?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge tag 'v3.3.7' into 3.3/master
[zen-stable.git] / drivers / net / xen-netfront.c
blob698b905058dd23435b3f13b40ab4e0b7c870b5c5
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 #include <linux/module.h>
33 #include <linux/kernel.h>
34 #include <linux/netdevice.h>
35 #include <linux/etherdevice.h>
36 #include <linux/skbuff.h>
37 #include <linux/ethtool.h>
38 #include <linux/if_ether.h>
39 #include <linux/tcp.h>
40 #include <linux/udp.h>
41 #include <linux/moduleparam.h>
42 #include <linux/mm.h>
43 #include <linux/slab.h>
44 #include <net/ip.h>
45
46 #include <xen/xen.h>
47 #include <xen/xenbus.h>
48 #include <xen/events.h>
49 #include <xen/page.h>
50 #include <xen/grant_table.h>
51
52 #include <xen/interface/io/netif.h>
53 #include <xen/interface/memory.h>
54 #include <xen/interface/grant_table.h>
55
56 static const struct ethtool_ops xennet_ethtool_ops;
57
58 struct netfront_cb {
59 struct page *page;
60 unsigned offset;
61 };
62
63 #define NETFRONT_SKB_CB(skb) ((struct netfront_cb *)((skb)->cb))
64
65 #define RX_COPY_THRESHOLD 256
66
67 #define GRANT_INVALID_REF 0
68
69 #define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, PAGE_SIZE)
70 #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, PAGE_SIZE)
71 #define TX_MAX_TARGET min_t(int, NET_TX_RING_SIZE, 256)
72
73 struct netfront_stats {
74 u64 rx_packets;
75 u64 tx_packets;
76 u64 rx_bytes;
77 u64 tx_bytes;
78 struct u64_stats_sync syncp;
79 };
80
81 struct netfront_info {
82 struct list_head list;
83 struct net_device *netdev;
84
85 struct napi_struct napi;
86
87 unsigned int evtchn;
88 struct xenbus_device *xbdev;
89
90 spinlock_t tx_lock;
91 struct xen_netif_tx_front_ring tx;
92 int tx_ring_ref;
93
94 /*
95 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
96 * are linked from tx_skb_freelist through skb_entry.link.
97 *
98 * NB. Freelist index entries are always going to be less than
99 * PAGE_OFFSET, whereas pointers to skbs will always be equal or
100 * greater than PAGE_OFFSET: we use this property to distinguish
101 * them.
102 */
103 union skb_entry {
104 struct sk_buff *skb;
105 unsigned long link;
106 } tx_skbs[NET_TX_RING_SIZE];
107 grant_ref_t gref_tx_head;
108 grant_ref_t grant_tx_ref[NET_TX_RING_SIZE];
109 unsigned tx_skb_freelist;
110
111 spinlock_t rx_lock ____cacheline_aligned_in_smp;
112 struct xen_netif_rx_front_ring rx;
113 int rx_ring_ref;
114
115 /* Receive-ring batched refills. */
116 #define RX_MIN_TARGET 8
117 #define RX_DFL_MIN_TARGET 64
118 #define RX_MAX_TARGET min_t(int, NET_RX_RING_SIZE, 256)
119 unsigned rx_min_target, rx_max_target, rx_target;
120 struct sk_buff_head rx_batch;
121
122 struct timer_list rx_refill_timer;
123
124 struct sk_buff *rx_skbs[NET_RX_RING_SIZE];
125 grant_ref_t gref_rx_head;
126 grant_ref_t grant_rx_ref[NET_RX_RING_SIZE];
127
128 unsigned long rx_pfn_array[NET_RX_RING_SIZE];
129 struct multicall_entry rx_mcl[NET_RX_RING_SIZE+1];
130 struct mmu_update rx_mmu[NET_RX_RING_SIZE];
131
132 /* Statistics */
133 struct netfront_stats __percpu *stats;
134
135 unsigned long rx_gso_checksum_fixup;
136 };
137
138 struct netfront_rx_info {
139 struct xen_netif_rx_response rx;
140 struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
141 };
142
143 static void skb_entry_set_link(union skb_entry *list, unsigned short id)
144 {
145 list->link = id;
146 }
147
148 static int skb_entry_is_link(const union skb_entry *list)
149 {
150 BUILD_BUG_ON(sizeof(list->skb) != sizeof(list->link));
151 return (unsigned long)list->skb < PAGE_OFFSET;
152 }
153
154 /*
155 * Access macros for acquiring freeing slots in tx_skbs[].
156 */
157
158 static void add_id_to_freelist(unsigned *head, union skb_entry *list,
159 unsigned short id)
160 {
161 skb_entry_set_link(&list[id], *head);
162 *head = id;
163 }
164
165 static unsigned short get_id_from_freelist(unsigned *head,
166 union skb_entry *list)
167 {
168 unsigned int id = *head;
169 *head = list[id].link;
170 return id;
171 }
172
173 static int xennet_rxidx(RING_IDX idx)
174 {
175 return idx & (NET_RX_RING_SIZE - 1);
176 }
177
178 static struct sk_buff *xennet_get_rx_skb(struct netfront_info *np,
179 RING_IDX ri)
180 {
181 int i = xennet_rxidx(ri);
182 struct sk_buff *skb = np->rx_skbs[i];
183 np->rx_skbs[i] = NULL;
184 return skb;
185 }
186
187 static grant_ref_t xennet_get_rx_ref(struct netfront_info *np,
188 RING_IDX ri)
189 {
190 int i = xennet_rxidx(ri);
191 grant_ref_t ref = np->grant_rx_ref[i];
192 np->grant_rx_ref[i] = GRANT_INVALID_REF;
193 return ref;
194 }
195
196 #ifdef CONFIG_SYSFS
197 static int xennet_sysfs_addif(struct net_device *netdev);
198 static void xennet_sysfs_delif(struct net_device *netdev);
199 #else /* !CONFIG_SYSFS */
200 #define xennet_sysfs_addif(dev) (0)
201 #define xennet_sysfs_delif(dev) do { } while (0)
202 #endif
203
204 static bool xennet_can_sg(struct net_device *dev)
205 {
206 return dev->features & NETIF_F_SG;
207 }
208
209
210 static void rx_refill_timeout(unsigned long data)
211 {
212 struct net_device *dev = (struct net_device *)data;
213 struct netfront_info *np = netdev_priv(dev);
214 napi_schedule(&np->napi);
215 }
216
217 static int netfront_tx_slot_available(struct netfront_info *np)
218 {
219 return (np->tx.req_prod_pvt - np->tx.rsp_cons) <
220 (TX_MAX_TARGET - MAX_SKB_FRAGS - 2);
221 }
222
223 static void xennet_maybe_wake_tx(struct net_device *dev)
224 {
225 struct netfront_info *np = netdev_priv(dev);
226
227 if (unlikely(netif_queue_stopped(dev)) &&
228 netfront_tx_slot_available(np) &&
229 likely(netif_running(dev)))
230 netif_wake_queue(dev);
231 }
232
233 static void xennet_alloc_rx_buffers(struct net_device *dev)
234 {
235 unsigned short id;
236 struct netfront_info *np = netdev_priv(dev);
237 struct sk_buff *skb;
238 struct page *page;
239 int i, batch_target, notify;
240 RING_IDX req_prod = np->rx.req_prod_pvt;
241 grant_ref_t ref;
242 unsigned long pfn;
243 void *vaddr;
244 struct xen_netif_rx_request *req;
245
246 if (unlikely(!netif_carrier_ok(dev)))
247 return;
248
249 /*
250 * Allocate skbuffs greedily, even though we batch updates to the
251 * receive ring. This creates a less bursty demand on the memory
252 * allocator, so should reduce the chance of failed allocation requests
253 * both for ourself and for other kernel subsystems.
254 */
255 batch_target = np->rx_target - (req_prod - np->rx.rsp_cons);
256 for (i = skb_queue_len(&np->rx_batch); i < batch_target; i++) {
257 skb = __netdev_alloc_skb(dev, RX_COPY_THRESHOLD + NET_IP_ALIGN,
258 GFP_ATOMIC | __GFP_NOWARN);
259 if (unlikely(!skb))
260 goto no_skb;
261
262 /* Align ip header to a 16 bytes boundary */
263 skb_reserve(skb, NET_IP_ALIGN);
264
265 page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
266 if (!page) {
267 kfree_skb(skb);
268 no_skb:
269 /* Any skbuffs queued for refill? Force them out. */
270 if (i != 0)
271 goto refill;
272 /* Could not allocate any skbuffs. Try again later. */
273 mod_timer(&np->rx_refill_timer,
274 jiffies + (HZ/10));
275 break;
276 }
277
278 __skb_fill_page_desc(skb, 0, page, 0, 0);
279 skb_shinfo(skb)->nr_frags = 1;
280 __skb_queue_tail(&np->rx_batch, skb);
281 }
282
283 /* Is the batch large enough to be worthwhile? */
284 if (i < (np->rx_target/2)) {
285 if (req_prod > np->rx.sring->req_prod)
286 goto push;
287 return;
288 }
289
290 /* Adjust our fill target if we risked running out of buffers. */
291 if (((req_prod - np->rx.sring->rsp_prod) < (np->rx_target / 4)) &&
292 ((np->rx_target *= 2) > np->rx_max_target))
293 np->rx_target = np->rx_max_target;
294
295 refill:
296 for (i = 0; ; i++) {
297 skb = __skb_dequeue(&np->rx_batch);
298 if (skb == NULL)
299 break;
300
301 skb->dev = dev;
302
303 id = xennet_rxidx(req_prod + i);
304
305 BUG_ON(np->rx_skbs[id]);
306 np->rx_skbs[id] = skb;
307
308 ref = gnttab_claim_grant_reference(&np->gref_rx_head);
309 BUG_ON((signed short)ref < 0);
310 np->grant_rx_ref[id] = ref;
311
312 pfn = page_to_pfn(skb_frag_page(&skb_shinfo(skb)->frags[0]));
313 vaddr = page_address(skb_frag_page(&skb_shinfo(skb)->frags[0]));
314
315 req = RING_GET_REQUEST(&np->rx, req_prod + i);
316 gnttab_grant_foreign_access_ref(ref,
317 np->xbdev->otherend_id,
318 pfn_to_mfn(pfn),
319 0);
320
321 req->id = id;
322 req->gref = ref;
323 }
324
325 wmb(); /* barrier so backend seens requests */
326
327 /* Above is a suitable barrier to ensure backend will see requests. */
328 np->rx.req_prod_pvt = req_prod + i;
329 push:
330 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->rx, notify);
331 if (notify)
332 notify_remote_via_irq(np->netdev->irq);
333 }
334
335 static int xennet_open(struct net_device *dev)
336 {
337 struct netfront_info *np = netdev_priv(dev);
338
339 napi_enable(&np->napi);
340
341 spin_lock_bh(&np->rx_lock);
342 if (netif_carrier_ok(dev)) {
343 xennet_alloc_rx_buffers(dev);
344 np->rx.sring->rsp_event = np->rx.rsp_cons + 1;
345 if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx))
346 napi_schedule(&np->napi);
347 }
348 spin_unlock_bh(&np->rx_lock);
349
350 netif_start_queue(dev);
351
352 return 0;
353 }
354
355 static void xennet_tx_buf_gc(struct net_device *dev)
356 {
357 RING_IDX cons, prod;
358 unsigned short id;
359 struct netfront_info *np = netdev_priv(dev);
360 struct sk_buff *skb;
361
362 BUG_ON(!netif_carrier_ok(dev));
363
364 do {
365 prod = np->tx.sring->rsp_prod;
366 rmb(); /* Ensure we see responses up to 'rp'. */
367
368 for (cons = np->tx.rsp_cons; cons != prod; cons++) {
369 struct xen_netif_tx_response *txrsp;
370
371 txrsp = RING_GET_RESPONSE(&np->tx, cons);
372 if (txrsp->status == XEN_NETIF_RSP_NULL)
373 continue;
374
375 id = txrsp->id;
376 skb = np->tx_skbs[id].skb;
377 if (unlikely(gnttab_query_foreign_access(
378 np->grant_tx_ref[id]) != 0)) {
379 printk(KERN_ALERT "xennet_tx_buf_gc: warning "
380 "-- grant still in use by backend "
381 "domain.\n");
382 BUG();
383 }
384 gnttab_end_foreign_access_ref(
385 np->grant_tx_ref[id], GNTMAP_readonly);
386 gnttab_release_grant_reference(
387 &np->gref_tx_head, np->grant_tx_ref[id]);
388 np->grant_tx_ref[id] = GRANT_INVALID_REF;
389 add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, id);
390 dev_kfree_skb_irq(skb);
391 }
392
393 np->tx.rsp_cons = prod;
394
395 /*
396 * Set a new event, then check for race with update of tx_cons.
397 * Note that it is essential to schedule a callback, no matter
398 * how few buffers are pending. Even if there is space in the
399 * transmit ring, higher layers may be blocked because too much
400 * data is outstanding: in such cases notification from Xen is
401 * likely to be the only kick that we'll get.
402 */
403 np->tx.sring->rsp_event =
404 prod + ((np->tx.sring->req_prod - prod) >> 1) + 1;
405 mb(); /* update shared area */
406 } while ((cons == prod) && (prod != np->tx.sring->rsp_prod));
407
408 xennet_maybe_wake_tx(dev);
409 }
410
411 static void xennet_make_frags(struct sk_buff *skb, struct net_device *dev,
412 struct xen_netif_tx_request *tx)
413 {
414 struct netfront_info *np = netdev_priv(dev);
415 char *data = skb->data;
416 unsigned long mfn;
417 RING_IDX prod = np->tx.req_prod_pvt;
418 int frags = skb_shinfo(skb)->nr_frags;
419 unsigned int offset = offset_in_page(data);
420 unsigned int len = skb_headlen(skb);
421 unsigned int id;
422 grant_ref_t ref;
423 int i;
424
425 /* While the header overlaps a page boundary (including being
426 larger than a page), split it it into page-sized chunks. */
427 while (len > PAGE_SIZE - offset) {
428 tx->size = PAGE_SIZE - offset;
429 tx->flags |= XEN_NETTXF_more_data;
430 len -= tx->size;
431 data += tx->size;
432 offset = 0;
433
434 id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
435 np->tx_skbs[id].skb = skb_get(skb);
436 tx = RING_GET_REQUEST(&np->tx, prod++);
437 tx->id = id;
438 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
439 BUG_ON((signed short)ref < 0);
440
441 mfn = virt_to_mfn(data);
442 gnttab_grant_foreign_access_ref(ref, np->xbdev->otherend_id,
443 mfn, GNTMAP_readonly);
444
445 tx->gref = np->grant_tx_ref[id] = ref;
446 tx->offset = offset;
447 tx->size = len;
448 tx->flags = 0;
449 }
450
451 /* Grant backend access to each skb fragment page. */
452 for (i = 0; i < frags; i++) {
453 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
454
455 tx->flags |= XEN_NETTXF_more_data;
456
457 id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
458 np->tx_skbs[id].skb = skb_get(skb);
459 tx = RING_GET_REQUEST(&np->tx, prod++);
460 tx->id = id;
461 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
462 BUG_ON((signed short)ref < 0);
463
464 mfn = pfn_to_mfn(page_to_pfn(skb_frag_page(frag)));
465 gnttab_grant_foreign_access_ref(ref, np->xbdev->otherend_id,
466 mfn, GNTMAP_readonly);
467
468 tx->gref = np->grant_tx_ref[id] = ref;
469 tx->offset = frag->page_offset;
470 tx->size = skb_frag_size(frag);
471 tx->flags = 0;
472 }
473
474 np->tx.req_prod_pvt = prod;
475 }
476
477 static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
478 {
479 unsigned short id;
480 struct netfront_info *np = netdev_priv(dev);
481 struct netfront_stats *stats = this_cpu_ptr(np->stats);
482 struct xen_netif_tx_request *tx;
483 struct xen_netif_extra_info *extra;
484 char *data = skb->data;
485 RING_IDX i;
486 grant_ref_t ref;
487 unsigned long mfn;
488 int notify;
489 int frags = skb_shinfo(skb)->nr_frags;
490 unsigned int offset = offset_in_page(data);
491 unsigned int len = skb_headlen(skb);
492
493 frags += DIV_ROUND_UP(offset + len, PAGE_SIZE);
494 if (unlikely(frags > MAX_SKB_FRAGS + 1)) {
495 printk(KERN_ALERT "xennet: skb rides the rocket: %d frags\n",
496 frags);
497 dump_stack();
498 goto drop;
499 }
500
501 spin_lock_irq(&np->tx_lock);
502
503 if (unlikely(!netif_carrier_ok(dev) ||
504 (frags > 1 && !xennet_can_sg(dev)) ||
505 netif_needs_gso(skb, netif_skb_features(skb)))) {
506 spin_unlock_irq(&np->tx_lock);
507 goto drop;
508 }
509
510 i = np->tx.req_prod_pvt;
511
512 id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
513 np->tx_skbs[id].skb = skb;
514
515 tx = RING_GET_REQUEST(&np->tx, i);
516
517 tx->id = id;
518 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
519 BUG_ON((signed short)ref < 0);
520 mfn = virt_to_mfn(data);
521 gnttab_grant_foreign_access_ref(
522 ref, np->xbdev->otherend_id, mfn, GNTMAP_readonly);
523 tx->gref = np->grant_tx_ref[id] = ref;
524 tx->offset = offset;
525 tx->size = len;
526 extra = NULL;
527
528 tx->flags = 0;
529 if (skb->ip_summed == CHECKSUM_PARTIAL)
530 /* local packet? */
531 tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated;
532 else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
533 /* remote but checksummed. */
534 tx->flags |= XEN_NETTXF_data_validated;
535
536 if (skb_shinfo(skb)->gso_size) {
537 struct xen_netif_extra_info *gso;
538
539 gso = (struct xen_netif_extra_info *)
540 RING_GET_REQUEST(&np->tx, ++i);
541
542 if (extra)
543 extra->flags |= XEN_NETIF_EXTRA_FLAG_MORE;
544 else
545 tx->flags |= XEN_NETTXF_extra_info;
546
547 gso->u.gso.size = skb_shinfo(skb)->gso_size;
548 gso->u.gso.type = XEN_NETIF_GSO_TYPE_TCPV4;
549 gso->u.gso.pad = 0;
550 gso->u.gso.features = 0;
551
552 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
553 gso->flags = 0;
554 extra = gso;
555 }
556
557 np->tx.req_prod_pvt = i + 1;
558
559 xennet_make_frags(skb, dev, tx);
560 tx->size = skb->len;
561
562 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->tx, notify);
563 if (notify)
564 notify_remote_via_irq(np->netdev->irq);
565
566 u64_stats_update_begin(&stats->syncp);
567 stats->tx_bytes += skb->len;
568 stats->tx_packets++;
569 u64_stats_update_end(&stats->syncp);
570
571 /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
572 xennet_tx_buf_gc(dev);
573
574 if (!netfront_tx_slot_available(np))
575 netif_stop_queue(dev);
576
577 spin_unlock_irq(&np->tx_lock);
578
579 return NETDEV_TX_OK;
580
581 drop:
582 dev->stats.tx_dropped++;
583 dev_kfree_skb(skb);
584 return NETDEV_TX_OK;
585 }
586
587 static int xennet_close(struct net_device *dev)
588 {
589 struct netfront_info *np = netdev_priv(dev);
590 netif_stop_queue(np->netdev);
591 napi_disable(&np->napi);
592 return 0;
593 }
594
595 static void xennet_move_rx_slot(struct netfront_info *np, struct sk_buff *skb,
596 grant_ref_t ref)
597 {
598 int new = xennet_rxidx(np->rx.req_prod_pvt);
599
600 BUG_ON(np->rx_skbs[new]);
601 np->rx_skbs[new] = skb;
602 np->grant_rx_ref[new] = ref;
603 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->id = new;
604 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->gref = ref;
605 np->rx.req_prod_pvt++;
606 }
607
608 static int xennet_get_extras(struct netfront_info *np,
609 struct xen_netif_extra_info *extras,
610 RING_IDX rp)
611
612 {
613 struct xen_netif_extra_info *extra;
614 struct device *dev = &np->netdev->dev;
615 RING_IDX cons = np->rx.rsp_cons;
616 int err = 0;
617
618 do {
619 struct sk_buff *skb;
620 grant_ref_t ref;
621
622 if (unlikely(cons + 1 == rp)) {
623 if (net_ratelimit())
624 dev_warn(dev, "Missing extra info\n");
625 err = -EBADR;
626 break;
627 }
628
629 extra = (struct xen_netif_extra_info *)
630 RING_GET_RESPONSE(&np->rx, ++cons);
631
632 if (unlikely(!extra->type ||
633 extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
634 if (net_ratelimit())
635 dev_warn(dev, "Invalid extra type: %d\n",
636 extra->type);
637 err = -EINVAL;
638 } else {
639 memcpy(&extras[extra->type - 1], extra,
640 sizeof(*extra));
641 }
642
643 skb = xennet_get_rx_skb(np, cons);
644 ref = xennet_get_rx_ref(np, cons);
645 xennet_move_rx_slot(np, skb, ref);
646 } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
647
648 np->rx.rsp_cons = cons;
649 return err;
650 }
651
652 static int xennet_get_responses(struct netfront_info *np,
653 struct netfront_rx_info *rinfo, RING_IDX rp,
654 struct sk_buff_head *list)
655 {
656 struct xen_netif_rx_response *rx = &rinfo->rx;
657 struct xen_netif_extra_info *extras = rinfo->extras;
658 struct device *dev = &np->netdev->dev;
659 RING_IDX cons = np->rx.rsp_cons;
660 struct sk_buff *skb = xennet_get_rx_skb(np, cons);
661 grant_ref_t ref = xennet_get_rx_ref(np, cons);
662 int max = MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD);
663 int frags = 1;
664 int err = 0;
665 unsigned long ret;
666
667 if (rx->flags & XEN_NETRXF_extra_info) {
668 err = xennet_get_extras(np, extras, rp);
669 cons = np->rx.rsp_cons;
670 }
671
672 for (;;) {
673 if (unlikely(rx->status < 0 ||
674 rx->offset + rx->status > PAGE_SIZE)) {
675 if (net_ratelimit())
676 dev_warn(dev, "rx->offset: %x, size: %u\n",
677 rx->offset, rx->status);
678 xennet_move_rx_slot(np, skb, ref);
679 err = -EINVAL;
680 goto next;
681 }
682
683 /*
684 * This definitely indicates a bug, either in this driver or in
685 * the backend driver. In future this should flag the bad
686 * situation to the system controller to reboot the backed.
687 */
688 if (ref == GRANT_INVALID_REF) {
689 if (net_ratelimit())
690 dev_warn(dev, "Bad rx response id %d.\n",
691 rx->id);
692 err = -EINVAL;
693 goto next;
694 }
695
696 ret = gnttab_end_foreign_access_ref(ref, 0);
697 BUG_ON(!ret);
698
699 gnttab_release_grant_reference(&np->gref_rx_head, ref);
700
701 __skb_queue_tail(list, skb);
702
703 next:
704 if (!(rx->flags & XEN_NETRXF_more_data))
705 break;
706
707 if (cons + frags == rp) {
708 if (net_ratelimit())
709 dev_warn(dev, "Need more frags\n");
710 err = -ENOENT;
711 break;
712 }
713
714 rx = RING_GET_RESPONSE(&np->rx, cons + frags);
715 skb = xennet_get_rx_skb(np, cons + frags);
716 ref = xennet_get_rx_ref(np, cons + frags);
717 frags++;
718 }
719
720 if (unlikely(frags > max)) {
721 if (net_ratelimit())
722 dev_warn(dev, "Too many frags\n");
723 err = -E2BIG;
724 }
725
726 if (unlikely(err))
727 np->rx.rsp_cons = cons + frags;
728
729 return err;
730 }
731
732 static int xennet_set_skb_gso(struct sk_buff *skb,
733 struct xen_netif_extra_info *gso)
734 {
735 if (!gso->u.gso.size) {
736 if (net_ratelimit())
737 printk(KERN_WARNING "GSO size must not be zero.\n");
738 return -EINVAL;
739 }
740
741 /* Currently only TCPv4 S.O. is supported. */
742 if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4) {
743 if (net_ratelimit())
744 printk(KERN_WARNING "Bad GSO type %d.\n", gso->u.gso.type);
745 return -EINVAL;
746 }
747
748 skb_shinfo(skb)->gso_size = gso->u.gso.size;
749 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
750
751 /* Header must be checked, and gso_segs computed. */
752 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
753 skb_shinfo(skb)->gso_segs = 0;
754
755 return 0;
756 }
757
758 static RING_IDX xennet_fill_frags(struct netfront_info *np,
759 struct sk_buff *skb,
760 struct sk_buff_head *list)
761 {
762 struct skb_shared_info *shinfo = skb_shinfo(skb);
763 int nr_frags = shinfo->nr_frags;
764 RING_IDX cons = np->rx.rsp_cons;
765 struct sk_buff *nskb;
766
767 while ((nskb = __skb_dequeue(list))) {
768 struct xen_netif_rx_response *rx =
769 RING_GET_RESPONSE(&np->rx, ++cons);
770 skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0];
771
772 __skb_fill_page_desc(skb, nr_frags,
773 skb_frag_page(nfrag),
774 rx->offset, rx->status);
775
776 skb->data_len += rx->status;
777
778 skb_shinfo(nskb)->nr_frags = 0;
779 kfree_skb(nskb);
780
781 nr_frags++;
782 }
783
784 shinfo->nr_frags = nr_frags;
785 return cons;
786 }
787
788 static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
789 {
790 struct iphdr *iph;
791 unsigned char *th;
792 int err = -EPROTO;
793 int recalculate_partial_csum = 0;
794
795 /*
796 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
797 * peers can fail to set NETRXF_csum_blank when sending a GSO
798 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
799 * recalculate the partial checksum.
800 */
801 if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
802 struct netfront_info *np = netdev_priv(dev);
803 np->rx_gso_checksum_fixup++;
804 skb->ip_summed = CHECKSUM_PARTIAL;
805 recalculate_partial_csum = 1;
806 }
807
808 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
809 if (skb->ip_summed != CHECKSUM_PARTIAL)
810 return 0;
811
812 if (skb->protocol != htons(ETH_P_IP))
813 goto out;
814
815 iph = (void *)skb->data;
816 th = skb->data + 4 * iph->ihl;
817 if (th >= skb_tail_pointer(skb))
818 goto out;
819
820 skb->csum_start = th - skb->head;
821 switch (iph->protocol) {
822 case IPPROTO_TCP:
823 skb->csum_offset = offsetof(struct tcphdr, check);
824
825 if (recalculate_partial_csum) {
826 struct tcphdr *tcph = (struct tcphdr *)th;
827 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
828 skb->len - iph->ihl*4,
829 IPPROTO_TCP, 0);
830 }
831 break;
832 case IPPROTO_UDP:
833 skb->csum_offset = offsetof(struct udphdr, check);
834
835 if (recalculate_partial_csum) {
836 struct udphdr *udph = (struct udphdr *)th;
837 udph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
838 skb->len - iph->ihl*4,
839 IPPROTO_UDP, 0);
840 }
841 break;
842 default:
843 if (net_ratelimit())
844 printk(KERN_ERR "Attempting to checksum a non-"
845 "TCP/UDP packet, dropping a protocol"
846 " %d packet", iph->protocol);
847 goto out;
848 }
849
850 if ((th + skb->csum_offset + 2) > skb_tail_pointer(skb))
851 goto out;
852
853 err = 0;
854
855 out:
856 return err;
857 }
858
859 static int handle_incoming_queue(struct net_device *dev,
860 struct sk_buff_head *rxq)
861 {
862 struct netfront_info *np = netdev_priv(dev);
863 struct netfront_stats *stats = this_cpu_ptr(np->stats);
864 int packets_dropped = 0;
865 struct sk_buff *skb;
866
867 while ((skb = __skb_dequeue(rxq)) != NULL) {
868 struct page *page = NETFRONT_SKB_CB(skb)->page;
869 void *vaddr = page_address(page);
870 unsigned offset = NETFRONT_SKB_CB(skb)->offset;
871
872 memcpy(skb->data, vaddr + offset,
873 skb_headlen(skb));
874
875 if (page != skb_frag_page(&skb_shinfo(skb)->frags[0]))
876 __free_page(page);
877
878 /* Ethernet work: Delayed to here as it peeks the header. */
879 skb->protocol = eth_type_trans(skb, dev);
880
881 if (checksum_setup(dev, skb)) {
882 kfree_skb(skb);
883 packets_dropped++;
884 dev->stats.rx_errors++;
885 continue;
886 }
887
888 u64_stats_update_begin(&stats->syncp);
889 stats->rx_packets++;
890 stats->rx_bytes += skb->len;
891 u64_stats_update_end(&stats->syncp);
892
893 /* Pass it up. */
894 netif_receive_skb(skb);
895 }
896
897 return packets_dropped;
898 }
899
900 static int xennet_poll(struct napi_struct *napi, int budget)
901 {
902 struct netfront_info *np = container_of(napi, struct netfront_info, napi);
903 struct net_device *dev = np->netdev;
904 struct sk_buff *skb;
905 struct netfront_rx_info rinfo;
906 struct xen_netif_rx_response *rx = &rinfo.rx;
907 struct xen_netif_extra_info *extras = rinfo.extras;
908 RING_IDX i, rp;
909 int work_done;
910 struct sk_buff_head rxq;
911 struct sk_buff_head errq;
912 struct sk_buff_head tmpq;
913 unsigned long flags;
914 unsigned int len;
915 int err;
916
917 spin_lock(&np->rx_lock);
918
919 skb_queue_head_init(&rxq);
920 skb_queue_head_init(&errq);
921 skb_queue_head_init(&tmpq);
922
923 rp = np->rx.sring->rsp_prod;
924 rmb(); /* Ensure we see queued responses up to 'rp'. */
925
926 i = np->rx.rsp_cons;
927 work_done = 0;
928 while ((i != rp) && (work_done < budget)) {
929 memcpy(rx, RING_GET_RESPONSE(&np->rx, i), sizeof(*rx));
930 memset(extras, 0, sizeof(rinfo.extras));
931
932 err = xennet_get_responses(np, &rinfo, rp, &tmpq);
933
934 if (unlikely(err)) {
935 err:
936 while ((skb = __skb_dequeue(&tmpq)))
937 __skb_queue_tail(&errq, skb);
938 dev->stats.rx_errors++;
939 i = np->rx.rsp_cons;
940 continue;
941 }
942
943 skb = __skb_dequeue(&tmpq);
944
945 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
946 struct xen_netif_extra_info *gso;
947 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
948
949 if (unlikely(xennet_set_skb_gso(skb, gso))) {
950 __skb_queue_head(&tmpq, skb);
951 np->rx.rsp_cons += skb_queue_len(&tmpq);
952 goto err;
953 }
954 }
955
956 NETFRONT_SKB_CB(skb)->page =
957 skb_frag_page(&skb_shinfo(skb)->frags[0]);
958 NETFRONT_SKB_CB(skb)->offset = rx->offset;
959
960 len = rx->status;
961 if (len > RX_COPY_THRESHOLD)
962 len = RX_COPY_THRESHOLD;
963 skb_put(skb, len);
964
965 if (rx->status > len) {
966 skb_shinfo(skb)->frags[0].page_offset =
967 rx->offset + len;
968 skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status - len);
969 skb->data_len = rx->status - len;
970 } else {
971 __skb_fill_page_desc(skb, 0, NULL, 0, 0);
972 skb_shinfo(skb)->nr_frags = 0;
973 }
974
975 i = xennet_fill_frags(np, skb, &tmpq);
976
977 /*
978 * Truesize approximates the size of true data plus
979 * any supervisor overheads. Adding hypervisor
980 * overheads has been shown to significantly reduce
981 * achievable bandwidth with the default receive
982 * buffer size. It is therefore not wise to account
983 * for it here.
984 *
985 * After alloc_skb(RX_COPY_THRESHOLD), truesize is set
986 * to RX_COPY_THRESHOLD + the supervisor
987 * overheads. Here, we add the size of the data pulled
988 * in xennet_fill_frags().
989 *
990 * We also adjust for any unused space in the main
991 * data area by subtracting (RX_COPY_THRESHOLD -
992 * len). This is especially important with drivers
993 * which split incoming packets into header and data,
994 * using only 66 bytes of the main data area (see the
995 * e1000 driver for example.) On such systems,
996 * without this last adjustement, our achievable
997 * receive throughout using the standard receive
998 * buffer size was cut by 25%(!!!).
999 */
1000 skb->truesize += skb->data_len - (RX_COPY_THRESHOLD - len);
1001 skb->len += skb->data_len;
1002
1003 if (rx->flags & XEN_NETRXF_csum_blank)
1004 skb->ip_summed = CHECKSUM_PARTIAL;
1005 else if (rx->flags & XEN_NETRXF_data_validated)
1006 skb->ip_summed = CHECKSUM_UNNECESSARY;
1007
1008 __skb_queue_tail(&rxq, skb);
1009
1010 np->rx.rsp_cons = ++i;
1011 work_done++;
1012 }
1013
1014 __skb_queue_purge(&errq);
1015
1016 work_done -= handle_incoming_queue(dev, &rxq);
1017
1018 /* If we get a callback with very few responses, reduce fill target. */
1019 /* NB. Note exponential increase, linear decrease. */
1020 if (((np->rx.req_prod_pvt - np->rx.sring->rsp_prod) >
1021 ((3*np->rx_target) / 4)) &&
1022 (--np->rx_target < np->rx_min_target))
1023 np->rx_target = np->rx_min_target;
1024
1025 xennet_alloc_rx_buffers(dev);
1026
1027 if (work_done < budget) {
1028 int more_to_do = 0;
1029
1030 local_irq_save(flags);
1031
1032 RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, more_to_do);
1033 if (!more_to_do)
1034 __napi_complete(napi);
1035
1036 local_irq_restore(flags);
1037 }
1038
1039 spin_unlock(&np->rx_lock);
1040
1041 return work_done;
1042 }
1043
1044 static int xennet_change_mtu(struct net_device *dev, int mtu)
1045 {
1046 int max = xennet_can_sg(dev) ? 65535 - ETH_HLEN : ETH_DATA_LEN;
1047
1048 if (mtu > max)
1049 return -EINVAL;
1050 dev->mtu = mtu;
1051 return 0;
1052 }
1053
1054 static struct rtnl_link_stats64 *xennet_get_stats64(struct net_device *dev,
1055 struct rtnl_link_stats64 *tot)
1056 {
1057 struct netfront_info *np = netdev_priv(dev);
1058 int cpu;
1059
1060 for_each_possible_cpu(cpu) {
1061 struct netfront_stats *stats = per_cpu_ptr(np->stats, cpu);
1062 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1063 unsigned int start;
1064
1065 do {
1066 start = u64_stats_fetch_begin_bh(&stats->syncp);
1067
1068 rx_packets = stats->rx_packets;
1069 tx_packets = stats->tx_packets;
1070 rx_bytes = stats->rx_bytes;
1071 tx_bytes = stats->tx_bytes;
1072 } while (u64_stats_fetch_retry_bh(&stats->syncp, start));
1073
1074 tot->rx_packets += rx_packets;
1075 tot->tx_packets += tx_packets;
1076 tot->rx_bytes += rx_bytes;
1077 tot->tx_bytes += tx_bytes;
1078 }
1079
1080 tot->rx_errors = dev->stats.rx_errors;
1081 tot->tx_dropped = dev->stats.tx_dropped;
1082
1083 return tot;
1084 }
1085
1086 static void xennet_release_tx_bufs(struct netfront_info *np)
1087 {
1088 struct sk_buff *skb;
1089 int i;
1090
1091 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1092 /* Skip over entries which are actually freelist references */
1093 if (skb_entry_is_link(&np->tx_skbs[i]))
1094 continue;
1095
1096 skb = np->tx_skbs[i].skb;
1097 gnttab_end_foreign_access_ref(np->grant_tx_ref[i],
1098 GNTMAP_readonly);
1099 gnttab_release_grant_reference(&np->gref_tx_head,
1100 np->grant_tx_ref[i]);
1101 np->grant_tx_ref[i] = GRANT_INVALID_REF;
1102 add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, i);
1103 dev_kfree_skb_irq(skb);
1104 }
1105 }
1106
1107 static void xennet_release_rx_bufs(struct netfront_info *np)
1108 {
1109 struct mmu_update *mmu = np->rx_mmu;
1110 struct multicall_entry *mcl = np->rx_mcl;
1111 struct sk_buff_head free_list;
1112 struct sk_buff *skb;
1113 unsigned long mfn;
1114 int xfer = 0, noxfer = 0, unused = 0;
1115 int id, ref;
1116
1117 dev_warn(&np->netdev->dev, "%s: fix me for copying receiver.\n",
1118 __func__);
1119 return;
1120
1121 skb_queue_head_init(&free_list);
1122
1123 spin_lock_bh(&np->rx_lock);
1124
1125 for (id = 0; id < NET_RX_RING_SIZE; id++) {
1126 ref = np->grant_rx_ref[id];
1127 if (ref == GRANT_INVALID_REF) {
1128 unused++;
1129 continue;
1130 }
1131
1132 skb = np->rx_skbs[id];
1133 mfn = gnttab_end_foreign_transfer_ref(ref);
1134 gnttab_release_grant_reference(&np->gref_rx_head, ref);
1135 np->grant_rx_ref[id] = GRANT_INVALID_REF;
1136
1137 if (0 == mfn) {
1138 skb_shinfo(skb)->nr_frags = 0;
1139 dev_kfree_skb(skb);
1140 noxfer++;
1141 continue;
1142 }
1143
1144 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1145 /* Remap the page. */
1146 const struct page *page =
1147 skb_frag_page(&skb_shinfo(skb)->frags[0]);
1148 unsigned long pfn = page_to_pfn(page);
1149 void *vaddr = page_address(page);
1150
1151 MULTI_update_va_mapping(mcl, (unsigned long)vaddr,
1152 mfn_pte(mfn, PAGE_KERNEL),
1153 0);
1154 mcl++;
1155 mmu->ptr = ((u64)mfn << PAGE_SHIFT)
1156 | MMU_MACHPHYS_UPDATE;
1157 mmu->val = pfn;
1158 mmu++;
1159
1160 set_phys_to_machine(pfn, mfn);
1161 }
1162 __skb_queue_tail(&free_list, skb);
1163 xfer++;
1164 }
1165
1166 dev_info(&np->netdev->dev, "%s: %d xfer, %d noxfer, %d unused\n",
1167 __func__, xfer, noxfer, unused);
1168
1169 if (xfer) {
1170 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1171 /* Do all the remapping work and M2P updates. */
1172 MULTI_mmu_update(mcl, np->rx_mmu, mmu - np->rx_mmu,
1173 NULL, DOMID_SELF);
1174 mcl++;
1175 HYPERVISOR_multicall(np->rx_mcl, mcl - np->rx_mcl);
1176 }
1177 }
1178
1179 __skb_queue_purge(&free_list);
1180
1181 spin_unlock_bh(&np->rx_lock);
1182 }
1183
1184 static void xennet_uninit(struct net_device *dev)
1185 {
1186 struct netfront_info *np = netdev_priv(dev);
1187 xennet_release_tx_bufs(np);
1188 xennet_release_rx_bufs(np);
1189 gnttab_free_grant_references(np->gref_tx_head);
1190 gnttab_free_grant_references(np->gref_rx_head);
1191 }
1192
1193 static netdev_features_t xennet_fix_features(struct net_device *dev,
1194 netdev_features_t features)
1195 {
1196 struct netfront_info *np = netdev_priv(dev);
1197 int val;
1198
1199 if (features & NETIF_F_SG) {
1200 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, "feature-sg",
1201 "%d", &val) < 0)
1202 val = 0;
1203
1204 if (!val)
1205 features &= ~NETIF_F_SG;
1206 }
1207
1208 if (features & NETIF_F_TSO) {
1209 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1210 "feature-gso-tcpv4", "%d", &val) < 0)
1211 val = 0;
1212
1213 if (!val)
1214 features &= ~NETIF_F_TSO;
1215 }
1216
1217 return features;
1218 }
1219
1220 static int xennet_set_features(struct net_device *dev,
1221 netdev_features_t features)
1222 {
1223 if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1224 netdev_info(dev, "Reducing MTU because no SG offload");
1225 dev->mtu = ETH_DATA_LEN;
1226 }
1227
1228 return 0;
1229 }
1230
1231 static const struct net_device_ops xennet_netdev_ops = {
1232 .ndo_open = xennet_open,
1233 .ndo_uninit = xennet_uninit,
1234 .ndo_stop = xennet_close,
1235 .ndo_start_xmit = xennet_start_xmit,
1236 .ndo_change_mtu = xennet_change_mtu,
1237 .ndo_get_stats64 = xennet_get_stats64,
1238 .ndo_set_mac_address = eth_mac_addr,
1239 .ndo_validate_addr = eth_validate_addr,
1240 .ndo_fix_features = xennet_fix_features,
1241 .ndo_set_features = xennet_set_features,
1242 };
1243
1244 static struct net_device * __devinit xennet_create_dev(struct xenbus_device *dev)
1245 {
1246 int i, err;
1247 struct net_device *netdev;
1248 struct netfront_info *np;
1249
1250 netdev = alloc_etherdev(sizeof(struct netfront_info));
1251 if (!netdev) {
1252 printk(KERN_WARNING "%s> alloc_etherdev failed.\n",
1253 __func__);
1254 return ERR_PTR(-ENOMEM);
1255 }
1256
1257 np = netdev_priv(netdev);
1258 np->xbdev = dev;
1259
1260 spin_lock_init(&np->tx_lock);
1261 spin_lock_init(&np->rx_lock);
1262
1263 skb_queue_head_init(&np->rx_batch);
1264 np->rx_target = RX_DFL_MIN_TARGET;
1265 np->rx_min_target = RX_DFL_MIN_TARGET;
1266 np->rx_max_target = RX_MAX_TARGET;
1267
1268 init_timer(&np->rx_refill_timer);
1269 np->rx_refill_timer.data = (unsigned long)netdev;
1270 np->rx_refill_timer.function = rx_refill_timeout;
1271
1272 err = -ENOMEM;
1273 np->stats = alloc_percpu(struct netfront_stats);
1274 if (np->stats == NULL)
1275 goto exit;
1276
1277 /* Initialise tx_skbs as a free chain containing every entry. */
1278 np->tx_skb_freelist = 0;
1279 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1280 skb_entry_set_link(&np->tx_skbs[i], i+1);
1281 np->grant_tx_ref[i] = GRANT_INVALID_REF;
1282 }
1283
1284 /* Clear out rx_skbs */
1285 for (i = 0; i < NET_RX_RING_SIZE; i++) {
1286 np->rx_skbs[i] = NULL;
1287 np->grant_rx_ref[i] = GRANT_INVALID_REF;
1288 }
1289
1290 /* A grant for every tx ring slot */
1291 if (gnttab_alloc_grant_references(TX_MAX_TARGET,
1292 &np->gref_tx_head) < 0) {
1293 printk(KERN_ALERT "#### netfront can't alloc tx grant refs\n");
1294 err = -ENOMEM;
1295 goto exit_free_stats;
1296 }
1297 /* A grant for every rx ring slot */
1298 if (gnttab_alloc_grant_references(RX_MAX_TARGET,
1299 &np->gref_rx_head) < 0) {
1300 printk(KERN_ALERT "#### netfront can't alloc rx grant refs\n");
1301 err = -ENOMEM;
1302 goto exit_free_tx;
1303 }
1304
1305 netdev->netdev_ops = &xennet_netdev_ops;
1306
1307 netif_napi_add(netdev, &np->napi, xennet_poll, 64);
1308 netdev->features = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1309 NETIF_F_GSO_ROBUST;
1310 netdev->hw_features = NETIF_F_IP_CSUM | NETIF_F_SG | NETIF_F_TSO;
1311
1312 /*
1313 * Assume that all hw features are available for now. This set
1314 * will be adjusted by the call to netdev_update_features() in
1315 * xennet_connect() which is the earliest point where we can
1316 * negotiate with the backend regarding supported features.
1317 */
1318 netdev->features |= netdev->hw_features;
1319
1320 SET_ETHTOOL_OPS(netdev, &xennet_ethtool_ops);
1321 SET_NETDEV_DEV(netdev, &dev->dev);
1322
1323 np->netdev = netdev;
1324
1325 netif_carrier_off(netdev);
1326
1327 return netdev;
1328
1329 exit_free_tx:
1330 gnttab_free_grant_references(np->gref_tx_head);
1331 exit_free_stats:
1332 free_percpu(np->stats);
1333 exit:
1334 free_netdev(netdev);
1335 return ERR_PTR(err);
1336 }
1337
1338 /**
1339 * Entry point to this code when a new device is created. Allocate the basic
1340 * structures and the ring buffers for communication with the backend, and
1341 * inform the backend of the appropriate details for those.
1342 */
1343 static int __devinit netfront_probe(struct xenbus_device *dev,
1344 const struct xenbus_device_id *id)
1345 {
1346 int err;
1347 struct net_device *netdev;
1348 struct netfront_info *info;
1349
1350 netdev = xennet_create_dev(dev);
1351 if (IS_ERR(netdev)) {
1352 err = PTR_ERR(netdev);
1353 xenbus_dev_fatal(dev, err, "creating netdev");
1354 return err;
1355 }
1356
1357 info = netdev_priv(netdev);
1358 dev_set_drvdata(&dev->dev, info);
1359
1360 err = register_netdev(info->netdev);
1361 if (err) {
1362 printk(KERN_WARNING "%s: register_netdev err=%d\n",
1363 __func__, err);
1364 goto fail;
1365 }
1366
1367 err = xennet_sysfs_addif(info->netdev);
1368 if (err) {
1369 unregister_netdev(info->netdev);
1370 printk(KERN_WARNING "%s: add sysfs failed err=%d\n",
1371 __func__, err);
1372 goto fail;
1373 }
1374
1375 return 0;
1376
1377 fail:
1378 free_netdev(netdev);
1379 dev_set_drvdata(&dev->dev, NULL);
1380 return err;
1381 }
1382
1383 static void xennet_end_access(int ref, void *page)
1384 {
1385 /* This frees the page as a side-effect */
1386 if (ref != GRANT_INVALID_REF)
1387 gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1388 }
1389
1390 static void xennet_disconnect_backend(struct netfront_info *info)
1391 {
1392 /* Stop old i/f to prevent errors whilst we rebuild the state. */
1393 spin_lock_bh(&info->rx_lock);
1394 spin_lock_irq(&info->tx_lock);
1395 netif_carrier_off(info->netdev);
1396 spin_unlock_irq(&info->tx_lock);
1397 spin_unlock_bh(&info->rx_lock);
1398
1399 if (info->netdev->irq)
1400 unbind_from_irqhandler(info->netdev->irq, info->netdev);
1401 info->evtchn = info->netdev->irq = 0;
1402
1403 /* End access and free the pages */
1404 xennet_end_access(info->tx_ring_ref, info->tx.sring);
1405 xennet_end_access(info->rx_ring_ref, info->rx.sring);
1406
1407 info->tx_ring_ref = GRANT_INVALID_REF;
1408 info->rx_ring_ref = GRANT_INVALID_REF;
1409 info->tx.sring = NULL;
1410 info->rx.sring = NULL;
1411 }
1412
1413 /**
1414 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1415 * driver restart. We tear down our netif structure and recreate it, but
1416 * leave the device-layer structures intact so that this is transparent to the
1417 * rest of the kernel.
1418 */
1419 static int netfront_resume(struct xenbus_device *dev)
1420 {
1421 struct netfront_info *info = dev_get_drvdata(&dev->dev);
1422
1423 dev_dbg(&dev->dev, "%s\n", dev->nodename);
1424
1425 xennet_disconnect_backend(info);
1426 return 0;
1427 }
1428
1429 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1430 {
1431 char *s, *e, *macstr;
1432 int i;
1433
1434 macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1435 if (IS_ERR(macstr))
1436 return PTR_ERR(macstr);
1437
1438 for (i = 0; i < ETH_ALEN; i++) {
1439 mac[i] = simple_strtoul(s, &e, 16);
1440 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1441 kfree(macstr);
1442 return -ENOENT;
1443 }
1444 s = e+1;
1445 }
1446
1447 kfree(macstr);
1448 return 0;
1449 }
1450
1451 static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1452 {
1453 struct net_device *dev = dev_id;
1454 struct netfront_info *np = netdev_priv(dev);
1455 unsigned long flags;
1456
1457 spin_lock_irqsave(&np->tx_lock, flags);
1458
1459 if (likely(netif_carrier_ok(dev))) {
1460 xennet_tx_buf_gc(dev);
1461 /* Under tx_lock: protects access to rx shared-ring indexes. */
1462 if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx))
1463 napi_schedule(&np->napi);
1464 }
1465
1466 spin_unlock_irqrestore(&np->tx_lock, flags);
1467
1468 return IRQ_HANDLED;
1469 }
1470
1471 static int setup_netfront(struct xenbus_device *dev, struct netfront_info *info)
1472 {
1473 struct xen_netif_tx_sring *txs;
1474 struct xen_netif_rx_sring *rxs;
1475 int err;
1476 struct net_device *netdev = info->netdev;
1477
1478 info->tx_ring_ref = GRANT_INVALID_REF;
1479 info->rx_ring_ref = GRANT_INVALID_REF;
1480 info->rx.sring = NULL;
1481 info->tx.sring = NULL;
1482 netdev->irq = 0;
1483
1484 err = xen_net_read_mac(dev, netdev->dev_addr);
1485 if (err) {
1486 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
1487 goto fail;
1488 }
1489
1490 txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1491 if (!txs) {
1492 err = -ENOMEM;
1493 xenbus_dev_fatal(dev, err, "allocating tx ring page");
1494 goto fail;
1495 }
1496 SHARED_RING_INIT(txs);
1497 FRONT_RING_INIT(&info->tx, txs, PAGE_SIZE);
1498
1499 err = xenbus_grant_ring(dev, virt_to_mfn(txs));
1500 if (err < 0) {
1501 free_page((unsigned long)txs);
1502 goto fail;
1503 }
1504
1505 info->tx_ring_ref = err;
1506 rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1507 if (!rxs) {
1508 err = -ENOMEM;
1509 xenbus_dev_fatal(dev, err, "allocating rx ring page");
1510 goto fail;
1511 }
1512 SHARED_RING_INIT(rxs);
1513 FRONT_RING_INIT(&info->rx, rxs, PAGE_SIZE);
1514
1515 err = xenbus_grant_ring(dev, virt_to_mfn(rxs));
1516 if (err < 0) {
1517 free_page((unsigned long)rxs);
1518 goto fail;
1519 }
1520 info->rx_ring_ref = err;
1521
1522 err = xenbus_alloc_evtchn(dev, &info->evtchn);
1523 if (err)
1524 goto fail;
1525
1526 err = bind_evtchn_to_irqhandler(info->evtchn, xennet_interrupt,
1527 0, netdev->name, netdev);
1528 if (err < 0)
1529 goto fail;
1530 netdev->irq = err;
1531 return 0;
1532
1533 fail:
1534 return err;
1535 }
1536
1537 /* Common code used when first setting up, and when resuming. */
1538 static int talk_to_netback(struct xenbus_device *dev,
1539 struct netfront_info *info)
1540 {
1541 const char *message;
1542 struct xenbus_transaction xbt;
1543 int err;
1544
1545 /* Create shared ring, alloc event channel. */
1546 err = setup_netfront(dev, info);
1547 if (err)
1548 goto out;
1549
1550 again:
1551 err = xenbus_transaction_start(&xbt);
1552 if (err) {
1553 xenbus_dev_fatal(dev, err, "starting transaction");
1554 goto destroy_ring;
1555 }
1556
1557 err = xenbus_printf(xbt, dev->nodename, "tx-ring-ref", "%u",
1558 info->tx_ring_ref);
1559 if (err) {
1560 message = "writing tx ring-ref";
1561 goto abort_transaction;
1562 }
1563 err = xenbus_printf(xbt, dev->nodename, "rx-ring-ref", "%u",
1564 info->rx_ring_ref);
1565 if (err) {
1566 message = "writing rx ring-ref";
1567 goto abort_transaction;
1568 }
1569 err = xenbus_printf(xbt, dev->nodename,
1570 "event-channel", "%u", info->evtchn);
1571 if (err) {
1572 message = "writing event-channel";
1573 goto abort_transaction;
1574 }
1575
1576 err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
1577 1);
1578 if (err) {
1579 message = "writing request-rx-copy";
1580 goto abort_transaction;
1581 }
1582
1583 err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
1584 if (err) {
1585 message = "writing feature-rx-notify";
1586 goto abort_transaction;
1587 }
1588
1589 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
1590 if (err) {
1591 message = "writing feature-sg";
1592 goto abort_transaction;
1593 }
1594
1595 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
1596 if (err) {
1597 message = "writing feature-gso-tcpv4";
1598 goto abort_transaction;
1599 }
1600
1601 err = xenbus_transaction_end(xbt, 0);
1602 if (err) {
1603 if (err == -EAGAIN)
1604 goto again;
1605 xenbus_dev_fatal(dev, err, "completing transaction");
1606 goto destroy_ring;
1607 }
1608
1609 return 0;
1610
1611 abort_transaction:
1612 xenbus_transaction_end(xbt, 1);
1613 xenbus_dev_fatal(dev, err, "%s", message);
1614 destroy_ring:
1615 xennet_disconnect_backend(info);
1616 out:
1617 return err;
1618 }
1619
1620 static int xennet_connect(struct net_device *dev)
1621 {
1622 struct netfront_info *np = netdev_priv(dev);
1623 int i, requeue_idx, err;
1624 struct sk_buff *skb;
1625 grant_ref_t ref;
1626 struct xen_netif_rx_request *req;
1627 unsigned int feature_rx_copy;
1628
1629 err = xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1630 "feature-rx-copy", "%u", &feature_rx_copy);
1631 if (err != 1)
1632 feature_rx_copy = 0;
1633
1634 if (!feature_rx_copy) {
1635 dev_info(&dev->dev,
1636 "backend does not support copying receive path\n");
1637 return -ENODEV;
1638 }
1639
1640 err = talk_to_netback(np->xbdev, np);
1641 if (err)
1642 return err;
1643
1644 rtnl_lock();
1645 netdev_update_features(dev);
1646 rtnl_unlock();
1647
1648 spin_lock_bh(&np->rx_lock);
1649 spin_lock_irq(&np->tx_lock);
1650
1651 /* Step 1: Discard all pending TX packet fragments. */
1652 xennet_release_tx_bufs(np);
1653
1654 /* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */
1655 for (requeue_idx = 0, i = 0; i < NET_RX_RING_SIZE; i++) {
1656 skb_frag_t *frag;
1657 const struct page *page;
1658 if (!np->rx_skbs[i])
1659 continue;
1660
1661 skb = np->rx_skbs[requeue_idx] = xennet_get_rx_skb(np, i);
1662 ref = np->grant_rx_ref[requeue_idx] = xennet_get_rx_ref(np, i);
1663 req = RING_GET_REQUEST(&np->rx, requeue_idx);
1664
1665 frag = &skb_shinfo(skb)->frags[0];
1666 page = skb_frag_page(frag);
1667 gnttab_grant_foreign_access_ref(
1668 ref, np->xbdev->otherend_id,
1669 pfn_to_mfn(page_to_pfn(page)),
1670 0);
1671 req->gref = ref;
1672 req->id = requeue_idx;
1673
1674 requeue_idx++;
1675 }
1676
1677 np->rx.req_prod_pvt = requeue_idx;
1678
1679 /*
1680 * Step 3: All public and private state should now be sane. Get
1681 * ready to start sending and receiving packets and give the driver
1682 * domain a kick because we've probably just requeued some
1683 * packets.
1684 */
1685 netif_carrier_on(np->netdev);
1686 notify_remote_via_irq(np->netdev->irq);
1687 xennet_tx_buf_gc(dev);
1688 xennet_alloc_rx_buffers(dev);
1689
1690 spin_unlock_irq(&np->tx_lock);
1691 spin_unlock_bh(&np->rx_lock);
1692
1693 return 0;
1694 }
1695
1696 /**
1697 * Callback received when the backend's state changes.
1698 */
1699 static void netback_changed(struct xenbus_device *dev,
1700 enum xenbus_state backend_state)
1701 {
1702 struct netfront_info *np = dev_get_drvdata(&dev->dev);
1703 struct net_device *netdev = np->netdev;
1704
1705 dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
1706
1707 switch (backend_state) {
1708 case XenbusStateInitialising:
1709 case XenbusStateInitialised:
1710 case XenbusStateReconfiguring:
1711 case XenbusStateReconfigured:
1712 case XenbusStateUnknown:
1713 case XenbusStateClosed:
1714 break;
1715
1716 case XenbusStateInitWait:
1717 if (dev->state != XenbusStateInitialising)
1718 break;
1719 if (xennet_connect(netdev) != 0)
1720 break;
1721 xenbus_switch_state(dev, XenbusStateConnected);
1722 break;
1723
1724 case XenbusStateConnected:
1725 netif_notify_peers(netdev);
1726 break;
1727
1728 case XenbusStateClosing:
1729 xenbus_frontend_closed(dev);
1730 break;
1731 }
1732 }
1733
1734 static const struct xennet_stat {
1735 char name[ETH_GSTRING_LEN];
1736 u16 offset;
1737 } xennet_stats[] = {
1738 {
1739 "rx_gso_checksum_fixup",
1740 offsetof(struct netfront_info, rx_gso_checksum_fixup)
1741 },
1742 };
1743
1744 static int xennet_get_sset_count(struct net_device *dev, int string_set)
1745 {
1746 switch (string_set) {
1747 case ETH_SS_STATS:
1748 return ARRAY_SIZE(xennet_stats);
1749 default:
1750 return -EINVAL;
1751 }
1752 }
1753
1754 static void xennet_get_ethtool_stats(struct net_device *dev,
1755 struct ethtool_stats *stats, u64 * data)
1756 {
1757 void *np = netdev_priv(dev);
1758 int i;
1759
1760 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
1761 data[i] = *(unsigned long *)(np + xennet_stats[i].offset);
1762 }
1763
1764 static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
1765 {
1766 int i;
1767
1768 switch (stringset) {
1769 case ETH_SS_STATS:
1770 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
1771 memcpy(data + i * ETH_GSTRING_LEN,
1772 xennet_stats[i].name, ETH_GSTRING_LEN);
1773 break;
1774 }
1775 }
1776
1777 static const struct ethtool_ops xennet_ethtool_ops =
1778 {
1779 .get_link = ethtool_op_get_link,
1780
1781 .get_sset_count = xennet_get_sset_count,
1782 .get_ethtool_stats = xennet_get_ethtool_stats,
1783 .get_strings = xennet_get_strings,
1784 };
1785
1786 #ifdef CONFIG_SYSFS
1787 static ssize_t show_rxbuf_min(struct device *dev,
1788 struct device_attribute *attr, char *buf)
1789 {
1790 struct net_device *netdev = to_net_dev(dev);
1791 struct netfront_info *info = netdev_priv(netdev);
1792
1793 return sprintf(buf, "%u\n", info->rx_min_target);
1794 }
1795
1796 static ssize_t store_rxbuf_min(struct device *dev,
1797 struct device_attribute *attr,
1798 const char *buf, size_t len)
1799 {
1800 struct net_device *netdev = to_net_dev(dev);
1801 struct netfront_info *np = netdev_priv(netdev);
1802 char *endp;
1803 unsigned long target;
1804
1805 if (!capable(CAP_NET_ADMIN))
1806 return -EPERM;
1807
1808 target = simple_strtoul(buf, &endp, 0);
1809 if (endp == buf)
1810 return -EBADMSG;
1811
1812 if (target < RX_MIN_TARGET)
1813 target = RX_MIN_TARGET;
1814 if (target > RX_MAX_TARGET)
1815 target = RX_MAX_TARGET;
1816
1817 spin_lock_bh(&np->rx_lock);
1818 if (target > np->rx_max_target)
1819 np->rx_max_target = target;
1820 np->rx_min_target = target;
1821 if (target > np->rx_target)
1822 np->rx_target = target;
1823
1824 xennet_alloc_rx_buffers(netdev);
1825
1826 spin_unlock_bh(&np->rx_lock);
1827 return len;
1828 }
1829
1830 static ssize_t show_rxbuf_max(struct device *dev,
1831 struct device_attribute *attr, char *buf)
1832 {
1833 struct net_device *netdev = to_net_dev(dev);
1834 struct netfront_info *info = netdev_priv(netdev);
1835
1836 return sprintf(buf, "%u\n", info->rx_max_target);
1837 }
1838
1839 static ssize_t store_rxbuf_max(struct device *dev,
1840 struct device_attribute *attr,
1841 const char *buf, size_t len)
1842 {
1843 struct net_device *netdev = to_net_dev(dev);
1844 struct netfront_info *np = netdev_priv(netdev);
1845 char *endp;
1846 unsigned long target;
1847
1848 if (!capable(CAP_NET_ADMIN))
1849 return -EPERM;
1850
1851 target = simple_strtoul(buf, &endp, 0);
1852 if (endp == buf)
1853 return -EBADMSG;
1854
1855 if (target < RX_MIN_TARGET)
1856 target = RX_MIN_TARGET;
1857 if (target > RX_MAX_TARGET)
1858 target = RX_MAX_TARGET;
1859
1860 spin_lock_bh(&np->rx_lock);
1861 if (target < np->rx_min_target)
1862 np->rx_min_target = target;
1863 np->rx_max_target = target;
1864 if (target < np->rx_target)
1865 np->rx_target = target;
1866
1867 xennet_alloc_rx_buffers(netdev);
1868
1869 spin_unlock_bh(&np->rx_lock);
1870 return len;
1871 }
1872
1873 static ssize_t show_rxbuf_cur(struct device *dev,
1874 struct device_attribute *attr, char *buf)
1875 {
1876 struct net_device *netdev = to_net_dev(dev);
1877 struct netfront_info *info = netdev_priv(netdev);
1878
1879 return sprintf(buf, "%u\n", info->rx_target);
1880 }
1881
1882 static struct device_attribute xennet_attrs[] = {
1883 __ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf_min, store_rxbuf_min),
1884 __ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf_max, store_rxbuf_max),
1885 __ATTR(rxbuf_cur, S_IRUGO, show_rxbuf_cur, NULL),
1886 };
1887
1888 static int xennet_sysfs_addif(struct net_device *netdev)
1889 {
1890 int i;
1891 int err;
1892
1893 for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++) {
1894 err = device_create_file(&netdev->dev,
1895 &xennet_attrs[i]);
1896 if (err)
1897 goto fail;
1898 }
1899 return 0;
1900
1901 fail:
1902 while (--i >= 0)
1903 device_remove_file(&netdev->dev, &xennet_attrs[i]);
1904 return err;
1905 }
1906
1907 static void xennet_sysfs_delif(struct net_device *netdev)
1908 {
1909 int i;
1910
1911 for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++)
1912 device_remove_file(&netdev->dev, &xennet_attrs[i]);
1913 }
1914
1915 #endif /* CONFIG_SYSFS */
1916
1917 static const struct xenbus_device_id netfront_ids[] = {
1918 { "vif" },
1919 { "" }
1920 };
1921
1922
1923 static int __devexit xennet_remove(struct xenbus_device *dev)
1924 {
1925 struct netfront_info *info = dev_get_drvdata(&dev->dev);
1926
1927 dev_dbg(&dev->dev, "%s\n", dev->nodename);
1928
1929 unregister_netdev(info->netdev);
1930
1931 xennet_disconnect_backend(info);
1932
1933 del_timer_sync(&info->rx_refill_timer);
1934
1935 xennet_sysfs_delif(info->netdev);
1936
1937 free_percpu(info->stats);
1938
1939 free_netdev(info->netdev);
1940
1941 return 0;
1942 }
1943
1944 static DEFINE_XENBUS_DRIVER(netfront, ,
1945 .probe = netfront_probe,
1946 .remove = __devexit_p(xennet_remove),
1947 .resume = netfront_resume,
1948 .otherend_changed = netback_changed,
1949 );
1950
1951 static int __init netif_init(void)
1952 {
1953 if (!xen_domain())
1954 return -ENODEV;
1955
1956 if (xen_initial_domain())
1957 return 0;
1958
1959 printk(KERN_INFO "Initialising Xen virtual ethernet driver.\n");
1960
1961 return xenbus_register_frontend(&netfront_driver);
1962 }
1963 module_init(netif_init);
1964
1965
1966 static void __exit netif_exit(void)
1967 {
1968 if (xen_initial_domain())
1969 return;
1970
1971 xenbus_unregister_driver(&netfront_driver);
1972 }
1973 module_exit(netif_exit);
1974
1975 MODULE_DESCRIPTION("Xen virtual network device frontend");
1976 MODULE_LICENSE("GPL");
1977 MODULE_ALIAS("xen:vif");
1978 MODULE_ALIAS("xennet");