?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
mfd: wm8350-i2c: Make sure the i2c regmap functions are compiled
[linux/fpc-iii.git] / drivers / net / xen-netfront.c
blobfd3c1da144952932f2e843a48391cfdee668b703
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 <asm/xen/page.h>
49 #include <xen/xen.h>
50 #include <xen/xenbus.h>
51 #include <xen/events.h>
52 #include <xen/page.h>
53 #include <xen/platform_pci.h>
54 #include <xen/grant_table.h>
55
56 #include <xen/interface/io/netif.h>
57 #include <xen/interface/memory.h>
58 #include <xen/interface/grant_table.h>
59
60 static const struct ethtool_ops xennet_ethtool_ops;
61
62 struct netfront_cb {
63 int pull_to;
64 };
65
66 #define NETFRONT_SKB_CB(skb) ((struct netfront_cb *)((skb)->cb))
67
68 #define RX_COPY_THRESHOLD 256
69
70 #define GRANT_INVALID_REF 0
71
72 #define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, PAGE_SIZE)
73 #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, PAGE_SIZE)
74 #define TX_MAX_TARGET min_t(int, NET_TX_RING_SIZE, 256)
75
76 struct netfront_stats {
77 u64 rx_packets;
78 u64 tx_packets;
79 u64 rx_bytes;
80 u64 tx_bytes;
81 struct u64_stats_sync syncp;
82 };
83
84 struct netfront_info {
85 struct list_head list;
86 struct net_device *netdev;
87
88 struct napi_struct napi;
89
90 /* Split event channels support, tx_* == rx_* when using
91 * single event channel.
92 */
93 unsigned int tx_evtchn, rx_evtchn;
94 unsigned int tx_irq, rx_irq;
95 /* Only used when split event channels support is enabled */
96 char tx_irq_name[IFNAMSIZ+4]; /* DEVNAME-tx */
97 char rx_irq_name[IFNAMSIZ+4]; /* DEVNAME-rx */
98
99 struct xenbus_device *xbdev;
100
101 spinlock_t tx_lock;
102 struct xen_netif_tx_front_ring tx;
103 int tx_ring_ref;
104
105 /*
106 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
107 * are linked from tx_skb_freelist through skb_entry.link.
108 *
109 * NB. Freelist index entries are always going to be less than
110 * PAGE_OFFSET, whereas pointers to skbs will always be equal or
111 * greater than PAGE_OFFSET: we use this property to distinguish
112 * them.
113 */
114 union skb_entry {
115 struct sk_buff *skb;
116 unsigned long link;
117 } tx_skbs[NET_TX_RING_SIZE];
118 grant_ref_t gref_tx_head;
119 grant_ref_t grant_tx_ref[NET_TX_RING_SIZE];
120 struct page *grant_tx_page[NET_TX_RING_SIZE];
121 unsigned tx_skb_freelist;
122
123 spinlock_t rx_lock ____cacheline_aligned_in_smp;
124 struct xen_netif_rx_front_ring rx;
125 int rx_ring_ref;
126
127 /* Receive-ring batched refills. */
128 #define RX_MIN_TARGET 8
129 #define RX_DFL_MIN_TARGET 64
130 #define RX_MAX_TARGET min_t(int, NET_RX_RING_SIZE, 256)
131 unsigned rx_min_target, rx_max_target, rx_target;
132 struct sk_buff_head rx_batch;
133
134 struct timer_list rx_refill_timer;
135
136 struct sk_buff *rx_skbs[NET_RX_RING_SIZE];
137 grant_ref_t gref_rx_head;
138 grant_ref_t grant_rx_ref[NET_RX_RING_SIZE];
139
140 unsigned long rx_pfn_array[NET_RX_RING_SIZE];
141 struct multicall_entry rx_mcl[NET_RX_RING_SIZE+1];
142 struct mmu_update rx_mmu[NET_RX_RING_SIZE];
143
144 /* Statistics */
145 struct netfront_stats __percpu *stats;
146
147 unsigned long rx_gso_checksum_fixup;
148 };
149
150 struct netfront_rx_info {
151 struct xen_netif_rx_response rx;
152 struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
153 };
154
155 static void skb_entry_set_link(union skb_entry *list, unsigned short id)
156 {
157 list->link = id;
158 }
159
160 static int skb_entry_is_link(const union skb_entry *list)
161 {
162 BUILD_BUG_ON(sizeof(list->skb) != sizeof(list->link));
163 return (unsigned long)list->skb < PAGE_OFFSET;
164 }
165
166 /*
167 * Access macros for acquiring freeing slots in tx_skbs[].
168 */
169
170 static void add_id_to_freelist(unsigned *head, union skb_entry *list,
171 unsigned short id)
172 {
173 skb_entry_set_link(&list[id], *head);
174 *head = id;
175 }
176
177 static unsigned short get_id_from_freelist(unsigned *head,
178 union skb_entry *list)
179 {
180 unsigned int id = *head;
181 *head = list[id].link;
182 return id;
183 }
184
185 static int xennet_rxidx(RING_IDX idx)
186 {
187 return idx & (NET_RX_RING_SIZE - 1);
188 }
189
190 static struct sk_buff *xennet_get_rx_skb(struct netfront_info *np,
191 RING_IDX ri)
192 {
193 int i = xennet_rxidx(ri);
194 struct sk_buff *skb = np->rx_skbs[i];
195 np->rx_skbs[i] = NULL;
196 return skb;
197 }
198
199 static grant_ref_t xennet_get_rx_ref(struct netfront_info *np,
200 RING_IDX ri)
201 {
202 int i = xennet_rxidx(ri);
203 grant_ref_t ref = np->grant_rx_ref[i];
204 np->grant_rx_ref[i] = GRANT_INVALID_REF;
205 return ref;
206 }
207
208 #ifdef CONFIG_SYSFS
209 static int xennet_sysfs_addif(struct net_device *netdev);
210 static void xennet_sysfs_delif(struct net_device *netdev);
211 #else /* !CONFIG_SYSFS */
212 #define xennet_sysfs_addif(dev) (0)
213 #define xennet_sysfs_delif(dev) do { } while (0)
214 #endif
215
216 static bool xennet_can_sg(struct net_device *dev)
217 {
218 return dev->features & NETIF_F_SG;
219 }
220
221
222 static void rx_refill_timeout(unsigned long data)
223 {
224 struct net_device *dev = (struct net_device *)data;
225 struct netfront_info *np = netdev_priv(dev);
226 napi_schedule(&np->napi);
227 }
228
229 static int netfront_tx_slot_available(struct netfront_info *np)
230 {
231 return (np->tx.req_prod_pvt - np->tx.rsp_cons) <
232 (TX_MAX_TARGET - MAX_SKB_FRAGS - 2);
233 }
234
235 static void xennet_maybe_wake_tx(struct net_device *dev)
236 {
237 struct netfront_info *np = netdev_priv(dev);
238
239 if (unlikely(netif_queue_stopped(dev)) &&
240 netfront_tx_slot_available(np) &&
241 likely(netif_running(dev)))
242 netif_wake_queue(dev);
243 }
244
245 static void xennet_alloc_rx_buffers(struct net_device *dev)
246 {
247 unsigned short id;
248 struct netfront_info *np = netdev_priv(dev);
249 struct sk_buff *skb;
250 struct page *page;
251 int i, batch_target, notify;
252 RING_IDX req_prod = np->rx.req_prod_pvt;
253 grant_ref_t ref;
254 unsigned long pfn;
255 void *vaddr;
256 struct xen_netif_rx_request *req;
257
258 if (unlikely(!netif_carrier_ok(dev)))
259 return;
260
261 /*
262 * Allocate skbuffs greedily, even though we batch updates to the
263 * receive ring. This creates a less bursty demand on the memory
264 * allocator, so should reduce the chance of failed allocation requests
265 * both for ourself and for other kernel subsystems.
266 */
267 batch_target = np->rx_target - (req_prod - np->rx.rsp_cons);
268 for (i = skb_queue_len(&np->rx_batch); i < batch_target; i++) {
269 skb = __netdev_alloc_skb(dev, RX_COPY_THRESHOLD + NET_IP_ALIGN,
270 GFP_ATOMIC | __GFP_NOWARN);
271 if (unlikely(!skb))
272 goto no_skb;
273
274 /* Align ip header to a 16 bytes boundary */
275 skb_reserve(skb, NET_IP_ALIGN);
276
277 page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
278 if (!page) {
279 kfree_skb(skb);
280 no_skb:
281 /* Any skbuffs queued for refill? Force them out. */
282 if (i != 0)
283 goto refill;
284 /* Could not allocate any skbuffs. Try again later. */
285 mod_timer(&np->rx_refill_timer,
286 jiffies + (HZ/10));
287 break;
288 }
289
290 skb_add_rx_frag(skb, 0, page, 0, 0, PAGE_SIZE);
291 __skb_queue_tail(&np->rx_batch, skb);
292 }
293
294 /* Is the batch large enough to be worthwhile? */
295 if (i < (np->rx_target/2)) {
296 if (req_prod > np->rx.sring->req_prod)
297 goto push;
298 return;
299 }
300
301 /* Adjust our fill target if we risked running out of buffers. */
302 if (((req_prod - np->rx.sring->rsp_prod) < (np->rx_target / 4)) &&
303 ((np->rx_target *= 2) > np->rx_max_target))
304 np->rx_target = np->rx_max_target;
305
306 refill:
307 for (i = 0; ; i++) {
308 skb = __skb_dequeue(&np->rx_batch);
309 if (skb == NULL)
310 break;
311
312 skb->dev = dev;
313
314 id = xennet_rxidx(req_prod + i);
315
316 BUG_ON(np->rx_skbs[id]);
317 np->rx_skbs[id] = skb;
318
319 ref = gnttab_claim_grant_reference(&np->gref_rx_head);
320 BUG_ON((signed short)ref < 0);
321 np->grant_rx_ref[id] = ref;
322
323 pfn = page_to_pfn(skb_frag_page(&skb_shinfo(skb)->frags[0]));
324 vaddr = page_address(skb_frag_page(&skb_shinfo(skb)->frags[0]));
325
326 req = RING_GET_REQUEST(&np->rx, req_prod + i);
327 gnttab_grant_foreign_access_ref(ref,
328 np->xbdev->otherend_id,
329 pfn_to_mfn(pfn),
330 0);
331
332 req->id = id;
333 req->gref = ref;
334 }
335
336 wmb(); /* barrier so backend seens requests */
337
338 /* Above is a suitable barrier to ensure backend will see requests. */
339 np->rx.req_prod_pvt = req_prod + i;
340 push:
341 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->rx, notify);
342 if (notify)
343 notify_remote_via_irq(np->rx_irq);
344 }
345
346 static int xennet_open(struct net_device *dev)
347 {
348 struct netfront_info *np = netdev_priv(dev);
349
350 napi_enable(&np->napi);
351
352 spin_lock_bh(&np->rx_lock);
353 if (netif_carrier_ok(dev)) {
354 xennet_alloc_rx_buffers(dev);
355 np->rx.sring->rsp_event = np->rx.rsp_cons + 1;
356 if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx))
357 napi_schedule(&np->napi);
358 }
359 spin_unlock_bh(&np->rx_lock);
360
361 netif_start_queue(dev);
362
363 return 0;
364 }
365
366 static void xennet_tx_buf_gc(struct net_device *dev)
367 {
368 RING_IDX cons, prod;
369 unsigned short id;
370 struct netfront_info *np = netdev_priv(dev);
371 struct sk_buff *skb;
372
373 BUG_ON(!netif_carrier_ok(dev));
374
375 do {
376 prod = np->tx.sring->rsp_prod;
377 rmb(); /* Ensure we see responses up to 'rp'. */
378
379 for (cons = np->tx.rsp_cons; cons != prod; cons++) {
380 struct xen_netif_tx_response *txrsp;
381
382 txrsp = RING_GET_RESPONSE(&np->tx, cons);
383 if (txrsp->status == XEN_NETIF_RSP_NULL)
384 continue;
385
386 id = txrsp->id;
387 skb = np->tx_skbs[id].skb;
388 if (unlikely(gnttab_query_foreign_access(
389 np->grant_tx_ref[id]) != 0)) {
390 pr_alert("%s: warning -- grant still in use by backend domain\n",
391 __func__);
392 BUG();
393 }
394 gnttab_end_foreign_access_ref(
395 np->grant_tx_ref[id], GNTMAP_readonly);
396 gnttab_release_grant_reference(
397 &np->gref_tx_head, np->grant_tx_ref[id]);
398 np->grant_tx_ref[id] = GRANT_INVALID_REF;
399 np->grant_tx_page[id] = NULL;
400 add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, id);
401 dev_kfree_skb_irq(skb);
402 }
403
404 np->tx.rsp_cons = prod;
405
406 /*
407 * Set a new event, then check for race with update of tx_cons.
408 * Note that it is essential to schedule a callback, no matter
409 * how few buffers are pending. Even if there is space in the
410 * transmit ring, higher layers may be blocked because too much
411 * data is outstanding: in such cases notification from Xen is
412 * likely to be the only kick that we'll get.
413 */
414 np->tx.sring->rsp_event =
415 prod + ((np->tx.sring->req_prod - prod) >> 1) + 1;
416 mb(); /* update shared area */
417 } while ((cons == prod) && (prod != np->tx.sring->rsp_prod));
418
419 xennet_maybe_wake_tx(dev);
420 }
421
422 static void xennet_make_frags(struct sk_buff *skb, struct net_device *dev,
423 struct xen_netif_tx_request *tx)
424 {
425 struct netfront_info *np = netdev_priv(dev);
426 char *data = skb->data;
427 unsigned long mfn;
428 RING_IDX prod = np->tx.req_prod_pvt;
429 int frags = skb_shinfo(skb)->nr_frags;
430 unsigned int offset = offset_in_page(data);
431 unsigned int len = skb_headlen(skb);
432 unsigned int id;
433 grant_ref_t ref;
434 int i;
435
436 /* While the header overlaps a page boundary (including being
437 larger than a page), split it it into page-sized chunks. */
438 while (len > PAGE_SIZE - offset) {
439 tx->size = PAGE_SIZE - offset;
440 tx->flags |= XEN_NETTXF_more_data;
441 len -= tx->size;
442 data += tx->size;
443 offset = 0;
444
445 id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
446 np->tx_skbs[id].skb = skb_get(skb);
447 tx = RING_GET_REQUEST(&np->tx, prod++);
448 tx->id = id;
449 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
450 BUG_ON((signed short)ref < 0);
451
452 mfn = virt_to_mfn(data);
453 gnttab_grant_foreign_access_ref(ref, np->xbdev->otherend_id,
454 mfn, GNTMAP_readonly);
455
456 np->grant_tx_page[id] = virt_to_page(data);
457 tx->gref = np->grant_tx_ref[id] = ref;
458 tx->offset = offset;
459 tx->size = len;
460 tx->flags = 0;
461 }
462
463 /* Grant backend access to each skb fragment page. */
464 for (i = 0; i < frags; i++) {
465 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
466 struct page *page = skb_frag_page(frag);
467
468 len = skb_frag_size(frag);
469 offset = frag->page_offset;
470
471 /* Skip unused frames from start of page */
472 page += offset >> PAGE_SHIFT;
473 offset &= ~PAGE_MASK;
474
475 while (len > 0) {
476 unsigned long bytes;
477
478 bytes = PAGE_SIZE - offset;
479 if (bytes > len)
480 bytes = len;
481
482 tx->flags |= XEN_NETTXF_more_data;
483
484 id = get_id_from_freelist(&np->tx_skb_freelist,
485 np->tx_skbs);
486 np->tx_skbs[id].skb = skb_get(skb);
487 tx = RING_GET_REQUEST(&np->tx, prod++);
488 tx->id = id;
489 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
490 BUG_ON((signed short)ref < 0);
491
492 mfn = pfn_to_mfn(page_to_pfn(page));
493 gnttab_grant_foreign_access_ref(ref,
494 np->xbdev->otherend_id,
495 mfn, GNTMAP_readonly);
496
497 np->grant_tx_page[id] = page;
498 tx->gref = np->grant_tx_ref[id] = ref;
499 tx->offset = offset;
500 tx->size = bytes;
501 tx->flags = 0;
502
503 offset += bytes;
504 len -= bytes;
505
506 /* Next frame */
507 if (offset == PAGE_SIZE && len) {
508 BUG_ON(!PageCompound(page));
509 page++;
510 offset = 0;
511 }
512 }
513 }
514
515 np->tx.req_prod_pvt = prod;
516 }
517
518 /*
519 * Count how many ring slots are required to send the frags of this
520 * skb. Each frag might be a compound page.
521 */
522 static int xennet_count_skb_frag_slots(struct sk_buff *skb)
523 {
524 int i, frags = skb_shinfo(skb)->nr_frags;
525 int pages = 0;
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 pages += PFN_UP(offset + size);
536 }
537
538 return pages;
539 }
540
541 static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
542 {
543 unsigned short id;
544 struct netfront_info *np = netdev_priv(dev);
545 struct netfront_stats *stats = this_cpu_ptr(np->stats);
546 struct xen_netif_tx_request *tx;
547 char *data = skb->data;
548 RING_IDX i;
549 grant_ref_t ref;
550 unsigned long mfn;
551 int notify;
552 int slots;
553 unsigned int offset = offset_in_page(data);
554 unsigned int len = skb_headlen(skb);
555 unsigned long flags;
556
557 /* If skb->len is too big for wire format, drop skb and alert
558 * user about misconfiguration.
559 */
560 if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) {
561 net_alert_ratelimited(
562 "xennet: skb->len = %u, too big for wire format\n",
563 skb->len);
564 goto drop;
565 }
566
567 slots = DIV_ROUND_UP(offset + len, PAGE_SIZE) +
568 xennet_count_skb_frag_slots(skb);
569 if (unlikely(slots > MAX_SKB_FRAGS + 1)) {
570 net_alert_ratelimited(
571 "xennet: skb rides the rocket: %d slots\n", slots);
572 goto drop;
573 }
574
575 spin_lock_irqsave(&np->tx_lock, flags);
576
577 if (unlikely(!netif_carrier_ok(dev) ||
578 (slots > 1 && !xennet_can_sg(dev)) ||
579 netif_needs_gso(skb, netif_skb_features(skb)))) {
580 spin_unlock_irqrestore(&np->tx_lock, flags);
581 goto drop;
582 }
583
584 i = np->tx.req_prod_pvt;
585
586 id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
587 np->tx_skbs[id].skb = skb;
588
589 tx = RING_GET_REQUEST(&np->tx, i);
590
591 tx->id = id;
592 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
593 BUG_ON((signed short)ref < 0);
594 mfn = virt_to_mfn(data);
595 gnttab_grant_foreign_access_ref(
596 ref, np->xbdev->otherend_id, mfn, GNTMAP_readonly);
597 np->grant_tx_page[id] = virt_to_page(data);
598 tx->gref = np->grant_tx_ref[id] = ref;
599 tx->offset = offset;
600 tx->size = len;
601
602 tx->flags = 0;
603 if (skb->ip_summed == CHECKSUM_PARTIAL)
604 /* local packet? */
605 tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated;
606 else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
607 /* remote but checksummed. */
608 tx->flags |= XEN_NETTXF_data_validated;
609
610 if (skb_shinfo(skb)->gso_size) {
611 struct xen_netif_extra_info *gso;
612
613 gso = (struct xen_netif_extra_info *)
614 RING_GET_REQUEST(&np->tx, ++i);
615
616 tx->flags |= XEN_NETTXF_extra_info;
617
618 gso->u.gso.size = skb_shinfo(skb)->gso_size;
619 gso->u.gso.type = XEN_NETIF_GSO_TYPE_TCPV4;
620 gso->u.gso.pad = 0;
621 gso->u.gso.features = 0;
622
623 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
624 gso->flags = 0;
625 }
626
627 np->tx.req_prod_pvt = i + 1;
628
629 xennet_make_frags(skb, dev, tx);
630 tx->size = skb->len;
631
632 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->tx, notify);
633 if (notify)
634 notify_remote_via_irq(np->tx_irq);
635
636 u64_stats_update_begin(&stats->syncp);
637 stats->tx_bytes += skb->len;
638 stats->tx_packets++;
639 u64_stats_update_end(&stats->syncp);
640
641 /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
642 xennet_tx_buf_gc(dev);
643
644 if (!netfront_tx_slot_available(np))
645 netif_stop_queue(dev);
646
647 spin_unlock_irqrestore(&np->tx_lock, flags);
648
649 return NETDEV_TX_OK;
650
651 drop:
652 dev->stats.tx_dropped++;
653 dev_kfree_skb(skb);
654 return NETDEV_TX_OK;
655 }
656
657 static int xennet_close(struct net_device *dev)
658 {
659 struct netfront_info *np = netdev_priv(dev);
660 netif_stop_queue(np->netdev);
661 napi_disable(&np->napi);
662 return 0;
663 }
664
665 static void xennet_move_rx_slot(struct netfront_info *np, struct sk_buff *skb,
666 grant_ref_t ref)
667 {
668 int new = xennet_rxidx(np->rx.req_prod_pvt);
669
670 BUG_ON(np->rx_skbs[new]);
671 np->rx_skbs[new] = skb;
672 np->grant_rx_ref[new] = ref;
673 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->id = new;
674 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->gref = ref;
675 np->rx.req_prod_pvt++;
676 }
677
678 static int xennet_get_extras(struct netfront_info *np,
679 struct xen_netif_extra_info *extras,
680 RING_IDX rp)
681
682 {
683 struct xen_netif_extra_info *extra;
684 struct device *dev = &np->netdev->dev;
685 RING_IDX cons = np->rx.rsp_cons;
686 int err = 0;
687
688 do {
689 struct sk_buff *skb;
690 grant_ref_t ref;
691
692 if (unlikely(cons + 1 == rp)) {
693 if (net_ratelimit())
694 dev_warn(dev, "Missing extra info\n");
695 err = -EBADR;
696 break;
697 }
698
699 extra = (struct xen_netif_extra_info *)
700 RING_GET_RESPONSE(&np->rx, ++cons);
701
702 if (unlikely(!extra->type ||
703 extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
704 if (net_ratelimit())
705 dev_warn(dev, "Invalid extra type: %d\n",
706 extra->type);
707 err = -EINVAL;
708 } else {
709 memcpy(&extras[extra->type - 1], extra,
710 sizeof(*extra));
711 }
712
713 skb = xennet_get_rx_skb(np, cons);
714 ref = xennet_get_rx_ref(np, cons);
715 xennet_move_rx_slot(np, skb, ref);
716 } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
717
718 np->rx.rsp_cons = cons;
719 return err;
720 }
721
722 static int xennet_get_responses(struct netfront_info *np,
723 struct netfront_rx_info *rinfo, RING_IDX rp,
724 struct sk_buff_head *list)
725 {
726 struct xen_netif_rx_response *rx = &rinfo->rx;
727 struct xen_netif_extra_info *extras = rinfo->extras;
728 struct device *dev = &np->netdev->dev;
729 RING_IDX cons = np->rx.rsp_cons;
730 struct sk_buff *skb = xennet_get_rx_skb(np, cons);
731 grant_ref_t ref = xennet_get_rx_ref(np, cons);
732 int max = MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD);
733 int slots = 1;
734 int err = 0;
735 unsigned long ret;
736
737 if (rx->flags & XEN_NETRXF_extra_info) {
738 err = xennet_get_extras(np, extras, rp);
739 cons = np->rx.rsp_cons;
740 }
741
742 for (;;) {
743 if (unlikely(rx->status < 0 ||
744 rx->offset + rx->status > PAGE_SIZE)) {
745 if (net_ratelimit())
746 dev_warn(dev, "rx->offset: %x, size: %u\n",
747 rx->offset, rx->status);
748 xennet_move_rx_slot(np, skb, ref);
749 err = -EINVAL;
750 goto next;
751 }
752
753 /*
754 * This definitely indicates a bug, either in this driver or in
755 * the backend driver. In future this should flag the bad
756 * situation to the system controller to reboot the backend.
757 */
758 if (ref == GRANT_INVALID_REF) {
759 if (net_ratelimit())
760 dev_warn(dev, "Bad rx response id %d.\n",
761 rx->id);
762 err = -EINVAL;
763 goto next;
764 }
765
766 ret = gnttab_end_foreign_access_ref(ref, 0);
767 BUG_ON(!ret);
768
769 gnttab_release_grant_reference(&np->gref_rx_head, ref);
770
771 __skb_queue_tail(list, skb);
772
773 next:
774 if (!(rx->flags & XEN_NETRXF_more_data))
775 break;
776
777 if (cons + slots == rp) {
778 if (net_ratelimit())
779 dev_warn(dev, "Need more slots\n");
780 err = -ENOENT;
781 break;
782 }
783
784 rx = RING_GET_RESPONSE(&np->rx, cons + slots);
785 skb = xennet_get_rx_skb(np, cons + slots);
786 ref = xennet_get_rx_ref(np, cons + slots);
787 slots++;
788 }
789
790 if (unlikely(slots > max)) {
791 if (net_ratelimit())
792 dev_warn(dev, "Too many slots\n");
793 err = -E2BIG;
794 }
795
796 if (unlikely(err))
797 np->rx.rsp_cons = cons + slots;
798
799 return err;
800 }
801
802 static int xennet_set_skb_gso(struct sk_buff *skb,
803 struct xen_netif_extra_info *gso)
804 {
805 if (!gso->u.gso.size) {
806 if (net_ratelimit())
807 pr_warn("GSO size must not be zero\n");
808 return -EINVAL;
809 }
810
811 /* Currently only TCPv4 S.O. is supported. */
812 if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4) {
813 if (net_ratelimit())
814 pr_warn("Bad GSO type %d\n", gso->u.gso.type);
815 return -EINVAL;
816 }
817
818 skb_shinfo(skb)->gso_size = gso->u.gso.size;
819 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
820
821 /* Header must be checked, and gso_segs computed. */
822 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
823 skb_shinfo(skb)->gso_segs = 0;
824
825 return 0;
826 }
827
828 static RING_IDX xennet_fill_frags(struct netfront_info *np,
829 struct sk_buff *skb,
830 struct sk_buff_head *list)
831 {
832 struct skb_shared_info *shinfo = skb_shinfo(skb);
833 RING_IDX cons = np->rx.rsp_cons;
834 struct sk_buff *nskb;
835
836 while ((nskb = __skb_dequeue(list))) {
837 struct xen_netif_rx_response *rx =
838 RING_GET_RESPONSE(&np->rx, ++cons);
839 skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0];
840
841 if (shinfo->nr_frags == MAX_SKB_FRAGS) {
842 unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
843
844 BUG_ON(pull_to <= skb_headlen(skb));
845 __pskb_pull_tail(skb, pull_to - skb_headlen(skb));
846 }
847 BUG_ON(shinfo->nr_frags >= MAX_SKB_FRAGS);
848
849 skb_add_rx_frag(skb, shinfo->nr_frags, skb_frag_page(nfrag),
850 rx->offset, rx->status, PAGE_SIZE);
851
852 skb_shinfo(nskb)->nr_frags = 0;
853 kfree_skb(nskb);
854 }
855
856 return cons;
857 }
858
859 static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
860 {
861 struct iphdr *iph;
862 int err = -EPROTO;
863 int recalculate_partial_csum = 0;
864
865 /*
866 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
867 * peers can fail to set NETRXF_csum_blank when sending a GSO
868 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
869 * recalculate the partial checksum.
870 */
871 if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
872 struct netfront_info *np = netdev_priv(dev);
873 np->rx_gso_checksum_fixup++;
874 skb->ip_summed = CHECKSUM_PARTIAL;
875 recalculate_partial_csum = 1;
876 }
877
878 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
879 if (skb->ip_summed != CHECKSUM_PARTIAL)
880 return 0;
881
882 if (skb->protocol != htons(ETH_P_IP))
883 goto out;
884
885 iph = (void *)skb->data;
886
887 switch (iph->protocol) {
888 case IPPROTO_TCP:
889 if (!skb_partial_csum_set(skb, 4 * iph->ihl,
890 offsetof(struct tcphdr, check)))
891 goto out;
892
893 if (recalculate_partial_csum) {
894 struct tcphdr *tcph = tcp_hdr(skb);
895 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
896 skb->len - iph->ihl*4,
897 IPPROTO_TCP, 0);
898 }
899 break;
900 case IPPROTO_UDP:
901 if (!skb_partial_csum_set(skb, 4 * iph->ihl,
902 offsetof(struct udphdr, check)))
903 goto out;
904
905 if (recalculate_partial_csum) {
906 struct udphdr *udph = udp_hdr(skb);
907 udph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
908 skb->len - iph->ihl*4,
909 IPPROTO_UDP, 0);
910 }
911 break;
912 default:
913 if (net_ratelimit())
914 pr_err("Attempting to checksum a non-TCP/UDP packet, dropping a protocol %d packet\n",
915 iph->protocol);
916 goto out;
917 }
918
919 err = 0;
920
921 out:
922 return err;
923 }
924
925 static int handle_incoming_queue(struct net_device *dev,
926 struct sk_buff_head *rxq)
927 {
928 struct netfront_info *np = netdev_priv(dev);
929 struct netfront_stats *stats = this_cpu_ptr(np->stats);
930 int packets_dropped = 0;
931 struct sk_buff *skb;
932
933 while ((skb = __skb_dequeue(rxq)) != NULL) {
934 int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
935
936 if (pull_to > skb_headlen(skb))
937 __pskb_pull_tail(skb, pull_to - skb_headlen(skb));
938
939 /* Ethernet work: Delayed to here as it peeks the header. */
940 skb->protocol = eth_type_trans(skb, dev);
941
942 if (checksum_setup(dev, skb)) {
943 kfree_skb(skb);
944 packets_dropped++;
945 dev->stats.rx_errors++;
946 continue;
947 }
948
949 u64_stats_update_begin(&stats->syncp);
950 stats->rx_packets++;
951 stats->rx_bytes += skb->len;
952 u64_stats_update_end(&stats->syncp);
953
954 /* Pass it up. */
955 netif_receive_skb(skb);
956 }
957
958 return packets_dropped;
959 }
960
961 static int xennet_poll(struct napi_struct *napi, int budget)
962 {
963 struct netfront_info *np = container_of(napi, struct netfront_info, napi);
964 struct net_device *dev = np->netdev;
965 struct sk_buff *skb;
966 struct netfront_rx_info rinfo;
967 struct xen_netif_rx_response *rx = &rinfo.rx;
968 struct xen_netif_extra_info *extras = rinfo.extras;
969 RING_IDX i, rp;
970 int work_done;
971 struct sk_buff_head rxq;
972 struct sk_buff_head errq;
973 struct sk_buff_head tmpq;
974 unsigned long flags;
975 int err;
976
977 spin_lock(&np->rx_lock);
978
979 skb_queue_head_init(&rxq);
980 skb_queue_head_init(&errq);
981 skb_queue_head_init(&tmpq);
982
983 rp = np->rx.sring->rsp_prod;
984 rmb(); /* Ensure we see queued responses up to 'rp'. */
985
986 i = np->rx.rsp_cons;
987 work_done = 0;
988 while ((i != rp) && (work_done < budget)) {
989 memcpy(rx, RING_GET_RESPONSE(&np->rx, i), sizeof(*rx));
990 memset(extras, 0, sizeof(rinfo.extras));
991
992 err = xennet_get_responses(np, &rinfo, rp, &tmpq);
993
994 if (unlikely(err)) {
995 err:
996 while ((skb = __skb_dequeue(&tmpq)))
997 __skb_queue_tail(&errq, skb);
998 dev->stats.rx_errors++;
999 i = np->rx.rsp_cons;
1000 continue;
1001 }
1002
1003 skb = __skb_dequeue(&tmpq);
1004
1005 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1006 struct xen_netif_extra_info *gso;
1007 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1008
1009 if (unlikely(xennet_set_skb_gso(skb, gso))) {
1010 __skb_queue_head(&tmpq, skb);
1011 np->rx.rsp_cons += skb_queue_len(&tmpq);
1012 goto err;
1013 }
1014 }
1015
1016 NETFRONT_SKB_CB(skb)->pull_to = rx->status;
1017 if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD)
1018 NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD;
1019
1020 skb_shinfo(skb)->frags[0].page_offset = rx->offset;
1021 skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status);
1022 skb->data_len = rx->status;
1023 skb->len += rx->status;
1024
1025 i = xennet_fill_frags(np, skb, &tmpq);
1026
1027 if (rx->flags & XEN_NETRXF_csum_blank)
1028 skb->ip_summed = CHECKSUM_PARTIAL;
1029 else if (rx->flags & XEN_NETRXF_data_validated)
1030 skb->ip_summed = CHECKSUM_UNNECESSARY;
1031
1032 __skb_queue_tail(&rxq, skb);
1033
1034 np->rx.rsp_cons = ++i;
1035 work_done++;
1036 }
1037
1038 __skb_queue_purge(&errq);
1039
1040 work_done -= handle_incoming_queue(dev, &rxq);
1041
1042 /* If we get a callback with very few responses, reduce fill target. */
1043 /* NB. Note exponential increase, linear decrease. */
1044 if (((np->rx.req_prod_pvt - np->rx.sring->rsp_prod) >
1045 ((3*np->rx_target) / 4)) &&
1046 (--np->rx_target < np->rx_min_target))
1047 np->rx_target = np->rx_min_target;
1048
1049 xennet_alloc_rx_buffers(dev);
1050
1051 if (work_done < budget) {
1052 int more_to_do = 0;
1053
1054 local_irq_save(flags);
1055
1056 RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, more_to_do);
1057 if (!more_to_do)
1058 __napi_complete(napi);
1059
1060 local_irq_restore(flags);
1061 }
1062
1063 spin_unlock(&np->rx_lock);
1064
1065 return work_done;
1066 }
1067
1068 static int xennet_change_mtu(struct net_device *dev, int mtu)
1069 {
1070 int max = xennet_can_sg(dev) ?
1071 XEN_NETIF_MAX_TX_SIZE - MAX_TCP_HEADER : ETH_DATA_LEN;
1072
1073 if (mtu > max)
1074 return -EINVAL;
1075 dev->mtu = mtu;
1076 return 0;
1077 }
1078
1079 static struct rtnl_link_stats64 *xennet_get_stats64(struct net_device *dev,
1080 struct rtnl_link_stats64 *tot)
1081 {
1082 struct netfront_info *np = netdev_priv(dev);
1083 int cpu;
1084
1085 for_each_possible_cpu(cpu) {
1086 struct netfront_stats *stats = per_cpu_ptr(np->stats, cpu);
1087 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1088 unsigned int start;
1089
1090 do {
1091 start = u64_stats_fetch_begin_bh(&stats->syncp);
1092
1093 rx_packets = stats->rx_packets;
1094 tx_packets = stats->tx_packets;
1095 rx_bytes = stats->rx_bytes;
1096 tx_bytes = stats->tx_bytes;
1097 } while (u64_stats_fetch_retry_bh(&stats->syncp, start));
1098
1099 tot->rx_packets += rx_packets;
1100 tot->tx_packets += tx_packets;
1101 tot->rx_bytes += rx_bytes;
1102 tot->tx_bytes += tx_bytes;
1103 }
1104
1105 tot->rx_errors = dev->stats.rx_errors;
1106 tot->tx_dropped = dev->stats.tx_dropped;
1107
1108 return tot;
1109 }
1110
1111 static void xennet_release_tx_bufs(struct netfront_info *np)
1112 {
1113 struct sk_buff *skb;
1114 int i;
1115
1116 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1117 /* Skip over entries which are actually freelist references */
1118 if (skb_entry_is_link(&np->tx_skbs[i]))
1119 continue;
1120
1121 skb = np->tx_skbs[i].skb;
1122 get_page(np->grant_tx_page[i]);
1123 gnttab_end_foreign_access(np->grant_tx_ref[i],
1124 GNTMAP_readonly,
1125 (unsigned long)page_address(np->grant_tx_page[i]));
1126 np->grant_tx_page[i] = NULL;
1127 np->grant_tx_ref[i] = GRANT_INVALID_REF;
1128 add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, i);
1129 dev_kfree_skb_irq(skb);
1130 }
1131 }
1132
1133 static void xennet_release_rx_bufs(struct netfront_info *np)
1134 {
1135 int id, ref;
1136
1137 spin_lock_bh(&np->rx_lock);
1138
1139 for (id = 0; id < NET_RX_RING_SIZE; id++) {
1140 struct sk_buff *skb;
1141 struct page *page;
1142
1143 skb = np->rx_skbs[id];
1144 if (!skb)
1145 continue;
1146
1147 ref = np->grant_rx_ref[id];
1148 if (ref == GRANT_INVALID_REF)
1149 continue;
1150
1151 page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
1152
1153 /* gnttab_end_foreign_access() needs a page ref until
1154 * foreign access is ended (which may be deferred).
1155 */
1156 get_page(page);
1157 gnttab_end_foreign_access(ref, 0,
1158 (unsigned long)page_address(page));
1159 np->grant_rx_ref[id] = GRANT_INVALID_REF;
1160
1161 kfree_skb(skb);
1162 }
1163
1164 spin_unlock_bh(&np->rx_lock);
1165 }
1166
1167 static void xennet_uninit(struct net_device *dev)
1168 {
1169 struct netfront_info *np = netdev_priv(dev);
1170 xennet_release_tx_bufs(np);
1171 xennet_release_rx_bufs(np);
1172 gnttab_free_grant_references(np->gref_tx_head);
1173 gnttab_free_grant_references(np->gref_rx_head);
1174 }
1175
1176 static netdev_features_t xennet_fix_features(struct net_device *dev,
1177 netdev_features_t features)
1178 {
1179 struct netfront_info *np = netdev_priv(dev);
1180 int val;
1181
1182 if (features & NETIF_F_SG) {
1183 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, "feature-sg",
1184 "%d", &val) < 0)
1185 val = 0;
1186
1187 if (!val)
1188 features &= ~NETIF_F_SG;
1189 }
1190
1191 if (features & NETIF_F_TSO) {
1192 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1193 "feature-gso-tcpv4", "%d", &val) < 0)
1194 val = 0;
1195
1196 if (!val)
1197 features &= ~NETIF_F_TSO;
1198 }
1199
1200 return features;
1201 }
1202
1203 static int xennet_set_features(struct net_device *dev,
1204 netdev_features_t features)
1205 {
1206 if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1207 netdev_info(dev, "Reducing MTU because no SG offload");
1208 dev->mtu = ETH_DATA_LEN;
1209 }
1210
1211 return 0;
1212 }
1213
1214 static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id)
1215 {
1216 struct netfront_info *np = dev_id;
1217 struct net_device *dev = np->netdev;
1218 unsigned long flags;
1219
1220 spin_lock_irqsave(&np->tx_lock, flags);
1221 xennet_tx_buf_gc(dev);
1222 spin_unlock_irqrestore(&np->tx_lock, flags);
1223
1224 return IRQ_HANDLED;
1225 }
1226
1227 static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id)
1228 {
1229 struct netfront_info *np = dev_id;
1230 struct net_device *dev = np->netdev;
1231
1232 if (likely(netif_carrier_ok(dev) &&
1233 RING_HAS_UNCONSUMED_RESPONSES(&np->rx)))
1234 napi_schedule(&np->napi);
1235
1236 return IRQ_HANDLED;
1237 }
1238
1239 static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1240 {
1241 xennet_tx_interrupt(irq, dev_id);
1242 xennet_rx_interrupt(irq, dev_id);
1243 return IRQ_HANDLED;
1244 }
1245
1246 #ifdef CONFIG_NET_POLL_CONTROLLER
1247 static void xennet_poll_controller(struct net_device *dev)
1248 {
1249 xennet_interrupt(0, dev);
1250 }
1251 #endif
1252
1253 static const struct net_device_ops xennet_netdev_ops = {
1254 .ndo_open = xennet_open,
1255 .ndo_uninit = xennet_uninit,
1256 .ndo_stop = xennet_close,
1257 .ndo_start_xmit = xennet_start_xmit,
1258 .ndo_change_mtu = xennet_change_mtu,
1259 .ndo_get_stats64 = xennet_get_stats64,
1260 .ndo_set_mac_address = eth_mac_addr,
1261 .ndo_validate_addr = eth_validate_addr,
1262 .ndo_fix_features = xennet_fix_features,
1263 .ndo_set_features = xennet_set_features,
1264 #ifdef CONFIG_NET_POLL_CONTROLLER
1265 .ndo_poll_controller = xennet_poll_controller,
1266 #endif
1267 };
1268
1269 static struct net_device *xennet_create_dev(struct xenbus_device *dev)
1270 {
1271 int i, err;
1272 struct net_device *netdev;
1273 struct netfront_info *np;
1274
1275 netdev = alloc_etherdev(sizeof(struct netfront_info));
1276 if (!netdev)
1277 return ERR_PTR(-ENOMEM);
1278
1279 np = netdev_priv(netdev);
1280 np->xbdev = dev;
1281
1282 spin_lock_init(&np->tx_lock);
1283 spin_lock_init(&np->rx_lock);
1284
1285 skb_queue_head_init(&np->rx_batch);
1286 np->rx_target = RX_DFL_MIN_TARGET;
1287 np->rx_min_target = RX_DFL_MIN_TARGET;
1288 np->rx_max_target = RX_MAX_TARGET;
1289
1290 init_timer(&np->rx_refill_timer);
1291 np->rx_refill_timer.data = (unsigned long)netdev;
1292 np->rx_refill_timer.function = rx_refill_timeout;
1293
1294 err = -ENOMEM;
1295 np->stats = alloc_percpu(struct netfront_stats);
1296 if (np->stats == NULL)
1297 goto exit;
1298
1299 /* Initialise tx_skbs as a free chain containing every entry. */
1300 np->tx_skb_freelist = 0;
1301 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1302 skb_entry_set_link(&np->tx_skbs[i], i+1);
1303 np->grant_tx_ref[i] = GRANT_INVALID_REF;
1304 }
1305
1306 /* Clear out rx_skbs */
1307 for (i = 0; i < NET_RX_RING_SIZE; i++) {
1308 np->rx_skbs[i] = NULL;
1309 np->grant_rx_ref[i] = GRANT_INVALID_REF;
1310 np->grant_tx_page[i] = NULL;
1311 }
1312
1313 /* A grant for every tx ring slot */
1314 if (gnttab_alloc_grant_references(TX_MAX_TARGET,
1315 &np->gref_tx_head) < 0) {
1316 pr_alert("can't alloc tx grant refs\n");
1317 err = -ENOMEM;
1318 goto exit_free_stats;
1319 }
1320 /* A grant for every rx ring slot */
1321 if (gnttab_alloc_grant_references(RX_MAX_TARGET,
1322 &np->gref_rx_head) < 0) {
1323 pr_alert("can't alloc rx grant refs\n");
1324 err = -ENOMEM;
1325 goto exit_free_tx;
1326 }
1327
1328 netdev->netdev_ops = &xennet_netdev_ops;
1329
1330 netif_napi_add(netdev, &np->napi, xennet_poll, 64);
1331 netdev->features = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1332 NETIF_F_GSO_ROBUST;
1333 netdev->hw_features = NETIF_F_IP_CSUM | NETIF_F_SG | NETIF_F_TSO;
1334
1335 /*
1336 * Assume that all hw features are available for now. This set
1337 * will be adjusted by the call to netdev_update_features() in
1338 * xennet_connect() which is the earliest point where we can
1339 * negotiate with the backend regarding supported features.
1340 */
1341 netdev->features |= netdev->hw_features;
1342
1343 SET_ETHTOOL_OPS(netdev, &xennet_ethtool_ops);
1344 SET_NETDEV_DEV(netdev, &dev->dev);
1345
1346 netif_set_gso_max_size(netdev, XEN_NETIF_MAX_TX_SIZE - MAX_TCP_HEADER);
1347
1348 np->netdev = netdev;
1349
1350 netif_carrier_off(netdev);
1351
1352 return netdev;
1353
1354 exit_free_tx:
1355 gnttab_free_grant_references(np->gref_tx_head);
1356 exit_free_stats:
1357 free_percpu(np->stats);
1358 exit:
1359 free_netdev(netdev);
1360 return ERR_PTR(err);
1361 }
1362
1363 /**
1364 * Entry point to this code when a new device is created. Allocate the basic
1365 * structures and the ring buffers for communication with the backend, and
1366 * inform the backend of the appropriate details for those.
1367 */
1368 static int netfront_probe(struct xenbus_device *dev,
1369 const struct xenbus_device_id *id)
1370 {
1371 int err;
1372 struct net_device *netdev;
1373 struct netfront_info *info;
1374
1375 netdev = xennet_create_dev(dev);
1376 if (IS_ERR(netdev)) {
1377 err = PTR_ERR(netdev);
1378 xenbus_dev_fatal(dev, err, "creating netdev");
1379 return err;
1380 }
1381
1382 info = netdev_priv(netdev);
1383 dev_set_drvdata(&dev->dev, info);
1384
1385 err = register_netdev(info->netdev);
1386 if (err) {
1387 pr_warn("%s: register_netdev err=%d\n", __func__, err);
1388 goto fail;
1389 }
1390
1391 err = xennet_sysfs_addif(info->netdev);
1392 if (err) {
1393 unregister_netdev(info->netdev);
1394 pr_warn("%s: add sysfs failed err=%d\n", __func__, err);
1395 goto fail;
1396 }
1397
1398 return 0;
1399
1400 fail:
1401 free_netdev(netdev);
1402 dev_set_drvdata(&dev->dev, NULL);
1403 return err;
1404 }
1405
1406 static void xennet_end_access(int ref, void *page)
1407 {
1408 /* This frees the page as a side-effect */
1409 if (ref != GRANT_INVALID_REF)
1410 gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1411 }
1412
1413 static void xennet_disconnect_backend(struct netfront_info *info)
1414 {
1415 /* Stop old i/f to prevent errors whilst we rebuild the state. */
1416 spin_lock_bh(&info->rx_lock);
1417 spin_lock_irq(&info->tx_lock);
1418 netif_carrier_off(info->netdev);
1419 spin_unlock_irq(&info->tx_lock);
1420 spin_unlock_bh(&info->rx_lock);
1421
1422 if (info->tx_irq && (info->tx_irq == info->rx_irq))
1423 unbind_from_irqhandler(info->tx_irq, info);
1424 if (info->tx_irq && (info->tx_irq != info->rx_irq)) {
1425 unbind_from_irqhandler(info->tx_irq, info);
1426 unbind_from_irqhandler(info->rx_irq, info);
1427 }
1428 info->tx_evtchn = info->rx_evtchn = 0;
1429 info->tx_irq = info->rx_irq = 0;
1430
1431 /* End access and free the pages */
1432 xennet_end_access(info->tx_ring_ref, info->tx.sring);
1433 xennet_end_access(info->rx_ring_ref, info->rx.sring);
1434
1435 info->tx_ring_ref = GRANT_INVALID_REF;
1436 info->rx_ring_ref = GRANT_INVALID_REF;
1437 info->tx.sring = NULL;
1438 info->rx.sring = NULL;
1439 }
1440
1441 /**
1442 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1443 * driver restart. We tear down our netif structure and recreate it, but
1444 * leave the device-layer structures intact so that this is transparent to the
1445 * rest of the kernel.
1446 */
1447 static int netfront_resume(struct xenbus_device *dev)
1448 {
1449 struct netfront_info *info = dev_get_drvdata(&dev->dev);
1450
1451 dev_dbg(&dev->dev, "%s\n", dev->nodename);
1452
1453 xennet_disconnect_backend(info);
1454 return 0;
1455 }
1456
1457 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1458 {
1459 char *s, *e, *macstr;
1460 int i;
1461
1462 macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1463 if (IS_ERR(macstr))
1464 return PTR_ERR(macstr);
1465
1466 for (i = 0; i < ETH_ALEN; i++) {
1467 mac[i] = simple_strtoul(s, &e, 16);
1468 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1469 kfree(macstr);
1470 return -ENOENT;
1471 }
1472 s = e+1;
1473 }
1474
1475 kfree(macstr);
1476 return 0;
1477 }
1478
1479 static int setup_netfront_single(struct netfront_info *info)
1480 {
1481 int err;
1482
1483 err = xenbus_alloc_evtchn(info->xbdev, &info->tx_evtchn);
1484 if (err < 0)
1485 goto fail;
1486
1487 err = bind_evtchn_to_irqhandler(info->tx_evtchn,
1488 xennet_interrupt,
1489 0, info->netdev->name, info);
1490 if (err < 0)
1491 goto bind_fail;
1492 info->rx_evtchn = info->tx_evtchn;
1493 info->rx_irq = info->tx_irq = err;
1494
1495 return 0;
1496
1497 bind_fail:
1498 xenbus_free_evtchn(info->xbdev, info->tx_evtchn);
1499 info->tx_evtchn = 0;
1500 fail:
1501 return err;
1502 }
1503
1504 static int setup_netfront_split(struct netfront_info *info)
1505 {
1506 int err;
1507
1508 err = xenbus_alloc_evtchn(info->xbdev, &info->tx_evtchn);
1509 if (err < 0)
1510 goto fail;
1511 err = xenbus_alloc_evtchn(info->xbdev, &info->rx_evtchn);
1512 if (err < 0)
1513 goto alloc_rx_evtchn_fail;
1514
1515 snprintf(info->tx_irq_name, sizeof(info->tx_irq_name),
1516 "%s-tx", info->netdev->name);
1517 err = bind_evtchn_to_irqhandler(info->tx_evtchn,
1518 xennet_tx_interrupt,
1519 0, info->tx_irq_name, info);
1520 if (err < 0)
1521 goto bind_tx_fail;
1522 info->tx_irq = err;
1523
1524 snprintf(info->rx_irq_name, sizeof(info->rx_irq_name),
1525 "%s-rx", info->netdev->name);
1526 err = bind_evtchn_to_irqhandler(info->rx_evtchn,
1527 xennet_rx_interrupt,
1528 0, info->rx_irq_name, info);
1529 if (err < 0)
1530 goto bind_rx_fail;
1531 info->rx_irq = err;
1532
1533 return 0;
1534
1535 bind_rx_fail:
1536 unbind_from_irqhandler(info->tx_irq, info);
1537 info->tx_irq = 0;
1538 bind_tx_fail:
1539 xenbus_free_evtchn(info->xbdev, info->rx_evtchn);
1540 info->rx_evtchn = 0;
1541 alloc_rx_evtchn_fail:
1542 xenbus_free_evtchn(info->xbdev, info->tx_evtchn);
1543 info->tx_evtchn = 0;
1544 fail:
1545 return err;
1546 }
1547
1548 static int setup_netfront(struct xenbus_device *dev, struct netfront_info *info)
1549 {
1550 struct xen_netif_tx_sring *txs;
1551 struct xen_netif_rx_sring *rxs;
1552 int err;
1553 struct net_device *netdev = info->netdev;
1554 unsigned int feature_split_evtchn;
1555
1556 info->tx_ring_ref = GRANT_INVALID_REF;
1557 info->rx_ring_ref = GRANT_INVALID_REF;
1558 info->rx.sring = NULL;
1559 info->tx.sring = NULL;
1560 netdev->irq = 0;
1561
1562 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1563 "feature-split-event-channels", "%u",
1564 &feature_split_evtchn);
1565 if (err < 0)
1566 feature_split_evtchn = 0;
1567
1568 err = xen_net_read_mac(dev, netdev->dev_addr);
1569 if (err) {
1570 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
1571 goto fail;
1572 }
1573
1574 txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1575 if (!txs) {
1576 err = -ENOMEM;
1577 xenbus_dev_fatal(dev, err, "allocating tx ring page");
1578 goto fail;
1579 }
1580 SHARED_RING_INIT(txs);
1581 FRONT_RING_INIT(&info->tx, txs, PAGE_SIZE);
1582
1583 err = xenbus_grant_ring(dev, virt_to_mfn(txs));
1584 if (err < 0)
1585 goto grant_tx_ring_fail;
1586
1587 info->tx_ring_ref = err;
1588 rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1589 if (!rxs) {
1590 err = -ENOMEM;
1591 xenbus_dev_fatal(dev, err, "allocating rx ring page");
1592 goto alloc_rx_ring_fail;
1593 }
1594 SHARED_RING_INIT(rxs);
1595 FRONT_RING_INIT(&info->rx, rxs, PAGE_SIZE);
1596
1597 err = xenbus_grant_ring(dev, virt_to_mfn(rxs));
1598 if (err < 0)
1599 goto grant_rx_ring_fail;
1600 info->rx_ring_ref = err;
1601
1602 if (feature_split_evtchn)
1603 err = setup_netfront_split(info);
1604 /* setup single event channel if
1605 * a) feature-split-event-channels == 0
1606 * b) feature-split-event-channels == 1 but failed to setup
1607 */
1608 if (!feature_split_evtchn || (feature_split_evtchn && err))
1609 err = setup_netfront_single(info);
1610
1611 if (err)
1612 goto alloc_evtchn_fail;
1613
1614 return 0;
1615
1616 /* If we fail to setup netfront, it is safe to just revoke access to
1617 * granted pages because backend is not accessing it at this point.
1618 */
1619 alloc_evtchn_fail:
1620 gnttab_end_foreign_access_ref(info->rx_ring_ref, 0);
1621 grant_rx_ring_fail:
1622 free_page((unsigned long)rxs);
1623 alloc_rx_ring_fail:
1624 gnttab_end_foreign_access_ref(info->tx_ring_ref, 0);
1625 grant_tx_ring_fail:
1626 free_page((unsigned long)txs);
1627 fail:
1628 return err;
1629 }
1630
1631 /* Common code used when first setting up, and when resuming. */
1632 static int talk_to_netback(struct xenbus_device *dev,
1633 struct netfront_info *info)
1634 {
1635 const char *message;
1636 struct xenbus_transaction xbt;
1637 int err;
1638
1639 /* Create shared ring, alloc event channel. */
1640 err = setup_netfront(dev, info);
1641 if (err)
1642 goto out;
1643
1644 again:
1645 err = xenbus_transaction_start(&xbt);
1646 if (err) {
1647 xenbus_dev_fatal(dev, err, "starting transaction");
1648 goto destroy_ring;
1649 }
1650
1651 err = xenbus_printf(xbt, dev->nodename, "tx-ring-ref", "%u",
1652 info->tx_ring_ref);
1653 if (err) {
1654 message = "writing tx ring-ref";
1655 goto abort_transaction;
1656 }
1657 err = xenbus_printf(xbt, dev->nodename, "rx-ring-ref", "%u",
1658 info->rx_ring_ref);
1659 if (err) {
1660 message = "writing rx ring-ref";
1661 goto abort_transaction;
1662 }
1663
1664 if (info->tx_evtchn == info->rx_evtchn) {
1665 err = xenbus_printf(xbt, dev->nodename,
1666 "event-channel", "%u", info->tx_evtchn);
1667 if (err) {
1668 message = "writing event-channel";
1669 goto abort_transaction;
1670 }
1671 } else {
1672 err = xenbus_printf(xbt, dev->nodename,
1673 "event-channel-tx", "%u", info->tx_evtchn);
1674 if (err) {
1675 message = "writing event-channel-tx";
1676 goto abort_transaction;
1677 }
1678 err = xenbus_printf(xbt, dev->nodename,
1679 "event-channel-rx", "%u", info->rx_evtchn);
1680 if (err) {
1681 message = "writing event-channel-rx";
1682 goto abort_transaction;
1683 }
1684 }
1685
1686 err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
1687 1);
1688 if (err) {
1689 message = "writing request-rx-copy";
1690 goto abort_transaction;
1691 }
1692
1693 err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
1694 if (err) {
1695 message = "writing feature-rx-notify";
1696 goto abort_transaction;
1697 }
1698
1699 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
1700 if (err) {
1701 message = "writing feature-sg";
1702 goto abort_transaction;
1703 }
1704
1705 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
1706 if (err) {
1707 message = "writing feature-gso-tcpv4";
1708 goto abort_transaction;
1709 }
1710
1711 err = xenbus_transaction_end(xbt, 0);
1712 if (err) {
1713 if (err == -EAGAIN)
1714 goto again;
1715 xenbus_dev_fatal(dev, err, "completing transaction");
1716 goto destroy_ring;
1717 }
1718
1719 return 0;
1720
1721 abort_transaction:
1722 xenbus_transaction_end(xbt, 1);
1723 xenbus_dev_fatal(dev, err, "%s", message);
1724 destroy_ring:
1725 xennet_disconnect_backend(info);
1726 out:
1727 return err;
1728 }
1729
1730 static int xennet_connect(struct net_device *dev)
1731 {
1732 struct netfront_info *np = netdev_priv(dev);
1733 int i, requeue_idx, err;
1734 struct sk_buff *skb;
1735 grant_ref_t ref;
1736 struct xen_netif_rx_request *req;
1737 unsigned int feature_rx_copy;
1738
1739 err = xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1740 "feature-rx-copy", "%u", &feature_rx_copy);
1741 if (err != 1)
1742 feature_rx_copy = 0;
1743
1744 if (!feature_rx_copy) {
1745 dev_info(&dev->dev,
1746 "backend does not support copying receive path\n");
1747 return -ENODEV;
1748 }
1749
1750 err = talk_to_netback(np->xbdev, np);
1751 if (err)
1752 return err;
1753
1754 rtnl_lock();
1755 netdev_update_features(dev);
1756 rtnl_unlock();
1757
1758 spin_lock_bh(&np->rx_lock);
1759 spin_lock_irq(&np->tx_lock);
1760
1761 /* Step 1: Discard all pending TX packet fragments. */
1762 xennet_release_tx_bufs(np);
1763
1764 /* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */
1765 for (requeue_idx = 0, i = 0; i < NET_RX_RING_SIZE; i++) {
1766 skb_frag_t *frag;
1767 const struct page *page;
1768 if (!np->rx_skbs[i])
1769 continue;
1770
1771 skb = np->rx_skbs[requeue_idx] = xennet_get_rx_skb(np, i);
1772 ref = np->grant_rx_ref[requeue_idx] = xennet_get_rx_ref(np, i);
1773 req = RING_GET_REQUEST(&np->rx, requeue_idx);
1774
1775 frag = &skb_shinfo(skb)->frags[0];
1776 page = skb_frag_page(frag);
1777 gnttab_grant_foreign_access_ref(
1778 ref, np->xbdev->otherend_id,
1779 pfn_to_mfn(page_to_pfn(page)),
1780 0);
1781 req->gref = ref;
1782 req->id = requeue_idx;
1783
1784 requeue_idx++;
1785 }
1786
1787 np->rx.req_prod_pvt = requeue_idx;
1788
1789 /*
1790 * Step 3: All public and private state should now be sane. Get
1791 * ready to start sending and receiving packets and give the driver
1792 * domain a kick because we've probably just requeued some
1793 * packets.
1794 */
1795 netif_carrier_on(np->netdev);
1796 notify_remote_via_irq(np->tx_irq);
1797 if (np->tx_irq != np->rx_irq)
1798 notify_remote_via_irq(np->rx_irq);
1799 xennet_tx_buf_gc(dev);
1800 xennet_alloc_rx_buffers(dev);
1801
1802 spin_unlock_irq(&np->tx_lock);
1803 spin_unlock_bh(&np->rx_lock);
1804
1805 return 0;
1806 }
1807
1808 /**
1809 * Callback received when the backend's state changes.
1810 */
1811 static void netback_changed(struct xenbus_device *dev,
1812 enum xenbus_state backend_state)
1813 {
1814 struct netfront_info *np = dev_get_drvdata(&dev->dev);
1815 struct net_device *netdev = np->netdev;
1816
1817 dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
1818
1819 switch (backend_state) {
1820 case XenbusStateInitialising:
1821 case XenbusStateInitialised:
1822 case XenbusStateReconfiguring:
1823 case XenbusStateReconfigured:
1824 case XenbusStateUnknown:
1825 case XenbusStateClosed:
1826 break;
1827
1828 case XenbusStateInitWait:
1829 if (dev->state != XenbusStateInitialising)
1830 break;
1831 if (xennet_connect(netdev) != 0)
1832 break;
1833 xenbus_switch_state(dev, XenbusStateConnected);
1834 break;
1835
1836 case XenbusStateConnected:
1837 netdev_notify_peers(netdev);
1838 break;
1839
1840 case XenbusStateClosing:
1841 xenbus_frontend_closed(dev);
1842 break;
1843 }
1844 }
1845
1846 static const struct xennet_stat {
1847 char name[ETH_GSTRING_LEN];
1848 u16 offset;
1849 } xennet_stats[] = {
1850 {
1851 "rx_gso_checksum_fixup",
1852 offsetof(struct netfront_info, rx_gso_checksum_fixup)
1853 },
1854 };
1855
1856 static int xennet_get_sset_count(struct net_device *dev, int string_set)
1857 {
1858 switch (string_set) {
1859 case ETH_SS_STATS:
1860 return ARRAY_SIZE(xennet_stats);
1861 default:
1862 return -EINVAL;
1863 }
1864 }
1865
1866 static void xennet_get_ethtool_stats(struct net_device *dev,
1867 struct ethtool_stats *stats, u64 * data)
1868 {
1869 void *np = netdev_priv(dev);
1870 int i;
1871
1872 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
1873 data[i] = *(unsigned long *)(np + xennet_stats[i].offset);
1874 }
1875
1876 static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
1877 {
1878 int i;
1879
1880 switch (stringset) {
1881 case ETH_SS_STATS:
1882 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
1883 memcpy(data + i * ETH_GSTRING_LEN,
1884 xennet_stats[i].name, ETH_GSTRING_LEN);
1885 break;
1886 }
1887 }
1888
1889 static const struct ethtool_ops xennet_ethtool_ops =
1890 {
1891 .get_link = ethtool_op_get_link,
1892
1893 .get_sset_count = xennet_get_sset_count,
1894 .get_ethtool_stats = xennet_get_ethtool_stats,
1895 .get_strings = xennet_get_strings,
1896 };
1897
1898 #ifdef CONFIG_SYSFS
1899 static ssize_t show_rxbuf_min(struct device *dev,
1900 struct device_attribute *attr, char *buf)
1901 {
1902 struct net_device *netdev = to_net_dev(dev);
1903 struct netfront_info *info = netdev_priv(netdev);
1904
1905 return sprintf(buf, "%u\n", info->rx_min_target);
1906 }
1907
1908 static ssize_t store_rxbuf_min(struct device *dev,
1909 struct device_attribute *attr,
1910 const char *buf, size_t len)
1911 {
1912 struct net_device *netdev = to_net_dev(dev);
1913 struct netfront_info *np = netdev_priv(netdev);
1914 char *endp;
1915 unsigned long target;
1916
1917 if (!capable(CAP_NET_ADMIN))
1918 return -EPERM;
1919
1920 target = simple_strtoul(buf, &endp, 0);
1921 if (endp == buf)
1922 return -EBADMSG;
1923
1924 if (target < RX_MIN_TARGET)
1925 target = RX_MIN_TARGET;
1926 if (target > RX_MAX_TARGET)
1927 target = RX_MAX_TARGET;
1928
1929 spin_lock_bh(&np->rx_lock);
1930 if (target > np->rx_max_target)
1931 np->rx_max_target = target;
1932 np->rx_min_target = target;
1933 if (target > np->rx_target)
1934 np->rx_target = target;
1935
1936 xennet_alloc_rx_buffers(netdev);
1937
1938 spin_unlock_bh(&np->rx_lock);
1939 return len;
1940 }
1941
1942 static ssize_t show_rxbuf_max(struct device *dev,
1943 struct device_attribute *attr, char *buf)
1944 {
1945 struct net_device *netdev = to_net_dev(dev);
1946 struct netfront_info *info = netdev_priv(netdev);
1947
1948 return sprintf(buf, "%u\n", info->rx_max_target);
1949 }
1950
1951 static ssize_t store_rxbuf_max(struct device *dev,
1952 struct device_attribute *attr,
1953 const char *buf, size_t len)
1954 {
1955 struct net_device *netdev = to_net_dev(dev);
1956 struct netfront_info *np = netdev_priv(netdev);
1957 char *endp;
1958 unsigned long target;
1959
1960 if (!capable(CAP_NET_ADMIN))
1961 return -EPERM;
1962
1963 target = simple_strtoul(buf, &endp, 0);
1964 if (endp == buf)
1965 return -EBADMSG;
1966
1967 if (target < RX_MIN_TARGET)
1968 target = RX_MIN_TARGET;
1969 if (target > RX_MAX_TARGET)
1970 target = RX_MAX_TARGET;
1971
1972 spin_lock_bh(&np->rx_lock);
1973 if (target < np->rx_min_target)
1974 np->rx_min_target = target;
1975 np->rx_max_target = target;
1976 if (target < np->rx_target)
1977 np->rx_target = target;
1978
1979 xennet_alloc_rx_buffers(netdev);
1980
1981 spin_unlock_bh(&np->rx_lock);
1982 return len;
1983 }
1984
1985 static ssize_t show_rxbuf_cur(struct device *dev,
1986 struct device_attribute *attr, char *buf)
1987 {
1988 struct net_device *netdev = to_net_dev(dev);
1989 struct netfront_info *info = netdev_priv(netdev);
1990
1991 return sprintf(buf, "%u\n", info->rx_target);
1992 }
1993
1994 static struct device_attribute xennet_attrs[] = {
1995 __ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf_min, store_rxbuf_min),
1996 __ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf_max, store_rxbuf_max),
1997 __ATTR(rxbuf_cur, S_IRUGO, show_rxbuf_cur, NULL),
1998 };
1999
2000 static int xennet_sysfs_addif(struct net_device *netdev)
2001 {
2002 int i;
2003 int err;
2004
2005 for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++) {
2006 err = device_create_file(&netdev->dev,
2007 &xennet_attrs[i]);
2008 if (err)
2009 goto fail;
2010 }
2011 return 0;
2012
2013 fail:
2014 while (--i >= 0)
2015 device_remove_file(&netdev->dev, &xennet_attrs[i]);
2016 return err;
2017 }
2018
2019 static void xennet_sysfs_delif(struct net_device *netdev)
2020 {
2021 int i;
2022
2023 for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++)
2024 device_remove_file(&netdev->dev, &xennet_attrs[i]);
2025 }
2026
2027 #endif /* CONFIG_SYSFS */
2028
2029 static const struct xenbus_device_id netfront_ids[] = {
2030 { "vif" },
2031 { "" }
2032 };
2033
2034
2035 static int xennet_remove(struct xenbus_device *dev)
2036 {
2037 struct netfront_info *info = dev_get_drvdata(&dev->dev);
2038
2039 dev_dbg(&dev->dev, "%s\n", dev->nodename);
2040
2041 xennet_disconnect_backend(info);
2042
2043 xennet_sysfs_delif(info->netdev);
2044
2045 unregister_netdev(info->netdev);
2046
2047 del_timer_sync(&info->rx_refill_timer);
2048
2049 free_percpu(info->stats);
2050
2051 free_netdev(info->netdev);
2052
2053 return 0;
2054 }
2055
2056 static DEFINE_XENBUS_DRIVER(netfront, ,
2057 .probe = netfront_probe,
2058 .remove = xennet_remove,
2059 .resume = netfront_resume,
2060 .otherend_changed = netback_changed,
2061 );
2062
2063 static int __init netif_init(void)
2064 {
2065 if (!xen_domain())
2066 return -ENODEV;
2067
2068 if (!xen_has_pv_nic_devices())
2069 return -ENODEV;
2070
2071 pr_info("Initialising Xen virtual ethernet driver\n");
2072
2073 return xenbus_register_frontend(&netfront_driver);
2074 }
2075 module_init(netif_init);
2076
2077
2078 static void __exit netif_exit(void)
2079 {
2080 xenbus_unregister_driver(&netfront_driver);
2081 }
2082 module_exit(netif_exit);
2083
2084 MODULE_DESCRIPTION("Xen virtual network device frontend");
2085 MODULE_LICENSE("GPL");
2086 MODULE_ALIAS("xen:vif");
2087 MODULE_ALIAS("xennet");
Linux with added FPC-III support