?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'sched/urgent'
[linux-2.6/x86.git] / drivers / net / xen-netfront.c
blobd7c8a98daff60a56e2cb403548261a21aff1d22f
1 /*
2 * Virtual network driver for conversing with remote driver backends.
3 *
4 * Copyright (c) 2002-2005, K A Fraser
5 * Copyright (c) 2005, XenSource Ltd
6 *
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License version 2
9 * as published by the Free Software Foundation; or, when distributed
10 * separately from the Linux kernel or incorporated into other
11 * software packages, subject to the following license:
12 *
13 * Permission is hereby granted, free of charge, to any person obtaining a copy
14 * of this source file (the "Software"), to deal in the Software without
15 * restriction, including without limitation the rights to use, copy, modify,
16 * merge, publish, distribute, sublicense, and/or sell copies of the Software,
17 * and to permit persons to whom the Software is furnished to do so, subject to
18 * the following conditions:
19 *
20 * The above copyright notice and this permission notice shall be included in
21 * all copies or substantial portions of the Software.
22 *
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
24 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
25 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
26 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
27 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
28 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
29 * IN THE SOFTWARE.
30 */
31
32 #include <linux/module.h>
33 #include <linux/kernel.h>
34 #include <linux/netdevice.h>
35 #include <linux/etherdevice.h>
36 #include <linux/skbuff.h>
37 #include <linux/ethtool.h>
38 #include <linux/if_ether.h>
39 #include <linux/tcp.h>
40 #include <linux/udp.h>
41 #include <linux/moduleparam.h>
42 #include <linux/mm.h>
43 #include <linux/slab.h>
44 #include <net/ip.h>
45
46 #include <xen/xen.h>
47 #include <xen/xenbus.h>
48 #include <xen/events.h>
49 #include <xen/page.h>
50 #include <xen/grant_table.h>
51
52 #include <xen/interface/io/netif.h>
53 #include <xen/interface/memory.h>
54 #include <xen/interface/grant_table.h>
55
56 static const struct ethtool_ops xennet_ethtool_ops;
57
58 struct netfront_cb {
59 struct page *page;
60 unsigned offset;
61 };
62
63 #define NETFRONT_SKB_CB(skb) ((struct netfront_cb *)((skb)->cb))
64
65 #define RX_COPY_THRESHOLD 256
66
67 #define GRANT_INVALID_REF 0
68
69 #define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, PAGE_SIZE)
70 #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, PAGE_SIZE)
71 #define TX_MAX_TARGET min_t(int, NET_RX_RING_SIZE, 256)
72
73 struct netfront_stats {
74 u64 rx_packets;
75 u64 tx_packets;
76 u64 rx_bytes;
77 u64 tx_bytes;
78 struct u64_stats_sync syncp;
79 };
80
81 struct netfront_info {
82 struct list_head list;
83 struct net_device *netdev;
84
85 struct napi_struct napi;
86
87 unsigned int evtchn;
88 struct xenbus_device *xbdev;
89
90 spinlock_t tx_lock;
91 struct xen_netif_tx_front_ring tx;
92 int tx_ring_ref;
93
94 /*
95 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
96 * are linked from tx_skb_freelist through skb_entry.link.
97 *
98 * NB. Freelist index entries are always going to be less than
99 * PAGE_OFFSET, whereas pointers to skbs will always be equal or
100 * greater than PAGE_OFFSET: we use this property to distinguish
101 * them.
102 */
103 union skb_entry {
104 struct sk_buff *skb;
105 unsigned long link;
106 } tx_skbs[NET_TX_RING_SIZE];
107 grant_ref_t gref_tx_head;
108 grant_ref_t grant_tx_ref[NET_TX_RING_SIZE];
109 unsigned tx_skb_freelist;
110
111 spinlock_t rx_lock ____cacheline_aligned_in_smp;
112 struct xen_netif_rx_front_ring rx;
113 int rx_ring_ref;
114
115 /* Receive-ring batched refills. */
116 #define RX_MIN_TARGET 8
117 #define RX_DFL_MIN_TARGET 64
118 #define RX_MAX_TARGET min_t(int, NET_RX_RING_SIZE, 256)
119 unsigned rx_min_target, rx_max_target, rx_target;
120 struct sk_buff_head rx_batch;
121
122 struct timer_list rx_refill_timer;
123
124 struct sk_buff *rx_skbs[NET_RX_RING_SIZE];
125 grant_ref_t gref_rx_head;
126 grant_ref_t grant_rx_ref[NET_RX_RING_SIZE];
127
128 unsigned long rx_pfn_array[NET_RX_RING_SIZE];
129 struct multicall_entry rx_mcl[NET_RX_RING_SIZE+1];
130 struct mmu_update rx_mmu[NET_RX_RING_SIZE];
131
132 /* Statistics */
133 struct netfront_stats __percpu *stats;
134
135 unsigned long rx_gso_checksum_fixup;
136 };
137
138 struct netfront_rx_info {
139 struct xen_netif_rx_response rx;
140 struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
141 };
142
143 static void skb_entry_set_link(union skb_entry *list, unsigned short id)
144 {
145 list->link = id;
146 }
147
148 static int skb_entry_is_link(const union skb_entry *list)
149 {
150 BUILD_BUG_ON(sizeof(list->skb) != sizeof(list->link));
151 return (unsigned long)list->skb < PAGE_OFFSET;
152 }
153
154 /*
155 * Access macros for acquiring freeing slots in tx_skbs[].
156 */
157
158 static void add_id_to_freelist(unsigned *head, union skb_entry *list,
159 unsigned short id)
160 {
161 skb_entry_set_link(&list[id], *head);
162 *head = id;
163 }
164
165 static unsigned short get_id_from_freelist(unsigned *head,
166 union skb_entry *list)
167 {
168 unsigned int id = *head;
169 *head = list[id].link;
170 return id;
171 }
172
173 static int xennet_rxidx(RING_IDX idx)
174 {
175 return idx & (NET_RX_RING_SIZE - 1);
176 }
177
178 static struct sk_buff *xennet_get_rx_skb(struct netfront_info *np,
179 RING_IDX ri)
180 {
181 int i = xennet_rxidx(ri);
182 struct sk_buff *skb = np->rx_skbs[i];
183 np->rx_skbs[i] = NULL;
184 return skb;
185 }
186
187 static grant_ref_t xennet_get_rx_ref(struct netfront_info *np,
188 RING_IDX ri)
189 {
190 int i = xennet_rxidx(ri);
191 grant_ref_t ref = np->grant_rx_ref[i];
192 np->grant_rx_ref[i] = GRANT_INVALID_REF;
193 return ref;
194 }
195
196 #ifdef CONFIG_SYSFS
197 static int xennet_sysfs_addif(struct net_device *netdev);
198 static void xennet_sysfs_delif(struct net_device *netdev);
199 #else /* !CONFIG_SYSFS */
200 #define xennet_sysfs_addif(dev) (0)
201 #define xennet_sysfs_delif(dev) do { } while (0)
202 #endif
203
204 static int xennet_can_sg(struct net_device *dev)
205 {
206 return dev->features & NETIF_F_SG;
207 }
208
209
210 static void rx_refill_timeout(unsigned long data)
211 {
212 struct net_device *dev = (struct net_device *)data;
213 struct netfront_info *np = netdev_priv(dev);
214 napi_schedule(&np->napi);
215 }
216
217 static int netfront_tx_slot_available(struct netfront_info *np)
218 {
219 return (np->tx.req_prod_pvt - np->tx.rsp_cons) <
220 (TX_MAX_TARGET - MAX_SKB_FRAGS - 2);
221 }
222
223 static void xennet_maybe_wake_tx(struct net_device *dev)
224 {
225 struct netfront_info *np = netdev_priv(dev);
226
227 if (unlikely(netif_queue_stopped(dev)) &&
228 netfront_tx_slot_available(np) &&
229 likely(netif_running(dev)))
230 netif_wake_queue(dev);
231 }
232
233 static void xennet_alloc_rx_buffers(struct net_device *dev)
234 {
235 unsigned short id;
236 struct netfront_info *np = netdev_priv(dev);
237 struct sk_buff *skb;
238 struct page *page;
239 int i, batch_target, notify;
240 RING_IDX req_prod = np->rx.req_prod_pvt;
241 grant_ref_t ref;
242 unsigned long pfn;
243 void *vaddr;
244 struct xen_netif_rx_request *req;
245
246 if (unlikely(!netif_carrier_ok(dev)))
247 return;
248
249 /*
250 * Allocate skbuffs greedily, even though we batch updates to the
251 * receive ring. This creates a less bursty demand on the memory
252 * allocator, so should reduce the chance of failed allocation requests
253 * both for ourself and for other kernel subsystems.
254 */
255 batch_target = np->rx_target - (req_prod - np->rx.rsp_cons);
256 for (i = skb_queue_len(&np->rx_batch); i < batch_target; i++) {
257 skb = __netdev_alloc_skb(dev, RX_COPY_THRESHOLD + NET_IP_ALIGN,
258 GFP_ATOMIC | __GFP_NOWARN);
259 if (unlikely(!skb))
260 goto no_skb;
261
262 /* Align ip header to a 16 bytes boundary */
263 skb_reserve(skb, NET_IP_ALIGN);
264
265 page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
266 if (!page) {
267 kfree_skb(skb);
268 no_skb:
269 /* Any skbuffs queued for refill? Force them out. */
270 if (i != 0)
271 goto refill;
272 /* Could not allocate any skbuffs. Try again later. */
273 mod_timer(&np->rx_refill_timer,
274 jiffies + (HZ/10));
275 break;
276 }
277
278 skb_shinfo(skb)->frags[0].page = page;
279 skb_shinfo(skb)->nr_frags = 1;
280 __skb_queue_tail(&np->rx_batch, skb);
281 }
282
283 /* Is the batch large enough to be worthwhile? */
284 if (i < (np->rx_target/2)) {
285 if (req_prod > np->rx.sring->req_prod)
286 goto push;
287 return;
288 }
289
290 /* Adjust our fill target if we risked running out of buffers. */
291 if (((req_prod - np->rx.sring->rsp_prod) < (np->rx_target / 4)) &&
292 ((np->rx_target *= 2) > np->rx_max_target))
293 np->rx_target = np->rx_max_target;
294
295 refill:
296 for (i = 0; ; i++) {
297 skb = __skb_dequeue(&np->rx_batch);
298 if (skb == NULL)
299 break;
300
301 skb->dev = dev;
302
303 id = xennet_rxidx(req_prod + i);
304
305 BUG_ON(np->rx_skbs[id]);
306 np->rx_skbs[id] = skb;
307
308 ref = gnttab_claim_grant_reference(&np->gref_rx_head);
309 BUG_ON((signed short)ref < 0);
310 np->grant_rx_ref[id] = ref;
311
312 pfn = page_to_pfn(skb_shinfo(skb)->frags[0].page);
313 vaddr = page_address(skb_shinfo(skb)->frags[0].page);
314
315 req = RING_GET_REQUEST(&np->rx, req_prod + i);
316 gnttab_grant_foreign_access_ref(ref,
317 np->xbdev->otherend_id,
318 pfn_to_mfn(pfn),
319 0);
320
321 req->id = id;
322 req->gref = ref;
323 }
324
325 wmb(); /* barrier so backend seens requests */
326
327 /* Above is a suitable barrier to ensure backend will see requests. */
328 np->rx.req_prod_pvt = req_prod + i;
329 push:
330 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->rx, notify);
331 if (notify)
332 notify_remote_via_irq(np->netdev->irq);
333 }
334
335 static int xennet_open(struct net_device *dev)
336 {
337 struct netfront_info *np = netdev_priv(dev);
338
339 napi_enable(&np->napi);
340
341 spin_lock_bh(&np->rx_lock);
342 if (netif_carrier_ok(dev)) {
343 xennet_alloc_rx_buffers(dev);
344 np->rx.sring->rsp_event = np->rx.rsp_cons + 1;
345 if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx))
346 napi_schedule(&np->napi);
347 }
348 spin_unlock_bh(&np->rx_lock);
349
350 netif_start_queue(dev);
351
352 return 0;
353 }
354
355 static void xennet_tx_buf_gc(struct net_device *dev)
356 {
357 RING_IDX cons, prod;
358 unsigned short id;
359 struct netfront_info *np = netdev_priv(dev);
360 struct sk_buff *skb;
361
362 BUG_ON(!netif_carrier_ok(dev));
363
364 do {
365 prod = np->tx.sring->rsp_prod;
366 rmb(); /* Ensure we see responses up to 'rp'. */
367
368 for (cons = np->tx.rsp_cons; cons != prod; cons++) {
369 struct xen_netif_tx_response *txrsp;
370
371 txrsp = RING_GET_RESPONSE(&np->tx, cons);
372 if (txrsp->status == XEN_NETIF_RSP_NULL)
373 continue;
374
375 id = txrsp->id;
376 skb = np->tx_skbs[id].skb;
377 if (unlikely(gnttab_query_foreign_access(
378 np->grant_tx_ref[id]) != 0)) {
379 printk(KERN_ALERT "xennet_tx_buf_gc: warning "
380 "-- grant still in use by backend "
381 "domain.\n");
382 BUG();
383 }
384 gnttab_end_foreign_access_ref(
385 np->grant_tx_ref[id], GNTMAP_readonly);
386 gnttab_release_grant_reference(
387 &np->gref_tx_head, np->grant_tx_ref[id]);
388 np->grant_tx_ref[id] = GRANT_INVALID_REF;
389 add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, id);
390 dev_kfree_skb_irq(skb);
391 }
392
393 np->tx.rsp_cons = prod;
394
395 /*
396 * Set a new event, then check for race with update of tx_cons.
397 * Note that it is essential to schedule a callback, no matter
398 * how few buffers are pending. Even if there is space in the
399 * transmit ring, higher layers may be blocked because too much
400 * data is outstanding: in such cases notification from Xen is
401 * likely to be the only kick that we'll get.
402 */
403 np->tx.sring->rsp_event =
404 prod + ((np->tx.sring->req_prod - prod) >> 1) + 1;
405 mb(); /* update shared area */
406 } while ((cons == prod) && (prod != np->tx.sring->rsp_prod));
407
408 xennet_maybe_wake_tx(dev);
409 }
410
411 static void xennet_make_frags(struct sk_buff *skb, struct net_device *dev,
412 struct xen_netif_tx_request *tx)
413 {
414 struct netfront_info *np = netdev_priv(dev);
415 char *data = skb->data;
416 unsigned long mfn;
417 RING_IDX prod = np->tx.req_prod_pvt;
418 int frags = skb_shinfo(skb)->nr_frags;
419 unsigned int offset = offset_in_page(data);
420 unsigned int len = skb_headlen(skb);
421 unsigned int id;
422 grant_ref_t ref;
423 int i;
424
425 /* While the header overlaps a page boundary (including being
426 larger than a page), split it it into page-sized chunks. */
427 while (len > PAGE_SIZE - offset) {
428 tx->size = PAGE_SIZE - offset;
429 tx->flags |= XEN_NETTXF_more_data;
430 len -= tx->size;
431 data += tx->size;
432 offset = 0;
433
434 id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
435 np->tx_skbs[id].skb = skb_get(skb);
436 tx = RING_GET_REQUEST(&np->tx, prod++);
437 tx->id = id;
438 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
439 BUG_ON((signed short)ref < 0);
440
441 mfn = virt_to_mfn(data);
442 gnttab_grant_foreign_access_ref(ref, np->xbdev->otherend_id,
443 mfn, GNTMAP_readonly);
444
445 tx->gref = np->grant_tx_ref[id] = ref;
446 tx->offset = offset;
447 tx->size = len;
448 tx->flags = 0;
449 }
450
451 /* Grant backend access to each skb fragment page. */
452 for (i = 0; i < frags; i++) {
453 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
454
455 tx->flags |= XEN_NETTXF_more_data;
456
457 id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
458 np->tx_skbs[id].skb = skb_get(skb);
459 tx = RING_GET_REQUEST(&np->tx, prod++);
460 tx->id = id;
461 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
462 BUG_ON((signed short)ref < 0);
463
464 mfn = pfn_to_mfn(page_to_pfn(frag->page));
465 gnttab_grant_foreign_access_ref(ref, np->xbdev->otherend_id,
466 mfn, GNTMAP_readonly);
467
468 tx->gref = np->grant_tx_ref[id] = ref;
469 tx->offset = frag->page_offset;
470 tx->size = frag->size;
471 tx->flags = 0;
472 }
473
474 np->tx.req_prod_pvt = prod;
475 }
476
477 static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
478 {
479 unsigned short id;
480 struct netfront_info *np = netdev_priv(dev);
481 struct netfront_stats *stats = this_cpu_ptr(np->stats);
482 struct xen_netif_tx_request *tx;
483 struct xen_netif_extra_info *extra;
484 char *data = skb->data;
485 RING_IDX i;
486 grant_ref_t ref;
487 unsigned long mfn;
488 int notify;
489 int frags = skb_shinfo(skb)->nr_frags;
490 unsigned int offset = offset_in_page(data);
491 unsigned int len = skb_headlen(skb);
492
493 frags += DIV_ROUND_UP(offset + len, PAGE_SIZE);
494 if (unlikely(frags > MAX_SKB_FRAGS + 1)) {
495 printk(KERN_ALERT "xennet: skb rides the rocket: %d frags\n",
496 frags);
497 dump_stack();
498 goto drop;
499 }
500
501 spin_lock_irq(&np->tx_lock);
502
503 if (unlikely(!netif_carrier_ok(dev) ||
504 (frags > 1 && !xennet_can_sg(dev)) ||
505 netif_needs_gso(skb, netif_skb_features(skb)))) {
506 spin_unlock_irq(&np->tx_lock);
507 goto drop;
508 }
509
510 i = np->tx.req_prod_pvt;
511
512 id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
513 np->tx_skbs[id].skb = skb;
514
515 tx = RING_GET_REQUEST(&np->tx, i);
516
517 tx->id = id;
518 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
519 BUG_ON((signed short)ref < 0);
520 mfn = virt_to_mfn(data);
521 gnttab_grant_foreign_access_ref(
522 ref, np->xbdev->otherend_id, mfn, GNTMAP_readonly);
523 tx->gref = np->grant_tx_ref[id] = ref;
524 tx->offset = offset;
525 tx->size = len;
526 extra = NULL;
527
528 tx->flags = 0;
529 if (skb->ip_summed == CHECKSUM_PARTIAL)
530 /* local packet? */
531 tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated;
532 else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
533 /* remote but checksummed. */
534 tx->flags |= XEN_NETTXF_data_validated;
535
536 if (skb_shinfo(skb)->gso_size) {
537 struct xen_netif_extra_info *gso;
538
539 gso = (struct xen_netif_extra_info *)
540 RING_GET_REQUEST(&np->tx, ++i);
541
542 if (extra)
543 extra->flags |= XEN_NETIF_EXTRA_FLAG_MORE;
544 else
545 tx->flags |= XEN_NETTXF_extra_info;
546
547 gso->u.gso.size = skb_shinfo(skb)->gso_size;
548 gso->u.gso.type = XEN_NETIF_GSO_TYPE_TCPV4;
549 gso->u.gso.pad = 0;
550 gso->u.gso.features = 0;
551
552 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
553 gso->flags = 0;
554 extra = gso;
555 }
556
557 np->tx.req_prod_pvt = i + 1;
558
559 xennet_make_frags(skb, dev, tx);
560 tx->size = skb->len;
561
562 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->tx, notify);
563 if (notify)
564 notify_remote_via_irq(np->netdev->irq);
565
566 u64_stats_update_begin(&stats->syncp);
567 stats->tx_bytes += skb->len;
568 stats->tx_packets++;
569 u64_stats_update_end(&stats->syncp);
570
571 /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
572 xennet_tx_buf_gc(dev);
573
574 if (!netfront_tx_slot_available(np))
575 netif_stop_queue(dev);
576
577 spin_unlock_irq(&np->tx_lock);
578
579 return NETDEV_TX_OK;
580
581 drop:
582 dev->stats.tx_dropped++;
583 dev_kfree_skb(skb);
584 return NETDEV_TX_OK;
585 }
586
587 static int xennet_close(struct net_device *dev)
588 {
589 struct netfront_info *np = netdev_priv(dev);
590 netif_stop_queue(np->netdev);
591 napi_disable(&np->napi);
592 return 0;
593 }
594
595 static void xennet_move_rx_slot(struct netfront_info *np, struct sk_buff *skb,
596 grant_ref_t ref)
597 {
598 int new = xennet_rxidx(np->rx.req_prod_pvt);
599
600 BUG_ON(np->rx_skbs[new]);
601 np->rx_skbs[new] = skb;
602 np->grant_rx_ref[new] = ref;
603 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->id = new;
604 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->gref = ref;
605 np->rx.req_prod_pvt++;
606 }
607
608 static int xennet_get_extras(struct netfront_info *np,
609 struct xen_netif_extra_info *extras,
610 RING_IDX rp)
611
612 {
613 struct xen_netif_extra_info *extra;
614 struct device *dev = &np->netdev->dev;
615 RING_IDX cons = np->rx.rsp_cons;
616 int err = 0;
617
618 do {
619 struct sk_buff *skb;
620 grant_ref_t ref;
621
622 if (unlikely(cons + 1 == rp)) {
623 if (net_ratelimit())
624 dev_warn(dev, "Missing extra info\n");
625 err = -EBADR;
626 break;
627 }
628
629 extra = (struct xen_netif_extra_info *)
630 RING_GET_RESPONSE(&np->rx, ++cons);
631
632 if (unlikely(!extra->type ||
633 extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
634 if (net_ratelimit())
635 dev_warn(dev, "Invalid extra type: %d\n",
636 extra->type);
637 err = -EINVAL;
638 } else {
639 memcpy(&extras[extra->type - 1], extra,
640 sizeof(*extra));
641 }
642
643 skb = xennet_get_rx_skb(np, cons);
644 ref = xennet_get_rx_ref(np, cons);
645 xennet_move_rx_slot(np, skb, ref);
646 } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
647
648 np->rx.rsp_cons = cons;
649 return err;
650 }
651
652 static int xennet_get_responses(struct netfront_info *np,
653 struct netfront_rx_info *rinfo, RING_IDX rp,
654 struct sk_buff_head *list)
655 {
656 struct xen_netif_rx_response *rx = &rinfo->rx;
657 struct xen_netif_extra_info *extras = rinfo->extras;
658 struct device *dev = &np->netdev->dev;
659 RING_IDX cons = np->rx.rsp_cons;
660 struct sk_buff *skb = xennet_get_rx_skb(np, cons);
661 grant_ref_t ref = xennet_get_rx_ref(np, cons);
662 int max = MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD);
663 int frags = 1;
664 int err = 0;
665 unsigned long ret;
666
667 if (rx->flags & XEN_NETRXF_extra_info) {
668 err = xennet_get_extras(np, extras, rp);
669 cons = np->rx.rsp_cons;
670 }
671
672 for (;;) {
673 if (unlikely(rx->status < 0 ||
674 rx->offset + rx->status > PAGE_SIZE)) {
675 if (net_ratelimit())
676 dev_warn(dev, "rx->offset: %x, size: %u\n",
677 rx->offset, rx->status);
678 xennet_move_rx_slot(np, skb, ref);
679 err = -EINVAL;
680 goto next;
681 }
682
683 /*
684 * This definitely indicates a bug, either in this driver or in
685 * the backend driver. In future this should flag the bad
686 * situation to the system controller to reboot the backed.
687 */
688 if (ref == GRANT_INVALID_REF) {
689 if (net_ratelimit())
690 dev_warn(dev, "Bad rx response id %d.\n",
691 rx->id);
692 err = -EINVAL;
693 goto next;
694 }
695
696 ret = gnttab_end_foreign_access_ref(ref, 0);
697 BUG_ON(!ret);
698
699 gnttab_release_grant_reference(&np->gref_rx_head, ref);
700
701 __skb_queue_tail(list, skb);
702
703 next:
704 if (!(rx->flags & XEN_NETRXF_more_data))
705 break;
706
707 if (cons + frags == rp) {
708 if (net_ratelimit())
709 dev_warn(dev, "Need more frags\n");
710 err = -ENOENT;
711 break;
712 }
713
714 rx = RING_GET_RESPONSE(&np->rx, cons + frags);
715 skb = xennet_get_rx_skb(np, cons + frags);
716 ref = xennet_get_rx_ref(np, cons + frags);
717 frags++;
718 }
719
720 if (unlikely(frags > max)) {
721 if (net_ratelimit())
722 dev_warn(dev, "Too many frags\n");
723 err = -E2BIG;
724 }
725
726 if (unlikely(err))
727 np->rx.rsp_cons = cons + frags;
728
729 return err;
730 }
731
732 static int xennet_set_skb_gso(struct sk_buff *skb,
733 struct xen_netif_extra_info *gso)
734 {
735 if (!gso->u.gso.size) {
736 if (net_ratelimit())
737 printk(KERN_WARNING "GSO size must not be zero.\n");
738 return -EINVAL;
739 }
740
741 /* Currently only TCPv4 S.O. is supported. */
742 if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4) {
743 if (net_ratelimit())
744 printk(KERN_WARNING "Bad GSO type %d.\n", gso->u.gso.type);
745 return -EINVAL;
746 }
747
748 skb_shinfo(skb)->gso_size = gso->u.gso.size;
749 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
750
751 /* Header must be checked, and gso_segs computed. */
752 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
753 skb_shinfo(skb)->gso_segs = 0;
754
755 return 0;
756 }
757
758 static RING_IDX xennet_fill_frags(struct netfront_info *np,
759 struct sk_buff *skb,
760 struct sk_buff_head *list)
761 {
762 struct skb_shared_info *shinfo = skb_shinfo(skb);
763 int nr_frags = shinfo->nr_frags;
764 RING_IDX cons = np->rx.rsp_cons;
765 skb_frag_t *frag = shinfo->frags + nr_frags;
766 struct sk_buff *nskb;
767
768 while ((nskb = __skb_dequeue(list))) {
769 struct xen_netif_rx_response *rx =
770 RING_GET_RESPONSE(&np->rx, ++cons);
771
772 frag->page = skb_shinfo(nskb)->frags[0].page;
773 frag->page_offset = rx->offset;
774 frag->size = rx->status;
775
776 skb->data_len += rx->status;
777
778 skb_shinfo(nskb)->nr_frags = 0;
779 kfree_skb(nskb);
780
781 frag++;
782 nr_frags++;
783 }
784
785 shinfo->nr_frags = nr_frags;
786 return cons;
787 }
788
789 static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
790 {
791 struct iphdr *iph;
792 unsigned char *th;
793 int err = -EPROTO;
794 int recalculate_partial_csum = 0;
795
796 /*
797 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
798 * peers can fail to set NETRXF_csum_blank when sending a GSO
799 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
800 * recalculate the partial checksum.
801 */
802 if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
803 struct netfront_info *np = netdev_priv(dev);
804 np->rx_gso_checksum_fixup++;
805 skb->ip_summed = CHECKSUM_PARTIAL;
806 recalculate_partial_csum = 1;
807 }
808
809 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
810 if (skb->ip_summed != CHECKSUM_PARTIAL)
811 return 0;
812
813 if (skb->protocol != htons(ETH_P_IP))
814 goto out;
815
816 iph = (void *)skb->data;
817 th = skb->data + 4 * iph->ihl;
818 if (th >= skb_tail_pointer(skb))
819 goto out;
820
821 skb->csum_start = th - skb->head;
822 switch (iph->protocol) {
823 case IPPROTO_TCP:
824 skb->csum_offset = offsetof(struct tcphdr, check);
825
826 if (recalculate_partial_csum) {
827 struct tcphdr *tcph = (struct tcphdr *)th;
828 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
829 skb->len - iph->ihl*4,
830 IPPROTO_TCP, 0);
831 }
832 break;
833 case IPPROTO_UDP:
834 skb->csum_offset = offsetof(struct udphdr, check);
835
836 if (recalculate_partial_csum) {
837 struct udphdr *udph = (struct udphdr *)th;
838 udph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
839 skb->len - iph->ihl*4,
840 IPPROTO_UDP, 0);
841 }
842 break;
843 default:
844 if (net_ratelimit())
845 printk(KERN_ERR "Attempting to checksum a non-"
846 "TCP/UDP packet, dropping a protocol"
847 " %d packet", iph->protocol);
848 goto out;
849 }
850
851 if ((th + skb->csum_offset + 2) > skb_tail_pointer(skb))
852 goto out;
853
854 err = 0;
855
856 out:
857 return err;
858 }
859
860 static int handle_incoming_queue(struct net_device *dev,
861 struct sk_buff_head *rxq)
862 {
863 struct netfront_info *np = netdev_priv(dev);
864 struct netfront_stats *stats = this_cpu_ptr(np->stats);
865 int packets_dropped = 0;
866 struct sk_buff *skb;
867
868 while ((skb = __skb_dequeue(rxq)) != NULL) {
869 struct page *page = NETFRONT_SKB_CB(skb)->page;
870 void *vaddr = page_address(page);
871 unsigned offset = NETFRONT_SKB_CB(skb)->offset;
872
873 memcpy(skb->data, vaddr + offset,
874 skb_headlen(skb));
875
876 if (page != skb_shinfo(skb)->frags[0].page)
877 __free_page(page);
878
879 /* Ethernet work: Delayed to here as it peeks the header. */
880 skb->protocol = eth_type_trans(skb, dev);
881
882 if (checksum_setup(dev, skb)) {
883 kfree_skb(skb);
884 packets_dropped++;
885 dev->stats.rx_errors++;
886 continue;
887 }
888
889 u64_stats_update_begin(&stats->syncp);
890 stats->rx_packets++;
891 stats->rx_bytes += skb->len;
892 u64_stats_update_end(&stats->syncp);
893
894 /* Pass it up. */
895 netif_receive_skb(skb);
896 }
897
898 return packets_dropped;
899 }
900
901 static int xennet_poll(struct napi_struct *napi, int budget)
902 {
903 struct netfront_info *np = container_of(napi, struct netfront_info, napi);
904 struct net_device *dev = np->netdev;
905 struct sk_buff *skb;
906 struct netfront_rx_info rinfo;
907 struct xen_netif_rx_response *rx = &rinfo.rx;
908 struct xen_netif_extra_info *extras = rinfo.extras;
909 RING_IDX i, rp;
910 int work_done;
911 struct sk_buff_head rxq;
912 struct sk_buff_head errq;
913 struct sk_buff_head tmpq;
914 unsigned long flags;
915 unsigned int len;
916 int err;
917
918 spin_lock(&np->rx_lock);
919
920 skb_queue_head_init(&rxq);
921 skb_queue_head_init(&errq);
922 skb_queue_head_init(&tmpq);
923
924 rp = np->rx.sring->rsp_prod;
925 rmb(); /* Ensure we see queued responses up to 'rp'. */
926
927 i = np->rx.rsp_cons;
928 work_done = 0;
929 while ((i != rp) && (work_done < budget)) {
930 memcpy(rx, RING_GET_RESPONSE(&np->rx, i), sizeof(*rx));
931 memset(extras, 0, sizeof(rinfo.extras));
932
933 err = xennet_get_responses(np, &rinfo, rp, &tmpq);
934
935 if (unlikely(err)) {
936 err:
937 while ((skb = __skb_dequeue(&tmpq)))
938 __skb_queue_tail(&errq, skb);
939 dev->stats.rx_errors++;
940 i = np->rx.rsp_cons;
941 continue;
942 }
943
944 skb = __skb_dequeue(&tmpq);
945
946 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
947 struct xen_netif_extra_info *gso;
948 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
949
950 if (unlikely(xennet_set_skb_gso(skb, gso))) {
951 __skb_queue_head(&tmpq, skb);
952 np->rx.rsp_cons += skb_queue_len(&tmpq);
953 goto err;
954 }
955 }
956
957 NETFRONT_SKB_CB(skb)->page = skb_shinfo(skb)->frags[0].page;
958 NETFRONT_SKB_CB(skb)->offset = rx->offset;
959
960 len = rx->status;
961 if (len > RX_COPY_THRESHOLD)
962 len = RX_COPY_THRESHOLD;
963 skb_put(skb, len);
964
965 if (rx->status > len) {
966 skb_shinfo(skb)->frags[0].page_offset =
967 rx->offset + len;
968 skb_shinfo(skb)->frags[0].size = rx->status - len;
969 skb->data_len = rx->status - len;
970 } else {
971 skb_shinfo(skb)->frags[0].page = NULL;
972 skb_shinfo(skb)->nr_frags = 0;
973 }
974
975 i = xennet_fill_frags(np, skb, &tmpq);
976
977 /*
978 * Truesize approximates the size of true data plus
979 * any supervisor overheads. Adding hypervisor
980 * overheads has been shown to significantly reduce
981 * achievable bandwidth with the default receive
982 * buffer size. It is therefore not wise to account
983 * for it here.
984 *
985 * After alloc_skb(RX_COPY_THRESHOLD), truesize is set
986 * to RX_COPY_THRESHOLD + the supervisor
987 * overheads. Here, we add the size of the data pulled
988 * in xennet_fill_frags().
989 *
990 * We also adjust for any unused space in the main
991 * data area by subtracting (RX_COPY_THRESHOLD -
992 * len). This is especially important with drivers
993 * which split incoming packets into header and data,
994 * using only 66 bytes of the main data area (see the
995 * e1000 driver for example.) On such systems,
996 * without this last adjustement, our achievable
997 * receive throughout using the standard receive
998 * buffer size was cut by 25%(!!!).
999 */
1000 skb->truesize += skb->data_len - (RX_COPY_THRESHOLD - len);
1001 skb->len += skb->data_len;
1002
1003 if (rx->flags & XEN_NETRXF_csum_blank)
1004 skb->ip_summed = CHECKSUM_PARTIAL;
1005 else if (rx->flags & XEN_NETRXF_data_validated)
1006 skb->ip_summed = CHECKSUM_UNNECESSARY;
1007
1008 __skb_queue_tail(&rxq, skb);
1009
1010 np->rx.rsp_cons = ++i;
1011 work_done++;
1012 }
1013
1014 __skb_queue_purge(&errq);
1015
1016 work_done -= handle_incoming_queue(dev, &rxq);
1017
1018 /* If we get a callback with very few responses, reduce fill target. */
1019 /* NB. Note exponential increase, linear decrease. */
1020 if (((np->rx.req_prod_pvt - np->rx.sring->rsp_prod) >
1021 ((3*np->rx_target) / 4)) &&
1022 (--np->rx_target < np->rx_min_target))
1023 np->rx_target = np->rx_min_target;
1024
1025 xennet_alloc_rx_buffers(dev);
1026
1027 if (work_done < budget) {
1028 int more_to_do = 0;
1029
1030 local_irq_save(flags);
1031
1032 RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, more_to_do);
1033 if (!more_to_do)
1034 __napi_complete(napi);
1035
1036 local_irq_restore(flags);
1037 }
1038
1039 spin_unlock(&np->rx_lock);
1040
1041 return work_done;
1042 }
1043
1044 static int xennet_change_mtu(struct net_device *dev, int mtu)
1045 {
1046 int max = xennet_can_sg(dev) ? 65535 - ETH_HLEN : ETH_DATA_LEN;
1047
1048 if (mtu > max)
1049 return -EINVAL;
1050 dev->mtu = mtu;
1051 return 0;
1052 }
1053
1054 static struct rtnl_link_stats64 *xennet_get_stats64(struct net_device *dev,
1055 struct rtnl_link_stats64 *tot)
1056 {
1057 struct netfront_info *np = netdev_priv(dev);
1058 int cpu;
1059
1060 for_each_possible_cpu(cpu) {
1061 struct netfront_stats *stats = per_cpu_ptr(np->stats, cpu);
1062 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1063 unsigned int start;
1064
1065 do {
1066 start = u64_stats_fetch_begin_bh(&stats->syncp);
1067
1068 rx_packets = stats->rx_packets;
1069 tx_packets = stats->tx_packets;
1070 rx_bytes = stats->rx_bytes;
1071 tx_bytes = stats->tx_bytes;
1072 } while (u64_stats_fetch_retry_bh(&stats->syncp, start));
1073
1074 tot->rx_packets += rx_packets;
1075 tot->tx_packets += tx_packets;
1076 tot->rx_bytes += rx_bytes;
1077 tot->tx_bytes += tx_bytes;
1078 }
1079
1080 tot->rx_errors = dev->stats.rx_errors;
1081 tot->tx_dropped = dev->stats.tx_dropped;
1082
1083 return tot;
1084 }
1085
1086 static void xennet_release_tx_bufs(struct netfront_info *np)
1087 {
1088 struct sk_buff *skb;
1089 int i;
1090
1091 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1092 /* Skip over entries which are actually freelist references */
1093 if (skb_entry_is_link(&np->tx_skbs[i]))
1094 continue;
1095
1096 skb = np->tx_skbs[i].skb;
1097 gnttab_end_foreign_access_ref(np->grant_tx_ref[i],
1098 GNTMAP_readonly);
1099 gnttab_release_grant_reference(&np->gref_tx_head,
1100 np->grant_tx_ref[i]);
1101 np->grant_tx_ref[i] = GRANT_INVALID_REF;
1102 add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, i);
1103 dev_kfree_skb_irq(skb);
1104 }
1105 }
1106
1107 static void xennet_release_rx_bufs(struct netfront_info *np)
1108 {
1109 struct mmu_update *mmu = np->rx_mmu;
1110 struct multicall_entry *mcl = np->rx_mcl;
1111 struct sk_buff_head free_list;
1112 struct sk_buff *skb;
1113 unsigned long mfn;
1114 int xfer = 0, noxfer = 0, unused = 0;
1115 int id, ref;
1116
1117 dev_warn(&np->netdev->dev, "%s: fix me for copying receiver.\n",
1118 __func__);
1119 return;
1120
1121 skb_queue_head_init(&free_list);
1122
1123 spin_lock_bh(&np->rx_lock);
1124
1125 for (id = 0; id < NET_RX_RING_SIZE; id++) {
1126 ref = np->grant_rx_ref[id];
1127 if (ref == GRANT_INVALID_REF) {
1128 unused++;
1129 continue;
1130 }
1131
1132 skb = np->rx_skbs[id];
1133 mfn = gnttab_end_foreign_transfer_ref(ref);
1134 gnttab_release_grant_reference(&np->gref_rx_head, ref);
1135 np->grant_rx_ref[id] = GRANT_INVALID_REF;
1136
1137 if (0 == mfn) {
1138 skb_shinfo(skb)->nr_frags = 0;
1139 dev_kfree_skb(skb);
1140 noxfer++;
1141 continue;
1142 }
1143
1144 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1145 /* Remap the page. */
1146 struct page *page = skb_shinfo(skb)->frags[0].page;
1147 unsigned long pfn = page_to_pfn(page);
1148 void *vaddr = page_address(page);
1149
1150 MULTI_update_va_mapping(mcl, (unsigned long)vaddr,
1151 mfn_pte(mfn, PAGE_KERNEL),
1152 0);
1153 mcl++;
1154 mmu->ptr = ((u64)mfn << PAGE_SHIFT)
1155 | MMU_MACHPHYS_UPDATE;
1156 mmu->val = pfn;
1157 mmu++;
1158
1159 set_phys_to_machine(pfn, mfn);
1160 }
1161 __skb_queue_tail(&free_list, skb);
1162 xfer++;
1163 }
1164
1165 dev_info(&np->netdev->dev, "%s: %d xfer, %d noxfer, %d unused\n",
1166 __func__, xfer, noxfer, unused);
1167
1168 if (xfer) {
1169 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1170 /* Do all the remapping work and M2P updates. */
1171 MULTI_mmu_update(mcl, np->rx_mmu, mmu - np->rx_mmu,
1172 NULL, DOMID_SELF);
1173 mcl++;
1174 HYPERVISOR_multicall(np->rx_mcl, mcl - np->rx_mcl);
1175 }
1176 }
1177
1178 __skb_queue_purge(&free_list);
1179
1180 spin_unlock_bh(&np->rx_lock);
1181 }
1182
1183 static void xennet_uninit(struct net_device *dev)
1184 {
1185 struct netfront_info *np = netdev_priv(dev);
1186 xennet_release_tx_bufs(np);
1187 xennet_release_rx_bufs(np);
1188 gnttab_free_grant_references(np->gref_tx_head);
1189 gnttab_free_grant_references(np->gref_rx_head);
1190 }
1191
1192 static u32 xennet_fix_features(struct net_device *dev, u32 features)
1193 {
1194 struct netfront_info *np = netdev_priv(dev);
1195 int val;
1196
1197 if (features & NETIF_F_SG) {
1198 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, "feature-sg",
1199 "%d", &val) < 0)
1200 val = 0;
1201
1202 if (!val)
1203 features &= ~NETIF_F_SG;
1204 }
1205
1206 if (features & NETIF_F_TSO) {
1207 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1208 "feature-gso-tcpv4", "%d", &val) < 0)
1209 val = 0;
1210
1211 if (!val)
1212 features &= ~NETIF_F_TSO;
1213 }
1214
1215 return features;
1216 }
1217
1218 static int xennet_set_features(struct net_device *dev, u32 features)
1219 {
1220 if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1221 netdev_info(dev, "Reducing MTU because no SG offload");
1222 dev->mtu = ETH_DATA_LEN;
1223 }
1224
1225 return 0;
1226 }
1227
1228 static const struct net_device_ops xennet_netdev_ops = {
1229 .ndo_open = xennet_open,
1230 .ndo_uninit = xennet_uninit,
1231 .ndo_stop = xennet_close,
1232 .ndo_start_xmit = xennet_start_xmit,
1233 .ndo_change_mtu = xennet_change_mtu,
1234 .ndo_get_stats64 = xennet_get_stats64,
1235 .ndo_set_mac_address = eth_mac_addr,
1236 .ndo_validate_addr = eth_validate_addr,
1237 .ndo_fix_features = xennet_fix_features,
1238 .ndo_set_features = xennet_set_features,
1239 };
1240
1241 static struct net_device * __devinit xennet_create_dev(struct xenbus_device *dev)
1242 {
1243 int i, err;
1244 struct net_device *netdev;
1245 struct netfront_info *np;
1246
1247 netdev = alloc_etherdev(sizeof(struct netfront_info));
1248 if (!netdev) {
1249 printk(KERN_WARNING "%s> alloc_etherdev failed.\n",
1250 __func__);
1251 return ERR_PTR(-ENOMEM);
1252 }
1253
1254 np = netdev_priv(netdev);
1255 np->xbdev = dev;
1256
1257 spin_lock_init(&np->tx_lock);
1258 spin_lock_init(&np->rx_lock);
1259
1260 skb_queue_head_init(&np->rx_batch);
1261 np->rx_target = RX_DFL_MIN_TARGET;
1262 np->rx_min_target = RX_DFL_MIN_TARGET;
1263 np->rx_max_target = RX_MAX_TARGET;
1264
1265 init_timer(&np->rx_refill_timer);
1266 np->rx_refill_timer.data = (unsigned long)netdev;
1267 np->rx_refill_timer.function = rx_refill_timeout;
1268
1269 err = -ENOMEM;
1270 np->stats = alloc_percpu(struct netfront_stats);
1271 if (np->stats == NULL)
1272 goto exit;
1273
1274 /* Initialise tx_skbs as a free chain containing every entry. */
1275 np->tx_skb_freelist = 0;
1276 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1277 skb_entry_set_link(&np->tx_skbs[i], i+1);
1278 np->grant_tx_ref[i] = GRANT_INVALID_REF;
1279 }
1280
1281 /* Clear out rx_skbs */
1282 for (i = 0; i < NET_RX_RING_SIZE; i++) {
1283 np->rx_skbs[i] = NULL;
1284 np->grant_rx_ref[i] = GRANT_INVALID_REF;
1285 }
1286
1287 /* A grant for every tx ring slot */
1288 if (gnttab_alloc_grant_references(TX_MAX_TARGET,
1289 &np->gref_tx_head) < 0) {
1290 printk(KERN_ALERT "#### netfront can't alloc tx grant refs\n");
1291 err = -ENOMEM;
1292 goto exit_free_stats;
1293 }
1294 /* A grant for every rx ring slot */
1295 if (gnttab_alloc_grant_references(RX_MAX_TARGET,
1296 &np->gref_rx_head) < 0) {
1297 printk(KERN_ALERT "#### netfront can't alloc rx grant refs\n");
1298 err = -ENOMEM;
1299 goto exit_free_tx;
1300 }
1301
1302 netdev->netdev_ops = &xennet_netdev_ops;
1303
1304 netif_napi_add(netdev, &np->napi, xennet_poll, 64);
1305 netdev->features = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1306 NETIF_F_GSO_ROBUST;
1307 netdev->hw_features = NETIF_F_IP_CSUM | NETIF_F_SG | NETIF_F_TSO;
1308
1309 /*
1310 * Assume that all hw features are available for now. This set
1311 * will be adjusted by the call to netdev_update_features() in
1312 * xennet_connect() which is the earliest point where we can
1313 * negotiate with the backend regarding supported features.
1314 */
1315 netdev->features |= netdev->hw_features;
1316
1317 SET_ETHTOOL_OPS(netdev, &xennet_ethtool_ops);
1318 SET_NETDEV_DEV(netdev, &dev->dev);
1319
1320 np->netdev = netdev;
1321
1322 netif_carrier_off(netdev);
1323
1324 return netdev;
1325
1326 exit_free_tx:
1327 gnttab_free_grant_references(np->gref_tx_head);
1328 exit_free_stats:
1329 free_percpu(np->stats);
1330 exit:
1331 free_netdev(netdev);
1332 return ERR_PTR(err);
1333 }
1334
1335 /**
1336 * Entry point to this code when a new device is created. Allocate the basic
1337 * structures and the ring buffers for communication with the backend, and
1338 * inform the backend of the appropriate details for those.
1339 */
1340 static int __devinit netfront_probe(struct xenbus_device *dev,
1341 const struct xenbus_device_id *id)
1342 {
1343 int err;
1344 struct net_device *netdev;
1345 struct netfront_info *info;
1346
1347 netdev = xennet_create_dev(dev);
1348 if (IS_ERR(netdev)) {
1349 err = PTR_ERR(netdev);
1350 xenbus_dev_fatal(dev, err, "creating netdev");
1351 return err;
1352 }
1353
1354 info = netdev_priv(netdev);
1355 dev_set_drvdata(&dev->dev, info);
1356
1357 err = register_netdev(info->netdev);
1358 if (err) {
1359 printk(KERN_WARNING "%s: register_netdev err=%d\n",
1360 __func__, err);
1361 goto fail;
1362 }
1363
1364 err = xennet_sysfs_addif(info->netdev);
1365 if (err) {
1366 unregister_netdev(info->netdev);
1367 printk(KERN_WARNING "%s: add sysfs failed err=%d\n",
1368 __func__, err);
1369 goto fail;
1370 }
1371
1372 return 0;
1373
1374 fail:
1375 free_netdev(netdev);
1376 dev_set_drvdata(&dev->dev, NULL);
1377 return err;
1378 }
1379
1380 static void xennet_end_access(int ref, void *page)
1381 {
1382 /* This frees the page as a side-effect */
1383 if (ref != GRANT_INVALID_REF)
1384 gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1385 }
1386
1387 static void xennet_disconnect_backend(struct netfront_info *info)
1388 {
1389 /* Stop old i/f to prevent errors whilst we rebuild the state. */
1390 spin_lock_bh(&info->rx_lock);
1391 spin_lock_irq(&info->tx_lock);
1392 netif_carrier_off(info->netdev);
1393 spin_unlock_irq(&info->tx_lock);
1394 spin_unlock_bh(&info->rx_lock);
1395
1396 if (info->netdev->irq)
1397 unbind_from_irqhandler(info->netdev->irq, info->netdev);
1398 info->evtchn = info->netdev->irq = 0;
1399
1400 /* End access and free the pages */
1401 xennet_end_access(info->tx_ring_ref, info->tx.sring);
1402 xennet_end_access(info->rx_ring_ref, info->rx.sring);
1403
1404 info->tx_ring_ref = GRANT_INVALID_REF;
1405 info->rx_ring_ref = GRANT_INVALID_REF;
1406 info->tx.sring = NULL;
1407 info->rx.sring = NULL;
1408 }
1409
1410 /**
1411 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1412 * driver restart. We tear down our netif structure and recreate it, but
1413 * leave the device-layer structures intact so that this is transparent to the
1414 * rest of the kernel.
1415 */
1416 static int netfront_resume(struct xenbus_device *dev)
1417 {
1418 struct netfront_info *info = dev_get_drvdata(&dev->dev);
1419
1420 dev_dbg(&dev->dev, "%s\n", dev->nodename);
1421
1422 xennet_disconnect_backend(info);
1423 return 0;
1424 }
1425
1426 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1427 {
1428 char *s, *e, *macstr;
1429 int i;
1430
1431 macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1432 if (IS_ERR(macstr))
1433 return PTR_ERR(macstr);
1434
1435 for (i = 0; i < ETH_ALEN; i++) {
1436 mac[i] = simple_strtoul(s, &e, 16);
1437 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1438 kfree(macstr);
1439 return -ENOENT;
1440 }
1441 s = e+1;
1442 }
1443
1444 kfree(macstr);
1445 return 0;
1446 }
1447
1448 static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1449 {
1450 struct net_device *dev = dev_id;
1451 struct netfront_info *np = netdev_priv(dev);
1452 unsigned long flags;
1453
1454 spin_lock_irqsave(&np->tx_lock, flags);
1455
1456 if (likely(netif_carrier_ok(dev))) {
1457 xennet_tx_buf_gc(dev);
1458 /* Under tx_lock: protects access to rx shared-ring indexes. */
1459 if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx))
1460 napi_schedule(&np->napi);
1461 }
1462
1463 spin_unlock_irqrestore(&np->tx_lock, flags);
1464
1465 return IRQ_HANDLED;
1466 }
1467
1468 static int setup_netfront(struct xenbus_device *dev, struct netfront_info *info)
1469 {
1470 struct xen_netif_tx_sring *txs;
1471 struct xen_netif_rx_sring *rxs;
1472 int err;
1473 struct net_device *netdev = info->netdev;
1474
1475 info->tx_ring_ref = GRANT_INVALID_REF;
1476 info->rx_ring_ref = GRANT_INVALID_REF;
1477 info->rx.sring = NULL;
1478 info->tx.sring = NULL;
1479 netdev->irq = 0;
1480
1481 err = xen_net_read_mac(dev, netdev->dev_addr);
1482 if (err) {
1483 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
1484 goto fail;
1485 }
1486
1487 txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1488 if (!txs) {
1489 err = -ENOMEM;
1490 xenbus_dev_fatal(dev, err, "allocating tx ring page");
1491 goto fail;
1492 }
1493 SHARED_RING_INIT(txs);
1494 FRONT_RING_INIT(&info->tx, txs, PAGE_SIZE);
1495
1496 err = xenbus_grant_ring(dev, virt_to_mfn(txs));
1497 if (err < 0) {
1498 free_page((unsigned long)txs);
1499 goto fail;
1500 }
1501
1502 info->tx_ring_ref = err;
1503 rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1504 if (!rxs) {
1505 err = -ENOMEM;
1506 xenbus_dev_fatal(dev, err, "allocating rx ring page");
1507 goto fail;
1508 }
1509 SHARED_RING_INIT(rxs);
1510 FRONT_RING_INIT(&info->rx, rxs, PAGE_SIZE);
1511
1512 err = xenbus_grant_ring(dev, virt_to_mfn(rxs));
1513 if (err < 0) {
1514 free_page((unsigned long)rxs);
1515 goto fail;
1516 }
1517 info->rx_ring_ref = err;
1518
1519 err = xenbus_alloc_evtchn(dev, &info->evtchn);
1520 if (err)
1521 goto fail;
1522
1523 err = bind_evtchn_to_irqhandler(info->evtchn, xennet_interrupt,
1524 0, netdev->name, netdev);
1525 if (err < 0)
1526 goto fail;
1527 netdev->irq = err;
1528 return 0;
1529
1530 fail:
1531 return err;
1532 }
1533
1534 /* Common code used when first setting up, and when resuming. */
1535 static int talk_to_netback(struct xenbus_device *dev,
1536 struct netfront_info *info)
1537 {
1538 const char *message;
1539 struct xenbus_transaction xbt;
1540 int err;
1541
1542 /* Create shared ring, alloc event channel. */
1543 err = setup_netfront(dev, info);
1544 if (err)
1545 goto out;
1546
1547 again:
1548 err = xenbus_transaction_start(&xbt);
1549 if (err) {
1550 xenbus_dev_fatal(dev, err, "starting transaction");
1551 goto destroy_ring;
1552 }
1553
1554 err = xenbus_printf(xbt, dev->nodename, "tx-ring-ref", "%u",
1555 info->tx_ring_ref);
1556 if (err) {
1557 message = "writing tx ring-ref";
1558 goto abort_transaction;
1559 }
1560 err = xenbus_printf(xbt, dev->nodename, "rx-ring-ref", "%u",
1561 info->rx_ring_ref);
1562 if (err) {
1563 message = "writing rx ring-ref";
1564 goto abort_transaction;
1565 }
1566 err = xenbus_printf(xbt, dev->nodename,
1567 "event-channel", "%u", info->evtchn);
1568 if (err) {
1569 message = "writing event-channel";
1570 goto abort_transaction;
1571 }
1572
1573 err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
1574 1);
1575 if (err) {
1576 message = "writing request-rx-copy";
1577 goto abort_transaction;
1578 }
1579
1580 err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
1581 if (err) {
1582 message = "writing feature-rx-notify";
1583 goto abort_transaction;
1584 }
1585
1586 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
1587 if (err) {
1588 message = "writing feature-sg";
1589 goto abort_transaction;
1590 }
1591
1592 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
1593 if (err) {
1594 message = "writing feature-gso-tcpv4";
1595 goto abort_transaction;
1596 }
1597
1598 err = xenbus_transaction_end(xbt, 0);
1599 if (err) {
1600 if (err == -EAGAIN)
1601 goto again;
1602 xenbus_dev_fatal(dev, err, "completing transaction");
1603 goto destroy_ring;
1604 }
1605
1606 return 0;
1607
1608 abort_transaction:
1609 xenbus_transaction_end(xbt, 1);
1610 xenbus_dev_fatal(dev, err, "%s", message);
1611 destroy_ring:
1612 xennet_disconnect_backend(info);
1613 out:
1614 return err;
1615 }
1616
1617 static int xennet_connect(struct net_device *dev)
1618 {
1619 struct netfront_info *np = netdev_priv(dev);
1620 int i, requeue_idx, err;
1621 struct sk_buff *skb;
1622 grant_ref_t ref;
1623 struct xen_netif_rx_request *req;
1624 unsigned int feature_rx_copy;
1625
1626 err = xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1627 "feature-rx-copy", "%u", &feature_rx_copy);
1628 if (err != 1)
1629 feature_rx_copy = 0;
1630
1631 if (!feature_rx_copy) {
1632 dev_info(&dev->dev,
1633 "backend does not support copying receive path\n");
1634 return -ENODEV;
1635 }
1636
1637 err = talk_to_netback(np->xbdev, np);
1638 if (err)
1639 return err;
1640
1641 rtnl_lock();
1642 netdev_update_features(dev);
1643 rtnl_unlock();
1644
1645 spin_lock_bh(&np->rx_lock);
1646 spin_lock_irq(&np->tx_lock);
1647
1648 /* Step 1: Discard all pending TX packet fragments. */
1649 xennet_release_tx_bufs(np);
1650
1651 /* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */
1652 for (requeue_idx = 0, i = 0; i < NET_RX_RING_SIZE; i++) {
1653 if (!np->rx_skbs[i])
1654 continue;
1655
1656 skb = np->rx_skbs[requeue_idx] = xennet_get_rx_skb(np, i);
1657 ref = np->grant_rx_ref[requeue_idx] = xennet_get_rx_ref(np, i);
1658 req = RING_GET_REQUEST(&np->rx, requeue_idx);
1659
1660 gnttab_grant_foreign_access_ref(
1661 ref, np->xbdev->otherend_id,
1662 pfn_to_mfn(page_to_pfn(skb_shinfo(skb)->
1663 frags->page)),
1664 0);
1665 req->gref = ref;
1666 req->id = requeue_idx;
1667
1668 requeue_idx++;
1669 }
1670
1671 np->rx.req_prod_pvt = requeue_idx;
1672
1673 /*
1674 * Step 3: All public and private state should now be sane. Get
1675 * ready to start sending and receiving packets and give the driver
1676 * domain a kick because we've probably just requeued some
1677 * packets.
1678 */
1679 netif_carrier_on(np->netdev);
1680 notify_remote_via_irq(np->netdev->irq);
1681 xennet_tx_buf_gc(dev);
1682 xennet_alloc_rx_buffers(dev);
1683
1684 spin_unlock_irq(&np->tx_lock);
1685 spin_unlock_bh(&np->rx_lock);
1686
1687 return 0;
1688 }
1689
1690 /**
1691 * Callback received when the backend's state changes.
1692 */
1693 static void netback_changed(struct xenbus_device *dev,
1694 enum xenbus_state backend_state)
1695 {
1696 struct netfront_info *np = dev_get_drvdata(&dev->dev);
1697 struct net_device *netdev = np->netdev;
1698
1699 dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
1700
1701 switch (backend_state) {
1702 case XenbusStateInitialising:
1703 case XenbusStateInitialised:
1704 case XenbusStateReconfiguring:
1705 case XenbusStateReconfigured:
1706 case XenbusStateConnected:
1707 case XenbusStateUnknown:
1708 case XenbusStateClosed:
1709 break;
1710
1711 case XenbusStateInitWait:
1712 if (dev->state != XenbusStateInitialising)
1713 break;
1714 if (xennet_connect(netdev) != 0)
1715 break;
1716 xenbus_switch_state(dev, XenbusStateConnected);
1717 netif_notify_peers(netdev);
1718 break;
1719
1720 case XenbusStateClosing:
1721 xenbus_frontend_closed(dev);
1722 break;
1723 }
1724 }
1725
1726 static const struct xennet_stat {
1727 char name[ETH_GSTRING_LEN];
1728 u16 offset;
1729 } xennet_stats[] = {
1730 {
1731 "rx_gso_checksum_fixup",
1732 offsetof(struct netfront_info, rx_gso_checksum_fixup)
1733 },
1734 };
1735
1736 static int xennet_get_sset_count(struct net_device *dev, int string_set)
1737 {
1738 switch (string_set) {
1739 case ETH_SS_STATS:
1740 return ARRAY_SIZE(xennet_stats);
1741 default:
1742 return -EINVAL;
1743 }
1744 }
1745
1746 static void xennet_get_ethtool_stats(struct net_device *dev,
1747 struct ethtool_stats *stats, u64 * data)
1748 {
1749 void *np = netdev_priv(dev);
1750 int i;
1751
1752 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
1753 data[i] = *(unsigned long *)(np + xennet_stats[i].offset);
1754 }
1755
1756 static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
1757 {
1758 int i;
1759
1760 switch (stringset) {
1761 case ETH_SS_STATS:
1762 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
1763 memcpy(data + i * ETH_GSTRING_LEN,
1764 xennet_stats[i].name, ETH_GSTRING_LEN);
1765 break;
1766 }
1767 }
1768
1769 static const struct ethtool_ops xennet_ethtool_ops =
1770 {
1771 .get_link = ethtool_op_get_link,
1772
1773 .get_sset_count = xennet_get_sset_count,
1774 .get_ethtool_stats = xennet_get_ethtool_stats,
1775 .get_strings = xennet_get_strings,
1776 };
1777
1778 #ifdef CONFIG_SYSFS
1779 static ssize_t show_rxbuf_min(struct device *dev,
1780 struct device_attribute *attr, char *buf)
1781 {
1782 struct net_device *netdev = to_net_dev(dev);
1783 struct netfront_info *info = netdev_priv(netdev);
1784
1785 return sprintf(buf, "%u\n", info->rx_min_target);
1786 }
1787
1788 static ssize_t store_rxbuf_min(struct device *dev,
1789 struct device_attribute *attr,
1790 const char *buf, size_t len)
1791 {
1792 struct net_device *netdev = to_net_dev(dev);
1793 struct netfront_info *np = netdev_priv(netdev);
1794 char *endp;
1795 unsigned long target;
1796
1797 if (!capable(CAP_NET_ADMIN))
1798 return -EPERM;
1799
1800 target = simple_strtoul(buf, &endp, 0);
1801 if (endp == buf)
1802 return -EBADMSG;
1803
1804 if (target < RX_MIN_TARGET)
1805 target = RX_MIN_TARGET;
1806 if (target > RX_MAX_TARGET)
1807 target = RX_MAX_TARGET;
1808
1809 spin_lock_bh(&np->rx_lock);
1810 if (target > np->rx_max_target)
1811 np->rx_max_target = target;
1812 np->rx_min_target = target;
1813 if (target > np->rx_target)
1814 np->rx_target = target;
1815
1816 xennet_alloc_rx_buffers(netdev);
1817
1818 spin_unlock_bh(&np->rx_lock);
1819 return len;
1820 }
1821
1822 static ssize_t show_rxbuf_max(struct device *dev,
1823 struct device_attribute *attr, char *buf)
1824 {
1825 struct net_device *netdev = to_net_dev(dev);
1826 struct netfront_info *info = netdev_priv(netdev);
1827
1828 return sprintf(buf, "%u\n", info->rx_max_target);
1829 }
1830
1831 static ssize_t store_rxbuf_max(struct device *dev,
1832 struct device_attribute *attr,
1833 const char *buf, size_t len)
1834 {
1835 struct net_device *netdev = to_net_dev(dev);
1836 struct netfront_info *np = netdev_priv(netdev);
1837 char *endp;
1838 unsigned long target;
1839
1840 if (!capable(CAP_NET_ADMIN))
1841 return -EPERM;
1842
1843 target = simple_strtoul(buf, &endp, 0);
1844 if (endp == buf)
1845 return -EBADMSG;
1846
1847 if (target < RX_MIN_TARGET)
1848 target = RX_MIN_TARGET;
1849 if (target > RX_MAX_TARGET)
1850 target = RX_MAX_TARGET;
1851
1852 spin_lock_bh(&np->rx_lock);
1853 if (target < np->rx_min_target)
1854 np->rx_min_target = target;
1855 np->rx_max_target = target;
1856 if (target < np->rx_target)
1857 np->rx_target = target;
1858
1859 xennet_alloc_rx_buffers(netdev);
1860
1861 spin_unlock_bh(&np->rx_lock);
1862 return len;
1863 }
1864
1865 static ssize_t show_rxbuf_cur(struct device *dev,
1866 struct device_attribute *attr, char *buf)
1867 {
1868 struct net_device *netdev = to_net_dev(dev);
1869 struct netfront_info *info = netdev_priv(netdev);
1870
1871 return sprintf(buf, "%u\n", info->rx_target);
1872 }
1873
1874 static struct device_attribute xennet_attrs[] = {
1875 __ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf_min, store_rxbuf_min),
1876 __ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf_max, store_rxbuf_max),
1877 __ATTR(rxbuf_cur, S_IRUGO, show_rxbuf_cur, NULL),
1878 };
1879
1880 static int xennet_sysfs_addif(struct net_device *netdev)
1881 {
1882 int i;
1883 int err;
1884
1885 for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++) {
1886 err = device_create_file(&netdev->dev,
1887 &xennet_attrs[i]);
1888 if (err)
1889 goto fail;
1890 }
1891 return 0;
1892
1893 fail:
1894 while (--i >= 0)
1895 device_remove_file(&netdev->dev, &xennet_attrs[i]);
1896 return err;
1897 }
1898
1899 static void xennet_sysfs_delif(struct net_device *netdev)
1900 {
1901 int i;
1902
1903 for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++)
1904 device_remove_file(&netdev->dev, &xennet_attrs[i]);
1905 }
1906
1907 #endif /* CONFIG_SYSFS */
1908
1909 static struct xenbus_device_id netfront_ids[] = {
1910 { "vif" },
1911 { "" }
1912 };
1913
1914
1915 static int __devexit xennet_remove(struct xenbus_device *dev)
1916 {
1917 struct netfront_info *info = dev_get_drvdata(&dev->dev);
1918
1919 dev_dbg(&dev->dev, "%s\n", dev->nodename);
1920
1921 unregister_netdev(info->netdev);
1922
1923 xennet_disconnect_backend(info);
1924
1925 del_timer_sync(&info->rx_refill_timer);
1926
1927 xennet_sysfs_delif(info->netdev);
1928
1929 free_percpu(info->stats);
1930
1931 free_netdev(info->netdev);
1932
1933 return 0;
1934 }
1935
1936 static struct xenbus_driver netfront_driver = {
1937 .name = "vif",
1938 .owner = THIS_MODULE,
1939 .ids = netfront_ids,
1940 .probe = netfront_probe,
1941 .remove = __devexit_p(xennet_remove),
1942 .resume = netfront_resume,
1943 .otherend_changed = netback_changed,
1944 };
1945
1946 static int __init netif_init(void)
1947 {
1948 if (!xen_domain())
1949 return -ENODEV;
1950
1951 if (xen_initial_domain())
1952 return 0;
1953
1954 printk(KERN_INFO "Initialising Xen virtual ethernet driver.\n");
1955
1956 return xenbus_register_frontend(&netfront_driver);
1957 }
1958 module_init(netif_init);
1959
1960
1961 static void __exit netif_exit(void)
1962 {
1963 if (xen_initial_domain())
1964 return;
1965
1966 xenbus_unregister_driver(&netfront_driver);
1967 }
1968 module_exit(netif_exit);
1969
1970 MODULE_DESCRIPTION("Xen virtual network device frontend");
1971 MODULE_LICENSE("GPL");
1972 MODULE_ALIAS("xen:vif");
1973 MODULE_ALIAS("xennet");
x86 "subsystem" repository