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