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