?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
b43legacy: Do not return TX_BUSY from op_tx
[linux-2.6/verdex.git] / drivers / net / xen-netfront.c
blobd26f69b0184f05bac7ec906d5d4f50c11b99b458
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 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 /*
129 * Access macros for acquiring freeing slots in tx_skbs[].
130 */
131
132 static void add_id_to_freelist(unsigned *head, union skb_entry *list,
133 unsigned short id)
134 {
135 list[id].link = *head;
136 *head = id;
137 }
138
139 static unsigned short get_id_from_freelist(unsigned *head,
140 union skb_entry *list)
141 {
142 unsigned int id = *head;
143 *head = list[id].link;
144 return id;
145 }
146
147 static int xennet_rxidx(RING_IDX idx)
148 {
149 return idx & (NET_RX_RING_SIZE - 1);
150 }
151
152 static struct sk_buff *xennet_get_rx_skb(struct netfront_info *np,
153 RING_IDX ri)
154 {
155 int i = xennet_rxidx(ri);
156 struct sk_buff *skb = np->rx_skbs[i];
157 np->rx_skbs[i] = NULL;
158 return skb;
159 }
160
161 static grant_ref_t xennet_get_rx_ref(struct netfront_info *np,
162 RING_IDX ri)
163 {
164 int i = xennet_rxidx(ri);
165 grant_ref_t ref = np->grant_rx_ref[i];
166 np->grant_rx_ref[i] = GRANT_INVALID_REF;
167 return ref;
168 }
169
170 #ifdef CONFIG_SYSFS
171 static int xennet_sysfs_addif(struct net_device *netdev);
172 static void xennet_sysfs_delif(struct net_device *netdev);
173 #else /* !CONFIG_SYSFS */
174 #define xennet_sysfs_addif(dev) (0)
175 #define xennet_sysfs_delif(dev) do { } while (0)
176 #endif
177
178 static int xennet_can_sg(struct net_device *dev)
179 {
180 return dev->features & NETIF_F_SG;
181 }
182
183
184 static void rx_refill_timeout(unsigned long data)
185 {
186 struct net_device *dev = (struct net_device *)data;
187 struct netfront_info *np = netdev_priv(dev);
188 netif_rx_schedule(dev, &np->napi);
189 }
190
191 static int netfront_tx_slot_available(struct netfront_info *np)
192 {
193 return ((np->tx.req_prod_pvt - np->tx.rsp_cons) <
194 (TX_MAX_TARGET - MAX_SKB_FRAGS - 2));
195 }
196
197 static void xennet_maybe_wake_tx(struct net_device *dev)
198 {
199 struct netfront_info *np = netdev_priv(dev);
200
201 if (unlikely(netif_queue_stopped(dev)) &&
202 netfront_tx_slot_available(np) &&
203 likely(netif_running(dev)))
204 netif_wake_queue(dev);
205 }
206
207 static void xennet_alloc_rx_buffers(struct net_device *dev)
208 {
209 unsigned short id;
210 struct netfront_info *np = netdev_priv(dev);
211 struct sk_buff *skb;
212 struct page *page;
213 int i, batch_target, notify;
214 RING_IDX req_prod = np->rx.req_prod_pvt;
215 grant_ref_t ref;
216 unsigned long pfn;
217 void *vaddr;
218 struct xen_netif_rx_request *req;
219
220 if (unlikely(!netif_carrier_ok(dev)))
221 return;
222
223 /*
224 * Allocate skbuffs greedily, even though we batch updates to the
225 * receive ring. This creates a less bursty demand on the memory
226 * allocator, so should reduce the chance of failed allocation requests
227 * both for ourself and for other kernel subsystems.
228 */
229 batch_target = np->rx_target - (req_prod - np->rx.rsp_cons);
230 for (i = skb_queue_len(&np->rx_batch); i < batch_target; i++) {
231 skb = __netdev_alloc_skb(dev, RX_COPY_THRESHOLD,
232 GFP_ATOMIC | __GFP_NOWARN);
233 if (unlikely(!skb))
234 goto no_skb;
235
236 page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
237 if (!page) {
238 kfree_skb(skb);
239 no_skb:
240 /* Any skbuffs queued for refill? Force them out. */
241 if (i != 0)
242 goto refill;
243 /* Could not allocate any skbuffs. Try again later. */
244 mod_timer(&np->rx_refill_timer,
245 jiffies + (HZ/10));
246 break;
247 }
248
249 skb_shinfo(skb)->frags[0].page = page;
250 skb_shinfo(skb)->nr_frags = 1;
251 __skb_queue_tail(&np->rx_batch, skb);
252 }
253
254 /* Is the batch large enough to be worthwhile? */
255 if (i < (np->rx_target/2)) {
256 if (req_prod > np->rx.sring->req_prod)
257 goto push;
258 return;
259 }
260
261 /* Adjust our fill target if we risked running out of buffers. */
262 if (((req_prod - np->rx.sring->rsp_prod) < (np->rx_target / 4)) &&
263 ((np->rx_target *= 2) > np->rx_max_target))
264 np->rx_target = np->rx_max_target;
265
266 refill:
267 for (i = 0; ; i++) {
268 skb = __skb_dequeue(&np->rx_batch);
269 if (skb == NULL)
270 break;
271
272 skb->dev = dev;
273
274 id = xennet_rxidx(req_prod + i);
275
276 BUG_ON(np->rx_skbs[id]);
277 np->rx_skbs[id] = skb;
278
279 ref = gnttab_claim_grant_reference(&np->gref_rx_head);
280 BUG_ON((signed short)ref < 0);
281 np->grant_rx_ref[id] = ref;
282
283 pfn = page_to_pfn(skb_shinfo(skb)->frags[0].page);
284 vaddr = page_address(skb_shinfo(skb)->frags[0].page);
285
286 req = RING_GET_REQUEST(&np->rx, req_prod + i);
287 gnttab_grant_foreign_access_ref(ref,
288 np->xbdev->otherend_id,
289 pfn_to_mfn(pfn),
290 0);
291
292 req->id = id;
293 req->gref = ref;
294 }
295
296 wmb(); /* barrier so backend seens requests */
297
298 /* Above is a suitable barrier to ensure backend will see requests. */
299 np->rx.req_prod_pvt = req_prod + i;
300 push:
301 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->rx, notify);
302 if (notify)
303 notify_remote_via_irq(np->netdev->irq);
304 }
305
306 static int xennet_open(struct net_device *dev)
307 {
308 struct netfront_info *np = netdev_priv(dev);
309
310 napi_enable(&np->napi);
311
312 spin_lock_bh(&np->rx_lock);
313 if (netif_carrier_ok(dev)) {
314 xennet_alloc_rx_buffers(dev);
315 np->rx.sring->rsp_event = np->rx.rsp_cons + 1;
316 if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx))
317 netif_rx_schedule(dev, &np->napi);
318 }
319 spin_unlock_bh(&np->rx_lock);
320
321 xennet_maybe_wake_tx(dev);
322
323 return 0;
324 }
325
326 static void xennet_tx_buf_gc(struct net_device *dev)
327 {
328 RING_IDX cons, prod;
329 unsigned short id;
330 struct netfront_info *np = netdev_priv(dev);
331 struct sk_buff *skb;
332
333 BUG_ON(!netif_carrier_ok(dev));
334
335 do {
336 prod = np->tx.sring->rsp_prod;
337 rmb(); /* Ensure we see responses up to 'rp'. */
338
339 for (cons = np->tx.rsp_cons; cons != prod; cons++) {
340 struct xen_netif_tx_response *txrsp;
341
342 txrsp = RING_GET_RESPONSE(&np->tx, cons);
343 if (txrsp->status == NETIF_RSP_NULL)
344 continue;
345
346 id = txrsp->id;
347 skb = np->tx_skbs[id].skb;
348 if (unlikely(gnttab_query_foreign_access(
349 np->grant_tx_ref[id]) != 0)) {
350 printk(KERN_ALERT "xennet_tx_buf_gc: warning "
351 "-- grant still in use by backend "
352 "domain.\n");
353 BUG();
354 }
355 gnttab_end_foreign_access_ref(
356 np->grant_tx_ref[id], GNTMAP_readonly);
357 gnttab_release_grant_reference(
358 &np->gref_tx_head, np->grant_tx_ref[id]);
359 np->grant_tx_ref[id] = GRANT_INVALID_REF;
360 add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, id);
361 dev_kfree_skb_irq(skb);
362 }
363
364 np->tx.rsp_cons = prod;
365
366 /*
367 * Set a new event, then check for race with update of tx_cons.
368 * Note that it is essential to schedule a callback, no matter
369 * how few buffers are pending. Even if there is space in the
370 * transmit ring, higher layers may be blocked because too much
371 * data is outstanding: in such cases notification from Xen is
372 * likely to be the only kick that we'll get.
373 */
374 np->tx.sring->rsp_event =
375 prod + ((np->tx.sring->req_prod - prod) >> 1) + 1;
376 mb(); /* update shared area */
377 } while ((cons == prod) && (prod != np->tx.sring->rsp_prod));
378
379 xennet_maybe_wake_tx(dev);
380 }
381
382 static void xennet_make_frags(struct sk_buff *skb, struct net_device *dev,
383 struct xen_netif_tx_request *tx)
384 {
385 struct netfront_info *np = netdev_priv(dev);
386 char *data = skb->data;
387 unsigned long mfn;
388 RING_IDX prod = np->tx.req_prod_pvt;
389 int frags = skb_shinfo(skb)->nr_frags;
390 unsigned int offset = offset_in_page(data);
391 unsigned int len = skb_headlen(skb);
392 unsigned int id;
393 grant_ref_t ref;
394 int i;
395
396 /* While the header overlaps a page boundary (including being
397 larger than a page), split it it into page-sized chunks. */
398 while (len > PAGE_SIZE - offset) {
399 tx->size = PAGE_SIZE - offset;
400 tx->flags |= NETTXF_more_data;
401 len -= tx->size;
402 data += tx->size;
403 offset = 0;
404
405 id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
406 np->tx_skbs[id].skb = skb_get(skb);
407 tx = RING_GET_REQUEST(&np->tx, prod++);
408 tx->id = id;
409 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
410 BUG_ON((signed short)ref < 0);
411
412 mfn = virt_to_mfn(data);
413 gnttab_grant_foreign_access_ref(ref, np->xbdev->otherend_id,
414 mfn, GNTMAP_readonly);
415
416 tx->gref = np->grant_tx_ref[id] = ref;
417 tx->offset = offset;
418 tx->size = len;
419 tx->flags = 0;
420 }
421
422 /* Grant backend access to each skb fragment page. */
423 for (i = 0; i < frags; i++) {
424 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
425
426 tx->flags |= NETTXF_more_data;
427
428 id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
429 np->tx_skbs[id].skb = skb_get(skb);
430 tx = RING_GET_REQUEST(&np->tx, prod++);
431 tx->id = id;
432 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
433 BUG_ON((signed short)ref < 0);
434
435 mfn = pfn_to_mfn(page_to_pfn(frag->page));
436 gnttab_grant_foreign_access_ref(ref, np->xbdev->otherend_id,
437 mfn, GNTMAP_readonly);
438
439 tx->gref = np->grant_tx_ref[id] = ref;
440 tx->offset = frag->page_offset;
441 tx->size = frag->size;
442 tx->flags = 0;
443 }
444
445 np->tx.req_prod_pvt = prod;
446 }
447
448 static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
449 {
450 unsigned short id;
451 struct netfront_info *np = netdev_priv(dev);
452 struct xen_netif_tx_request *tx;
453 struct xen_netif_extra_info *extra;
454 char *data = skb->data;
455 RING_IDX i;
456 grant_ref_t ref;
457 unsigned long mfn;
458 int notify;
459 int frags = skb_shinfo(skb)->nr_frags;
460 unsigned int offset = offset_in_page(data);
461 unsigned int len = skb_headlen(skb);
462
463 frags += (offset + len + PAGE_SIZE - 1) / PAGE_SIZE;
464 if (unlikely(frags > MAX_SKB_FRAGS + 1)) {
465 printk(KERN_ALERT "xennet: skb rides the rocket: %d frags\n",
466 frags);
467 dump_stack();
468 goto drop;
469 }
470
471 spin_lock_irq(&np->tx_lock);
472
473 if (unlikely(!netif_carrier_ok(dev) ||
474 (frags > 1 && !xennet_can_sg(dev)) ||
475 netif_needs_gso(dev, skb))) {
476 spin_unlock_irq(&np->tx_lock);
477 goto drop;
478 }
479
480 i = np->tx.req_prod_pvt;
481
482 id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
483 np->tx_skbs[id].skb = skb;
484
485 tx = RING_GET_REQUEST(&np->tx, i);
486
487 tx->id = id;
488 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
489 BUG_ON((signed short)ref < 0);
490 mfn = virt_to_mfn(data);
491 gnttab_grant_foreign_access_ref(
492 ref, np->xbdev->otherend_id, mfn, GNTMAP_readonly);
493 tx->gref = np->grant_tx_ref[id] = ref;
494 tx->offset = offset;
495 tx->size = len;
496 extra = NULL;
497
498 tx->flags = 0;
499 if (skb->ip_summed == CHECKSUM_PARTIAL)
500 /* local packet? */
501 tx->flags |= NETTXF_csum_blank | NETTXF_data_validated;
502 else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
503 /* remote but checksummed. */
504 tx->flags |= NETTXF_data_validated;
505
506 if (skb_shinfo(skb)->gso_size) {
507 struct xen_netif_extra_info *gso;
508
509 gso = (struct xen_netif_extra_info *)
510 RING_GET_REQUEST(&np->tx, ++i);
511
512 if (extra)
513 extra->flags |= XEN_NETIF_EXTRA_FLAG_MORE;
514 else
515 tx->flags |= NETTXF_extra_info;
516
517 gso->u.gso.size = skb_shinfo(skb)->gso_size;
518 gso->u.gso.type = XEN_NETIF_GSO_TYPE_TCPV4;
519 gso->u.gso.pad = 0;
520 gso->u.gso.features = 0;
521
522 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
523 gso->flags = 0;
524 extra = gso;
525 }
526
527 np->tx.req_prod_pvt = i + 1;
528
529 xennet_make_frags(skb, dev, tx);
530 tx->size = skb->len;
531
532 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->tx, notify);
533 if (notify)
534 notify_remote_via_irq(np->netdev->irq);
535
536 dev->stats.tx_bytes += skb->len;
537 dev->stats.tx_packets++;
538
539 /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
540 xennet_tx_buf_gc(dev);
541
542 if (!netfront_tx_slot_available(np))
543 netif_stop_queue(dev);
544
545 spin_unlock_irq(&np->tx_lock);
546
547 return 0;
548
549 drop:
550 dev->stats.tx_dropped++;
551 dev_kfree_skb(skb);
552 return 0;
553 }
554
555 static int xennet_close(struct net_device *dev)
556 {
557 struct netfront_info *np = netdev_priv(dev);
558 netif_stop_queue(np->netdev);
559 napi_disable(&np->napi);
560 return 0;
561 }
562
563 static void xennet_move_rx_slot(struct netfront_info *np, struct sk_buff *skb,
564 grant_ref_t ref)
565 {
566 int new = xennet_rxidx(np->rx.req_prod_pvt);
567
568 BUG_ON(np->rx_skbs[new]);
569 np->rx_skbs[new] = skb;
570 np->grant_rx_ref[new] = ref;
571 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->id = new;
572 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->gref = ref;
573 np->rx.req_prod_pvt++;
574 }
575
576 static int xennet_get_extras(struct netfront_info *np,
577 struct xen_netif_extra_info *extras,
578 RING_IDX rp)
579
580 {
581 struct xen_netif_extra_info *extra;
582 struct device *dev = &np->netdev->dev;
583 RING_IDX cons = np->rx.rsp_cons;
584 int err = 0;
585
586 do {
587 struct sk_buff *skb;
588 grant_ref_t ref;
589
590 if (unlikely(cons + 1 == rp)) {
591 if (net_ratelimit())
592 dev_warn(dev, "Missing extra info\n");
593 err = -EBADR;
594 break;
595 }
596
597 extra = (struct xen_netif_extra_info *)
598 RING_GET_RESPONSE(&np->rx, ++cons);
599
600 if (unlikely(!extra->type ||
601 extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
602 if (net_ratelimit())
603 dev_warn(dev, "Invalid extra type: %d\n",
604 extra->type);
605 err = -EINVAL;
606 } else {
607 memcpy(&extras[extra->type - 1], extra,
608 sizeof(*extra));
609 }
610
611 skb = xennet_get_rx_skb(np, cons);
612 ref = xennet_get_rx_ref(np, cons);
613 xennet_move_rx_slot(np, skb, ref);
614 } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
615
616 np->rx.rsp_cons = cons;
617 return err;
618 }
619
620 static int xennet_get_responses(struct netfront_info *np,
621 struct netfront_rx_info *rinfo, RING_IDX rp,
622 struct sk_buff_head *list)
623 {
624 struct xen_netif_rx_response *rx = &rinfo->rx;
625 struct xen_netif_extra_info *extras = rinfo->extras;
626 struct device *dev = &np->netdev->dev;
627 RING_IDX cons = np->rx.rsp_cons;
628 struct sk_buff *skb = xennet_get_rx_skb(np, cons);
629 grant_ref_t ref = xennet_get_rx_ref(np, cons);
630 int max = MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD);
631 int frags = 1;
632 int err = 0;
633 unsigned long ret;
634
635 if (rx->flags & NETRXF_extra_info) {
636 err = xennet_get_extras(np, extras, rp);
637 cons = np->rx.rsp_cons;
638 }
639
640 for (;;) {
641 if (unlikely(rx->status < 0 ||
642 rx->offset + rx->status > PAGE_SIZE)) {
643 if (net_ratelimit())
644 dev_warn(dev, "rx->offset: %x, size: %u\n",
645 rx->offset, rx->status);
646 xennet_move_rx_slot(np, skb, ref);
647 err = -EINVAL;
648 goto next;
649 }
650
651 /*
652 * This definitely indicates a bug, either in this driver or in
653 * the backend driver. In future this should flag the bad
654 * situation to the system controller to reboot the backed.
655 */
656 if (ref == GRANT_INVALID_REF) {
657 if (net_ratelimit())
658 dev_warn(dev, "Bad rx response id %d.\n",
659 rx->id);
660 err = -EINVAL;
661 goto next;
662 }
663
664 ret = gnttab_end_foreign_access_ref(ref, 0);
665 BUG_ON(!ret);
666
667 gnttab_release_grant_reference(&np->gref_rx_head, ref);
668
669 __skb_queue_tail(list, skb);
670
671 next:
672 if (!(rx->flags & NETRXF_more_data))
673 break;
674
675 if (cons + frags == rp) {
676 if (net_ratelimit())
677 dev_warn(dev, "Need more frags\n");
678 err = -ENOENT;
679 break;
680 }
681
682 rx = RING_GET_RESPONSE(&np->rx, cons + frags);
683 skb = xennet_get_rx_skb(np, cons + frags);
684 ref = xennet_get_rx_ref(np, cons + frags);
685 frags++;
686 }
687
688 if (unlikely(frags > max)) {
689 if (net_ratelimit())
690 dev_warn(dev, "Too many frags\n");
691 err = -E2BIG;
692 }
693
694 if (unlikely(err))
695 np->rx.rsp_cons = cons + frags;
696
697 return err;
698 }
699
700 static int xennet_set_skb_gso(struct sk_buff *skb,
701 struct xen_netif_extra_info *gso)
702 {
703 if (!gso->u.gso.size) {
704 if (net_ratelimit())
705 printk(KERN_WARNING "GSO size must not be zero.\n");
706 return -EINVAL;
707 }
708
709 /* Currently only TCPv4 S.O. is supported. */
710 if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4) {
711 if (net_ratelimit())
712 printk(KERN_WARNING "Bad GSO type %d.\n", gso->u.gso.type);
713 return -EINVAL;
714 }
715
716 skb_shinfo(skb)->gso_size = gso->u.gso.size;
717 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
718
719 /* Header must be checked, and gso_segs computed. */
720 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
721 skb_shinfo(skb)->gso_segs = 0;
722
723 return 0;
724 }
725
726 static RING_IDX xennet_fill_frags(struct netfront_info *np,
727 struct sk_buff *skb,
728 struct sk_buff_head *list)
729 {
730 struct skb_shared_info *shinfo = skb_shinfo(skb);
731 int nr_frags = shinfo->nr_frags;
732 RING_IDX cons = np->rx.rsp_cons;
733 skb_frag_t *frag = shinfo->frags + nr_frags;
734 struct sk_buff *nskb;
735
736 while ((nskb = __skb_dequeue(list))) {
737 struct xen_netif_rx_response *rx =
738 RING_GET_RESPONSE(&np->rx, ++cons);
739
740 frag->page = skb_shinfo(nskb)->frags[0].page;
741 frag->page_offset = rx->offset;
742 frag->size = rx->status;
743
744 skb->data_len += rx->status;
745
746 skb_shinfo(nskb)->nr_frags = 0;
747 kfree_skb(nskb);
748
749 frag++;
750 nr_frags++;
751 }
752
753 shinfo->nr_frags = nr_frags;
754 return cons;
755 }
756
757 static int skb_checksum_setup(struct sk_buff *skb)
758 {
759 struct iphdr *iph;
760 unsigned char *th;
761 int err = -EPROTO;
762
763 if (skb->protocol != htons(ETH_P_IP))
764 goto out;
765
766 iph = (void *)skb->data;
767 th = skb->data + 4 * iph->ihl;
768 if (th >= skb_tail_pointer(skb))
769 goto out;
770
771 skb->csum_start = th - skb->head;
772 switch (iph->protocol) {
773 case IPPROTO_TCP:
774 skb->csum_offset = offsetof(struct tcphdr, check);
775 break;
776 case IPPROTO_UDP:
777 skb->csum_offset = offsetof(struct udphdr, check);
778 break;
779 default:
780 if (net_ratelimit())
781 printk(KERN_ERR "Attempting to checksum a non-"
782 "TCP/UDP packet, dropping a protocol"
783 " %d packet", iph->protocol);
784 goto out;
785 }
786
787 if ((th + skb->csum_offset + 2) > skb_tail_pointer(skb))
788 goto out;
789
790 err = 0;
791
792 out:
793 return err;
794 }
795
796 static int handle_incoming_queue(struct net_device *dev,
797 struct sk_buff_head *rxq)
798 {
799 int packets_dropped = 0;
800 struct sk_buff *skb;
801
802 while ((skb = __skb_dequeue(rxq)) != NULL) {
803 struct page *page = NETFRONT_SKB_CB(skb)->page;
804 void *vaddr = page_address(page);
805 unsigned offset = NETFRONT_SKB_CB(skb)->offset;
806
807 memcpy(skb->data, vaddr + offset,
808 skb_headlen(skb));
809
810 if (page != skb_shinfo(skb)->frags[0].page)
811 __free_page(page);
812
813 /* Ethernet work: Delayed to here as it peeks the header. */
814 skb->protocol = eth_type_trans(skb, dev);
815
816 if (skb->ip_summed == CHECKSUM_PARTIAL) {
817 if (skb_checksum_setup(skb)) {
818 kfree_skb(skb);
819 packets_dropped++;
820 dev->stats.rx_errors++;
821 continue;
822 }
823 }
824
825 dev->stats.rx_packets++;
826 dev->stats.rx_bytes += skb->len;
827
828 /* Pass it up. */
829 netif_receive_skb(skb);
830 dev->last_rx = jiffies;
831 }
832
833 return packets_dropped;
834 }
835
836 static int xennet_poll(struct napi_struct *napi, int budget)
837 {
838 struct netfront_info *np = container_of(napi, struct netfront_info, napi);
839 struct net_device *dev = np->netdev;
840 struct sk_buff *skb;
841 struct netfront_rx_info rinfo;
842 struct xen_netif_rx_response *rx = &rinfo.rx;
843 struct xen_netif_extra_info *extras = rinfo.extras;
844 RING_IDX i, rp;
845 int work_done;
846 struct sk_buff_head rxq;
847 struct sk_buff_head errq;
848 struct sk_buff_head tmpq;
849 unsigned long flags;
850 unsigned int len;
851 int err;
852
853 spin_lock(&np->rx_lock);
854
855 skb_queue_head_init(&rxq);
856 skb_queue_head_init(&errq);
857 skb_queue_head_init(&tmpq);
858
859 rp = np->rx.sring->rsp_prod;
860 rmb(); /* Ensure we see queued responses up to 'rp'. */
861
862 i = np->rx.rsp_cons;
863 work_done = 0;
864 while ((i != rp) && (work_done < budget)) {
865 memcpy(rx, RING_GET_RESPONSE(&np->rx, i), sizeof(*rx));
866 memset(extras, 0, sizeof(rinfo.extras));
867
868 err = xennet_get_responses(np, &rinfo, rp, &tmpq);
869
870 if (unlikely(err)) {
871 err:
872 while ((skb = __skb_dequeue(&tmpq)))
873 __skb_queue_tail(&errq, skb);
874 dev->stats.rx_errors++;
875 i = np->rx.rsp_cons;
876 continue;
877 }
878
879 skb = __skb_dequeue(&tmpq);
880
881 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
882 struct xen_netif_extra_info *gso;
883 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
884
885 if (unlikely(xennet_set_skb_gso(skb, gso))) {
886 __skb_queue_head(&tmpq, skb);
887 np->rx.rsp_cons += skb_queue_len(&tmpq);
888 goto err;
889 }
890 }
891
892 NETFRONT_SKB_CB(skb)->page = skb_shinfo(skb)->frags[0].page;
893 NETFRONT_SKB_CB(skb)->offset = rx->offset;
894
895 len = rx->status;
896 if (len > RX_COPY_THRESHOLD)
897 len = RX_COPY_THRESHOLD;
898 skb_put(skb, len);
899
900 if (rx->status > len) {
901 skb_shinfo(skb)->frags[0].page_offset =
902 rx->offset + len;
903 skb_shinfo(skb)->frags[0].size = rx->status - len;
904 skb->data_len = rx->status - len;
905 } else {
906 skb_shinfo(skb)->frags[0].page = NULL;
907 skb_shinfo(skb)->nr_frags = 0;
908 }
909
910 i = xennet_fill_frags(np, skb, &tmpq);
911
912 /*
913 * Truesize approximates the size of true data plus
914 * any supervisor overheads. Adding hypervisor
915 * overheads has been shown to significantly reduce
916 * achievable bandwidth with the default receive
917 * buffer size. It is therefore not wise to account
918 * for it here.
919 *
920 * After alloc_skb(RX_COPY_THRESHOLD), truesize is set
921 * to RX_COPY_THRESHOLD + the supervisor
922 * overheads. Here, we add the size of the data pulled
923 * in xennet_fill_frags().
924 *
925 * We also adjust for any unused space in the main
926 * data area by subtracting (RX_COPY_THRESHOLD -
927 * len). This is especially important with drivers
928 * which split incoming packets into header and data,
929 * using only 66 bytes of the main data area (see the
930 * e1000 driver for example.) On such systems,
931 * without this last adjustement, our achievable
932 * receive throughout using the standard receive
933 * buffer size was cut by 25%(!!!).
934 */
935 skb->truesize += skb->data_len - (RX_COPY_THRESHOLD - len);
936 skb->len += skb->data_len;
937
938 if (rx->flags & NETRXF_csum_blank)
939 skb->ip_summed = CHECKSUM_PARTIAL;
940 else if (rx->flags & NETRXF_data_validated)
941 skb->ip_summed = CHECKSUM_UNNECESSARY;
942
943 __skb_queue_tail(&rxq, skb);
944
945 np->rx.rsp_cons = ++i;
946 work_done++;
947 }
948
949 __skb_queue_purge(&errq);
950
951 work_done -= handle_incoming_queue(dev, &rxq);
952
953 /* If we get a callback with very few responses, reduce fill target. */
954 /* NB. Note exponential increase, linear decrease. */
955 if (((np->rx.req_prod_pvt - np->rx.sring->rsp_prod) >
956 ((3*np->rx_target) / 4)) &&
957 (--np->rx_target < np->rx_min_target))
958 np->rx_target = np->rx_min_target;
959
960 xennet_alloc_rx_buffers(dev);
961
962 if (work_done < budget) {
963 int more_to_do = 0;
964
965 local_irq_save(flags);
966
967 RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, more_to_do);
968 if (!more_to_do)
969 __netif_rx_complete(dev, napi);
970
971 local_irq_restore(flags);
972 }
973
974 spin_unlock(&np->rx_lock);
975
976 return work_done;
977 }
978
979 static int xennet_change_mtu(struct net_device *dev, int mtu)
980 {
981 int max = xennet_can_sg(dev) ? 65535 - ETH_HLEN : ETH_DATA_LEN;
982
983 if (mtu > max)
984 return -EINVAL;
985 dev->mtu = mtu;
986 return 0;
987 }
988
989 static void xennet_release_tx_bufs(struct netfront_info *np)
990 {
991 struct sk_buff *skb;
992 int i;
993
994 for (i = 0; i < NET_TX_RING_SIZE; i++) {
995 /* Skip over entries which are actually freelist references */
996 if ((unsigned long)np->tx_skbs[i].skb < PAGE_OFFSET)
997 continue;
998
999 skb = np->tx_skbs[i].skb;
1000 gnttab_end_foreign_access_ref(np->grant_tx_ref[i],
1001 GNTMAP_readonly);
1002 gnttab_release_grant_reference(&np->gref_tx_head,
1003 np->grant_tx_ref[i]);
1004 np->grant_tx_ref[i] = GRANT_INVALID_REF;
1005 add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, i);
1006 dev_kfree_skb_irq(skb);
1007 }
1008 }
1009
1010 static void xennet_release_rx_bufs(struct netfront_info *np)
1011 {
1012 struct mmu_update *mmu = np->rx_mmu;
1013 struct multicall_entry *mcl = np->rx_mcl;
1014 struct sk_buff_head free_list;
1015 struct sk_buff *skb;
1016 unsigned long mfn;
1017 int xfer = 0, noxfer = 0, unused = 0;
1018 int id, ref;
1019
1020 dev_warn(&np->netdev->dev, "%s: fix me for copying receiver.\n",
1021 __func__);
1022 return;
1023
1024 skb_queue_head_init(&free_list);
1025
1026 spin_lock_bh(&np->rx_lock);
1027
1028 for (id = 0; id < NET_RX_RING_SIZE; id++) {
1029 ref = np->grant_rx_ref[id];
1030 if (ref == GRANT_INVALID_REF) {
1031 unused++;
1032 continue;
1033 }
1034
1035 skb = np->rx_skbs[id];
1036 mfn = gnttab_end_foreign_transfer_ref(ref);
1037 gnttab_release_grant_reference(&np->gref_rx_head, ref);
1038 np->grant_rx_ref[id] = GRANT_INVALID_REF;
1039
1040 if (0 == mfn) {
1041 skb_shinfo(skb)->nr_frags = 0;
1042 dev_kfree_skb(skb);
1043 noxfer++;
1044 continue;
1045 }
1046
1047 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1048 /* Remap the page. */
1049 struct page *page = skb_shinfo(skb)->frags[0].page;
1050 unsigned long pfn = page_to_pfn(page);
1051 void *vaddr = page_address(page);
1052
1053 MULTI_update_va_mapping(mcl, (unsigned long)vaddr,
1054 mfn_pte(mfn, PAGE_KERNEL),
1055 0);
1056 mcl++;
1057 mmu->ptr = ((u64)mfn << PAGE_SHIFT)
1058 | MMU_MACHPHYS_UPDATE;
1059 mmu->val = pfn;
1060 mmu++;
1061
1062 set_phys_to_machine(pfn, mfn);
1063 }
1064 __skb_queue_tail(&free_list, skb);
1065 xfer++;
1066 }
1067
1068 dev_info(&np->netdev->dev, "%s: %d xfer, %d noxfer, %d unused\n",
1069 __func__, xfer, noxfer, unused);
1070
1071 if (xfer) {
1072 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1073 /* Do all the remapping work and M2P updates. */
1074 MULTI_mmu_update(mcl, np->rx_mmu, mmu - np->rx_mmu,
1075 NULL, DOMID_SELF);
1076 mcl++;
1077 HYPERVISOR_multicall(np->rx_mcl, mcl - np->rx_mcl);
1078 }
1079 }
1080
1081 __skb_queue_purge(&free_list);
1082
1083 spin_unlock_bh(&np->rx_lock);
1084 }
1085
1086 static void xennet_uninit(struct net_device *dev)
1087 {
1088 struct netfront_info *np = netdev_priv(dev);
1089 xennet_release_tx_bufs(np);
1090 xennet_release_rx_bufs(np);
1091 gnttab_free_grant_references(np->gref_tx_head);
1092 gnttab_free_grant_references(np->gref_rx_head);
1093 }
1094
1095 static struct net_device * __devinit xennet_create_dev(struct xenbus_device *dev)
1096 {
1097 int i, err;
1098 struct net_device *netdev;
1099 struct netfront_info *np;
1100
1101 netdev = alloc_etherdev(sizeof(struct netfront_info));
1102 if (!netdev) {
1103 printk(KERN_WARNING "%s> alloc_etherdev failed.\n",
1104 __func__);
1105 return ERR_PTR(-ENOMEM);
1106 }
1107
1108 np = netdev_priv(netdev);
1109 np->xbdev = dev;
1110
1111 spin_lock_init(&np->tx_lock);
1112 spin_lock_init(&np->rx_lock);
1113
1114 skb_queue_head_init(&np->rx_batch);
1115 np->rx_target = RX_DFL_MIN_TARGET;
1116 np->rx_min_target = RX_DFL_MIN_TARGET;
1117 np->rx_max_target = RX_MAX_TARGET;
1118
1119 init_timer(&np->rx_refill_timer);
1120 np->rx_refill_timer.data = (unsigned long)netdev;
1121 np->rx_refill_timer.function = rx_refill_timeout;
1122
1123 /* Initialise tx_skbs as a free chain containing every entry. */
1124 np->tx_skb_freelist = 0;
1125 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1126 np->tx_skbs[i].link = i+1;
1127 np->grant_tx_ref[i] = GRANT_INVALID_REF;
1128 }
1129
1130 /* Clear out rx_skbs */
1131 for (i = 0; i < NET_RX_RING_SIZE; i++) {
1132 np->rx_skbs[i] = NULL;
1133 np->grant_rx_ref[i] = GRANT_INVALID_REF;
1134 }
1135
1136 /* A grant for every tx ring slot */
1137 if (gnttab_alloc_grant_references(TX_MAX_TARGET,
1138 &np->gref_tx_head) < 0) {
1139 printk(KERN_ALERT "#### netfront can't alloc tx grant refs\n");
1140 err = -ENOMEM;
1141 goto exit;
1142 }
1143 /* A grant for every rx ring slot */
1144 if (gnttab_alloc_grant_references(RX_MAX_TARGET,
1145 &np->gref_rx_head) < 0) {
1146 printk(KERN_ALERT "#### netfront can't alloc rx grant refs\n");
1147 err = -ENOMEM;
1148 goto exit_free_tx;
1149 }
1150
1151 netdev->open = xennet_open;
1152 netdev->hard_start_xmit = xennet_start_xmit;
1153 netdev->stop = xennet_close;
1154 netif_napi_add(netdev, &np->napi, xennet_poll, 64);
1155 netdev->uninit = xennet_uninit;
1156 netdev->change_mtu = xennet_change_mtu;
1157 netdev->features = NETIF_F_IP_CSUM;
1158
1159 SET_ETHTOOL_OPS(netdev, &xennet_ethtool_ops);
1160 SET_NETDEV_DEV(netdev, &dev->dev);
1161
1162 np->netdev = netdev;
1163
1164 netif_carrier_off(netdev);
1165
1166 return netdev;
1167
1168 exit_free_tx:
1169 gnttab_free_grant_references(np->gref_tx_head);
1170 exit:
1171 free_netdev(netdev);
1172 return ERR_PTR(err);
1173 }
1174
1175 /**
1176 * Entry point to this code when a new device is created. Allocate the basic
1177 * structures and the ring buffers for communication with the backend, and
1178 * inform the backend of the appropriate details for those.
1179 */
1180 static int __devinit netfront_probe(struct xenbus_device *dev,
1181 const struct xenbus_device_id *id)
1182 {
1183 int err;
1184 struct net_device *netdev;
1185 struct netfront_info *info;
1186
1187 netdev = xennet_create_dev(dev);
1188 if (IS_ERR(netdev)) {
1189 err = PTR_ERR(netdev);
1190 xenbus_dev_fatal(dev, err, "creating netdev");
1191 return err;
1192 }
1193
1194 info = netdev_priv(netdev);
1195 dev->dev.driver_data = info;
1196
1197 err = register_netdev(info->netdev);
1198 if (err) {
1199 printk(KERN_WARNING "%s: register_netdev err=%d\n",
1200 __func__, err);
1201 goto fail;
1202 }
1203
1204 err = xennet_sysfs_addif(info->netdev);
1205 if (err) {
1206 unregister_netdev(info->netdev);
1207 printk(KERN_WARNING "%s: add sysfs failed err=%d\n",
1208 __func__, err);
1209 goto fail;
1210 }
1211
1212 return 0;
1213
1214 fail:
1215 free_netdev(netdev);
1216 dev->dev.driver_data = NULL;
1217 return err;
1218 }
1219
1220 static void xennet_end_access(int ref, void *page)
1221 {
1222 /* This frees the page as a side-effect */
1223 if (ref != GRANT_INVALID_REF)
1224 gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1225 }
1226
1227 static void xennet_disconnect_backend(struct netfront_info *info)
1228 {
1229 /* Stop old i/f to prevent errors whilst we rebuild the state. */
1230 spin_lock_bh(&info->rx_lock);
1231 spin_lock_irq(&info->tx_lock);
1232 netif_carrier_off(info->netdev);
1233 spin_unlock_irq(&info->tx_lock);
1234 spin_unlock_bh(&info->rx_lock);
1235
1236 if (info->netdev->irq)
1237 unbind_from_irqhandler(info->netdev->irq, info->netdev);
1238 info->evtchn = info->netdev->irq = 0;
1239
1240 /* End access and free the pages */
1241 xennet_end_access(info->tx_ring_ref, info->tx.sring);
1242 xennet_end_access(info->rx_ring_ref, info->rx.sring);
1243
1244 info->tx_ring_ref = GRANT_INVALID_REF;
1245 info->rx_ring_ref = GRANT_INVALID_REF;
1246 info->tx.sring = NULL;
1247 info->rx.sring = NULL;
1248 }
1249
1250 /**
1251 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1252 * driver restart. We tear down our netif structure and recreate it, but
1253 * leave the device-layer structures intact so that this is transparent to the
1254 * rest of the kernel.
1255 */
1256 static int netfront_resume(struct xenbus_device *dev)
1257 {
1258 struct netfront_info *info = dev->dev.driver_data;
1259
1260 dev_dbg(&dev->dev, "%s\n", dev->nodename);
1261
1262 xennet_disconnect_backend(info);
1263 return 0;
1264 }
1265
1266 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1267 {
1268 char *s, *e, *macstr;
1269 int i;
1270
1271 macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1272 if (IS_ERR(macstr))
1273 return PTR_ERR(macstr);
1274
1275 for (i = 0; i < ETH_ALEN; i++) {
1276 mac[i] = simple_strtoul(s, &e, 16);
1277 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1278 kfree(macstr);
1279 return -ENOENT;
1280 }
1281 s = e+1;
1282 }
1283
1284 kfree(macstr);
1285 return 0;
1286 }
1287
1288 static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1289 {
1290 struct net_device *dev = dev_id;
1291 struct netfront_info *np = netdev_priv(dev);
1292 unsigned long flags;
1293
1294 spin_lock_irqsave(&np->tx_lock, flags);
1295
1296 if (likely(netif_carrier_ok(dev))) {
1297 xennet_tx_buf_gc(dev);
1298 /* Under tx_lock: protects access to rx shared-ring indexes. */
1299 if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx))
1300 netif_rx_schedule(dev, &np->napi);
1301 }
1302
1303 spin_unlock_irqrestore(&np->tx_lock, flags);
1304
1305 return IRQ_HANDLED;
1306 }
1307
1308 static int setup_netfront(struct xenbus_device *dev, struct netfront_info *info)
1309 {
1310 struct xen_netif_tx_sring *txs;
1311 struct xen_netif_rx_sring *rxs;
1312 int err;
1313 struct net_device *netdev = info->netdev;
1314
1315 info->tx_ring_ref = GRANT_INVALID_REF;
1316 info->rx_ring_ref = GRANT_INVALID_REF;
1317 info->rx.sring = NULL;
1318 info->tx.sring = NULL;
1319 netdev->irq = 0;
1320
1321 err = xen_net_read_mac(dev, netdev->dev_addr);
1322 if (err) {
1323 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
1324 goto fail;
1325 }
1326
1327 txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_KERNEL);
1328 if (!txs) {
1329 err = -ENOMEM;
1330 xenbus_dev_fatal(dev, err, "allocating tx ring page");
1331 goto fail;
1332 }
1333 SHARED_RING_INIT(txs);
1334 FRONT_RING_INIT(&info->tx, txs, PAGE_SIZE);
1335
1336 err = xenbus_grant_ring(dev, virt_to_mfn(txs));
1337 if (err < 0) {
1338 free_page((unsigned long)txs);
1339 goto fail;
1340 }
1341
1342 info->tx_ring_ref = err;
1343 rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_KERNEL);
1344 if (!rxs) {
1345 err = -ENOMEM;
1346 xenbus_dev_fatal(dev, err, "allocating rx ring page");
1347 goto fail;
1348 }
1349 SHARED_RING_INIT(rxs);
1350 FRONT_RING_INIT(&info->rx, rxs, PAGE_SIZE);
1351
1352 err = xenbus_grant_ring(dev, virt_to_mfn(rxs));
1353 if (err < 0) {
1354 free_page((unsigned long)rxs);
1355 goto fail;
1356 }
1357 info->rx_ring_ref = err;
1358
1359 err = xenbus_alloc_evtchn(dev, &info->evtchn);
1360 if (err)
1361 goto fail;
1362
1363 err = bind_evtchn_to_irqhandler(info->evtchn, xennet_interrupt,
1364 IRQF_SAMPLE_RANDOM, netdev->name,
1365 netdev);
1366 if (err < 0)
1367 goto fail;
1368 netdev->irq = err;
1369 return 0;
1370
1371 fail:
1372 return err;
1373 }
1374
1375 /* Common code used when first setting up, and when resuming. */
1376 static int talk_to_backend(struct xenbus_device *dev,
1377 struct netfront_info *info)
1378 {
1379 const char *message;
1380 struct xenbus_transaction xbt;
1381 int err;
1382
1383 /* Create shared ring, alloc event channel. */
1384 err = setup_netfront(dev, info);
1385 if (err)
1386 goto out;
1387
1388 again:
1389 err = xenbus_transaction_start(&xbt);
1390 if (err) {
1391 xenbus_dev_fatal(dev, err, "starting transaction");
1392 goto destroy_ring;
1393 }
1394
1395 err = xenbus_printf(xbt, dev->nodename, "tx-ring-ref", "%u",
1396 info->tx_ring_ref);
1397 if (err) {
1398 message = "writing tx ring-ref";
1399 goto abort_transaction;
1400 }
1401 err = xenbus_printf(xbt, dev->nodename, "rx-ring-ref", "%u",
1402 info->rx_ring_ref);
1403 if (err) {
1404 message = "writing rx ring-ref";
1405 goto abort_transaction;
1406 }
1407 err = xenbus_printf(xbt, dev->nodename,
1408 "event-channel", "%u", info->evtchn);
1409 if (err) {
1410 message = "writing event-channel";
1411 goto abort_transaction;
1412 }
1413
1414 err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
1415 1);
1416 if (err) {
1417 message = "writing request-rx-copy";
1418 goto abort_transaction;
1419 }
1420
1421 err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
1422 if (err) {
1423 message = "writing feature-rx-notify";
1424 goto abort_transaction;
1425 }
1426
1427 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
1428 if (err) {
1429 message = "writing feature-sg";
1430 goto abort_transaction;
1431 }
1432
1433 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
1434 if (err) {
1435 message = "writing feature-gso-tcpv4";
1436 goto abort_transaction;
1437 }
1438
1439 err = xenbus_transaction_end(xbt, 0);
1440 if (err) {
1441 if (err == -EAGAIN)
1442 goto again;
1443 xenbus_dev_fatal(dev, err, "completing transaction");
1444 goto destroy_ring;
1445 }
1446
1447 return 0;
1448
1449 abort_transaction:
1450 xenbus_transaction_end(xbt, 1);
1451 xenbus_dev_fatal(dev, err, "%s", message);
1452 destroy_ring:
1453 xennet_disconnect_backend(info);
1454 out:
1455 return err;
1456 }
1457
1458 static int xennet_set_sg(struct net_device *dev, u32 data)
1459 {
1460 if (data) {
1461 struct netfront_info *np = netdev_priv(dev);
1462 int val;
1463
1464 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, "feature-sg",
1465 "%d", &val) < 0)
1466 val = 0;
1467 if (!val)
1468 return -ENOSYS;
1469 } else if (dev->mtu > ETH_DATA_LEN)
1470 dev->mtu = ETH_DATA_LEN;
1471
1472 return ethtool_op_set_sg(dev, data);
1473 }
1474
1475 static int xennet_set_tso(struct net_device *dev, u32 data)
1476 {
1477 if (data) {
1478 struct netfront_info *np = netdev_priv(dev);
1479 int val;
1480
1481 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1482 "feature-gso-tcpv4", "%d", &val) < 0)
1483 val = 0;
1484 if (!val)
1485 return -ENOSYS;
1486 }
1487
1488 return ethtool_op_set_tso(dev, data);
1489 }
1490
1491 static void xennet_set_features(struct net_device *dev)
1492 {
1493 /* Turn off all GSO bits except ROBUST. */
1494 dev->features &= (1 << NETIF_F_GSO_SHIFT) - 1;
1495 dev->features |= NETIF_F_GSO_ROBUST;
1496 xennet_set_sg(dev, 0);
1497
1498 /* We need checksum offload to enable scatter/gather and TSO. */
1499 if (!(dev->features & NETIF_F_IP_CSUM))
1500 return;
1501
1502 if (!xennet_set_sg(dev, 1))
1503 xennet_set_tso(dev, 1);
1504 }
1505
1506 static int xennet_connect(struct net_device *dev)
1507 {
1508 struct netfront_info *np = netdev_priv(dev);
1509 int i, requeue_idx, err;
1510 struct sk_buff *skb;
1511 grant_ref_t ref;
1512 struct xen_netif_rx_request *req;
1513 unsigned int feature_rx_copy;
1514
1515 err = xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1516 "feature-rx-copy", "%u", &feature_rx_copy);
1517 if (err != 1)
1518 feature_rx_copy = 0;
1519
1520 if (!feature_rx_copy) {
1521 dev_info(&dev->dev,
1522 "backend does not support copying receive path\n");
1523 return -ENODEV;
1524 }
1525
1526 err = talk_to_backend(np->xbdev, np);
1527 if (err)
1528 return err;
1529
1530 xennet_set_features(dev);
1531
1532 spin_lock_bh(&np->rx_lock);
1533 spin_lock_irq(&np->tx_lock);
1534
1535 /* Step 1: Discard all pending TX packet fragments. */
1536 xennet_release_tx_bufs(np);
1537
1538 /* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */
1539 for (requeue_idx = 0, i = 0; i < NET_RX_RING_SIZE; i++) {
1540 if (!np->rx_skbs[i])
1541 continue;
1542
1543 skb = np->rx_skbs[requeue_idx] = xennet_get_rx_skb(np, i);
1544 ref = np->grant_rx_ref[requeue_idx] = xennet_get_rx_ref(np, i);
1545 req = RING_GET_REQUEST(&np->rx, requeue_idx);
1546
1547 gnttab_grant_foreign_access_ref(
1548 ref, np->xbdev->otherend_id,
1549 pfn_to_mfn(page_to_pfn(skb_shinfo(skb)->
1550 frags->page)),
1551 0);
1552 req->gref = ref;
1553 req->id = requeue_idx;
1554
1555 requeue_idx++;
1556 }
1557
1558 np->rx.req_prod_pvt = requeue_idx;
1559
1560 /*
1561 * Step 3: All public and private state should now be sane. Get
1562 * ready to start sending and receiving packets and give the driver
1563 * domain a kick because we've probably just requeued some
1564 * packets.
1565 */
1566 netif_carrier_on(np->netdev);
1567 notify_remote_via_irq(np->netdev->irq);
1568 xennet_tx_buf_gc(dev);
1569 xennet_alloc_rx_buffers(dev);
1570
1571 spin_unlock_irq(&np->tx_lock);
1572 spin_unlock_bh(&np->rx_lock);
1573
1574 return 0;
1575 }
1576
1577 /**
1578 * Callback received when the backend's state changes.
1579 */
1580 static void backend_changed(struct xenbus_device *dev,
1581 enum xenbus_state backend_state)
1582 {
1583 struct netfront_info *np = dev->dev.driver_data;
1584 struct net_device *netdev = np->netdev;
1585
1586 dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
1587
1588 switch (backend_state) {
1589 case XenbusStateInitialising:
1590 case XenbusStateInitialised:
1591 case XenbusStateConnected:
1592 case XenbusStateUnknown:
1593 case XenbusStateClosed:
1594 break;
1595
1596 case XenbusStateInitWait:
1597 if (dev->state != XenbusStateInitialising)
1598 break;
1599 if (xennet_connect(netdev) != 0)
1600 break;
1601 xenbus_switch_state(dev, XenbusStateConnected);
1602 break;
1603
1604 case XenbusStateClosing:
1605 xenbus_frontend_closed(dev);
1606 break;
1607 }
1608 }
1609
1610 static struct ethtool_ops xennet_ethtool_ops =
1611 {
1612 .set_tx_csum = ethtool_op_set_tx_csum,
1613 .set_sg = xennet_set_sg,
1614 .set_tso = xennet_set_tso,
1615 .get_link = ethtool_op_get_link,
1616 };
1617
1618 #ifdef CONFIG_SYSFS
1619 static ssize_t show_rxbuf_min(struct device *dev,
1620 struct device_attribute *attr, char *buf)
1621 {
1622 struct net_device *netdev = to_net_dev(dev);
1623 struct netfront_info *info = netdev_priv(netdev);
1624
1625 return sprintf(buf, "%u\n", info->rx_min_target);
1626 }
1627
1628 static ssize_t store_rxbuf_min(struct device *dev,
1629 struct device_attribute *attr,
1630 const char *buf, size_t len)
1631 {
1632 struct net_device *netdev = to_net_dev(dev);
1633 struct netfront_info *np = netdev_priv(netdev);
1634 char *endp;
1635 unsigned long target;
1636
1637 if (!capable(CAP_NET_ADMIN))
1638 return -EPERM;
1639
1640 target = simple_strtoul(buf, &endp, 0);
1641 if (endp == buf)
1642 return -EBADMSG;
1643
1644 if (target < RX_MIN_TARGET)
1645 target = RX_MIN_TARGET;
1646 if (target > RX_MAX_TARGET)
1647 target = RX_MAX_TARGET;
1648
1649 spin_lock_bh(&np->rx_lock);
1650 if (target > np->rx_max_target)
1651 np->rx_max_target = target;
1652 np->rx_min_target = target;
1653 if (target > np->rx_target)
1654 np->rx_target = target;
1655
1656 xennet_alloc_rx_buffers(netdev);
1657
1658 spin_unlock_bh(&np->rx_lock);
1659 return len;
1660 }
1661
1662 static ssize_t show_rxbuf_max(struct device *dev,
1663 struct device_attribute *attr, char *buf)
1664 {
1665 struct net_device *netdev = to_net_dev(dev);
1666 struct netfront_info *info = netdev_priv(netdev);
1667
1668 return sprintf(buf, "%u\n", info->rx_max_target);
1669 }
1670
1671 static ssize_t store_rxbuf_max(struct device *dev,
1672 struct device_attribute *attr,
1673 const char *buf, size_t len)
1674 {
1675 struct net_device *netdev = to_net_dev(dev);
1676 struct netfront_info *np = netdev_priv(netdev);
1677 char *endp;
1678 unsigned long target;
1679
1680 if (!capable(CAP_NET_ADMIN))
1681 return -EPERM;
1682
1683 target = simple_strtoul(buf, &endp, 0);
1684 if (endp == buf)
1685 return -EBADMSG;
1686
1687 if (target < RX_MIN_TARGET)
1688 target = RX_MIN_TARGET;
1689 if (target > RX_MAX_TARGET)
1690 target = RX_MAX_TARGET;
1691
1692 spin_lock_bh(&np->rx_lock);
1693 if (target < np->rx_min_target)
1694 np->rx_min_target = target;
1695 np->rx_max_target = target;
1696 if (target < np->rx_target)
1697 np->rx_target = target;
1698
1699 xennet_alloc_rx_buffers(netdev);
1700
1701 spin_unlock_bh(&np->rx_lock);
1702 return len;
1703 }
1704
1705 static ssize_t show_rxbuf_cur(struct device *dev,
1706 struct device_attribute *attr, char *buf)
1707 {
1708 struct net_device *netdev = to_net_dev(dev);
1709 struct netfront_info *info = netdev_priv(netdev);
1710
1711 return sprintf(buf, "%u\n", info->rx_target);
1712 }
1713
1714 static struct device_attribute xennet_attrs[] = {
1715 __ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf_min, store_rxbuf_min),
1716 __ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf_max, store_rxbuf_max),
1717 __ATTR(rxbuf_cur, S_IRUGO, show_rxbuf_cur, NULL),
1718 };
1719
1720 static int xennet_sysfs_addif(struct net_device *netdev)
1721 {
1722 int i;
1723 int err;
1724
1725 for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++) {
1726 err = device_create_file(&netdev->dev,
1727 &xennet_attrs[i]);
1728 if (err)
1729 goto fail;
1730 }
1731 return 0;
1732
1733 fail:
1734 while (--i >= 0)
1735 device_remove_file(&netdev->dev, &xennet_attrs[i]);
1736 return err;
1737 }
1738
1739 static void xennet_sysfs_delif(struct net_device *netdev)
1740 {
1741 int i;
1742
1743 for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++)
1744 device_remove_file(&netdev->dev, &xennet_attrs[i]);
1745 }
1746
1747 #endif /* CONFIG_SYSFS */
1748
1749 static struct xenbus_device_id netfront_ids[] = {
1750 { "vif" },
1751 { "" }
1752 };
1753
1754
1755 static int __devexit xennet_remove(struct xenbus_device *dev)
1756 {
1757 struct netfront_info *info = dev->dev.driver_data;
1758
1759 dev_dbg(&dev->dev, "%s\n", dev->nodename);
1760
1761 unregister_netdev(info->netdev);
1762
1763 xennet_disconnect_backend(info);
1764
1765 del_timer_sync(&info->rx_refill_timer);
1766
1767 xennet_sysfs_delif(info->netdev);
1768
1769 free_netdev(info->netdev);
1770
1771 return 0;
1772 }
1773
1774 static struct xenbus_driver netfront = {
1775 .name = "vif",
1776 .owner = THIS_MODULE,
1777 .ids = netfront_ids,
1778 .probe = netfront_probe,
1779 .remove = __devexit_p(xennet_remove),
1780 .resume = netfront_resume,
1781 .otherend_changed = backend_changed,
1782 };
1783
1784 static int __init netif_init(void)
1785 {
1786 if (!is_running_on_xen())
1787 return -ENODEV;
1788
1789 if (is_initial_xendomain())
1790 return 0;
1791
1792 printk(KERN_INFO "Initialising Xen virtual ethernet driver.\n");
1793
1794 return xenbus_register_frontend(&netfront);
1795 }
1796 module_init(netif_init);
1797
1798
1799 static void __exit netif_exit(void)
1800 {
1801 if (is_initial_xendomain())
1802 return;
1803
1804 xenbus_unregister_driver(&netfront);
1805 }
1806 module_exit(netif_exit);
1807
1808 MODULE_DESCRIPTION("Xen virtual network device frontend");
1809 MODULE_LICENSE("GPL");
1810 MODULE_ALIAS("xen:vif");
1811 MODULE_ALIAS("xennet");
Verdex Pro support