Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux
[muen/linux.git] / 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_unload_q);
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 - MAX_SKB_FRAGS - 2);
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         wmb();          /* barrier so backend seens requests */
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         for (i = 0; i < num_queues; ++i) {
355                 queue = &np->queues[i];
356                 napi_enable(&queue->napi);
357
358                 spin_lock_bh(&queue->rx_lock);
359                 if (netif_carrier_ok(dev)) {
360                         xennet_alloc_rx_buffers(queue);
361                         queue->rx.sring->rsp_event = queue->rx.rsp_cons + 1;
362                         if (RING_HAS_UNCONSUMED_RESPONSES(&queue->rx))
363                                 napi_schedule(&queue->napi);
364                 }
365                 spin_unlock_bh(&queue->rx_lock);
366         }
367
368         netif_tx_start_all_queues(dev);
369
370         return 0;
371 }
372
373 static void xennet_tx_buf_gc(struct netfront_queue *queue)
374 {
375         RING_IDX cons, prod;
376         unsigned short id;
377         struct sk_buff *skb;
378         bool more_to_do;
379
380         BUG_ON(!netif_carrier_ok(queue->info->netdev));
381
382         do {
383                 prod = queue->tx.sring->rsp_prod;
384                 rmb(); /* Ensure we see responses up to 'rp'. */
385
386                 for (cons = queue->tx.rsp_cons; cons != prod; cons++) {
387                         struct xen_netif_tx_response *txrsp;
388
389                         txrsp = RING_GET_RESPONSE(&queue->tx, cons);
390                         if (txrsp->status == XEN_NETIF_RSP_NULL)
391                                 continue;
392
393                         id  = txrsp->id;
394                         skb = queue->tx_skbs[id].skb;
395                         if (unlikely(gnttab_query_foreign_access(
396                                 queue->grant_tx_ref[id]) != 0)) {
397                                 pr_alert("%s: warning -- grant still in use by backend domain\n",
398                                          __func__);
399                                 BUG();
400                         }
401                         gnttab_end_foreign_access_ref(
402                                 queue->grant_tx_ref[id], GNTMAP_readonly);
403                         gnttab_release_grant_reference(
404                                 &queue->gref_tx_head, queue->grant_tx_ref[id]);
405                         queue->grant_tx_ref[id] = GRANT_INVALID_REF;
406                         queue->grant_tx_page[id] = NULL;
407                         add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, id);
408                         dev_kfree_skb_irq(skb);
409                 }
410
411                 queue->tx.rsp_cons = prod;
412
413                 RING_FINAL_CHECK_FOR_RESPONSES(&queue->tx, more_to_do);
414         } while (more_to_do);
415
416         xennet_maybe_wake_tx(queue);
417 }
418
419 struct xennet_gnttab_make_txreq {
420         struct netfront_queue *queue;
421         struct sk_buff *skb;
422         struct page *page;
423         struct xen_netif_tx_request *tx; /* Last request */
424         unsigned int size;
425 };
426
427 static void xennet_tx_setup_grant(unsigned long gfn, unsigned int offset,
428                                   unsigned int len, void *data)
429 {
430         struct xennet_gnttab_make_txreq *info = data;
431         unsigned int id;
432         struct xen_netif_tx_request *tx;
433         grant_ref_t ref;
434         /* convenient aliases */
435         struct page *page = info->page;
436         struct netfront_queue *queue = info->queue;
437         struct sk_buff *skb = info->skb;
438
439         id = get_id_from_freelist(&queue->tx_skb_freelist, queue->tx_skbs);
440         tx = RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
441         ref = gnttab_claim_grant_reference(&queue->gref_tx_head);
442         WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
443
444         gnttab_grant_foreign_access_ref(ref, queue->info->xbdev->otherend_id,
445                                         gfn, GNTMAP_readonly);
446
447         queue->tx_skbs[id].skb = skb;
448         queue->grant_tx_page[id] = page;
449         queue->grant_tx_ref[id] = ref;
450
451         tx->id = id;
452         tx->gref = ref;
453         tx->offset = offset;
454         tx->size = len;
455         tx->flags = 0;
456
457         info->tx = tx;
458         info->size += tx->size;
459 }
460
461 static struct xen_netif_tx_request *xennet_make_first_txreq(
462         struct netfront_queue *queue, struct sk_buff *skb,
463         struct page *page, unsigned int offset, unsigned int len)
464 {
465         struct xennet_gnttab_make_txreq info = {
466                 .queue = queue,
467                 .skb = skb,
468                 .page = page,
469                 .size = 0,
470         };
471
472         gnttab_for_one_grant(page, offset, len, xennet_tx_setup_grant, &info);
473
474         return info.tx;
475 }
476
477 static void xennet_make_one_txreq(unsigned long gfn, unsigned int offset,
478                                   unsigned int len, void *data)
479 {
480         struct xennet_gnttab_make_txreq *info = data;
481
482         info->tx->flags |= XEN_NETTXF_more_data;
483         skb_get(info->skb);
484         xennet_tx_setup_grant(gfn, offset, len, data);
485 }
486
487 static struct xen_netif_tx_request *xennet_make_txreqs(
488         struct netfront_queue *queue, struct xen_netif_tx_request *tx,
489         struct sk_buff *skb, struct page *page,
490         unsigned int offset, unsigned int len)
491 {
492         struct xennet_gnttab_make_txreq info = {
493                 .queue = queue,
494                 .skb = skb,
495                 .tx = tx,
496         };
497
498         /* Skip unused frames from start of page */
499         page += offset >> PAGE_SHIFT;
500         offset &= ~PAGE_MASK;
501
502         while (len) {
503                 info.page = page;
504                 info.size = 0;
505
506                 gnttab_foreach_grant_in_range(page, offset, len,
507                                               xennet_make_one_txreq,
508                                               &info);
509
510                 page++;
511                 offset = 0;
512                 len -= info.size;
513         }
514
515         return info.tx;
516 }
517
518 /*
519  * Count how many ring slots are required to send this skb. Each frag
520  * might be a compound page.
521  */
522 static int xennet_count_skb_slots(struct sk_buff *skb)
523 {
524         int i, frags = skb_shinfo(skb)->nr_frags;
525         int slots;
526
527         slots = gnttab_count_grant(offset_in_page(skb->data),
528                                    skb_headlen(skb));
529
530         for (i = 0; i < frags; i++) {
531                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
532                 unsigned long size = skb_frag_size(frag);
533                 unsigned long offset = frag->page_offset;
534
535                 /* Skip unused frames from start of page */
536                 offset &= ~PAGE_MASK;
537
538                 slots += gnttab_count_grant(offset, size);
539         }
540
541         return slots;
542 }
543
544 static u16 xennet_select_queue(struct net_device *dev, struct sk_buff *skb,
545                                void *accel_priv, select_queue_fallback_t fallback)
546 {
547         unsigned int num_queues = dev->real_num_tx_queues;
548         u32 hash;
549         u16 queue_idx;
550
551         /* First, check if there is only one queue */
552         if (num_queues == 1) {
553                 queue_idx = 0;
554         } else {
555                 hash = skb_get_hash(skb);
556                 queue_idx = hash % num_queues;
557         }
558
559         return queue_idx;
560 }
561
562 #define MAX_XEN_SKB_FRAGS (65536 / XEN_PAGE_SIZE + 1)
563
564 static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
565 {
566         struct netfront_info *np = netdev_priv(dev);
567         struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats);
568         struct xen_netif_tx_request *tx, *first_tx;
569         unsigned int i;
570         int notify;
571         int slots;
572         struct page *page;
573         unsigned int offset;
574         unsigned int len;
575         unsigned long flags;
576         struct netfront_queue *queue = NULL;
577         unsigned int num_queues = dev->real_num_tx_queues;
578         u16 queue_index;
579         struct sk_buff *nskb;
580
581         /* Drop the packet if no queues are set up */
582         if (num_queues < 1)
583                 goto drop;
584         /* Determine which queue to transmit this SKB on */
585         queue_index = skb_get_queue_mapping(skb);
586         queue = &np->queues[queue_index];
587
588         /* If skb->len is too big for wire format, drop skb and alert
589          * user about misconfiguration.
590          */
591         if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) {
592                 net_alert_ratelimited(
593                         "xennet: skb->len = %u, too big for wire format\n",
594                         skb->len);
595                 goto drop;
596         }
597
598         slots = xennet_count_skb_slots(skb);
599         if (unlikely(slots > MAX_XEN_SKB_FRAGS + 1)) {
600                 net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n",
601                                     slots, skb->len);
602                 if (skb_linearize(skb))
603                         goto drop;
604         }
605
606         page = virt_to_page(skb->data);
607         offset = offset_in_page(skb->data);
608
609         /* The first req should be at least ETH_HLEN size or the packet will be
610          * dropped by netback.
611          */
612         if (unlikely(PAGE_SIZE - offset < ETH_HLEN)) {
613                 nskb = skb_copy(skb, GFP_ATOMIC);
614                 if (!nskb)
615                         goto drop;
616                 dev_consume_skb_any(skb);
617                 skb = nskb;
618                 page = virt_to_page(skb->data);
619                 offset = offset_in_page(skb->data);
620         }
621
622         len = skb_headlen(skb);
623
624         spin_lock_irqsave(&queue->tx_lock, flags);
625
626         if (unlikely(!netif_carrier_ok(dev) ||
627                      (slots > 1 && !xennet_can_sg(dev)) ||
628                      netif_needs_gso(skb, netif_skb_features(skb)))) {
629                 spin_unlock_irqrestore(&queue->tx_lock, flags);
630                 goto drop;
631         }
632
633         /* First request for the linear area. */
634         first_tx = tx = xennet_make_first_txreq(queue, skb,
635                                                 page, offset, len);
636         offset += tx->size;
637         if (offset == PAGE_SIZE) {
638                 page++;
639                 offset = 0;
640         }
641         len -= tx->size;
642
643         if (skb->ip_summed == CHECKSUM_PARTIAL)
644                 /* local packet? */
645                 tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated;
646         else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
647                 /* remote but checksummed. */
648                 tx->flags |= XEN_NETTXF_data_validated;
649
650         /* Optional extra info after the first request. */
651         if (skb_shinfo(skb)->gso_size) {
652                 struct xen_netif_extra_info *gso;
653
654                 gso = (struct xen_netif_extra_info *)
655                         RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
656
657                 tx->flags |= XEN_NETTXF_extra_info;
658
659                 gso->u.gso.size = skb_shinfo(skb)->gso_size;
660                 gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ?
661                         XEN_NETIF_GSO_TYPE_TCPV6 :
662                         XEN_NETIF_GSO_TYPE_TCPV4;
663                 gso->u.gso.pad = 0;
664                 gso->u.gso.features = 0;
665
666                 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
667                 gso->flags = 0;
668         }
669
670         /* Requests for the rest of the linear area. */
671         tx = xennet_make_txreqs(queue, tx, skb, page, offset, len);
672
673         /* Requests for all the frags. */
674         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
675                 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
676                 tx = xennet_make_txreqs(queue, tx, skb,
677                                         skb_frag_page(frag), frag->page_offset,
678                                         skb_frag_size(frag));
679         }
680
681         /* First request has the packet length. */
682         first_tx->size = skb->len;
683
684         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
685         if (notify)
686                 notify_remote_via_irq(queue->tx_irq);
687
688         u64_stats_update_begin(&tx_stats->syncp);
689         tx_stats->bytes += skb->len;
690         tx_stats->packets++;
691         u64_stats_update_end(&tx_stats->syncp);
692
693         /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
694         xennet_tx_buf_gc(queue);
695
696         if (!netfront_tx_slot_available(queue))
697                 netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id));
698
699         spin_unlock_irqrestore(&queue->tx_lock, flags);
700
701         return NETDEV_TX_OK;
702
703  drop:
704         dev->stats.tx_dropped++;
705         dev_kfree_skb_any(skb);
706         return NETDEV_TX_OK;
707 }
708
709 static int xennet_close(struct net_device *dev)
710 {
711         struct netfront_info *np = netdev_priv(dev);
712         unsigned int num_queues = dev->real_num_tx_queues;
713         unsigned int i;
714         struct netfront_queue *queue;
715         netif_tx_stop_all_queues(np->netdev);
716         for (i = 0; i < num_queues; ++i) {
717                 queue = &np->queues[i];
718                 napi_disable(&queue->napi);
719         }
720         return 0;
721 }
722
723 static void xennet_move_rx_slot(struct netfront_queue *queue, struct sk_buff *skb,
724                                 grant_ref_t ref)
725 {
726         int new = xennet_rxidx(queue->rx.req_prod_pvt);
727
728         BUG_ON(queue->rx_skbs[new]);
729         queue->rx_skbs[new] = skb;
730         queue->grant_rx_ref[new] = ref;
731         RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->id = new;
732         RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->gref = ref;
733         queue->rx.req_prod_pvt++;
734 }
735
736 static int xennet_get_extras(struct netfront_queue *queue,
737                              struct xen_netif_extra_info *extras,
738                              RING_IDX rp)
739
740 {
741         struct xen_netif_extra_info *extra;
742         struct device *dev = &queue->info->netdev->dev;
743         RING_IDX cons = queue->rx.rsp_cons;
744         int err = 0;
745
746         do {
747                 struct sk_buff *skb;
748                 grant_ref_t ref;
749
750                 if (unlikely(cons + 1 == rp)) {
751                         if (net_ratelimit())
752                                 dev_warn(dev, "Missing extra info\n");
753                         err = -EBADR;
754                         break;
755                 }
756
757                 extra = (struct xen_netif_extra_info *)
758                         RING_GET_RESPONSE(&queue->rx, ++cons);
759
760                 if (unlikely(!extra->type ||
761                              extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
762                         if (net_ratelimit())
763                                 dev_warn(dev, "Invalid extra type: %d\n",
764                                         extra->type);
765                         err = -EINVAL;
766                 } else {
767                         memcpy(&extras[extra->type - 1], extra,
768                                sizeof(*extra));
769                 }
770
771                 skb = xennet_get_rx_skb(queue, cons);
772                 ref = xennet_get_rx_ref(queue, cons);
773                 xennet_move_rx_slot(queue, skb, ref);
774         } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
775
776         queue->rx.rsp_cons = cons;
777         return err;
778 }
779
780 static int xennet_get_responses(struct netfront_queue *queue,
781                                 struct netfront_rx_info *rinfo, RING_IDX rp,
782                                 struct sk_buff_head *list)
783 {
784         struct xen_netif_rx_response *rx = &rinfo->rx;
785         struct xen_netif_extra_info *extras = rinfo->extras;
786         struct device *dev = &queue->info->netdev->dev;
787         RING_IDX cons = queue->rx.rsp_cons;
788         struct sk_buff *skb = xennet_get_rx_skb(queue, cons);
789         grant_ref_t ref = xennet_get_rx_ref(queue, cons);
790         int max = MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD);
791         int slots = 1;
792         int err = 0;
793         unsigned long ret;
794
795         if (rx->flags & XEN_NETRXF_extra_info) {
796                 err = xennet_get_extras(queue, extras, rp);
797                 cons = queue->rx.rsp_cons;
798         }
799
800         for (;;) {
801                 if (unlikely(rx->status < 0 ||
802                              rx->offset + rx->status > XEN_PAGE_SIZE)) {
803                         if (net_ratelimit())
804                                 dev_warn(dev, "rx->offset: %u, size: %d\n",
805                                          rx->offset, rx->status);
806                         xennet_move_rx_slot(queue, skb, ref);
807                         err = -EINVAL;
808                         goto next;
809                 }
810
811                 /*
812                  * This definitely indicates a bug, either in this driver or in
813                  * the backend driver. In future this should flag the bad
814                  * situation to the system controller to reboot the backend.
815                  */
816                 if (ref == GRANT_INVALID_REF) {
817                         if (net_ratelimit())
818                                 dev_warn(dev, "Bad rx response id %d.\n",
819                                          rx->id);
820                         err = -EINVAL;
821                         goto next;
822                 }
823
824                 ret = gnttab_end_foreign_access_ref(ref, 0);
825                 BUG_ON(!ret);
826
827                 gnttab_release_grant_reference(&queue->gref_rx_head, ref);
828
829                 __skb_queue_tail(list, skb);
830
831 next:
832                 if (!(rx->flags & XEN_NETRXF_more_data))
833                         break;
834
835                 if (cons + slots == rp) {
836                         if (net_ratelimit())
837                                 dev_warn(dev, "Need more slots\n");
838                         err = -ENOENT;
839                         break;
840                 }
841
842                 rx = RING_GET_RESPONSE(&queue->rx, cons + slots);
843                 skb = xennet_get_rx_skb(queue, cons + slots);
844                 ref = xennet_get_rx_ref(queue, cons + slots);
845                 slots++;
846         }
847
848         if (unlikely(slots > max)) {
849                 if (net_ratelimit())
850                         dev_warn(dev, "Too many slots\n");
851                 err = -E2BIG;
852         }
853
854         if (unlikely(err))
855                 queue->rx.rsp_cons = cons + slots;
856
857         return err;
858 }
859
860 static int xennet_set_skb_gso(struct sk_buff *skb,
861                               struct xen_netif_extra_info *gso)
862 {
863         if (!gso->u.gso.size) {
864                 if (net_ratelimit())
865                         pr_warn("GSO size must not be zero\n");
866                 return -EINVAL;
867         }
868
869         if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 &&
870             gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) {
871                 if (net_ratelimit())
872                         pr_warn("Bad GSO type %d\n", gso->u.gso.type);
873                 return -EINVAL;
874         }
875
876         skb_shinfo(skb)->gso_size = gso->u.gso.size;
877         skb_shinfo(skb)->gso_type =
878                 (gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ?
879                 SKB_GSO_TCPV4 :
880                 SKB_GSO_TCPV6;
881
882         /* Header must be checked, and gso_segs computed. */
883         skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
884         skb_shinfo(skb)->gso_segs = 0;
885
886         return 0;
887 }
888
889 static RING_IDX xennet_fill_frags(struct netfront_queue *queue,
890                                   struct sk_buff *skb,
891                                   struct sk_buff_head *list)
892 {
893         struct skb_shared_info *shinfo = skb_shinfo(skb);
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 (shinfo->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                 BUG_ON(shinfo->nr_frags >= MAX_SKB_FRAGS);
909
910                 skb_add_rx_frag(skb, shinfo->nr_frags, skb_frag_page(nfrag),
911                                 rx->offset, rx->status, PAGE_SIZE);
912
913                 skb_shinfo(nskb)->nr_frags = 0;
914                 kfree_skb(nskb);
915         }
916
917         return cons;
918 }
919
920 static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
921 {
922         bool recalculate_partial_csum = false;
923
924         /*
925          * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
926          * peers can fail to set NETRXF_csum_blank when sending a GSO
927          * frame. In this case force the SKB to CHECKSUM_PARTIAL and
928          * recalculate the partial checksum.
929          */
930         if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
931                 struct netfront_info *np = netdev_priv(dev);
932                 atomic_inc(&np->rx_gso_checksum_fixup);
933                 skb->ip_summed = CHECKSUM_PARTIAL;
934                 recalculate_partial_csum = true;
935         }
936
937         /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
938         if (skb->ip_summed != CHECKSUM_PARTIAL)
939                 return 0;
940
941         return skb_checksum_setup(skb, recalculate_partial_csum);
942 }
943
944 static int handle_incoming_queue(struct netfront_queue *queue,
945                                  struct sk_buff_head *rxq)
946 {
947         struct netfront_stats *rx_stats = this_cpu_ptr(queue->info->rx_stats);
948         int packets_dropped = 0;
949         struct sk_buff *skb;
950
951         while ((skb = __skb_dequeue(rxq)) != NULL) {
952                 int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
953
954                 if (pull_to > skb_headlen(skb))
955                         __pskb_pull_tail(skb, pull_to - skb_headlen(skb));
956
957                 /* Ethernet work: Delayed to here as it peeks the header. */
958                 skb->protocol = eth_type_trans(skb, queue->info->netdev);
959                 skb_reset_network_header(skb);
960
961                 if (checksum_setup(queue->info->netdev, skb)) {
962                         kfree_skb(skb);
963                         packets_dropped++;
964                         queue->info->netdev->stats.rx_errors++;
965                         continue;
966                 }
967
968                 u64_stats_update_begin(&rx_stats->syncp);
969                 rx_stats->packets++;
970                 rx_stats->bytes += skb->len;
971                 u64_stats_update_end(&rx_stats->syncp);
972
973                 /* Pass it up. */
974                 napi_gro_receive(&queue->napi, skb);
975         }
976
977         return packets_dropped;
978 }
979
980 static int xennet_poll(struct napi_struct *napi, int budget)
981 {
982         struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi);
983         struct net_device *dev = queue->info->netdev;
984         struct sk_buff *skb;
985         struct netfront_rx_info rinfo;
986         struct xen_netif_rx_response *rx = &rinfo.rx;
987         struct xen_netif_extra_info *extras = rinfo.extras;
988         RING_IDX i, rp;
989         int work_done;
990         struct sk_buff_head rxq;
991         struct sk_buff_head errq;
992         struct sk_buff_head tmpq;
993         int err;
994
995         spin_lock(&queue->rx_lock);
996
997         skb_queue_head_init(&rxq);
998         skb_queue_head_init(&errq);
999         skb_queue_head_init(&tmpq);
1000
1001         rp = queue->rx.sring->rsp_prod;
1002         rmb(); /* Ensure we see queued responses up to 'rp'. */
1003
1004         i = queue->rx.rsp_cons;
1005         work_done = 0;
1006         while ((i != rp) && (work_done < budget)) {
1007                 memcpy(rx, RING_GET_RESPONSE(&queue->rx, i), sizeof(*rx));
1008                 memset(extras, 0, sizeof(rinfo.extras));
1009
1010                 err = xennet_get_responses(queue, &rinfo, rp, &tmpq);
1011
1012                 if (unlikely(err)) {
1013 err:
1014                         while ((skb = __skb_dequeue(&tmpq)))
1015                                 __skb_queue_tail(&errq, skb);
1016                         dev->stats.rx_errors++;
1017                         i = queue->rx.rsp_cons;
1018                         continue;
1019                 }
1020
1021                 skb = __skb_dequeue(&tmpq);
1022
1023                 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1024                         struct xen_netif_extra_info *gso;
1025                         gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1026
1027                         if (unlikely(xennet_set_skb_gso(skb, gso))) {
1028                                 __skb_queue_head(&tmpq, skb);
1029                                 queue->rx.rsp_cons += skb_queue_len(&tmpq);
1030                                 goto err;
1031                         }
1032                 }
1033
1034                 NETFRONT_SKB_CB(skb)->pull_to = rx->status;
1035                 if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD)
1036                         NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD;
1037
1038                 skb_shinfo(skb)->frags[0].page_offset = rx->offset;
1039                 skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status);
1040                 skb->data_len = rx->status;
1041                 skb->len += rx->status;
1042
1043                 i = xennet_fill_frags(queue, skb, &tmpq);
1044
1045                 if (rx->flags & XEN_NETRXF_csum_blank)
1046                         skb->ip_summed = CHECKSUM_PARTIAL;
1047                 else if (rx->flags & XEN_NETRXF_data_validated)
1048                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1049
1050                 __skb_queue_tail(&rxq, skb);
1051
1052                 queue->rx.rsp_cons = ++i;
1053                 work_done++;
1054         }
1055
1056         __skb_queue_purge(&errq);
1057
1058         work_done -= handle_incoming_queue(queue, &rxq);
1059
1060         xennet_alloc_rx_buffers(queue);
1061
1062         if (work_done < budget) {
1063                 int more_to_do = 0;
1064
1065                 napi_complete_done(napi, work_done);
1066
1067                 RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do);
1068                 if (more_to_do)
1069                         napi_schedule(napi);
1070         }
1071
1072         spin_unlock(&queue->rx_lock);
1073
1074         return work_done;
1075 }
1076
1077 static int xennet_change_mtu(struct net_device *dev, int mtu)
1078 {
1079         int max = xennet_can_sg(dev) ? XEN_NETIF_MAX_TX_SIZE : ETH_DATA_LEN;
1080
1081         if (mtu > max)
1082                 return -EINVAL;
1083         dev->mtu = mtu;
1084         return 0;
1085 }
1086
1087 static void xennet_get_stats64(struct net_device *dev,
1088                                struct rtnl_link_stats64 *tot)
1089 {
1090         struct netfront_info *np = netdev_priv(dev);
1091         int cpu;
1092
1093         for_each_possible_cpu(cpu) {
1094                 struct netfront_stats *rx_stats = per_cpu_ptr(np->rx_stats, cpu);
1095                 struct netfront_stats *tx_stats = per_cpu_ptr(np->tx_stats, cpu);
1096                 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1097                 unsigned int start;
1098
1099                 do {
1100                         start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
1101                         tx_packets = tx_stats->packets;
1102                         tx_bytes = tx_stats->bytes;
1103                 } while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
1104
1105                 do {
1106                         start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
1107                         rx_packets = rx_stats->packets;
1108                         rx_bytes = rx_stats->bytes;
1109                 } while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
1110
1111                 tot->rx_packets += rx_packets;
1112                 tot->tx_packets += tx_packets;
1113                 tot->rx_bytes   += rx_bytes;
1114                 tot->tx_bytes   += tx_bytes;
1115         }
1116
1117         tot->rx_errors  = dev->stats.rx_errors;
1118         tot->tx_dropped = dev->stats.tx_dropped;
1119 }
1120
1121 static void xennet_release_tx_bufs(struct netfront_queue *queue)
1122 {
1123         struct sk_buff *skb;
1124         int i;
1125
1126         for (i = 0; i < NET_TX_RING_SIZE; i++) {
1127                 /* Skip over entries which are actually freelist references */
1128                 if (skb_entry_is_link(&queue->tx_skbs[i]))
1129                         continue;
1130
1131                 skb = queue->tx_skbs[i].skb;
1132                 get_page(queue->grant_tx_page[i]);
1133                 gnttab_end_foreign_access(queue->grant_tx_ref[i],
1134                                           GNTMAP_readonly,
1135                                           (unsigned long)page_address(queue->grant_tx_page[i]));
1136                 queue->grant_tx_page[i] = NULL;
1137                 queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1138                 add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, i);
1139                 dev_kfree_skb_irq(skb);
1140         }
1141 }
1142
1143 static void xennet_release_rx_bufs(struct netfront_queue *queue)
1144 {
1145         int id, ref;
1146
1147         spin_lock_bh(&queue->rx_lock);
1148
1149         for (id = 0; id < NET_RX_RING_SIZE; id++) {
1150                 struct sk_buff *skb;
1151                 struct page *page;
1152
1153                 skb = queue->rx_skbs[id];
1154                 if (!skb)
1155                         continue;
1156
1157                 ref = queue->grant_rx_ref[id];
1158                 if (ref == GRANT_INVALID_REF)
1159                         continue;
1160
1161                 page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
1162
1163                 /* gnttab_end_foreign_access() needs a page ref until
1164                  * foreign access is ended (which may be deferred).
1165                  */
1166                 get_page(page);
1167                 gnttab_end_foreign_access(ref, 0,
1168                                           (unsigned long)page_address(page));
1169                 queue->grant_rx_ref[id] = GRANT_INVALID_REF;
1170
1171                 kfree_skb(skb);
1172         }
1173
1174         spin_unlock_bh(&queue->rx_lock);
1175 }
1176
1177 static netdev_features_t xennet_fix_features(struct net_device *dev,
1178         netdev_features_t features)
1179 {
1180         struct netfront_info *np = netdev_priv(dev);
1181
1182         if (features & NETIF_F_SG &&
1183             !xenbus_read_unsigned(np->xbdev->otherend, "feature-sg", 0))
1184                 features &= ~NETIF_F_SG;
1185
1186         if (features & NETIF_F_IPV6_CSUM &&
1187             !xenbus_read_unsigned(np->xbdev->otherend,
1188                                   "feature-ipv6-csum-offload", 0))
1189                 features &= ~NETIF_F_IPV6_CSUM;
1190
1191         if (features & NETIF_F_TSO &&
1192             !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv4", 0))
1193                 features &= ~NETIF_F_TSO;
1194
1195         if (features & NETIF_F_TSO6 &&
1196             !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv6", 0))
1197                 features &= ~NETIF_F_TSO6;
1198
1199         return features;
1200 }
1201
1202 static int xennet_set_features(struct net_device *dev,
1203         netdev_features_t features)
1204 {
1205         if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1206                 netdev_info(dev, "Reducing MTU because no SG offload");
1207                 dev->mtu = ETH_DATA_LEN;
1208         }
1209
1210         return 0;
1211 }
1212
1213 static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id)
1214 {
1215         struct netfront_queue *queue = dev_id;
1216         unsigned long flags;
1217
1218         spin_lock_irqsave(&queue->tx_lock, flags);
1219         xennet_tx_buf_gc(queue);
1220         spin_unlock_irqrestore(&queue->tx_lock, flags);
1221
1222         return IRQ_HANDLED;
1223 }
1224
1225 static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id)
1226 {
1227         struct netfront_queue *queue = dev_id;
1228         struct net_device *dev = queue->info->netdev;
1229
1230         if (likely(netif_carrier_ok(dev) &&
1231                    RING_HAS_UNCONSUMED_RESPONSES(&queue->rx)))
1232                 napi_schedule(&queue->napi);
1233
1234         return IRQ_HANDLED;
1235 }
1236
1237 static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1238 {
1239         xennet_tx_interrupt(irq, dev_id);
1240         xennet_rx_interrupt(irq, dev_id);
1241         return IRQ_HANDLED;
1242 }
1243
1244 #ifdef CONFIG_NET_POLL_CONTROLLER
1245 static void xennet_poll_controller(struct net_device *dev)
1246 {
1247         /* Poll each queue */
1248         struct netfront_info *info = netdev_priv(dev);
1249         unsigned int num_queues = dev->real_num_tx_queues;
1250         unsigned int i;
1251         for (i = 0; i < num_queues; ++i)
1252                 xennet_interrupt(0, &info->queues[i]);
1253 }
1254 #endif
1255
1256 static const struct net_device_ops xennet_netdev_ops = {
1257         .ndo_open            = xennet_open,
1258         .ndo_stop            = xennet_close,
1259         .ndo_start_xmit      = xennet_start_xmit,
1260         .ndo_change_mtu      = xennet_change_mtu,
1261         .ndo_get_stats64     = xennet_get_stats64,
1262         .ndo_set_mac_address = eth_mac_addr,
1263         .ndo_validate_addr   = eth_validate_addr,
1264         .ndo_fix_features    = xennet_fix_features,
1265         .ndo_set_features    = xennet_set_features,
1266         .ndo_select_queue    = xennet_select_queue,
1267 #ifdef CONFIG_NET_POLL_CONTROLLER
1268         .ndo_poll_controller = xennet_poll_controller,
1269 #endif
1270 };
1271
1272 static void xennet_free_netdev(struct net_device *netdev)
1273 {
1274         struct netfront_info *np = netdev_priv(netdev);
1275
1276         free_percpu(np->rx_stats);
1277         free_percpu(np->tx_stats);
1278         free_netdev(netdev);
1279 }
1280
1281 static struct net_device *xennet_create_dev(struct xenbus_device *dev)
1282 {
1283         int err;
1284         struct net_device *netdev;
1285         struct netfront_info *np;
1286
1287         netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues);
1288         if (!netdev)
1289                 return ERR_PTR(-ENOMEM);
1290
1291         np                   = netdev_priv(netdev);
1292         np->xbdev            = dev;
1293
1294         np->queues = NULL;
1295
1296         err = -ENOMEM;
1297         np->rx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1298         if (np->rx_stats == NULL)
1299                 goto exit;
1300         np->tx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1301         if (np->tx_stats == NULL)
1302                 goto exit;
1303
1304         netdev->netdev_ops      = &xennet_netdev_ops;
1305
1306         netdev->features        = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1307                                   NETIF_F_GSO_ROBUST;
1308         netdev->hw_features     = NETIF_F_SG |
1309                                   NETIF_F_IPV6_CSUM |
1310                                   NETIF_F_TSO | NETIF_F_TSO6;
1311
1312         /*
1313          * Assume that all hw features are available for now. This set
1314          * will be adjusted by the call to netdev_update_features() in
1315          * xennet_connect() which is the earliest point where we can
1316          * negotiate with the backend regarding supported features.
1317          */
1318         netdev->features |= netdev->hw_features;
1319
1320         netdev->ethtool_ops = &xennet_ethtool_ops;
1321         netdev->min_mtu = ETH_MIN_MTU;
1322         netdev->max_mtu = XEN_NETIF_MAX_TX_SIZE;
1323         SET_NETDEV_DEV(netdev, &dev->dev);
1324
1325         np->netdev = netdev;
1326
1327         netif_carrier_off(netdev);
1328
1329         xenbus_switch_state(dev, XenbusStateInitialising);
1330         return netdev;
1331
1332  exit:
1333         xennet_free_netdev(netdev);
1334         return ERR_PTR(err);
1335 }
1336
1337 /**
1338  * Entry point to this code when a new device is created.  Allocate the basic
1339  * structures and the ring buffers for communication with the backend, and
1340  * inform the backend of the appropriate details for those.
1341  */
1342 static int netfront_probe(struct xenbus_device *dev,
1343                           const struct xenbus_device_id *id)
1344 {
1345         int err;
1346         struct net_device *netdev;
1347         struct netfront_info *info;
1348
1349         netdev = xennet_create_dev(dev);
1350         if (IS_ERR(netdev)) {
1351                 err = PTR_ERR(netdev);
1352                 xenbus_dev_fatal(dev, err, "creating netdev");
1353                 return err;
1354         }
1355
1356         info = netdev_priv(netdev);
1357         dev_set_drvdata(&dev->dev, info);
1358 #ifdef CONFIG_SYSFS
1359         info->netdev->sysfs_groups[0] = &xennet_dev_group;
1360 #endif
1361         err = register_netdev(info->netdev);
1362         if (err) {
1363                 pr_warn("%s: register_netdev err=%d\n", __func__, err);
1364                 goto fail;
1365         }
1366
1367         return 0;
1368
1369  fail:
1370         xennet_free_netdev(netdev);
1371         dev_set_drvdata(&dev->dev, NULL);
1372         return err;
1373 }
1374
1375 static void xennet_end_access(int ref, void *page)
1376 {
1377         /* This frees the page as a side-effect */
1378         if (ref != GRANT_INVALID_REF)
1379                 gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1380 }
1381
1382 static void xennet_disconnect_backend(struct netfront_info *info)
1383 {
1384         unsigned int i = 0;
1385         unsigned int num_queues = info->netdev->real_num_tx_queues;
1386
1387         netif_carrier_off(info->netdev);
1388
1389         for (i = 0; i < num_queues && info->queues; ++i) {
1390                 struct netfront_queue *queue = &info->queues[i];
1391
1392                 del_timer_sync(&queue->rx_refill_timer);
1393
1394                 if (queue->tx_irq && (queue->tx_irq == queue->rx_irq))
1395                         unbind_from_irqhandler(queue->tx_irq, queue);
1396                 if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) {
1397                         unbind_from_irqhandler(queue->tx_irq, queue);
1398                         unbind_from_irqhandler(queue->rx_irq, queue);
1399                 }
1400                 queue->tx_evtchn = queue->rx_evtchn = 0;
1401                 queue->tx_irq = queue->rx_irq = 0;
1402
1403                 if (netif_running(info->netdev))
1404                         napi_synchronize(&queue->napi);
1405
1406                 xennet_release_tx_bufs(queue);
1407                 xennet_release_rx_bufs(queue);
1408                 gnttab_free_grant_references(queue->gref_tx_head);
1409                 gnttab_free_grant_references(queue->gref_rx_head);
1410
1411                 /* End access and free the pages */
1412                 xennet_end_access(queue->tx_ring_ref, queue->tx.sring);
1413                 xennet_end_access(queue->rx_ring_ref, queue->rx.sring);
1414
1415                 queue->tx_ring_ref = GRANT_INVALID_REF;
1416                 queue->rx_ring_ref = GRANT_INVALID_REF;
1417                 queue->tx.sring = NULL;
1418                 queue->rx.sring = NULL;
1419         }
1420 }
1421
1422 /**
1423  * We are reconnecting to the backend, due to a suspend/resume, or a backend
1424  * driver restart.  We tear down our netif structure and recreate it, but
1425  * leave the device-layer structures intact so that this is transparent to the
1426  * rest of the kernel.
1427  */
1428 static int netfront_resume(struct xenbus_device *dev)
1429 {
1430         struct netfront_info *info = dev_get_drvdata(&dev->dev);
1431
1432         dev_dbg(&dev->dev, "%s\n", dev->nodename);
1433
1434         xennet_disconnect_backend(info);
1435         return 0;
1436 }
1437
1438 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1439 {
1440         char *s, *e, *macstr;
1441         int i;
1442
1443         macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1444         if (IS_ERR(macstr))
1445                 return PTR_ERR(macstr);
1446
1447         for (i = 0; i < ETH_ALEN; i++) {
1448                 mac[i] = simple_strtoul(s, &e, 16);
1449                 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1450                         kfree(macstr);
1451                         return -ENOENT;
1452                 }
1453                 s = e+1;
1454         }
1455
1456         kfree(macstr);
1457         return 0;
1458 }
1459
1460 static int setup_netfront_single(struct netfront_queue *queue)
1461 {
1462         int err;
1463
1464         err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1465         if (err < 0)
1466                 goto fail;
1467
1468         err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1469                                         xennet_interrupt,
1470                                         0, queue->info->netdev->name, queue);
1471         if (err < 0)
1472                 goto bind_fail;
1473         queue->rx_evtchn = queue->tx_evtchn;
1474         queue->rx_irq = queue->tx_irq = err;
1475
1476         return 0;
1477
1478 bind_fail:
1479         xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1480         queue->tx_evtchn = 0;
1481 fail:
1482         return err;
1483 }
1484
1485 static int setup_netfront_split(struct netfront_queue *queue)
1486 {
1487         int err;
1488
1489         err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1490         if (err < 0)
1491                 goto fail;
1492         err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn);
1493         if (err < 0)
1494                 goto alloc_rx_evtchn_fail;
1495
1496         snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
1497                  "%s-tx", queue->name);
1498         err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1499                                         xennet_tx_interrupt,
1500                                         0, queue->tx_irq_name, queue);
1501         if (err < 0)
1502                 goto bind_tx_fail;
1503         queue->tx_irq = err;
1504
1505         snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
1506                  "%s-rx", queue->name);
1507         err = bind_evtchn_to_irqhandler(queue->rx_evtchn,
1508                                         xennet_rx_interrupt,
1509                                         0, queue->rx_irq_name, queue);
1510         if (err < 0)
1511                 goto bind_rx_fail;
1512         queue->rx_irq = err;
1513
1514         return 0;
1515
1516 bind_rx_fail:
1517         unbind_from_irqhandler(queue->tx_irq, queue);
1518         queue->tx_irq = 0;
1519 bind_tx_fail:
1520         xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn);
1521         queue->rx_evtchn = 0;
1522 alloc_rx_evtchn_fail:
1523         xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1524         queue->tx_evtchn = 0;
1525 fail:
1526         return err;
1527 }
1528
1529 static int setup_netfront(struct xenbus_device *dev,
1530                         struct netfront_queue *queue, unsigned int feature_split_evtchn)
1531 {
1532         struct xen_netif_tx_sring *txs;
1533         struct xen_netif_rx_sring *rxs;
1534         grant_ref_t gref;
1535         int err;
1536
1537         queue->tx_ring_ref = GRANT_INVALID_REF;
1538         queue->rx_ring_ref = GRANT_INVALID_REF;
1539         queue->rx.sring = NULL;
1540         queue->tx.sring = NULL;
1541
1542         txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1543         if (!txs) {
1544                 err = -ENOMEM;
1545                 xenbus_dev_fatal(dev, err, "allocating tx ring page");
1546                 goto fail;
1547         }
1548         SHARED_RING_INIT(txs);
1549         FRONT_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE);
1550
1551         err = xenbus_grant_ring(dev, txs, 1, &gref);
1552         if (err < 0)
1553                 goto grant_tx_ring_fail;
1554         queue->tx_ring_ref = gref;
1555
1556         rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1557         if (!rxs) {
1558                 err = -ENOMEM;
1559                 xenbus_dev_fatal(dev, err, "allocating rx ring page");
1560                 goto alloc_rx_ring_fail;
1561         }
1562         SHARED_RING_INIT(rxs);
1563         FRONT_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE);
1564
1565         err = xenbus_grant_ring(dev, rxs, 1, &gref);
1566         if (err < 0)
1567                 goto grant_rx_ring_fail;
1568         queue->rx_ring_ref = gref;
1569
1570         if (feature_split_evtchn)
1571                 err = setup_netfront_split(queue);
1572         /* setup single event channel if
1573          *  a) feature-split-event-channels == 0
1574          *  b) feature-split-event-channels == 1 but failed to setup
1575          */
1576         if (!feature_split_evtchn || (feature_split_evtchn && err))
1577                 err = setup_netfront_single(queue);
1578
1579         if (err)
1580                 goto alloc_evtchn_fail;
1581
1582         return 0;
1583
1584         /* If we fail to setup netfront, it is safe to just revoke access to
1585          * granted pages because backend is not accessing it at this point.
1586          */
1587 alloc_evtchn_fail:
1588         gnttab_end_foreign_access_ref(queue->rx_ring_ref, 0);
1589 grant_rx_ring_fail:
1590         free_page((unsigned long)rxs);
1591 alloc_rx_ring_fail:
1592         gnttab_end_foreign_access_ref(queue->tx_ring_ref, 0);
1593 grant_tx_ring_fail:
1594         free_page((unsigned long)txs);
1595 fail:
1596         return err;
1597 }
1598
1599 /* Queue-specific initialisation
1600  * This used to be done in xennet_create_dev() but must now
1601  * be run per-queue.
1602  */
1603 static int xennet_init_queue(struct netfront_queue *queue)
1604 {
1605         unsigned short i;
1606         int err = 0;
1607
1608         spin_lock_init(&queue->tx_lock);
1609         spin_lock_init(&queue->rx_lock);
1610
1611         timer_setup(&queue->rx_refill_timer, rx_refill_timeout, 0);
1612
1613         snprintf(queue->name, sizeof(queue->name), "%s-q%u",
1614                  queue->info->netdev->name, queue->id);
1615
1616         /* Initialise tx_skbs as a free chain containing every entry. */
1617         queue->tx_skb_freelist = 0;
1618         for (i = 0; i < NET_TX_RING_SIZE; i++) {
1619                 skb_entry_set_link(&queue->tx_skbs[i], i+1);
1620                 queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1621                 queue->grant_tx_page[i] = NULL;
1622         }
1623
1624         /* Clear out rx_skbs */
1625         for (i = 0; i < NET_RX_RING_SIZE; i++) {
1626                 queue->rx_skbs[i] = NULL;
1627                 queue->grant_rx_ref[i] = GRANT_INVALID_REF;
1628         }
1629
1630         /* A grant for every tx ring slot */
1631         if (gnttab_alloc_grant_references(NET_TX_RING_SIZE,
1632                                           &queue->gref_tx_head) < 0) {
1633                 pr_alert("can't alloc tx grant refs\n");
1634                 err = -ENOMEM;
1635                 goto exit;
1636         }
1637
1638         /* A grant for every rx ring slot */
1639         if (gnttab_alloc_grant_references(NET_RX_RING_SIZE,
1640                                           &queue->gref_rx_head) < 0) {
1641                 pr_alert("can't alloc rx grant refs\n");
1642                 err = -ENOMEM;
1643                 goto exit_free_tx;
1644         }
1645
1646         return 0;
1647
1648  exit_free_tx:
1649         gnttab_free_grant_references(queue->gref_tx_head);
1650  exit:
1651         return err;
1652 }
1653
1654 static int write_queue_xenstore_keys(struct netfront_queue *queue,
1655                            struct xenbus_transaction *xbt, int write_hierarchical)
1656 {
1657         /* Write the queue-specific keys into XenStore in the traditional
1658          * way for a single queue, or in a queue subkeys for multiple
1659          * queues.
1660          */
1661         struct xenbus_device *dev = queue->info->xbdev;
1662         int err;
1663         const char *message;
1664         char *path;
1665         size_t pathsize;
1666
1667         /* Choose the correct place to write the keys */
1668         if (write_hierarchical) {
1669                 pathsize = strlen(dev->nodename) + 10;
1670                 path = kzalloc(pathsize, GFP_KERNEL);
1671                 if (!path) {
1672                         err = -ENOMEM;
1673                         message = "out of memory while writing ring references";
1674                         goto error;
1675                 }
1676                 snprintf(path, pathsize, "%s/queue-%u",
1677                                 dev->nodename, queue->id);
1678         } else {
1679                 path = (char *)dev->nodename;
1680         }
1681
1682         /* Write ring references */
1683         err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u",
1684                         queue->tx_ring_ref);
1685         if (err) {
1686                 message = "writing tx-ring-ref";
1687                 goto error;
1688         }
1689
1690         err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u",
1691                         queue->rx_ring_ref);
1692         if (err) {
1693                 message = "writing rx-ring-ref";
1694                 goto error;
1695         }
1696
1697         /* Write event channels; taking into account both shared
1698          * and split event channel scenarios.
1699          */
1700         if (queue->tx_evtchn == queue->rx_evtchn) {
1701                 /* Shared event channel */
1702                 err = xenbus_printf(*xbt, path,
1703                                 "event-channel", "%u", queue->tx_evtchn);
1704                 if (err) {
1705                         message = "writing event-channel";
1706                         goto error;
1707                 }
1708         } else {
1709                 /* Split event channels */
1710                 err = xenbus_printf(*xbt, path,
1711                                 "event-channel-tx", "%u", queue->tx_evtchn);
1712                 if (err) {
1713                         message = "writing event-channel-tx";
1714                         goto error;
1715                 }
1716
1717                 err = xenbus_printf(*xbt, path,
1718                                 "event-channel-rx", "%u", queue->rx_evtchn);
1719                 if (err) {
1720                         message = "writing event-channel-rx";
1721                         goto error;
1722                 }
1723         }
1724
1725         if (write_hierarchical)
1726                 kfree(path);
1727         return 0;
1728
1729 error:
1730         if (write_hierarchical)
1731                 kfree(path);
1732         xenbus_dev_fatal(dev, err, "%s", message);
1733         return err;
1734 }
1735
1736 static void xennet_destroy_queues(struct netfront_info *info)
1737 {
1738         unsigned int i;
1739
1740         rtnl_lock();
1741
1742         for (i = 0; i < info->netdev->real_num_tx_queues; i++) {
1743                 struct netfront_queue *queue = &info->queues[i];
1744
1745                 if (netif_running(info->netdev))
1746                         napi_disable(&queue->napi);
1747                 netif_napi_del(&queue->napi);
1748         }
1749
1750         rtnl_unlock();
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         rtnl_lock();
1768
1769         for (i = 0; i < *num_queues; i++) {
1770                 struct netfront_queue *queue = &info->queues[i];
1771
1772                 queue->id = i;
1773                 queue->info = info;
1774
1775                 ret = xennet_init_queue(queue);
1776                 if (ret < 0) {
1777                         dev_warn(&info->netdev->dev,
1778                                  "only created %d queues\n", i);
1779                         *num_queues = i;
1780                         break;
1781                 }
1782
1783                 netif_napi_add(queue->info->netdev, &queue->napi,
1784                                xennet_poll, 64);
1785                 if (netif_running(info->netdev))
1786                         napi_enable(&queue->napi);
1787         }
1788
1789         netif_set_real_num_tx_queues(info->netdev, *num_queues);
1790
1791         rtnl_unlock();
1792
1793         if (*num_queues == 0) {
1794                 dev_err(&info->netdev->dev, "no queues\n");
1795                 return -EINVAL;
1796         }
1797         return 0;
1798 }
1799
1800 /* Common code used when first setting up, and when resuming. */
1801 static int talk_to_netback(struct xenbus_device *dev,
1802                            struct netfront_info *info)
1803 {
1804         const char *message;
1805         struct xenbus_transaction xbt;
1806         int err;
1807         unsigned int feature_split_evtchn;
1808         unsigned int i = 0;
1809         unsigned int max_queues = 0;
1810         struct netfront_queue *queue = NULL;
1811         unsigned int num_queues = 1;
1812
1813         info->netdev->irq = 0;
1814
1815         /* Check if backend supports multiple queues */
1816         max_queues = xenbus_read_unsigned(info->xbdev->otherend,
1817                                           "multi-queue-max-queues", 1);
1818         num_queues = min(max_queues, xennet_max_queues);
1819
1820         /* Check feature-split-event-channels */
1821         feature_split_evtchn = xenbus_read_unsigned(info->xbdev->otherend,
1822                                         "feature-split-event-channels", 0);
1823
1824         /* Read mac addr. */
1825         err = xen_net_read_mac(dev, info->netdev->dev_addr);
1826         if (err) {
1827                 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
1828                 goto out;
1829         }
1830
1831         if (info->queues)
1832                 xennet_destroy_queues(info);
1833
1834         err = xennet_create_queues(info, &num_queues);
1835         if (err < 0) {
1836                 xenbus_dev_fatal(dev, err, "creating queues");
1837                 kfree(info->queues);
1838                 info->queues = NULL;
1839                 goto out;
1840         }
1841
1842         /* Create shared ring, alloc event channel -- for each queue */
1843         for (i = 0; i < num_queues; ++i) {
1844                 queue = &info->queues[i];
1845                 err = setup_netfront(dev, queue, feature_split_evtchn);
1846                 if (err)
1847                         goto destroy_ring;
1848         }
1849
1850 again:
1851         err = xenbus_transaction_start(&xbt);
1852         if (err) {
1853                 xenbus_dev_fatal(dev, err, "starting transaction");
1854                 goto destroy_ring;
1855         }
1856
1857         if (xenbus_exists(XBT_NIL,
1858                           info->xbdev->otherend, "multi-queue-max-queues")) {
1859                 /* Write the number of queues */
1860                 err = xenbus_printf(xbt, dev->nodename,
1861                                     "multi-queue-num-queues", "%u", num_queues);
1862                 if (err) {
1863                         message = "writing multi-queue-num-queues";
1864                         goto abort_transaction_no_dev_fatal;
1865                 }
1866         }
1867
1868         if (num_queues == 1) {
1869                 err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */
1870                 if (err)
1871                         goto abort_transaction_no_dev_fatal;
1872         } else {
1873                 /* Write the keys for each queue */
1874                 for (i = 0; i < num_queues; ++i) {
1875                         queue = &info->queues[i];
1876                         err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */
1877                         if (err)
1878                                 goto abort_transaction_no_dev_fatal;
1879                 }
1880         }
1881
1882         /* The remaining keys are not queue-specific */
1883         err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
1884                             1);
1885         if (err) {
1886                 message = "writing request-rx-copy";
1887                 goto abort_transaction;
1888         }
1889
1890         err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
1891         if (err) {
1892                 message = "writing feature-rx-notify";
1893                 goto abort_transaction;
1894         }
1895
1896         err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
1897         if (err) {
1898                 message = "writing feature-sg";
1899                 goto abort_transaction;
1900         }
1901
1902         err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
1903         if (err) {
1904                 message = "writing feature-gso-tcpv4";
1905                 goto abort_transaction;
1906         }
1907
1908         err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1");
1909         if (err) {
1910                 message = "writing feature-gso-tcpv6";
1911                 goto abort_transaction;
1912         }
1913
1914         err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload",
1915                            "1");
1916         if (err) {
1917                 message = "writing feature-ipv6-csum-offload";
1918                 goto abort_transaction;
1919         }
1920
1921         err = xenbus_transaction_end(xbt, 0);
1922         if (err) {
1923                 if (err == -EAGAIN)
1924                         goto again;
1925                 xenbus_dev_fatal(dev, err, "completing transaction");
1926                 goto destroy_ring;
1927         }
1928
1929         return 0;
1930
1931  abort_transaction:
1932         xenbus_dev_fatal(dev, err, "%s", message);
1933 abort_transaction_no_dev_fatal:
1934         xenbus_transaction_end(xbt, 1);
1935  destroy_ring:
1936         xennet_disconnect_backend(info);
1937         xennet_destroy_queues(info);
1938  out:
1939         device_unregister(&dev->dev);
1940         return err;
1941 }
1942
1943 static int xennet_connect(struct net_device *dev)
1944 {
1945         struct netfront_info *np = netdev_priv(dev);
1946         unsigned int num_queues = 0;
1947         int err;
1948         unsigned int j = 0;
1949         struct netfront_queue *queue = NULL;
1950
1951         if (!xenbus_read_unsigned(np->xbdev->otherend, "feature-rx-copy", 0)) {
1952                 dev_info(&dev->dev,
1953                          "backend does not support copying receive path\n");
1954                 return -ENODEV;
1955         }
1956
1957         err = talk_to_netback(np->xbdev, np);
1958         if (err)
1959                 return err;
1960
1961         /* talk_to_netback() sets the correct number of queues */
1962         num_queues = dev->real_num_tx_queues;
1963
1964         rtnl_lock();
1965         netdev_update_features(dev);
1966         rtnl_unlock();
1967
1968         /*
1969          * All public and private state should now be sane.  Get
1970          * ready to start sending and receiving packets and give the driver
1971          * domain a kick because we've probably just requeued some
1972          * packets.
1973          */
1974         netif_carrier_on(np->netdev);
1975         for (j = 0; j < num_queues; ++j) {
1976                 queue = &np->queues[j];
1977
1978                 notify_remote_via_irq(queue->tx_irq);
1979                 if (queue->tx_irq != queue->rx_irq)
1980                         notify_remote_via_irq(queue->rx_irq);
1981
1982                 spin_lock_irq(&queue->tx_lock);
1983                 xennet_tx_buf_gc(queue);
1984                 spin_unlock_irq(&queue->tx_lock);
1985
1986                 spin_lock_bh(&queue->rx_lock);
1987                 xennet_alloc_rx_buffers(queue);
1988                 spin_unlock_bh(&queue->rx_lock);
1989         }
1990
1991         return 0;
1992 }
1993
1994 /**
1995  * Callback received when the backend's state changes.
1996  */
1997 static void netback_changed(struct xenbus_device *dev,
1998                             enum xenbus_state backend_state)
1999 {
2000         struct netfront_info *np = dev_get_drvdata(&dev->dev);
2001         struct net_device *netdev = np->netdev;
2002
2003         dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
2004
2005         switch (backend_state) {
2006         case XenbusStateInitialising:
2007         case XenbusStateInitialised:
2008         case XenbusStateReconfiguring:
2009         case XenbusStateReconfigured:
2010         case XenbusStateUnknown:
2011                 break;
2012
2013         case XenbusStateInitWait:
2014                 if (dev->state != XenbusStateInitialising)
2015                         break;
2016                 if (xennet_connect(netdev) != 0)
2017                         break;
2018                 xenbus_switch_state(dev, XenbusStateConnected);
2019                 break;
2020
2021         case XenbusStateConnected:
2022                 netdev_notify_peers(netdev);
2023                 break;
2024
2025         case XenbusStateClosed:
2026                 wake_up_all(&module_unload_q);
2027                 if (dev->state == XenbusStateClosed)
2028                         break;
2029                 /* Missed the backend's CLOSING state -- fallthrough */
2030         case XenbusStateClosing:
2031                 wake_up_all(&module_unload_q);
2032                 xenbus_frontend_closed(dev);
2033                 break;
2034         }
2035 }
2036
2037 static const struct xennet_stat {
2038         char name[ETH_GSTRING_LEN];
2039         u16 offset;
2040 } xennet_stats[] = {
2041         {
2042                 "rx_gso_checksum_fixup",
2043                 offsetof(struct netfront_info, rx_gso_checksum_fixup)
2044         },
2045 };
2046
2047 static int xennet_get_sset_count(struct net_device *dev, int string_set)
2048 {
2049         switch (string_set) {
2050         case ETH_SS_STATS:
2051                 return ARRAY_SIZE(xennet_stats);
2052         default:
2053                 return -EINVAL;
2054         }
2055 }
2056
2057 static void xennet_get_ethtool_stats(struct net_device *dev,
2058                                      struct ethtool_stats *stats, u64 * data)
2059 {
2060         void *np = netdev_priv(dev);
2061         int i;
2062
2063         for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2064                 data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset));
2065 }
2066
2067 static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
2068 {
2069         int i;
2070
2071         switch (stringset) {
2072         case ETH_SS_STATS:
2073                 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2074                         memcpy(data + i * ETH_GSTRING_LEN,
2075                                xennet_stats[i].name, ETH_GSTRING_LEN);
2076                 break;
2077         }
2078 }
2079
2080 static const struct ethtool_ops xennet_ethtool_ops =
2081 {
2082         .get_link = ethtool_op_get_link,
2083
2084         .get_sset_count = xennet_get_sset_count,
2085         .get_ethtool_stats = xennet_get_ethtool_stats,
2086         .get_strings = xennet_get_strings,
2087 };
2088
2089 #ifdef CONFIG_SYSFS
2090 static ssize_t show_rxbuf(struct device *dev,
2091                           struct device_attribute *attr, char *buf)
2092 {
2093         return sprintf(buf, "%lu\n", NET_RX_RING_SIZE);
2094 }
2095
2096 static ssize_t store_rxbuf(struct device *dev,
2097                            struct device_attribute *attr,
2098                            const char *buf, size_t len)
2099 {
2100         char *endp;
2101         unsigned long target;
2102
2103         if (!capable(CAP_NET_ADMIN))
2104                 return -EPERM;
2105
2106         target = simple_strtoul(buf, &endp, 0);
2107         if (endp == buf)
2108                 return -EBADMSG;
2109
2110         /* rxbuf_min and rxbuf_max are no longer configurable. */
2111
2112         return len;
2113 }
2114
2115 static DEVICE_ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf, store_rxbuf);
2116 static DEVICE_ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf, store_rxbuf);
2117 static DEVICE_ATTR(rxbuf_cur, S_IRUGO, show_rxbuf, NULL);
2118
2119 static struct attribute *xennet_dev_attrs[] = {
2120         &dev_attr_rxbuf_min.attr,
2121         &dev_attr_rxbuf_max.attr,
2122         &dev_attr_rxbuf_cur.attr,
2123         NULL
2124 };
2125
2126 static const struct attribute_group xennet_dev_group = {
2127         .attrs = xennet_dev_attrs
2128 };
2129 #endif /* CONFIG_SYSFS */
2130
2131 static int xennet_remove(struct xenbus_device *dev)
2132 {
2133         struct netfront_info *info = dev_get_drvdata(&dev->dev);
2134
2135         dev_dbg(&dev->dev, "%s\n", dev->nodename);
2136
2137         if (xenbus_read_driver_state(dev->otherend) != XenbusStateClosed) {
2138                 xenbus_switch_state(dev, XenbusStateClosing);
2139                 wait_event(module_unload_q,
2140                            xenbus_read_driver_state(dev->otherend) ==
2141                            XenbusStateClosing);
2142
2143                 xenbus_switch_state(dev, XenbusStateClosed);
2144                 wait_event(module_unload_q,
2145                            xenbus_read_driver_state(dev->otherend) ==
2146                            XenbusStateClosed ||
2147                            xenbus_read_driver_state(dev->otherend) ==
2148                            XenbusStateUnknown);
2149         }
2150
2151         xennet_disconnect_backend(info);
2152
2153         unregister_netdev(info->netdev);
2154
2155         if (info->queues)
2156                 xennet_destroy_queues(info);
2157         xennet_free_netdev(info->netdev);
2158
2159         return 0;
2160 }
2161
2162 static const struct xenbus_device_id netfront_ids[] = {
2163         { "vif" },
2164         { "" }
2165 };
2166
2167 static struct xenbus_driver netfront_driver = {
2168         .ids = netfront_ids,
2169         .probe = netfront_probe,
2170         .remove = xennet_remove,
2171         .resume = netfront_resume,
2172         .otherend_changed = netback_changed,
2173 };
2174
2175 static int __init netif_init(void)
2176 {
2177         if (!xen_domain())
2178                 return -ENODEV;
2179
2180         if (!xen_has_pv_nic_devices())
2181                 return -ENODEV;
2182
2183         pr_info("Initialising Xen virtual ethernet driver\n");
2184
2185         /* Allow as many queues as there are CPUs inut max. 8 if user has not
2186          * specified a value.
2187          */
2188         if (xennet_max_queues == 0)
2189                 xennet_max_queues = min_t(unsigned int, MAX_QUEUES_DEFAULT,
2190                                           num_online_cpus());
2191
2192         return xenbus_register_frontend(&netfront_driver);
2193 }
2194 module_init(netif_init);
2195
2196
2197 static void __exit netif_exit(void)
2198 {
2199         xenbus_unregister_driver(&netfront_driver);
2200 }
2201 module_exit(netif_exit);
2202
2203 MODULE_DESCRIPTION("Xen virtual network device frontend");
2204 MODULE_LICENSE("GPL");
2205 MODULE_ALIAS("xen:vif");
2206 MODULE_ALIAS("xennet");