?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
x86/amd-iommu: Add per IOMMU reference counting
[linux/fpc-iii.git] / drivers / net / xen-netfront.c
blobbaa051d5bfbe0bcd84bd9f28530ac73f6d4863d8
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 const 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 napi_schedule(&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 + NET_IP_ALIGN,
243 GFP_ATOMIC | __GFP_NOWARN);
244 if (unlikely(!skb))
245 goto no_skb;
246
247 /* Align ip header to a 16 bytes boundary */
248 skb_reserve(skb, NET_IP_ALIGN);
249
250 page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
251 if (!page) {
252 kfree_skb(skb);
253 no_skb:
254 /* Any skbuffs queued for refill? Force them out. */
255 if (i != 0)
256 goto refill;
257 /* Could not allocate any skbuffs. Try again later. */
258 mod_timer(&np->rx_refill_timer,
259 jiffies + (HZ/10));
260 break;
261 }
262
263 skb_shinfo(skb)->frags[0].page = page;
264 skb_shinfo(skb)->nr_frags = 1;
265 __skb_queue_tail(&np->rx_batch, skb);
266 }
267
268 /* Is the batch large enough to be worthwhile? */
269 if (i < (np->rx_target/2)) {
270 if (req_prod > np->rx.sring->req_prod)
271 goto push;
272 return;
273 }
274
275 /* Adjust our fill target if we risked running out of buffers. */
276 if (((req_prod - np->rx.sring->rsp_prod) < (np->rx_target / 4)) &&
277 ((np->rx_target *= 2) > np->rx_max_target))
278 np->rx_target = np->rx_max_target;
279
280 refill:
281 for (i = 0; ; i++) {
282 skb = __skb_dequeue(&np->rx_batch);
283 if (skb == NULL)
284 break;
285
286 skb->dev = dev;
287
288 id = xennet_rxidx(req_prod + i);
289
290 BUG_ON(np->rx_skbs[id]);
291 np->rx_skbs[id] = skb;
292
293 ref = gnttab_claim_grant_reference(&np->gref_rx_head);
294 BUG_ON((signed short)ref < 0);
295 np->grant_rx_ref[id] = ref;
296
297 pfn = page_to_pfn(skb_shinfo(skb)->frags[0].page);
298 vaddr = page_address(skb_shinfo(skb)->frags[0].page);
299
300 req = RING_GET_REQUEST(&np->rx, req_prod + i);
301 gnttab_grant_foreign_access_ref(ref,
302 np->xbdev->otherend_id,
303 pfn_to_mfn(pfn),
304 0);
305
306 req->id = id;
307 req->gref = ref;
308 }
309
310 wmb(); /* barrier so backend seens requests */
311
312 /* Above is a suitable barrier to ensure backend will see requests. */
313 np->rx.req_prod_pvt = req_prod + i;
314 push:
315 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->rx, notify);
316 if (notify)
317 notify_remote_via_irq(np->netdev->irq);
318 }
319
320 static int xennet_open(struct net_device *dev)
321 {
322 struct netfront_info *np = netdev_priv(dev);
323
324 napi_enable(&np->napi);
325
326 spin_lock_bh(&np->rx_lock);
327 if (netif_carrier_ok(dev)) {
328 xennet_alloc_rx_buffers(dev);
329 np->rx.sring->rsp_event = np->rx.rsp_cons + 1;
330 if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx))
331 napi_schedule(&np->napi);
332 }
333 spin_unlock_bh(&np->rx_lock);
334
335 netif_start_queue(dev);
336
337 return 0;
338 }
339
340 static void xennet_tx_buf_gc(struct net_device *dev)
341 {
342 RING_IDX cons, prod;
343 unsigned short id;
344 struct netfront_info *np = netdev_priv(dev);
345 struct sk_buff *skb;
346
347 BUG_ON(!netif_carrier_ok(dev));
348
349 do {
350 prod = np->tx.sring->rsp_prod;
351 rmb(); /* Ensure we see responses up to 'rp'. */
352
353 for (cons = np->tx.rsp_cons; cons != prod; cons++) {
354 struct xen_netif_tx_response *txrsp;
355
356 txrsp = RING_GET_RESPONSE(&np->tx, cons);
357 if (txrsp->status == NETIF_RSP_NULL)
358 continue;
359
360 id = txrsp->id;
361 skb = np->tx_skbs[id].skb;
362 if (unlikely(gnttab_query_foreign_access(
363 np->grant_tx_ref[id]) != 0)) {
364 printk(KERN_ALERT "xennet_tx_buf_gc: warning "
365 "-- grant still in use by backend "
366 "domain.\n");
367 BUG();
368 }
369 gnttab_end_foreign_access_ref(
370 np->grant_tx_ref[id], GNTMAP_readonly);
371 gnttab_release_grant_reference(
372 &np->gref_tx_head, np->grant_tx_ref[id]);
373 np->grant_tx_ref[id] = GRANT_INVALID_REF;
374 add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, id);
375 dev_kfree_skb_irq(skb);
376 }
377
378 np->tx.rsp_cons = prod;
379
380 /*
381 * Set a new event, then check for race with update of tx_cons.
382 * Note that it is essential to schedule a callback, no matter
383 * how few buffers are pending. Even if there is space in the
384 * transmit ring, higher layers may be blocked because too much
385 * data is outstanding: in such cases notification from Xen is
386 * likely to be the only kick that we'll get.
387 */
388 np->tx.sring->rsp_event =
389 prod + ((np->tx.sring->req_prod - prod) >> 1) + 1;
390 mb(); /* update shared area */
391 } while ((cons == prod) && (prod != np->tx.sring->rsp_prod));
392
393 xennet_maybe_wake_tx(dev);
394 }
395
396 static void xennet_make_frags(struct sk_buff *skb, struct net_device *dev,
397 struct xen_netif_tx_request *tx)
398 {
399 struct netfront_info *np = netdev_priv(dev);
400 char *data = skb->data;
401 unsigned long mfn;
402 RING_IDX prod = np->tx.req_prod_pvt;
403 int frags = skb_shinfo(skb)->nr_frags;
404 unsigned int offset = offset_in_page(data);
405 unsigned int len = skb_headlen(skb);
406 unsigned int id;
407 grant_ref_t ref;
408 int i;
409
410 /* While the header overlaps a page boundary (including being
411 larger than a page), split it it into page-sized chunks. */
412 while (len > PAGE_SIZE - offset) {
413 tx->size = PAGE_SIZE - offset;
414 tx->flags |= NETTXF_more_data;
415 len -= tx->size;
416 data += tx->size;
417 offset = 0;
418
419 id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
420 np->tx_skbs[id].skb = skb_get(skb);
421 tx = RING_GET_REQUEST(&np->tx, prod++);
422 tx->id = id;
423 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
424 BUG_ON((signed short)ref < 0);
425
426 mfn = virt_to_mfn(data);
427 gnttab_grant_foreign_access_ref(ref, np->xbdev->otherend_id,
428 mfn, GNTMAP_readonly);
429
430 tx->gref = np->grant_tx_ref[id] = ref;
431 tx->offset = offset;
432 tx->size = len;
433 tx->flags = 0;
434 }
435
436 /* Grant backend access to each skb fragment page. */
437 for (i = 0; i < frags; i++) {
438 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
439
440 tx->flags |= NETTXF_more_data;
441
442 id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
443 np->tx_skbs[id].skb = skb_get(skb);
444 tx = RING_GET_REQUEST(&np->tx, prod++);
445 tx->id = id;
446 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
447 BUG_ON((signed short)ref < 0);
448
449 mfn = pfn_to_mfn(page_to_pfn(frag->page));
450 gnttab_grant_foreign_access_ref(ref, np->xbdev->otherend_id,
451 mfn, GNTMAP_readonly);
452
453 tx->gref = np->grant_tx_ref[id] = ref;
454 tx->offset = frag->page_offset;
455 tx->size = frag->size;
456 tx->flags = 0;
457 }
458
459 np->tx.req_prod_pvt = prod;
460 }
461
462 static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
463 {
464 unsigned short id;
465 struct netfront_info *np = netdev_priv(dev);
466 struct xen_netif_tx_request *tx;
467 struct xen_netif_extra_info *extra;
468 char *data = skb->data;
469 RING_IDX i;
470 grant_ref_t ref;
471 unsigned long mfn;
472 int notify;
473 int frags = skb_shinfo(skb)->nr_frags;
474 unsigned int offset = offset_in_page(data);
475 unsigned int len = skb_headlen(skb);
476
477 frags += DIV_ROUND_UP(offset + len, PAGE_SIZE);
478 if (unlikely(frags > MAX_SKB_FRAGS + 1)) {
479 printk(KERN_ALERT "xennet: skb rides the rocket: %d frags\n",
480 frags);
481 dump_stack();
482 goto drop;
483 }
484
485 spin_lock_irq(&np->tx_lock);
486
487 if (unlikely(!netif_carrier_ok(dev) ||
488 (frags > 1 && !xennet_can_sg(dev)) ||
489 netif_needs_gso(dev, skb))) {
490 spin_unlock_irq(&np->tx_lock);
491 goto drop;
492 }
493
494 i = np->tx.req_prod_pvt;
495
496 id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
497 np->tx_skbs[id].skb = skb;
498
499 tx = RING_GET_REQUEST(&np->tx, i);
500
501 tx->id = id;
502 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
503 BUG_ON((signed short)ref < 0);
504 mfn = virt_to_mfn(data);
505 gnttab_grant_foreign_access_ref(
506 ref, np->xbdev->otherend_id, mfn, GNTMAP_readonly);
507 tx->gref = np->grant_tx_ref[id] = ref;
508 tx->offset = offset;
509 tx->size = len;
510 extra = NULL;
511
512 tx->flags = 0;
513 if (skb->ip_summed == CHECKSUM_PARTIAL)
514 /* local packet? */
515 tx->flags |= NETTXF_csum_blank | NETTXF_data_validated;
516 else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
517 /* remote but checksummed. */
518 tx->flags |= NETTXF_data_validated;
519
520 if (skb_shinfo(skb)->gso_size) {
521 struct xen_netif_extra_info *gso;
522
523 gso = (struct xen_netif_extra_info *)
524 RING_GET_REQUEST(&np->tx, ++i);
525
526 if (extra)
527 extra->flags |= XEN_NETIF_EXTRA_FLAG_MORE;
528 else
529 tx->flags |= NETTXF_extra_info;
530
531 gso->u.gso.size = skb_shinfo(skb)->gso_size;
532 gso->u.gso.type = XEN_NETIF_GSO_TYPE_TCPV4;
533 gso->u.gso.pad = 0;
534 gso->u.gso.features = 0;
535
536 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
537 gso->flags = 0;
538 extra = gso;
539 }
540
541 np->tx.req_prod_pvt = i + 1;
542
543 xennet_make_frags(skb, dev, tx);
544 tx->size = skb->len;
545
546 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->tx, notify);
547 if (notify)
548 notify_remote_via_irq(np->netdev->irq);
549
550 dev->stats.tx_bytes += skb->len;
551 dev->stats.tx_packets++;
552
553 /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
554 xennet_tx_buf_gc(dev);
555
556 if (!netfront_tx_slot_available(np))
557 netif_stop_queue(dev);
558
559 spin_unlock_irq(&np->tx_lock);
560
561 return NETDEV_TX_OK;
562
563 drop:
564 dev->stats.tx_dropped++;
565 dev_kfree_skb(skb);
566 return NETDEV_TX_OK;
567 }
568
569 static int xennet_close(struct net_device *dev)
570 {
571 struct netfront_info *np = netdev_priv(dev);
572 netif_stop_queue(np->netdev);
573 napi_disable(&np->napi);
574 return 0;
575 }
576
577 static void xennet_move_rx_slot(struct netfront_info *np, struct sk_buff *skb,
578 grant_ref_t ref)
579 {
580 int new = xennet_rxidx(np->rx.req_prod_pvt);
581
582 BUG_ON(np->rx_skbs[new]);
583 np->rx_skbs[new] = skb;
584 np->grant_rx_ref[new] = ref;
585 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->id = new;
586 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->gref = ref;
587 np->rx.req_prod_pvt++;
588 }
589
590 static int xennet_get_extras(struct netfront_info *np,
591 struct xen_netif_extra_info *extras,
592 RING_IDX rp)
593
594 {
595 struct xen_netif_extra_info *extra;
596 struct device *dev = &np->netdev->dev;
597 RING_IDX cons = np->rx.rsp_cons;
598 int err = 0;
599
600 do {
601 struct sk_buff *skb;
602 grant_ref_t ref;
603
604 if (unlikely(cons + 1 == rp)) {
605 if (net_ratelimit())
606 dev_warn(dev, "Missing extra info\n");
607 err = -EBADR;
608 break;
609 }
610
611 extra = (struct xen_netif_extra_info *)
612 RING_GET_RESPONSE(&np->rx, ++cons);
613
614 if (unlikely(!extra->type ||
615 extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
616 if (net_ratelimit())
617 dev_warn(dev, "Invalid extra type: %d\n",
618 extra->type);
619 err = -EINVAL;
620 } else {
621 memcpy(&extras[extra->type - 1], extra,
622 sizeof(*extra));
623 }
624
625 skb = xennet_get_rx_skb(np, cons);
626 ref = xennet_get_rx_ref(np, cons);
627 xennet_move_rx_slot(np, skb, ref);
628 } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
629
630 np->rx.rsp_cons = cons;
631 return err;
632 }
633
634 static int xennet_get_responses(struct netfront_info *np,
635 struct netfront_rx_info *rinfo, RING_IDX rp,
636 struct sk_buff_head *list)
637 {
638 struct xen_netif_rx_response *rx = &rinfo->rx;
639 struct xen_netif_extra_info *extras = rinfo->extras;
640 struct device *dev = &np->netdev->dev;
641 RING_IDX cons = np->rx.rsp_cons;
642 struct sk_buff *skb = xennet_get_rx_skb(np, cons);
643 grant_ref_t ref = xennet_get_rx_ref(np, cons);
644 int max = MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD);
645 int frags = 1;
646 int err = 0;
647 unsigned long ret;
648
649 if (rx->flags & NETRXF_extra_info) {
650 err = xennet_get_extras(np, extras, rp);
651 cons = np->rx.rsp_cons;
652 }
653
654 for (;;) {
655 if (unlikely(rx->status < 0 ||
656 rx->offset + rx->status > PAGE_SIZE)) {
657 if (net_ratelimit())
658 dev_warn(dev, "rx->offset: %x, size: %u\n",
659 rx->offset, rx->status);
660 xennet_move_rx_slot(np, skb, ref);
661 err = -EINVAL;
662 goto next;
663 }
664
665 /*
666 * This definitely indicates a bug, either in this driver or in
667 * the backend driver. In future this should flag the bad
668 * situation to the system controller to reboot the backed.
669 */
670 if (ref == GRANT_INVALID_REF) {
671 if (net_ratelimit())
672 dev_warn(dev, "Bad rx response id %d.\n",
673 rx->id);
674 err = -EINVAL;
675 goto next;
676 }
677
678 ret = gnttab_end_foreign_access_ref(ref, 0);
679 BUG_ON(!ret);
680
681 gnttab_release_grant_reference(&np->gref_rx_head, ref);
682
683 __skb_queue_tail(list, skb);
684
685 next:
686 if (!(rx->flags & NETRXF_more_data))
687 break;
688
689 if (cons + frags == rp) {
690 if (net_ratelimit())
691 dev_warn(dev, "Need more frags\n");
692 err = -ENOENT;
693 break;
694 }
695
696 rx = RING_GET_RESPONSE(&np->rx, cons + frags);
697 skb = xennet_get_rx_skb(np, cons + frags);
698 ref = xennet_get_rx_ref(np, cons + frags);
699 frags++;
700 }
701
702 if (unlikely(frags > max)) {
703 if (net_ratelimit())
704 dev_warn(dev, "Too many frags\n");
705 err = -E2BIG;
706 }
707
708 if (unlikely(err))
709 np->rx.rsp_cons = cons + frags;
710
711 return err;
712 }
713
714 static int xennet_set_skb_gso(struct sk_buff *skb,
715 struct xen_netif_extra_info *gso)
716 {
717 if (!gso->u.gso.size) {
718 if (net_ratelimit())
719 printk(KERN_WARNING "GSO size must not be zero.\n");
720 return -EINVAL;
721 }
722
723 /* Currently only TCPv4 S.O. is supported. */
724 if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4) {
725 if (net_ratelimit())
726 printk(KERN_WARNING "Bad GSO type %d.\n", gso->u.gso.type);
727 return -EINVAL;
728 }
729
730 skb_shinfo(skb)->gso_size = gso->u.gso.size;
731 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
732
733 /* Header must be checked, and gso_segs computed. */
734 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
735 skb_shinfo(skb)->gso_segs = 0;
736
737 return 0;
738 }
739
740 static RING_IDX xennet_fill_frags(struct netfront_info *np,
741 struct sk_buff *skb,
742 struct sk_buff_head *list)
743 {
744 struct skb_shared_info *shinfo = skb_shinfo(skb);
745 int nr_frags = shinfo->nr_frags;
746 RING_IDX cons = np->rx.rsp_cons;
747 skb_frag_t *frag = shinfo->frags + nr_frags;
748 struct sk_buff *nskb;
749
750 while ((nskb = __skb_dequeue(list))) {
751 struct xen_netif_rx_response *rx =
752 RING_GET_RESPONSE(&np->rx, ++cons);
753
754 frag->page = skb_shinfo(nskb)->frags[0].page;
755 frag->page_offset = rx->offset;
756 frag->size = rx->status;
757
758 skb->data_len += rx->status;
759
760 skb_shinfo(nskb)->nr_frags = 0;
761 kfree_skb(nskb);
762
763 frag++;
764 nr_frags++;
765 }
766
767 shinfo->nr_frags = nr_frags;
768 return cons;
769 }
770
771 static int skb_checksum_setup(struct sk_buff *skb)
772 {
773 struct iphdr *iph;
774 unsigned char *th;
775 int err = -EPROTO;
776
777 if (skb->protocol != htons(ETH_P_IP))
778 goto out;
779
780 iph = (void *)skb->data;
781 th = skb->data + 4 * iph->ihl;
782 if (th >= skb_tail_pointer(skb))
783 goto out;
784
785 skb->csum_start = th - skb->head;
786 switch (iph->protocol) {
787 case IPPROTO_TCP:
788 skb->csum_offset = offsetof(struct tcphdr, check);
789 break;
790 case IPPROTO_UDP:
791 skb->csum_offset = offsetof(struct udphdr, check);
792 break;
793 default:
794 if (net_ratelimit())
795 printk(KERN_ERR "Attempting to checksum a non-"
796 "TCP/UDP packet, dropping a protocol"
797 " %d packet", iph->protocol);
798 goto out;
799 }
800
801 if ((th + skb->csum_offset + 2) > skb_tail_pointer(skb))
802 goto out;
803
804 err = 0;
805
806 out:
807 return err;
808 }
809
810 static int handle_incoming_queue(struct net_device *dev,
811 struct sk_buff_head *rxq)
812 {
813 int packets_dropped = 0;
814 struct sk_buff *skb;
815
816 while ((skb = __skb_dequeue(rxq)) != NULL) {
817 struct page *page = NETFRONT_SKB_CB(skb)->page;
818 void *vaddr = page_address(page);
819 unsigned offset = NETFRONT_SKB_CB(skb)->offset;
820
821 memcpy(skb->data, vaddr + offset,
822 skb_headlen(skb));
823
824 if (page != skb_shinfo(skb)->frags[0].page)
825 __free_page(page);
826
827 /* Ethernet work: Delayed to here as it peeks the header. */
828 skb->protocol = eth_type_trans(skb, dev);
829
830 if (skb->ip_summed == CHECKSUM_PARTIAL) {
831 if (skb_checksum_setup(skb)) {
832 kfree_skb(skb);
833 packets_dropped++;
834 dev->stats.rx_errors++;
835 continue;
836 }
837 }
838
839 dev->stats.rx_packets++;
840 dev->stats.rx_bytes += skb->len;
841
842 /* Pass it up. */
843 netif_receive_skb(skb);
844 }
845
846 return packets_dropped;
847 }
848
849 static int xennet_poll(struct napi_struct *napi, int budget)
850 {
851 struct netfront_info *np = container_of(napi, struct netfront_info, napi);
852 struct net_device *dev = np->netdev;
853 struct sk_buff *skb;
854 struct netfront_rx_info rinfo;
855 struct xen_netif_rx_response *rx = &rinfo.rx;
856 struct xen_netif_extra_info *extras = rinfo.extras;
857 RING_IDX i, rp;
858 int work_done;
859 struct sk_buff_head rxq;
860 struct sk_buff_head errq;
861 struct sk_buff_head tmpq;
862 unsigned long flags;
863 unsigned int len;
864 int err;
865
866 spin_lock(&np->rx_lock);
867
868 skb_queue_head_init(&rxq);
869 skb_queue_head_init(&errq);
870 skb_queue_head_init(&tmpq);
871
872 rp = np->rx.sring->rsp_prod;
873 rmb(); /* Ensure we see queued responses up to 'rp'. */
874
875 i = np->rx.rsp_cons;
876 work_done = 0;
877 while ((i != rp) && (work_done < budget)) {
878 memcpy(rx, RING_GET_RESPONSE(&np->rx, i), sizeof(*rx));
879 memset(extras, 0, sizeof(rinfo.extras));
880
881 err = xennet_get_responses(np, &rinfo, rp, &tmpq);
882
883 if (unlikely(err)) {
884 err:
885 while ((skb = __skb_dequeue(&tmpq)))
886 __skb_queue_tail(&errq, skb);
887 dev->stats.rx_errors++;
888 i = np->rx.rsp_cons;
889 continue;
890 }
891
892 skb = __skb_dequeue(&tmpq);
893
894 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
895 struct xen_netif_extra_info *gso;
896 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
897
898 if (unlikely(xennet_set_skb_gso(skb, gso))) {
899 __skb_queue_head(&tmpq, skb);
900 np->rx.rsp_cons += skb_queue_len(&tmpq);
901 goto err;
902 }
903 }
904
905 NETFRONT_SKB_CB(skb)->page = skb_shinfo(skb)->frags[0].page;
906 NETFRONT_SKB_CB(skb)->offset = rx->offset;
907
908 len = rx->status;
909 if (len > RX_COPY_THRESHOLD)
910 len = RX_COPY_THRESHOLD;
911 skb_put(skb, len);
912
913 if (rx->status > len) {
914 skb_shinfo(skb)->frags[0].page_offset =
915 rx->offset + len;
916 skb_shinfo(skb)->frags[0].size = rx->status - len;
917 skb->data_len = rx->status - len;
918 } else {
919 skb_shinfo(skb)->frags[0].page = NULL;
920 skb_shinfo(skb)->nr_frags = 0;
921 }
922
923 i = xennet_fill_frags(np, skb, &tmpq);
924
925 /*
926 * Truesize approximates the size of true data plus
927 * any supervisor overheads. Adding hypervisor
928 * overheads has been shown to significantly reduce
929 * achievable bandwidth with the default receive
930 * buffer size. It is therefore not wise to account
931 * for it here.
932 *
933 * After alloc_skb(RX_COPY_THRESHOLD), truesize is set
934 * to RX_COPY_THRESHOLD + the supervisor
935 * overheads. Here, we add the size of the data pulled
936 * in xennet_fill_frags().
937 *
938 * We also adjust for any unused space in the main
939 * data area by subtracting (RX_COPY_THRESHOLD -
940 * len). This is especially important with drivers
941 * which split incoming packets into header and data,
942 * using only 66 bytes of the main data area (see the
943 * e1000 driver for example.) On such systems,
944 * without this last adjustement, our achievable
945 * receive throughout using the standard receive
946 * buffer size was cut by 25%(!!!).
947 */
948 skb->truesize += skb->data_len - (RX_COPY_THRESHOLD - len);
949 skb->len += skb->data_len;
950
951 if (rx->flags & NETRXF_csum_blank)
952 skb->ip_summed = CHECKSUM_PARTIAL;
953 else if (rx->flags & NETRXF_data_validated)
954 skb->ip_summed = CHECKSUM_UNNECESSARY;
955
956 __skb_queue_tail(&rxq, skb);
957
958 np->rx.rsp_cons = ++i;
959 work_done++;
960 }
961
962 __skb_queue_purge(&errq);
963
964 work_done -= handle_incoming_queue(dev, &rxq);
965
966 /* If we get a callback with very few responses, reduce fill target. */
967 /* NB. Note exponential increase, linear decrease. */
968 if (((np->rx.req_prod_pvt - np->rx.sring->rsp_prod) >
969 ((3*np->rx_target) / 4)) &&
970 (--np->rx_target < np->rx_min_target))
971 np->rx_target = np->rx_min_target;
972
973 xennet_alloc_rx_buffers(dev);
974
975 if (work_done < budget) {
976 int more_to_do = 0;
977
978 local_irq_save(flags);
979
980 RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, more_to_do);
981 if (!more_to_do)
982 __napi_complete(napi);
983
984 local_irq_restore(flags);
985 }
986
987 spin_unlock(&np->rx_lock);
988
989 return work_done;
990 }
991
992 static int xennet_change_mtu(struct net_device *dev, int mtu)
993 {
994 int max = xennet_can_sg(dev) ? 65535 - ETH_HLEN : ETH_DATA_LEN;
995
996 if (mtu > max)
997 return -EINVAL;
998 dev->mtu = mtu;
999 return 0;
1000 }
1001
1002 static void xennet_release_tx_bufs(struct netfront_info *np)
1003 {
1004 struct sk_buff *skb;
1005 int i;
1006
1007 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1008 /* Skip over entries which are actually freelist references */
1009 if (skb_entry_is_link(&np->tx_skbs[i]))
1010 continue;
1011
1012 skb = np->tx_skbs[i].skb;
1013 gnttab_end_foreign_access_ref(np->grant_tx_ref[i],
1014 GNTMAP_readonly);
1015 gnttab_release_grant_reference(&np->gref_tx_head,
1016 np->grant_tx_ref[i]);
1017 np->grant_tx_ref[i] = GRANT_INVALID_REF;
1018 add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, i);
1019 dev_kfree_skb_irq(skb);
1020 }
1021 }
1022
1023 static void xennet_release_rx_bufs(struct netfront_info *np)
1024 {
1025 struct mmu_update *mmu = np->rx_mmu;
1026 struct multicall_entry *mcl = np->rx_mcl;
1027 struct sk_buff_head free_list;
1028 struct sk_buff *skb;
1029 unsigned long mfn;
1030 int xfer = 0, noxfer = 0, unused = 0;
1031 int id, ref;
1032
1033 dev_warn(&np->netdev->dev, "%s: fix me for copying receiver.\n",
1034 __func__);
1035 return;
1036
1037 skb_queue_head_init(&free_list);
1038
1039 spin_lock_bh(&np->rx_lock);
1040
1041 for (id = 0; id < NET_RX_RING_SIZE; id++) {
1042 ref = np->grant_rx_ref[id];
1043 if (ref == GRANT_INVALID_REF) {
1044 unused++;
1045 continue;
1046 }
1047
1048 skb = np->rx_skbs[id];
1049 mfn = gnttab_end_foreign_transfer_ref(ref);
1050 gnttab_release_grant_reference(&np->gref_rx_head, ref);
1051 np->grant_rx_ref[id] = GRANT_INVALID_REF;
1052
1053 if (0 == mfn) {
1054 skb_shinfo(skb)->nr_frags = 0;
1055 dev_kfree_skb(skb);
1056 noxfer++;
1057 continue;
1058 }
1059
1060 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1061 /* Remap the page. */
1062 struct page *page = skb_shinfo(skb)->frags[0].page;
1063 unsigned long pfn = page_to_pfn(page);
1064 void *vaddr = page_address(page);
1065
1066 MULTI_update_va_mapping(mcl, (unsigned long)vaddr,
1067 mfn_pte(mfn, PAGE_KERNEL),
1068 0);
1069 mcl++;
1070 mmu->ptr = ((u64)mfn << PAGE_SHIFT)
1071 | MMU_MACHPHYS_UPDATE;
1072 mmu->val = pfn;
1073 mmu++;
1074
1075 set_phys_to_machine(pfn, mfn);
1076 }
1077 __skb_queue_tail(&free_list, skb);
1078 xfer++;
1079 }
1080
1081 dev_info(&np->netdev->dev, "%s: %d xfer, %d noxfer, %d unused\n",
1082 __func__, xfer, noxfer, unused);
1083
1084 if (xfer) {
1085 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1086 /* Do all the remapping work and M2P updates. */
1087 MULTI_mmu_update(mcl, np->rx_mmu, mmu - np->rx_mmu,
1088 NULL, DOMID_SELF);
1089 mcl++;
1090 HYPERVISOR_multicall(np->rx_mcl, mcl - np->rx_mcl);
1091 }
1092 }
1093
1094 __skb_queue_purge(&free_list);
1095
1096 spin_unlock_bh(&np->rx_lock);
1097 }
1098
1099 static void xennet_uninit(struct net_device *dev)
1100 {
1101 struct netfront_info *np = netdev_priv(dev);
1102 xennet_release_tx_bufs(np);
1103 xennet_release_rx_bufs(np);
1104 gnttab_free_grant_references(np->gref_tx_head);
1105 gnttab_free_grant_references(np->gref_rx_head);
1106 }
1107
1108 static const struct net_device_ops xennet_netdev_ops = {
1109 .ndo_open = xennet_open,
1110 .ndo_uninit = xennet_uninit,
1111 .ndo_stop = xennet_close,
1112 .ndo_start_xmit = xennet_start_xmit,
1113 .ndo_change_mtu = xennet_change_mtu,
1114 .ndo_set_mac_address = eth_mac_addr,
1115 .ndo_validate_addr = eth_validate_addr,
1116 };
1117
1118 static struct net_device * __devinit xennet_create_dev(struct xenbus_device *dev)
1119 {
1120 int i, err;
1121 struct net_device *netdev;
1122 struct netfront_info *np;
1123
1124 netdev = alloc_etherdev(sizeof(struct netfront_info));
1125 if (!netdev) {
1126 printk(KERN_WARNING "%s> alloc_etherdev failed.\n",
1127 __func__);
1128 return ERR_PTR(-ENOMEM);
1129 }
1130
1131 np = netdev_priv(netdev);
1132 np->xbdev = dev;
1133
1134 spin_lock_init(&np->tx_lock);
1135 spin_lock_init(&np->rx_lock);
1136
1137 skb_queue_head_init(&np->rx_batch);
1138 np->rx_target = RX_DFL_MIN_TARGET;
1139 np->rx_min_target = RX_DFL_MIN_TARGET;
1140 np->rx_max_target = RX_MAX_TARGET;
1141
1142 init_timer(&np->rx_refill_timer);
1143 np->rx_refill_timer.data = (unsigned long)netdev;
1144 np->rx_refill_timer.function = rx_refill_timeout;
1145
1146 /* Initialise tx_skbs as a free chain containing every entry. */
1147 np->tx_skb_freelist = 0;
1148 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1149 skb_entry_set_link(&np->tx_skbs[i], i+1);
1150 np->grant_tx_ref[i] = GRANT_INVALID_REF;
1151 }
1152
1153 /* Clear out rx_skbs */
1154 for (i = 0; i < NET_RX_RING_SIZE; i++) {
1155 np->rx_skbs[i] = NULL;
1156 np->grant_rx_ref[i] = GRANT_INVALID_REF;
1157 }
1158
1159 /* A grant for every tx ring slot */
1160 if (gnttab_alloc_grant_references(TX_MAX_TARGET,
1161 &np->gref_tx_head) < 0) {
1162 printk(KERN_ALERT "#### netfront can't alloc tx grant refs\n");
1163 err = -ENOMEM;
1164 goto exit;
1165 }
1166 /* A grant for every rx ring slot */
1167 if (gnttab_alloc_grant_references(RX_MAX_TARGET,
1168 &np->gref_rx_head) < 0) {
1169 printk(KERN_ALERT "#### netfront can't alloc rx grant refs\n");
1170 err = -ENOMEM;
1171 goto exit_free_tx;
1172 }
1173
1174 netdev->netdev_ops = &xennet_netdev_ops;
1175
1176 netif_napi_add(netdev, &np->napi, xennet_poll, 64);
1177 netdev->features = NETIF_F_IP_CSUM;
1178
1179 SET_ETHTOOL_OPS(netdev, &xennet_ethtool_ops);
1180 SET_NETDEV_DEV(netdev, &dev->dev);
1181
1182 np->netdev = netdev;
1183
1184 netif_carrier_off(netdev);
1185
1186 return netdev;
1187
1188 exit_free_tx:
1189 gnttab_free_grant_references(np->gref_tx_head);
1190 exit:
1191 free_netdev(netdev);
1192 return ERR_PTR(err);
1193 }
1194
1195 /**
1196 * Entry point to this code when a new device is created. Allocate the basic
1197 * structures and the ring buffers for communication with the backend, and
1198 * inform the backend of the appropriate details for those.
1199 */
1200 static int __devinit netfront_probe(struct xenbus_device *dev,
1201 const struct xenbus_device_id *id)
1202 {
1203 int err;
1204 struct net_device *netdev;
1205 struct netfront_info *info;
1206
1207 netdev = xennet_create_dev(dev);
1208 if (IS_ERR(netdev)) {
1209 err = PTR_ERR(netdev);
1210 xenbus_dev_fatal(dev, err, "creating netdev");
1211 return err;
1212 }
1213
1214 info = netdev_priv(netdev);
1215 dev_set_drvdata(&dev->dev, info);
1216
1217 err = register_netdev(info->netdev);
1218 if (err) {
1219 printk(KERN_WARNING "%s: register_netdev err=%d\n",
1220 __func__, err);
1221 goto fail;
1222 }
1223
1224 err = xennet_sysfs_addif(info->netdev);
1225 if (err) {
1226 unregister_netdev(info->netdev);
1227 printk(KERN_WARNING "%s: add sysfs failed err=%d\n",
1228 __func__, err);
1229 goto fail;
1230 }
1231
1232 return 0;
1233
1234 fail:
1235 free_netdev(netdev);
1236 dev_set_drvdata(&dev->dev, NULL);
1237 return err;
1238 }
1239
1240 static void xennet_end_access(int ref, void *page)
1241 {
1242 /* This frees the page as a side-effect */
1243 if (ref != GRANT_INVALID_REF)
1244 gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1245 }
1246
1247 static void xennet_disconnect_backend(struct netfront_info *info)
1248 {
1249 /* Stop old i/f to prevent errors whilst we rebuild the state. */
1250 spin_lock_bh(&info->rx_lock);
1251 spin_lock_irq(&info->tx_lock);
1252 netif_carrier_off(info->netdev);
1253 spin_unlock_irq(&info->tx_lock);
1254 spin_unlock_bh(&info->rx_lock);
1255
1256 if (info->netdev->irq)
1257 unbind_from_irqhandler(info->netdev->irq, info->netdev);
1258 info->evtchn = info->netdev->irq = 0;
1259
1260 /* End access and free the pages */
1261 xennet_end_access(info->tx_ring_ref, info->tx.sring);
1262 xennet_end_access(info->rx_ring_ref, info->rx.sring);
1263
1264 info->tx_ring_ref = GRANT_INVALID_REF;
1265 info->rx_ring_ref = GRANT_INVALID_REF;
1266 info->tx.sring = NULL;
1267 info->rx.sring = NULL;
1268 }
1269
1270 /**
1271 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1272 * driver restart. We tear down our netif structure and recreate it, but
1273 * leave the device-layer structures intact so that this is transparent to the
1274 * rest of the kernel.
1275 */
1276 static int netfront_resume(struct xenbus_device *dev)
1277 {
1278 struct netfront_info *info = dev_get_drvdata(&dev->dev);
1279
1280 dev_dbg(&dev->dev, "%s\n", dev->nodename);
1281
1282 xennet_disconnect_backend(info);
1283 return 0;
1284 }
1285
1286 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1287 {
1288 char *s, *e, *macstr;
1289 int i;
1290
1291 macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1292 if (IS_ERR(macstr))
1293 return PTR_ERR(macstr);
1294
1295 for (i = 0; i < ETH_ALEN; i++) {
1296 mac[i] = simple_strtoul(s, &e, 16);
1297 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1298 kfree(macstr);
1299 return -ENOENT;
1300 }
1301 s = e+1;
1302 }
1303
1304 kfree(macstr);
1305 return 0;
1306 }
1307
1308 static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1309 {
1310 struct net_device *dev = dev_id;
1311 struct netfront_info *np = netdev_priv(dev);
1312 unsigned long flags;
1313
1314 spin_lock_irqsave(&np->tx_lock, flags);
1315
1316 if (likely(netif_carrier_ok(dev))) {
1317 xennet_tx_buf_gc(dev);
1318 /* Under tx_lock: protects access to rx shared-ring indexes. */
1319 if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx))
1320 napi_schedule(&np->napi);
1321 }
1322
1323 spin_unlock_irqrestore(&np->tx_lock, flags);
1324
1325 return IRQ_HANDLED;
1326 }
1327
1328 static int setup_netfront(struct xenbus_device *dev, struct netfront_info *info)
1329 {
1330 struct xen_netif_tx_sring *txs;
1331 struct xen_netif_rx_sring *rxs;
1332 int err;
1333 struct net_device *netdev = info->netdev;
1334
1335 info->tx_ring_ref = GRANT_INVALID_REF;
1336 info->rx_ring_ref = GRANT_INVALID_REF;
1337 info->rx.sring = NULL;
1338 info->tx.sring = NULL;
1339 netdev->irq = 0;
1340
1341 err = xen_net_read_mac(dev, netdev->dev_addr);
1342 if (err) {
1343 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
1344 goto fail;
1345 }
1346
1347 txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1348 if (!txs) {
1349 err = -ENOMEM;
1350 xenbus_dev_fatal(dev, err, "allocating tx ring page");
1351 goto fail;
1352 }
1353 SHARED_RING_INIT(txs);
1354 FRONT_RING_INIT(&info->tx, txs, PAGE_SIZE);
1355
1356 err = xenbus_grant_ring(dev, virt_to_mfn(txs));
1357 if (err < 0) {
1358 free_page((unsigned long)txs);
1359 goto fail;
1360 }
1361
1362 info->tx_ring_ref = err;
1363 rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1364 if (!rxs) {
1365 err = -ENOMEM;
1366 xenbus_dev_fatal(dev, err, "allocating rx ring page");
1367 goto fail;
1368 }
1369 SHARED_RING_INIT(rxs);
1370 FRONT_RING_INIT(&info->rx, rxs, PAGE_SIZE);
1371
1372 err = xenbus_grant_ring(dev, virt_to_mfn(rxs));
1373 if (err < 0) {
1374 free_page((unsigned long)rxs);
1375 goto fail;
1376 }
1377 info->rx_ring_ref = err;
1378
1379 err = xenbus_alloc_evtchn(dev, &info->evtchn);
1380 if (err)
1381 goto fail;
1382
1383 err = bind_evtchn_to_irqhandler(info->evtchn, xennet_interrupt,
1384 IRQF_SAMPLE_RANDOM, netdev->name,
1385 netdev);
1386 if (err < 0)
1387 goto fail;
1388 netdev->irq = err;
1389 return 0;
1390
1391 fail:
1392 return err;
1393 }
1394
1395 /* Common code used when first setting up, and when resuming. */
1396 static int talk_to_backend(struct xenbus_device *dev,
1397 struct netfront_info *info)
1398 {
1399 const char *message;
1400 struct xenbus_transaction xbt;
1401 int err;
1402
1403 /* Create shared ring, alloc event channel. */
1404 err = setup_netfront(dev, info);
1405 if (err)
1406 goto out;
1407
1408 again:
1409 err = xenbus_transaction_start(&xbt);
1410 if (err) {
1411 xenbus_dev_fatal(dev, err, "starting transaction");
1412 goto destroy_ring;
1413 }
1414
1415 err = xenbus_printf(xbt, dev->nodename, "tx-ring-ref", "%u",
1416 info->tx_ring_ref);
1417 if (err) {
1418 message = "writing tx ring-ref";
1419 goto abort_transaction;
1420 }
1421 err = xenbus_printf(xbt, dev->nodename, "rx-ring-ref", "%u",
1422 info->rx_ring_ref);
1423 if (err) {
1424 message = "writing rx ring-ref";
1425 goto abort_transaction;
1426 }
1427 err = xenbus_printf(xbt, dev->nodename,
1428 "event-channel", "%u", info->evtchn);
1429 if (err) {
1430 message = "writing event-channel";
1431 goto abort_transaction;
1432 }
1433
1434 err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
1435 1);
1436 if (err) {
1437 message = "writing request-rx-copy";
1438 goto abort_transaction;
1439 }
1440
1441 err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
1442 if (err) {
1443 message = "writing feature-rx-notify";
1444 goto abort_transaction;
1445 }
1446
1447 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
1448 if (err) {
1449 message = "writing feature-sg";
1450 goto abort_transaction;
1451 }
1452
1453 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
1454 if (err) {
1455 message = "writing feature-gso-tcpv4";
1456 goto abort_transaction;
1457 }
1458
1459 err = xenbus_transaction_end(xbt, 0);
1460 if (err) {
1461 if (err == -EAGAIN)
1462 goto again;
1463 xenbus_dev_fatal(dev, err, "completing transaction");
1464 goto destroy_ring;
1465 }
1466
1467 return 0;
1468
1469 abort_transaction:
1470 xenbus_transaction_end(xbt, 1);
1471 xenbus_dev_fatal(dev, err, "%s", message);
1472 destroy_ring:
1473 xennet_disconnect_backend(info);
1474 out:
1475 return err;
1476 }
1477
1478 static int xennet_set_sg(struct net_device *dev, u32 data)
1479 {
1480 if (data) {
1481 struct netfront_info *np = netdev_priv(dev);
1482 int val;
1483
1484 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, "feature-sg",
1485 "%d", &val) < 0)
1486 val = 0;
1487 if (!val)
1488 return -ENOSYS;
1489 } else if (dev->mtu > ETH_DATA_LEN)
1490 dev->mtu = ETH_DATA_LEN;
1491
1492 return ethtool_op_set_sg(dev, data);
1493 }
1494
1495 static int xennet_set_tso(struct net_device *dev, u32 data)
1496 {
1497 if (data) {
1498 struct netfront_info *np = netdev_priv(dev);
1499 int val;
1500
1501 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1502 "feature-gso-tcpv4", "%d", &val) < 0)
1503 val = 0;
1504 if (!val)
1505 return -ENOSYS;
1506 }
1507
1508 return ethtool_op_set_tso(dev, data);
1509 }
1510
1511 static void xennet_set_features(struct net_device *dev)
1512 {
1513 /* Turn off all GSO bits except ROBUST. */
1514 dev->features &= ~NETIF_F_GSO_MASK;
1515 dev->features |= NETIF_F_GSO_ROBUST;
1516 xennet_set_sg(dev, 0);
1517
1518 /* We need checksum offload to enable scatter/gather and TSO. */
1519 if (!(dev->features & NETIF_F_IP_CSUM))
1520 return;
1521
1522 if (!xennet_set_sg(dev, 1))
1523 xennet_set_tso(dev, 1);
1524 }
1525
1526 static int xennet_connect(struct net_device *dev)
1527 {
1528 struct netfront_info *np = netdev_priv(dev);
1529 int i, requeue_idx, err;
1530 struct sk_buff *skb;
1531 grant_ref_t ref;
1532 struct xen_netif_rx_request *req;
1533 unsigned int feature_rx_copy;
1534
1535 err = xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1536 "feature-rx-copy", "%u", &feature_rx_copy);
1537 if (err != 1)
1538 feature_rx_copy = 0;
1539
1540 if (!feature_rx_copy) {
1541 dev_info(&dev->dev,
1542 "backend does not support copying receive path\n");
1543 return -ENODEV;
1544 }
1545
1546 err = talk_to_backend(np->xbdev, np);
1547 if (err)
1548 return err;
1549
1550 xennet_set_features(dev);
1551
1552 spin_lock_bh(&np->rx_lock);
1553 spin_lock_irq(&np->tx_lock);
1554
1555 /* Step 1: Discard all pending TX packet fragments. */
1556 xennet_release_tx_bufs(np);
1557
1558 /* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */
1559 for (requeue_idx = 0, i = 0; i < NET_RX_RING_SIZE; i++) {
1560 if (!np->rx_skbs[i])
1561 continue;
1562
1563 skb = np->rx_skbs[requeue_idx] = xennet_get_rx_skb(np, i);
1564 ref = np->grant_rx_ref[requeue_idx] = xennet_get_rx_ref(np, i);
1565 req = RING_GET_REQUEST(&np->rx, requeue_idx);
1566
1567 gnttab_grant_foreign_access_ref(
1568 ref, np->xbdev->otherend_id,
1569 pfn_to_mfn(page_to_pfn(skb_shinfo(skb)->
1570 frags->page)),
1571 0);
1572 req->gref = ref;
1573 req->id = requeue_idx;
1574
1575 requeue_idx++;
1576 }
1577
1578 np->rx.req_prod_pvt = requeue_idx;
1579
1580 /*
1581 * Step 3: All public and private state should now be sane. Get
1582 * ready to start sending and receiving packets and give the driver
1583 * domain a kick because we've probably just requeued some
1584 * packets.
1585 */
1586 netif_carrier_on(np->netdev);
1587 notify_remote_via_irq(np->netdev->irq);
1588 xennet_tx_buf_gc(dev);
1589 xennet_alloc_rx_buffers(dev);
1590
1591 spin_unlock_irq(&np->tx_lock);
1592 spin_unlock_bh(&np->rx_lock);
1593
1594 return 0;
1595 }
1596
1597 /**
1598 * Callback received when the backend's state changes.
1599 */
1600 static void backend_changed(struct xenbus_device *dev,
1601 enum xenbus_state backend_state)
1602 {
1603 struct netfront_info *np = dev_get_drvdata(&dev->dev);
1604 struct net_device *netdev = np->netdev;
1605
1606 dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
1607
1608 switch (backend_state) {
1609 case XenbusStateInitialising:
1610 case XenbusStateInitialised:
1611 case XenbusStateConnected:
1612 case XenbusStateUnknown:
1613 case XenbusStateClosed:
1614 break;
1615
1616 case XenbusStateInitWait:
1617 if (dev->state != XenbusStateInitialising)
1618 break;
1619 if (xennet_connect(netdev) != 0)
1620 break;
1621 xenbus_switch_state(dev, XenbusStateConnected);
1622 break;
1623
1624 case XenbusStateClosing:
1625 xenbus_frontend_closed(dev);
1626 break;
1627 }
1628 }
1629
1630 static const struct ethtool_ops xennet_ethtool_ops =
1631 {
1632 .set_tx_csum = ethtool_op_set_tx_csum,
1633 .set_sg = xennet_set_sg,
1634 .set_tso = xennet_set_tso,
1635 .get_link = ethtool_op_get_link,
1636 };
1637
1638 #ifdef CONFIG_SYSFS
1639 static ssize_t show_rxbuf_min(struct device *dev,
1640 struct device_attribute *attr, char *buf)
1641 {
1642 struct net_device *netdev = to_net_dev(dev);
1643 struct netfront_info *info = netdev_priv(netdev);
1644
1645 return sprintf(buf, "%u\n", info->rx_min_target);
1646 }
1647
1648 static ssize_t store_rxbuf_min(struct device *dev,
1649 struct device_attribute *attr,
1650 const char *buf, size_t len)
1651 {
1652 struct net_device *netdev = to_net_dev(dev);
1653 struct netfront_info *np = netdev_priv(netdev);
1654 char *endp;
1655 unsigned long target;
1656
1657 if (!capable(CAP_NET_ADMIN))
1658 return -EPERM;
1659
1660 target = simple_strtoul(buf, &endp, 0);
1661 if (endp == buf)
1662 return -EBADMSG;
1663
1664 if (target < RX_MIN_TARGET)
1665 target = RX_MIN_TARGET;
1666 if (target > RX_MAX_TARGET)
1667 target = RX_MAX_TARGET;
1668
1669 spin_lock_bh(&np->rx_lock);
1670 if (target > np->rx_max_target)
1671 np->rx_max_target = target;
1672 np->rx_min_target = target;
1673 if (target > np->rx_target)
1674 np->rx_target = target;
1675
1676 xennet_alloc_rx_buffers(netdev);
1677
1678 spin_unlock_bh(&np->rx_lock);
1679 return len;
1680 }
1681
1682 static ssize_t show_rxbuf_max(struct device *dev,
1683 struct device_attribute *attr, char *buf)
1684 {
1685 struct net_device *netdev = to_net_dev(dev);
1686 struct netfront_info *info = netdev_priv(netdev);
1687
1688 return sprintf(buf, "%u\n", info->rx_max_target);
1689 }
1690
1691 static ssize_t store_rxbuf_max(struct device *dev,
1692 struct device_attribute *attr,
1693 const char *buf, size_t len)
1694 {
1695 struct net_device *netdev = to_net_dev(dev);
1696 struct netfront_info *np = netdev_priv(netdev);
1697 char *endp;
1698 unsigned long target;
1699
1700 if (!capable(CAP_NET_ADMIN))
1701 return -EPERM;
1702
1703 target = simple_strtoul(buf, &endp, 0);
1704 if (endp == buf)
1705 return -EBADMSG;
1706
1707 if (target < RX_MIN_TARGET)
1708 target = RX_MIN_TARGET;
1709 if (target > RX_MAX_TARGET)
1710 target = RX_MAX_TARGET;
1711
1712 spin_lock_bh(&np->rx_lock);
1713 if (target < np->rx_min_target)
1714 np->rx_min_target = target;
1715 np->rx_max_target = target;
1716 if (target < np->rx_target)
1717 np->rx_target = target;
1718
1719 xennet_alloc_rx_buffers(netdev);
1720
1721 spin_unlock_bh(&np->rx_lock);
1722 return len;
1723 }
1724
1725 static ssize_t show_rxbuf_cur(struct device *dev,
1726 struct device_attribute *attr, char *buf)
1727 {
1728 struct net_device *netdev = to_net_dev(dev);
1729 struct netfront_info *info = netdev_priv(netdev);
1730
1731 return sprintf(buf, "%u\n", info->rx_target);
1732 }
1733
1734 static struct device_attribute xennet_attrs[] = {
1735 __ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf_min, store_rxbuf_min),
1736 __ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf_max, store_rxbuf_max),
1737 __ATTR(rxbuf_cur, S_IRUGO, show_rxbuf_cur, NULL),
1738 };
1739
1740 static int xennet_sysfs_addif(struct net_device *netdev)
1741 {
1742 int i;
1743 int err;
1744
1745 for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++) {
1746 err = device_create_file(&netdev->dev,
1747 &xennet_attrs[i]);
1748 if (err)
1749 goto fail;
1750 }
1751 return 0;
1752
1753 fail:
1754 while (--i >= 0)
1755 device_remove_file(&netdev->dev, &xennet_attrs[i]);
1756 return err;
1757 }
1758
1759 static void xennet_sysfs_delif(struct net_device *netdev)
1760 {
1761 int i;
1762
1763 for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++)
1764 device_remove_file(&netdev->dev, &xennet_attrs[i]);
1765 }
1766
1767 #endif /* CONFIG_SYSFS */
1768
1769 static struct xenbus_device_id netfront_ids[] = {
1770 { "vif" },
1771 { "" }
1772 };
1773
1774
1775 static int __devexit xennet_remove(struct xenbus_device *dev)
1776 {
1777 struct netfront_info *info = dev_get_drvdata(&dev->dev);
1778
1779 dev_dbg(&dev->dev, "%s\n", dev->nodename);
1780
1781 unregister_netdev(info->netdev);
1782
1783 xennet_disconnect_backend(info);
1784
1785 del_timer_sync(&info->rx_refill_timer);
1786
1787 xennet_sysfs_delif(info->netdev);
1788
1789 free_netdev(info->netdev);
1790
1791 return 0;
1792 }
1793
1794 static struct xenbus_driver netfront_driver = {
1795 .name = "vif",
1796 .owner = THIS_MODULE,
1797 .ids = netfront_ids,
1798 .probe = netfront_probe,
1799 .remove = __devexit_p(xennet_remove),
1800 .resume = netfront_resume,
1801 .otherend_changed = backend_changed,
1802 };
1803
1804 static int __init netif_init(void)
1805 {
1806 if (!xen_domain())
1807 return -ENODEV;
1808
1809 if (xen_initial_domain())
1810 return 0;
1811
1812 printk(KERN_INFO "Initialising Xen virtual ethernet driver.\n");
1813
1814 return xenbus_register_frontend(&netfront_driver);
1815 }
1816 module_init(netif_init);
1817
1818
1819 static void __exit netif_exit(void)
1820 {
1821 if (xen_initial_domain())
1822 return;
1823
1824 xenbus_unregister_driver(&netfront_driver);
1825 }
1826 module_exit(netif_exit);
1827
1828 MODULE_DESCRIPTION("Xen virtual network device frontend");
1829 MODULE_LICENSE("GPL");
1830 MODULE_ALIAS("xen:vif");
1831 MODULE_ALIAS("xennet");
Linux with added FPC-III support