?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 3.4.102
[linux/fpc-iii.git] / drivers / net / xen-netfront.c
blobfe50f14fba4e6f08227cb5cb05d5bf674eba38ef
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 <net/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 /* If skb->len is too big for wire format, drop skb and alert
496 * user about misconfiguration.
497 */
498 if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) {
499 net_alert_ratelimited(
500 "xennet: skb->len = %u, too big for wire format\n",
501 skb->len);
502 goto drop;
503 }
504
505 frags += DIV_ROUND_UP(offset + len, PAGE_SIZE);
506 if (unlikely(frags > MAX_SKB_FRAGS + 1)) {
507 printk(KERN_ALERT "xennet: skb rides the rocket: %d frags\n",
508 frags);
509 dump_stack();
510 goto drop;
511 }
512
513 spin_lock_irqsave(&np->tx_lock, flags);
514
515 if (unlikely(!netif_carrier_ok(dev) ||
516 (frags > 1 && !xennet_can_sg(dev)) ||
517 netif_needs_gso(skb, netif_skb_features(skb)))) {
518 spin_unlock_irqrestore(&np->tx_lock, flags);
519 goto drop;
520 }
521
522 i = np->tx.req_prod_pvt;
523
524 id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
525 np->tx_skbs[id].skb = skb;
526
527 tx = RING_GET_REQUEST(&np->tx, i);
528
529 tx->id = id;
530 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
531 BUG_ON((signed short)ref < 0);
532 mfn = virt_to_mfn(data);
533 gnttab_grant_foreign_access_ref(
534 ref, np->xbdev->otherend_id, mfn, GNTMAP_readonly);
535 tx->gref = np->grant_tx_ref[id] = ref;
536 tx->offset = offset;
537 tx->size = len;
538 extra = NULL;
539
540 tx->flags = 0;
541 if (skb->ip_summed == CHECKSUM_PARTIAL)
542 /* local packet? */
543 tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated;
544 else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
545 /* remote but checksummed. */
546 tx->flags |= XEN_NETTXF_data_validated;
547
548 if (skb_shinfo(skb)->gso_size) {
549 struct xen_netif_extra_info *gso;
550
551 gso = (struct xen_netif_extra_info *)
552 RING_GET_REQUEST(&np->tx, ++i);
553
554 if (extra)
555 extra->flags |= XEN_NETIF_EXTRA_FLAG_MORE;
556 else
557 tx->flags |= XEN_NETTXF_extra_info;
558
559 gso->u.gso.size = skb_shinfo(skb)->gso_size;
560 gso->u.gso.type = XEN_NETIF_GSO_TYPE_TCPV4;
561 gso->u.gso.pad = 0;
562 gso->u.gso.features = 0;
563
564 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
565 gso->flags = 0;
566 extra = gso;
567 }
568
569 np->tx.req_prod_pvt = i + 1;
570
571 xennet_make_frags(skb, dev, tx);
572 tx->size = skb->len;
573
574 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->tx, notify);
575 if (notify)
576 notify_remote_via_irq(np->netdev->irq);
577
578 u64_stats_update_begin(&stats->syncp);
579 stats->tx_bytes += skb->len;
580 stats->tx_packets++;
581 u64_stats_update_end(&stats->syncp);
582
583 /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
584 xennet_tx_buf_gc(dev);
585
586 if (!netfront_tx_slot_available(np))
587 netif_stop_queue(dev);
588
589 spin_unlock_irqrestore(&np->tx_lock, flags);
590
591 return NETDEV_TX_OK;
592
593 drop:
594 dev->stats.tx_dropped++;
595 dev_kfree_skb(skb);
596 return NETDEV_TX_OK;
597 }
598
599 static int xennet_close(struct net_device *dev)
600 {
601 struct netfront_info *np = netdev_priv(dev);
602 netif_stop_queue(np->netdev);
603 napi_disable(&np->napi);
604 return 0;
605 }
606
607 static void xennet_move_rx_slot(struct netfront_info *np, struct sk_buff *skb,
608 grant_ref_t ref)
609 {
610 int new = xennet_rxidx(np->rx.req_prod_pvt);
611
612 BUG_ON(np->rx_skbs[new]);
613 np->rx_skbs[new] = skb;
614 np->grant_rx_ref[new] = ref;
615 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->id = new;
616 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->gref = ref;
617 np->rx.req_prod_pvt++;
618 }
619
620 static int xennet_get_extras(struct netfront_info *np,
621 struct xen_netif_extra_info *extras,
622 RING_IDX rp)
623
624 {
625 struct xen_netif_extra_info *extra;
626 struct device *dev = &np->netdev->dev;
627 RING_IDX cons = np->rx.rsp_cons;
628 int err = 0;
629
630 do {
631 struct sk_buff *skb;
632 grant_ref_t ref;
633
634 if (unlikely(cons + 1 == rp)) {
635 if (net_ratelimit())
636 dev_warn(dev, "Missing extra info\n");
637 err = -EBADR;
638 break;
639 }
640
641 extra = (struct xen_netif_extra_info *)
642 RING_GET_RESPONSE(&np->rx, ++cons);
643
644 if (unlikely(!extra->type ||
645 extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
646 if (net_ratelimit())
647 dev_warn(dev, "Invalid extra type: %d\n",
648 extra->type);
649 err = -EINVAL;
650 } else {
651 memcpy(&extras[extra->type - 1], extra,
652 sizeof(*extra));
653 }
654
655 skb = xennet_get_rx_skb(np, cons);
656 ref = xennet_get_rx_ref(np, cons);
657 xennet_move_rx_slot(np, skb, ref);
658 } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
659
660 np->rx.rsp_cons = cons;
661 return err;
662 }
663
664 static int xennet_get_responses(struct netfront_info *np,
665 struct netfront_rx_info *rinfo, RING_IDX rp,
666 struct sk_buff_head *list)
667 {
668 struct xen_netif_rx_response *rx = &rinfo->rx;
669 struct xen_netif_extra_info *extras = rinfo->extras;
670 struct device *dev = &np->netdev->dev;
671 RING_IDX cons = np->rx.rsp_cons;
672 struct sk_buff *skb = xennet_get_rx_skb(np, cons);
673 grant_ref_t ref = xennet_get_rx_ref(np, cons);
674 int max = MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD);
675 int frags = 1;
676 int err = 0;
677 unsigned long ret;
678
679 if (rx->flags & XEN_NETRXF_extra_info) {
680 err = xennet_get_extras(np, extras, rp);
681 cons = np->rx.rsp_cons;
682 }
683
684 for (;;) {
685 if (unlikely(rx->status < 0 ||
686 rx->offset + rx->status > PAGE_SIZE)) {
687 if (net_ratelimit())
688 dev_warn(dev, "rx->offset: %x, size: %u\n",
689 rx->offset, rx->status);
690 xennet_move_rx_slot(np, skb, ref);
691 err = -EINVAL;
692 goto next;
693 }
694
695 /*
696 * This definitely indicates a bug, either in this driver or in
697 * the backend driver. In future this should flag the bad
698 * situation to the system controller to reboot the backed.
699 */
700 if (ref == GRANT_INVALID_REF) {
701 if (net_ratelimit())
702 dev_warn(dev, "Bad rx response id %d.\n",
703 rx->id);
704 err = -EINVAL;
705 goto next;
706 }
707
708 ret = gnttab_end_foreign_access_ref(ref, 0);
709 BUG_ON(!ret);
710
711 gnttab_release_grant_reference(&np->gref_rx_head, ref);
712
713 __skb_queue_tail(list, skb);
714
715 next:
716 if (!(rx->flags & XEN_NETRXF_more_data))
717 break;
718
719 if (cons + frags == rp) {
720 if (net_ratelimit())
721 dev_warn(dev, "Need more frags\n");
722 err = -ENOENT;
723 break;
724 }
725
726 rx = RING_GET_RESPONSE(&np->rx, cons + frags);
727 skb = xennet_get_rx_skb(np, cons + frags);
728 ref = xennet_get_rx_ref(np, cons + frags);
729 frags++;
730 }
731
732 if (unlikely(frags > max)) {
733 if (net_ratelimit())
734 dev_warn(dev, "Too many frags\n");
735 err = -E2BIG;
736 }
737
738 if (unlikely(err))
739 np->rx.rsp_cons = cons + frags;
740
741 return err;
742 }
743
744 static int xennet_set_skb_gso(struct sk_buff *skb,
745 struct xen_netif_extra_info *gso)
746 {
747 if (!gso->u.gso.size) {
748 if (net_ratelimit())
749 printk(KERN_WARNING "GSO size must not be zero.\n");
750 return -EINVAL;
751 }
752
753 /* Currently only TCPv4 S.O. is supported. */
754 if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4) {
755 if (net_ratelimit())
756 printk(KERN_WARNING "Bad GSO type %d.\n", gso->u.gso.type);
757 return -EINVAL;
758 }
759
760 skb_shinfo(skb)->gso_size = gso->u.gso.size;
761 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
762
763 /* Header must be checked, and gso_segs computed. */
764 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
765 skb_shinfo(skb)->gso_segs = 0;
766
767 return 0;
768 }
769
770 static RING_IDX xennet_fill_frags(struct netfront_info *np,
771 struct sk_buff *skb,
772 struct sk_buff_head *list)
773 {
774 struct skb_shared_info *shinfo = skb_shinfo(skb);
775 int nr_frags = shinfo->nr_frags;
776 RING_IDX cons = np->rx.rsp_cons;
777 struct sk_buff *nskb;
778
779 while ((nskb = __skb_dequeue(list))) {
780 struct xen_netif_rx_response *rx =
781 RING_GET_RESPONSE(&np->rx, ++cons);
782 skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0];
783
784 __skb_fill_page_desc(skb, nr_frags,
785 skb_frag_page(nfrag),
786 rx->offset, rx->status);
787
788 skb->data_len += rx->status;
789
790 skb_shinfo(nskb)->nr_frags = 0;
791 kfree_skb(nskb);
792
793 nr_frags++;
794 }
795
796 shinfo->nr_frags = nr_frags;
797 return cons;
798 }
799
800 static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
801 {
802 struct iphdr *iph;
803 unsigned char *th;
804 int err = -EPROTO;
805 int recalculate_partial_csum = 0;
806
807 /*
808 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
809 * peers can fail to set NETRXF_csum_blank when sending a GSO
810 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
811 * recalculate the partial checksum.
812 */
813 if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
814 struct netfront_info *np = netdev_priv(dev);
815 np->rx_gso_checksum_fixup++;
816 skb->ip_summed = CHECKSUM_PARTIAL;
817 recalculate_partial_csum = 1;
818 }
819
820 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
821 if (skb->ip_summed != CHECKSUM_PARTIAL)
822 return 0;
823
824 if (skb->protocol != htons(ETH_P_IP))
825 goto out;
826
827 iph = (void *)skb->data;
828 th = skb->data + 4 * iph->ihl;
829 if (th >= skb_tail_pointer(skb))
830 goto out;
831
832 skb->csum_start = th - skb->head;
833 switch (iph->protocol) {
834 case IPPROTO_TCP:
835 skb->csum_offset = offsetof(struct tcphdr, check);
836
837 if (recalculate_partial_csum) {
838 struct tcphdr *tcph = (struct tcphdr *)th;
839 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
840 skb->len - iph->ihl*4,
841 IPPROTO_TCP, 0);
842 }
843 break;
844 case IPPROTO_UDP:
845 skb->csum_offset = offsetof(struct udphdr, check);
846
847 if (recalculate_partial_csum) {
848 struct udphdr *udph = (struct udphdr *)th;
849 udph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
850 skb->len - iph->ihl*4,
851 IPPROTO_UDP, 0);
852 }
853 break;
854 default:
855 if (net_ratelimit())
856 printk(KERN_ERR "Attempting to checksum a non-"
857 "TCP/UDP packet, dropping a protocol"
858 " %d packet", iph->protocol);
859 goto out;
860 }
861
862 if ((th + skb->csum_offset + 2) > skb_tail_pointer(skb))
863 goto out;
864
865 err = 0;
866
867 out:
868 return err;
869 }
870
871 static int handle_incoming_queue(struct net_device *dev,
872 struct sk_buff_head *rxq)
873 {
874 struct netfront_info *np = netdev_priv(dev);
875 struct netfront_stats *stats = this_cpu_ptr(np->stats);
876 int packets_dropped = 0;
877 struct sk_buff *skb;
878
879 while ((skb = __skb_dequeue(rxq)) != NULL) {
880 struct page *page = NETFRONT_SKB_CB(skb)->page;
881 void *vaddr = page_address(page);
882 unsigned offset = NETFRONT_SKB_CB(skb)->offset;
883
884 memcpy(skb->data, vaddr + offset,
885 skb_headlen(skb));
886
887 if (page != skb_frag_page(&skb_shinfo(skb)->frags[0]))
888 __free_page(page);
889
890 /* Ethernet work: Delayed to here as it peeks the header. */
891 skb->protocol = eth_type_trans(skb, dev);
892
893 if (checksum_setup(dev, skb)) {
894 kfree_skb(skb);
895 packets_dropped++;
896 dev->stats.rx_errors++;
897 continue;
898 }
899
900 u64_stats_update_begin(&stats->syncp);
901 stats->rx_packets++;
902 stats->rx_bytes += skb->len;
903 u64_stats_update_end(&stats->syncp);
904
905 /* Pass it up. */
906 netif_receive_skb(skb);
907 }
908
909 return packets_dropped;
910 }
911
912 static int xennet_poll(struct napi_struct *napi, int budget)
913 {
914 struct netfront_info *np = container_of(napi, struct netfront_info, napi);
915 struct net_device *dev = np->netdev;
916 struct sk_buff *skb;
917 struct netfront_rx_info rinfo;
918 struct xen_netif_rx_response *rx = &rinfo.rx;
919 struct xen_netif_extra_info *extras = rinfo.extras;
920 RING_IDX i, rp;
921 int work_done;
922 struct sk_buff_head rxq;
923 struct sk_buff_head errq;
924 struct sk_buff_head tmpq;
925 unsigned long flags;
926 unsigned int len;
927 int err;
928
929 spin_lock(&np->rx_lock);
930
931 skb_queue_head_init(&rxq);
932 skb_queue_head_init(&errq);
933 skb_queue_head_init(&tmpq);
934
935 rp = np->rx.sring->rsp_prod;
936 rmb(); /* Ensure we see queued responses up to 'rp'. */
937
938 i = np->rx.rsp_cons;
939 work_done = 0;
940 while ((i != rp) && (work_done < budget)) {
941 memcpy(rx, RING_GET_RESPONSE(&np->rx, i), sizeof(*rx));
942 memset(extras, 0, sizeof(rinfo.extras));
943
944 err = xennet_get_responses(np, &rinfo, rp, &tmpq);
945
946 if (unlikely(err)) {
947 err:
948 while ((skb = __skb_dequeue(&tmpq)))
949 __skb_queue_tail(&errq, skb);
950 dev->stats.rx_errors++;
951 i = np->rx.rsp_cons;
952 continue;
953 }
954
955 skb = __skb_dequeue(&tmpq);
956
957 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
958 struct xen_netif_extra_info *gso;
959 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
960
961 if (unlikely(xennet_set_skb_gso(skb, gso))) {
962 __skb_queue_head(&tmpq, skb);
963 np->rx.rsp_cons += skb_queue_len(&tmpq);
964 goto err;
965 }
966 }
967
968 NETFRONT_SKB_CB(skb)->page =
969 skb_frag_page(&skb_shinfo(skb)->frags[0]);
970 NETFRONT_SKB_CB(skb)->offset = rx->offset;
971
972 len = rx->status;
973 if (len > RX_COPY_THRESHOLD)
974 len = RX_COPY_THRESHOLD;
975 skb_put(skb, len);
976
977 if (rx->status > len) {
978 skb_shinfo(skb)->frags[0].page_offset =
979 rx->offset + len;
980 skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status - len);
981 skb->data_len = rx->status - len;
982 } else {
983 __skb_fill_page_desc(skb, 0, NULL, 0, 0);
984 skb_shinfo(skb)->nr_frags = 0;
985 }
986
987 i = xennet_fill_frags(np, skb, &tmpq);
988
989 /*
990 * Truesize approximates the size of true data plus
991 * any supervisor overheads. Adding hypervisor
992 * overheads has been shown to significantly reduce
993 * achievable bandwidth with the default receive
994 * buffer size. It is therefore not wise to account
995 * for it here.
996 *
997 * After alloc_skb(RX_COPY_THRESHOLD), truesize is set
998 * to RX_COPY_THRESHOLD + the supervisor
999 * overheads. Here, we add the size of the data pulled
1000 * in xennet_fill_frags().
1001 *
1002 * We also adjust for any unused space in the main
1003 * data area by subtracting (RX_COPY_THRESHOLD -
1004 * len). This is especially important with drivers
1005 * which split incoming packets into header and data,
1006 * using only 66 bytes of the main data area (see the
1007 * e1000 driver for example.) On such systems,
1008 * without this last adjustement, our achievable
1009 * receive throughout using the standard receive
1010 * buffer size was cut by 25%(!!!).
1011 */
1012 skb->truesize += skb->data_len - (RX_COPY_THRESHOLD - len);
1013 skb->len += skb->data_len;
1014
1015 if (rx->flags & XEN_NETRXF_csum_blank)
1016 skb->ip_summed = CHECKSUM_PARTIAL;
1017 else if (rx->flags & XEN_NETRXF_data_validated)
1018 skb->ip_summed = CHECKSUM_UNNECESSARY;
1019
1020 __skb_queue_tail(&rxq, skb);
1021
1022 np->rx.rsp_cons = ++i;
1023 work_done++;
1024 }
1025
1026 __skb_queue_purge(&errq);
1027
1028 work_done -= handle_incoming_queue(dev, &rxq);
1029
1030 /* If we get a callback with very few responses, reduce fill target. */
1031 /* NB. Note exponential increase, linear decrease. */
1032 if (((np->rx.req_prod_pvt - np->rx.sring->rsp_prod) >
1033 ((3*np->rx_target) / 4)) &&
1034 (--np->rx_target < np->rx_min_target))
1035 np->rx_target = np->rx_min_target;
1036
1037 xennet_alloc_rx_buffers(dev);
1038
1039 if (work_done < budget) {
1040 int more_to_do = 0;
1041
1042 local_irq_save(flags);
1043
1044 RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, more_to_do);
1045 if (!more_to_do)
1046 __napi_complete(napi);
1047
1048 local_irq_restore(flags);
1049 }
1050
1051 spin_unlock(&np->rx_lock);
1052
1053 return work_done;
1054 }
1055
1056 static int xennet_change_mtu(struct net_device *dev, int mtu)
1057 {
1058 int max = xennet_can_sg(dev) ?
1059 XEN_NETIF_MAX_TX_SIZE - MAX_TCP_HEADER : ETH_DATA_LEN;
1060
1061 if (mtu > max)
1062 return -EINVAL;
1063 dev->mtu = mtu;
1064 return 0;
1065 }
1066
1067 static struct rtnl_link_stats64 *xennet_get_stats64(struct net_device *dev,
1068 struct rtnl_link_stats64 *tot)
1069 {
1070 struct netfront_info *np = netdev_priv(dev);
1071 int cpu;
1072
1073 for_each_possible_cpu(cpu) {
1074 struct netfront_stats *stats = per_cpu_ptr(np->stats, cpu);
1075 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1076 unsigned int start;
1077
1078 do {
1079 start = u64_stats_fetch_begin_bh(&stats->syncp);
1080
1081 rx_packets = stats->rx_packets;
1082 tx_packets = stats->tx_packets;
1083 rx_bytes = stats->rx_bytes;
1084 tx_bytes = stats->tx_bytes;
1085 } while (u64_stats_fetch_retry_bh(&stats->syncp, start));
1086
1087 tot->rx_packets += rx_packets;
1088 tot->tx_packets += tx_packets;
1089 tot->rx_bytes += rx_bytes;
1090 tot->tx_bytes += tx_bytes;
1091 }
1092
1093 tot->rx_errors = dev->stats.rx_errors;
1094 tot->tx_dropped = dev->stats.tx_dropped;
1095
1096 return tot;
1097 }
1098
1099 static void xennet_release_tx_bufs(struct netfront_info *np)
1100 {
1101 struct sk_buff *skb;
1102 int i;
1103
1104 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1105 /* Skip over entries which are actually freelist references */
1106 if (skb_entry_is_link(&np->tx_skbs[i]))
1107 continue;
1108
1109 skb = np->tx_skbs[i].skb;
1110 gnttab_end_foreign_access_ref(np->grant_tx_ref[i],
1111 GNTMAP_readonly);
1112 gnttab_release_grant_reference(&np->gref_tx_head,
1113 np->grant_tx_ref[i]);
1114 np->grant_tx_ref[i] = GRANT_INVALID_REF;
1115 add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, i);
1116 dev_kfree_skb_irq(skb);
1117 }
1118 }
1119
1120 static void xennet_release_rx_bufs(struct netfront_info *np)
1121 {
1122 struct mmu_update *mmu = np->rx_mmu;
1123 struct multicall_entry *mcl = np->rx_mcl;
1124 struct sk_buff_head free_list;
1125 struct sk_buff *skb;
1126 unsigned long mfn;
1127 int xfer = 0, noxfer = 0, unused = 0;
1128 int id, ref;
1129
1130 dev_warn(&np->netdev->dev, "%s: fix me for copying receiver.\n",
1131 __func__);
1132 return;
1133
1134 skb_queue_head_init(&free_list);
1135
1136 spin_lock_bh(&np->rx_lock);
1137
1138 for (id = 0; id < NET_RX_RING_SIZE; id++) {
1139 ref = np->grant_rx_ref[id];
1140 if (ref == GRANT_INVALID_REF) {
1141 unused++;
1142 continue;
1143 }
1144
1145 skb = np->rx_skbs[id];
1146 mfn = gnttab_end_foreign_transfer_ref(ref);
1147 gnttab_release_grant_reference(&np->gref_rx_head, ref);
1148 np->grant_rx_ref[id] = GRANT_INVALID_REF;
1149
1150 if (0 == mfn) {
1151 skb_shinfo(skb)->nr_frags = 0;
1152 dev_kfree_skb(skb);
1153 noxfer++;
1154 continue;
1155 }
1156
1157 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1158 /* Remap the page. */
1159 const struct page *page =
1160 skb_frag_page(&skb_shinfo(skb)->frags[0]);
1161 unsigned long pfn = page_to_pfn(page);
1162 void *vaddr = page_address(page);
1163
1164 MULTI_update_va_mapping(mcl, (unsigned long)vaddr,
1165 mfn_pte(mfn, PAGE_KERNEL),
1166 0);
1167 mcl++;
1168 mmu->ptr = ((u64)mfn << PAGE_SHIFT)
1169 | MMU_MACHPHYS_UPDATE;
1170 mmu->val = pfn;
1171 mmu++;
1172
1173 set_phys_to_machine(pfn, mfn);
1174 }
1175 __skb_queue_tail(&free_list, skb);
1176 xfer++;
1177 }
1178
1179 dev_info(&np->netdev->dev, "%s: %d xfer, %d noxfer, %d unused\n",
1180 __func__, xfer, noxfer, unused);
1181
1182 if (xfer) {
1183 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1184 /* Do all the remapping work and M2P updates. */
1185 MULTI_mmu_update(mcl, np->rx_mmu, mmu - np->rx_mmu,
1186 NULL, DOMID_SELF);
1187 mcl++;
1188 HYPERVISOR_multicall(np->rx_mcl, mcl - np->rx_mcl);
1189 }
1190 }
1191
1192 __skb_queue_purge(&free_list);
1193
1194 spin_unlock_bh(&np->rx_lock);
1195 }
1196
1197 static void xennet_uninit(struct net_device *dev)
1198 {
1199 struct netfront_info *np = netdev_priv(dev);
1200 xennet_release_tx_bufs(np);
1201 xennet_release_rx_bufs(np);
1202 gnttab_free_grant_references(np->gref_tx_head);
1203 gnttab_free_grant_references(np->gref_rx_head);
1204 }
1205
1206 static netdev_features_t xennet_fix_features(struct net_device *dev,
1207 netdev_features_t features)
1208 {
1209 struct netfront_info *np = netdev_priv(dev);
1210 int val;
1211
1212 if (features & NETIF_F_SG) {
1213 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, "feature-sg",
1214 "%d", &val) < 0)
1215 val = 0;
1216
1217 if (!val)
1218 features &= ~NETIF_F_SG;
1219 }
1220
1221 if (features & NETIF_F_TSO) {
1222 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1223 "feature-gso-tcpv4", "%d", &val) < 0)
1224 val = 0;
1225
1226 if (!val)
1227 features &= ~NETIF_F_TSO;
1228 }
1229
1230 return features;
1231 }
1232
1233 static int xennet_set_features(struct net_device *dev,
1234 netdev_features_t features)
1235 {
1236 if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1237 netdev_info(dev, "Reducing MTU because no SG offload");
1238 dev->mtu = ETH_DATA_LEN;
1239 }
1240
1241 return 0;
1242 }
1243
1244 static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1245 {
1246 struct net_device *dev = dev_id;
1247 struct netfront_info *np = netdev_priv(dev);
1248 unsigned long flags;
1249
1250 spin_lock_irqsave(&np->tx_lock, flags);
1251
1252 if (likely(netif_carrier_ok(dev))) {
1253 xennet_tx_buf_gc(dev);
1254 /* Under tx_lock: protects access to rx shared-ring indexes. */
1255 if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx))
1256 napi_schedule(&np->napi);
1257 }
1258
1259 spin_unlock_irqrestore(&np->tx_lock, flags);
1260
1261 return IRQ_HANDLED;
1262 }
1263
1264 #ifdef CONFIG_NET_POLL_CONTROLLER
1265 static void xennet_poll_controller(struct net_device *dev)
1266 {
1267 xennet_interrupt(0, dev);
1268 }
1269 #endif
1270
1271 static const struct net_device_ops xennet_netdev_ops = {
1272 .ndo_open = xennet_open,
1273 .ndo_uninit = xennet_uninit,
1274 .ndo_stop = xennet_close,
1275 .ndo_start_xmit = xennet_start_xmit,
1276 .ndo_change_mtu = xennet_change_mtu,
1277 .ndo_get_stats64 = xennet_get_stats64,
1278 .ndo_set_mac_address = eth_mac_addr,
1279 .ndo_validate_addr = eth_validate_addr,
1280 .ndo_fix_features = xennet_fix_features,
1281 .ndo_set_features = xennet_set_features,
1282 #ifdef CONFIG_NET_POLL_CONTROLLER
1283 .ndo_poll_controller = xennet_poll_controller,
1284 #endif
1285 };
1286
1287 static struct net_device * __devinit xennet_create_dev(struct xenbus_device *dev)
1288 {
1289 int i, err;
1290 struct net_device *netdev;
1291 struct netfront_info *np;
1292
1293 netdev = alloc_etherdev(sizeof(struct netfront_info));
1294 if (!netdev)
1295 return ERR_PTR(-ENOMEM);
1296
1297 np = netdev_priv(netdev);
1298 np->xbdev = dev;
1299
1300 spin_lock_init(&np->tx_lock);
1301 spin_lock_init(&np->rx_lock);
1302
1303 skb_queue_head_init(&np->rx_batch);
1304 np->rx_target = RX_DFL_MIN_TARGET;
1305 np->rx_min_target = RX_DFL_MIN_TARGET;
1306 np->rx_max_target = RX_MAX_TARGET;
1307
1308 init_timer(&np->rx_refill_timer);
1309 np->rx_refill_timer.data = (unsigned long)netdev;
1310 np->rx_refill_timer.function = rx_refill_timeout;
1311
1312 err = -ENOMEM;
1313 np->stats = alloc_percpu(struct netfront_stats);
1314 if (np->stats == NULL)
1315 goto exit;
1316
1317 /* Initialise tx_skbs as a free chain containing every entry. */
1318 np->tx_skb_freelist = 0;
1319 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1320 skb_entry_set_link(&np->tx_skbs[i], i+1);
1321 np->grant_tx_ref[i] = GRANT_INVALID_REF;
1322 }
1323
1324 /* Clear out rx_skbs */
1325 for (i = 0; i < NET_RX_RING_SIZE; i++) {
1326 np->rx_skbs[i] = NULL;
1327 np->grant_rx_ref[i] = GRANT_INVALID_REF;
1328 }
1329
1330 /* A grant for every tx ring slot */
1331 if (gnttab_alloc_grant_references(TX_MAX_TARGET,
1332 &np->gref_tx_head) < 0) {
1333 printk(KERN_ALERT "#### netfront can't alloc tx grant refs\n");
1334 err = -ENOMEM;
1335 goto exit_free_stats;
1336 }
1337 /* A grant for every rx ring slot */
1338 if (gnttab_alloc_grant_references(RX_MAX_TARGET,
1339 &np->gref_rx_head) < 0) {
1340 printk(KERN_ALERT "#### netfront can't alloc rx grant refs\n");
1341 err = -ENOMEM;
1342 goto exit_free_tx;
1343 }
1344
1345 netdev->netdev_ops = &xennet_netdev_ops;
1346
1347 netif_napi_add(netdev, &np->napi, xennet_poll, 64);
1348 netdev->features = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1349 NETIF_F_GSO_ROBUST;
1350 netdev->hw_features = NETIF_F_IP_CSUM | NETIF_F_SG | NETIF_F_TSO;
1351
1352 /*
1353 * Assume that all hw features are available for now. This set
1354 * will be adjusted by the call to netdev_update_features() in
1355 * xennet_connect() which is the earliest point where we can
1356 * negotiate with the backend regarding supported features.
1357 */
1358 netdev->features |= netdev->hw_features;
1359
1360 SET_ETHTOOL_OPS(netdev, &xennet_ethtool_ops);
1361 SET_NETDEV_DEV(netdev, &dev->dev);
1362
1363 netif_set_gso_max_size(netdev, XEN_NETIF_MAX_TX_SIZE - MAX_TCP_HEADER);
1364
1365 np->netdev = netdev;
1366
1367 netif_carrier_off(netdev);
1368
1369 return netdev;
1370
1371 exit_free_tx:
1372 gnttab_free_grant_references(np->gref_tx_head);
1373 exit_free_stats:
1374 free_percpu(np->stats);
1375 exit:
1376 free_netdev(netdev);
1377 return ERR_PTR(err);
1378 }
1379
1380 /**
1381 * Entry point to this code when a new device is created. Allocate the basic
1382 * structures and the ring buffers for communication with the backend, and
1383 * inform the backend of the appropriate details for those.
1384 */
1385 static int __devinit netfront_probe(struct xenbus_device *dev,
1386 const struct xenbus_device_id *id)
1387 {
1388 int err;
1389 struct net_device *netdev;
1390 struct netfront_info *info;
1391
1392 netdev = xennet_create_dev(dev);
1393 if (IS_ERR(netdev)) {
1394 err = PTR_ERR(netdev);
1395 xenbus_dev_fatal(dev, err, "creating netdev");
1396 return err;
1397 }
1398
1399 info = netdev_priv(netdev);
1400 dev_set_drvdata(&dev->dev, info);
1401
1402 err = register_netdev(info->netdev);
1403 if (err) {
1404 printk(KERN_WARNING "%s: register_netdev err=%d\n",
1405 __func__, err);
1406 goto fail;
1407 }
1408
1409 err = xennet_sysfs_addif(info->netdev);
1410 if (err) {
1411 unregister_netdev(info->netdev);
1412 printk(KERN_WARNING "%s: add sysfs failed err=%d\n",
1413 __func__, err);
1414 goto fail;
1415 }
1416
1417 return 0;
1418
1419 fail:
1420 free_netdev(netdev);
1421 dev_set_drvdata(&dev->dev, NULL);
1422 return err;
1423 }
1424
1425 static void xennet_end_access(int ref, void *page)
1426 {
1427 /* This frees the page as a side-effect */
1428 if (ref != GRANT_INVALID_REF)
1429 gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1430 }
1431
1432 static void xennet_disconnect_backend(struct netfront_info *info)
1433 {
1434 /* Stop old i/f to prevent errors whilst we rebuild the state. */
1435 spin_lock_bh(&info->rx_lock);
1436 spin_lock_irq(&info->tx_lock);
1437 netif_carrier_off(info->netdev);
1438 spin_unlock_irq(&info->tx_lock);
1439 spin_unlock_bh(&info->rx_lock);
1440
1441 if (info->netdev->irq)
1442 unbind_from_irqhandler(info->netdev->irq, info->netdev);
1443 info->evtchn = info->netdev->irq = 0;
1444
1445 /* End access and free the pages */
1446 xennet_end_access(info->tx_ring_ref, info->tx.sring);
1447 xennet_end_access(info->rx_ring_ref, info->rx.sring);
1448
1449 info->tx_ring_ref = GRANT_INVALID_REF;
1450 info->rx_ring_ref = GRANT_INVALID_REF;
1451 info->tx.sring = NULL;
1452 info->rx.sring = NULL;
1453 }
1454
1455 /**
1456 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1457 * driver restart. We tear down our netif structure and recreate it, but
1458 * leave the device-layer structures intact so that this is transparent to the
1459 * rest of the kernel.
1460 */
1461 static int netfront_resume(struct xenbus_device *dev)
1462 {
1463 struct netfront_info *info = dev_get_drvdata(&dev->dev);
1464
1465 dev_dbg(&dev->dev, "%s\n", dev->nodename);
1466
1467 xennet_disconnect_backend(info);
1468 return 0;
1469 }
1470
1471 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1472 {
1473 char *s, *e, *macstr;
1474 int i;
1475
1476 macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1477 if (IS_ERR(macstr))
1478 return PTR_ERR(macstr);
1479
1480 for (i = 0; i < ETH_ALEN; i++) {
1481 mac[i] = simple_strtoul(s, &e, 16);
1482 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1483 kfree(macstr);
1484 return -ENOENT;
1485 }
1486 s = e+1;
1487 }
1488
1489 kfree(macstr);
1490 return 0;
1491 }
1492
1493 static int setup_netfront(struct xenbus_device *dev, struct netfront_info *info)
1494 {
1495 struct xen_netif_tx_sring *txs;
1496 struct xen_netif_rx_sring *rxs;
1497 int err;
1498 struct net_device *netdev = info->netdev;
1499
1500 info->tx_ring_ref = GRANT_INVALID_REF;
1501 info->rx_ring_ref = GRANT_INVALID_REF;
1502 info->rx.sring = NULL;
1503 info->tx.sring = NULL;
1504 netdev->irq = 0;
1505
1506 err = xen_net_read_mac(dev, netdev->dev_addr);
1507 if (err) {
1508 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
1509 goto fail;
1510 }
1511
1512 txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1513 if (!txs) {
1514 err = -ENOMEM;
1515 xenbus_dev_fatal(dev, err, "allocating tx ring page");
1516 goto fail;
1517 }
1518 SHARED_RING_INIT(txs);
1519 FRONT_RING_INIT(&info->tx, txs, PAGE_SIZE);
1520
1521 err = xenbus_grant_ring(dev, virt_to_mfn(txs));
1522 if (err < 0) {
1523 free_page((unsigned long)txs);
1524 goto fail;
1525 }
1526
1527 info->tx_ring_ref = err;
1528 rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1529 if (!rxs) {
1530 err = -ENOMEM;
1531 xenbus_dev_fatal(dev, err, "allocating rx ring page");
1532 goto fail;
1533 }
1534 SHARED_RING_INIT(rxs);
1535 FRONT_RING_INIT(&info->rx, rxs, PAGE_SIZE);
1536
1537 err = xenbus_grant_ring(dev, virt_to_mfn(rxs));
1538 if (err < 0) {
1539 free_page((unsigned long)rxs);
1540 goto fail;
1541 }
1542 info->rx_ring_ref = err;
1543
1544 err = xenbus_alloc_evtchn(dev, &info->evtchn);
1545 if (err)
1546 goto fail;
1547
1548 err = bind_evtchn_to_irqhandler(info->evtchn, xennet_interrupt,
1549 0, netdev->name, netdev);
1550 if (err < 0)
1551 goto fail;
1552 netdev->irq = err;
1553 return 0;
1554
1555 fail:
1556 return err;
1557 }
1558
1559 /* Common code used when first setting up, and when resuming. */
1560 static int talk_to_netback(struct xenbus_device *dev,
1561 struct netfront_info *info)
1562 {
1563 const char *message;
1564 struct xenbus_transaction xbt;
1565 int err;
1566
1567 /* Create shared ring, alloc event channel. */
1568 err = setup_netfront(dev, info);
1569 if (err)
1570 goto out;
1571
1572 again:
1573 err = xenbus_transaction_start(&xbt);
1574 if (err) {
1575 xenbus_dev_fatal(dev, err, "starting transaction");
1576 goto destroy_ring;
1577 }
1578
1579 err = xenbus_printf(xbt, dev->nodename, "tx-ring-ref", "%u",
1580 info->tx_ring_ref);
1581 if (err) {
1582 message = "writing tx ring-ref";
1583 goto abort_transaction;
1584 }
1585 err = xenbus_printf(xbt, dev->nodename, "rx-ring-ref", "%u",
1586 info->rx_ring_ref);
1587 if (err) {
1588 message = "writing rx ring-ref";
1589 goto abort_transaction;
1590 }
1591 err = xenbus_printf(xbt, dev->nodename,
1592 "event-channel", "%u", info->evtchn);
1593 if (err) {
1594 message = "writing event-channel";
1595 goto abort_transaction;
1596 }
1597
1598 err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
1599 1);
1600 if (err) {
1601 message = "writing request-rx-copy";
1602 goto abort_transaction;
1603 }
1604
1605 err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
1606 if (err) {
1607 message = "writing feature-rx-notify";
1608 goto abort_transaction;
1609 }
1610
1611 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
1612 if (err) {
1613 message = "writing feature-sg";
1614 goto abort_transaction;
1615 }
1616
1617 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
1618 if (err) {
1619 message = "writing feature-gso-tcpv4";
1620 goto abort_transaction;
1621 }
1622
1623 err = xenbus_transaction_end(xbt, 0);
1624 if (err) {
1625 if (err == -EAGAIN)
1626 goto again;
1627 xenbus_dev_fatal(dev, err, "completing transaction");
1628 goto destroy_ring;
1629 }
1630
1631 return 0;
1632
1633 abort_transaction:
1634 xenbus_transaction_end(xbt, 1);
1635 xenbus_dev_fatal(dev, err, "%s", message);
1636 destroy_ring:
1637 xennet_disconnect_backend(info);
1638 out:
1639 return err;
1640 }
1641
1642 static int xennet_connect(struct net_device *dev)
1643 {
1644 struct netfront_info *np = netdev_priv(dev);
1645 int i, requeue_idx, err;
1646 struct sk_buff *skb;
1647 grant_ref_t ref;
1648 struct xen_netif_rx_request *req;
1649 unsigned int feature_rx_copy;
1650
1651 err = xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1652 "feature-rx-copy", "%u", &feature_rx_copy);
1653 if (err != 1)
1654 feature_rx_copy = 0;
1655
1656 if (!feature_rx_copy) {
1657 dev_info(&dev->dev,
1658 "backend does not support copying receive path\n");
1659 return -ENODEV;
1660 }
1661
1662 err = talk_to_netback(np->xbdev, np);
1663 if (err)
1664 return err;
1665
1666 rtnl_lock();
1667 netdev_update_features(dev);
1668 rtnl_unlock();
1669
1670 spin_lock_bh(&np->rx_lock);
1671 spin_lock_irq(&np->tx_lock);
1672
1673 /* Step 1: Discard all pending TX packet fragments. */
1674 xennet_release_tx_bufs(np);
1675
1676 /* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */
1677 for (requeue_idx = 0, i = 0; i < NET_RX_RING_SIZE; i++) {
1678 skb_frag_t *frag;
1679 const struct page *page;
1680 if (!np->rx_skbs[i])
1681 continue;
1682
1683 skb = np->rx_skbs[requeue_idx] = xennet_get_rx_skb(np, i);
1684 ref = np->grant_rx_ref[requeue_idx] = xennet_get_rx_ref(np, i);
1685 req = RING_GET_REQUEST(&np->rx, requeue_idx);
1686
1687 frag = &skb_shinfo(skb)->frags[0];
1688 page = skb_frag_page(frag);
1689 gnttab_grant_foreign_access_ref(
1690 ref, np->xbdev->otherend_id,
1691 pfn_to_mfn(page_to_pfn(page)),
1692 0);
1693 req->gref = ref;
1694 req->id = requeue_idx;
1695
1696 requeue_idx++;
1697 }
1698
1699 np->rx.req_prod_pvt = requeue_idx;
1700
1701 /*
1702 * Step 3: All public and private state should now be sane. Get
1703 * ready to start sending and receiving packets and give the driver
1704 * domain a kick because we've probably just requeued some
1705 * packets.
1706 */
1707 netif_carrier_on(np->netdev);
1708 notify_remote_via_irq(np->netdev->irq);
1709 xennet_tx_buf_gc(dev);
1710 xennet_alloc_rx_buffers(dev);
1711
1712 spin_unlock_irq(&np->tx_lock);
1713 spin_unlock_bh(&np->rx_lock);
1714
1715 return 0;
1716 }
1717
1718 /**
1719 * Callback received when the backend's state changes.
1720 */
1721 static void netback_changed(struct xenbus_device *dev,
1722 enum xenbus_state backend_state)
1723 {
1724 struct netfront_info *np = dev_get_drvdata(&dev->dev);
1725 struct net_device *netdev = np->netdev;
1726
1727 dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
1728
1729 switch (backend_state) {
1730 case XenbusStateInitialising:
1731 case XenbusStateInitialised:
1732 case XenbusStateReconfiguring:
1733 case XenbusStateReconfigured:
1734 case XenbusStateUnknown:
1735 case XenbusStateClosed:
1736 break;
1737
1738 case XenbusStateInitWait:
1739 if (dev->state != XenbusStateInitialising)
1740 break;
1741 if (xennet_connect(netdev) != 0)
1742 break;
1743 xenbus_switch_state(dev, XenbusStateConnected);
1744 break;
1745
1746 case XenbusStateConnected:
1747 netif_notify_peers(netdev);
1748 break;
1749
1750 case XenbusStateClosing:
1751 xenbus_frontend_closed(dev);
1752 break;
1753 }
1754 }
1755
1756 static const struct xennet_stat {
1757 char name[ETH_GSTRING_LEN];
1758 u16 offset;
1759 } xennet_stats[] = {
1760 {
1761 "rx_gso_checksum_fixup",
1762 offsetof(struct netfront_info, rx_gso_checksum_fixup)
1763 },
1764 };
1765
1766 static int xennet_get_sset_count(struct net_device *dev, int string_set)
1767 {
1768 switch (string_set) {
1769 case ETH_SS_STATS:
1770 return ARRAY_SIZE(xennet_stats);
1771 default:
1772 return -EINVAL;
1773 }
1774 }
1775
1776 static void xennet_get_ethtool_stats(struct net_device *dev,
1777 struct ethtool_stats *stats, u64 * data)
1778 {
1779 void *np = netdev_priv(dev);
1780 int i;
1781
1782 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
1783 data[i] = *(unsigned long *)(np + xennet_stats[i].offset);
1784 }
1785
1786 static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
1787 {
1788 int i;
1789
1790 switch (stringset) {
1791 case ETH_SS_STATS:
1792 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
1793 memcpy(data + i * ETH_GSTRING_LEN,
1794 xennet_stats[i].name, ETH_GSTRING_LEN);
1795 break;
1796 }
1797 }
1798
1799 static const struct ethtool_ops xennet_ethtool_ops =
1800 {
1801 .get_link = ethtool_op_get_link,
1802
1803 .get_sset_count = xennet_get_sset_count,
1804 .get_ethtool_stats = xennet_get_ethtool_stats,
1805 .get_strings = xennet_get_strings,
1806 };
1807
1808 #ifdef CONFIG_SYSFS
1809 static ssize_t show_rxbuf_min(struct device *dev,
1810 struct device_attribute *attr, char *buf)
1811 {
1812 struct net_device *netdev = to_net_dev(dev);
1813 struct netfront_info *info = netdev_priv(netdev);
1814
1815 return sprintf(buf, "%u\n", info->rx_min_target);
1816 }
1817
1818 static ssize_t store_rxbuf_min(struct device *dev,
1819 struct device_attribute *attr,
1820 const char *buf, size_t len)
1821 {
1822 struct net_device *netdev = to_net_dev(dev);
1823 struct netfront_info *np = netdev_priv(netdev);
1824 char *endp;
1825 unsigned long target;
1826
1827 if (!capable(CAP_NET_ADMIN))
1828 return -EPERM;
1829
1830 target = simple_strtoul(buf, &endp, 0);
1831 if (endp == buf)
1832 return -EBADMSG;
1833
1834 if (target < RX_MIN_TARGET)
1835 target = RX_MIN_TARGET;
1836 if (target > RX_MAX_TARGET)
1837 target = RX_MAX_TARGET;
1838
1839 spin_lock_bh(&np->rx_lock);
1840 if (target > np->rx_max_target)
1841 np->rx_max_target = target;
1842 np->rx_min_target = target;
1843 if (target > np->rx_target)
1844 np->rx_target = target;
1845
1846 xennet_alloc_rx_buffers(netdev);
1847
1848 spin_unlock_bh(&np->rx_lock);
1849 return len;
1850 }
1851
1852 static ssize_t show_rxbuf_max(struct device *dev,
1853 struct device_attribute *attr, char *buf)
1854 {
1855 struct net_device *netdev = to_net_dev(dev);
1856 struct netfront_info *info = netdev_priv(netdev);
1857
1858 return sprintf(buf, "%u\n", info->rx_max_target);
1859 }
1860
1861 static ssize_t store_rxbuf_max(struct device *dev,
1862 struct device_attribute *attr,
1863 const char *buf, size_t len)
1864 {
1865 struct net_device *netdev = to_net_dev(dev);
1866 struct netfront_info *np = netdev_priv(netdev);
1867 char *endp;
1868 unsigned long target;
1869
1870 if (!capable(CAP_NET_ADMIN))
1871 return -EPERM;
1872
1873 target = simple_strtoul(buf, &endp, 0);
1874 if (endp == buf)
1875 return -EBADMSG;
1876
1877 if (target < RX_MIN_TARGET)
1878 target = RX_MIN_TARGET;
1879 if (target > RX_MAX_TARGET)
1880 target = RX_MAX_TARGET;
1881
1882 spin_lock_bh(&np->rx_lock);
1883 if (target < np->rx_min_target)
1884 np->rx_min_target = target;
1885 np->rx_max_target = target;
1886 if (target < np->rx_target)
1887 np->rx_target = target;
1888
1889 xennet_alloc_rx_buffers(netdev);
1890
1891 spin_unlock_bh(&np->rx_lock);
1892 return len;
1893 }
1894
1895 static ssize_t show_rxbuf_cur(struct device *dev,
1896 struct device_attribute *attr, char *buf)
1897 {
1898 struct net_device *netdev = to_net_dev(dev);
1899 struct netfront_info *info = netdev_priv(netdev);
1900
1901 return sprintf(buf, "%u\n", info->rx_target);
1902 }
1903
1904 static struct device_attribute xennet_attrs[] = {
1905 __ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf_min, store_rxbuf_min),
1906 __ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf_max, store_rxbuf_max),
1907 __ATTR(rxbuf_cur, S_IRUGO, show_rxbuf_cur, NULL),
1908 };
1909
1910 static int xennet_sysfs_addif(struct net_device *netdev)
1911 {
1912 int i;
1913 int err;
1914
1915 for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++) {
1916 err = device_create_file(&netdev->dev,
1917 &xennet_attrs[i]);
1918 if (err)
1919 goto fail;
1920 }
1921 return 0;
1922
1923 fail:
1924 while (--i >= 0)
1925 device_remove_file(&netdev->dev, &xennet_attrs[i]);
1926 return err;
1927 }
1928
1929 static void xennet_sysfs_delif(struct net_device *netdev)
1930 {
1931 int i;
1932
1933 for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++)
1934 device_remove_file(&netdev->dev, &xennet_attrs[i]);
1935 }
1936
1937 #endif /* CONFIG_SYSFS */
1938
1939 static const struct xenbus_device_id netfront_ids[] = {
1940 { "vif" },
1941 { "" }
1942 };
1943
1944
1945 static int __devexit xennet_remove(struct xenbus_device *dev)
1946 {
1947 struct netfront_info *info = dev_get_drvdata(&dev->dev);
1948
1949 dev_dbg(&dev->dev, "%s\n", dev->nodename);
1950
1951 xennet_disconnect_backend(info);
1952
1953 xennet_sysfs_delif(info->netdev);
1954
1955 unregister_netdev(info->netdev);
1956
1957 del_timer_sync(&info->rx_refill_timer);
1958
1959 free_percpu(info->stats);
1960
1961 free_netdev(info->netdev);
1962
1963 return 0;
1964 }
1965
1966 static DEFINE_XENBUS_DRIVER(netfront, ,
1967 .probe = netfront_probe,
1968 .remove = __devexit_p(xennet_remove),
1969 .resume = netfront_resume,
1970 .otherend_changed = netback_changed,
1971 );
1972
1973 static int __init netif_init(void)
1974 {
1975 if (!xen_domain())
1976 return -ENODEV;
1977
1978 if (xen_initial_domain())
1979 return 0;
1980
1981 if (xen_hvm_domain() && !xen_platform_pci_unplug)
1982 return -ENODEV;
1983
1984 printk(KERN_INFO "Initialising Xen virtual ethernet driver.\n");
1985
1986 return xenbus_register_frontend(&netfront_driver);
1987 }
1988 module_init(netif_init);
1989
1990
1991 static void __exit netif_exit(void)
1992 {
1993 if (xen_initial_domain())
1994 return;
1995
1996 xenbus_unregister_driver(&netfront_driver);
1997 }
1998 module_exit(netif_exit);
1999
2000 MODULE_DESCRIPTION("Xen virtual network device frontend");
2001 MODULE_LICENSE("GPL");
2002 MODULE_ALIAS("xen:vif");
2003 MODULE_ALIAS("xennet");
Linux with added FPC-III support