?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
ath6kl: Normalize use of FW_DIR
[linux/fpc-iii.git] / drivers / net / xen-netfront.c
blob663b32c2e93185f60b391b2706cbc847e4c8ffd2
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/platform_pci.h>
51 #include <xen/grant_table.h>
52
53 #include <xen/interface/io/netif.h>
54 #include <xen/interface/memory.h>
55 #include <xen/interface/grant_table.h>
56
57 static const struct ethtool_ops xennet_ethtool_ops;
58
59 struct netfront_cb {
60 struct page *page;
61 unsigned offset;
62 };
63
64 #define NETFRONT_SKB_CB(skb) ((struct netfront_cb *)((skb)->cb))
65
66 #define RX_COPY_THRESHOLD 256
67
68 #define GRANT_INVALID_REF 0
69
70 #define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, PAGE_SIZE)
71 #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, PAGE_SIZE)
72 #define TX_MAX_TARGET min_t(int, NET_TX_RING_SIZE, 256)
73
74 struct netfront_stats {
75 u64 rx_packets;
76 u64 tx_packets;
77 u64 rx_bytes;
78 u64 tx_bytes;
79 struct u64_stats_sync syncp;
80 };
81
82 struct netfront_info {
83 struct list_head list;
84 struct net_device *netdev;
85
86 struct napi_struct napi;
87
88 unsigned int evtchn;
89 struct xenbus_device *xbdev;
90
91 spinlock_t tx_lock;
92 struct xen_netif_tx_front_ring tx;
93 int tx_ring_ref;
94
95 /*
96 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
97 * are linked from tx_skb_freelist through skb_entry.link.
98 *
99 * NB. Freelist index entries are always going to be less than
100 * PAGE_OFFSET, whereas pointers to skbs will always be equal or
101 * greater than PAGE_OFFSET: we use this property to distinguish
102 * them.
103 */
104 union skb_entry {
105 struct sk_buff *skb;
106 unsigned long link;
107 } tx_skbs[NET_TX_RING_SIZE];
108 grant_ref_t gref_tx_head;
109 grant_ref_t grant_tx_ref[NET_TX_RING_SIZE];
110 unsigned tx_skb_freelist;
111
112 spinlock_t rx_lock ____cacheline_aligned_in_smp;
113 struct xen_netif_rx_front_ring rx;
114 int rx_ring_ref;
115
116 /* Receive-ring batched refills. */
117 #define RX_MIN_TARGET 8
118 #define RX_DFL_MIN_TARGET 64
119 #define RX_MAX_TARGET min_t(int, NET_RX_RING_SIZE, 256)
120 unsigned rx_min_target, rx_max_target, rx_target;
121 struct sk_buff_head rx_batch;
122
123 struct timer_list rx_refill_timer;
124
125 struct sk_buff *rx_skbs[NET_RX_RING_SIZE];
126 grant_ref_t gref_rx_head;
127 grant_ref_t grant_rx_ref[NET_RX_RING_SIZE];
128
129 unsigned long rx_pfn_array[NET_RX_RING_SIZE];
130 struct multicall_entry rx_mcl[NET_RX_RING_SIZE+1];
131 struct mmu_update rx_mmu[NET_RX_RING_SIZE];
132
133 /* Statistics */
134 struct netfront_stats __percpu *stats;
135
136 unsigned long rx_gso_checksum_fixup;
137 };
138
139 struct netfront_rx_info {
140 struct xen_netif_rx_response rx;
141 struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
142 };
143
144 static void skb_entry_set_link(union skb_entry *list, unsigned short id)
145 {
146 list->link = id;
147 }
148
149 static int skb_entry_is_link(const union skb_entry *list)
150 {
151 BUILD_BUG_ON(sizeof(list->skb) != sizeof(list->link));
152 return (unsigned long)list->skb < PAGE_OFFSET;
153 }
154
155 /*
156 * Access macros for acquiring freeing slots in tx_skbs[].
157 */
158
159 static void add_id_to_freelist(unsigned *head, union skb_entry *list,
160 unsigned short id)
161 {
162 skb_entry_set_link(&list[id], *head);
163 *head = id;
164 }
165
166 static unsigned short get_id_from_freelist(unsigned *head,
167 union skb_entry *list)
168 {
169 unsigned int id = *head;
170 *head = list[id].link;
171 return id;
172 }
173
174 static int xennet_rxidx(RING_IDX idx)
175 {
176 return idx & (NET_RX_RING_SIZE - 1);
177 }
178
179 static struct sk_buff *xennet_get_rx_skb(struct netfront_info *np,
180 RING_IDX ri)
181 {
182 int i = xennet_rxidx(ri);
183 struct sk_buff *skb = np->rx_skbs[i];
184 np->rx_skbs[i] = NULL;
185 return skb;
186 }
187
188 static grant_ref_t xennet_get_rx_ref(struct netfront_info *np,
189 RING_IDX ri)
190 {
191 int i = xennet_rxidx(ri);
192 grant_ref_t ref = np->grant_rx_ref[i];
193 np->grant_rx_ref[i] = GRANT_INVALID_REF;
194 return ref;
195 }
196
197 #ifdef CONFIG_SYSFS
198 static int xennet_sysfs_addif(struct net_device *netdev);
199 static void xennet_sysfs_delif(struct net_device *netdev);
200 #else /* !CONFIG_SYSFS */
201 #define xennet_sysfs_addif(dev) (0)
202 #define xennet_sysfs_delif(dev) do { } while (0)
203 #endif
204
205 static bool xennet_can_sg(struct net_device *dev)
206 {
207 return dev->features & NETIF_F_SG;
208 }
209
210
211 static void rx_refill_timeout(unsigned long data)
212 {
213 struct net_device *dev = (struct net_device *)data;
214 struct netfront_info *np = netdev_priv(dev);
215 napi_schedule(&np->napi);
216 }
217
218 static int netfront_tx_slot_available(struct netfront_info *np)
219 {
220 return (np->tx.req_prod_pvt - np->tx.rsp_cons) <
221 (TX_MAX_TARGET - MAX_SKB_FRAGS - 2);
222 }
223
224 static void xennet_maybe_wake_tx(struct net_device *dev)
225 {
226 struct netfront_info *np = netdev_priv(dev);
227
228 if (unlikely(netif_queue_stopped(dev)) &&
229 netfront_tx_slot_available(np) &&
230 likely(netif_running(dev)))
231 netif_wake_queue(dev);
232 }
233
234 static void xennet_alloc_rx_buffers(struct net_device *dev)
235 {
236 unsigned short id;
237 struct netfront_info *np = netdev_priv(dev);
238 struct sk_buff *skb;
239 struct page *page;
240 int i, batch_target, notify;
241 RING_IDX req_prod = np->rx.req_prod_pvt;
242 grant_ref_t ref;
243 unsigned long pfn;
244 void *vaddr;
245 struct xen_netif_rx_request *req;
246
247 if (unlikely(!netif_carrier_ok(dev)))
248 return;
249
250 /*
251 * Allocate skbuffs greedily, even though we batch updates to the
252 * receive ring. This creates a less bursty demand on the memory
253 * allocator, so should reduce the chance of failed allocation requests
254 * both for ourself and for other kernel subsystems.
255 */
256 batch_target = np->rx_target - (req_prod - np->rx.rsp_cons);
257 for (i = skb_queue_len(&np->rx_batch); i < batch_target; i++) {
258 skb = __netdev_alloc_skb(dev, RX_COPY_THRESHOLD + NET_IP_ALIGN,
259 GFP_ATOMIC | __GFP_NOWARN);
260 if (unlikely(!skb))
261 goto no_skb;
262
263 /* Align ip header to a 16 bytes boundary */
264 skb_reserve(skb, NET_IP_ALIGN);
265
266 page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
267 if (!page) {
268 kfree_skb(skb);
269 no_skb:
270 /* Any skbuffs queued for refill? Force them out. */
271 if (i != 0)
272 goto refill;
273 /* Could not allocate any skbuffs. Try again later. */
274 mod_timer(&np->rx_refill_timer,
275 jiffies + (HZ/10));
276 break;
277 }
278
279 __skb_fill_page_desc(skb, 0, page, 0, 0);
280 skb_shinfo(skb)->nr_frags = 1;
281 __skb_queue_tail(&np->rx_batch, skb);
282 }
283
284 /* Is the batch large enough to be worthwhile? */
285 if (i < (np->rx_target/2)) {
286 if (req_prod > np->rx.sring->req_prod)
287 goto push;
288 return;
289 }
290
291 /* Adjust our fill target if we risked running out of buffers. */
292 if (((req_prod - np->rx.sring->rsp_prod) < (np->rx_target / 4)) &&
293 ((np->rx_target *= 2) > np->rx_max_target))
294 np->rx_target = np->rx_max_target;
295
296 refill:
297 for (i = 0; ; i++) {
298 skb = __skb_dequeue(&np->rx_batch);
299 if (skb == NULL)
300 break;
301
302 skb->dev = dev;
303
304 id = xennet_rxidx(req_prod + i);
305
306 BUG_ON(np->rx_skbs[id]);
307 np->rx_skbs[id] = skb;
308
309 ref = gnttab_claim_grant_reference(&np->gref_rx_head);
310 BUG_ON((signed short)ref < 0);
311 np->grant_rx_ref[id] = ref;
312
313 pfn = page_to_pfn(skb_frag_page(&skb_shinfo(skb)->frags[0]));
314 vaddr = page_address(skb_frag_page(&skb_shinfo(skb)->frags[0]));
315
316 req = RING_GET_REQUEST(&np->rx, req_prod + i);
317 gnttab_grant_foreign_access_ref(ref,
318 np->xbdev->otherend_id,
319 pfn_to_mfn(pfn),
320 0);
321
322 req->id = id;
323 req->gref = ref;
324 }
325
326 wmb(); /* barrier so backend seens requests */
327
328 /* Above is a suitable barrier to ensure backend will see requests. */
329 np->rx.req_prod_pvt = req_prod + i;
330 push:
331 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->rx, notify);
332 if (notify)
333 notify_remote_via_irq(np->netdev->irq);
334 }
335
336 static int xennet_open(struct net_device *dev)
337 {
338 struct netfront_info *np = netdev_priv(dev);
339
340 napi_enable(&np->napi);
341
342 spin_lock_bh(&np->rx_lock);
343 if (netif_carrier_ok(dev)) {
344 xennet_alloc_rx_buffers(dev);
345 np->rx.sring->rsp_event = np->rx.rsp_cons + 1;
346 if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx))
347 napi_schedule(&np->napi);
348 }
349 spin_unlock_bh(&np->rx_lock);
350
351 netif_start_queue(dev);
352
353 return 0;
354 }
355
356 static void xennet_tx_buf_gc(struct net_device *dev)
357 {
358 RING_IDX cons, prod;
359 unsigned short id;
360 struct netfront_info *np = netdev_priv(dev);
361 struct sk_buff *skb;
362
363 BUG_ON(!netif_carrier_ok(dev));
364
365 do {
366 prod = np->tx.sring->rsp_prod;
367 rmb(); /* Ensure we see responses up to 'rp'. */
368
369 for (cons = np->tx.rsp_cons; cons != prod; cons++) {
370 struct xen_netif_tx_response *txrsp;
371
372 txrsp = RING_GET_RESPONSE(&np->tx, cons);
373 if (txrsp->status == XEN_NETIF_RSP_NULL)
374 continue;
375
376 id = txrsp->id;
377 skb = np->tx_skbs[id].skb;
378 if (unlikely(gnttab_query_foreign_access(
379 np->grant_tx_ref[id]) != 0)) {
380 printk(KERN_ALERT "xennet_tx_buf_gc: warning "
381 "-- grant still in use by backend "
382 "domain.\n");
383 BUG();
384 }
385 gnttab_end_foreign_access_ref(
386 np->grant_tx_ref[id], GNTMAP_readonly);
387 gnttab_release_grant_reference(
388 &np->gref_tx_head, np->grant_tx_ref[id]);
389 np->grant_tx_ref[id] = GRANT_INVALID_REF;
390 add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, id);
391 dev_kfree_skb_irq(skb);
392 }
393
394 np->tx.rsp_cons = prod;
395
396 /*
397 * Set a new event, then check for race with update of tx_cons.
398 * Note that it is essential to schedule a callback, no matter
399 * how few buffers are pending. Even if there is space in the
400 * transmit ring, higher layers may be blocked because too much
401 * data is outstanding: in such cases notification from Xen is
402 * likely to be the only kick that we'll get.
403 */
404 np->tx.sring->rsp_event =
405 prod + ((np->tx.sring->req_prod - prod) >> 1) + 1;
406 mb(); /* update shared area */
407 } while ((cons == prod) && (prod != np->tx.sring->rsp_prod));
408
409 xennet_maybe_wake_tx(dev);
410 }
411
412 static void xennet_make_frags(struct sk_buff *skb, struct net_device *dev,
413 struct xen_netif_tx_request *tx)
414 {
415 struct netfront_info *np = netdev_priv(dev);
416 char *data = skb->data;
417 unsigned long mfn;
418 RING_IDX prod = np->tx.req_prod_pvt;
419 int frags = skb_shinfo(skb)->nr_frags;
420 unsigned int offset = offset_in_page(data);
421 unsigned int len = skb_headlen(skb);
422 unsigned int id;
423 grant_ref_t ref;
424 int i;
425
426 /* While the header overlaps a page boundary (including being
427 larger than a page), split it it into page-sized chunks. */
428 while (len > PAGE_SIZE - offset) {
429 tx->size = PAGE_SIZE - offset;
430 tx->flags |= XEN_NETTXF_more_data;
431 len -= tx->size;
432 data += tx->size;
433 offset = 0;
434
435 id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
436 np->tx_skbs[id].skb = skb_get(skb);
437 tx = RING_GET_REQUEST(&np->tx, prod++);
438 tx->id = id;
439 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
440 BUG_ON((signed short)ref < 0);
441
442 mfn = virt_to_mfn(data);
443 gnttab_grant_foreign_access_ref(ref, np->xbdev->otherend_id,
444 mfn, GNTMAP_readonly);
445
446 tx->gref = np->grant_tx_ref[id] = ref;
447 tx->offset = offset;
448 tx->size = len;
449 tx->flags = 0;
450 }
451
452 /* Grant backend access to each skb fragment page. */
453 for (i = 0; i < frags; i++) {
454 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
455
456 tx->flags |= XEN_NETTXF_more_data;
457
458 id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
459 np->tx_skbs[id].skb = skb_get(skb);
460 tx = RING_GET_REQUEST(&np->tx, prod++);
461 tx->id = id;
462 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
463 BUG_ON((signed short)ref < 0);
464
465 mfn = pfn_to_mfn(page_to_pfn(skb_frag_page(frag)));
466 gnttab_grant_foreign_access_ref(ref, np->xbdev->otherend_id,
467 mfn, GNTMAP_readonly);
468
469 tx->gref = np->grant_tx_ref[id] = ref;
470 tx->offset = frag->page_offset;
471 tx->size = skb_frag_size(frag);
472 tx->flags = 0;
473 }
474
475 np->tx.req_prod_pvt = prod;
476 }
477
478 static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
479 {
480 unsigned short id;
481 struct netfront_info *np = netdev_priv(dev);
482 struct netfront_stats *stats = this_cpu_ptr(np->stats);
483 struct xen_netif_tx_request *tx;
484 struct xen_netif_extra_info *extra;
485 char *data = skb->data;
486 RING_IDX i;
487 grant_ref_t ref;
488 unsigned long mfn;
489 int notify;
490 int frags = skb_shinfo(skb)->nr_frags;
491 unsigned int offset = offset_in_page(data);
492 unsigned int len = skb_headlen(skb);
493 unsigned long flags;
494
495 frags += DIV_ROUND_UP(offset + len, PAGE_SIZE);
496 if (unlikely(frags > MAX_SKB_FRAGS + 1)) {
497 printk(KERN_ALERT "xennet: skb rides the rocket: %d frags\n",
498 frags);
499 dump_stack();
500 goto drop;
501 }
502
503 spin_lock_irqsave(&np->tx_lock, flags);
504
505 if (unlikely(!netif_carrier_ok(dev) ||
506 (frags > 1 && !xennet_can_sg(dev)) ||
507 netif_needs_gso(skb, netif_skb_features(skb)))) {
508 spin_unlock_irqrestore(&np->tx_lock, flags);
509 goto drop;
510 }
511
512 i = np->tx.req_prod_pvt;
513
514 id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
515 np->tx_skbs[id].skb = skb;
516
517 tx = RING_GET_REQUEST(&np->tx, i);
518
519 tx->id = id;
520 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
521 BUG_ON((signed short)ref < 0);
522 mfn = virt_to_mfn(data);
523 gnttab_grant_foreign_access_ref(
524 ref, np->xbdev->otherend_id, mfn, GNTMAP_readonly);
525 tx->gref = np->grant_tx_ref[id] = ref;
526 tx->offset = offset;
527 tx->size = len;
528 extra = NULL;
529
530 tx->flags = 0;
531 if (skb->ip_summed == CHECKSUM_PARTIAL)
532 /* local packet? */
533 tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated;
534 else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
535 /* remote but checksummed. */
536 tx->flags |= XEN_NETTXF_data_validated;
537
538 if (skb_shinfo(skb)->gso_size) {
539 struct xen_netif_extra_info *gso;
540
541 gso = (struct xen_netif_extra_info *)
542 RING_GET_REQUEST(&np->tx, ++i);
543
544 if (extra)
545 extra->flags |= XEN_NETIF_EXTRA_FLAG_MORE;
546 else
547 tx->flags |= XEN_NETTXF_extra_info;
548
549 gso->u.gso.size = skb_shinfo(skb)->gso_size;
550 gso->u.gso.type = XEN_NETIF_GSO_TYPE_TCPV4;
551 gso->u.gso.pad = 0;
552 gso->u.gso.features = 0;
553
554 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
555 gso->flags = 0;
556 extra = gso;
557 }
558
559 np->tx.req_prod_pvt = i + 1;
560
561 xennet_make_frags(skb, dev, tx);
562 tx->size = skb->len;
563
564 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->tx, notify);
565 if (notify)
566 notify_remote_via_irq(np->netdev->irq);
567
568 u64_stats_update_begin(&stats->syncp);
569 stats->tx_bytes += skb->len;
570 stats->tx_packets++;
571 u64_stats_update_end(&stats->syncp);
572
573 /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
574 xennet_tx_buf_gc(dev);
575
576 if (!netfront_tx_slot_available(np))
577 netif_stop_queue(dev);
578
579 spin_unlock_irqrestore(&np->tx_lock, flags);
580
581 return NETDEV_TX_OK;
582
583 drop:
584 dev->stats.tx_dropped++;
585 dev_kfree_skb(skb);
586 return NETDEV_TX_OK;
587 }
588
589 static int xennet_close(struct net_device *dev)
590 {
591 struct netfront_info *np = netdev_priv(dev);
592 netif_stop_queue(np->netdev);
593 napi_disable(&np->napi);
594 return 0;
595 }
596
597 static void xennet_move_rx_slot(struct netfront_info *np, struct sk_buff *skb,
598 grant_ref_t ref)
599 {
600 int new = xennet_rxidx(np->rx.req_prod_pvt);
601
602 BUG_ON(np->rx_skbs[new]);
603 np->rx_skbs[new] = skb;
604 np->grant_rx_ref[new] = ref;
605 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->id = new;
606 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->gref = ref;
607 np->rx.req_prod_pvt++;
608 }
609
610 static int xennet_get_extras(struct netfront_info *np,
611 struct xen_netif_extra_info *extras,
612 RING_IDX rp)
613
614 {
615 struct xen_netif_extra_info *extra;
616 struct device *dev = &np->netdev->dev;
617 RING_IDX cons = np->rx.rsp_cons;
618 int err = 0;
619
620 do {
621 struct sk_buff *skb;
622 grant_ref_t ref;
623
624 if (unlikely(cons + 1 == rp)) {
625 if (net_ratelimit())
626 dev_warn(dev, "Missing extra info\n");
627 err = -EBADR;
628 break;
629 }
630
631 extra = (struct xen_netif_extra_info *)
632 RING_GET_RESPONSE(&np->rx, ++cons);
633
634 if (unlikely(!extra->type ||
635 extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
636 if (net_ratelimit())
637 dev_warn(dev, "Invalid extra type: %d\n",
638 extra->type);
639 err = -EINVAL;
640 } else {
641 memcpy(&extras[extra->type - 1], extra,
642 sizeof(*extra));
643 }
644
645 skb = xennet_get_rx_skb(np, cons);
646 ref = xennet_get_rx_ref(np, cons);
647 xennet_move_rx_slot(np, skb, ref);
648 } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
649
650 np->rx.rsp_cons = cons;
651 return err;
652 }
653
654 static int xennet_get_responses(struct netfront_info *np,
655 struct netfront_rx_info *rinfo, RING_IDX rp,
656 struct sk_buff_head *list)
657 {
658 struct xen_netif_rx_response *rx = &rinfo->rx;
659 struct xen_netif_extra_info *extras = rinfo->extras;
660 struct device *dev = &np->netdev->dev;
661 RING_IDX cons = np->rx.rsp_cons;
662 struct sk_buff *skb = xennet_get_rx_skb(np, cons);
663 grant_ref_t ref = xennet_get_rx_ref(np, cons);
664 int max = MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD);
665 int frags = 1;
666 int err = 0;
667 unsigned long ret;
668
669 if (rx->flags & XEN_NETRXF_extra_info) {
670 err = xennet_get_extras(np, extras, rp);
671 cons = np->rx.rsp_cons;
672 }
673
674 for (;;) {
675 if (unlikely(rx->status < 0 ||
676 rx->offset + rx->status > PAGE_SIZE)) {
677 if (net_ratelimit())
678 dev_warn(dev, "rx->offset: %x, size: %u\n",
679 rx->offset, rx->status);
680 xennet_move_rx_slot(np, skb, ref);
681 err = -EINVAL;
682 goto next;
683 }
684
685 /*
686 * This definitely indicates a bug, either in this driver or in
687 * the backend driver. In future this should flag the bad
688 * situation to the system controller to reboot the backed.
689 */
690 if (ref == GRANT_INVALID_REF) {
691 if (net_ratelimit())
692 dev_warn(dev, "Bad rx response id %d.\n",
693 rx->id);
694 err = -EINVAL;
695 goto next;
696 }
697
698 ret = gnttab_end_foreign_access_ref(ref, 0);
699 BUG_ON(!ret);
700
701 gnttab_release_grant_reference(&np->gref_rx_head, ref);
702
703 __skb_queue_tail(list, skb);
704
705 next:
706 if (!(rx->flags & XEN_NETRXF_more_data))
707 break;
708
709 if (cons + frags == rp) {
710 if (net_ratelimit())
711 dev_warn(dev, "Need more frags\n");
712 err = -ENOENT;
713 break;
714 }
715
716 rx = RING_GET_RESPONSE(&np->rx, cons + frags);
717 skb = xennet_get_rx_skb(np, cons + frags);
718 ref = xennet_get_rx_ref(np, cons + frags);
719 frags++;
720 }
721
722 if (unlikely(frags > max)) {
723 if (net_ratelimit())
724 dev_warn(dev, "Too many frags\n");
725 err = -E2BIG;
726 }
727
728 if (unlikely(err))
729 np->rx.rsp_cons = cons + frags;
730
731 return err;
732 }
733
734 static int xennet_set_skb_gso(struct sk_buff *skb,
735 struct xen_netif_extra_info *gso)
736 {
737 if (!gso->u.gso.size) {
738 if (net_ratelimit())
739 printk(KERN_WARNING "GSO size must not be zero.\n");
740 return -EINVAL;
741 }
742
743 /* Currently only TCPv4 S.O. is supported. */
744 if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4) {
745 if (net_ratelimit())
746 printk(KERN_WARNING "Bad GSO type %d.\n", gso->u.gso.type);
747 return -EINVAL;
748 }
749
750 skb_shinfo(skb)->gso_size = gso->u.gso.size;
751 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
752
753 /* Header must be checked, and gso_segs computed. */
754 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
755 skb_shinfo(skb)->gso_segs = 0;
756
757 return 0;
758 }
759
760 static RING_IDX xennet_fill_frags(struct netfront_info *np,
761 struct sk_buff *skb,
762 struct sk_buff_head *list)
763 {
764 struct skb_shared_info *shinfo = skb_shinfo(skb);
765 int nr_frags = shinfo->nr_frags;
766 RING_IDX cons = np->rx.rsp_cons;
767 struct sk_buff *nskb;
768
769 while ((nskb = __skb_dequeue(list))) {
770 struct xen_netif_rx_response *rx =
771 RING_GET_RESPONSE(&np->rx, ++cons);
772 skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0];
773
774 __skb_fill_page_desc(skb, nr_frags,
775 skb_frag_page(nfrag),
776 rx->offset, rx->status);
777
778 skb->data_len += rx->status;
779
780 skb_shinfo(nskb)->nr_frags = 0;
781 kfree_skb(nskb);
782
783 nr_frags++;
784 }
785
786 shinfo->nr_frags = nr_frags;
787 return cons;
788 }
789
790 static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
791 {
792 struct iphdr *iph;
793 unsigned char *th;
794 int err = -EPROTO;
795 int recalculate_partial_csum = 0;
796
797 /*
798 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
799 * peers can fail to set NETRXF_csum_blank when sending a GSO
800 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
801 * recalculate the partial checksum.
802 */
803 if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
804 struct netfront_info *np = netdev_priv(dev);
805 np->rx_gso_checksum_fixup++;
806 skb->ip_summed = CHECKSUM_PARTIAL;
807 recalculate_partial_csum = 1;
808 }
809
810 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
811 if (skb->ip_summed != CHECKSUM_PARTIAL)
812 return 0;
813
814 if (skb->protocol != htons(ETH_P_IP))
815 goto out;
816
817 iph = (void *)skb->data;
818 th = skb->data + 4 * iph->ihl;
819 if (th >= skb_tail_pointer(skb))
820 goto out;
821
822 skb->csum_start = th - skb->head;
823 switch (iph->protocol) {
824 case IPPROTO_TCP:
825 skb->csum_offset = offsetof(struct tcphdr, check);
826
827 if (recalculate_partial_csum) {
828 struct tcphdr *tcph = (struct tcphdr *)th;
829 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
830 skb->len - iph->ihl*4,
831 IPPROTO_TCP, 0);
832 }
833 break;
834 case IPPROTO_UDP:
835 skb->csum_offset = offsetof(struct udphdr, check);
836
837 if (recalculate_partial_csum) {
838 struct udphdr *udph = (struct udphdr *)th;
839 udph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
840 skb->len - iph->ihl*4,
841 IPPROTO_UDP, 0);
842 }
843 break;
844 default:
845 if (net_ratelimit())
846 printk(KERN_ERR "Attempting to checksum a non-"
847 "TCP/UDP packet, dropping a protocol"
848 " %d packet", iph->protocol);
849 goto out;
850 }
851
852 if ((th + skb->csum_offset + 2) > skb_tail_pointer(skb))
853 goto out;
854
855 err = 0;
856
857 out:
858 return err;
859 }
860
861 static int handle_incoming_queue(struct net_device *dev,
862 struct sk_buff_head *rxq)
863 {
864 struct netfront_info *np = netdev_priv(dev);
865 struct netfront_stats *stats = this_cpu_ptr(np->stats);
866 int packets_dropped = 0;
867 struct sk_buff *skb;
868
869 while ((skb = __skb_dequeue(rxq)) != NULL) {
870 struct page *page = NETFRONT_SKB_CB(skb)->page;
871 void *vaddr = page_address(page);
872 unsigned offset = NETFRONT_SKB_CB(skb)->offset;
873
874 memcpy(skb->data, vaddr + offset,
875 skb_headlen(skb));
876
877 if (page != skb_frag_page(&skb_shinfo(skb)->frags[0]))
878 __free_page(page);
879
880 /* Ethernet work: Delayed to here as it peeks the header. */
881 skb->protocol = eth_type_trans(skb, dev);
882
883 if (checksum_setup(dev, skb)) {
884 kfree_skb(skb);
885 packets_dropped++;
886 dev->stats.rx_errors++;
887 continue;
888 }
889
890 u64_stats_update_begin(&stats->syncp);
891 stats->rx_packets++;
892 stats->rx_bytes += skb->len;
893 u64_stats_update_end(&stats->syncp);
894
895 /* Pass it up. */
896 netif_receive_skb(skb);
897 }
898
899 return packets_dropped;
900 }
901
902 static int xennet_poll(struct napi_struct *napi, int budget)
903 {
904 struct netfront_info *np = container_of(napi, struct netfront_info, napi);
905 struct net_device *dev = np->netdev;
906 struct sk_buff *skb;
907 struct netfront_rx_info rinfo;
908 struct xen_netif_rx_response *rx = &rinfo.rx;
909 struct xen_netif_extra_info *extras = rinfo.extras;
910 RING_IDX i, rp;
911 int work_done;
912 struct sk_buff_head rxq;
913 struct sk_buff_head errq;
914 struct sk_buff_head tmpq;
915 unsigned long flags;
916 unsigned int len;
917 int err;
918
919 spin_lock(&np->rx_lock);
920
921 skb_queue_head_init(&rxq);
922 skb_queue_head_init(&errq);
923 skb_queue_head_init(&tmpq);
924
925 rp = np->rx.sring->rsp_prod;
926 rmb(); /* Ensure we see queued responses up to 'rp'. */
927
928 i = np->rx.rsp_cons;
929 work_done = 0;
930 while ((i != rp) && (work_done < budget)) {
931 memcpy(rx, RING_GET_RESPONSE(&np->rx, i), sizeof(*rx));
932 memset(extras, 0, sizeof(rinfo.extras));
933
934 err = xennet_get_responses(np, &rinfo, rp, &tmpq);
935
936 if (unlikely(err)) {
937 err:
938 while ((skb = __skb_dequeue(&tmpq)))
939 __skb_queue_tail(&errq, skb);
940 dev->stats.rx_errors++;
941 i = np->rx.rsp_cons;
942 continue;
943 }
944
945 skb = __skb_dequeue(&tmpq);
946
947 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
948 struct xen_netif_extra_info *gso;
949 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
950
951 if (unlikely(xennet_set_skb_gso(skb, gso))) {
952 __skb_queue_head(&tmpq, skb);
953 np->rx.rsp_cons += skb_queue_len(&tmpq);
954 goto err;
955 }
956 }
957
958 NETFRONT_SKB_CB(skb)->page =
959 skb_frag_page(&skb_shinfo(skb)->frags[0]);
960 NETFRONT_SKB_CB(skb)->offset = rx->offset;
961
962 len = rx->status;
963 if (len > RX_COPY_THRESHOLD)
964 len = RX_COPY_THRESHOLD;
965 skb_put(skb, len);
966
967 if (rx->status > len) {
968 skb_shinfo(skb)->frags[0].page_offset =
969 rx->offset + len;
970 skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status - len);
971 skb->data_len = rx->status - len;
972 } else {
973 __skb_fill_page_desc(skb, 0, NULL, 0, 0);
974 skb_shinfo(skb)->nr_frags = 0;
975 }
976
977 i = xennet_fill_frags(np, skb, &tmpq);
978
979 /*
980 * Truesize approximates the size of true data plus
981 * any supervisor overheads. Adding hypervisor
982 * overheads has been shown to significantly reduce
983 * achievable bandwidth with the default receive
984 * buffer size. It is therefore not wise to account
985 * for it here.
986 *
987 * After alloc_skb(RX_COPY_THRESHOLD), truesize is set
988 * to RX_COPY_THRESHOLD + the supervisor
989 * overheads. Here, we add the size of the data pulled
990 * in xennet_fill_frags().
991 *
992 * We also adjust for any unused space in the main
993 * data area by subtracting (RX_COPY_THRESHOLD -
994 * len). This is especially important with drivers
995 * which split incoming packets into header and data,
996 * using only 66 bytes of the main data area (see the
997 * e1000 driver for example.) On such systems,
998 * without this last adjustement, our achievable
999 * receive throughout using the standard receive
1000 * buffer size was cut by 25%(!!!).
1001 */
1002 skb->truesize += skb->data_len - (RX_COPY_THRESHOLD - len);
1003 skb->len += skb->data_len;
1004
1005 if (rx->flags & XEN_NETRXF_csum_blank)
1006 skb->ip_summed = CHECKSUM_PARTIAL;
1007 else if (rx->flags & XEN_NETRXF_data_validated)
1008 skb->ip_summed = CHECKSUM_UNNECESSARY;
1009
1010 __skb_queue_tail(&rxq, skb);
1011
1012 np->rx.rsp_cons = ++i;
1013 work_done++;
1014 }
1015
1016 __skb_queue_purge(&errq);
1017
1018 work_done -= handle_incoming_queue(dev, &rxq);
1019
1020 /* If we get a callback with very few responses, reduce fill target. */
1021 /* NB. Note exponential increase, linear decrease. */
1022 if (((np->rx.req_prod_pvt - np->rx.sring->rsp_prod) >
1023 ((3*np->rx_target) / 4)) &&
1024 (--np->rx_target < np->rx_min_target))
1025 np->rx_target = np->rx_min_target;
1026
1027 xennet_alloc_rx_buffers(dev);
1028
1029 if (work_done < budget) {
1030 int more_to_do = 0;
1031
1032 local_irq_save(flags);
1033
1034 RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, more_to_do);
1035 if (!more_to_do)
1036 __napi_complete(napi);
1037
1038 local_irq_restore(flags);
1039 }
1040
1041 spin_unlock(&np->rx_lock);
1042
1043 return work_done;
1044 }
1045
1046 static int xennet_change_mtu(struct net_device *dev, int mtu)
1047 {
1048 int max = xennet_can_sg(dev) ? 65535 - ETH_HLEN : ETH_DATA_LEN;
1049
1050 if (mtu > max)
1051 return -EINVAL;
1052 dev->mtu = mtu;
1053 return 0;
1054 }
1055
1056 static struct rtnl_link_stats64 *xennet_get_stats64(struct net_device *dev,
1057 struct rtnl_link_stats64 *tot)
1058 {
1059 struct netfront_info *np = netdev_priv(dev);
1060 int cpu;
1061
1062 for_each_possible_cpu(cpu) {
1063 struct netfront_stats *stats = per_cpu_ptr(np->stats, cpu);
1064 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1065 unsigned int start;
1066
1067 do {
1068 start = u64_stats_fetch_begin_bh(&stats->syncp);
1069
1070 rx_packets = stats->rx_packets;
1071 tx_packets = stats->tx_packets;
1072 rx_bytes = stats->rx_bytes;
1073 tx_bytes = stats->tx_bytes;
1074 } while (u64_stats_fetch_retry_bh(&stats->syncp, start));
1075
1076 tot->rx_packets += rx_packets;
1077 tot->tx_packets += tx_packets;
1078 tot->rx_bytes += rx_bytes;
1079 tot->tx_bytes += tx_bytes;
1080 }
1081
1082 tot->rx_errors = dev->stats.rx_errors;
1083 tot->tx_dropped = dev->stats.tx_dropped;
1084
1085 return tot;
1086 }
1087
1088 static void xennet_release_tx_bufs(struct netfront_info *np)
1089 {
1090 struct sk_buff *skb;
1091 int i;
1092
1093 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1094 /* Skip over entries which are actually freelist references */
1095 if (skb_entry_is_link(&np->tx_skbs[i]))
1096 continue;
1097
1098 skb = np->tx_skbs[i].skb;
1099 gnttab_end_foreign_access_ref(np->grant_tx_ref[i],
1100 GNTMAP_readonly);
1101 gnttab_release_grant_reference(&np->gref_tx_head,
1102 np->grant_tx_ref[i]);
1103 np->grant_tx_ref[i] = GRANT_INVALID_REF;
1104 add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, i);
1105 dev_kfree_skb_irq(skb);
1106 }
1107 }
1108
1109 static void xennet_release_rx_bufs(struct netfront_info *np)
1110 {
1111 struct mmu_update *mmu = np->rx_mmu;
1112 struct multicall_entry *mcl = np->rx_mcl;
1113 struct sk_buff_head free_list;
1114 struct sk_buff *skb;
1115 unsigned long mfn;
1116 int xfer = 0, noxfer = 0, unused = 0;
1117 int id, ref;
1118
1119 dev_warn(&np->netdev->dev, "%s: fix me for copying receiver.\n",
1120 __func__);
1121 return;
1122
1123 skb_queue_head_init(&free_list);
1124
1125 spin_lock_bh(&np->rx_lock);
1126
1127 for (id = 0; id < NET_RX_RING_SIZE; id++) {
1128 ref = np->grant_rx_ref[id];
1129 if (ref == GRANT_INVALID_REF) {
1130 unused++;
1131 continue;
1132 }
1133
1134 skb = np->rx_skbs[id];
1135 mfn = gnttab_end_foreign_transfer_ref(ref);
1136 gnttab_release_grant_reference(&np->gref_rx_head, ref);
1137 np->grant_rx_ref[id] = GRANT_INVALID_REF;
1138
1139 if (0 == mfn) {
1140 skb_shinfo(skb)->nr_frags = 0;
1141 dev_kfree_skb(skb);
1142 noxfer++;
1143 continue;
1144 }
1145
1146 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1147 /* Remap the page. */
1148 const struct page *page =
1149 skb_frag_page(&skb_shinfo(skb)->frags[0]);
1150 unsigned long pfn = page_to_pfn(page);
1151 void *vaddr = page_address(page);
1152
1153 MULTI_update_va_mapping(mcl, (unsigned long)vaddr,
1154 mfn_pte(mfn, PAGE_KERNEL),
1155 0);
1156 mcl++;
1157 mmu->ptr = ((u64)mfn << PAGE_SHIFT)
1158 | MMU_MACHPHYS_UPDATE;
1159 mmu->val = pfn;
1160 mmu++;
1161
1162 set_phys_to_machine(pfn, mfn);
1163 }
1164 __skb_queue_tail(&free_list, skb);
1165 xfer++;
1166 }
1167
1168 dev_info(&np->netdev->dev, "%s: %d xfer, %d noxfer, %d unused\n",
1169 __func__, xfer, noxfer, unused);
1170
1171 if (xfer) {
1172 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1173 /* Do all the remapping work and M2P updates. */
1174 MULTI_mmu_update(mcl, np->rx_mmu, mmu - np->rx_mmu,
1175 NULL, DOMID_SELF);
1176 mcl++;
1177 HYPERVISOR_multicall(np->rx_mcl, mcl - np->rx_mcl);
1178 }
1179 }
1180
1181 __skb_queue_purge(&free_list);
1182
1183 spin_unlock_bh(&np->rx_lock);
1184 }
1185
1186 static void xennet_uninit(struct net_device *dev)
1187 {
1188 struct netfront_info *np = netdev_priv(dev);
1189 xennet_release_tx_bufs(np);
1190 xennet_release_rx_bufs(np);
1191 gnttab_free_grant_references(np->gref_tx_head);
1192 gnttab_free_grant_references(np->gref_rx_head);
1193 }
1194
1195 static netdev_features_t xennet_fix_features(struct net_device *dev,
1196 netdev_features_t features)
1197 {
1198 struct netfront_info *np = netdev_priv(dev);
1199 int val;
1200
1201 if (features & NETIF_F_SG) {
1202 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, "feature-sg",
1203 "%d", &val) < 0)
1204 val = 0;
1205
1206 if (!val)
1207 features &= ~NETIF_F_SG;
1208 }
1209
1210 if (features & NETIF_F_TSO) {
1211 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1212 "feature-gso-tcpv4", "%d", &val) < 0)
1213 val = 0;
1214
1215 if (!val)
1216 features &= ~NETIF_F_TSO;
1217 }
1218
1219 return features;
1220 }
1221
1222 static int xennet_set_features(struct net_device *dev,
1223 netdev_features_t features)
1224 {
1225 if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1226 netdev_info(dev, "Reducing MTU because no SG offload");
1227 dev->mtu = ETH_DATA_LEN;
1228 }
1229
1230 return 0;
1231 }
1232
1233 static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1234 {
1235 struct net_device *dev = dev_id;
1236 struct netfront_info *np = netdev_priv(dev);
1237 unsigned long flags;
1238
1239 spin_lock_irqsave(&np->tx_lock, flags);
1240
1241 if (likely(netif_carrier_ok(dev))) {
1242 xennet_tx_buf_gc(dev);
1243 /* Under tx_lock: protects access to rx shared-ring indexes. */
1244 if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx))
1245 napi_schedule(&np->napi);
1246 }
1247
1248 spin_unlock_irqrestore(&np->tx_lock, flags);
1249
1250 return IRQ_HANDLED;
1251 }
1252
1253 #ifdef CONFIG_NET_POLL_CONTROLLER
1254 static void xennet_poll_controller(struct net_device *dev)
1255 {
1256 xennet_interrupt(0, dev);
1257 }
1258 #endif
1259
1260 static const struct net_device_ops xennet_netdev_ops = {
1261 .ndo_open = xennet_open,
1262 .ndo_uninit = xennet_uninit,
1263 .ndo_stop = xennet_close,
1264 .ndo_start_xmit = xennet_start_xmit,
1265 .ndo_change_mtu = xennet_change_mtu,
1266 .ndo_get_stats64 = xennet_get_stats64,
1267 .ndo_set_mac_address = eth_mac_addr,
1268 .ndo_validate_addr = eth_validate_addr,
1269 .ndo_fix_features = xennet_fix_features,
1270 .ndo_set_features = xennet_set_features,
1271 #ifdef CONFIG_NET_POLL_CONTROLLER
1272 .ndo_poll_controller = xennet_poll_controller,
1273 #endif
1274 };
1275
1276 static struct net_device * __devinit xennet_create_dev(struct xenbus_device *dev)
1277 {
1278 int i, err;
1279 struct net_device *netdev;
1280 struct netfront_info *np;
1281
1282 netdev = alloc_etherdev(sizeof(struct netfront_info));
1283 if (!netdev)
1284 return ERR_PTR(-ENOMEM);
1285
1286 np = netdev_priv(netdev);
1287 np->xbdev = dev;
1288
1289 spin_lock_init(&np->tx_lock);
1290 spin_lock_init(&np->rx_lock);
1291
1292 skb_queue_head_init(&np->rx_batch);
1293 np->rx_target = RX_DFL_MIN_TARGET;
1294 np->rx_min_target = RX_DFL_MIN_TARGET;
1295 np->rx_max_target = RX_MAX_TARGET;
1296
1297 init_timer(&np->rx_refill_timer);
1298 np->rx_refill_timer.data = (unsigned long)netdev;
1299 np->rx_refill_timer.function = rx_refill_timeout;
1300
1301 err = -ENOMEM;
1302 np->stats = alloc_percpu(struct netfront_stats);
1303 if (np->stats == NULL)
1304 goto exit;
1305
1306 /* Initialise tx_skbs as a free chain containing every entry. */
1307 np->tx_skb_freelist = 0;
1308 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1309 skb_entry_set_link(&np->tx_skbs[i], i+1);
1310 np->grant_tx_ref[i] = GRANT_INVALID_REF;
1311 }
1312
1313 /* Clear out rx_skbs */
1314 for (i = 0; i < NET_RX_RING_SIZE; i++) {
1315 np->rx_skbs[i] = NULL;
1316 np->grant_rx_ref[i] = GRANT_INVALID_REF;
1317 }
1318
1319 /* A grant for every tx ring slot */
1320 if (gnttab_alloc_grant_references(TX_MAX_TARGET,
1321 &np->gref_tx_head) < 0) {
1322 printk(KERN_ALERT "#### netfront can't alloc tx grant refs\n");
1323 err = -ENOMEM;
1324 goto exit_free_stats;
1325 }
1326 /* A grant for every rx ring slot */
1327 if (gnttab_alloc_grant_references(RX_MAX_TARGET,
1328 &np->gref_rx_head) < 0) {
1329 printk(KERN_ALERT "#### netfront can't alloc rx grant refs\n");
1330 err = -ENOMEM;
1331 goto exit_free_tx;
1332 }
1333
1334 netdev->netdev_ops = &xennet_netdev_ops;
1335
1336 netif_napi_add(netdev, &np->napi, xennet_poll, 64);
1337 netdev->features = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1338 NETIF_F_GSO_ROBUST;
1339 netdev->hw_features = NETIF_F_IP_CSUM | NETIF_F_SG | NETIF_F_TSO;
1340
1341 /*
1342 * Assume that all hw features are available for now. This set
1343 * will be adjusted by the call to netdev_update_features() in
1344 * xennet_connect() which is the earliest point where we can
1345 * negotiate with the backend regarding supported features.
1346 */
1347 netdev->features |= netdev->hw_features;
1348
1349 SET_ETHTOOL_OPS(netdev, &xennet_ethtool_ops);
1350 SET_NETDEV_DEV(netdev, &dev->dev);
1351
1352 np->netdev = netdev;
1353
1354 netif_carrier_off(netdev);
1355
1356 return netdev;
1357
1358 exit_free_tx:
1359 gnttab_free_grant_references(np->gref_tx_head);
1360 exit_free_stats:
1361 free_percpu(np->stats);
1362 exit:
1363 free_netdev(netdev);
1364 return ERR_PTR(err);
1365 }
1366
1367 /**
1368 * Entry point to this code when a new device is created. Allocate the basic
1369 * structures and the ring buffers for communication with the backend, and
1370 * inform the backend of the appropriate details for those.
1371 */
1372 static int __devinit netfront_probe(struct xenbus_device *dev,
1373 const struct xenbus_device_id *id)
1374 {
1375 int err;
1376 struct net_device *netdev;
1377 struct netfront_info *info;
1378
1379 netdev = xennet_create_dev(dev);
1380 if (IS_ERR(netdev)) {
1381 err = PTR_ERR(netdev);
1382 xenbus_dev_fatal(dev, err, "creating netdev");
1383 return err;
1384 }
1385
1386 info = netdev_priv(netdev);
1387 dev_set_drvdata(&dev->dev, info);
1388
1389 err = register_netdev(info->netdev);
1390 if (err) {
1391 printk(KERN_WARNING "%s: register_netdev err=%d\n",
1392 __func__, err);
1393 goto fail;
1394 }
1395
1396 err = xennet_sysfs_addif(info->netdev);
1397 if (err) {
1398 unregister_netdev(info->netdev);
1399 printk(KERN_WARNING "%s: add sysfs failed err=%d\n",
1400 __func__, err);
1401 goto fail;
1402 }
1403
1404 return 0;
1405
1406 fail:
1407 free_netdev(netdev);
1408 dev_set_drvdata(&dev->dev, NULL);
1409 return err;
1410 }
1411
1412 static void xennet_end_access(int ref, void *page)
1413 {
1414 /* This frees the page as a side-effect */
1415 if (ref != GRANT_INVALID_REF)
1416 gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1417 }
1418
1419 static void xennet_disconnect_backend(struct netfront_info *info)
1420 {
1421 /* Stop old i/f to prevent errors whilst we rebuild the state. */
1422 spin_lock_bh(&info->rx_lock);
1423 spin_lock_irq(&info->tx_lock);
1424 netif_carrier_off(info->netdev);
1425 spin_unlock_irq(&info->tx_lock);
1426 spin_unlock_bh(&info->rx_lock);
1427
1428 if (info->netdev->irq)
1429 unbind_from_irqhandler(info->netdev->irq, info->netdev);
1430 info->evtchn = info->netdev->irq = 0;
1431
1432 /* End access and free the pages */
1433 xennet_end_access(info->tx_ring_ref, info->tx.sring);
1434 xennet_end_access(info->rx_ring_ref, info->rx.sring);
1435
1436 info->tx_ring_ref = GRANT_INVALID_REF;
1437 info->rx_ring_ref = GRANT_INVALID_REF;
1438 info->tx.sring = NULL;
1439 info->rx.sring = NULL;
1440 }
1441
1442 /**
1443 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1444 * driver restart. We tear down our netif structure and recreate it, but
1445 * leave the device-layer structures intact so that this is transparent to the
1446 * rest of the kernel.
1447 */
1448 static int netfront_resume(struct xenbus_device *dev)
1449 {
1450 struct netfront_info *info = dev_get_drvdata(&dev->dev);
1451
1452 dev_dbg(&dev->dev, "%s\n", dev->nodename);
1453
1454 xennet_disconnect_backend(info);
1455 return 0;
1456 }
1457
1458 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1459 {
1460 char *s, *e, *macstr;
1461 int i;
1462
1463 macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1464 if (IS_ERR(macstr))
1465 return PTR_ERR(macstr);
1466
1467 for (i = 0; i < ETH_ALEN; i++) {
1468 mac[i] = simple_strtoul(s, &e, 16);
1469 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1470 kfree(macstr);
1471 return -ENOENT;
1472 }
1473 s = e+1;
1474 }
1475
1476 kfree(macstr);
1477 return 0;
1478 }
1479
1480 static int setup_netfront(struct xenbus_device *dev, struct netfront_info *info)
1481 {
1482 struct xen_netif_tx_sring *txs;
1483 struct xen_netif_rx_sring *rxs;
1484 int err;
1485 struct net_device *netdev = info->netdev;
1486
1487 info->tx_ring_ref = GRANT_INVALID_REF;
1488 info->rx_ring_ref = GRANT_INVALID_REF;
1489 info->rx.sring = NULL;
1490 info->tx.sring = NULL;
1491 netdev->irq = 0;
1492
1493 err = xen_net_read_mac(dev, netdev->dev_addr);
1494 if (err) {
1495 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
1496 goto fail;
1497 }
1498
1499 txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1500 if (!txs) {
1501 err = -ENOMEM;
1502 xenbus_dev_fatal(dev, err, "allocating tx ring page");
1503 goto fail;
1504 }
1505 SHARED_RING_INIT(txs);
1506 FRONT_RING_INIT(&info->tx, txs, PAGE_SIZE);
1507
1508 err = xenbus_grant_ring(dev, virt_to_mfn(txs));
1509 if (err < 0) {
1510 free_page((unsigned long)txs);
1511 goto fail;
1512 }
1513
1514 info->tx_ring_ref = err;
1515 rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1516 if (!rxs) {
1517 err = -ENOMEM;
1518 xenbus_dev_fatal(dev, err, "allocating rx ring page");
1519 goto fail;
1520 }
1521 SHARED_RING_INIT(rxs);
1522 FRONT_RING_INIT(&info->rx, rxs, PAGE_SIZE);
1523
1524 err = xenbus_grant_ring(dev, virt_to_mfn(rxs));
1525 if (err < 0) {
1526 free_page((unsigned long)rxs);
1527 goto fail;
1528 }
1529 info->rx_ring_ref = err;
1530
1531 err = xenbus_alloc_evtchn(dev, &info->evtchn);
1532 if (err)
1533 goto fail;
1534
1535 err = bind_evtchn_to_irqhandler(info->evtchn, xennet_interrupt,
1536 0, netdev->name, netdev);
1537 if (err < 0)
1538 goto fail;
1539 netdev->irq = err;
1540 return 0;
1541
1542 fail:
1543 return err;
1544 }
1545
1546 /* Common code used when first setting up, and when resuming. */
1547 static int talk_to_netback(struct xenbus_device *dev,
1548 struct netfront_info *info)
1549 {
1550 const char *message;
1551 struct xenbus_transaction xbt;
1552 int err;
1553
1554 /* Create shared ring, alloc event channel. */
1555 err = setup_netfront(dev, info);
1556 if (err)
1557 goto out;
1558
1559 again:
1560 err = xenbus_transaction_start(&xbt);
1561 if (err) {
1562 xenbus_dev_fatal(dev, err, "starting transaction");
1563 goto destroy_ring;
1564 }
1565
1566 err = xenbus_printf(xbt, dev->nodename, "tx-ring-ref", "%u",
1567 info->tx_ring_ref);
1568 if (err) {
1569 message = "writing tx ring-ref";
1570 goto abort_transaction;
1571 }
1572 err = xenbus_printf(xbt, dev->nodename, "rx-ring-ref", "%u",
1573 info->rx_ring_ref);
1574 if (err) {
1575 message = "writing rx ring-ref";
1576 goto abort_transaction;
1577 }
1578 err = xenbus_printf(xbt, dev->nodename,
1579 "event-channel", "%u", info->evtchn);
1580 if (err) {
1581 message = "writing event-channel";
1582 goto abort_transaction;
1583 }
1584
1585 err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
1586 1);
1587 if (err) {
1588 message = "writing request-rx-copy";
1589 goto abort_transaction;
1590 }
1591
1592 err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
1593 if (err) {
1594 message = "writing feature-rx-notify";
1595 goto abort_transaction;
1596 }
1597
1598 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
1599 if (err) {
1600 message = "writing feature-sg";
1601 goto abort_transaction;
1602 }
1603
1604 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
1605 if (err) {
1606 message = "writing feature-gso-tcpv4";
1607 goto abort_transaction;
1608 }
1609
1610 err = xenbus_transaction_end(xbt, 0);
1611 if (err) {
1612 if (err == -EAGAIN)
1613 goto again;
1614 xenbus_dev_fatal(dev, err, "completing transaction");
1615 goto destroy_ring;
1616 }
1617
1618 return 0;
1619
1620 abort_transaction:
1621 xenbus_transaction_end(xbt, 1);
1622 xenbus_dev_fatal(dev, err, "%s", message);
1623 destroy_ring:
1624 xennet_disconnect_backend(info);
1625 out:
1626 return err;
1627 }
1628
1629 static int xennet_connect(struct net_device *dev)
1630 {
1631 struct netfront_info *np = netdev_priv(dev);
1632 int i, requeue_idx, err;
1633 struct sk_buff *skb;
1634 grant_ref_t ref;
1635 struct xen_netif_rx_request *req;
1636 unsigned int feature_rx_copy;
1637
1638 err = xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1639 "feature-rx-copy", "%u", &feature_rx_copy);
1640 if (err != 1)
1641 feature_rx_copy = 0;
1642
1643 if (!feature_rx_copy) {
1644 dev_info(&dev->dev,
1645 "backend does not support copying receive path\n");
1646 return -ENODEV;
1647 }
1648
1649 err = talk_to_netback(np->xbdev, np);
1650 if (err)
1651 return err;
1652
1653 rtnl_lock();
1654 netdev_update_features(dev);
1655 rtnl_unlock();
1656
1657 spin_lock_bh(&np->rx_lock);
1658 spin_lock_irq(&np->tx_lock);
1659
1660 /* Step 1: Discard all pending TX packet fragments. */
1661 xennet_release_tx_bufs(np);
1662
1663 /* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */
1664 for (requeue_idx = 0, i = 0; i < NET_RX_RING_SIZE; i++) {
1665 skb_frag_t *frag;
1666 const struct page *page;
1667 if (!np->rx_skbs[i])
1668 continue;
1669
1670 skb = np->rx_skbs[requeue_idx] = xennet_get_rx_skb(np, i);
1671 ref = np->grant_rx_ref[requeue_idx] = xennet_get_rx_ref(np, i);
1672 req = RING_GET_REQUEST(&np->rx, requeue_idx);
1673
1674 frag = &skb_shinfo(skb)->frags[0];
1675 page = skb_frag_page(frag);
1676 gnttab_grant_foreign_access_ref(
1677 ref, np->xbdev->otherend_id,
1678 pfn_to_mfn(page_to_pfn(page)),
1679 0);
1680 req->gref = ref;
1681 req->id = requeue_idx;
1682
1683 requeue_idx++;
1684 }
1685
1686 np->rx.req_prod_pvt = requeue_idx;
1687
1688 /*
1689 * Step 3: All public and private state should now be sane. Get
1690 * ready to start sending and receiving packets and give the driver
1691 * domain a kick because we've probably just requeued some
1692 * packets.
1693 */
1694 netif_carrier_on(np->netdev);
1695 notify_remote_via_irq(np->netdev->irq);
1696 xennet_tx_buf_gc(dev);
1697 xennet_alloc_rx_buffers(dev);
1698
1699 spin_unlock_irq(&np->tx_lock);
1700 spin_unlock_bh(&np->rx_lock);
1701
1702 return 0;
1703 }
1704
1705 /**
1706 * Callback received when the backend's state changes.
1707 */
1708 static void netback_changed(struct xenbus_device *dev,
1709 enum xenbus_state backend_state)
1710 {
1711 struct netfront_info *np = dev_get_drvdata(&dev->dev);
1712 struct net_device *netdev = np->netdev;
1713
1714 dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
1715
1716 switch (backend_state) {
1717 case XenbusStateInitialising:
1718 case XenbusStateInitialised:
1719 case XenbusStateReconfiguring:
1720 case XenbusStateReconfigured:
1721 case XenbusStateUnknown:
1722 case XenbusStateClosed:
1723 break;
1724
1725 case XenbusStateInitWait:
1726 if (dev->state != XenbusStateInitialising)
1727 break;
1728 if (xennet_connect(netdev) != 0)
1729 break;
1730 xenbus_switch_state(dev, XenbusStateConnected);
1731 break;
1732
1733 case XenbusStateConnected:
1734 netif_notify_peers(netdev);
1735 break;
1736
1737 case XenbusStateClosing:
1738 xenbus_frontend_closed(dev);
1739 break;
1740 }
1741 }
1742
1743 static const struct xennet_stat {
1744 char name[ETH_GSTRING_LEN];
1745 u16 offset;
1746 } xennet_stats[] = {
1747 {
1748 "rx_gso_checksum_fixup",
1749 offsetof(struct netfront_info, rx_gso_checksum_fixup)
1750 },
1751 };
1752
1753 static int xennet_get_sset_count(struct net_device *dev, int string_set)
1754 {
1755 switch (string_set) {
1756 case ETH_SS_STATS:
1757 return ARRAY_SIZE(xennet_stats);
1758 default:
1759 return -EINVAL;
1760 }
1761 }
1762
1763 static void xennet_get_ethtool_stats(struct net_device *dev,
1764 struct ethtool_stats *stats, u64 * data)
1765 {
1766 void *np = netdev_priv(dev);
1767 int i;
1768
1769 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
1770 data[i] = *(unsigned long *)(np + xennet_stats[i].offset);
1771 }
1772
1773 static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
1774 {
1775 int i;
1776
1777 switch (stringset) {
1778 case ETH_SS_STATS:
1779 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
1780 memcpy(data + i * ETH_GSTRING_LEN,
1781 xennet_stats[i].name, ETH_GSTRING_LEN);
1782 break;
1783 }
1784 }
1785
1786 static const struct ethtool_ops xennet_ethtool_ops =
1787 {
1788 .get_link = ethtool_op_get_link,
1789
1790 .get_sset_count = xennet_get_sset_count,
1791 .get_ethtool_stats = xennet_get_ethtool_stats,
1792 .get_strings = xennet_get_strings,
1793 };
1794
1795 #ifdef CONFIG_SYSFS
1796 static ssize_t show_rxbuf_min(struct device *dev,
1797 struct device_attribute *attr, char *buf)
1798 {
1799 struct net_device *netdev = to_net_dev(dev);
1800 struct netfront_info *info = netdev_priv(netdev);
1801
1802 return sprintf(buf, "%u\n", info->rx_min_target);
1803 }
1804
1805 static ssize_t store_rxbuf_min(struct device *dev,
1806 struct device_attribute *attr,
1807 const char *buf, size_t len)
1808 {
1809 struct net_device *netdev = to_net_dev(dev);
1810 struct netfront_info *np = netdev_priv(netdev);
1811 char *endp;
1812 unsigned long target;
1813
1814 if (!capable(CAP_NET_ADMIN))
1815 return -EPERM;
1816
1817 target = simple_strtoul(buf, &endp, 0);
1818 if (endp == buf)
1819 return -EBADMSG;
1820
1821 if (target < RX_MIN_TARGET)
1822 target = RX_MIN_TARGET;
1823 if (target > RX_MAX_TARGET)
1824 target = RX_MAX_TARGET;
1825
1826 spin_lock_bh(&np->rx_lock);
1827 if (target > np->rx_max_target)
1828 np->rx_max_target = target;
1829 np->rx_min_target = target;
1830 if (target > np->rx_target)
1831 np->rx_target = target;
1832
1833 xennet_alloc_rx_buffers(netdev);
1834
1835 spin_unlock_bh(&np->rx_lock);
1836 return len;
1837 }
1838
1839 static ssize_t show_rxbuf_max(struct device *dev,
1840 struct device_attribute *attr, char *buf)
1841 {
1842 struct net_device *netdev = to_net_dev(dev);
1843 struct netfront_info *info = netdev_priv(netdev);
1844
1845 return sprintf(buf, "%u\n", info->rx_max_target);
1846 }
1847
1848 static ssize_t store_rxbuf_max(struct device *dev,
1849 struct device_attribute *attr,
1850 const char *buf, size_t len)
1851 {
1852 struct net_device *netdev = to_net_dev(dev);
1853 struct netfront_info *np = netdev_priv(netdev);
1854 char *endp;
1855 unsigned long target;
1856
1857 if (!capable(CAP_NET_ADMIN))
1858 return -EPERM;
1859
1860 target = simple_strtoul(buf, &endp, 0);
1861 if (endp == buf)
1862 return -EBADMSG;
1863
1864 if (target < RX_MIN_TARGET)
1865 target = RX_MIN_TARGET;
1866 if (target > RX_MAX_TARGET)
1867 target = RX_MAX_TARGET;
1868
1869 spin_lock_bh(&np->rx_lock);
1870 if (target < np->rx_min_target)
1871 np->rx_min_target = target;
1872 np->rx_max_target = target;
1873 if (target < np->rx_target)
1874 np->rx_target = target;
1875
1876 xennet_alloc_rx_buffers(netdev);
1877
1878 spin_unlock_bh(&np->rx_lock);
1879 return len;
1880 }
1881
1882 static ssize_t show_rxbuf_cur(struct device *dev,
1883 struct device_attribute *attr, char *buf)
1884 {
1885 struct net_device *netdev = to_net_dev(dev);
1886 struct netfront_info *info = netdev_priv(netdev);
1887
1888 return sprintf(buf, "%u\n", info->rx_target);
1889 }
1890
1891 static struct device_attribute xennet_attrs[] = {
1892 __ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf_min, store_rxbuf_min),
1893 __ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf_max, store_rxbuf_max),
1894 __ATTR(rxbuf_cur, S_IRUGO, show_rxbuf_cur, NULL),
1895 };
1896
1897 static int xennet_sysfs_addif(struct net_device *netdev)
1898 {
1899 int i;
1900 int err;
1901
1902 for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++) {
1903 err = device_create_file(&netdev->dev,
1904 &xennet_attrs[i]);
1905 if (err)
1906 goto fail;
1907 }
1908 return 0;
1909
1910 fail:
1911 while (--i >= 0)
1912 device_remove_file(&netdev->dev, &xennet_attrs[i]);
1913 return err;
1914 }
1915
1916 static void xennet_sysfs_delif(struct net_device *netdev)
1917 {
1918 int i;
1919
1920 for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++)
1921 device_remove_file(&netdev->dev, &xennet_attrs[i]);
1922 }
1923
1924 #endif /* CONFIG_SYSFS */
1925
1926 static const struct xenbus_device_id netfront_ids[] = {
1927 { "vif" },
1928 { "" }
1929 };
1930
1931
1932 static int __devexit xennet_remove(struct xenbus_device *dev)
1933 {
1934 struct netfront_info *info = dev_get_drvdata(&dev->dev);
1935
1936 dev_dbg(&dev->dev, "%s\n", dev->nodename);
1937
1938 unregister_netdev(info->netdev);
1939
1940 xennet_disconnect_backend(info);
1941
1942 del_timer_sync(&info->rx_refill_timer);
1943
1944 xennet_sysfs_delif(info->netdev);
1945
1946 free_percpu(info->stats);
1947
1948 free_netdev(info->netdev);
1949
1950 return 0;
1951 }
1952
1953 static DEFINE_XENBUS_DRIVER(netfront, ,
1954 .probe = netfront_probe,
1955 .remove = __devexit_p(xennet_remove),
1956 .resume = netfront_resume,
1957 .otherend_changed = netback_changed,
1958 );
1959
1960 static int __init netif_init(void)
1961 {
1962 if (!xen_domain())
1963 return -ENODEV;
1964
1965 if (xen_initial_domain())
1966 return 0;
1967
1968 if (!xen_platform_pci_unplug)
1969 return -ENODEV;
1970
1971 printk(KERN_INFO "Initialising Xen virtual ethernet driver.\n");
1972
1973 return xenbus_register_frontend(&netfront_driver);
1974 }
1975 module_init(netif_init);
1976
1977
1978 static void __exit netif_exit(void)
1979 {
1980 if (xen_initial_domain())
1981 return;
1982
1983 xenbus_unregister_driver(&netfront_driver);
1984 }
1985 module_exit(netif_exit);
1986
1987 MODULE_DESCRIPTION("Xen virtual network device frontend");
1988 MODULE_LICENSE("GPL");
1989 MODULE_ALIAS("xen:vif");
1990 MODULE_ALIAS("xennet");
Linux with added FPC-III support