?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
x86: make 32 bit to use sparse_irq
[linux-2.6/linux-2.6-openrd.git] / drivers / net / xen-netfront.c
blob3c3dd403f5ddd9f58bed9f632baad6b72409e0ee
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 <net/ip.h>
44
45 #include <xen/xenbus.h>
46 #include <xen/events.h>
47 #include <xen/page.h>
48 #include <xen/grant_table.h>
49
50 #include <xen/interface/io/netif.h>
51 #include <xen/interface/memory.h>
52 #include <xen/interface/grant_table.h>
53
54 static struct ethtool_ops xennet_ethtool_ops;
55
56 struct netfront_cb {
57 struct page *page;
58 unsigned offset;
59 };
60
61 #define NETFRONT_SKB_CB(skb) ((struct netfront_cb *)((skb)->cb))
62
63 #define RX_COPY_THRESHOLD 256
64
65 #define GRANT_INVALID_REF 0
66
67 #define NET_TX_RING_SIZE __RING_SIZE((struct xen_netif_tx_sring *)0, PAGE_SIZE)
68 #define NET_RX_RING_SIZE __RING_SIZE((struct xen_netif_rx_sring *)0, PAGE_SIZE)
69 #define TX_MAX_TARGET min_t(int, NET_RX_RING_SIZE, 256)
70
71 struct netfront_info {
72 struct list_head list;
73 struct net_device *netdev;
74
75 struct napi_struct napi;
76
77 unsigned int evtchn;
78 struct xenbus_device *xbdev;
79
80 spinlock_t tx_lock;
81 struct xen_netif_tx_front_ring tx;
82 int tx_ring_ref;
83
84 /*
85 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
86 * are linked from tx_skb_freelist through skb_entry.link.
87 *
88 * NB. Freelist index entries are always going to be less than
89 * PAGE_OFFSET, whereas pointers to skbs will always be equal or
90 * greater than PAGE_OFFSET: we use this property to distinguish
91 * them.
92 */
93 union skb_entry {
94 struct sk_buff *skb;
95 unsigned long link;
96 } tx_skbs[NET_TX_RING_SIZE];
97 grant_ref_t gref_tx_head;
98 grant_ref_t grant_tx_ref[NET_TX_RING_SIZE];
99 unsigned tx_skb_freelist;
100
101 spinlock_t rx_lock ____cacheline_aligned_in_smp;
102 struct xen_netif_rx_front_ring rx;
103 int rx_ring_ref;
104
105 /* Receive-ring batched refills. */
106 #define RX_MIN_TARGET 8
107 #define RX_DFL_MIN_TARGET 64
108 #define RX_MAX_TARGET min_t(int, NET_RX_RING_SIZE, 256)
109 unsigned rx_min_target, rx_max_target, rx_target;
110 struct sk_buff_head rx_batch;
111
112 struct timer_list rx_refill_timer;
113
114 struct sk_buff *rx_skbs[NET_RX_RING_SIZE];
115 grant_ref_t gref_rx_head;
116 grant_ref_t grant_rx_ref[NET_RX_RING_SIZE];
117
118 unsigned long rx_pfn_array[NET_RX_RING_SIZE];
119 struct multicall_entry rx_mcl[NET_RX_RING_SIZE+1];
120 struct mmu_update rx_mmu[NET_RX_RING_SIZE];
121 };
122
123 struct netfront_rx_info {
124 struct xen_netif_rx_response rx;
125 struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
126 };
127
128 static void skb_entry_set_link(union skb_entry *list, unsigned short id)
129 {
130 list->link = id;
131 }
132
133 static int skb_entry_is_link(const union skb_entry *list)
134 {
135 BUILD_BUG_ON(sizeof(list->skb) != sizeof(list->link));
136 return ((unsigned long)list->skb < PAGE_OFFSET);
137 }
138
139 /*
140 * Access macros for acquiring freeing slots in tx_skbs[].
141 */
142
143 static void add_id_to_freelist(unsigned *head, union skb_entry *list,
144 unsigned short id)
145 {
146 skb_entry_set_link(&list[id], *head);
147 *head = id;
148 }
149
150 static unsigned short get_id_from_freelist(unsigned *head,
151 union skb_entry *list)
152 {
153 unsigned int id = *head;
154 *head = list[id].link;
155 return id;
156 }
157
158 static int xennet_rxidx(RING_IDX idx)
159 {
160 return idx & (NET_RX_RING_SIZE - 1);
161 }
162
163 static struct sk_buff *xennet_get_rx_skb(struct netfront_info *np,
164 RING_IDX ri)
165 {
166 int i = xennet_rxidx(ri);
167 struct sk_buff *skb = np->rx_skbs[i];
168 np->rx_skbs[i] = NULL;
169 return skb;
170 }
171
172 static grant_ref_t xennet_get_rx_ref(struct netfront_info *np,
173 RING_IDX ri)
174 {
175 int i = xennet_rxidx(ri);
176 grant_ref_t ref = np->grant_rx_ref[i];
177 np->grant_rx_ref[i] = GRANT_INVALID_REF;
178 return ref;
179 }
180
181 #ifdef CONFIG_SYSFS
182 static int xennet_sysfs_addif(struct net_device *netdev);
183 static void xennet_sysfs_delif(struct net_device *netdev);
184 #else /* !CONFIG_SYSFS */
185 #define xennet_sysfs_addif(dev) (0)
186 #define xennet_sysfs_delif(dev) do { } while (0)
187 #endif
188
189 static int xennet_can_sg(struct net_device *dev)
190 {
191 return dev->features & NETIF_F_SG;
192 }
193
194
195 static void rx_refill_timeout(unsigned long data)
196 {
197 struct net_device *dev = (struct net_device *)data;
198 struct netfront_info *np = netdev_priv(dev);
199 netif_rx_schedule(dev, &np->napi);
200 }
201
202 static int netfront_tx_slot_available(struct netfront_info *np)
203 {
204 return ((np->tx.req_prod_pvt - np->tx.rsp_cons) <
205 (TX_MAX_TARGET - MAX_SKB_FRAGS - 2));
206 }
207
208 static void xennet_maybe_wake_tx(struct net_device *dev)
209 {
210 struct netfront_info *np = netdev_priv(dev);
211
212 if (unlikely(netif_queue_stopped(dev)) &&
213 netfront_tx_slot_available(np) &&
214 likely(netif_running(dev)))
215 netif_wake_queue(dev);
216 }
217
218 static void xennet_alloc_rx_buffers(struct net_device *dev)
219 {
220 unsigned short id;
221 struct netfront_info *np = netdev_priv(dev);
222 struct sk_buff *skb;
223 struct page *page;
224 int i, batch_target, notify;
225 RING_IDX req_prod = np->rx.req_prod_pvt;
226 grant_ref_t ref;
227 unsigned long pfn;
228 void *vaddr;
229 struct xen_netif_rx_request *req;
230
231 if (unlikely(!netif_carrier_ok(dev)))
232 return;
233
234 /*
235 * Allocate skbuffs greedily, even though we batch updates to the
236 * receive ring. This creates a less bursty demand on the memory
237 * allocator, so should reduce the chance of failed allocation requests
238 * both for ourself and for other kernel subsystems.
239 */
240 batch_target = np->rx_target - (req_prod - np->rx.rsp_cons);
241 for (i = skb_queue_len(&np->rx_batch); i < batch_target; i++) {
242 skb = __netdev_alloc_skb(dev, RX_COPY_THRESHOLD,
243 GFP_ATOMIC | __GFP_NOWARN);
244 if (unlikely(!skb))
245 goto no_skb;
246
247 page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
248 if (!page) {
249 kfree_skb(skb);
250 no_skb:
251 /* Any skbuffs queued for refill? Force them out. */
252 if (i != 0)
253 goto refill;
254 /* Could not allocate any skbuffs. Try again later. */
255 mod_timer(&np->rx_refill_timer,
256 jiffies + (HZ/10));
257 break;
258 }
259
260 skb_shinfo(skb)->frags[0].page = page;
261 skb_shinfo(skb)->nr_frags = 1;
262 __skb_queue_tail(&np->rx_batch, skb);
263 }
264
265 /* Is the batch large enough to be worthwhile? */
266 if (i < (np->rx_target/2)) {
267 if (req_prod > np->rx.sring->req_prod)
268 goto push;
269 return;
270 }
271
272 /* Adjust our fill target if we risked running out of buffers. */
273 if (((req_prod - np->rx.sring->rsp_prod) < (np->rx_target / 4)) &&
274 ((np->rx_target *= 2) > np->rx_max_target))
275 np->rx_target = np->rx_max_target;
276
277 refill:
278 for (i = 0; ; i++) {
279 skb = __skb_dequeue(&np->rx_batch);
280 if (skb == NULL)
281 break;
282
283 skb->dev = dev;
284
285 id = xennet_rxidx(req_prod + i);
286
287 BUG_ON(np->rx_skbs[id]);
288 np->rx_skbs[id] = skb;
289
290 ref = gnttab_claim_grant_reference(&np->gref_rx_head);
291 BUG_ON((signed short)ref < 0);
292 np->grant_rx_ref[id] = ref;
293
294 pfn = page_to_pfn(skb_shinfo(skb)->frags[0].page);
295 vaddr = page_address(skb_shinfo(skb)->frags[0].page);
296
297 req = RING_GET_REQUEST(&np->rx, req_prod + i);
298 gnttab_grant_foreign_access_ref(ref,
299 np->xbdev->otherend_id,
300 pfn_to_mfn(pfn),
301 0);
302
303 req->id = id;
304 req->gref = ref;
305 }
306
307 wmb(); /* barrier so backend seens requests */
308
309 /* Above is a suitable barrier to ensure backend will see requests. */
310 np->rx.req_prod_pvt = req_prod + i;
311 push:
312 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->rx, notify);
313 if (notify)
314 notify_remote_via_irq(np->netdev->irq);
315 }
316
317 static int xennet_open(struct net_device *dev)
318 {
319 struct netfront_info *np = netdev_priv(dev);
320
321 napi_enable(&np->napi);
322
323 spin_lock_bh(&np->rx_lock);
324 if (netif_carrier_ok(dev)) {
325 xennet_alloc_rx_buffers(dev);
326 np->rx.sring->rsp_event = np->rx.rsp_cons + 1;
327 if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx))
328 netif_rx_schedule(dev, &np->napi);
329 }
330 spin_unlock_bh(&np->rx_lock);
331
332 netif_start_queue(dev);
333
334 return 0;
335 }
336
337 static void xennet_tx_buf_gc(struct net_device *dev)
338 {
339 RING_IDX cons, prod;
340 unsigned short id;
341 struct netfront_info *np = netdev_priv(dev);
342 struct sk_buff *skb;
343
344 BUG_ON(!netif_carrier_ok(dev));
345
346 do {
347 prod = np->tx.sring->rsp_prod;
348 rmb(); /* Ensure we see responses up to 'rp'. */
349
350 for (cons = np->tx.rsp_cons; cons != prod; cons++) {
351 struct xen_netif_tx_response *txrsp;
352
353 txrsp = RING_GET_RESPONSE(&np->tx, cons);
354 if (txrsp->status == NETIF_RSP_NULL)
355 continue;
356
357 id = txrsp->id;
358 skb = np->tx_skbs[id].skb;
359 if (unlikely(gnttab_query_foreign_access(
360 np->grant_tx_ref[id]) != 0)) {
361 printk(KERN_ALERT "xennet_tx_buf_gc: warning "
362 "-- grant still in use by backend "
363 "domain.\n");
364 BUG();
365 }
366 gnttab_end_foreign_access_ref(
367 np->grant_tx_ref[id], GNTMAP_readonly);
368 gnttab_release_grant_reference(
369 &np->gref_tx_head, np->grant_tx_ref[id]);
370 np->grant_tx_ref[id] = GRANT_INVALID_REF;
371 add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, id);
372 dev_kfree_skb_irq(skb);
373 }
374
375 np->tx.rsp_cons = prod;
376
377 /*
378 * Set a new event, then check for race with update of tx_cons.
379 * Note that it is essential to schedule a callback, no matter
380 * how few buffers are pending. Even if there is space in the
381 * transmit ring, higher layers may be blocked because too much
382 * data is outstanding: in such cases notification from Xen is
383 * likely to be the only kick that we'll get.
384 */
385 np->tx.sring->rsp_event =
386 prod + ((np->tx.sring->req_prod - prod) >> 1) + 1;
387 mb(); /* update shared area */
388 } while ((cons == prod) && (prod != np->tx.sring->rsp_prod));
389
390 xennet_maybe_wake_tx(dev);
391 }
392
393 static void xennet_make_frags(struct sk_buff *skb, struct net_device *dev,
394 struct xen_netif_tx_request *tx)
395 {
396 struct netfront_info *np = netdev_priv(dev);
397 char *data = skb->data;
398 unsigned long mfn;
399 RING_IDX prod = np->tx.req_prod_pvt;
400 int frags = skb_shinfo(skb)->nr_frags;
401 unsigned int offset = offset_in_page(data);
402 unsigned int len = skb_headlen(skb);
403 unsigned int id;
404 grant_ref_t ref;
405 int i;
406
407 /* While the header overlaps a page boundary (including being
408 larger than a page), split it it into page-sized chunks. */
409 while (len > PAGE_SIZE - offset) {
410 tx->size = PAGE_SIZE - offset;
411 tx->flags |= NETTXF_more_data;
412 len -= tx->size;
413 data += tx->size;
414 offset = 0;
415
416 id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
417 np->tx_skbs[id].skb = skb_get(skb);
418 tx = RING_GET_REQUEST(&np->tx, prod++);
419 tx->id = id;
420 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
421 BUG_ON((signed short)ref < 0);
422
423 mfn = virt_to_mfn(data);
424 gnttab_grant_foreign_access_ref(ref, np->xbdev->otherend_id,
425 mfn, GNTMAP_readonly);
426
427 tx->gref = np->grant_tx_ref[id] = ref;
428 tx->offset = offset;
429 tx->size = len;
430 tx->flags = 0;
431 }
432
433 /* Grant backend access to each skb fragment page. */
434 for (i = 0; i < frags; i++) {
435 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
436
437 tx->flags |= NETTXF_more_data;
438
439 id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
440 np->tx_skbs[id].skb = skb_get(skb);
441 tx = RING_GET_REQUEST(&np->tx, prod++);
442 tx->id = id;
443 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
444 BUG_ON((signed short)ref < 0);
445
446 mfn = pfn_to_mfn(page_to_pfn(frag->page));
447 gnttab_grant_foreign_access_ref(ref, np->xbdev->otherend_id,
448 mfn, GNTMAP_readonly);
449
450 tx->gref = np->grant_tx_ref[id] = ref;
451 tx->offset = frag->page_offset;
452 tx->size = frag->size;
453 tx->flags = 0;
454 }
455
456 np->tx.req_prod_pvt = prod;
457 }
458
459 static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
460 {
461 unsigned short id;
462 struct netfront_info *np = netdev_priv(dev);
463 struct xen_netif_tx_request *tx;
464 struct xen_netif_extra_info *extra;
465 char *data = skb->data;
466 RING_IDX i;
467 grant_ref_t ref;
468 unsigned long mfn;
469 int notify;
470 int frags = skb_shinfo(skb)->nr_frags;
471 unsigned int offset = offset_in_page(data);
472 unsigned int len = skb_headlen(skb);
473
474 frags += (offset + len + PAGE_SIZE - 1) / PAGE_SIZE;
475 if (unlikely(frags > MAX_SKB_FRAGS + 1)) {
476 printk(KERN_ALERT "xennet: skb rides the rocket: %d frags\n",
477 frags);
478 dump_stack();
479 goto drop;
480 }
481
482 spin_lock_irq(&np->tx_lock);
483
484 if (unlikely(!netif_carrier_ok(dev) ||
485 (frags > 1 && !xennet_can_sg(dev)) ||
486 netif_needs_gso(dev, skb))) {
487 spin_unlock_irq(&np->tx_lock);
488 goto drop;
489 }
490
491 i = np->tx.req_prod_pvt;
492
493 id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
494 np->tx_skbs[id].skb = skb;
495
496 tx = RING_GET_REQUEST(&np->tx, i);
497
498 tx->id = id;
499 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
500 BUG_ON((signed short)ref < 0);
501 mfn = virt_to_mfn(data);
502 gnttab_grant_foreign_access_ref(
503 ref, np->xbdev->otherend_id, mfn, GNTMAP_readonly);
504 tx->gref = np->grant_tx_ref[id] = ref;
505 tx->offset = offset;
506 tx->size = len;
507 extra = NULL;
508
509 tx->flags = 0;
510 if (skb->ip_summed == CHECKSUM_PARTIAL)
511 /* local packet? */
512 tx->flags |= NETTXF_csum_blank | NETTXF_data_validated;
513 else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
514 /* remote but checksummed. */
515 tx->flags |= NETTXF_data_validated;
516
517 if (skb_shinfo(skb)->gso_size) {
518 struct xen_netif_extra_info *gso;
519
520 gso = (struct xen_netif_extra_info *)
521 RING_GET_REQUEST(&np->tx, ++i);
522
523 if (extra)
524 extra->flags |= XEN_NETIF_EXTRA_FLAG_MORE;
525 else
526 tx->flags |= NETTXF_extra_info;
527
528 gso->u.gso.size = skb_shinfo(skb)->gso_size;
529 gso->u.gso.type = XEN_NETIF_GSO_TYPE_TCPV4;
530 gso->u.gso.pad = 0;
531 gso->u.gso.features = 0;
532
533 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
534 gso->flags = 0;
535 extra = gso;
536 }
537
538 np->tx.req_prod_pvt = i + 1;
539
540 xennet_make_frags(skb, dev, tx);
541 tx->size = skb->len;
542
543 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->tx, notify);
544 if (notify)
545 notify_remote_via_irq(np->netdev->irq);
546
547 dev->stats.tx_bytes += skb->len;
548 dev->stats.tx_packets++;
549
550 /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
551 xennet_tx_buf_gc(dev);
552
553 if (!netfront_tx_slot_available(np))
554 netif_stop_queue(dev);
555
556 spin_unlock_irq(&np->tx_lock);
557
558 return 0;
559
560 drop:
561 dev->stats.tx_dropped++;
562 dev_kfree_skb(skb);
563 return 0;
564 }
565
566 static int xennet_close(struct net_device *dev)
567 {
568 struct netfront_info *np = netdev_priv(dev);
569 netif_stop_queue(np->netdev);
570 napi_disable(&np->napi);
571 return 0;
572 }
573
574 static void xennet_move_rx_slot(struct netfront_info *np, struct sk_buff *skb,
575 grant_ref_t ref)
576 {
577 int new = xennet_rxidx(np->rx.req_prod_pvt);
578
579 BUG_ON(np->rx_skbs[new]);
580 np->rx_skbs[new] = skb;
581 np->grant_rx_ref[new] = ref;
582 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->id = new;
583 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->gref = ref;
584 np->rx.req_prod_pvt++;
585 }
586
587 static int xennet_get_extras(struct netfront_info *np,
588 struct xen_netif_extra_info *extras,
589 RING_IDX rp)
590
591 {
592 struct xen_netif_extra_info *extra;
593 struct device *dev = &np->netdev->dev;
594 RING_IDX cons = np->rx.rsp_cons;
595 int err = 0;
596
597 do {
598 struct sk_buff *skb;
599 grant_ref_t ref;
600
601 if (unlikely(cons + 1 == rp)) {
602 if (net_ratelimit())
603 dev_warn(dev, "Missing extra info\n");
604 err = -EBADR;
605 break;
606 }
607
608 extra = (struct xen_netif_extra_info *)
609 RING_GET_RESPONSE(&np->rx, ++cons);
610
611 if (unlikely(!extra->type ||
612 extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
613 if (net_ratelimit())
614 dev_warn(dev, "Invalid extra type: %d\n",
615 extra->type);
616 err = -EINVAL;
617 } else {
618 memcpy(&extras[extra->type - 1], extra,
619 sizeof(*extra));
620 }
621
622 skb = xennet_get_rx_skb(np, cons);
623 ref = xennet_get_rx_ref(np, cons);
624 xennet_move_rx_slot(np, skb, ref);
625 } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
626
627 np->rx.rsp_cons = cons;
628 return err;
629 }
630
631 static int xennet_get_responses(struct netfront_info *np,
632 struct netfront_rx_info *rinfo, RING_IDX rp,
633 struct sk_buff_head *list)
634 {
635 struct xen_netif_rx_response *rx = &rinfo->rx;
636 struct xen_netif_extra_info *extras = rinfo->extras;
637 struct device *dev = &np->netdev->dev;
638 RING_IDX cons = np->rx.rsp_cons;
639 struct sk_buff *skb = xennet_get_rx_skb(np, cons);
640 grant_ref_t ref = xennet_get_rx_ref(np, cons);
641 int max = MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD);
642 int frags = 1;
643 int err = 0;
644 unsigned long ret;
645
646 if (rx->flags & NETRXF_extra_info) {
647 err = xennet_get_extras(np, extras, rp);
648 cons = np->rx.rsp_cons;
649 }
650
651 for (;;) {
652 if (unlikely(rx->status < 0 ||
653 rx->offset + rx->status > PAGE_SIZE)) {
654 if (net_ratelimit())
655 dev_warn(dev, "rx->offset: %x, size: %u\n",
656 rx->offset, rx->status);
657 xennet_move_rx_slot(np, skb, ref);
658 err = -EINVAL;
659 goto next;
660 }
661
662 /*
663 * This definitely indicates a bug, either in this driver or in
664 * the backend driver. In future this should flag the bad
665 * situation to the system controller to reboot the backed.
666 */
667 if (ref == GRANT_INVALID_REF) {
668 if (net_ratelimit())
669 dev_warn(dev, "Bad rx response id %d.\n",
670 rx->id);
671 err = -EINVAL;
672 goto next;
673 }
674
675 ret = gnttab_end_foreign_access_ref(ref, 0);
676 BUG_ON(!ret);
677
678 gnttab_release_grant_reference(&np->gref_rx_head, ref);
679
680 __skb_queue_tail(list, skb);
681
682 next:
683 if (!(rx->flags & NETRXF_more_data))
684 break;
685
686 if (cons + frags == rp) {
687 if (net_ratelimit())
688 dev_warn(dev, "Need more frags\n");
689 err = -ENOENT;
690 break;
691 }
692
693 rx = RING_GET_RESPONSE(&np->rx, cons + frags);
694 skb = xennet_get_rx_skb(np, cons + frags);
695 ref = xennet_get_rx_ref(np, cons + frags);
696 frags++;
697 }
698
699 if (unlikely(frags > max)) {
700 if (net_ratelimit())
701 dev_warn(dev, "Too many frags\n");
702 err = -E2BIG;
703 }
704
705 if (unlikely(err))
706 np->rx.rsp_cons = cons + frags;
707
708 return err;
709 }
710
711 static int xennet_set_skb_gso(struct sk_buff *skb,
712 struct xen_netif_extra_info *gso)
713 {
714 if (!gso->u.gso.size) {
715 if (net_ratelimit())
716 printk(KERN_WARNING "GSO size must not be zero.\n");
717 return -EINVAL;
718 }
719
720 /* Currently only TCPv4 S.O. is supported. */
721 if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4) {
722 if (net_ratelimit())
723 printk(KERN_WARNING "Bad GSO type %d.\n", gso->u.gso.type);
724 return -EINVAL;
725 }
726
727 skb_shinfo(skb)->gso_size = gso->u.gso.size;
728 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
729
730 /* Header must be checked, and gso_segs computed. */
731 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
732 skb_shinfo(skb)->gso_segs = 0;
733
734 return 0;
735 }
736
737 static RING_IDX xennet_fill_frags(struct netfront_info *np,
738 struct sk_buff *skb,
739 struct sk_buff_head *list)
740 {
741 struct skb_shared_info *shinfo = skb_shinfo(skb);
742 int nr_frags = shinfo->nr_frags;
743 RING_IDX cons = np->rx.rsp_cons;
744 skb_frag_t *frag = shinfo->frags + nr_frags;
745 struct sk_buff *nskb;
746
747 while ((nskb = __skb_dequeue(list))) {
748 struct xen_netif_rx_response *rx =
749 RING_GET_RESPONSE(&np->rx, ++cons);
750
751 frag->page = skb_shinfo(nskb)->frags[0].page;
752 frag->page_offset = rx->offset;
753 frag->size = rx->status;
754
755 skb->data_len += rx->status;
756
757 skb_shinfo(nskb)->nr_frags = 0;
758 kfree_skb(nskb);
759
760 frag++;
761 nr_frags++;
762 }
763
764 shinfo->nr_frags = nr_frags;
765 return cons;
766 }
767
768 static int skb_checksum_setup(struct sk_buff *skb)
769 {
770 struct iphdr *iph;
771 unsigned char *th;
772 int err = -EPROTO;
773
774 if (skb->protocol != htons(ETH_P_IP))
775 goto out;
776
777 iph = (void *)skb->data;
778 th = skb->data + 4 * iph->ihl;
779 if (th >= skb_tail_pointer(skb))
780 goto out;
781
782 skb->csum_start = th - skb->head;
783 switch (iph->protocol) {
784 case IPPROTO_TCP:
785 skb->csum_offset = offsetof(struct tcphdr, check);
786 break;
787 case IPPROTO_UDP:
788 skb->csum_offset = offsetof(struct udphdr, check);
789 break;
790 default:
791 if (net_ratelimit())
792 printk(KERN_ERR "Attempting to checksum a non-"
793 "TCP/UDP packet, dropping a protocol"
794 " %d packet", iph->protocol);
795 goto out;
796 }
797
798 if ((th + skb->csum_offset + 2) > skb_tail_pointer(skb))
799 goto out;
800
801 err = 0;
802
803 out:
804 return err;
805 }
806
807 static int handle_incoming_queue(struct net_device *dev,
808 struct sk_buff_head *rxq)
809 {
810 int packets_dropped = 0;
811 struct sk_buff *skb;
812
813 while ((skb = __skb_dequeue(rxq)) != NULL) {
814 struct page *page = NETFRONT_SKB_CB(skb)->page;
815 void *vaddr = page_address(page);
816 unsigned offset = NETFRONT_SKB_CB(skb)->offset;
817
818 memcpy(skb->data, vaddr + offset,
819 skb_headlen(skb));
820
821 if (page != skb_shinfo(skb)->frags[0].page)
822 __free_page(page);
823
824 /* Ethernet work: Delayed to here as it peeks the header. */
825 skb->protocol = eth_type_trans(skb, dev);
826
827 if (skb->ip_summed == CHECKSUM_PARTIAL) {
828 if (skb_checksum_setup(skb)) {
829 kfree_skb(skb);
830 packets_dropped++;
831 dev->stats.rx_errors++;
832 continue;
833 }
834 }
835
836 dev->stats.rx_packets++;
837 dev->stats.rx_bytes += skb->len;
838
839 /* Pass it up. */
840 netif_receive_skb(skb);
841 dev->last_rx = jiffies;
842 }
843
844 return packets_dropped;
845 }
846
847 static int xennet_poll(struct napi_struct *napi, int budget)
848 {
849 struct netfront_info *np = container_of(napi, struct netfront_info, napi);
850 struct net_device *dev = np->netdev;
851 struct sk_buff *skb;
852 struct netfront_rx_info rinfo;
853 struct xen_netif_rx_response *rx = &rinfo.rx;
854 struct xen_netif_extra_info *extras = rinfo.extras;
855 RING_IDX i, rp;
856 int work_done;
857 struct sk_buff_head rxq;
858 struct sk_buff_head errq;
859 struct sk_buff_head tmpq;
860 unsigned long flags;
861 unsigned int len;
862 int err;
863
864 spin_lock(&np->rx_lock);
865
866 skb_queue_head_init(&rxq);
867 skb_queue_head_init(&errq);
868 skb_queue_head_init(&tmpq);
869
870 rp = np->rx.sring->rsp_prod;
871 rmb(); /* Ensure we see queued responses up to 'rp'. */
872
873 i = np->rx.rsp_cons;
874 work_done = 0;
875 while ((i != rp) && (work_done < budget)) {
876 memcpy(rx, RING_GET_RESPONSE(&np->rx, i), sizeof(*rx));
877 memset(extras, 0, sizeof(rinfo.extras));
878
879 err = xennet_get_responses(np, &rinfo, rp, &tmpq);
880
881 if (unlikely(err)) {
882 err:
883 while ((skb = __skb_dequeue(&tmpq)))
884 __skb_queue_tail(&errq, skb);
885 dev->stats.rx_errors++;
886 i = np->rx.rsp_cons;
887 continue;
888 }
889
890 skb = __skb_dequeue(&tmpq);
891
892 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
893 struct xen_netif_extra_info *gso;
894 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
895
896 if (unlikely(xennet_set_skb_gso(skb, gso))) {
897 __skb_queue_head(&tmpq, skb);
898 np->rx.rsp_cons += skb_queue_len(&tmpq);
899 goto err;
900 }
901 }
902
903 NETFRONT_SKB_CB(skb)->page = skb_shinfo(skb)->frags[0].page;
904 NETFRONT_SKB_CB(skb)->offset = rx->offset;
905
906 len = rx->status;
907 if (len > RX_COPY_THRESHOLD)
908 len = RX_COPY_THRESHOLD;
909 skb_put(skb, len);
910
911 if (rx->status > len) {
912 skb_shinfo(skb)->frags[0].page_offset =
913 rx->offset + len;
914 skb_shinfo(skb)->frags[0].size = rx->status - len;
915 skb->data_len = rx->status - len;
916 } else {
917 skb_shinfo(skb)->frags[0].page = NULL;
918 skb_shinfo(skb)->nr_frags = 0;
919 }
920
921 i = xennet_fill_frags(np, skb, &tmpq);
922
923 /*
924 * Truesize approximates the size of true data plus
925 * any supervisor overheads. Adding hypervisor
926 * overheads has been shown to significantly reduce
927 * achievable bandwidth with the default receive
928 * buffer size. It is therefore not wise to account
929 * for it here.
930 *
931 * After alloc_skb(RX_COPY_THRESHOLD), truesize is set
932 * to RX_COPY_THRESHOLD + the supervisor
933 * overheads. Here, we add the size of the data pulled
934 * in xennet_fill_frags().
935 *
936 * We also adjust for any unused space in the main
937 * data area by subtracting (RX_COPY_THRESHOLD -
938 * len). This is especially important with drivers
939 * which split incoming packets into header and data,
940 * using only 66 bytes of the main data area (see the
941 * e1000 driver for example.) On such systems,
942 * without this last adjustement, our achievable
943 * receive throughout using the standard receive
944 * buffer size was cut by 25%(!!!).
945 */
946 skb->truesize += skb->data_len - (RX_COPY_THRESHOLD - len);
947 skb->len += skb->data_len;
948
949 if (rx->flags & NETRXF_csum_blank)
950 skb->ip_summed = CHECKSUM_PARTIAL;
951 else if (rx->flags & NETRXF_data_validated)
952 skb->ip_summed = CHECKSUM_UNNECESSARY;
953
954 __skb_queue_tail(&rxq, skb);
955
956 np->rx.rsp_cons = ++i;
957 work_done++;
958 }
959
960 __skb_queue_purge(&errq);
961
962 work_done -= handle_incoming_queue(dev, &rxq);
963
964 /* If we get a callback with very few responses, reduce fill target. */
965 /* NB. Note exponential increase, linear decrease. */
966 if (((np->rx.req_prod_pvt - np->rx.sring->rsp_prod) >
967 ((3*np->rx_target) / 4)) &&
968 (--np->rx_target < np->rx_min_target))
969 np->rx_target = np->rx_min_target;
970
971 xennet_alloc_rx_buffers(dev);
972
973 if (work_done < budget) {
974 int more_to_do = 0;
975
976 local_irq_save(flags);
977
978 RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, more_to_do);
979 if (!more_to_do)
980 __netif_rx_complete(dev, napi);
981
982 local_irq_restore(flags);
983 }
984
985 spin_unlock(&np->rx_lock);
986
987 return work_done;
988 }
989
990 static int xennet_change_mtu(struct net_device *dev, int mtu)
991 {
992 int max = xennet_can_sg(dev) ? 65535 - ETH_HLEN : ETH_DATA_LEN;
993
994 if (mtu > max)
995 return -EINVAL;
996 dev->mtu = mtu;
997 return 0;
998 }
999
1000 static void xennet_release_tx_bufs(struct netfront_info *np)
1001 {
1002 struct sk_buff *skb;
1003 int i;
1004
1005 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1006 /* Skip over entries which are actually freelist references */
1007 if (skb_entry_is_link(&np->tx_skbs[i]))
1008 continue;
1009
1010 skb = np->tx_skbs[i].skb;
1011 gnttab_end_foreign_access_ref(np->grant_tx_ref[i],
1012 GNTMAP_readonly);
1013 gnttab_release_grant_reference(&np->gref_tx_head,
1014 np->grant_tx_ref[i]);
1015 np->grant_tx_ref[i] = GRANT_INVALID_REF;
1016 add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, i);
1017 dev_kfree_skb_irq(skb);
1018 }
1019 }
1020
1021 static void xennet_release_rx_bufs(struct netfront_info *np)
1022 {
1023 struct mmu_update *mmu = np->rx_mmu;
1024 struct multicall_entry *mcl = np->rx_mcl;
1025 struct sk_buff_head free_list;
1026 struct sk_buff *skb;
1027 unsigned long mfn;
1028 int xfer = 0, noxfer = 0, unused = 0;
1029 int id, ref;
1030
1031 dev_warn(&np->netdev->dev, "%s: fix me for copying receiver.\n",
1032 __func__);
1033 return;
1034
1035 skb_queue_head_init(&free_list);
1036
1037 spin_lock_bh(&np->rx_lock);
1038
1039 for (id = 0; id < NET_RX_RING_SIZE; id++) {
1040 ref = np->grant_rx_ref[id];
1041 if (ref == GRANT_INVALID_REF) {
1042 unused++;
1043 continue;
1044 }
1045
1046 skb = np->rx_skbs[id];
1047 mfn = gnttab_end_foreign_transfer_ref(ref);
1048 gnttab_release_grant_reference(&np->gref_rx_head, ref);
1049 np->grant_rx_ref[id] = GRANT_INVALID_REF;
1050
1051 if (0 == mfn) {
1052 skb_shinfo(skb)->nr_frags = 0;
1053 dev_kfree_skb(skb);
1054 noxfer++;
1055 continue;
1056 }
1057
1058 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1059 /* Remap the page. */
1060 struct page *page = skb_shinfo(skb)->frags[0].page;
1061 unsigned long pfn = page_to_pfn(page);
1062 void *vaddr = page_address(page);
1063
1064 MULTI_update_va_mapping(mcl, (unsigned long)vaddr,
1065 mfn_pte(mfn, PAGE_KERNEL),
1066 0);
1067 mcl++;
1068 mmu->ptr = ((u64)mfn << PAGE_SHIFT)
1069 | MMU_MACHPHYS_UPDATE;
1070 mmu->val = pfn;
1071 mmu++;
1072
1073 set_phys_to_machine(pfn, mfn);
1074 }
1075 __skb_queue_tail(&free_list, skb);
1076 xfer++;
1077 }
1078
1079 dev_info(&np->netdev->dev, "%s: %d xfer, %d noxfer, %d unused\n",
1080 __func__, xfer, noxfer, unused);
1081
1082 if (xfer) {
1083 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1084 /* Do all the remapping work and M2P updates. */
1085 MULTI_mmu_update(mcl, np->rx_mmu, mmu - np->rx_mmu,
1086 NULL, DOMID_SELF);
1087 mcl++;
1088 HYPERVISOR_multicall(np->rx_mcl, mcl - np->rx_mcl);
1089 }
1090 }
1091
1092 __skb_queue_purge(&free_list);
1093
1094 spin_unlock_bh(&np->rx_lock);
1095 }
1096
1097 static void xennet_uninit(struct net_device *dev)
1098 {
1099 struct netfront_info *np = netdev_priv(dev);
1100 xennet_release_tx_bufs(np);
1101 xennet_release_rx_bufs(np);
1102 gnttab_free_grant_references(np->gref_tx_head);
1103 gnttab_free_grant_references(np->gref_rx_head);
1104 }
1105
1106 static struct net_device * __devinit xennet_create_dev(struct xenbus_device *dev)
1107 {
1108 int i, err;
1109 struct net_device *netdev;
1110 struct netfront_info *np;
1111
1112 netdev = alloc_etherdev(sizeof(struct netfront_info));
1113 if (!netdev) {
1114 printk(KERN_WARNING "%s> alloc_etherdev failed.\n",
1115 __func__);
1116 return ERR_PTR(-ENOMEM);
1117 }
1118
1119 np = netdev_priv(netdev);
1120 np->xbdev = dev;
1121
1122 spin_lock_init(&np->tx_lock);
1123 spin_lock_init(&np->rx_lock);
1124
1125 skb_queue_head_init(&np->rx_batch);
1126 np->rx_target = RX_DFL_MIN_TARGET;
1127 np->rx_min_target = RX_DFL_MIN_TARGET;
1128 np->rx_max_target = RX_MAX_TARGET;
1129
1130 init_timer(&np->rx_refill_timer);
1131 np->rx_refill_timer.data = (unsigned long)netdev;
1132 np->rx_refill_timer.function = rx_refill_timeout;
1133
1134 /* Initialise tx_skbs as a free chain containing every entry. */
1135 np->tx_skb_freelist = 0;
1136 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1137 skb_entry_set_link(&np->tx_skbs[i], i+1);
1138 np->grant_tx_ref[i] = GRANT_INVALID_REF;
1139 }
1140
1141 /* Clear out rx_skbs */
1142 for (i = 0; i < NET_RX_RING_SIZE; i++) {
1143 np->rx_skbs[i] = NULL;
1144 np->grant_rx_ref[i] = GRANT_INVALID_REF;
1145 }
1146
1147 /* A grant for every tx ring slot */
1148 if (gnttab_alloc_grant_references(TX_MAX_TARGET,
1149 &np->gref_tx_head) < 0) {
1150 printk(KERN_ALERT "#### netfront can't alloc tx grant refs\n");
1151 err = -ENOMEM;
1152 goto exit;
1153 }
1154 /* A grant for every rx ring slot */
1155 if (gnttab_alloc_grant_references(RX_MAX_TARGET,
1156 &np->gref_rx_head) < 0) {
1157 printk(KERN_ALERT "#### netfront can't alloc rx grant refs\n");
1158 err = -ENOMEM;
1159 goto exit_free_tx;
1160 }
1161
1162 netdev->open = xennet_open;
1163 netdev->hard_start_xmit = xennet_start_xmit;
1164 netdev->stop = xennet_close;
1165 netif_napi_add(netdev, &np->napi, xennet_poll, 64);
1166 netdev->uninit = xennet_uninit;
1167 netdev->change_mtu = xennet_change_mtu;
1168 netdev->features = NETIF_F_IP_CSUM;
1169
1170 SET_ETHTOOL_OPS(netdev, &xennet_ethtool_ops);
1171 SET_NETDEV_DEV(netdev, &dev->dev);
1172
1173 np->netdev = netdev;
1174
1175 netif_carrier_off(netdev);
1176
1177 return netdev;
1178
1179 exit_free_tx:
1180 gnttab_free_grant_references(np->gref_tx_head);
1181 exit:
1182 free_netdev(netdev);
1183 return ERR_PTR(err);
1184 }
1185
1186 /**
1187 * Entry point to this code when a new device is created. Allocate the basic
1188 * structures and the ring buffers for communication with the backend, and
1189 * inform the backend of the appropriate details for those.
1190 */
1191 static int __devinit netfront_probe(struct xenbus_device *dev,
1192 const struct xenbus_device_id *id)
1193 {
1194 int err;
1195 struct net_device *netdev;
1196 struct netfront_info *info;
1197
1198 netdev = xennet_create_dev(dev);
1199 if (IS_ERR(netdev)) {
1200 err = PTR_ERR(netdev);
1201 xenbus_dev_fatal(dev, err, "creating netdev");
1202 return err;
1203 }
1204
1205 info = netdev_priv(netdev);
1206 dev->dev.driver_data = info;
1207
1208 err = register_netdev(info->netdev);
1209 if (err) {
1210 printk(KERN_WARNING "%s: register_netdev err=%d\n",
1211 __func__, err);
1212 goto fail;
1213 }
1214
1215 err = xennet_sysfs_addif(info->netdev);
1216 if (err) {
1217 unregister_netdev(info->netdev);
1218 printk(KERN_WARNING "%s: add sysfs failed err=%d\n",
1219 __func__, err);
1220 goto fail;
1221 }
1222
1223 return 0;
1224
1225 fail:
1226 free_netdev(netdev);
1227 dev->dev.driver_data = NULL;
1228 return err;
1229 }
1230
1231 static void xennet_end_access(int ref, void *page)
1232 {
1233 /* This frees the page as a side-effect */
1234 if (ref != GRANT_INVALID_REF)
1235 gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1236 }
1237
1238 static void xennet_disconnect_backend(struct netfront_info *info)
1239 {
1240 /* Stop old i/f to prevent errors whilst we rebuild the state. */
1241 spin_lock_bh(&info->rx_lock);
1242 spin_lock_irq(&info->tx_lock);
1243 netif_carrier_off(info->netdev);
1244 spin_unlock_irq(&info->tx_lock);
1245 spin_unlock_bh(&info->rx_lock);
1246
1247 if (info->netdev->irq)
1248 unbind_from_irqhandler(info->netdev->irq, info->netdev);
1249 info->evtchn = info->netdev->irq = 0;
1250
1251 /* End access and free the pages */
1252 xennet_end_access(info->tx_ring_ref, info->tx.sring);
1253 xennet_end_access(info->rx_ring_ref, info->rx.sring);
1254
1255 info->tx_ring_ref = GRANT_INVALID_REF;
1256 info->rx_ring_ref = GRANT_INVALID_REF;
1257 info->tx.sring = NULL;
1258 info->rx.sring = NULL;
1259 }
1260
1261 /**
1262 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1263 * driver restart. We tear down our netif structure and recreate it, but
1264 * leave the device-layer structures intact so that this is transparent to the
1265 * rest of the kernel.
1266 */
1267 static int netfront_resume(struct xenbus_device *dev)
1268 {
1269 struct netfront_info *info = dev->dev.driver_data;
1270
1271 dev_dbg(&dev->dev, "%s\n", dev->nodename);
1272
1273 xennet_disconnect_backend(info);
1274 return 0;
1275 }
1276
1277 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1278 {
1279 char *s, *e, *macstr;
1280 int i;
1281
1282 macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1283 if (IS_ERR(macstr))
1284 return PTR_ERR(macstr);
1285
1286 for (i = 0; i < ETH_ALEN; i++) {
1287 mac[i] = simple_strtoul(s, &e, 16);
1288 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1289 kfree(macstr);
1290 return -ENOENT;
1291 }
1292 s = e+1;
1293 }
1294
1295 kfree(macstr);
1296 return 0;
1297 }
1298
1299 static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1300 {
1301 struct net_device *dev = dev_id;
1302 struct netfront_info *np = netdev_priv(dev);
1303 unsigned long flags;
1304
1305 spin_lock_irqsave(&np->tx_lock, flags);
1306
1307 if (likely(netif_carrier_ok(dev))) {
1308 xennet_tx_buf_gc(dev);
1309 /* Under tx_lock: protects access to rx shared-ring indexes. */
1310 if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx))
1311 netif_rx_schedule(dev, &np->napi);
1312 }
1313
1314 spin_unlock_irqrestore(&np->tx_lock, flags);
1315
1316 return IRQ_HANDLED;
1317 }
1318
1319 static int setup_netfront(struct xenbus_device *dev, struct netfront_info *info)
1320 {
1321 struct xen_netif_tx_sring *txs;
1322 struct xen_netif_rx_sring *rxs;
1323 int err;
1324 struct net_device *netdev = info->netdev;
1325
1326 info->tx_ring_ref = GRANT_INVALID_REF;
1327 info->rx_ring_ref = GRANT_INVALID_REF;
1328 info->rx.sring = NULL;
1329 info->tx.sring = NULL;
1330 netdev->irq = 0;
1331
1332 err = xen_net_read_mac(dev, netdev->dev_addr);
1333 if (err) {
1334 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
1335 goto fail;
1336 }
1337
1338 txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1339 if (!txs) {
1340 err = -ENOMEM;
1341 xenbus_dev_fatal(dev, err, "allocating tx ring page");
1342 goto fail;
1343 }
1344 SHARED_RING_INIT(txs);
1345 FRONT_RING_INIT(&info->tx, txs, PAGE_SIZE);
1346
1347 err = xenbus_grant_ring(dev, virt_to_mfn(txs));
1348 if (err < 0) {
1349 free_page((unsigned long)txs);
1350 goto fail;
1351 }
1352
1353 info->tx_ring_ref = err;
1354 rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1355 if (!rxs) {
1356 err = -ENOMEM;
1357 xenbus_dev_fatal(dev, err, "allocating rx ring page");
1358 goto fail;
1359 }
1360 SHARED_RING_INIT(rxs);
1361 FRONT_RING_INIT(&info->rx, rxs, PAGE_SIZE);
1362
1363 err = xenbus_grant_ring(dev, virt_to_mfn(rxs));
1364 if (err < 0) {
1365 free_page((unsigned long)rxs);
1366 goto fail;
1367 }
1368 info->rx_ring_ref = err;
1369
1370 err = xenbus_alloc_evtchn(dev, &info->evtchn);
1371 if (err)
1372 goto fail;
1373
1374 err = bind_evtchn_to_irqhandler(info->evtchn, xennet_interrupt,
1375 IRQF_SAMPLE_RANDOM, netdev->name,
1376 netdev);
1377 if (err < 0)
1378 goto fail;
1379 netdev->irq = err;
1380 return 0;
1381
1382 fail:
1383 return err;
1384 }
1385
1386 /* Common code used when first setting up, and when resuming. */
1387 static int talk_to_backend(struct xenbus_device *dev,
1388 struct netfront_info *info)
1389 {
1390 const char *message;
1391 struct xenbus_transaction xbt;
1392 int err;
1393
1394 /* Create shared ring, alloc event channel. */
1395 err = setup_netfront(dev, info);
1396 if (err)
1397 goto out;
1398
1399 again:
1400 err = xenbus_transaction_start(&xbt);
1401 if (err) {
1402 xenbus_dev_fatal(dev, err, "starting transaction");
1403 goto destroy_ring;
1404 }
1405
1406 err = xenbus_printf(xbt, dev->nodename, "tx-ring-ref", "%u",
1407 info->tx_ring_ref);
1408 if (err) {
1409 message = "writing tx ring-ref";
1410 goto abort_transaction;
1411 }
1412 err = xenbus_printf(xbt, dev->nodename, "rx-ring-ref", "%u",
1413 info->rx_ring_ref);
1414 if (err) {
1415 message = "writing rx ring-ref";
1416 goto abort_transaction;
1417 }
1418 err = xenbus_printf(xbt, dev->nodename,
1419 "event-channel", "%u", info->evtchn);
1420 if (err) {
1421 message = "writing event-channel";
1422 goto abort_transaction;
1423 }
1424
1425 err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
1426 1);
1427 if (err) {
1428 message = "writing request-rx-copy";
1429 goto abort_transaction;
1430 }
1431
1432 err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
1433 if (err) {
1434 message = "writing feature-rx-notify";
1435 goto abort_transaction;
1436 }
1437
1438 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
1439 if (err) {
1440 message = "writing feature-sg";
1441 goto abort_transaction;
1442 }
1443
1444 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
1445 if (err) {
1446 message = "writing feature-gso-tcpv4";
1447 goto abort_transaction;
1448 }
1449
1450 err = xenbus_transaction_end(xbt, 0);
1451 if (err) {
1452 if (err == -EAGAIN)
1453 goto again;
1454 xenbus_dev_fatal(dev, err, "completing transaction");
1455 goto destroy_ring;
1456 }
1457
1458 return 0;
1459
1460 abort_transaction:
1461 xenbus_transaction_end(xbt, 1);
1462 xenbus_dev_fatal(dev, err, "%s", message);
1463 destroy_ring:
1464 xennet_disconnect_backend(info);
1465 out:
1466 return err;
1467 }
1468
1469 static int xennet_set_sg(struct net_device *dev, u32 data)
1470 {
1471 if (data) {
1472 struct netfront_info *np = netdev_priv(dev);
1473 int val;
1474
1475 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, "feature-sg",
1476 "%d", &val) < 0)
1477 val = 0;
1478 if (!val)
1479 return -ENOSYS;
1480 } else if (dev->mtu > ETH_DATA_LEN)
1481 dev->mtu = ETH_DATA_LEN;
1482
1483 return ethtool_op_set_sg(dev, data);
1484 }
1485
1486 static int xennet_set_tso(struct net_device *dev, u32 data)
1487 {
1488 if (data) {
1489 struct netfront_info *np = netdev_priv(dev);
1490 int val;
1491
1492 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1493 "feature-gso-tcpv4", "%d", &val) < 0)
1494 val = 0;
1495 if (!val)
1496 return -ENOSYS;
1497 }
1498
1499 return ethtool_op_set_tso(dev, data);
1500 }
1501
1502 static void xennet_set_features(struct net_device *dev)
1503 {
1504 /* Turn off all GSO bits except ROBUST. */
1505 dev->features &= (1 << NETIF_F_GSO_SHIFT) - 1;
1506 dev->features |= NETIF_F_GSO_ROBUST;
1507 xennet_set_sg(dev, 0);
1508
1509 /* We need checksum offload to enable scatter/gather and TSO. */
1510 if (!(dev->features & NETIF_F_IP_CSUM))
1511 return;
1512
1513 if (!xennet_set_sg(dev, 1))
1514 xennet_set_tso(dev, 1);
1515 }
1516
1517 static int xennet_connect(struct net_device *dev)
1518 {
1519 struct netfront_info *np = netdev_priv(dev);
1520 int i, requeue_idx, err;
1521 struct sk_buff *skb;
1522 grant_ref_t ref;
1523 struct xen_netif_rx_request *req;
1524 unsigned int feature_rx_copy;
1525
1526 err = xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1527 "feature-rx-copy", "%u", &feature_rx_copy);
1528 if (err != 1)
1529 feature_rx_copy = 0;
1530
1531 if (!feature_rx_copy) {
1532 dev_info(&dev->dev,
1533 "backend does not support copying receive path\n");
1534 return -ENODEV;
1535 }
1536
1537 err = talk_to_backend(np->xbdev, np);
1538 if (err)
1539 return err;
1540
1541 xennet_set_features(dev);
1542
1543 spin_lock_bh(&np->rx_lock);
1544 spin_lock_irq(&np->tx_lock);
1545
1546 /* Step 1: Discard all pending TX packet fragments. */
1547 xennet_release_tx_bufs(np);
1548
1549 /* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */
1550 for (requeue_idx = 0, i = 0; i < NET_RX_RING_SIZE; i++) {
1551 if (!np->rx_skbs[i])
1552 continue;
1553
1554 skb = np->rx_skbs[requeue_idx] = xennet_get_rx_skb(np, i);
1555 ref = np->grant_rx_ref[requeue_idx] = xennet_get_rx_ref(np, i);
1556 req = RING_GET_REQUEST(&np->rx, requeue_idx);
1557
1558 gnttab_grant_foreign_access_ref(
1559 ref, np->xbdev->otherend_id,
1560 pfn_to_mfn(page_to_pfn(skb_shinfo(skb)->
1561 frags->page)),
1562 0);
1563 req->gref = ref;
1564 req->id = requeue_idx;
1565
1566 requeue_idx++;
1567 }
1568
1569 np->rx.req_prod_pvt = requeue_idx;
1570
1571 /*
1572 * Step 3: All public and private state should now be sane. Get
1573 * ready to start sending and receiving packets and give the driver
1574 * domain a kick because we've probably just requeued some
1575 * packets.
1576 */
1577 netif_carrier_on(np->netdev);
1578 notify_remote_via_irq(np->netdev->irq);
1579 xennet_tx_buf_gc(dev);
1580 xennet_alloc_rx_buffers(dev);
1581
1582 spin_unlock_irq(&np->tx_lock);
1583 spin_unlock_bh(&np->rx_lock);
1584
1585 return 0;
1586 }
1587
1588 /**
1589 * Callback received when the backend's state changes.
1590 */
1591 static void backend_changed(struct xenbus_device *dev,
1592 enum xenbus_state backend_state)
1593 {
1594 struct netfront_info *np = dev->dev.driver_data;
1595 struct net_device *netdev = np->netdev;
1596
1597 dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
1598
1599 switch (backend_state) {
1600 case XenbusStateInitialising:
1601 case XenbusStateInitialised:
1602 case XenbusStateConnected:
1603 case XenbusStateUnknown:
1604 case XenbusStateClosed:
1605 break;
1606
1607 case XenbusStateInitWait:
1608 if (dev->state != XenbusStateInitialising)
1609 break;
1610 if (xennet_connect(netdev) != 0)
1611 break;
1612 xenbus_switch_state(dev, XenbusStateConnected);
1613 break;
1614
1615 case XenbusStateClosing:
1616 xenbus_frontend_closed(dev);
1617 break;
1618 }
1619 }
1620
1621 static struct ethtool_ops xennet_ethtool_ops =
1622 {
1623 .set_tx_csum = ethtool_op_set_tx_csum,
1624 .set_sg = xennet_set_sg,
1625 .set_tso = xennet_set_tso,
1626 .get_link = ethtool_op_get_link,
1627 };
1628
1629 #ifdef CONFIG_SYSFS
1630 static ssize_t show_rxbuf_min(struct device *dev,
1631 struct device_attribute *attr, char *buf)
1632 {
1633 struct net_device *netdev = to_net_dev(dev);
1634 struct netfront_info *info = netdev_priv(netdev);
1635
1636 return sprintf(buf, "%u\n", info->rx_min_target);
1637 }
1638
1639 static ssize_t store_rxbuf_min(struct device *dev,
1640 struct device_attribute *attr,
1641 const char *buf, size_t len)
1642 {
1643 struct net_device *netdev = to_net_dev(dev);
1644 struct netfront_info *np = netdev_priv(netdev);
1645 char *endp;
1646 unsigned long target;
1647
1648 if (!capable(CAP_NET_ADMIN))
1649 return -EPERM;
1650
1651 target = simple_strtoul(buf, &endp, 0);
1652 if (endp == buf)
1653 return -EBADMSG;
1654
1655 if (target < RX_MIN_TARGET)
1656 target = RX_MIN_TARGET;
1657 if (target > RX_MAX_TARGET)
1658 target = RX_MAX_TARGET;
1659
1660 spin_lock_bh(&np->rx_lock);
1661 if (target > np->rx_max_target)
1662 np->rx_max_target = target;
1663 np->rx_min_target = target;
1664 if (target > np->rx_target)
1665 np->rx_target = target;
1666
1667 xennet_alloc_rx_buffers(netdev);
1668
1669 spin_unlock_bh(&np->rx_lock);
1670 return len;
1671 }
1672
1673 static ssize_t show_rxbuf_max(struct device *dev,
1674 struct device_attribute *attr, char *buf)
1675 {
1676 struct net_device *netdev = to_net_dev(dev);
1677 struct netfront_info *info = netdev_priv(netdev);
1678
1679 return sprintf(buf, "%u\n", info->rx_max_target);
1680 }
1681
1682 static ssize_t store_rxbuf_max(struct device *dev,
1683 struct device_attribute *attr,
1684 const char *buf, size_t len)
1685 {
1686 struct net_device *netdev = to_net_dev(dev);
1687 struct netfront_info *np = netdev_priv(netdev);
1688 char *endp;
1689 unsigned long target;
1690
1691 if (!capable(CAP_NET_ADMIN))
1692 return -EPERM;
1693
1694 target = simple_strtoul(buf, &endp, 0);
1695 if (endp == buf)
1696 return -EBADMSG;
1697
1698 if (target < RX_MIN_TARGET)
1699 target = RX_MIN_TARGET;
1700 if (target > RX_MAX_TARGET)
1701 target = RX_MAX_TARGET;
1702
1703 spin_lock_bh(&np->rx_lock);
1704 if (target < np->rx_min_target)
1705 np->rx_min_target = target;
1706 np->rx_max_target = target;
1707 if (target < np->rx_target)
1708 np->rx_target = target;
1709
1710 xennet_alloc_rx_buffers(netdev);
1711
1712 spin_unlock_bh(&np->rx_lock);
1713 return len;
1714 }
1715
1716 static ssize_t show_rxbuf_cur(struct device *dev,
1717 struct device_attribute *attr, char *buf)
1718 {
1719 struct net_device *netdev = to_net_dev(dev);
1720 struct netfront_info *info = netdev_priv(netdev);
1721
1722 return sprintf(buf, "%u\n", info->rx_target);
1723 }
1724
1725 static struct device_attribute xennet_attrs[] = {
1726 __ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf_min, store_rxbuf_min),
1727 __ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf_max, store_rxbuf_max),
1728 __ATTR(rxbuf_cur, S_IRUGO, show_rxbuf_cur, NULL),
1729 };
1730
1731 static int xennet_sysfs_addif(struct net_device *netdev)
1732 {
1733 int i;
1734 int err;
1735
1736 for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++) {
1737 err = device_create_file(&netdev->dev,
1738 &xennet_attrs[i]);
1739 if (err)
1740 goto fail;
1741 }
1742 return 0;
1743
1744 fail:
1745 while (--i >= 0)
1746 device_remove_file(&netdev->dev, &xennet_attrs[i]);
1747 return err;
1748 }
1749
1750 static void xennet_sysfs_delif(struct net_device *netdev)
1751 {
1752 int i;
1753
1754 for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++)
1755 device_remove_file(&netdev->dev, &xennet_attrs[i]);
1756 }
1757
1758 #endif /* CONFIG_SYSFS */
1759
1760 static struct xenbus_device_id netfront_ids[] = {
1761 { "vif" },
1762 { "" }
1763 };
1764
1765
1766 static int __devexit xennet_remove(struct xenbus_device *dev)
1767 {
1768 struct netfront_info *info = dev->dev.driver_data;
1769
1770 dev_dbg(&dev->dev, "%s\n", dev->nodename);
1771
1772 unregister_netdev(info->netdev);
1773
1774 xennet_disconnect_backend(info);
1775
1776 del_timer_sync(&info->rx_refill_timer);
1777
1778 xennet_sysfs_delif(info->netdev);
1779
1780 free_netdev(info->netdev);
1781
1782 return 0;
1783 }
1784
1785 static struct xenbus_driver netfront = {
1786 .name = "vif",
1787 .owner = THIS_MODULE,
1788 .ids = netfront_ids,
1789 .probe = netfront_probe,
1790 .remove = __devexit_p(xennet_remove),
1791 .resume = netfront_resume,
1792 .otherend_changed = backend_changed,
1793 };
1794
1795 static int __init netif_init(void)
1796 {
1797 if (!xen_domain())
1798 return -ENODEV;
1799
1800 if (xen_initial_domain())
1801 return 0;
1802
1803 printk(KERN_INFO "Initialising Xen virtual ethernet driver.\n");
1804
1805 return xenbus_register_frontend(&netfront);
1806 }
1807 module_init(netif_init);
1808
1809
1810 static void __exit netif_exit(void)
1811 {
1812 if (xen_initial_domain())
1813 return;
1814
1815 xenbus_unregister_driver(&netfront);
1816 }
1817 module_exit(netif_exit);
1818
1819 MODULE_DESCRIPTION("Xen virtual network device frontend");
1820 MODULE_LICENSE("GPL");
1821 MODULE_ALIAS("xen:vif");
1822 MODULE_ALIAS("xennet");
linux 2.6 git repository for openrd