the packet length. */
 795         first_tx->size = skb->len;
 796
 797         /* timestamp packet in software */
 798         skb_tx_timestamp(skb);
 799
 800         xennet_mark_tx_pending(queue);
 801
 802         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
 803         if (notify)
 804                 notify_remote_via_irq(queue->tx_irq);
 805
 806         u64_stats_update_begin(&tx_stats->syncp);
 807         tx_stats->bytes += skb->len;
 808         tx_stats->packets++;
 809         u64_stats_update_end(&tx_stats->syncp);
 810
 811         /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
 812         xennet_tx_buf_gc(queue);
 813
 814         if (!netfront_tx_slot_available(queue))
 815                 netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id));
 816
 817         spin_unlock_irqrestore(&queue->tx_lock, flags);
 818
 819         return NETDEV_TX_OK;
 820
 821  drop:
 822         dev->stats.tx_dropped++;
 823         dev_kfree_skb_any(skb);
 824         return NETDEV_TX_OK;
 825 }
 826
 827 static int xennet_close(struct net_device *dev)
 828 {
 829         struct netfront_info *np = netdev_priv(dev);
 830         unsigned int num_queues = dev->real_num_tx_queues;
 831         unsigned int i;
 832         struct netfront_queue *queue;
 833         netif_tx_stop_all_queues(np->netdev);
 834         for (i = 0; i < num_queues; ++i) {
 835                 queue = &np->queues[i];
 836                 napi_disable(&queue->napi);
 837         }
 838         return 0;
 839 }
 840
 841 static void xennet_destroy_queues(struct netfront_info *info)
 842 {
 843         unsigned int i;
 844
 845         for (i = 0; i < info->netdev->real_num_tx_queues; i++) {
 846                 struct netfront_queue *queue = &info->queues[i];
 847
 848                 if (netif_running(info->netdev))
 849                         napi_disable(&queue->napi);
 850                 netif_napi_del(&queue->napi);
 851         }
 852
 853         kfree(info->queues);
 854         info->queues = NULL;
 855 }
 856
 857 static void xennet_uninit(struct net_device *dev)
 858 {
 859         struct netfront_info *np = netdev_priv(dev);
 860         xennet_destroy_queues(np);
 861 }
 862
 863 static void xennet_set_rx_rsp_cons(struct netfront_queue *queue, RING_IDX val)
 864 {
 865         unsigned long flags;
 866
 867         spin_lock_irqsave(&queue->rx_cons_lock, flags);
 868         queue->rx.rsp_cons = val;
 869         queue->rx_rsp_unconsumed = XEN_RING_NR_UNCONSUMED_RESPONSES(&queue->rx);
 870         spin_unlock_irqrestore(&queue->rx_cons_lock, flags);
 871 }
 872
 873 static void xennet_move_rx_slot(struct netfront_queue *queue, struct sk_buff *skb,
 874                                 grant_ref_t ref)
 875 {
 876         int new = xennet_rxidx(queue->rx.req_prod_pvt);
 877
 878         BUG_ON(queue->rx_skbs[new]);
 879         queue->rx_skbs[new] = skb;
 880         queue->grant_rx_ref[new] = ref;
 881         RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->id = new;
 882         RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->gref = ref;
 883         queue->rx.req_prod_pvt++;
 884 }
 885
 886 static int xennet_get_extras(struct netfront_queue *queue,
 887                              struct xen_netif_extra_info *extras,
 888                              RING_IDX rp)
 889
 890 {
 891         struct xen_netif_extra_info extra;
 892         struct device *dev = &queue->info->netdev->dev;
 893         RING_IDX cons = queue->rx.rsp_cons;
 894         int err = 0;
 895
 896         do {
 897                 struct sk_buff *skb;
 898                 grant_ref_t ref;
 899
 900                 if (unlikely(cons + 1 == rp)) {
 901                         if (net_ratelimit())
 902                                 dev_warn(dev, "Missing extra info\n");
 903                         err = -EBADR;
 904                         break;
 905                 }
 906
 907                 RING_COPY_RESPONSE(&queue->rx, ++cons, &extra);
 908
 909                 if (unlikely(!extra.type ||
 910                              extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
 911                         if (net_ratelimit())
 912                                 dev_warn(dev, "Invalid extra type: %d\n",
 913                                          extra.type);
 914                         err = -EINVAL;
 915                 } else {
 916                         extras[extra.type - 1] = extra;
 917                 }
 918
 919                 skb = xennet_get_rx_skb(queue, cons);
 920                 ref = xennet_get_rx_ref(queue, cons);
 921                 xennet_move_rx_slot(queue, skb, ref);
 922         } while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE);
 923
 924         xennet_set_rx_rsp_cons(queue, cons);
 925         return err;
 926 }
 927
 928 static u32 xennet_run_xdp(struct netfront_queue *queue, struct page *pdata,
 929                    struct xen_netif_rx_response *rx, struct bpf_prog *prog,
 930                    struct xdp_buff *xdp, bool *need_xdp_flush)
 931 {
 932         struct xdp_frame *xdpf;
 933         u32 len = rx->status;
 934         u32 act;
 935         int err;
 936
 937         xdp_init_buff(xdp, XEN_PAGE_SIZE - XDP_PACKET_HEADROOM,
 938                       &queue->xdp_rxq);
 939         xdp_prepare_buff(xdp, page_address(pdata), XDP_PACKET_HEADROOM,
 940                          len, false);
 941
 942         act = bpf_prog_run_xdp(prog, xdp);
 943         switch (act) {
 944         case XDP_TX:
 945                 get_page(pdata);
 946                 xdpf = xdp_convert_buff_to_frame(xdp);
 947                 err = xennet_xdp_xmit(queue->info->netdev, 1, &xdpf, 0);
 948                 if (unlikely(!err))
 949                         xdp_return_frame_rx_napi(xdpf);
 950                 else if (unlikely(err < 0))
 951                         trace_xdp_exception(queue->info->netdev, prog, act);
 952                 break;
 953         case XDP_REDIRECT:
 954                 get_page(pdata);
 955                 err = xdp_do_redirect(queue->info->netdev, xdp, prog);
 956                 *need_xdp_flush = true;
 957                 if (unlikely(err))
 958                         trace_xdp_exception(queue->info->netdev, prog, act);
 959                 break;
 960         case XDP_PASS:
 961         case XDP_DROP:
 962                 break;
 963
 964         case XDP_ABORTED:
 965                 trace_xdp_exception(queue->info->netdev, prog, act);
 966                 break;
 967
 968         default:
 969                 bpf_warn_invalid_xdp_action(queue->info->netdev, prog, act);
 970         }
 971
 972         return act;
 973 }
 974
 975 static int xennet_get_responses(struct netfront_queue *queue,
 976                                 struct netfront_rx_info *rinfo, RING_IDX rp,
 977                                 struct sk_buff_head *list,
 978                                 bool *need_xdp_flush)
 979 {
 980         struct xen_netif_rx_response *rx = &rinfo->rx, rx_local;
 981         int max = XEN_NETIF_NR_SLOTS_MIN + (rx->status <= RX_COPY_THRESHOLD);
 982         RING_IDX cons = queue->rx.rsp_cons;
 983         struct sk_buff *skb = xennet_get_rx_skb(queue, cons);
 984         struct xen_netif_extra_info *extras = rinfo->extras;
 985         grant_ref_t ref = xennet_get_rx_ref(queue, cons);
 986         struct device *dev = &queue->info->netdev->dev;
 987         struct bpf_prog *xdp_prog;
 988         struct xdp_buff xdp;
 989         int slots = 1;
 990         int err = 0;
 991         u32 verdict;
 992
 993         if (rx->flags & XEN_NETRXF_extra_info) {
 994                 err = xennet_get_extras(queue, extras, rp);
 995                 if (!err) {
 996                         if (extras[XEN_NETIF_EXTRA_TYPE_XDP - 1].type) {
 997                                 struct xen_netif_extra_info *xdp;
 998
 999                                 xdp = &extras[XEN_NETIF_EXTRA_TYPE_XDP - 1];
1000                                 rx->offset = xdp->u.xdp.headroom;
1001                         }
1002                 }
1003                 cons = queue->rx.rsp_cons;
1004         }
1005
1006         for (;;) {
1007                 if (unlikely(rx->status < 0 ||
1008                              rx->offset + rx->status > XEN_PAGE_SIZE)) {
1009                         if (net_ratelimit())
1010                                 dev_warn(dev, "rx->offset: %u, size: %d\n",
1011                                          rx->offset, rx->status);
1012                         xennet_move_rx_slot(queue, skb, ref);
1013                         err = -EINVAL;
1014                         goto next;
1015                 }
1016
1017                 /*
1018                  * This definitely indicates a bug, either in this driver or in
1019                  * the backend driver. In future this should flag the bad
1020                  * situation to the system controller to reboot the backend.
1021                  */
1022                 if (ref == INVALID_GRANT_REF) {
1023                         if (net_ratelimit())
1024                                 dev_warn(dev, "Bad rx response id %d.\n",
1025                                          rx->id);
1026                         err = -EINVAL;
1027                         goto next;
1028                 }
1029
1030                 if (!gnttab_end_foreign_access_ref(ref)) {
1031                         dev_alert(dev,
1032                                   "Grant still in use by backend domain\n");
1033                         queue->info->broken = true;
1034                         dev_alert(dev, "Disabled for further use\n");
1035                         return -EINVAL;
1036                 }
1037
1038                 gnttab_release_grant_reference(&queue->gref_rx_head, ref);
1039
1040                 rcu_read_lock();
1041                 xdp_prog = rcu_dereference(queue->xdp_prog);
1042                 if (xdp_prog) {
1043                         if (!(rx->flags & XEN_NETRXF_more_data)) {
1044                                 /* currently only a single page contains data */
1045                                 verdict = xennet_run_xdp(queue,
1046                                                          skb_frag_page(&skb_shinfo(skb)->frags[0]),
1047                                                          rx, xdp_prog, &xdp, need_xdp_flush);
1048                                 if (verdict != XDP_PASS)
1049                                         err = -EINVAL;
1050                         } else {
1051                                 /* drop the frame */
1052                                 err = -EINVAL;
1053                         }
1054                 }
1055                 rcu_read_unlock();
1056 next:
1057                 __skb_queue_tail(list, skb);
1058                 if (!(rx->flags & XEN_NETRXF_more_data))
1059                         break;
1060
1061                 if (cons + slots == rp) {
1062                         if (net_ratelimit())
1063                                 dev_warn(dev, "Need more slots\n");
1064                         err = -ENOENT;
1065                         break;
1066                 }
1067
1068                 RING_COPY_RESPONSE(&queue->rx, cons + slots, &rx_local);
1069                 rx = &rx_local;
1070                 skb = xennet_get_rx_skb(queue, cons + slots);
1071                 ref = xennet_get_rx_ref(queue, cons + slots);
1072                 slots++;
1073         }
1074
1075         if (unlikely(slots > max)) {
1076                 if (net_ratelimit())
1077                         dev_warn(dev, "Too many slots\n");
1078                 err = -E2BIG;
1079         }
1080
1081         if (unlikely(err))
1082                 xennet_set_rx_rsp_cons(queue, cons + slots);
1083
1084         return err;
1085 }
1086
1087 static int xennet_set_skb_gso(struct sk_buff *skb,
1088                               struct xen_netif_extra_info *gso)
1089 {
1090         if (!gso->u.gso.size) {
1091                 if (net_ratelimit())
1092                         pr_warn("GSO size must not be zero\n");
1093                 return -EINVAL;
1094         }
1095
1096         if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 &&
1097             gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) {
1098                 if (net_ratelimit())
1099                         pr_warn("Bad GSO type %d\n", gso->u.gso.type);
1100                 return -EINVAL;
1101         }
1102
1103         skb_shinfo(skb)->gso_size = gso->u.gso.size;
1104         skb_shinfo(skb)->gso_type =
1105                 (gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ?
1106                 SKB_GSO_TCPV4 :
1107                 SKB_GSO_TCPV6;
1108
1109         /* Header must be checked, and gso_segs computed. */
1110         skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1111         skb_shinfo(skb)->gso_segs = 0;
1112
1113         return 0;
1114 }
1115
1116 static int xennet_fill_frags(struct netfront_queue *queue,
1117                              struct sk_buff *skb,
1118                              struct sk_buff_head *list)
1119 {
1120         RING_IDX cons = queue->rx.rsp_cons;
1121         struct sk_buff *nskb;
1122
1123         while ((nskb = __skb_dequeue(list))) {
1124                 struct xen_netif_rx_response rx;
1125                 skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0];
1126
1127                 RING_COPY_RESPONSE(&queue->rx, ++cons, &rx);
1128
1129                 if (skb_shinfo(skb)->nr_frags == MAX_SKB_FRAGS) {
1130                         unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
1131
1132                         BUG_ON(pull_to < skb_headlen(skb));
1133                         __pskb_pull_tail(skb, pull_to - skb_headlen(skb));
1134                 }
1135                 if (unlikely(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS)) {
1136                         xennet_set_rx_rsp_cons(queue,
1137                                                ++cons + skb_queue_len(list));
1138                         kfree_skb(nskb);
1139                         return -ENOENT;
1140                 }
1141
1142                 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags,
1143                                 skb_frag_page(nfrag),
1144                                 rx.offset, rx.status, PAGE_SIZE);
1145
1146                 skb_shinfo(nskb)->nr_frags = 0;
1147                 kfree_skb(nskb);
1148         }
1149
1150         xennet_set_rx_rsp_cons(queue, cons);
1151
1152         return 0;
1153 }
1154
1155 static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
1156 {
1157         bool recalculate_partial_csum = false;
1158
1159         /*
1160          * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
1161          * peers can fail to set NETRXF_csum_blank when sending a GSO
1162          * frame. In this case force the SKB to CHECKSUM_PARTIAL and
1163          * recalculate the partial checksum.
1164          */
1165         if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
1166                 struct netfront_info *np = netdev_priv(dev);
1167                 atomic_inc(&np->rx_gso_checksum_fixup);
1168                 skb->ip_summed = CHECKSUM_PARTIAL;
1169                 recalculate_partial_csum = true;
1170         }
1171
1172         /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
1173         if (skb->ip_summed != CHECKSUM_PARTIAL)
1174                 return 0;
1175
1176         return skb_checksum_setup(skb, recalculate_partial_csum);
1177 }
1178
1179 static int handle_incoming_queue(struct netfront_queue *queue,
1180                                  struct sk_buff_head *rxq)
1181 {
1182         struct netfront_stats *rx_stats = this_cpu_ptr(queue->info->rx_stats);
1183         int packets_dropped = 0;
1184         struct sk_buff *skb;
1185
1186         while ((skb = __skb_dequeue(rxq)) != NULL) {
1187                 int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
1188
1189                 if (pull_to > skb_headlen(skb))
1190                         __pskb_pull_tail(skb, pull_to - skb_headlen(skb));
1191
1192                 /* Ethernet work: Delayed to here as it peeks the header. */
1193                 skb->protocol = eth_type_trans(skb, queue->info->netdev);
1194                 skb_reset_network_header(skb);
1195
1196                 if (checksum_setup(queue->info->netdev, skb)) {
1197                         kfree_skb(skb);
1198                         packets_dropped++;
1199                         queue->info->netdev->stats.rx_errors++;
1200                         continue;
1201                 }
1202
1203                 u64_stats_update_begin(&rx_stats->syncp);
1204                 rx_stats->packets++;
1205                 rx_stats->bytes += skb->len;
1206                 u64_stats_update_end(&rx_stats->syncp);
1207
1208                 /* Pass it up. */
1209                 napi_gro_receive(&queue->napi, skb);
1210         }
1211
1212         return packets_dropped;
1213 }
1214
1215 static int xennet_poll(struct napi_struct *napi, int budget)
1216 {
1217         struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi);
1218         struct net_device *dev = queue->info->netdev;
1219         struct sk_buff *skb;
1220         struct netfront_rx_info rinfo;
1221         struct xen_netif_rx_response *rx = &rinfo.rx;
1222         struct xen_netif_extra_info *extras = rinfo.extras;
1223         RING_IDX i, rp;
1224         int work_done;
1225         struct sk_buff_head rxq;
1226         struct sk_buff_head errq;
1227         struct sk_buff_head tmpq;
1228         int err;
1229         bool need_xdp_flush = false;
1230
1231         spin_lock(&queue->rx_lock);
1232
1233         skb_queue_head_init(&rxq);
1234         skb_queue_head_init(&errq);
1235         skb_queue_head_init(&tmpq);
1236
1237         rp = queue->rx.sring->rsp_prod;
1238         if (RING_RESPONSE_PROD_OVERFLOW(&queue->rx, rp)) {
1239                 dev_alert(&dev->dev, "Illegal number of responses %u\n",
1240                           rp - queue->rx.rsp_cons);
1241                 queue->info->broken = true;
1242                 spin_unlock(&queue->rx_lock);
1243                 return 0;
1244         }
1245         rmb(); /* Ensure we see queued responses up to 'rp'. */
1246
1247         i = queue->rx.rsp_cons;
1248         work_done = 0;
1249         while ((i != rp) && (work_done < budget)) {
1250                 RING_COPY_RESPONSE(&queue->rx, i, rx);
1251                 memset(extras, 0, sizeof(rinfo.extras));
1252
1253                 err = xennet_get_responses(queue, &rinfo, rp, &tmpq,
1254                                            &need_xdp_flush);
1255
1256                 if (unlikely(err)) {
1257                         if (queue->info->broken) {
1258                                 spin_unlock(&queue->rx_lock);
1259                                 return 0;
1260                         }
1261 err:
1262                         while ((skb = __skb_dequeue(&tmpq)))
1263                                 __skb_queue_tail(&errq, skb);
1264                         dev->stats.rx_errors++;
1265                         i = queue->rx.rsp_cons;
1266                         continue;
1267                 }
1268
1269                 skb = __skb_dequeue(&tmpq);
1270
1271                 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1272                         struct xen_netif_extra_info *gso;
1273                         gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1274
1275                         if (unlikely(xennet_set_skb_gso(skb, gso))) {
1276                                 __skb_queue_head(&tmpq, skb);
1277                                 xennet_set_rx_rsp_cons(queue,
1278                                                        queue->rx.rsp_cons +
1279                                                        skb_queue_len(&tmpq));
1280                                 goto err;
1281                         }
1282                 }
1283
1284                 NETFRONT_SKB_CB(skb)->pull_to = rx->status;
1285                 if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD)
1286                         NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD;
1287
1288                 skb_frag_off_set(&skb_shinfo(skb)->frags[0], rx->offset);
1289                 skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status);
1290                 skb->data_len = rx->status;
1291                 skb->len += rx->status;
1292
1293                 if (unlikely(xennet_fill_frags(queue, skb, &tmpq)))
1294                         goto err;
1295
1296                 if (rx->flags & XEN_NETRXF_csum_blank)
1297                         skb->ip_summed = CHECKSUM_PARTIAL;
1298                 else if (rx->flags & XEN_NETRXF_data_validated)
1299                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1300
1301                 __skb_queue_tail(&rxq, skb);
1302
1303                 i = queue->rx.rsp_cons + 1;
1304                 xennet_set_rx_rsp_cons(queue, i);
1305                 work_done++;
1306         }
1307         if (need_xdp_flush)
1308                 xdp_do_flush();
1309
1310         __skb_queue_purge(&errq);
1311
1312         work_done -= handle_incoming_queue(queue, &rxq);
1313
1314         xennet_alloc_rx_buffers(queue);
1315
1316         if (work_done < budget) {
1317                 int more_to_do = 0;
1318
1319                 napi_complete_done(napi, work_done);
1320
1321                 RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do);
1322                 if (more_to_do)
1323                         napi_schedule(napi);
1324         }
1325
1326         spin_unlock(&queue->rx_lock);
1327
1328         return work_done;
1329 }
1330
1331 static int xennet_change_mtu(struct net_device *dev, int mtu)
1332 {
1333         int max = xennet_can_sg(dev) ? XEN_NETIF_MAX_TX_SIZE : ETH_DATA_LEN;
1334
1335         if (mtu > max)
1336                 return -EINVAL;
1337         dev->mtu = mtu;
1338         return 0;
1339 }
1340
1341 static void xennet_get_stats64(struct net_device *dev,
1342                                struct rtnl_link_stats64 *tot)
1343 {
1344         struct netfront_info *np = netdev_priv(dev);
1345         int cpu;
1346
1347         for_each_possible_cpu(cpu) {
1348                 struct netfront_stats *rx_stats = per_cpu_ptr(np->rx_stats, cpu);
1349                 struct netfront_stats *tx_stats = per_cpu_ptr(np->tx_stats, cpu);
1350                 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1351                 unsigned int start;
1352
1353                 do {
1354                         start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
1355                         tx_packets = tx_stats->packets;
1356                         tx_bytes = tx_stats->bytes;
1357                 } while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
1358
1359                 do {
1360                         start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
1361                         rx_packets = rx_stats->packets;
1362                         rx_bytes = rx_stats->bytes;
1363                 } while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
1364
1365                 tot->rx_packets += rx_packets;
1366                 tot->tx_packets += tx_packets;
1367                 tot->rx_bytes   += rx_bytes;
1368                 tot->tx_bytes   += tx_bytes;
1369         }
1370
1371         tot->rx_errors  = dev->stats.rx_errors;
1372         tot->tx_dropped = dev->stats.tx_dropped;
1373 }
1374
1375 static void xennet_release_tx_bufs(struct netfront_queue *queue)
1376 {
1377         struct sk_buff *skb;
1378         int i;
1379
1380         for (i = 0; i < NET_TX_RING_SIZE; i++) {
1381                 /* Skip over entries which are actually freelist references */
1382                 if (!queue->tx_skbs[i])
1383                         continue;
1384
1385                 skb = queue->tx_skbs[i];
1386                 queue->tx_skbs[i] = NULL;
1387                 get_page(queue->grant_tx_page[i]);
1388                 gnttab_end_foreign_access(queue->grant_tx_ref[i],
1389                                           (unsigned long)page_address(queue->grant_tx_page[i]));
1390                 queue->grant_tx_page[i] = NULL;
1391                 queue->grant_tx_ref[i] = INVALID_GRANT_REF;
1392                 add_id_to_list(&queue->tx_skb_freelist, queue->tx_link, i);
1393                 dev_kfree_skb_irq(skb);
1394         }
1395 }
1396
1397 static void xennet_release_rx_bufs(struct netfront_queue *queue)
1398 {
1399         int id, ref;
1400
1401         spin_lock_bh(&queue->rx_lock);
1402
1403         for (id = 0; id < NET_RX_RING_SIZE; id++) {
1404                 struct sk_buff *skb;
1405                 struct page *page;
1406
1407                 skb = queue->rx_skbs[id];
1408                 if (!skb)
1409                         continue;
1410
1411                 ref = queue->grant_rx_ref[id];
1412                 if (ref == INVALID_GRANT_REF)
1413                         continue;
1414
1415                 page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
1416
1417                 /* gnttab_end_foreign_access() needs a page ref until
1418                  * foreign access is ended (which may be deferred).
1419                  */
1420                 get_page(page);
1421                 gnttab_end_foreign_access(ref,
1422                                           (unsigned long)page_address(page));
1423                 queue->grant_rx_ref[id] = INVALID_GRANT_REF;
1424
1425                 kfree_skb(skb);
1426         }
1427
1428         spin_unlock_bh(&queue->rx_lock);
1429 }
1430
1431 static netdev_features_t xennet_fix_features(struct net_device *dev,
1432         netdev_features_t features)
1433 {
1434         struct netfront_info *np = netdev_priv(dev);
1435
1436         if (features & NETIF_F_SG &&
1437             !xenbus_read_unsigned(np->xbdev->otherend, "feature-sg", 0))
1438                 features &= ~NETIF_F_SG;
1439
1440         if (features & NETIF_F_IPV6_CSUM &&
1441             !xenbus_read_unsigned(np->xbdev->otherend,
1442                                   "feature-ipv6-csum-offload", 0))
1443                 features &= ~NETIF_F_IPV6_CSUM;
1444
1445         if (features & NETIF_F_TSO &&
1446             !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv4", 0))
1447                 features &= ~NETIF_F_TSO;
1448
1449         if (features & NETIF_F_TSO6 &&
1450             !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv6", 0))
1451                 features &= ~NETIF_F_TSO6;
1452
1453         return features;
1454 }
1455
1456 static int xennet_set_features(struct net_device *dev,
1457         netdev_features_t features)
1458 {
1459         if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1460                 netdev_info(dev, "Reducing MTU because no SG offload");
1461                 dev->mtu = ETH_DATA_LEN;
1462         }
1463
1464         return 0;
1465 }
1466
1467 static bool xennet_handle_tx(struct netfront_queue *queue, unsigned int *eoi)
1468 {
1469         unsigned long flags;
1470
1471         if (unlikely(queue->info->broken))
1472                 return false;
1473
1474         spin_lock_irqsave(&queue->tx_lock, flags);
1475         if (xennet_tx_buf_gc(queue))
1476                 *eoi = 0;
1477         spin_unlock_irqrestore(&queue->tx_lock, flags);
1478
1479         return true;
1480 }
1481
1482 static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id)
1483 {
1484         unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS;
1485
1486         if (likely(xennet_handle_tx(dev_id, &eoiflag)))
1487                 xen_irq_lateeoi(irq, eoiflag);
1488
1489         return IRQ_HANDLED;
1490 }
1491
1492 static bool xennet_handle_rx(struct netfront_queue *queue, unsigned int *eoi)
1493 {
1494         unsigned int work_queued;
1495         unsigned long flags;
1496
1497         if (unlikely(queue->info->broken))
1498                 return false;
1499
1500         spin_lock_irqsave(&queue->rx_cons_lock, flags);
1501         work_queued = XEN_RING_NR_UNCONSUMED_RESPONSES(&queue->rx);
1502         if (work_queued > queue->rx_rsp_unconsumed) {
1503                 queue->rx_rsp_unconsumed = work_queued;
1504                 *eoi = 0;
1505         } else if (unlikely(work_queued < queue->rx_rsp_unconsumed)) {
1506                 const struct device *dev = &queue->info->netdev->dev;
1507
1508                 spin_unlock_irqrestore(&queue->rx_cons_lock, flags);
1509                 dev_alert(dev, "RX producer index going backwards\n");
1510                 dev_alert(dev, "Disabled for further use\n");
1511                 queue->info->broken = true;
1512                 return false;
1513         }
1514         spin_unlock_irqrestore(&queue->rx_cons_lock, flags);
1515
1516         if (likely(netif_carrier_ok(queue->info->netdev) && work_queued))
1517                 napi_schedule(&queue->napi);
1518
1519         return true;
1520 }
1521
1522 static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id)
1523 {
1524         unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS;
1525
1526         if (likely(xennet_handle_rx(dev_id, &eoiflag)))
1527                 xen_irq_lateeoi(irq, eoiflag);
1528
1529         return IRQ_HANDLED;
1530 }
1531
1532 static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1533 {
1534         unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS;
1535
1536         if (xennet_handle_tx(dev_id, &eoiflag) &&
1537             xennet_handle_rx(dev_id, &eoiflag))
1538                 xen_irq_lateeoi(irq, eoiflag);
1539
1540         return IRQ_HANDLED;
1541 }
1542
1543 #ifdef CONFIG_NET_POLL_CONTROLLER
1544 static void xennet_poll_controller(struct net_device *dev)
1545 {
1546         /* Poll each queue */
1547         struct netfront_info *info = netdev_priv(dev);
1548         unsigned int num_queues = dev->real_num_tx_queues;
1549         unsigned int i;
1550
1551         if (info->broken)
1552                 return;
1553
1554         for (i = 0; i < num_queues; ++i)
1555                 xennet_interrupt(0, &info->queues[i]);
1556 }
1557 #endif
1558
1559 #define NETBACK_XDP_HEADROOM_DISABLE    0
1560 #define NETBACK_XDP_HEADROOM_ENABLE     1
1561
1562 static int talk_to_netback_xdp(struct netfront_info *np, int xdp)
1563 {
1564         int err;
1565         unsigned short headroom;
1566
1567         headroom = xdp ? XDP_PACKET_HEADROOM : 0;
1568         err = xenbus_printf(XBT_NIL, np->xbdev->nodename,
1569                             "xdp-headroom", "%hu",
1570                             headroom);
1571         if (err)
1572                 pr_warn("Error writing xdp-headroom\n");
1573
1574         return err;
1575 }
1576
1577 static int xennet_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1578                           struct netlink_ext_ack *extack)
1579 {
1580         unsigned long max_mtu = XEN_PAGE_SIZE - XDP_PACKET_HEADROOM;
1581         struct netfront_info *np = netdev_priv(dev);
1582         struct bpf_prog *old_prog;
1583         unsigned int i, err;
1584
1585         if (dev->mtu > max_mtu) {
1586                 netdev_warn(dev, "XDP requires MTU less than %lu\n", max_mtu);
1587                 return -EINVAL;
1588         }
1589
1590         if (!np->netback_has_xdp_headroom)
1591                 return 0;
1592
1593         xenbus_switch_state(np->xbdev, XenbusStateReconfiguring);
1594
1595         err = talk_to_netback_xdp(np, prog ? NETBACK_XDP_HEADROOM_ENABLE :
1596                                   NETBACK_XDP_HEADROOM_DISABLE);
1597         if (err)
1598                 return err;
1599
1600         /* avoid the race with XDP headroom adjustment */
1601         wait_event(module_wq,
1602                    xenbus_read_driver_state(np->xbdev->otherend) ==
1603                    XenbusStateReconfigured);
1604         np->netfront_xdp_enabled = true;
1605
1606         old_prog = rtnl_dereference(np->queues[0].xdp_prog);
1607
1608         if (prog)
1609                 bpf_prog_add(prog, dev->real_num_tx_queues);
1610
1611         for (i = 0; i < dev->real_num_tx_queues; ++i)
1612                 rcu_assign_pointer(np->queues[i].xdp_prog, prog);
1613
1614         if (old_prog)
1615                 for (i = 0; i < dev->real_num_tx_queues; ++i)
1616                         bpf_prog_put(old_prog);
1617
1618         xenbus_switch_state(np->xbdev, XenbusStateConnected);
1619
1620         return 0;
1621 }
1622
1623 static int xennet_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1624 {
1625         struct netfront_info *np = netdev_priv(dev);
1626
1627         if (np->broken)
1628                 return -ENODEV;
1629
1630         switch (xdp->command) {
1631         case XDP_SETUP_PROG:
1632                 return xennet_xdp_set(dev, xdp->prog, xdp->extack);
1633         default:
1634                 return -EINVAL;
1635         }
1636 }
1637
1638 static const struct net_device_ops xennet_netdev_ops = {
1639         .ndo_uninit          = xennet_uninit,
1640         .ndo_open            = xennet_open,
1641         .ndo_stop            = xennet_close,
1642         .ndo_start_xmit      = xennet_start_xmit,
1643         .ndo_change_mtu      = xennet_change_mtu,
1644         .ndo_get_stats64     = xennet_get_stats64,
1645         .ndo_set_mac_address = eth_mac_addr,
1646         .ndo_validate_addr   = eth_validate_addr,
1647         .ndo_fix_features    = xennet_fix_features,
1648         .ndo_set_features    = xennet_set_features,
1649         .ndo_select_queue    = xennet_select_queue,
1650         .ndo_bpf            = xennet_xdp,
1651         .ndo_xdp_xmit       = xennet_xdp_xmit,
1652 #ifdef CONFIG_NET_POLL_CONTROLLER
1653         .ndo_poll_controller = xennet_poll_controller,
1654 #endif
1655 };
1656
1657 static void xennet_free_netdev(struct net_device *netdev)
1658 {
1659         struct netfront_info *np = netdev_priv(netdev);
1660
1661         free_percpu(np->rx_stats);
1662         free_percpu(np->tx_stats);
1663         free_netdev(netdev);
1664 }
1665
1666 static struct net_device *xennet_create_dev(struct xenbus_device *dev)
1667 {
1668         int err;
1669         struct net_device *netdev;
1670         struct netfront_info *np;
1671
1672         netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues);
1673         if (!netdev)
1674                 return ERR_PTR(-ENOMEM);
1675
1676         np                   = netdev_priv(netdev);
1677         np->xbdev            = dev;
1678
1679         np->queues = NULL;
1680
1681         err = -ENOMEM;
1682         np->rx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1683         if (np->rx_stats == NULL)
1684                 goto exit;
1685         np->tx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1686         if (np->tx_stats == NULL)
1687                 goto exit;
1688
1689         netdev->netdev_ops      = &xennet_netdev_ops;
1690
1691         netdev->features        = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1692                                   NETIF_F_GSO_ROBUST;
1693         netdev->hw_features     = NETIF_F_SG |
1694                                   NETIF_F_IPV6_CSUM |
1695                                   NETIF_F_TSO | NETIF_F_TSO6;
1696
1697         /*
1698          * Assume that all hw features are available for now. This set
1699          * will be adjusted by the call to netdev_update_features() in
1700          * xennet_connect() which is the earliest point where we can
1701          * negotiate with the backend regarding supported features.
1702          */
1703         netdev->features |= netdev->hw_features;
1704
1705         netdev->ethtool_ops = &xennet_ethtool_ops;
1706         netdev->min_mtu = ETH_MIN_MTU;
1707         netdev->max_mtu = XEN_NETIF_MAX_TX_SIZE;
1708         SET_NETDEV_DEV(netdev, &dev->dev);
1709
1710         np->netdev = netdev;
1711         np->netfront_xdp_enabled = false;
1712
1713         netif_carrier_off(netdev);
1714
1715         do {
1716                 xenbus_switch_state(dev, XenbusStateInitialising);
1717                 err = wait_event_timeout(module_wq,
1718                                  xenbus_read_driver_state(dev->otherend) !=
1719                                  XenbusStateClosed &&
1720                                  xenbus_read_driver_state(dev->otherend) !=
1721                                  XenbusStateUnknown, XENNET_TIMEOUT);
1722         } while (!err);
1723
1724         return netdev;
1725
1726  exit:
1727         xennet_free_netdev(netdev);
1728         return ERR_PTR(err);
1729 }
1730
1731 /*
1732  * Entry point to this code when a new device is created.  Allocate the basic
1733  * structures and the ring buffers for communication with the backend, and
1734  * inform the backend of the appropriate details for those.
1735  */
1736 static int netfront_probe(struct xenbus_device *dev,
1737                           const struct xenbus_device_id *id)
1738 {
1739         int err;
1740         struct net_device *netdev;
1741         struct netfront_info *info;
1742
1743         netdev = xennet_create_dev(dev);
1744         if (IS_ERR(netdev)) {
1745                 err = PTR_ERR(netdev);
1746                 xenbus_dev_fatal(dev, err, "creating netdev");
1747                 return err;
1748         }
1749
1750         info = netdev_priv(netdev);
1751         dev_set_drvdata(&dev->dev, info);
1752 #ifdef CONFIG_SYSFS
1753         info->netdev->sysfs_groups[0] = &xennet_dev_group;
1754 #endif
1755
1756         return 0;
1757 }
1758
1759 static void xennet_end_access(int ref, void *page)
1760 {
1761         /* This frees the page as a side-effect */
1762         if (ref != INVALID_GRANT_REF)
1763                 gnttab_end_foreign_access(ref, (unsigned long)page);
1764 }
1765
1766 static void xennet_disconnect_backend(struct netfront_info *info)
1767 {
1768         unsigned int i = 0;
1769         unsigned int num_queues = info->netdev->real_num_tx_queues;
1770
1771         netif_carrier_off(info->netdev);
1772
1773         for (i = 0; i < num_queues && info->queues; ++i) {
1774                 struct netfront_queue *queue = &info->queues[i];
1775
1776                 del_timer_sync(&queue->rx_refill_timer);
1777
1778                 if (queue->tx_irq && (queue->tx_irq == queue->rx_irq))
1779                         unbind_from_irqhandler(queue->tx_irq, queue);
1780                 if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) {
1781                         unbind_from_irqhandler(queue->tx_irq, queue);
1782                         unbind_from_irqhandler(queue->rx_irq, queue);
1783                 }
1784                 queue->tx_evtchn = queue->rx_evtchn = 0;
1785                 queue->tx_irq = queue->rx_irq = 0;
1786
1787                 if (netif_running(info->netdev))
1788                         napi_synchronize(&queue->napi);
1789
1790                 xennet_release_tx_bufs(queue);
1791                 xennet_release_rx_bufs(queue);
1792                 gnttab_free_grant_references(queue->gref_tx_head);
1793                 gnttab_free_grant_references(queue->gref_rx_head);
1794
1795                 /* End access and free the pages */
1796                 xennet_end_access(queue->tx_ring_ref, queue->tx.sring);
1797                 xennet_end_access(queue->rx_ring_ref, queue->rx.sring);
1798
1799                 queue->tx_ring_ref = INVALID_GRANT_REF;
1800                 queue->rx_ring_ref = INVALID_GRANT_REF;
1801                 queue->tx.sring = NULL;
1802                 queue->rx.sring = NULL;
1803
1804                 page_pool_destroy(queue->page_pool);
1805         }
1806 }
1807
1808 /*
1809  * We are reconnecting to the backend, due to a suspend/resume, or a backend
1810  * driver restart.  We tear down our netif structure and recreate it, but
1811  * leave the device-layer structures intact so that this is transparent to the
1812  * rest of the kernel.
1813  */
1814 static int netfront_resume(struct xenbus_device *dev)
1815 {
1816         struct netfront_info *info = dev_get_drvdata(&dev->dev);
1817
1818         dev_dbg(&dev->dev, "%s\n", dev->nodename);
1819
1820         netif_tx_lock_bh(info->netdev);
1821         netif_device_detach(info->netdev);
1822         netif_tx_unlock_bh(info->netdev);
1823
1824         xennet_disconnect_backend(info);
1825         return 0;
1826 }
1827
1828 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1829 {
1830         char *s, *e, *macstr;
1831         int i;
1832
1833         macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1834         if (IS_ERR(macstr))
1835                 return PTR_ERR(macstr);
1836
1837         for (i = 0; i < ETH_ALEN; i++) {
1838                 mac[i] = simple_strtoul(s, &e, 16);
1839                 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1840                         kfree(macstr);
1841                         return -ENOENT;
1842                 }
1843                 s = e+1;
1844         }
1845
1846         kfree(macstr);
1847         return 0;
1848 }
1849
1850 static int setup_netfront_single(struct netfront_queue *queue)
1851 {
1852         int err;
1853
1854         err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1855         if (err < 0)
1856                 goto fail;
1857
1858         err = bind_evtchn_to_irqhandler_lateeoi(queue->tx_evtchn,
1859                                                 xennet_interrupt, 0,
1860                                                 queue->info->netdev->name,
1861                                                 queue);
1862         if (err < 0)
1863                 goto bind_fail;
1864         queue->rx_evtchn = queue->tx_evtchn;
1865         queue->rx_irq = queue->tx_irq = err;
1866
1867         return 0;
1868
1869 bind_fail:
1870         xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1871         queue->tx_evtchn = 0;
1872 fail:
1873         return err;
1874 }
1875
1876 static int setup_netfront_split(struct netfront_queue *queue)
1877 {
1878         int err;
1879
1880         err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1881         if (err < 0)
1882                 goto fail;
1883         err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn);
1884         if (err < 0)
1885                 goto alloc_rx_evtchn_fail;
1886
1887         snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
1888                  "%s-tx", queue->name);
1889         err = bind_evtchn_to_irqhandler_lateeoi(queue->tx_evtchn,
1890                                                 xennet_tx_interrupt, 0,
1891                                                 queue->tx_irq_name, queue);
1892         if (err < 0)
1893                 goto bind_tx_fail;
1894         queue->tx_irq = err;
1895
1896         snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
1897                  "%s-rx", queue->name);
1898         err = bind_evtchn_to_irqhandler_lateeoi(queue->rx_evtchn,
1899                                                 xennet_rx_interrupt, 0,
1900                                                 queue->rx_irq_name, queue);
1901         if (err < 0)
1902                 goto bind_rx_fail;
1903         queue->rx_irq = err;
1904
1905         return 0;
1906
1907 bind_rx_fail:
1908         unbind_from_irqhandler(queue->tx_irq, queue);
1909         queue->tx_irq = 0;
1910 bind_tx_fail:
1911         xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn);
1912         queue->rx_evtchn = 0;
1913 alloc_rx_evtchn_fail:
1914         xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1915         queue->tx_evtchn = 0;
1916 fail:
1917         return err;
1918 }
1919
1920 static int setup_netfront(struct xenbus_device *dev,
1921                         struct netfront_queue *queue, unsigned int feature_split_evtchn)
1922 {
1923         struct xen_netif_tx_sring *txs;
1924         struct xen_netif_rx_sring *rxs;
1925         int err;
1926
1927         queue->tx_ring_ref = INVALID_GRANT_REF;
1928         queue->rx_ring_ref = INVALID_GRANT_REF;
1929         queue->rx.sring = NULL;
1930         queue->tx.sring = NULL;
1931
1932         err = xenbus_setup_ring(dev, GFP_NOIO | __GFP_HIGH, (void **)&txs,
1933                                 1, &queue->tx_ring_ref);
1934         if (err)
1935                 goto fail;
1936
1937         XEN_FRONT_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE);
1938
1939         err = xenbus_setup_ring(dev, GFP_NOIO | __GFP_HIGH, (void **)&rxs,
1940                                 1, &queue->rx_ring_ref);
1941         if (err)
1942                 goto fail;
1943
1944         XEN_FRONT_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE);
1945
1946         if (feature_split_evtchn)
1947                 err = setup_netfront_split(queue);
1948         /* setup single event channel if
1949          *  a) feature-split-event-channels == 0
1950          *  b) feature-split-event-channels == 1 but failed to setup
1951          */
1952         if (!feature_split_evtchn || err)
1953                 err = setup_netfront_single(queue);
1954
1955         if (err)
1956                 goto fail;
1957
1958         return 0;
1959
1960  fail:
1961         xenbus_teardown_ring((void **)&queue->rx.sring, 1, &queue->rx_ring_ref);
1962         xenbus_teardown_ring((void **)&queue->tx.sring, 1, &queue->tx_ring_ref);
1963
1964         return err;
1965 }
1966
1967 /* Queue-specific initialisation
1968  * This used to be done in xennet_create_dev() but must now
1969  * be run per-queue.
1970  */
1971 static int xennet_init_queue(struct netfront_queue *queue)
1972 {
1973         unsigned short i;
1974         int err = 0;
1975         char *devid;
1976
1977         spin_lock_init(&queue->tx_lock);
1978         spin_lock_init(&queue->rx_lock);
1979         spin_lock_init(&queue->rx_cons_lock);
1980
1981         timer_setup(&queue->rx_refill_timer, rx_refill_timeout, 0);
1982
1983         devid = strrchr(queue->info->xbdev->nodename, '/') + 1;
1984         snprintf(queue->name, sizeof(queue->name), "vif%s-q%u",
1985                  devid, queue->id);
1986
1987         /* Initialise tx_skb_freelist as a free chain containing every entry. */
1988         queue->tx_skb_freelist = 0;
1989         queue->tx_pend_queue = TX_LINK_NONE;
1990         for (i = 0; i < NET_TX_RING_SIZE; i++) {
1991                 queue->tx_link[i] = i + 1;
1992                 queue->grant_tx_ref[i] = INVALID_GRANT_REF;
1993                 queue->grant_tx_page[i] = NULL;
1994         }
1995         queue->tx_link[NET_TX_RING_SIZE - 1] = TX_LINK_NONE;
1996
1997         /* Clear out rx_skbs */
1998         for (i = 0; i < NET_RX_RING_SIZE; i++) {
1999                 queue->rx_skbs[i] = NULL;
2000                 queue->grant_rx_ref[i] = INVALID_GRANT_REF;
2001         }
2002
2003         /* A grant for every tx ring slot */
2004         if (gnttab_alloc_grant_references(NET_TX_RING_SIZE,
2005                                           &queue->gref_tx_head) < 0) {
2006                 pr_alert("can't alloc tx grant refs\n");
2007                 err = -ENOMEM;
2008                 goto exit;
2009         }
2010
2011         /* A grant for every rx ring slot */
2012         if (gnttab_alloc_grant_references(NET_RX_RING_SIZE,
2013                                           &queue->gref_rx_head) < 0) {
2014                 pr_alert("can't alloc rx grant refs\n");
2015                 err = -ENOMEM;
2016                 goto exit_free_tx;
2017         }
2018
2019         return 0;
2020
2021  exit_free_tx:
2022         gnttab_free_grant_references(queue->gref_tx_head);
2023  exit:
2024         return err;
2025 }
2026
2027 static int write_queue_xenstore_keys(struct netfront_queue *queue,
2028                            struct xenbus_transaction *xbt, int write_hierarchical)
2029 {
2030         /* Write the queue-specific keys into XenStore in the traditional
2031          * way for a single queue, or in a queue subkeys for multiple
2032          * queues.
2033          */
2034         struct xenbus_device *dev = queue->info->xbdev;
2035         int err;
2036         const char *message;
2037         char *path;
2038         size_t pathsize;
2039
2040         /* Choose the correct place to write the keys */
2041         if (write_hierarchical) {
2042                 pathsize = strlen(dev->nodename) + 10;
2043                 path = kzalloc(pathsize, GFP_KERNEL);
2044                 if (!path) {
2045                         err = -ENOMEM;
2046                         message = "out of memory while writing ring references";
2047                         goto error;
2048                 }
2049                 snprintf(path, pathsize, "%s/queue-%u",
2050                                 dev->nodename, queue->id);
2051         } else {
2052                 path = (char *)dev->nodename;
2053         }
2054
2055         /* Write ring references */
2056         err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u",
2057                         queue->tx_ring_ref);
2058         if (err) {
2059                 message = "writing tx-ring-ref";
2060                 goto error;
2061         }
2062
2063         err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u",
2064                         queue->rx_ring_ref);
2065         if (err) {
2066                 message = "writing rx-ring-ref";
2067                 goto error;
2068         }
2069
2070         /* Write event channels; taking into account both shared
2071          * and split event channel scenarios.
2072          */
2073         if (queue->tx_evtchn == queue->rx_evtchn) {
2074                 /* Shared event channel */
2075                 err = xenbus_printf(*xbt, path,
2076                                 "event-channel", "%u", queue->tx_evtchn);
2077                 if (err) {
2078                         message = "writing event-channel";
2079                         goto error;
2080                 }
2081         } else {
2082                 /* Split event channels */
2083                 err = xenbus_printf(*xbt, path,
2084                                 "event-channel-tx", "%u", queue->tx_evtchn);
2085                 if (err) {
2086                         message = "writing event-channel-tx";
2087                         goto error;
2088                 }
2089
2090                 err = xenbus_printf(*xbt, path,
2091                                 "event-channel-rx", "%u", queue->rx_evtchn);
2092                 if (err) {
2093                         message = "writing event-channel-rx";
2094                         goto error;
2095                 }
2096         }
2097
2098         if (write_hierarchical)
2099                 kfree(path);
2100         return 0;
2101
2102 error:
2103         if (write_hierarchical)
2104                 kfree(path);
2105         xenbus_dev_fatal(dev, err, "%s", message);
2106         return err;
2107 }
2108
2109
2110
2111 static int xennet_create_page_pool(struct netfront_queue *queue)
2112 {
2113         int err;
2114         struct page_pool_params pp_params = {
2115                 .order = 0,
2116                 .flags = 0,
2117                 .pool_size = NET_RX_RING_SIZE,
2118                 .nid = NUMA_NO_NODE,
2119                 .dev = &queue->info->netdev->dev,
2120                 .offset = XDP_PACKET_HEADROOM,
2121                 .max_len = XEN_PAGE_SIZE - XDP_PACKET_HEADROOM,
2122         };
2123
2124         queue->page_pool = page_pool_create(&pp_params);
2125         if (IS_ERR(queue->page_pool)) {
2126                 err = PTR_ERR(queue->page_pool);
2127                 queue->page_pool = NULL;
2128                 return err;
2129         }
2130
2131         err = xdp_rxq_info_reg(&queue->xdp_rxq, queue->info->netdev,
2132                                queue->id, 0);
2133         if (err) {
2134                 netdev_err(queue->info->netdev, "xdp_rxq_info_reg failed\n");
2135                 goto err_free_pp;
2136         }
2137
2138         err = xdp_rxq_info_reg_mem_model(&queue->xdp_rxq,
2139                                          MEM_TYPE_PAGE_POOL, queue->page_pool);
2140         if (err) {
2141                 netdev_err(queue->info->netdev, "xdp_rxq_info_reg_mem_model failed\n");
2142                 goto err_unregister_rxq;
2143         }
2144         return 0;
2145
2146 err_unregister_rxq:
2147         xdp_rxq_info_unreg(&queue->xdp_rxq);
2148 err_free_pp:
2149         page_pool_destroy(queue->page_pool);
2150         queue->page_pool = NULL;
2151         return err;
2152 }
2153
2154 static int xennet_create_queues(struct netfront_info *info,
2155                                 unsigned int *num_queues)
2156 {
2157         unsigned int i;
2158         int ret;
2159
2160         info->queues = kcalloc(*num_queues, sizeof(struct netfront_queue),
2161                                GFP_KERNEL);
2162         if (!info->queues)
2163                 return -ENOMEM;
2164
2165         for (i = 0; i < *num_queues; i++) {
2166                 struct netfront_queue *queue = &info->queues[i];
2167
2168                 queue->id = i;
2169                 queue->info = info;
2170
2171                 ret = xennet_init_queue(queue);
2172                 if (ret < 0) {
2173                         dev_warn(&info->xbdev->dev,
2174                                  "only created %d queues\n", i);
2175                         *num_queues = i;
2176                         break;
2177                 }
2178
2179                 /* use page pool recycling instead of buddy allocator */
2180                 ret = xennet_create_page_pool(queue);
2181                 if (ret < 0) {
2182                         dev_err(&info->xbdev->dev, "can't allocate page pool\n");
2183                         *num_queues = i;
2184                         return ret;
2185                 }
2186
2187                 netif_napi_add(queue->info->netdev, &queue->napi,
2188                                xennet_poll, 64);
2189                 if (netif_running(info->netdev))
2190                         napi_enable(&queue->napi);
2191         }
2192
2193         netif_set_real_num_tx_queues(info->netdev, *num_queues);
2194
2195         if (*num_queues == 0) {
2196                 dev_err(&info->xbdev->dev, "no queues\n");
2197                 return -EINVAL;
2198         }
2199         return 0;
2200 }
2201
2202 /* Common code used when first setting up, and when resuming. */
2203 static int talk_to_netback(struct xenbus_device *dev,
2204                            struct netfront_info *info)
2205 {
2206         const char *message;
2207         struct xenbus_transaction xbt;
2208         int err;
2209         unsigned int feature_split_evtchn;
2210         unsigned int i = 0;
2211         unsigned int max_queues = 0;
2212         struct netfront_queue *queue = NULL;
2213         unsigned int num_queues = 1;
2214         u8 addr[ETH_ALEN];
2215
2216         info->netdev->irq = 0;
2217
2218         /* Check if backend supports multiple queues */
2219         max_queues = xenbus_read_unsigned(info->xbdev->otherend,
2220                                           "multi-queue-max-queues", 1);
2221         num_queues = min(max_queues, xennet_max_queues);
2222
2223         /* Check feature-split-event-channels */
2224         feature_split_evtchn = xenbus_read_unsigned(info->xbdev->otherend,
2225                                         "feature-split-event-channels", 0);
2226
2227         /* Read mac addr. */
2228         err = xen_net_read_mac(dev, addr);
2229         if (err) {
2230                 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
2231                 goto out_unlocked;
2232         }
2233         eth_hw_addr_set(info->netdev, addr);
2234
2235         info->netback_has_xdp_headroom = xenbus_read_unsigned(info->xbdev->otherend,
2236                                                               "feature-xdp-headroom", 0);
2237         if (info->netback_has_xdp_headroom) {
2238                 /* set the current xen-netfront xdp state */
2239                 err = talk_to_netback_xdp(info, info->netfront_xdp_enabled ?
2240                                           NETBACK_XDP_HEADROOM_ENABLE :
2241                                           NETBACK_XDP_HEADROOM_DISABLE);
2242                 if (err)
2243                         goto out_unlocked;
2244         }
2245
2246         rtnl_lock();
2247         if (info->queues)
2248                 xennet_destroy_queues(info);
2249
2250         /* For the case of a reconnect reset the "broken" indicator. */
2251         info->broken = false;
2252
2253         err = xennet_create_queues(info, &num_queues);
2254         if (err < 0) {
2255                 xenbus_dev_fatal(dev, err, "creating queues");
2256                 kfree(info->queues);
2257                 info->queues = NULL;
2258                 goto out;
2259         }
2260         rtnl_unlock();
2261
2262         /* Create shared ring, alloc event channel -- for each queue */
2263         for (i = 0; i < num_queues; ++i) {
2264                 queue = &info->queues[i];
2265                 err = setup_netfront(dev, queue, feature_split_evtchn);
2266                 if (err)
2267                         goto destroy_ring;
2268         }
2269
2270 again:
2271         err = xenbus_transaction_start(&xbt);
2272         if (err) {
2273                 xenbus_dev_fatal(dev, err, "starting transaction");
2274                 goto destroy_ring;
2275         }
2276
2277         if (xenbus_exists(XBT_NIL,
2278                           info->xbdev->otherend, "multi-queue-max-queues")) {
2279                 /* Write the number of queues */
2280                 err = xenbus_printf(xbt, dev->nodename,
2281                                     "multi-queue-num-queues", "%u", num_queues);
2282                 if (err) {
2283                         message = "writing multi-queue-num-queues";
2284                         goto abort_transaction_no_dev_fatal;
2285                 }
2286         }
2287
2288         if (num_queues == 1) {
2289                 err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */
2290                 if (err)
2291                         goto abort_transaction_no_dev_fatal;
2292         } else {
2293                 /* Write the keys for each queue */
2294                 for (i = 0; i < num_queues; ++i) {
2295                         queue = &info->queues[i];
2296                         err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */
2297                         if (err)
2298                                 goto abort_transaction_no_dev_fatal;
2299                 }
2300         }
2301
2302         /* The remaining keys are not queue-specific */
2303         err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
2304                             1);
2305         if (err) {
2306                 message = "writing request-rx-copy";
2307                 goto abort_transaction;
2308         }
2309
2310         err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
2311         if (err) {
2312                 message = "writing feature-rx-notify";
2313                 goto abort_transaction;
2314         }
2315
2316         err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
2317         if (err) {
2318                 message = "writing feature-sg";
2319                 goto abort_transaction;
2320         }
2321
2322         err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
2323         if (err) {
2324                 message = "writing feature-gso-tcpv4";
2325                 goto abort_transaction;
2326         }
2327
2328         err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1");
2329         if (err) {
2330                 message = "writing feature-gso-tcpv6";
2331                 goto abort_transaction;
2332         }
2333
2334         err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload",
2335                            "1");
2336         if (err) {
2337                 message = "writing feature-ipv6-csum-offload";
2338                 goto abort_transaction;
2339         }
2340
2341         err = xenbus_transaction_end(xbt, 0);
2342         if (err) {
2343                 if (err == -EAGAIN)
2344                         goto again;
2345                 xenbus_dev_fatal(dev, err, "completing transaction");
2346                 goto destroy_ring;
2347         }
2348
2349         return 0;
2350
2351  abort_transaction:
2352         xenbus_dev_fatal(dev, err, "%s", message);
2353 abort_transaction_no_dev_fatal:
2354         xenbus_transaction_end(xbt, 1);
2355  destroy_ring:
2356         xennet_disconnect_backend(info);
2357         rtnl_lock();
2358         xennet_destroy_queues(info);
2359  out:
2360         rtnl_unlock();
2361 out_unlocked:
2362         device_unregister(&dev->dev);
2363         return err;
2364 }
2365
2366 static int xennet_connect(struct net_device *dev)
2367 {
2368         struct netfront_info *np = netdev_priv(dev);
2369         unsigned int num_queues = 0;
2370         int err;
2371         unsigned int j = 0;
2372         struct netfront_queue *queue = NULL;
2373
2374         if (!xenbus_read_unsigned(np->xbdev->otherend, "feature-rx-copy", 0)) {
2375                 dev_info(&dev->dev,
2376                          "backend does not support copying receive path\n");
2377                 return -ENODEV;
2378         }
2379
2380         err = talk_to_netback(np->xbdev, np);
2381         if (err)
2382                 return err;
2383         if (np->netback_has_xdp_headroom)
2384                 pr_info("backend supports XDP headroom\n");
2385
2386         /* talk_to_netback() sets the correct number of queues */
2387         num_queues = dev->real_num_tx_queues;
2388
2389         if (dev->reg_state == NETREG_UNINITIALIZED) {
2390                 err = register_netdev(dev);
2391                 if (err) {
2392                         pr_warn("%s: register_netdev err=%d\n", __func__, err);
2393                         device_unregister(&np->xbdev->dev);
2394                         return err;
2395                 }
2396         }
2397
2398         rtnl_lock();
2399         netdev_update_features(dev);
2400         rtnl_unlock();
2401
2402         /*
2403          * All public and private state should now be sane.  Get
2404          * ready to start sending and receiving packets and give the driver
2405          * domain a kick because we've probably just requeued some
2406          * packets.
2407          */
2408         netif_tx_lock_bh(np->netdev);
2409         netif_device_attach(np->netdev);
2410         netif_tx_unlock_bh(np->netdev);
2411
2412         netif_carrier_on(np->netdev);
2413         for (j = 0; j < num_queues; ++j) {
2414                 queue = &np->queues[j];
2415
2416                 notify_remote_via_irq(queue->tx_irq);
2417                 if (queue->tx_irq != queue->rx_irq)
2418                         notify_remote_via_irq(queue->rx_irq);
2419
2420                 spin_lock_irq(&queue->tx_lock);
2421                 xennet_tx_buf_gc(queue);
2422                 spin_unlock_irq(&queue->tx_lock);
2423
2424                 spin_lock_bh(&queue->rx_lock);
2425                 xennet_alloc_rx_buffers(queue);
2426                 spin_unlock_bh(&queue->rx_lock);
2427         }
2428
2429         return 0;
2430 }
2431
2432 /*
2433  * Callback received when the backend's state changes.
2434  */
2435 static void netback_changed(struct xenbus_device *dev,
2436                             enum xenbus_state backend_state)
2437 {
2438         struct netfront_info *np = dev_get_drvdata(&dev->dev);
2439         struct net_device *netdev = np->netdev;
2440
2441         dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
2442
2443         wake_up_all(&module_wq);
2444
2445         switch (backend_state) {
2446         case XenbusStateInitialising:
2447         case XenbusStateInitialised:
2448         case XenbusStateReconfiguring:
2449         case XenbusStateReconfigured:
2450         case XenbusStateUnknown:
2451                 break;
2452
2453         case XenbusStateInitWait:
2454                 if (dev->state != XenbusStateInitialising)
2455                         break;
2456                 if (xennet_connect(netdev) != 0)
2457                         break;
2458                 xenbus_switch_state(dev, XenbusStateConnected);
2459                 break;
2460
2461         case XenbusStateConnected:
2462                 netdev_notify_peers(netdev);
2463                 break;
2464
2465         case XenbusStateClosed:
2466                 if (dev->state == XenbusStateClosed)
2467                         break;
2468                 fallthrough;    /* Missed the backend's CLOSING state */
2469         case XenbusStateClosing:
2470                 xenbus_frontend_closed(dev);
2471                 break;
2472         }
2473 }
2474
2475 static const struct xennet_stat {
2476         char name[ETH_GSTRING_LEN];
2477         u16 offset;
2478 } xennet_stats[] = {
2479         {
2480                 "rx_gso_checksum_fixup",
2481                 offsetof(struct netfront_info, rx_gso_checksum_fixup)
2482         },
2483 };
2484
2485 static int xennet_get_sset_count(struct net_device *dev, int string_set)
2486 {
2487         switch (string_set) {
2488         case ETH_SS_STATS:
2489                 return ARRAY_SIZE(xennet_stats);
2490         default:
2491                 return -EINVAL;
2492         }
2493 }
2494
2495 static void xennet_get_ethtool_stats(struct net_device *dev,
2496                                      struct ethtool_stats *stats, u64 * data)
2497 {
2498         void *np = netdev_priv(dev);
2499         int i;
2500
2501         for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2502                 data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset));
2503 }
2504
2505 static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
2506 {
2507         int i;
2508
2509         switch (stringset) {
2510         case ETH_SS_STATS:
2511                 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2512                         memcpy(data + i * ETH_GSTRING_LEN,
2513                                xennet_stats[i].name, ETH_GSTRING_LEN);
2514                 break;
2515         }
2516 }
2517
2518 static const struct ethtool_ops xennet_ethtool_ops =
2519 {
2520         .get_link = ethtool_op_get_link,
2521
2522         .get_sset_count = xennet_get_sset_count,
2523         .get_ethtool_stats = xennet_get_ethtool_stats,
2524         .get_strings = xennet_get_strings,
2525         .get_ts_info = ethtool_op_get_ts_info,
2526 };
2527
2528 #ifdef CONFIG_SYSFS
2529 static ssize_t show_rxbuf(struct device *dev,
2530                           struct device_attribute *attr, char *buf)
2531 {
2532         return sprintf(buf, "%lu\n", NET_RX_RING_SIZE);
2533 }
2534
2535 static ssize_t store_rxbuf(struct device *dev,
2536                            struct device_attribute *attr,
2537                            const char *buf, size_t len)
2538 {
2539         char *endp;
2540
2541         if (!capable(CAP_NET_ADMIN))
2542                 return -EPERM;
2543
2544         simple_strtoul(buf, &endp, 0);
2545         if (endp == buf)
2546                 return -EBADMSG;
2547
2548         /* rxbuf_min and rxbuf_max are no longer configurable. */
2549
2550         return len;
2551 }
2552
2553 static DEVICE_ATTR(rxbuf_min, 0644, show_rxbuf, store_rxbuf);
2554 static DEVICE_ATTR(rxbuf_max, 0644, show_rxbuf, store_rxbuf);
2555 static DEVICE_ATTR(rxbuf_cur, 0444, show_rxbuf, NULL);
2556
2557 static struct attribute *xennet_dev_attrs[] = {
2558         &dev_attr_rxbuf_min.attr,
2559         &dev_attr_rxbuf_max.attr,
2560         &dev_attr_rxbuf_cur.attr,
2561         NULL
2562 };
2563
2564 static const struct attribute_group xennet_dev_group = {
2565         .attrs = xennet_dev_attrs
2566 };
2567 #endif /* CONFIG_SYSFS */
2568
2569 static void xennet_bus_close(struct xenbus_device *dev)
2570 {
2571         int ret;
2572
2573         if (xenbus_read_driver_state(dev->otherend) == XenbusStateClosed)
2574                 return;
2575         do {
2576                 xenbus_switch_state(dev, XenbusStateClosing);
2577                 ret = wait_event_timeout(module_wq,
2578                                    xenbus_read_driver_state(dev->otherend) ==
2579                                    XenbusStateClosing ||
2580                                    xenbus_read_driver_state(dev->otherend) ==
2581                                    XenbusStateClosed ||
2582                                    xenbus_read_driver_state(dev->otherend) ==
2583                                    XenbusStateUnknown,
2584                                    XENNET_TIMEOUT);
2585         } while (!ret);
2586
2587         if (xenbus_read_driver_state(dev->otherend) == XenbusStateClosed)
2588                 return;
2589
2590         do {
2591                 xenbus_switch_state(dev, XenbusStateClosed);
2592                 ret = wait_event_timeout(module_wq,
2593                                    xenbus_read_driver_state(dev->otherend) ==
2594                                    XenbusStateClosed ||
2595                                    xenbus_read_driver_state(dev->otherend) ==
2596                                    XenbusStateUnknown,
2597                                    XENNET_TIMEOUT);
2598         } while (!ret);
2599 }
2600
2601 static int xennet_remove(struct xenbus_device *dev)
2602 {
2603         struct netfront_info *info = dev_get_drvdata(&dev->dev);
2604
2605         xennet_bus_close(dev);
2606         xennet_disconnect_backend(info);
2607
2608         if (info->netdev->reg_state == NETREG_REGISTERED)
2609                 unregister_netdev(info->netdev);
2610
2611         if (info->queues) {
2612                 rtnl_lock();
2613                 xennet_destroy_queues(info);
2614                 rtnl_unlock();
2615         }
2616         xennet_free_netdev(info->netdev);
2617
2618         return 0;
2619 }
2620
2621 static const struct xenbus_device_id netfront_ids[] = {
2622         { "vif" },
2623         { "" }
2624 };
2625
2626 static struct xenbus_driver netfront_driver = {
2627         .ids = netfront_ids,
2628         .probe = netfront_probe,
2629         .remove = xennet_remove,
2630         .resume = netfront_resume,
2631         .otherend_changed = netback_changed,
2632 };
2633
2634 static int __init netif_init(void)
2635 {
2636         if (!xen_domain())
2637                 return -ENODEV;
2638
2639         if (!xen_has_pv_nic_devices())
2640                 return -ENODEV;
2641
2642         pr_info("Initialising Xen virtual ethernet driver\n");
2643
2644         /* Allow as many queues as there are CPUs inut max. 8 if user has not
2645          * specified a value.
2646          */
2647         if (xennet_max_queues == 0)
2648                 xennet_max_queues = min_t(unsigned int, MAX_QUEUES_DEFAULT,
2649                                           num_online_cpus());
2650
2651         return xenbus_register_frontend(&netfront_driver);
2652 }
2653 module_init(netif_init);
2654
2655
2656 static void __exit netif_exit(void)
2657 {
2658         xenbus_unregister_driver(&netfront_driver);
2659 }
2660 module_exit(netif_exit);
2661
2662 MODULE_DESCRIPTION("Xen virtual network device frontend");
2663 MODULE_LICENSE("GPL");
2664 MODULE_ALIAS("xen:vif");
2665 MODULE_ALIAS("xennet");