]> Git Repo - linux.git/blob - drivers/net/xen-netfront.c
Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
[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 static unsigned int xennet_max_queues;
61 module_param_named(max_queues, xennet_max_queues, uint, 0644);
62 MODULE_PARM_DESC(max_queues,
63                  "Maximum number of queues per virtual interface");
64
65 static const struct ethtool_ops xennet_ethtool_ops;
66
67 struct netfront_cb {
68         int pull_to;
69 };
70
71 #define NETFRONT_SKB_CB(skb)    ((struct netfront_cb *)((skb)->cb))
72
73 #define RX_COPY_THRESHOLD 256
74
75 #define GRANT_INVALID_REF       0
76
77 #define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, XEN_PAGE_SIZE)
78 #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, XEN_PAGE_SIZE)
79
80 /* Minimum number of Rx slots (includes slot for GSO metadata). */
81 #define NET_RX_SLOTS_MIN (XEN_NETIF_NR_SLOTS_MIN + 1)
82
83 /* Queue name is interface name with "-qNNN" appended */
84 #define QUEUE_NAME_SIZE (IFNAMSIZ + 6)
85
86 /* IRQ name is queue name with "-tx" or "-rx" appended */
87 #define IRQ_NAME_SIZE (QUEUE_NAME_SIZE + 3)
88
89 struct netfront_stats {
90         u64                     packets;
91         u64                     bytes;
92         struct u64_stats_sync   syncp;
93 };
94
95 struct netfront_info;
96
97 struct netfront_queue {
98         unsigned int id; /* Queue ID, 0-based */
99         char name[QUEUE_NAME_SIZE]; /* DEVNAME-qN */
100         struct netfront_info *info;
101
102         struct napi_struct napi;
103
104         /* Split event channels support, tx_* == rx_* when using
105          * single event channel.
106          */
107         unsigned int tx_evtchn, rx_evtchn;
108         unsigned int tx_irq, rx_irq;
109         /* Only used when split event channels support is enabled */
110         char tx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-tx */
111         char rx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-rx */
112
113         spinlock_t   tx_lock;
114         struct xen_netif_tx_front_ring tx;
115         int tx_ring_ref;
116
117         /*
118          * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
119          * are linked from tx_skb_freelist through skb_entry.link.
120          *
121          *  NB. Freelist index entries are always going to be less than
122          *  PAGE_OFFSET, whereas pointers to skbs will always be equal or
123          *  greater than PAGE_OFFSET: we use this property to distinguish
124          *  them.
125          */
126         union skb_entry {
127                 struct sk_buff *skb;
128                 unsigned long link;
129         } tx_skbs[NET_TX_RING_SIZE];
130         grant_ref_t gref_tx_head;
131         grant_ref_t grant_tx_ref[NET_TX_RING_SIZE];
132         struct page *grant_tx_page[NET_TX_RING_SIZE];
133         unsigned tx_skb_freelist;
134
135         spinlock_t   rx_lock ____cacheline_aligned_in_smp;
136         struct xen_netif_rx_front_ring rx;
137         int rx_ring_ref;
138
139         struct timer_list rx_refill_timer;
140
141         struct sk_buff *rx_skbs[NET_RX_RING_SIZE];
142         grant_ref_t gref_rx_head;
143         grant_ref_t grant_rx_ref[NET_RX_RING_SIZE];
144 };
145
146 struct netfront_info {
147         struct list_head list;
148         struct net_device *netdev;
149
150         struct xenbus_device *xbdev;
151
152         /* Multi-queue support */
153         struct netfront_queue *queues;
154
155         /* Statistics */
156         struct netfront_stats __percpu *rx_stats;
157         struct netfront_stats __percpu *tx_stats;
158
159         atomic_t rx_gso_checksum_fixup;
160 };
161
162 struct netfront_rx_info {
163         struct xen_netif_rx_response rx;
164         struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
165 };
166
167 static void skb_entry_set_link(union skb_entry *list, unsigned short id)
168 {
169         list->link = id;
170 }
171
172 static int skb_entry_is_link(const union skb_entry *list)
173 {
174         BUILD_BUG_ON(sizeof(list->skb) != sizeof(list->link));
175         return (unsigned long)list->skb < PAGE_OFFSET;
176 }
177
178 /*
179  * Access macros for acquiring freeing slots in tx_skbs[].
180  */
181
182 static void add_id_to_freelist(unsigned *head, union skb_entry *list,
183                                unsigned short id)
184 {
185         skb_entry_set_link(&list[id], *head);
186         *head = id;
187 }
188
189 static unsigned short get_id_from_freelist(unsigned *head,
190                                            union skb_entry *list)
191 {
192         unsigned int id = *head;
193         *head = list[id].link;
194         return id;
195 }
196
197 static int xennet_rxidx(RING_IDX idx)
198 {
199         return idx & (NET_RX_RING_SIZE - 1);
200 }
201
202 static struct sk_buff *xennet_get_rx_skb(struct netfront_queue *queue,
203                                          RING_IDX ri)
204 {
205         int i = xennet_rxidx(ri);
206         struct sk_buff *skb = queue->rx_skbs[i];
207         queue->rx_skbs[i] = NULL;
208         return skb;
209 }
210
211 static grant_ref_t xennet_get_rx_ref(struct netfront_queue *queue,
212                                             RING_IDX ri)
213 {
214         int i = xennet_rxidx(ri);
215         grant_ref_t ref = queue->grant_rx_ref[i];
216         queue->grant_rx_ref[i] = GRANT_INVALID_REF;
217         return ref;
218 }
219
220 #ifdef CONFIG_SYSFS
221 static const struct attribute_group xennet_dev_group;
222 #endif
223
224 static bool xennet_can_sg(struct net_device *dev)
225 {
226         return dev->features & NETIF_F_SG;
227 }
228
229
230 static void rx_refill_timeout(unsigned long data)
231 {
232         struct netfront_queue *queue = (struct netfront_queue *)data;
233         napi_schedule(&queue->napi);
234 }
235
236 static int netfront_tx_slot_available(struct netfront_queue *queue)
237 {
238         return (queue->tx.req_prod_pvt - queue->tx.rsp_cons) <
239                 (NET_TX_RING_SIZE - MAX_SKB_FRAGS - 2);
240 }
241
242 static void xennet_maybe_wake_tx(struct netfront_queue *queue)
243 {
244         struct net_device *dev = queue->info->netdev;
245         struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, queue->id);
246
247         if (unlikely(netif_tx_queue_stopped(dev_queue)) &&
248             netfront_tx_slot_available(queue) &&
249             likely(netif_running(dev)))
250                 netif_tx_wake_queue(netdev_get_tx_queue(dev, queue->id));
251 }
252
253
254 static struct sk_buff *xennet_alloc_one_rx_buffer(struct netfront_queue *queue)
255 {
256         struct sk_buff *skb;
257         struct page *page;
258
259         skb = __netdev_alloc_skb(queue->info->netdev,
260                                  RX_COPY_THRESHOLD + NET_IP_ALIGN,
261                                  GFP_ATOMIC | __GFP_NOWARN);
262         if (unlikely(!skb))
263                 return NULL;
264
265         page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
266         if (!page) {
267                 kfree_skb(skb);
268                 return NULL;
269         }
270         skb_add_rx_frag(skb, 0, page, 0, 0, PAGE_SIZE);
271
272         /* Align ip header to a 16 bytes boundary */
273         skb_reserve(skb, NET_IP_ALIGN);
274         skb->dev = queue->info->netdev;
275
276         return skb;
277 }
278
279
280 static void xennet_alloc_rx_buffers(struct netfront_queue *queue)
281 {
282         RING_IDX req_prod = queue->rx.req_prod_pvt;
283         int notify;
284         int err = 0;
285
286         if (unlikely(!netif_carrier_ok(queue->info->netdev)))
287                 return;
288
289         for (req_prod = queue->rx.req_prod_pvt;
290              req_prod - queue->rx.rsp_cons < NET_RX_RING_SIZE;
291              req_prod++) {
292                 struct sk_buff *skb;
293                 unsigned short id;
294                 grant_ref_t ref;
295                 struct page *page;
296                 struct xen_netif_rx_request *req;
297
298                 skb = xennet_alloc_one_rx_buffer(queue);
299                 if (!skb) {
300                         err = -ENOMEM;
301                         break;
302                 }
303
304                 id = xennet_rxidx(req_prod);
305
306                 BUG_ON(queue->rx_skbs[id]);
307                 queue->rx_skbs[id] = skb;
308
309                 ref = gnttab_claim_grant_reference(&queue->gref_rx_head);
310                 WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
311                 queue->grant_rx_ref[id] = ref;
312
313                 page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
314
315                 req = RING_GET_REQUEST(&queue->rx, req_prod);
316                 gnttab_page_grant_foreign_access_ref_one(ref,
317                                                          queue->info->xbdev->otherend_id,
318                                                          page,
319                                                          0);
320                 req->id = id;
321                 req->gref = ref;
322         }
323
324         queue->rx.req_prod_pvt = req_prod;
325
326         /* Try again later if there are not enough requests or skb allocation
327          * failed.
328          * Enough requests is quantified as the sum of newly created slots and
329          * the unconsumed slots at the backend.
330          */
331         if (req_prod - queue->rx.rsp_cons < NET_RX_SLOTS_MIN ||
332             unlikely(err)) {
333                 mod_timer(&queue->rx_refill_timer, jiffies + (HZ/10));
334                 return;
335         }
336
337         wmb();          /* barrier so backend seens requests */
338
339         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->rx, notify);
340         if (notify)
341                 notify_remote_via_irq(queue->rx_irq);
342 }
343
344 static int xennet_open(struct net_device *dev)
345 {
346         struct netfront_info *np = netdev_priv(dev);
347         unsigned int num_queues = dev->real_num_tx_queues;
348         unsigned int i = 0;
349         struct netfront_queue *queue = NULL;
350
351         for (i = 0; i < num_queues; ++i) {
352                 queue = &np->queues[i];
353                 napi_enable(&queue->napi);
354
355                 spin_lock_bh(&queue->rx_lock);
356                 if (netif_carrier_ok(dev)) {
357                         xennet_alloc_rx_buffers(queue);
358                         queue->rx.sring->rsp_event = queue->rx.rsp_cons + 1;
359                         if (RING_HAS_UNCONSUMED_RESPONSES(&queue->rx))
360                                 napi_schedule(&queue->napi);
361                 }
362                 spin_unlock_bh(&queue->rx_lock);
363         }
364
365         netif_tx_start_all_queues(dev);
366
367         return 0;
368 }
369
370 static void xennet_tx_buf_gc(struct netfront_queue *queue)
371 {
372         RING_IDX cons, prod;
373         unsigned short id;
374         struct sk_buff *skb;
375         bool more_to_do;
376
377         BUG_ON(!netif_carrier_ok(queue->info->netdev));
378
379         do {
380                 prod = queue->tx.sring->rsp_prod;
381                 rmb(); /* Ensure we see responses up to 'rp'. */
382
383                 for (cons = queue->tx.rsp_cons; cons != prod; cons++) {
384                         struct xen_netif_tx_response *txrsp;
385
386                         txrsp = RING_GET_RESPONSE(&queue->tx, cons);
387                         if (txrsp->status == XEN_NETIF_RSP_NULL)
388                                 continue;
389
390                         id  = txrsp->id;
391                         skb = queue->tx_skbs[id].skb;
392                         if (unlikely(gnttab_query_foreign_access(
393                                 queue->grant_tx_ref[id]) != 0)) {
394                                 pr_alert("%s: warning -- grant still in use by backend domain\n",
395                                          __func__);
396                                 BUG();
397                         }
398                         gnttab_end_foreign_access_ref(
399                                 queue->grant_tx_ref[id], GNTMAP_readonly);
400                         gnttab_release_grant_reference(
401                                 &queue->gref_tx_head, queue->grant_tx_ref[id]);
402                         queue->grant_tx_ref[id] = GRANT_INVALID_REF;
403                         queue->grant_tx_page[id] = NULL;
404                         add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, id);
405                         dev_kfree_skb_irq(skb);
406                 }
407
408                 queue->tx.rsp_cons = prod;
409
410                 RING_FINAL_CHECK_FOR_RESPONSES(&queue->tx, more_to_do);
411         } while (more_to_do);
412
413         xennet_maybe_wake_tx(queue);
414 }
415
416 struct xennet_gnttab_make_txreq {
417         struct netfront_queue *queue;
418         struct sk_buff *skb;
419         struct page *page;
420         struct xen_netif_tx_request *tx; /* Last request */
421         unsigned int size;
422 };
423
424 static void xennet_tx_setup_grant(unsigned long gfn, unsigned int offset,
425                                   unsigned int len, void *data)
426 {
427         struct xennet_gnttab_make_txreq *info = data;
428         unsigned int id;
429         struct xen_netif_tx_request *tx;
430         grant_ref_t ref;
431         /* convenient aliases */
432         struct page *page = info->page;
433         struct netfront_queue *queue = info->queue;
434         struct sk_buff *skb = info->skb;
435
436         id = get_id_from_freelist(&queue->tx_skb_freelist, queue->tx_skbs);
437         tx = RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
438         ref = gnttab_claim_grant_reference(&queue->gref_tx_head);
439         WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
440
441         gnttab_grant_foreign_access_ref(ref, queue->info->xbdev->otherend_id,
442                                         gfn, GNTMAP_readonly);
443
444         queue->tx_skbs[id].skb = skb;
445         queue->grant_tx_page[id] = page;
446         queue->grant_tx_ref[id] = ref;
447
448         tx->id = id;
449         tx->gref = ref;
450         tx->offset = offset;
451         tx->size = len;
452         tx->flags = 0;
453
454         info->tx = tx;
455         info->size += tx->size;
456 }
457
458 static struct xen_netif_tx_request *xennet_make_first_txreq(
459         struct netfront_queue *queue, struct sk_buff *skb,
460         struct page *page, unsigned int offset, unsigned int len)
461 {
462         struct xennet_gnttab_make_txreq info = {
463                 .queue = queue,
464                 .skb = skb,
465                 .page = page,
466                 .size = 0,
467         };
468
469         gnttab_for_one_grant(page, offset, len, xennet_tx_setup_grant, &info);
470
471         return info.tx;
472 }
473
474 static void xennet_make_one_txreq(unsigned long gfn, unsigned int offset,
475                                   unsigned int len, void *data)
476 {
477         struct xennet_gnttab_make_txreq *info = data;
478
479         info->tx->flags |= XEN_NETTXF_more_data;
480         skb_get(info->skb);
481         xennet_tx_setup_grant(gfn, offset, len, data);
482 }
483
484 static struct xen_netif_tx_request *xennet_make_txreqs(
485         struct netfront_queue *queue, struct xen_netif_tx_request *tx,
486         struct sk_buff *skb, struct page *page,
487         unsigned int offset, unsigned int len)
488 {
489         struct xennet_gnttab_make_txreq info = {
490                 .queue = queue,
491                 .skb = skb,
492                 .tx = tx,
493         };
494
495         /* Skip unused frames from start of page */
496         page += offset >> PAGE_SHIFT;
497         offset &= ~PAGE_MASK;
498
499         while (len) {
500                 info.page = page;
501                 info.size = 0;
502
503                 gnttab_foreach_grant_in_range(page, offset, len,
504                                               xennet_make_one_txreq,
505                                               &info);
506
507                 page++;
508                 offset = 0;
509                 len -= info.size;
510         }
511
512         return info.tx;
513 }
514
515 /*
516  * Count how many ring slots are required to send this skb. Each frag
517  * might be a compound page.
518  */
519 static int xennet_count_skb_slots(struct sk_buff *skb)
520 {
521         int i, frags = skb_shinfo(skb)->nr_frags;
522         int slots;
523
524         slots = gnttab_count_grant(offset_in_page(skb->data),
525                                    skb_headlen(skb));
526
527         for (i = 0; i < frags; i++) {
528                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
529                 unsigned long size = skb_frag_size(frag);
530                 unsigned long offset = frag->page_offset;
531
532                 /* Skip unused frames from start of page */
533                 offset &= ~PAGE_MASK;
534
535                 slots += gnttab_count_grant(offset, size);
536         }
537
538         return slots;
539 }
540
541 static u16 xennet_select_queue(struct net_device *dev, struct sk_buff *skb,
542                                void *accel_priv, select_queue_fallback_t fallback)
543 {
544         unsigned int num_queues = dev->real_num_tx_queues;
545         u32 hash;
546         u16 queue_idx;
547
548         /* First, check if there is only one queue */
549         if (num_queues == 1) {
550                 queue_idx = 0;
551         } else {
552                 hash = skb_get_hash(skb);
553                 queue_idx = hash % num_queues;
554         }
555
556         return queue_idx;
557 }
558
559 #define MAX_XEN_SKB_FRAGS (65536 / XEN_PAGE_SIZE + 1)
560
561 static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
562 {
563         struct netfront_info *np = netdev_priv(dev);
564         struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats);
565         struct xen_netif_tx_request *tx, *first_tx;
566         unsigned int i;
567         int notify;
568         int slots;
569         struct page *page;
570         unsigned int offset;
571         unsigned int len;
572         unsigned long flags;
573         struct netfront_queue *queue = NULL;
574         unsigned int num_queues = dev->real_num_tx_queues;
575         u16 queue_index;
576         struct sk_buff *nskb;
577
578         /* Drop the packet if no queues are set up */
579         if (num_queues < 1)
580                 goto drop;
581         /* Determine which queue to transmit this SKB on */
582         queue_index = skb_get_queue_mapping(skb);
583         queue = &np->queues[queue_index];
584
585         /* If skb->len is too big for wire format, drop skb and alert
586          * user about misconfiguration.
587          */
588         if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) {
589                 net_alert_ratelimited(
590                         "xennet: skb->len = %u, too big for wire format\n",
591                         skb->len);
592                 goto drop;
593         }
594
595         slots = xennet_count_skb_slots(skb);
596         if (unlikely(slots > MAX_XEN_SKB_FRAGS + 1)) {
597                 net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n",
598                                     slots, skb->len);
599                 if (skb_linearize(skb))
600                         goto drop;
601         }
602
603         page = virt_to_page(skb->data);
604         offset = offset_in_page(skb->data);
605
606         /* The first req should be at least ETH_HLEN size or the packet will be
607          * dropped by netback.
608          */
609         if (unlikely(PAGE_SIZE - offset < ETH_HLEN)) {
610                 nskb = skb_copy(skb, GFP_ATOMIC);
611                 if (!nskb)
612                         goto drop;
613                 dev_kfree_skb_any(skb);
614                 skb = nskb;
615                 page = virt_to_page(skb->data);
616                 offset = offset_in_page(skb->data);
617         }
618
619         len = skb_headlen(skb);
620
621         spin_lock_irqsave(&queue->tx_lock, flags);
622
623         if (unlikely(!netif_carrier_ok(dev) ||
624                      (slots > 1 && !xennet_can_sg(dev)) ||
625                      netif_needs_gso(skb, netif_skb_features(skb)))) {
626                 spin_unlock_irqrestore(&queue->tx_lock, flags);
627                 goto drop;
628         }
629
630         /* First request for the linear area. */
631         first_tx = tx = xennet_make_first_txreq(queue, skb,
632                                                 page, offset, len);
633         offset += tx->size;
634         if (offset == PAGE_SIZE) {
635                 page++;
636                 offset = 0;
637         }
638         len -= tx->size;
639
640         if (skb->ip_summed == CHECKSUM_PARTIAL)
641                 /* local packet? */
642                 tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated;
643         else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
644                 /* remote but checksummed. */
645                 tx->flags |= XEN_NETTXF_data_validated;
646
647         /* Optional extra info after the first request. */
648         if (skb_shinfo(skb)->gso_size) {
649                 struct xen_netif_extra_info *gso;
650
651                 gso = (struct xen_netif_extra_info *)
652                         RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
653
654                 tx->flags |= XEN_NETTXF_extra_info;
655
656                 gso->u.gso.size = skb_shinfo(skb)->gso_size;
657                 gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ?
658                         XEN_NETIF_GSO_TYPE_TCPV6 :
659                         XEN_NETIF_GSO_TYPE_TCPV4;
660                 gso->u.gso.pad = 0;
661                 gso->u.gso.features = 0;
662
663                 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
664                 gso->flags = 0;
665         }
666
667         /* Requests for the rest of the linear area. */
668         tx = xennet_make_txreqs(queue, tx, skb, page, offset, len);
669
670         /* Requests for all the frags. */
671         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
672                 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
673                 tx = xennet_make_txreqs(queue, tx, skb,
674                                         skb_frag_page(frag), frag->page_offset,
675                                         skb_frag_size(frag));
676         }
677
678         /* First request has the packet length. */
679         first_tx->size = skb->len;
680
681         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
682         if (notify)
683                 notify_remote_via_irq(queue->tx_irq);
684
685         u64_stats_update_begin(&tx_stats->syncp);
686         tx_stats->bytes += skb->len;
687         tx_stats->packets++;
688         u64_stats_update_end(&tx_stats->syncp);
689
690         /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
691         xennet_tx_buf_gc(queue);
692
693         if (!netfront_tx_slot_available(queue))
694                 netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id));
695
696         spin_unlock_irqrestore(&queue->tx_lock, flags);
697
698         return NETDEV_TX_OK;
699
700  drop:
701         dev->stats.tx_dropped++;
702         dev_kfree_skb_any(skb);
703         return NETDEV_TX_OK;
704 }
705
706 static int xennet_close(struct net_device *dev)
707 {
708         struct netfront_info *np = netdev_priv(dev);
709         unsigned int num_queues = dev->real_num_tx_queues;
710         unsigned int i;
711         struct netfront_queue *queue;
712         netif_tx_stop_all_queues(np->netdev);
713         for (i = 0; i < num_queues; ++i) {
714                 queue = &np->queues[i];
715                 napi_disable(&queue->napi);
716         }
717         return 0;
718 }
719
720 static void xennet_move_rx_slot(struct netfront_queue *queue, struct sk_buff *skb,
721                                 grant_ref_t ref)
722 {
723         int new = xennet_rxidx(queue->rx.req_prod_pvt);
724
725         BUG_ON(queue->rx_skbs[new]);
726         queue->rx_skbs[new] = skb;
727         queue->grant_rx_ref[new] = ref;
728         RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->id = new;
729         RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->gref = ref;
730         queue->rx.req_prod_pvt++;
731 }
732
733 static int xennet_get_extras(struct netfront_queue *queue,
734                              struct xen_netif_extra_info *extras,
735                              RING_IDX rp)
736
737 {
738         struct xen_netif_extra_info *extra;
739         struct device *dev = &queue->info->netdev->dev;
740         RING_IDX cons = queue->rx.rsp_cons;
741         int err = 0;
742
743         do {
744                 struct sk_buff *skb;
745                 grant_ref_t ref;
746
747                 if (unlikely(cons + 1 == rp)) {
748                         if (net_ratelimit())
749                                 dev_warn(dev, "Missing extra info\n");
750                         err = -EBADR;
751                         break;
752                 }
753
754                 extra = (struct xen_netif_extra_info *)
755                         RING_GET_RESPONSE(&queue->rx, ++cons);
756
757                 if (unlikely(!extra->type ||
758                              extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
759                         if (net_ratelimit())
760                                 dev_warn(dev, "Invalid extra type: %d\n",
761                                         extra->type);
762                         err = -EINVAL;
763                 } else {
764                         memcpy(&extras[extra->type - 1], extra,
765                                sizeof(*extra));
766                 }
767
768                 skb = xennet_get_rx_skb(queue, cons);
769                 ref = xennet_get_rx_ref(queue, cons);
770                 xennet_move_rx_slot(queue, skb, ref);
771         } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
772
773         queue->rx.rsp_cons = cons;
774         return err;
775 }
776
777 static int xennet_get_responses(struct netfront_queue *queue,
778                                 struct netfront_rx_info *rinfo, RING_IDX rp,
779                                 struct sk_buff_head *list)
780 {
781         struct xen_netif_rx_response *rx = &rinfo->rx;
782         struct xen_netif_extra_info *extras = rinfo->extras;
783         struct device *dev = &queue->info->netdev->dev;
784         RING_IDX cons = queue->rx.rsp_cons;
785         struct sk_buff *skb = xennet_get_rx_skb(queue, cons);
786         grant_ref_t ref = xennet_get_rx_ref(queue, cons);
787         int max = MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD);
788         int slots = 1;
789         int err = 0;
790         unsigned long ret;
791
792         if (rx->flags & XEN_NETRXF_extra_info) {
793                 err = xennet_get_extras(queue, extras, rp);
794                 cons = queue->rx.rsp_cons;
795         }
796
797         for (;;) {
798                 if (unlikely(rx->status < 0 ||
799                              rx->offset + rx->status > XEN_PAGE_SIZE)) {
800                         if (net_ratelimit())
801                                 dev_warn(dev, "rx->offset: %u, size: %d\n",
802                                          rx->offset, rx->status);
803                         xennet_move_rx_slot(queue, skb, ref);
804                         err = -EINVAL;
805                         goto next;
806                 }
807
808                 /*
809                  * This definitely indicates a bug, either in this driver or in
810                  * the backend driver. In future this should flag the bad
811                  * situation to the system controller to reboot the backend.
812                  */
813                 if (ref == GRANT_INVALID_REF) {
814                         if (net_ratelimit())
815                                 dev_warn(dev, "Bad rx response id %d.\n",
816                                          rx->id);
817                         err = -EINVAL;
818                         goto next;
819                 }
820
821                 ret = gnttab_end_foreign_access_ref(ref, 0);
822                 BUG_ON(!ret);
823
824                 gnttab_release_grant_reference(&queue->gref_rx_head, ref);
825
826                 __skb_queue_tail(list, skb);
827
828 next:
829                 if (!(rx->flags & XEN_NETRXF_more_data))
830                         break;
831
832                 if (cons + slots == rp) {
833                         if (net_ratelimit())
834                                 dev_warn(dev, "Need more slots\n");
835                         err = -ENOENT;
836                         break;
837                 }
838
839                 rx = RING_GET_RESPONSE(&queue->rx, cons + slots);
840                 skb = xennet_get_rx_skb(queue, cons + slots);
841                 ref = xennet_get_rx_ref(queue, cons + slots);
842                 slots++;
843         }
844
845         if (unlikely(slots > max)) {
846                 if (net_ratelimit())
847                         dev_warn(dev, "Too many slots\n");
848                 err = -E2BIG;
849         }
850
851         if (unlikely(err))
852                 queue->rx.rsp_cons = cons + slots;
853
854         return err;
855 }
856
857 static int xennet_set_skb_gso(struct sk_buff *skb,
858                               struct xen_netif_extra_info *gso)
859 {
860         if (!gso->u.gso.size) {
861                 if (net_ratelimit())
862                         pr_warn("GSO size must not be zero\n");
863                 return -EINVAL;
864         }
865
866         if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 &&
867             gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) {
868                 if (net_ratelimit())
869                         pr_warn("Bad GSO type %d\n", gso->u.gso.type);
870                 return -EINVAL;
871         }
872
873         skb_shinfo(skb)->gso_size = gso->u.gso.size;
874         skb_shinfo(skb)->gso_type =
875                 (gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ?
876                 SKB_GSO_TCPV4 :
877                 SKB_GSO_TCPV6;
878
879         /* Header must be checked, and gso_segs computed. */
880         skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
881         skb_shinfo(skb)->gso_segs = 0;
882
883         return 0;
884 }
885
886 static RING_IDX xennet_fill_frags(struct netfront_queue *queue,
887                                   struct sk_buff *skb,
888                                   struct sk_buff_head *list)
889 {
890         struct skb_shared_info *shinfo = skb_shinfo(skb);
891         RING_IDX cons = queue->rx.rsp_cons;
892         struct sk_buff *nskb;
893
894         while ((nskb = __skb_dequeue(list))) {
895                 struct xen_netif_rx_response *rx =
896                         RING_GET_RESPONSE(&queue->rx, ++cons);
897                 skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0];
898
899                 if (shinfo->nr_frags == MAX_SKB_FRAGS) {
900                         unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
901
902                         BUG_ON(pull_to <= skb_headlen(skb));
903                         __pskb_pull_tail(skb, pull_to - skb_headlen(skb));
904                 }
905                 BUG_ON(shinfo->nr_frags >= MAX_SKB_FRAGS);
906
907                 skb_add_rx_frag(skb, shinfo->nr_frags, skb_frag_page(nfrag),
908                                 rx->offset, rx->status, PAGE_SIZE);
909
910                 skb_shinfo(nskb)->nr_frags = 0;
911                 kfree_skb(nskb);
912         }
913
914         return cons;
915 }
916
917 static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
918 {
919         bool recalculate_partial_csum = false;
920
921         /*
922          * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
923          * peers can fail to set NETRXF_csum_blank when sending a GSO
924          * frame. In this case force the SKB to CHECKSUM_PARTIAL and
925          * recalculate the partial checksum.
926          */
927         if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
928                 struct netfront_info *np = netdev_priv(dev);
929                 atomic_inc(&np->rx_gso_checksum_fixup);
930                 skb->ip_summed = CHECKSUM_PARTIAL;
931                 recalculate_partial_csum = true;
932         }
933
934         /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
935         if (skb->ip_summed != CHECKSUM_PARTIAL)
936                 return 0;
937
938         return skb_checksum_setup(skb, recalculate_partial_csum);
939 }
940
941 static int handle_incoming_queue(struct netfront_queue *queue,
942                                  struct sk_buff_head *rxq)
943 {
944         struct netfront_stats *rx_stats = this_cpu_ptr(queue->info->rx_stats);
945         int packets_dropped = 0;
946         struct sk_buff *skb;
947
948         while ((skb = __skb_dequeue(rxq)) != NULL) {
949                 int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
950
951                 if (pull_to > skb_headlen(skb))
952                         __pskb_pull_tail(skb, pull_to - skb_headlen(skb));
953
954                 /* Ethernet work: Delayed to here as it peeks the header. */
955                 skb->protocol = eth_type_trans(skb, queue->info->netdev);
956                 skb_reset_network_header(skb);
957
958                 if (checksum_setup(queue->info->netdev, skb)) {
959                         kfree_skb(skb);
960                         packets_dropped++;
961                         queue->info->netdev->stats.rx_errors++;
962                         continue;
963                 }
964
965                 u64_stats_update_begin(&rx_stats->syncp);
966                 rx_stats->packets++;
967                 rx_stats->bytes += skb->len;
968                 u64_stats_update_end(&rx_stats->syncp);
969
970                 /* Pass it up. */
971                 napi_gro_receive(&queue->napi, skb);
972         }
973
974         return packets_dropped;
975 }
976
977 static int xennet_poll(struct napi_struct *napi, int budget)
978 {
979         struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi);
980         struct net_device *dev = queue->info->netdev;
981         struct sk_buff *skb;
982         struct netfront_rx_info rinfo;
983         struct xen_netif_rx_response *rx = &rinfo.rx;
984         struct xen_netif_extra_info *extras = rinfo.extras;
985         RING_IDX i, rp;
986         int work_done;
987         struct sk_buff_head rxq;
988         struct sk_buff_head errq;
989         struct sk_buff_head tmpq;
990         int err;
991
992         spin_lock(&queue->rx_lock);
993
994         skb_queue_head_init(&rxq);
995         skb_queue_head_init(&errq);
996         skb_queue_head_init(&tmpq);
997
998         rp = queue->rx.sring->rsp_prod;
999         rmb(); /* Ensure we see queued responses up to 'rp'. */
1000
1001         i = queue->rx.rsp_cons;
1002         work_done = 0;
1003         while ((i != rp) && (work_done < budget)) {
1004                 memcpy(rx, RING_GET_RESPONSE(&queue->rx, i), sizeof(*rx));
1005                 memset(extras, 0, sizeof(rinfo.extras));
1006
1007                 err = xennet_get_responses(queue, &rinfo, rp, &tmpq);
1008
1009                 if (unlikely(err)) {
1010 err:
1011                         while ((skb = __skb_dequeue(&tmpq)))
1012                                 __skb_queue_tail(&errq, skb);
1013                         dev->stats.rx_errors++;
1014                         i = queue->rx.rsp_cons;
1015                         continue;
1016                 }
1017
1018                 skb = __skb_dequeue(&tmpq);
1019
1020                 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1021                         struct xen_netif_extra_info *gso;
1022                         gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1023
1024                         if (unlikely(xennet_set_skb_gso(skb, gso))) {
1025                                 __skb_queue_head(&tmpq, skb);
1026                                 queue->rx.rsp_cons += skb_queue_len(&tmpq);
1027                                 goto err;
1028                         }
1029                 }
1030
1031                 NETFRONT_SKB_CB(skb)->pull_to = rx->status;
1032                 if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD)
1033                         NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD;
1034
1035                 skb_shinfo(skb)->frags[0].page_offset = rx->offset;
1036                 skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status);
1037                 skb->data_len = rx->status;
1038                 skb->len += rx->status;
1039
1040                 i = xennet_fill_frags(queue, skb, &tmpq);
1041
1042                 if (rx->flags & XEN_NETRXF_csum_blank)
1043                         skb->ip_summed = CHECKSUM_PARTIAL;
1044                 else if (rx->flags & XEN_NETRXF_data_validated)
1045                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1046
1047                 __skb_queue_tail(&rxq, skb);
1048
1049                 queue->rx.rsp_cons = ++i;
1050                 work_done++;
1051         }
1052
1053         __skb_queue_purge(&errq);
1054
1055         work_done -= handle_incoming_queue(queue, &rxq);
1056
1057         xennet_alloc_rx_buffers(queue);
1058
1059         if (work_done < budget) {
1060                 int more_to_do = 0;
1061
1062                 napi_complete_done(napi, work_done);
1063
1064                 RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do);
1065                 if (more_to_do)
1066                         napi_schedule(napi);
1067         }
1068
1069         spin_unlock(&queue->rx_lock);
1070
1071         return work_done;
1072 }
1073
1074 static int xennet_change_mtu(struct net_device *dev, int mtu)
1075 {
1076         int max = xennet_can_sg(dev) ? XEN_NETIF_MAX_TX_SIZE : ETH_DATA_LEN;
1077
1078         if (mtu > max)
1079                 return -EINVAL;
1080         dev->mtu = mtu;
1081         return 0;
1082 }
1083
1084 static void xennet_get_stats64(struct net_device *dev,
1085                                struct rtnl_link_stats64 *tot)
1086 {
1087         struct netfront_info *np = netdev_priv(dev);
1088         int cpu;
1089
1090         for_each_possible_cpu(cpu) {
1091                 struct netfront_stats *rx_stats = per_cpu_ptr(np->rx_stats, cpu);
1092                 struct netfront_stats *tx_stats = per_cpu_ptr(np->tx_stats, cpu);
1093                 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1094                 unsigned int start;
1095
1096                 do {
1097                         start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
1098                         tx_packets = tx_stats->packets;
1099                         tx_bytes = tx_stats->bytes;
1100                 } while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
1101
1102                 do {
1103                         start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
1104                         rx_packets = rx_stats->packets;
1105                         rx_bytes = rx_stats->bytes;
1106                 } while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
1107
1108                 tot->rx_packets += rx_packets;
1109                 tot->tx_packets += tx_packets;
1110                 tot->rx_bytes   += rx_bytes;
1111                 tot->tx_bytes   += tx_bytes;
1112         }
1113
1114         tot->rx_errors  = dev->stats.rx_errors;
1115         tot->tx_dropped = dev->stats.tx_dropped;
1116 }
1117
1118 static void xennet_release_tx_bufs(struct netfront_queue *queue)
1119 {
1120         struct sk_buff *skb;
1121         int i;
1122
1123         for (i = 0; i < NET_TX_RING_SIZE; i++) {
1124                 /* Skip over entries which are actually freelist references */
1125                 if (skb_entry_is_link(&queue->tx_skbs[i]))
1126                         continue;
1127
1128                 skb = queue->tx_skbs[i].skb;
1129                 get_page(queue->grant_tx_page[i]);
1130                 gnttab_end_foreign_access(queue->grant_tx_ref[i],
1131                                           GNTMAP_readonly,
1132                                           (unsigned long)page_address(queue->grant_tx_page[i]));
1133                 queue->grant_tx_page[i] = NULL;
1134                 queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1135                 add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, i);
1136                 dev_kfree_skb_irq(skb);
1137         }
1138 }
1139
1140 static void xennet_release_rx_bufs(struct netfront_queue *queue)
1141 {
1142         int id, ref;
1143
1144         spin_lock_bh(&queue->rx_lock);
1145
1146         for (id = 0; id < NET_RX_RING_SIZE; id++) {
1147                 struct sk_buff *skb;
1148                 struct page *page;
1149
1150                 skb = queue->rx_skbs[id];
1151                 if (!skb)
1152                         continue;
1153
1154                 ref = queue->grant_rx_ref[id];
1155                 if (ref == GRANT_INVALID_REF)
1156                         continue;
1157
1158                 page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
1159
1160                 /* gnttab_end_foreign_access() needs a page ref until
1161                  * foreign access is ended (which may be deferred).
1162                  */
1163                 get_page(page);
1164                 gnttab_end_foreign_access(ref, 0,
1165                                           (unsigned long)page_address(page));
1166                 queue->grant_rx_ref[id] = GRANT_INVALID_REF;
1167
1168                 kfree_skb(skb);
1169         }
1170
1171         spin_unlock_bh(&queue->rx_lock);
1172 }
1173
1174 static netdev_features_t xennet_fix_features(struct net_device *dev,
1175         netdev_features_t features)
1176 {
1177         struct netfront_info *np = netdev_priv(dev);
1178
1179         if (features & NETIF_F_SG &&
1180             !xenbus_read_unsigned(np->xbdev->otherend, "feature-sg", 0))
1181                 features &= ~NETIF_F_SG;
1182
1183         if (features & NETIF_F_IPV6_CSUM &&
1184             !xenbus_read_unsigned(np->xbdev->otherend,
1185                                   "feature-ipv6-csum-offload", 0))
1186                 features &= ~NETIF_F_IPV6_CSUM;
1187
1188         if (features & NETIF_F_TSO &&
1189             !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv4", 0))
1190                 features &= ~NETIF_F_TSO;
1191
1192         if (features & NETIF_F_TSO6 &&
1193             !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv6", 0))
1194                 features &= ~NETIF_F_TSO6;
1195
1196         return features;
1197 }
1198
1199 static int xennet_set_features(struct net_device *dev,
1200         netdev_features_t features)
1201 {
1202         if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1203                 netdev_info(dev, "Reducing MTU because no SG offload");
1204                 dev->mtu = ETH_DATA_LEN;
1205         }
1206
1207         return 0;
1208 }
1209
1210 static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id)
1211 {
1212         struct netfront_queue *queue = dev_id;
1213         unsigned long flags;
1214
1215         spin_lock_irqsave(&queue->tx_lock, flags);
1216         xennet_tx_buf_gc(queue);
1217         spin_unlock_irqrestore(&queue->tx_lock, flags);
1218
1219         return IRQ_HANDLED;
1220 }
1221
1222 static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id)
1223 {
1224         struct netfront_queue *queue = dev_id;
1225         struct net_device *dev = queue->info->netdev;
1226
1227         if (likely(netif_carrier_ok(dev) &&
1228                    RING_HAS_UNCONSUMED_RESPONSES(&queue->rx)))
1229                 napi_schedule(&queue->napi);
1230
1231         return IRQ_HANDLED;
1232 }
1233
1234 static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1235 {
1236         xennet_tx_interrupt(irq, dev_id);
1237         xennet_rx_interrupt(irq, dev_id);
1238         return IRQ_HANDLED;
1239 }
1240
1241 #ifdef CONFIG_NET_POLL_CONTROLLER
1242 static void xennet_poll_controller(struct net_device *dev)
1243 {
1244         /* Poll each queue */
1245         struct netfront_info *info = netdev_priv(dev);
1246         unsigned int num_queues = dev->real_num_tx_queues;
1247         unsigned int i;
1248         for (i = 0; i < num_queues; ++i)
1249                 xennet_interrupt(0, &info->queues[i]);
1250 }
1251 #endif
1252
1253 static const struct net_device_ops xennet_netdev_ops = {
1254         .ndo_open            = xennet_open,
1255         .ndo_stop            = xennet_close,
1256         .ndo_start_xmit      = xennet_start_xmit,
1257         .ndo_change_mtu      = xennet_change_mtu,
1258         .ndo_get_stats64     = xennet_get_stats64,
1259         .ndo_set_mac_address = eth_mac_addr,
1260         .ndo_validate_addr   = eth_validate_addr,
1261         .ndo_fix_features    = xennet_fix_features,
1262         .ndo_set_features    = xennet_set_features,
1263         .ndo_select_queue    = xennet_select_queue,
1264 #ifdef CONFIG_NET_POLL_CONTROLLER
1265         .ndo_poll_controller = xennet_poll_controller,
1266 #endif
1267 };
1268
1269 static void xennet_free_netdev(struct net_device *netdev)
1270 {
1271         struct netfront_info *np = netdev_priv(netdev);
1272
1273         free_percpu(np->rx_stats);
1274         free_percpu(np->tx_stats);
1275         free_netdev(netdev);
1276 }
1277
1278 static struct net_device *xennet_create_dev(struct xenbus_device *dev)
1279 {
1280         int err;
1281         struct net_device *netdev;
1282         struct netfront_info *np;
1283
1284         netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues);
1285         if (!netdev)
1286                 return ERR_PTR(-ENOMEM);
1287
1288         np                   = netdev_priv(netdev);
1289         np->xbdev            = dev;
1290
1291         np->queues = NULL;
1292
1293         err = -ENOMEM;
1294         np->rx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1295         if (np->rx_stats == NULL)
1296                 goto exit;
1297         np->tx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1298         if (np->tx_stats == NULL)
1299                 goto exit;
1300
1301         netdev->netdev_ops      = &xennet_netdev_ops;
1302
1303         netdev->features        = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1304                                   NETIF_F_GSO_ROBUST;
1305         netdev->hw_features     = NETIF_F_SG |
1306                                   NETIF_F_IPV6_CSUM |
1307                                   NETIF_F_TSO | NETIF_F_TSO6;
1308
1309         /*
1310          * Assume that all hw features are available for now. This set
1311          * will be adjusted by the call to netdev_update_features() in
1312          * xennet_connect() which is the earliest point where we can
1313          * negotiate with the backend regarding supported features.
1314          */
1315         netdev->features |= netdev->hw_features;
1316
1317         netdev->ethtool_ops = &xennet_ethtool_ops;
1318         netdev->min_mtu = 0;
1319         netdev->max_mtu = XEN_NETIF_MAX_TX_SIZE;
1320         SET_NETDEV_DEV(netdev, &dev->dev);
1321
1322         np->netdev = netdev;
1323
1324         netif_carrier_off(netdev);
1325
1326         return netdev;
1327
1328  exit:
1329         xennet_free_netdev(netdev);
1330         return ERR_PTR(err);
1331 }
1332
1333 /**
1334  * Entry point to this code when a new device is created.  Allocate the basic
1335  * structures and the ring buffers for communication with the backend, and
1336  * inform the backend of the appropriate details for those.
1337  */
1338 static int netfront_probe(struct xenbus_device *dev,
1339                           const struct xenbus_device_id *id)
1340 {
1341         int err;
1342         struct net_device *netdev;
1343         struct netfront_info *info;
1344
1345         netdev = xennet_create_dev(dev);
1346         if (IS_ERR(netdev)) {
1347                 err = PTR_ERR(netdev);
1348                 xenbus_dev_fatal(dev, err, "creating netdev");
1349                 return err;
1350         }
1351
1352         info = netdev_priv(netdev);
1353         dev_set_drvdata(&dev->dev, info);
1354 #ifdef CONFIG_SYSFS
1355         info->netdev->sysfs_groups[0] = &xennet_dev_group;
1356 #endif
1357         err = register_netdev(info->netdev);
1358         if (err) {
1359                 pr_warn("%s: register_netdev err=%d\n", __func__, err);
1360                 goto fail;
1361         }
1362
1363         return 0;
1364
1365  fail:
1366         xennet_free_netdev(netdev);
1367         dev_set_drvdata(&dev->dev, NULL);
1368         return err;
1369 }
1370
1371 static void xennet_end_access(int ref, void *page)
1372 {
1373         /* This frees the page as a side-effect */
1374         if (ref != GRANT_INVALID_REF)
1375                 gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1376 }
1377
1378 static void xennet_disconnect_backend(struct netfront_info *info)
1379 {
1380         unsigned int i = 0;
1381         unsigned int num_queues = info->netdev->real_num_tx_queues;
1382
1383         netif_carrier_off(info->netdev);
1384
1385         for (i = 0; i < num_queues && info->queues; ++i) {
1386                 struct netfront_queue *queue = &info->queues[i];
1387
1388                 del_timer_sync(&queue->rx_refill_timer);
1389
1390                 if (queue->tx_irq && (queue->tx_irq == queue->rx_irq))
1391                         unbind_from_irqhandler(queue->tx_irq, queue);
1392                 if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) {
1393                         unbind_from_irqhandler(queue->tx_irq, queue);
1394                         unbind_from_irqhandler(queue->rx_irq, queue);
1395                 }
1396                 queue->tx_evtchn = queue->rx_evtchn = 0;
1397                 queue->tx_irq = queue->rx_irq = 0;
1398
1399                 if (netif_running(info->netdev))
1400                         napi_synchronize(&queue->napi);
1401
1402                 xennet_release_tx_bufs(queue);
1403                 xennet_release_rx_bufs(queue);
1404                 gnttab_free_grant_references(queue->gref_tx_head);
1405                 gnttab_free_grant_references(queue->gref_rx_head);
1406
1407                 /* End access and free the pages */
1408                 xennet_end_access(queue->tx_ring_ref, queue->tx.sring);
1409                 xennet_end_access(queue->rx_ring_ref, queue->rx.sring);
1410
1411                 queue->tx_ring_ref = GRANT_INVALID_REF;
1412                 queue->rx_ring_ref = GRANT_INVALID_REF;
1413                 queue->tx.sring = NULL;
1414                 queue->rx.sring = NULL;
1415         }
1416 }
1417
1418 /**
1419  * We are reconnecting to the backend, due to a suspend/resume, or a backend
1420  * driver restart.  We tear down our netif structure and recreate it, but
1421  * leave the device-layer structures intact so that this is transparent to the
1422  * rest of the kernel.
1423  */
1424 static int netfront_resume(struct xenbus_device *dev)
1425 {
1426         struct netfront_info *info = dev_get_drvdata(&dev->dev);
1427
1428         dev_dbg(&dev->dev, "%s\n", dev->nodename);
1429
1430         xennet_disconnect_backend(info);
1431         return 0;
1432 }
1433
1434 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1435 {
1436         char *s, *e, *macstr;
1437         int i;
1438
1439         macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1440         if (IS_ERR(macstr))
1441                 return PTR_ERR(macstr);
1442
1443         for (i = 0; i < ETH_ALEN; i++) {
1444                 mac[i] = simple_strtoul(s, &e, 16);
1445                 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1446                         kfree(macstr);
1447                         return -ENOENT;
1448                 }
1449                 s = e+1;
1450         }
1451
1452         kfree(macstr);
1453         return 0;
1454 }
1455
1456 static int setup_netfront_single(struct netfront_queue *queue)
1457 {
1458         int err;
1459
1460         err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1461         if (err < 0)
1462                 goto fail;
1463
1464         err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1465                                         xennet_interrupt,
1466                                         0, queue->info->netdev->name, queue);
1467         if (err < 0)
1468                 goto bind_fail;
1469         queue->rx_evtchn = queue->tx_evtchn;
1470         queue->rx_irq = queue->tx_irq = err;
1471
1472         return 0;
1473
1474 bind_fail:
1475         xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1476         queue->tx_evtchn = 0;
1477 fail:
1478         return err;
1479 }
1480
1481 static int setup_netfront_split(struct netfront_queue *queue)
1482 {
1483         int err;
1484
1485         err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1486         if (err < 0)
1487                 goto fail;
1488         err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn);
1489         if (err < 0)
1490                 goto alloc_rx_evtchn_fail;
1491
1492         snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
1493                  "%s-tx", queue->name);
1494         err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1495                                         xennet_tx_interrupt,
1496                                         0, queue->tx_irq_name, queue);
1497         if (err < 0)
1498                 goto bind_tx_fail;
1499         queue->tx_irq = err;
1500
1501         snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
1502                  "%s-rx", queue->name);
1503         err = bind_evtchn_to_irqhandler(queue->rx_evtchn,
1504                                         xennet_rx_interrupt,
1505                                         0, queue->rx_irq_name, queue);
1506         if (err < 0)
1507                 goto bind_rx_fail;
1508         queue->rx_irq = err;
1509
1510         return 0;
1511
1512 bind_rx_fail:
1513         unbind_from_irqhandler(queue->tx_irq, queue);
1514         queue->tx_irq = 0;
1515 bind_tx_fail:
1516         xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn);
1517         queue->rx_evtchn = 0;
1518 alloc_rx_evtchn_fail:
1519         xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1520         queue->tx_evtchn = 0;
1521 fail:
1522         return err;
1523 }
1524
1525 static int setup_netfront(struct xenbus_device *dev,
1526                         struct netfront_queue *queue, unsigned int feature_split_evtchn)
1527 {
1528         struct xen_netif_tx_sring *txs;
1529         struct xen_netif_rx_sring *rxs;
1530         grant_ref_t gref;
1531         int err;
1532
1533         queue->tx_ring_ref = GRANT_INVALID_REF;
1534         queue->rx_ring_ref = GRANT_INVALID_REF;
1535         queue->rx.sring = NULL;
1536         queue->tx.sring = NULL;
1537
1538         txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1539         if (!txs) {
1540                 err = -ENOMEM;
1541                 xenbus_dev_fatal(dev, err, "allocating tx ring page");
1542                 goto fail;
1543         }
1544         SHARED_RING_INIT(txs);
1545         FRONT_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE);
1546
1547         err = xenbus_grant_ring(dev, txs, 1, &gref);
1548         if (err < 0)
1549                 goto grant_tx_ring_fail;
1550         queue->tx_ring_ref = gref;
1551
1552         rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1553         if (!rxs) {
1554                 err = -ENOMEM;
1555                 xenbus_dev_fatal(dev, err, "allocating rx ring page");
1556                 goto alloc_rx_ring_fail;
1557         }
1558         SHARED_RING_INIT(rxs);
1559         FRONT_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE);
1560
1561         err = xenbus_grant_ring(dev, rxs, 1, &gref);
1562         if (err < 0)
1563                 goto grant_rx_ring_fail;
1564         queue->rx_ring_ref = gref;
1565
1566         if (feature_split_evtchn)
1567                 err = setup_netfront_split(queue);
1568         /* setup single event channel if
1569          *  a) feature-split-event-channels == 0
1570          *  b) feature-split-event-channels == 1 but failed to setup
1571          */
1572         if (!feature_split_evtchn || (feature_split_evtchn && err))
1573                 err = setup_netfront_single(queue);
1574
1575         if (err)
1576                 goto alloc_evtchn_fail;
1577
1578         return 0;
1579
1580         /* If we fail to setup netfront, it is safe to just revoke access to
1581          * granted pages because backend is not accessing it at this point.
1582          */
1583 alloc_evtchn_fail:
1584         gnttab_end_foreign_access_ref(queue->rx_ring_ref, 0);
1585 grant_rx_ring_fail:
1586         free_page((unsigned long)rxs);
1587 alloc_rx_ring_fail:
1588         gnttab_end_foreign_access_ref(queue->tx_ring_ref, 0);
1589 grant_tx_ring_fail:
1590         free_page((unsigned long)txs);
1591 fail:
1592         return err;
1593 }
1594
1595 /* Queue-specific initialisation
1596  * This used to be done in xennet_create_dev() but must now
1597  * be run per-queue.
1598  */
1599 static int xennet_init_queue(struct netfront_queue *queue)
1600 {
1601         unsigned short i;
1602         int err = 0;
1603
1604         spin_lock_init(&queue->tx_lock);
1605         spin_lock_init(&queue->rx_lock);
1606
1607         setup_timer(&queue->rx_refill_timer, rx_refill_timeout,
1608                     (unsigned long)queue);
1609
1610         snprintf(queue->name, sizeof(queue->name), "%s-q%u",
1611                  queue->info->netdev->name, queue->id);
1612
1613         /* Initialise tx_skbs as a free chain containing every entry. */
1614         queue->tx_skb_freelist = 0;
1615         for (i = 0; i < NET_TX_RING_SIZE; i++) {
1616                 skb_entry_set_link(&queue->tx_skbs[i], i+1);
1617                 queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1618                 queue->grant_tx_page[i] = NULL;
1619         }
1620
1621         /* Clear out rx_skbs */
1622         for (i = 0; i < NET_RX_RING_SIZE; i++) {
1623                 queue->rx_skbs[i] = NULL;
1624                 queue->grant_rx_ref[i] = GRANT_INVALID_REF;
1625         }
1626
1627         /* A grant for every tx ring slot */
1628         if (gnttab_alloc_grant_references(NET_TX_RING_SIZE,
1629                                           &queue->gref_tx_head) < 0) {
1630                 pr_alert("can't alloc tx grant refs\n");
1631                 err = -ENOMEM;
1632                 goto exit;
1633         }
1634
1635         /* A grant for every rx ring slot */
1636         if (gnttab_alloc_grant_references(NET_RX_RING_SIZE,
1637                                           &queue->gref_rx_head) < 0) {
1638                 pr_alert("can't alloc rx grant refs\n");
1639                 err = -ENOMEM;
1640                 goto exit_free_tx;
1641         }
1642
1643         return 0;
1644
1645  exit_free_tx:
1646         gnttab_free_grant_references(queue->gref_tx_head);
1647  exit:
1648         return err;
1649 }
1650
1651 static int write_queue_xenstore_keys(struct netfront_queue *queue,
1652                            struct xenbus_transaction *xbt, int write_hierarchical)
1653 {
1654         /* Write the queue-specific keys into XenStore in the traditional
1655          * way for a single queue, or in a queue subkeys for multiple
1656          * queues.
1657          */
1658         struct xenbus_device *dev = queue->info->xbdev;
1659         int err;
1660         const char *message;
1661         char *path;
1662         size_t pathsize;
1663
1664         /* Choose the correct place to write the keys */
1665         if (write_hierarchical) {
1666                 pathsize = strlen(dev->nodename) + 10;
1667                 path = kzalloc(pathsize, GFP_KERNEL);
1668                 if (!path) {
1669                         err = -ENOMEM;
1670                         message = "out of memory while writing ring references";
1671                         goto error;
1672                 }
1673                 snprintf(path, pathsize, "%s/queue-%u",
1674                                 dev->nodename, queue->id);
1675         } else {
1676                 path = (char *)dev->nodename;
1677         }
1678
1679         /* Write ring references */
1680         err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u",
1681                         queue->tx_ring_ref);
1682         if (err) {
1683                 message = "writing tx-ring-ref";
1684                 goto error;
1685         }
1686
1687         err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u",
1688                         queue->rx_ring_ref);
1689         if (err) {
1690                 message = "writing rx-ring-ref";
1691                 goto error;
1692         }
1693
1694         /* Write event channels; taking into account both shared
1695          * and split event channel scenarios.
1696          */
1697         if (queue->tx_evtchn == queue->rx_evtchn) {
1698                 /* Shared event channel */
1699                 err = xenbus_printf(*xbt, path,
1700                                 "event-channel", "%u", queue->tx_evtchn);
1701                 if (err) {
1702                         message = "writing event-channel";
1703                         goto error;
1704                 }
1705         } else {
1706                 /* Split event channels */
1707                 err = xenbus_printf(*xbt, path,
1708                                 "event-channel-tx", "%u", queue->tx_evtchn);
1709                 if (err) {
1710                         message = "writing event-channel-tx";
1711                         goto error;
1712                 }
1713
1714                 err = xenbus_printf(*xbt, path,
1715                                 "event-channel-rx", "%u", queue->rx_evtchn);
1716                 if (err) {
1717                         message = "writing event-channel-rx";
1718                         goto error;
1719                 }
1720         }
1721
1722         if (write_hierarchical)
1723                 kfree(path);
1724         return 0;
1725
1726 error:
1727         if (write_hierarchical)
1728                 kfree(path);
1729         xenbus_dev_fatal(dev, err, "%s", message);
1730         return err;
1731 }
1732
1733 static void xennet_destroy_queues(struct netfront_info *info)
1734 {
1735         unsigned int i;
1736
1737         rtnl_lock();
1738
1739         for (i = 0; i < info->netdev->real_num_tx_queues; i++) {
1740                 struct netfront_queue *queue = &info->queues[i];
1741
1742                 if (netif_running(info->netdev))
1743                         napi_disable(&queue->napi);
1744                 netif_napi_del(&queue->napi);
1745         }
1746
1747         rtnl_unlock();
1748
1749         kfree(info->queues);
1750         info->queues = NULL;
1751 }
1752
1753 static int xennet_create_queues(struct netfront_info *info,
1754                                 unsigned int *num_queues)
1755 {
1756         unsigned int i;
1757         int ret;
1758
1759         info->queues = kcalloc(*num_queues, sizeof(struct netfront_queue),
1760                                GFP_KERNEL);
1761         if (!info->queues)
1762                 return -ENOMEM;
1763
1764         rtnl_lock();
1765
1766         for (i = 0; i < *num_queues; i++) {
1767                 struct netfront_queue *queue = &info->queues[i];
1768
1769                 queue->id = i;
1770                 queue->info = info;
1771
1772                 ret = xennet_init_queue(queue);
1773                 if (ret < 0) {
1774                         dev_warn(&info->netdev->dev,
1775                                  "only created %d queues\n", i);
1776                         *num_queues = i;
1777                         break;
1778                 }
1779
1780                 netif_napi_add(queue->info->netdev, &queue->napi,
1781                                xennet_poll, 64);
1782                 if (netif_running(info->netdev))
1783                         napi_enable(&queue->napi);
1784         }
1785
1786         netif_set_real_num_tx_queues(info->netdev, *num_queues);
1787
1788         rtnl_unlock();
1789
1790         if (*num_queues == 0) {
1791                 dev_err(&info->netdev->dev, "no queues\n");
1792                 return -EINVAL;
1793         }
1794         return 0;
1795 }
1796
1797 /* Common code used when first setting up, and when resuming. */
1798 static int talk_to_netback(struct xenbus_device *dev,
1799                            struct netfront_info *info)
1800 {
1801         const char *message;
1802         struct xenbus_transaction xbt;
1803         int err;
1804         unsigned int feature_split_evtchn;
1805         unsigned int i = 0;
1806         unsigned int max_queues = 0;
1807         struct netfront_queue *queue = NULL;
1808         unsigned int num_queues = 1;
1809
1810         info->netdev->irq = 0;
1811
1812         /* Check if backend supports multiple queues */
1813         max_queues = xenbus_read_unsigned(info->xbdev->otherend,
1814                                           "multi-queue-max-queues", 1);
1815         num_queues = min(max_queues, xennet_max_queues);
1816
1817         /* Check feature-split-event-channels */
1818         feature_split_evtchn = xenbus_read_unsigned(info->xbdev->otherend,
1819                                         "feature-split-event-channels", 0);
1820
1821         /* Read mac addr. */
1822         err = xen_net_read_mac(dev, info->netdev->dev_addr);
1823         if (err) {
1824                 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
1825                 goto out;
1826         }
1827
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
1839         /* Create shared ring, alloc event channel -- for each queue */
1840         for (i = 0; i < num_queues; ++i) {
1841                 queue = &info->queues[i];
1842                 err = setup_netfront(dev, queue, feature_split_evtchn);
1843                 if (err)
1844                         goto destroy_ring;
1845         }
1846
1847 again:
1848         err = xenbus_transaction_start(&xbt);
1849         if (err) {
1850                 xenbus_dev_fatal(dev, err, "starting transaction");
1851                 goto destroy_ring;
1852         }
1853
1854         if (xenbus_exists(XBT_NIL,
1855                           info->xbdev->otherend, "multi-queue-max-queues")) {
1856                 /* Write the number of queues */
1857                 err = xenbus_printf(xbt, dev->nodename,
1858                                     "multi-queue-num-queues", "%u", num_queues);
1859                 if (err) {
1860                         message = "writing multi-queue-num-queues";
1861                         goto abort_transaction_no_dev_fatal;
1862                 }
1863         }
1864
1865         if (num_queues == 1) {
1866                 err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */
1867                 if (err)
1868                         goto abort_transaction_no_dev_fatal;
1869         } else {
1870                 /* Write the keys for each queue */
1871                 for (i = 0; i < num_queues; ++i) {
1872                         queue = &info->queues[i];
1873                         err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */
1874                         if (err)
1875                                 goto abort_transaction_no_dev_fatal;
1876                 }
1877         }
1878
1879         /* The remaining keys are not queue-specific */
1880         err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
1881                             1);
1882         if (err) {
1883                 message = "writing request-rx-copy";
1884                 goto abort_transaction;
1885         }
1886
1887         err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
1888         if (err) {
1889                 message = "writing feature-rx-notify";
1890                 goto abort_transaction;
1891         }
1892
1893         err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
1894         if (err) {
1895                 message = "writing feature-sg";
1896                 goto abort_transaction;
1897         }
1898
1899         err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
1900         if (err) {
1901                 message = "writing feature-gso-tcpv4";
1902                 goto abort_transaction;
1903         }
1904
1905         err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1");
1906         if (err) {
1907                 message = "writing feature-gso-tcpv6";
1908                 goto abort_transaction;
1909         }
1910
1911         err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload",
1912                            "1");
1913         if (err) {
1914                 message = "writing feature-ipv6-csum-offload";
1915                 goto abort_transaction;
1916         }
1917
1918         err = xenbus_transaction_end(xbt, 0);
1919         if (err) {
1920                 if (err == -EAGAIN)
1921                         goto again;
1922                 xenbus_dev_fatal(dev, err, "completing transaction");
1923                 goto destroy_ring;
1924         }
1925
1926         return 0;
1927
1928  abort_transaction:
1929         xenbus_dev_fatal(dev, err, "%s", message);
1930 abort_transaction_no_dev_fatal:
1931         xenbus_transaction_end(xbt, 1);
1932  destroy_ring:
1933         xennet_disconnect_backend(info);
1934         xennet_destroy_queues(info);
1935  out:
1936         unregister_netdev(info->netdev);
1937         xennet_free_netdev(info->netdev);
1938         return err;
1939 }
1940
1941 static int xennet_connect(struct net_device *dev)
1942 {
1943         struct netfront_info *np = netdev_priv(dev);
1944         unsigned int num_queues = 0;
1945         int err;
1946         unsigned int j = 0;
1947         struct netfront_queue *queue = NULL;
1948
1949         if (!xenbus_read_unsigned(np->xbdev->otherend, "feature-rx-copy", 0)) {
1950                 dev_info(&dev->dev,
1951                          "backend does not support copying receive path\n");
1952                 return -ENODEV;
1953         }
1954
1955         err = talk_to_netback(np->xbdev, np);
1956         if (err)
1957                 return err;
1958
1959         /* talk_to_netback() sets the correct number of queues */
1960         num_queues = dev->real_num_tx_queues;
1961
1962         rtnl_lock();
1963         netdev_update_features(dev);
1964         rtnl_unlock();
1965
1966         /*
1967          * All public and private state should now be sane.  Get
1968          * ready to start sending and receiving packets and give the driver
1969          * domain a kick because we've probably just requeued some
1970          * packets.
1971          */
1972         netif_carrier_on(np->netdev);
1973         for (j = 0; j < num_queues; ++j) {
1974                 queue = &np->queues[j];
1975
1976                 notify_remote_via_irq(queue->tx_irq);
1977                 if (queue->tx_irq != queue->rx_irq)
1978                         notify_remote_via_irq(queue->rx_irq);
1979
1980                 spin_lock_irq(&queue->tx_lock);
1981                 xennet_tx_buf_gc(queue);
1982                 spin_unlock_irq(&queue->tx_lock);
1983
1984                 spin_lock_bh(&queue->rx_lock);
1985                 xennet_alloc_rx_buffers(queue);
1986                 spin_unlock_bh(&queue->rx_lock);
1987         }
1988
1989         return 0;
1990 }
1991
1992 /**
1993  * Callback received when the backend's state changes.
1994  */
1995 static void netback_changed(struct xenbus_device *dev,
1996                             enum xenbus_state backend_state)
1997 {
1998         struct netfront_info *np = dev_get_drvdata(&dev->dev);
1999         struct net_device *netdev = np->netdev;
2000
2001         dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
2002
2003         switch (backend_state) {
2004         case XenbusStateInitialising:
2005         case XenbusStateInitialised:
2006         case XenbusStateReconfiguring:
2007         case XenbusStateReconfigured:
2008         case XenbusStateUnknown:
2009                 break;
2010
2011         case XenbusStateInitWait:
2012                 if (dev->state != XenbusStateInitialising)
2013                         break;
2014                 if (xennet_connect(netdev) != 0)
2015                         break;
2016                 xenbus_switch_state(dev, XenbusStateConnected);
2017                 break;
2018
2019         case XenbusStateConnected:
2020                 netdev_notify_peers(netdev);
2021                 break;
2022
2023         case XenbusStateClosed:
2024                 if (dev->state == XenbusStateClosed)
2025                         break;
2026                 /* Missed the backend's CLOSING state -- fallthrough */
2027         case XenbusStateClosing:
2028                 xenbus_frontend_closed(dev);
2029                 break;
2030         }
2031 }
2032
2033 static const struct xennet_stat {
2034         char name[ETH_GSTRING_LEN];
2035         u16 offset;
2036 } xennet_stats[] = {
2037         {
2038                 "rx_gso_checksum_fixup",
2039                 offsetof(struct netfront_info, rx_gso_checksum_fixup)
2040         },
2041 };
2042
2043 static int xennet_get_sset_count(struct net_device *dev, int string_set)
2044 {
2045         switch (string_set) {
2046         case ETH_SS_STATS:
2047                 return ARRAY_SIZE(xennet_stats);
2048         default:
2049                 return -EINVAL;
2050         }
2051 }
2052
2053 static void xennet_get_ethtool_stats(struct net_device *dev,
2054                                      struct ethtool_stats *stats, u64 * data)
2055 {
2056         void *np = netdev_priv(dev);
2057         int i;
2058
2059         for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2060                 data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset));
2061 }
2062
2063 static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
2064 {
2065         int i;
2066
2067         switch (stringset) {
2068         case ETH_SS_STATS:
2069                 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2070                         memcpy(data + i * ETH_GSTRING_LEN,
2071                                xennet_stats[i].name, ETH_GSTRING_LEN);
2072                 break;
2073         }
2074 }
2075
2076 static const struct ethtool_ops xennet_ethtool_ops =
2077 {
2078         .get_link = ethtool_op_get_link,
2079
2080         .get_sset_count = xennet_get_sset_count,
2081         .get_ethtool_stats = xennet_get_ethtool_stats,
2082         .get_strings = xennet_get_strings,
2083 };
2084
2085 #ifdef CONFIG_SYSFS
2086 static ssize_t show_rxbuf(struct device *dev,
2087                           struct device_attribute *attr, char *buf)
2088 {
2089         return sprintf(buf, "%lu\n", NET_RX_RING_SIZE);
2090 }
2091
2092 static ssize_t store_rxbuf(struct device *dev,
2093                            struct device_attribute *attr,
2094                            const char *buf, size_t len)
2095 {
2096         char *endp;
2097         unsigned long target;
2098
2099         if (!capable(CAP_NET_ADMIN))
2100                 return -EPERM;
2101
2102         target = simple_strtoul(buf, &endp, 0);
2103         if (endp == buf)
2104                 return -EBADMSG;
2105
2106         /* rxbuf_min and rxbuf_max are no longer configurable. */
2107
2108         return len;
2109 }
2110
2111 static DEVICE_ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf, store_rxbuf);
2112 static DEVICE_ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf, store_rxbuf);
2113 static DEVICE_ATTR(rxbuf_cur, S_IRUGO, show_rxbuf, NULL);
2114
2115 static struct attribute *xennet_dev_attrs[] = {
2116         &dev_attr_rxbuf_min.attr,
2117         &dev_attr_rxbuf_max.attr,
2118         &dev_attr_rxbuf_cur.attr,
2119         NULL
2120 };
2121
2122 static const struct attribute_group xennet_dev_group = {
2123         .attrs = xennet_dev_attrs
2124 };
2125 #endif /* CONFIG_SYSFS */
2126
2127 static int xennet_remove(struct xenbus_device *dev)
2128 {
2129         struct netfront_info *info = dev_get_drvdata(&dev->dev);
2130
2131         dev_dbg(&dev->dev, "%s\n", dev->nodename);
2132
2133         xennet_disconnect_backend(info);
2134
2135         unregister_netdev(info->netdev);
2136
2137         if (info->queues)
2138                 xennet_destroy_queues(info);
2139         xennet_free_netdev(info->netdev);
2140
2141         return 0;
2142 }
2143
2144 static const struct xenbus_device_id netfront_ids[] = {
2145         { "vif" },
2146         { "" }
2147 };
2148
2149 static struct xenbus_driver netfront_driver = {
2150         .ids = netfront_ids,
2151         .probe = netfront_probe,
2152         .remove = xennet_remove,
2153         .resume = netfront_resume,
2154         .otherend_changed = netback_changed,
2155 };
2156
2157 static int __init netif_init(void)
2158 {
2159         if (!xen_domain())
2160                 return -ENODEV;
2161
2162         if (!xen_has_pv_nic_devices())
2163                 return -ENODEV;
2164
2165         pr_info("Initialising Xen virtual ethernet driver\n");
2166
2167         /* Allow as many queues as there are CPUs if user has not
2168          * specified a value.
2169          */
2170         if (xennet_max_queues == 0)
2171                 xennet_max_queues = num_online_cpus();
2172
2173         return xenbus_register_frontend(&netfront_driver);
2174 }
2175 module_init(netif_init);
2176
2177
2178 static void __exit netif_exit(void)
2179 {
2180         xenbus_unregister_driver(&netfront_driver);
2181 }
2182 module_exit(netif_exit);
2183
2184 MODULE_DESCRIPTION("Xen virtual network device frontend");
2185 MODULE_LICENSE("GPL");
2186 MODULE_ALIAS("xen:vif");
2187 MODULE_ALIAS("xennet");
This page took 0.161756 seconds and 4 git commands to generate.