?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
xenbus_client.c: correct exit path for xenbus_map_ring_valloc_hvm
[linux/fpc-iii.git] / drivers / net / xen-netfront.c
blob1db101415069726fc59d26443f773fc0432f52a6
1 /*
2 * Virtual network driver for conversing with remote driver backends.
3 *
4 * Copyright (c) 2002-2005, K A Fraser
5 * Copyright (c) 2005, XenSource Ltd
6 *
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License version 2
9 * as published by the Free Software Foundation; or, when distributed
10 * separately from the Linux kernel or incorporated into other
11 * software packages, subject to the following license:
12 *
13 * Permission is hereby granted, free of charge, to any person obtaining a copy
14 * of this source file (the "Software"), to deal in the Software without
15 * restriction, including without limitation the rights to use, copy, modify,
16 * merge, publish, distribute, sublicense, and/or sell copies of the Software,
17 * and to permit persons to whom the Software is furnished to do so, subject to
18 * the following conditions:
19 *
20 * The above copyright notice and this permission notice shall be included in
21 * all copies or substantial portions of the Software.
22 *
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
24 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
25 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
26 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
27 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
28 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
29 * IN THE SOFTWARE.
30 */
31
32 #include <linux/module.h>
33 #include <linux/kernel.h>
34 #include <linux/netdevice.h>
35 #include <linux/etherdevice.h>
36 #include <linux/skbuff.h>
37 #include <linux/ethtool.h>
38 #include <linux/if_ether.h>
39 #include <net/tcp.h>
40 #include <linux/udp.h>
41 #include <linux/moduleparam.h>
42 #include <linux/mm.h>
43 #include <linux/slab.h>
44 #include <net/ip.h>
45
46 #include <asm/xen/page.h>
47 #include <xen/xen.h>
48 #include <xen/xenbus.h>
49 #include <xen/events.h>
50 #include <xen/page.h>
51 #include <xen/platform_pci.h>
52 #include <xen/grant_table.h>
53
54 #include <xen/interface/io/netif.h>
55 #include <xen/interface/memory.h>
56 #include <xen/interface/grant_table.h>
57
58 static const struct ethtool_ops xennet_ethtool_ops;
59
60 struct netfront_cb {
61 int pull_to;
62 };
63
64 #define NETFRONT_SKB_CB(skb) ((struct netfront_cb *)((skb)->cb))
65
66 #define RX_COPY_THRESHOLD 256
67
68 #define GRANT_INVALID_REF 0
69
70 #define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, PAGE_SIZE)
71 #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, PAGE_SIZE)
72 #define TX_MAX_TARGET min_t(int, NET_TX_RING_SIZE, 256)
73
74 struct netfront_stats {
75 u64 rx_packets;
76 u64 tx_packets;
77 u64 rx_bytes;
78 u64 tx_bytes;
79 struct u64_stats_sync syncp;
80 };
81
82 struct netfront_info {
83 struct list_head list;
84 struct net_device *netdev;
85
86 struct napi_struct napi;
87
88 unsigned int evtchn;
89 struct xenbus_device *xbdev;
90
91 spinlock_t tx_lock;
92 struct xen_netif_tx_front_ring tx;
93 int tx_ring_ref;
94
95 /*
96 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
97 * are linked from tx_skb_freelist through skb_entry.link.
98 *
99 * NB. Freelist index entries are always going to be less than
100 * PAGE_OFFSET, whereas pointers to skbs will always be equal or
101 * greater than PAGE_OFFSET: we use this property to distinguish
102 * them.
103 */
104 union skb_entry {
105 struct sk_buff *skb;
106 unsigned long link;
107 } tx_skbs[NET_TX_RING_SIZE];
108 grant_ref_t gref_tx_head;
109 grant_ref_t grant_tx_ref[NET_TX_RING_SIZE];
110 unsigned tx_skb_freelist;
111
112 spinlock_t rx_lock ____cacheline_aligned_in_smp;
113 struct xen_netif_rx_front_ring rx;
114 int rx_ring_ref;
115
116 /* Receive-ring batched refills. */
117 #define RX_MIN_TARGET 8
118 #define RX_DFL_MIN_TARGET 64
119 #define RX_MAX_TARGET min_t(int, NET_RX_RING_SIZE, 256)
120 unsigned rx_min_target, rx_max_target, rx_target;
121 struct sk_buff_head rx_batch;
122
123 struct timer_list rx_refill_timer;
124
125 struct sk_buff *rx_skbs[NET_RX_RING_SIZE];
126 grant_ref_t gref_rx_head;
127 grant_ref_t grant_rx_ref[NET_RX_RING_SIZE];
128
129 unsigned long rx_pfn_array[NET_RX_RING_SIZE];
130 struct multicall_entry rx_mcl[NET_RX_RING_SIZE+1];
131 struct mmu_update rx_mmu[NET_RX_RING_SIZE];
132
133 /* Statistics */
134 struct netfront_stats __percpu *stats;
135
136 unsigned long rx_gso_checksum_fixup;
137 };
138
139 struct netfront_rx_info {
140 struct xen_netif_rx_response rx;
141 struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
142 };
143
144 static void skb_entry_set_link(union skb_entry *list, unsigned short id)
145 {
146 list->link = id;
147 }
148
149 static int skb_entry_is_link(const union skb_entry *list)
150 {
151 BUILD_BUG_ON(sizeof(list->skb) != sizeof(list->link));
152 return (unsigned long)list->skb < PAGE_OFFSET;
153 }
154
155 /*
156 * Access macros for acquiring freeing slots in tx_skbs[].
157 */
158
159 static void add_id_to_freelist(unsigned *head, union skb_entry *list,
160 unsigned short id)
161 {
162 skb_entry_set_link(&list[id], *head);
163 *head = id;
164 }
165
166 static unsigned short get_id_from_freelist(unsigned *head,
167 union skb_entry *list)
168 {
169 unsigned int id = *head;
170 *head = list[id].link;
171 return id;
172 }
173
174 static int xennet_rxidx(RING_IDX idx)
175 {
176 return idx & (NET_RX_RING_SIZE - 1);
177 }
178
179 static struct sk_buff *xennet_get_rx_skb(struct netfront_info *np,
180 RING_IDX ri)
181 {
182 int i = xennet_rxidx(ri);
183 struct sk_buff *skb = np->rx_skbs[i];
184 np->rx_skbs[i] = NULL;
185 return skb;
186 }
187
188 static grant_ref_t xennet_get_rx_ref(struct netfront_info *np,
189 RING_IDX ri)
190 {
191 int i = xennet_rxidx(ri);
192 grant_ref_t ref = np->grant_rx_ref[i];
193 np->grant_rx_ref[i] = GRANT_INVALID_REF;
194 return ref;
195 }
196
197 #ifdef CONFIG_SYSFS
198 static int xennet_sysfs_addif(struct net_device *netdev);
199 static void xennet_sysfs_delif(struct net_device *netdev);
200 #else /* !CONFIG_SYSFS */
201 #define xennet_sysfs_addif(dev) (0)
202 #define xennet_sysfs_delif(dev) do { } while (0)
203 #endif
204
205 static bool xennet_can_sg(struct net_device *dev)
206 {
207 return dev->features & NETIF_F_SG;
208 }
209
210
211 static void rx_refill_timeout(unsigned long data)
212 {
213 struct net_device *dev = (struct net_device *)data;
214 struct netfront_info *np = netdev_priv(dev);
215 napi_schedule(&np->napi);
216 }
217
218 static int netfront_tx_slot_available(struct netfront_info *np)
219 {
220 return (np->tx.req_prod_pvt - np->tx.rsp_cons) <
221 (TX_MAX_TARGET - MAX_SKB_FRAGS - 2);
222 }
223
224 static void xennet_maybe_wake_tx(struct net_device *dev)
225 {
226 struct netfront_info *np = netdev_priv(dev);
227
228 if (unlikely(netif_queue_stopped(dev)) &&
229 netfront_tx_slot_available(np) &&
230 likely(netif_running(dev)))
231 netif_wake_queue(dev);
232 }
233
234 static void xennet_alloc_rx_buffers(struct net_device *dev)
235 {
236 unsigned short id;
237 struct netfront_info *np = netdev_priv(dev);
238 struct sk_buff *skb;
239 struct page *page;
240 int i, batch_target, notify;
241 RING_IDX req_prod = np->rx.req_prod_pvt;
242 grant_ref_t ref;
243 unsigned long pfn;
244 void *vaddr;
245 struct xen_netif_rx_request *req;
246
247 if (unlikely(!netif_carrier_ok(dev)))
248 return;
249
250 /*
251 * Allocate skbuffs greedily, even though we batch updates to the
252 * receive ring. This creates a less bursty demand on the memory
253 * allocator, so should reduce the chance of failed allocation requests
254 * both for ourself and for other kernel subsystems.
255 */
256 batch_target = np->rx_target - (req_prod - np->rx.rsp_cons);
257 for (i = skb_queue_len(&np->rx_batch); i < batch_target; i++) {
258 skb = __netdev_alloc_skb(dev, RX_COPY_THRESHOLD + NET_IP_ALIGN,
259 GFP_ATOMIC | __GFP_NOWARN);
260 if (unlikely(!skb))
261 goto no_skb;
262
263 /* Align ip header to a 16 bytes boundary */
264 skb_reserve(skb, NET_IP_ALIGN);
265
266 page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
267 if (!page) {
268 kfree_skb(skb);
269 no_skb:
270 /* Any skbuffs queued for refill? Force them out. */
271 if (i != 0)
272 goto refill;
273 /* Could not allocate any skbuffs. Try again later. */
274 mod_timer(&np->rx_refill_timer,
275 jiffies + (HZ/10));
276 break;
277 }
278
279 __skb_fill_page_desc(skb, 0, page, 0, 0);
280 skb_shinfo(skb)->nr_frags = 1;
281 __skb_queue_tail(&np->rx_batch, skb);
282 }
283
284 /* Is the batch large enough to be worthwhile? */
285 if (i < (np->rx_target/2)) {
286 if (req_prod > np->rx.sring->req_prod)
287 goto push;
288 return;
289 }
290
291 /* Adjust our fill target if we risked running out of buffers. */
292 if (((req_prod - np->rx.sring->rsp_prod) < (np->rx_target / 4)) &&
293 ((np->rx_target *= 2) > np->rx_max_target))
294 np->rx_target = np->rx_max_target;
295
296 refill:
297 for (i = 0; ; i++) {
298 skb = __skb_dequeue(&np->rx_batch);
299 if (skb == NULL)
300 break;
301
302 skb->dev = dev;
303
304 id = xennet_rxidx(req_prod + i);
305
306 BUG_ON(np->rx_skbs[id]);
307 np->rx_skbs[id] = skb;
308
309 ref = gnttab_claim_grant_reference(&np->gref_rx_head);
310 BUG_ON((signed short)ref < 0);
311 np->grant_rx_ref[id] = ref;
312
313 pfn = page_to_pfn(skb_frag_page(&skb_shinfo(skb)->frags[0]));
314 vaddr = page_address(skb_frag_page(&skb_shinfo(skb)->frags[0]));
315
316 req = RING_GET_REQUEST(&np->rx, req_prod + i);
317 gnttab_grant_foreign_access_ref(ref,
318 np->xbdev->otherend_id,
319 pfn_to_mfn(pfn),
320 0);
321
322 req->id = id;
323 req->gref = ref;
324 }
325
326 wmb(); /* barrier so backend seens requests */
327
328 /* Above is a suitable barrier to ensure backend will see requests. */
329 np->rx.req_prod_pvt = req_prod + i;
330 push:
331 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->rx, notify);
332 if (notify)
333 notify_remote_via_irq(np->netdev->irq);
334 }
335
336 static int xennet_open(struct net_device *dev)
337 {
338 struct netfront_info *np = netdev_priv(dev);
339
340 napi_enable(&np->napi);
341
342 spin_lock_bh(&np->rx_lock);
343 if (netif_carrier_ok(dev)) {
344 xennet_alloc_rx_buffers(dev);
345 np->rx.sring->rsp_event = np->rx.rsp_cons + 1;
346 if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx))
347 napi_schedule(&np->napi);
348 }
349 spin_unlock_bh(&np->rx_lock);
350
351 netif_start_queue(dev);
352
353 return 0;
354 }
355
356 static void xennet_tx_buf_gc(struct net_device *dev)
357 {
358 RING_IDX cons, prod;
359 unsigned short id;
360 struct netfront_info *np = netdev_priv(dev);
361 struct sk_buff *skb;
362
363 BUG_ON(!netif_carrier_ok(dev));
364
365 do {
366 prod = np->tx.sring->rsp_prod;
367 rmb(); /* Ensure we see responses up to 'rp'. */
368
369 for (cons = np->tx.rsp_cons; cons != prod; cons++) {
370 struct xen_netif_tx_response *txrsp;
371
372 txrsp = RING_GET_RESPONSE(&np->tx, cons);
373 if (txrsp->status == XEN_NETIF_RSP_NULL)
374 continue;
375
376 id = txrsp->id;
377 skb = np->tx_skbs[id].skb;
378 if (unlikely(gnttab_query_foreign_access(
379 np->grant_tx_ref[id]) != 0)) {
380 printk(KERN_ALERT "xennet_tx_buf_gc: warning "
381 "-- grant still in use by backend "
382 "domain.\n");
383 BUG();
384 }
385 gnttab_end_foreign_access_ref(
386 np->grant_tx_ref[id], GNTMAP_readonly);
387 gnttab_release_grant_reference(
388 &np->gref_tx_head, np->grant_tx_ref[id]);
389 np->grant_tx_ref[id] = GRANT_INVALID_REF;
390 add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, id);
391 dev_kfree_skb_irq(skb);
392 }
393
394 np->tx.rsp_cons = prod;
395
396 /*
397 * Set a new event, then check for race with update of tx_cons.
398 * Note that it is essential to schedule a callback, no matter
399 * how few buffers are pending. Even if there is space in the
400 * transmit ring, higher layers may be blocked because too much
401 * data is outstanding: in such cases notification from Xen is
402 * likely to be the only kick that we'll get.
403 */
404 np->tx.sring->rsp_event =
405 prod + ((np->tx.sring->req_prod - prod) >> 1) + 1;
406 mb(); /* update shared area */
407 } while ((cons == prod) && (prod != np->tx.sring->rsp_prod));
408
409 xennet_maybe_wake_tx(dev);
410 }
411
412 static void xennet_make_frags(struct sk_buff *skb, struct net_device *dev,
413 struct xen_netif_tx_request *tx)
414 {
415 struct netfront_info *np = netdev_priv(dev);
416 char *data = skb->data;
417 unsigned long mfn;
418 RING_IDX prod = np->tx.req_prod_pvt;
419 int frags = skb_shinfo(skb)->nr_frags;
420 unsigned int offset = offset_in_page(data);
421 unsigned int len = skb_headlen(skb);
422 unsigned int id;
423 grant_ref_t ref;
424 int i;
425
426 /* While the header overlaps a page boundary (including being
427 larger than a page), split it it into page-sized chunks. */
428 while (len > PAGE_SIZE - offset) {
429 tx->size = PAGE_SIZE - offset;
430 tx->flags |= XEN_NETTXF_more_data;
431 len -= tx->size;
432 data += tx->size;
433 offset = 0;
434
435 id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
436 np->tx_skbs[id].skb = skb_get(skb);
437 tx = RING_GET_REQUEST(&np->tx, prod++);
438 tx->id = id;
439 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
440 BUG_ON((signed short)ref < 0);
441
442 mfn = virt_to_mfn(data);
443 gnttab_grant_foreign_access_ref(ref, np->xbdev->otherend_id,
444 mfn, GNTMAP_readonly);
445
446 tx->gref = np->grant_tx_ref[id] = ref;
447 tx->offset = offset;
448 tx->size = len;
449 tx->flags = 0;
450 }
451
452 /* Grant backend access to each skb fragment page. */
453 for (i = 0; i < frags; i++) {
454 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
455 struct page *page = skb_frag_page(frag);
456
457 len = skb_frag_size(frag);
458 offset = frag->page_offset;
459
460 /* Data must not cross a page boundary. */
461 BUG_ON(len + offset > PAGE_SIZE<<compound_order(page));
462
463 /* Skip unused frames from start of page */
464 page += offset >> PAGE_SHIFT;
465 offset &= ~PAGE_MASK;
466
467 while (len > 0) {
468 unsigned long bytes;
469
470 BUG_ON(offset >= PAGE_SIZE);
471
472 bytes = PAGE_SIZE - offset;
473 if (bytes > len)
474 bytes = len;
475
476 tx->flags |= XEN_NETTXF_more_data;
477
478 id = get_id_from_freelist(&np->tx_skb_freelist,
479 np->tx_skbs);
480 np->tx_skbs[id].skb = skb_get(skb);
481 tx = RING_GET_REQUEST(&np->tx, prod++);
482 tx->id = id;
483 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
484 BUG_ON((signed short)ref < 0);
485
486 mfn = pfn_to_mfn(page_to_pfn(page));
487 gnttab_grant_foreign_access_ref(ref,
488 np->xbdev->otherend_id,
489 mfn, GNTMAP_readonly);
490
491 tx->gref = np->grant_tx_ref[id] = ref;
492 tx->offset = offset;
493 tx->size = bytes;
494 tx->flags = 0;
495
496 offset += bytes;
497 len -= bytes;
498
499 /* Next frame */
500 if (offset == PAGE_SIZE && len) {
501 BUG_ON(!PageCompound(page));
502 page++;
503 offset = 0;
504 }
505 }
506 }
507
508 np->tx.req_prod_pvt = prod;
509 }
510
511 /*
512 * Count how many ring slots are required to send the frags of this
513 * skb. Each frag might be a compound page.
514 */
515 static int xennet_count_skb_frag_slots(struct sk_buff *skb)
516 {
517 int i, frags = skb_shinfo(skb)->nr_frags;
518 int pages = 0;
519
520 for (i = 0; i < frags; i++) {
521 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
522 unsigned long size = skb_frag_size(frag);
523 unsigned long offset = frag->page_offset;
524
525 /* Skip unused frames from start of page */
526 offset &= ~PAGE_MASK;
527
528 pages += PFN_UP(offset + size);
529 }
530
531 return pages;
532 }
533
534 static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
535 {
536 unsigned short id;
537 struct netfront_info *np = netdev_priv(dev);
538 struct netfront_stats *stats = this_cpu_ptr(np->stats);
539 struct xen_netif_tx_request *tx;
540 char *data = skb->data;
541 RING_IDX i;
542 grant_ref_t ref;
543 unsigned long mfn;
544 int notify;
545 int slots;
546 unsigned int offset = offset_in_page(data);
547 unsigned int len = skb_headlen(skb);
548 unsigned long flags;
549
550 /* If skb->len is too big for wire format, drop skb and alert
551 * user about misconfiguration.
552 */
553 if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) {
554 net_alert_ratelimited(
555 "xennet: skb->len = %u, too big for wire format\n",
556 skb->len);
557 goto drop;
558 }
559
560 slots = DIV_ROUND_UP(offset + len, PAGE_SIZE) +
561 xennet_count_skb_frag_slots(skb);
562 if (unlikely(slots > MAX_SKB_FRAGS + 1)) {
563 net_alert_ratelimited(
564 "xennet: skb rides the rocket: %d slots\n", slots);
565 goto drop;
566 }
567
568 spin_lock_irqsave(&np->tx_lock, flags);
569
570 if (unlikely(!netif_carrier_ok(dev) ||
571 (slots > 1 && !xennet_can_sg(dev)) ||
572 netif_needs_gso(skb, netif_skb_features(skb)))) {
573 spin_unlock_irqrestore(&np->tx_lock, flags);
574 goto drop;
575 }
576
577 i = np->tx.req_prod_pvt;
578
579 id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
580 np->tx_skbs[id].skb = skb;
581
582 tx = RING_GET_REQUEST(&np->tx, i);
583
584 tx->id = id;
585 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
586 BUG_ON((signed short)ref < 0);
587 mfn = virt_to_mfn(data);
588 gnttab_grant_foreign_access_ref(
589 ref, np->xbdev->otherend_id, mfn, GNTMAP_readonly);
590 tx->gref = np->grant_tx_ref[id] = ref;
591 tx->offset = offset;
592 tx->size = len;
593
594 tx->flags = 0;
595 if (skb->ip_summed == CHECKSUM_PARTIAL)
596 /* local packet? */
597 tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated;
598 else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
599 /* remote but checksummed. */
600 tx->flags |= XEN_NETTXF_data_validated;
601
602 if (skb_shinfo(skb)->gso_size) {
603 struct xen_netif_extra_info *gso;
604
605 gso = (struct xen_netif_extra_info *)
606 RING_GET_REQUEST(&np->tx, ++i);
607
608 tx->flags |= XEN_NETTXF_extra_info;
609
610 gso->u.gso.size = skb_shinfo(skb)->gso_size;
611 gso->u.gso.type = XEN_NETIF_GSO_TYPE_TCPV4;
612 gso->u.gso.pad = 0;
613 gso->u.gso.features = 0;
614
615 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
616 gso->flags = 0;
617 }
618
619 np->tx.req_prod_pvt = i + 1;
620
621 xennet_make_frags(skb, dev, tx);
622 tx->size = skb->len;
623
624 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->tx, notify);
625 if (notify)
626 notify_remote_via_irq(np->netdev->irq);
627
628 u64_stats_update_begin(&stats->syncp);
629 stats->tx_bytes += skb->len;
630 stats->tx_packets++;
631 u64_stats_update_end(&stats->syncp);
632
633 /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
634 xennet_tx_buf_gc(dev);
635
636 if (!netfront_tx_slot_available(np))
637 netif_stop_queue(dev);
638
639 spin_unlock_irqrestore(&np->tx_lock, flags);
640
641 return NETDEV_TX_OK;
642
643 drop:
644 dev->stats.tx_dropped++;
645 dev_kfree_skb(skb);
646 return NETDEV_TX_OK;
647 }
648
649 static int xennet_close(struct net_device *dev)
650 {
651 struct netfront_info *np = netdev_priv(dev);
652 netif_stop_queue(np->netdev);
653 napi_disable(&np->napi);
654 return 0;
655 }
656
657 static void xennet_move_rx_slot(struct netfront_info *np, struct sk_buff *skb,
658 grant_ref_t ref)
659 {
660 int new = xennet_rxidx(np->rx.req_prod_pvt);
661
662 BUG_ON(np->rx_skbs[new]);
663 np->rx_skbs[new] = skb;
664 np->grant_rx_ref[new] = ref;
665 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->id = new;
666 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->gref = ref;
667 np->rx.req_prod_pvt++;
668 }
669
670 static int xennet_get_extras(struct netfront_info *np,
671 struct xen_netif_extra_info *extras,
672 RING_IDX rp)
673
674 {
675 struct xen_netif_extra_info *extra;
676 struct device *dev = &np->netdev->dev;
677 RING_IDX cons = np->rx.rsp_cons;
678 int err = 0;
679
680 do {
681 struct sk_buff *skb;
682 grant_ref_t ref;
683
684 if (unlikely(cons + 1 == rp)) {
685 if (net_ratelimit())
686 dev_warn(dev, "Missing extra info\n");
687 err = -EBADR;
688 break;
689 }
690
691 extra = (struct xen_netif_extra_info *)
692 RING_GET_RESPONSE(&np->rx, ++cons);
693
694 if (unlikely(!extra->type ||
695 extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
696 if (net_ratelimit())
697 dev_warn(dev, "Invalid extra type: %d\n",
698 extra->type);
699 err = -EINVAL;
700 } else {
701 memcpy(&extras[extra->type - 1], extra,
702 sizeof(*extra));
703 }
704
705 skb = xennet_get_rx_skb(np, cons);
706 ref = xennet_get_rx_ref(np, cons);
707 xennet_move_rx_slot(np, skb, ref);
708 } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
709
710 np->rx.rsp_cons = cons;
711 return err;
712 }
713
714 static int xennet_get_responses(struct netfront_info *np,
715 struct netfront_rx_info *rinfo, RING_IDX rp,
716 struct sk_buff_head *list)
717 {
718 struct xen_netif_rx_response *rx = &rinfo->rx;
719 struct xen_netif_extra_info *extras = rinfo->extras;
720 struct device *dev = &np->netdev->dev;
721 RING_IDX cons = np->rx.rsp_cons;
722 struct sk_buff *skb = xennet_get_rx_skb(np, cons);
723 grant_ref_t ref = xennet_get_rx_ref(np, cons);
724 int max = MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD);
725 int slots = 1;
726 int err = 0;
727 unsigned long ret;
728
729 if (rx->flags & XEN_NETRXF_extra_info) {
730 err = xennet_get_extras(np, extras, rp);
731 cons = np->rx.rsp_cons;
732 }
733
734 for (;;) {
735 if (unlikely(rx->status < 0 ||
736 rx->offset + rx->status > PAGE_SIZE)) {
737 if (net_ratelimit())
738 dev_warn(dev, "rx->offset: %x, size: %u\n",
739 rx->offset, rx->status);
740 xennet_move_rx_slot(np, skb, ref);
741 err = -EINVAL;
742 goto next;
743 }
744
745 /*
746 * This definitely indicates a bug, either in this driver or in
747 * the backend driver. In future this should flag the bad
748 * situation to the system controller to reboot the backend.
749 */
750 if (ref == GRANT_INVALID_REF) {
751 if (net_ratelimit())
752 dev_warn(dev, "Bad rx response id %d.\n",
753 rx->id);
754 err = -EINVAL;
755 goto next;
756 }
757
758 ret = gnttab_end_foreign_access_ref(ref, 0);
759 BUG_ON(!ret);
760
761 gnttab_release_grant_reference(&np->gref_rx_head, ref);
762
763 __skb_queue_tail(list, skb);
764
765 next:
766 if (!(rx->flags & XEN_NETRXF_more_data))
767 break;
768
769 if (cons + slots == rp) {
770 if (net_ratelimit())
771 dev_warn(dev, "Need more slots\n");
772 err = -ENOENT;
773 break;
774 }
775
776 rx = RING_GET_RESPONSE(&np->rx, cons + slots);
777 skb = xennet_get_rx_skb(np, cons + slots);
778 ref = xennet_get_rx_ref(np, cons + slots);
779 slots++;
780 }
781
782 if (unlikely(slots > max)) {
783 if (net_ratelimit())
784 dev_warn(dev, "Too many slots\n");
785 err = -E2BIG;
786 }
787
788 if (unlikely(err))
789 np->rx.rsp_cons = cons + slots;
790
791 return err;
792 }
793
794 static int xennet_set_skb_gso(struct sk_buff *skb,
795 struct xen_netif_extra_info *gso)
796 {
797 if (!gso->u.gso.size) {
798 if (net_ratelimit())
799 printk(KERN_WARNING "GSO size must not be zero.\n");
800 return -EINVAL;
801 }
802
803 /* Currently only TCPv4 S.O. is supported. */
804 if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4) {
805 if (net_ratelimit())
806 printk(KERN_WARNING "Bad GSO type %d.\n", gso->u.gso.type);
807 return -EINVAL;
808 }
809
810 skb_shinfo(skb)->gso_size = gso->u.gso.size;
811 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
812
813 /* Header must be checked, and gso_segs computed. */
814 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
815 skb_shinfo(skb)->gso_segs = 0;
816
817 return 0;
818 }
819
820 static RING_IDX xennet_fill_frags(struct netfront_info *np,
821 struct sk_buff *skb,
822 struct sk_buff_head *list)
823 {
824 struct skb_shared_info *shinfo = skb_shinfo(skb);
825 int nr_frags = shinfo->nr_frags;
826 RING_IDX cons = np->rx.rsp_cons;
827 struct sk_buff *nskb;
828
829 while ((nskb = __skb_dequeue(list))) {
830 struct xen_netif_rx_response *rx =
831 RING_GET_RESPONSE(&np->rx, ++cons);
832 skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0];
833
834 __skb_fill_page_desc(skb, nr_frags,
835 skb_frag_page(nfrag),
836 rx->offset, rx->status);
837
838 skb->data_len += rx->status;
839
840 skb_shinfo(nskb)->nr_frags = 0;
841 kfree_skb(nskb);
842
843 nr_frags++;
844 }
845
846 shinfo->nr_frags = nr_frags;
847 return cons;
848 }
849
850 static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
851 {
852 struct iphdr *iph;
853 unsigned char *th;
854 int err = -EPROTO;
855 int recalculate_partial_csum = 0;
856
857 /*
858 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
859 * peers can fail to set NETRXF_csum_blank when sending a GSO
860 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
861 * recalculate the partial checksum.
862 */
863 if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
864 struct netfront_info *np = netdev_priv(dev);
865 np->rx_gso_checksum_fixup++;
866 skb->ip_summed = CHECKSUM_PARTIAL;
867 recalculate_partial_csum = 1;
868 }
869
870 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
871 if (skb->ip_summed != CHECKSUM_PARTIAL)
872 return 0;
873
874 if (skb->protocol != htons(ETH_P_IP))
875 goto out;
876
877 iph = (void *)skb->data;
878 th = skb->data + 4 * iph->ihl;
879 if (th >= skb_tail_pointer(skb))
880 goto out;
881
882 skb->csum_start = th - skb->head;
883 switch (iph->protocol) {
884 case IPPROTO_TCP:
885 skb->csum_offset = offsetof(struct tcphdr, check);
886
887 if (recalculate_partial_csum) {
888 struct tcphdr *tcph = (struct tcphdr *)th;
889 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
890 skb->len - iph->ihl*4,
891 IPPROTO_TCP, 0);
892 }
893 break;
894 case IPPROTO_UDP:
895 skb->csum_offset = offsetof(struct udphdr, check);
896
897 if (recalculate_partial_csum) {
898 struct udphdr *udph = (struct udphdr *)th;
899 udph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
900 skb->len - iph->ihl*4,
901 IPPROTO_UDP, 0);
902 }
903 break;
904 default:
905 if (net_ratelimit())
906 printk(KERN_ERR "Attempting to checksum a non-"
907 "TCP/UDP packet, dropping a protocol"
908 " %d packet", iph->protocol);
909 goto out;
910 }
911
912 if ((th + skb->csum_offset + 2) > skb_tail_pointer(skb))
913 goto out;
914
915 err = 0;
916
917 out:
918 return err;
919 }
920
921 static int handle_incoming_queue(struct net_device *dev,
922 struct sk_buff_head *rxq)
923 {
924 struct netfront_info *np = netdev_priv(dev);
925 struct netfront_stats *stats = this_cpu_ptr(np->stats);
926 int packets_dropped = 0;
927 struct sk_buff *skb;
928
929 while ((skb = __skb_dequeue(rxq)) != NULL) {
930 int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
931
932 __pskb_pull_tail(skb, pull_to - skb_headlen(skb));
933
934 /* Ethernet work: Delayed to here as it peeks the header. */
935 skb->protocol = eth_type_trans(skb, dev);
936
937 if (checksum_setup(dev, skb)) {
938 kfree_skb(skb);
939 packets_dropped++;
940 dev->stats.rx_errors++;
941 continue;
942 }
943
944 u64_stats_update_begin(&stats->syncp);
945 stats->rx_packets++;
946 stats->rx_bytes += skb->len;
947 u64_stats_update_end(&stats->syncp);
948
949 /* Pass it up. */
950 netif_receive_skb(skb);
951 }
952
953 return packets_dropped;
954 }
955
956 static int xennet_poll(struct napi_struct *napi, int budget)
957 {
958 struct netfront_info *np = container_of(napi, struct netfront_info, napi);
959 struct net_device *dev = np->netdev;
960 struct sk_buff *skb;
961 struct netfront_rx_info rinfo;
962 struct xen_netif_rx_response *rx = &rinfo.rx;
963 struct xen_netif_extra_info *extras = rinfo.extras;
964 RING_IDX i, rp;
965 int work_done;
966 struct sk_buff_head rxq;
967 struct sk_buff_head errq;
968 struct sk_buff_head tmpq;
969 unsigned long flags;
970 int err;
971
972 spin_lock(&np->rx_lock);
973
974 skb_queue_head_init(&rxq);
975 skb_queue_head_init(&errq);
976 skb_queue_head_init(&tmpq);
977
978 rp = np->rx.sring->rsp_prod;
979 rmb(); /* Ensure we see queued responses up to 'rp'. */
980
981 i = np->rx.rsp_cons;
982 work_done = 0;
983 while ((i != rp) && (work_done < budget)) {
984 memcpy(rx, RING_GET_RESPONSE(&np->rx, i), sizeof(*rx));
985 memset(extras, 0, sizeof(rinfo.extras));
986
987 err = xennet_get_responses(np, &rinfo, rp, &tmpq);
988
989 if (unlikely(err)) {
990 err:
991 while ((skb = __skb_dequeue(&tmpq)))
992 __skb_queue_tail(&errq, skb);
993 dev->stats.rx_errors++;
994 i = np->rx.rsp_cons;
995 continue;
996 }
997
998 skb = __skb_dequeue(&tmpq);
999
1000 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1001 struct xen_netif_extra_info *gso;
1002 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1003
1004 if (unlikely(xennet_set_skb_gso(skb, gso))) {
1005 __skb_queue_head(&tmpq, skb);
1006 np->rx.rsp_cons += skb_queue_len(&tmpq);
1007 goto err;
1008 }
1009 }
1010
1011 NETFRONT_SKB_CB(skb)->pull_to = rx->status;
1012 if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD)
1013 NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD;
1014
1015 skb_shinfo(skb)->frags[0].page_offset = rx->offset;
1016 skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status);
1017 skb->data_len = rx->status;
1018
1019 i = xennet_fill_frags(np, skb, &tmpq);
1020
1021 /*
1022 * Truesize is the actual allocation size, even if the
1023 * allocation is only partially used.
1024 */
1025 skb->truesize += PAGE_SIZE * skb_shinfo(skb)->nr_frags;
1026 skb->len += skb->data_len;
1027
1028 if (rx->flags & XEN_NETRXF_csum_blank)
1029 skb->ip_summed = CHECKSUM_PARTIAL;
1030 else if (rx->flags & XEN_NETRXF_data_validated)
1031 skb->ip_summed = CHECKSUM_UNNECESSARY;
1032
1033 __skb_queue_tail(&rxq, skb);
1034
1035 np->rx.rsp_cons = ++i;
1036 work_done++;
1037 }
1038
1039 __skb_queue_purge(&errq);
1040
1041 work_done -= handle_incoming_queue(dev, &rxq);
1042
1043 /* If we get a callback with very few responses, reduce fill target. */
1044 /* NB. Note exponential increase, linear decrease. */
1045 if (((np->rx.req_prod_pvt - np->rx.sring->rsp_prod) >
1046 ((3*np->rx_target) / 4)) &&
1047 (--np->rx_target < np->rx_min_target))
1048 np->rx_target = np->rx_min_target;
1049
1050 xennet_alloc_rx_buffers(dev);
1051
1052 if (work_done < budget) {
1053 int more_to_do = 0;
1054
1055 local_irq_save(flags);
1056
1057 RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, more_to_do);
1058 if (!more_to_do)
1059 __napi_complete(napi);
1060
1061 local_irq_restore(flags);
1062 }
1063
1064 spin_unlock(&np->rx_lock);
1065
1066 return work_done;
1067 }
1068
1069 static int xennet_change_mtu(struct net_device *dev, int mtu)
1070 {
1071 int max = xennet_can_sg(dev) ?
1072 XEN_NETIF_MAX_TX_SIZE - MAX_TCP_HEADER : ETH_DATA_LEN;
1073
1074 if (mtu > max)
1075 return -EINVAL;
1076 dev->mtu = mtu;
1077 return 0;
1078 }
1079
1080 static struct rtnl_link_stats64 *xennet_get_stats64(struct net_device *dev,
1081 struct rtnl_link_stats64 *tot)
1082 {
1083 struct netfront_info *np = netdev_priv(dev);
1084 int cpu;
1085
1086 for_each_possible_cpu(cpu) {
1087 struct netfront_stats *stats = per_cpu_ptr(np->stats, cpu);
1088 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1089 unsigned int start;
1090
1091 do {
1092 start = u64_stats_fetch_begin_bh(&stats->syncp);
1093
1094 rx_packets = stats->rx_packets;
1095 tx_packets = stats->tx_packets;
1096 rx_bytes = stats->rx_bytes;
1097 tx_bytes = stats->tx_bytes;
1098 } while (u64_stats_fetch_retry_bh(&stats->syncp, start));
1099
1100 tot->rx_packets += rx_packets;
1101 tot->tx_packets += tx_packets;
1102 tot->rx_bytes += rx_bytes;
1103 tot->tx_bytes += tx_bytes;
1104 }
1105
1106 tot->rx_errors = dev->stats.rx_errors;
1107 tot->tx_dropped = dev->stats.tx_dropped;
1108
1109 return tot;
1110 }
1111
1112 static void xennet_release_tx_bufs(struct netfront_info *np)
1113 {
1114 struct sk_buff *skb;
1115 int i;
1116
1117 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1118 /* Skip over entries which are actually freelist references */
1119 if (skb_entry_is_link(&np->tx_skbs[i]))
1120 continue;
1121
1122 skb = np->tx_skbs[i].skb;
1123 gnttab_end_foreign_access_ref(np->grant_tx_ref[i],
1124 GNTMAP_readonly);
1125 gnttab_release_grant_reference(&np->gref_tx_head,
1126 np->grant_tx_ref[i]);
1127 np->grant_tx_ref[i] = GRANT_INVALID_REF;
1128 add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, i);
1129 dev_kfree_skb_irq(skb);
1130 }
1131 }
1132
1133 static void xennet_release_rx_bufs(struct netfront_info *np)
1134 {
1135 struct mmu_update *mmu = np->rx_mmu;
1136 struct multicall_entry *mcl = np->rx_mcl;
1137 struct sk_buff_head free_list;
1138 struct sk_buff *skb;
1139 unsigned long mfn;
1140 int xfer = 0, noxfer = 0, unused = 0;
1141 int id, ref;
1142
1143 dev_warn(&np->netdev->dev, "%s: fix me for copying receiver.\n",
1144 __func__);
1145 return;
1146
1147 skb_queue_head_init(&free_list);
1148
1149 spin_lock_bh(&np->rx_lock);
1150
1151 for (id = 0; id < NET_RX_RING_SIZE; id++) {
1152 ref = np->grant_rx_ref[id];
1153 if (ref == GRANT_INVALID_REF) {
1154 unused++;
1155 continue;
1156 }
1157
1158 skb = np->rx_skbs[id];
1159 mfn = gnttab_end_foreign_transfer_ref(ref);
1160 gnttab_release_grant_reference(&np->gref_rx_head, ref);
1161 np->grant_rx_ref[id] = GRANT_INVALID_REF;
1162
1163 if (0 == mfn) {
1164 skb_shinfo(skb)->nr_frags = 0;
1165 dev_kfree_skb(skb);
1166 noxfer++;
1167 continue;
1168 }
1169
1170 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1171 /* Remap the page. */
1172 const struct page *page =
1173 skb_frag_page(&skb_shinfo(skb)->frags[0]);
1174 unsigned long pfn = page_to_pfn(page);
1175 void *vaddr = page_address(page);
1176
1177 MULTI_update_va_mapping(mcl, (unsigned long)vaddr,
1178 mfn_pte(mfn, PAGE_KERNEL),
1179 0);
1180 mcl++;
1181 mmu->ptr = ((u64)mfn << PAGE_SHIFT)
1182 | MMU_MACHPHYS_UPDATE;
1183 mmu->val = pfn;
1184 mmu++;
1185
1186 set_phys_to_machine(pfn, mfn);
1187 }
1188 __skb_queue_tail(&free_list, skb);
1189 xfer++;
1190 }
1191
1192 dev_info(&np->netdev->dev, "%s: %d xfer, %d noxfer, %d unused\n",
1193 __func__, xfer, noxfer, unused);
1194
1195 if (xfer) {
1196 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1197 /* Do all the remapping work and M2P updates. */
1198 MULTI_mmu_update(mcl, np->rx_mmu, mmu - np->rx_mmu,
1199 NULL, DOMID_SELF);
1200 mcl++;
1201 HYPERVISOR_multicall(np->rx_mcl, mcl - np->rx_mcl);
1202 }
1203 }
1204
1205 __skb_queue_purge(&free_list);
1206
1207 spin_unlock_bh(&np->rx_lock);
1208 }
1209
1210 static void xennet_uninit(struct net_device *dev)
1211 {
1212 struct netfront_info *np = netdev_priv(dev);
1213 xennet_release_tx_bufs(np);
1214 xennet_release_rx_bufs(np);
1215 gnttab_free_grant_references(np->gref_tx_head);
1216 gnttab_free_grant_references(np->gref_rx_head);
1217 }
1218
1219 static netdev_features_t xennet_fix_features(struct net_device *dev,
1220 netdev_features_t features)
1221 {
1222 struct netfront_info *np = netdev_priv(dev);
1223 int val;
1224
1225 if (features & NETIF_F_SG) {
1226 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, "feature-sg",
1227 "%d", &val) < 0)
1228 val = 0;
1229
1230 if (!val)
1231 features &= ~NETIF_F_SG;
1232 }
1233
1234 if (features & NETIF_F_TSO) {
1235 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1236 "feature-gso-tcpv4", "%d", &val) < 0)
1237 val = 0;
1238
1239 if (!val)
1240 features &= ~NETIF_F_TSO;
1241 }
1242
1243 return features;
1244 }
1245
1246 static int xennet_set_features(struct net_device *dev,
1247 netdev_features_t features)
1248 {
1249 if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1250 netdev_info(dev, "Reducing MTU because no SG offload");
1251 dev->mtu = ETH_DATA_LEN;
1252 }
1253
1254 return 0;
1255 }
1256
1257 static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1258 {
1259 struct net_device *dev = dev_id;
1260 struct netfront_info *np = netdev_priv(dev);
1261 unsigned long flags;
1262
1263 spin_lock_irqsave(&np->tx_lock, flags);
1264
1265 if (likely(netif_carrier_ok(dev))) {
1266 xennet_tx_buf_gc(dev);
1267 /* Under tx_lock: protects access to rx shared-ring indexes. */
1268 if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx))
1269 napi_schedule(&np->napi);
1270 }
1271
1272 spin_unlock_irqrestore(&np->tx_lock, flags);
1273
1274 return IRQ_HANDLED;
1275 }
1276
1277 #ifdef CONFIG_NET_POLL_CONTROLLER
1278 static void xennet_poll_controller(struct net_device *dev)
1279 {
1280 xennet_interrupt(0, dev);
1281 }
1282 #endif
1283
1284 static const struct net_device_ops xennet_netdev_ops = {
1285 .ndo_open = xennet_open,
1286 .ndo_uninit = xennet_uninit,
1287 .ndo_stop = xennet_close,
1288 .ndo_start_xmit = xennet_start_xmit,
1289 .ndo_change_mtu = xennet_change_mtu,
1290 .ndo_get_stats64 = xennet_get_stats64,
1291 .ndo_set_mac_address = eth_mac_addr,
1292 .ndo_validate_addr = eth_validate_addr,
1293 .ndo_fix_features = xennet_fix_features,
1294 .ndo_set_features = xennet_set_features,
1295 #ifdef CONFIG_NET_POLL_CONTROLLER
1296 .ndo_poll_controller = xennet_poll_controller,
1297 #endif
1298 };
1299
1300 static struct net_device *xennet_create_dev(struct xenbus_device *dev)
1301 {
1302 int i, err;
1303 struct net_device *netdev;
1304 struct netfront_info *np;
1305
1306 netdev = alloc_etherdev(sizeof(struct netfront_info));
1307 if (!netdev)
1308 return ERR_PTR(-ENOMEM);
1309
1310 np = netdev_priv(netdev);
1311 np->xbdev = dev;
1312
1313 spin_lock_init(&np->tx_lock);
1314 spin_lock_init(&np->rx_lock);
1315
1316 skb_queue_head_init(&np->rx_batch);
1317 np->rx_target = RX_DFL_MIN_TARGET;
1318 np->rx_min_target = RX_DFL_MIN_TARGET;
1319 np->rx_max_target = RX_MAX_TARGET;
1320
1321 init_timer(&np->rx_refill_timer);
1322 np->rx_refill_timer.data = (unsigned long)netdev;
1323 np->rx_refill_timer.function = rx_refill_timeout;
1324
1325 err = -ENOMEM;
1326 np->stats = alloc_percpu(struct netfront_stats);
1327 if (np->stats == NULL)
1328 goto exit;
1329
1330 /* Initialise tx_skbs as a free chain containing every entry. */
1331 np->tx_skb_freelist = 0;
1332 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1333 skb_entry_set_link(&np->tx_skbs[i], i+1);
1334 np->grant_tx_ref[i] = GRANT_INVALID_REF;
1335 }
1336
1337 /* Clear out rx_skbs */
1338 for (i = 0; i < NET_RX_RING_SIZE; i++) {
1339 np->rx_skbs[i] = NULL;
1340 np->grant_rx_ref[i] = GRANT_INVALID_REF;
1341 }
1342
1343 /* A grant for every tx ring slot */
1344 if (gnttab_alloc_grant_references(TX_MAX_TARGET,
1345 &np->gref_tx_head) < 0) {
1346 printk(KERN_ALERT "#### netfront can't alloc tx grant refs\n");
1347 err = -ENOMEM;
1348 goto exit_free_stats;
1349 }
1350 /* A grant for every rx ring slot */
1351 if (gnttab_alloc_grant_references(RX_MAX_TARGET,
1352 &np->gref_rx_head) < 0) {
1353 printk(KERN_ALERT "#### netfront can't alloc rx grant refs\n");
1354 err = -ENOMEM;
1355 goto exit_free_tx;
1356 }
1357
1358 netdev->netdev_ops = &xennet_netdev_ops;
1359
1360 netif_napi_add(netdev, &np->napi, xennet_poll, 64);
1361 netdev->features = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1362 NETIF_F_GSO_ROBUST;
1363 netdev->hw_features = NETIF_F_IP_CSUM | NETIF_F_SG | NETIF_F_TSO;
1364
1365 /*
1366 * Assume that all hw features are available for now. This set
1367 * will be adjusted by the call to netdev_update_features() in
1368 * xennet_connect() which is the earliest point where we can
1369 * negotiate with the backend regarding supported features.
1370 */
1371 netdev->features |= netdev->hw_features;
1372
1373 SET_ETHTOOL_OPS(netdev, &xennet_ethtool_ops);
1374 SET_NETDEV_DEV(netdev, &dev->dev);
1375
1376 netif_set_gso_max_size(netdev, XEN_NETIF_MAX_TX_SIZE - MAX_TCP_HEADER);
1377
1378 np->netdev = netdev;
1379
1380 netif_carrier_off(netdev);
1381
1382 return netdev;
1383
1384 exit_free_tx:
1385 gnttab_free_grant_references(np->gref_tx_head);
1386 exit_free_stats:
1387 free_percpu(np->stats);
1388 exit:
1389 free_netdev(netdev);
1390 return ERR_PTR(err);
1391 }
1392
1393 /**
1394 * Entry point to this code when a new device is created. Allocate the basic
1395 * structures and the ring buffers for communication with the backend, and
1396 * inform the backend of the appropriate details for those.
1397 */
1398 static int netfront_probe(struct xenbus_device *dev,
1399 const struct xenbus_device_id *id)
1400 {
1401 int err;
1402 struct net_device *netdev;
1403 struct netfront_info *info;
1404
1405 netdev = xennet_create_dev(dev);
1406 if (IS_ERR(netdev)) {
1407 err = PTR_ERR(netdev);
1408 xenbus_dev_fatal(dev, err, "creating netdev");
1409 return err;
1410 }
1411
1412 info = netdev_priv(netdev);
1413 dev_set_drvdata(&dev->dev, info);
1414
1415 err = register_netdev(info->netdev);
1416 if (err) {
1417 printk(KERN_WARNING "%s: register_netdev err=%d\n",
1418 __func__, err);
1419 goto fail;
1420 }
1421
1422 err = xennet_sysfs_addif(info->netdev);
1423 if (err) {
1424 unregister_netdev(info->netdev);
1425 printk(KERN_WARNING "%s: add sysfs failed err=%d\n",
1426 __func__, err);
1427 goto fail;
1428 }
1429
1430 return 0;
1431
1432 fail:
1433 free_netdev(netdev);
1434 dev_set_drvdata(&dev->dev, NULL);
1435 return err;
1436 }
1437
1438 static void xennet_end_access(int ref, void *page)
1439 {
1440 /* This frees the page as a side-effect */
1441 if (ref != GRANT_INVALID_REF)
1442 gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1443 }
1444
1445 static void xennet_disconnect_backend(struct netfront_info *info)
1446 {
1447 /* Stop old i/f to prevent errors whilst we rebuild the state. */
1448 spin_lock_bh(&info->rx_lock);
1449 spin_lock_irq(&info->tx_lock);
1450 netif_carrier_off(info->netdev);
1451 spin_unlock_irq(&info->tx_lock);
1452 spin_unlock_bh(&info->rx_lock);
1453
1454 if (info->netdev->irq)
1455 unbind_from_irqhandler(info->netdev->irq, info->netdev);
1456 info->evtchn = info->netdev->irq = 0;
1457
1458 /* End access and free the pages */
1459 xennet_end_access(info->tx_ring_ref, info->tx.sring);
1460 xennet_end_access(info->rx_ring_ref, info->rx.sring);
1461
1462 info->tx_ring_ref = GRANT_INVALID_REF;
1463 info->rx_ring_ref = GRANT_INVALID_REF;
1464 info->tx.sring = NULL;
1465 info->rx.sring = NULL;
1466 }
1467
1468 /**
1469 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1470 * driver restart. We tear down our netif structure and recreate it, but
1471 * leave the device-layer structures intact so that this is transparent to the
1472 * rest of the kernel.
1473 */
1474 static int netfront_resume(struct xenbus_device *dev)
1475 {
1476 struct netfront_info *info = dev_get_drvdata(&dev->dev);
1477
1478 dev_dbg(&dev->dev, "%s\n", dev->nodename);
1479
1480 xennet_disconnect_backend(info);
1481 return 0;
1482 }
1483
1484 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1485 {
1486 char *s, *e, *macstr;
1487 int i;
1488
1489 macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1490 if (IS_ERR(macstr))
1491 return PTR_ERR(macstr);
1492
1493 for (i = 0; i < ETH_ALEN; i++) {
1494 mac[i] = simple_strtoul(s, &e, 16);
1495 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1496 kfree(macstr);
1497 return -ENOENT;
1498 }
1499 s = e+1;
1500 }
1501
1502 kfree(macstr);
1503 return 0;
1504 }
1505
1506 static int setup_netfront(struct xenbus_device *dev, struct netfront_info *info)
1507 {
1508 struct xen_netif_tx_sring *txs;
1509 struct xen_netif_rx_sring *rxs;
1510 int err;
1511 struct net_device *netdev = info->netdev;
1512
1513 info->tx_ring_ref = GRANT_INVALID_REF;
1514 info->rx_ring_ref = GRANT_INVALID_REF;
1515 info->rx.sring = NULL;
1516 info->tx.sring = NULL;
1517 netdev->irq = 0;
1518
1519 err = xen_net_read_mac(dev, netdev->dev_addr);
1520 if (err) {
1521 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
1522 goto fail;
1523 }
1524
1525 txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1526 if (!txs) {
1527 err = -ENOMEM;
1528 xenbus_dev_fatal(dev, err, "allocating tx ring page");
1529 goto fail;
1530 }
1531 SHARED_RING_INIT(txs);
1532 FRONT_RING_INIT(&info->tx, txs, PAGE_SIZE);
1533
1534 err = xenbus_grant_ring(dev, virt_to_mfn(txs));
1535 if (err < 0) {
1536 free_page((unsigned long)txs);
1537 goto fail;
1538 }
1539
1540 info->tx_ring_ref = err;
1541 rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1542 if (!rxs) {
1543 err = -ENOMEM;
1544 xenbus_dev_fatal(dev, err, "allocating rx ring page");
1545 goto fail;
1546 }
1547 SHARED_RING_INIT(rxs);
1548 FRONT_RING_INIT(&info->rx, rxs, PAGE_SIZE);
1549
1550 err = xenbus_grant_ring(dev, virt_to_mfn(rxs));
1551 if (err < 0) {
1552 free_page((unsigned long)rxs);
1553 goto fail;
1554 }
1555 info->rx_ring_ref = err;
1556
1557 err = xenbus_alloc_evtchn(dev, &info->evtchn);
1558 if (err)
1559 goto fail;
1560
1561 err = bind_evtchn_to_irqhandler(info->evtchn, xennet_interrupt,
1562 0, netdev->name, netdev);
1563 if (err < 0)
1564 goto fail;
1565 netdev->irq = err;
1566 return 0;
1567
1568 fail:
1569 return err;
1570 }
1571
1572 /* Common code used when first setting up, and when resuming. */
1573 static int talk_to_netback(struct xenbus_device *dev,
1574 struct netfront_info *info)
1575 {
1576 const char *message;
1577 struct xenbus_transaction xbt;
1578 int err;
1579
1580 /* Create shared ring, alloc event channel. */
1581 err = setup_netfront(dev, info);
1582 if (err)
1583 goto out;
1584
1585 again:
1586 err = xenbus_transaction_start(&xbt);
1587 if (err) {
1588 xenbus_dev_fatal(dev, err, "starting transaction");
1589 goto destroy_ring;
1590 }
1591
1592 err = xenbus_printf(xbt, dev->nodename, "tx-ring-ref", "%u",
1593 info->tx_ring_ref);
1594 if (err) {
1595 message = "writing tx ring-ref";
1596 goto abort_transaction;
1597 }
1598 err = xenbus_printf(xbt, dev->nodename, "rx-ring-ref", "%u",
1599 info->rx_ring_ref);
1600 if (err) {
1601 message = "writing rx ring-ref";
1602 goto abort_transaction;
1603 }
1604 err = xenbus_printf(xbt, dev->nodename,
1605 "event-channel", "%u", info->evtchn);
1606 if (err) {
1607 message = "writing event-channel";
1608 goto abort_transaction;
1609 }
1610
1611 err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
1612 1);
1613 if (err) {
1614 message = "writing request-rx-copy";
1615 goto abort_transaction;
1616 }
1617
1618 err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
1619 if (err) {
1620 message = "writing feature-rx-notify";
1621 goto abort_transaction;
1622 }
1623
1624 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
1625 if (err) {
1626 message = "writing feature-sg";
1627 goto abort_transaction;
1628 }
1629
1630 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
1631 if (err) {
1632 message = "writing feature-gso-tcpv4";
1633 goto abort_transaction;
1634 }
1635
1636 err = xenbus_transaction_end(xbt, 0);
1637 if (err) {
1638 if (err == -EAGAIN)
1639 goto again;
1640 xenbus_dev_fatal(dev, err, "completing transaction");
1641 goto destroy_ring;
1642 }
1643
1644 return 0;
1645
1646 abort_transaction:
1647 xenbus_transaction_end(xbt, 1);
1648 xenbus_dev_fatal(dev, err, "%s", message);
1649 destroy_ring:
1650 xennet_disconnect_backend(info);
1651 out:
1652 return err;
1653 }
1654
1655 static int xennet_connect(struct net_device *dev)
1656 {
1657 struct netfront_info *np = netdev_priv(dev);
1658 int i, requeue_idx, err;
1659 struct sk_buff *skb;
1660 grant_ref_t ref;
1661 struct xen_netif_rx_request *req;
1662 unsigned int feature_rx_copy;
1663
1664 err = xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1665 "feature-rx-copy", "%u", &feature_rx_copy);
1666 if (err != 1)
1667 feature_rx_copy = 0;
1668
1669 if (!feature_rx_copy) {
1670 dev_info(&dev->dev,
1671 "backend does not support copying receive path\n");
1672 return -ENODEV;
1673 }
1674
1675 err = talk_to_netback(np->xbdev, np);
1676 if (err)
1677 return err;
1678
1679 rtnl_lock();
1680 netdev_update_features(dev);
1681 rtnl_unlock();
1682
1683 spin_lock_bh(&np->rx_lock);
1684 spin_lock_irq(&np->tx_lock);
1685
1686 /* Step 1: Discard all pending TX packet fragments. */
1687 xennet_release_tx_bufs(np);
1688
1689 /* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */
1690 for (requeue_idx = 0, i = 0; i < NET_RX_RING_SIZE; i++) {
1691 skb_frag_t *frag;
1692 const struct page *page;
1693 if (!np->rx_skbs[i])
1694 continue;
1695
1696 skb = np->rx_skbs[requeue_idx] = xennet_get_rx_skb(np, i);
1697 ref = np->grant_rx_ref[requeue_idx] = xennet_get_rx_ref(np, i);
1698 req = RING_GET_REQUEST(&np->rx, requeue_idx);
1699
1700 frag = &skb_shinfo(skb)->frags[0];
1701 page = skb_frag_page(frag);
1702 gnttab_grant_foreign_access_ref(
1703 ref, np->xbdev->otherend_id,
1704 pfn_to_mfn(page_to_pfn(page)),
1705 0);
1706 req->gref = ref;
1707 req->id = requeue_idx;
1708
1709 requeue_idx++;
1710 }
1711
1712 np->rx.req_prod_pvt = requeue_idx;
1713
1714 /*
1715 * Step 3: All public and private state should now be sane. Get
1716 * ready to start sending and receiving packets and give the driver
1717 * domain a kick because we've probably just requeued some
1718 * packets.
1719 */
1720 netif_carrier_on(np->netdev);
1721 notify_remote_via_irq(np->netdev->irq);
1722 xennet_tx_buf_gc(dev);
1723 xennet_alloc_rx_buffers(dev);
1724
1725 spin_unlock_irq(&np->tx_lock);
1726 spin_unlock_bh(&np->rx_lock);
1727
1728 return 0;
1729 }
1730
1731 /**
1732 * Callback received when the backend's state changes.
1733 */
1734 static void netback_changed(struct xenbus_device *dev,
1735 enum xenbus_state backend_state)
1736 {
1737 struct netfront_info *np = dev_get_drvdata(&dev->dev);
1738 struct net_device *netdev = np->netdev;
1739
1740 dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
1741
1742 switch (backend_state) {
1743 case XenbusStateInitialising:
1744 case XenbusStateInitialised:
1745 case XenbusStateReconfiguring:
1746 case XenbusStateReconfigured:
1747 case XenbusStateUnknown:
1748 case XenbusStateClosed:
1749 break;
1750
1751 case XenbusStateInitWait:
1752 if (dev->state != XenbusStateInitialising)
1753 break;
1754 if (xennet_connect(netdev) != 0)
1755 break;
1756 xenbus_switch_state(dev, XenbusStateConnected);
1757 break;
1758
1759 case XenbusStateConnected:
1760 netdev_notify_peers(netdev);
1761 break;
1762
1763 case XenbusStateClosing:
1764 xenbus_frontend_closed(dev);
1765 break;
1766 }
1767 }
1768
1769 static const struct xennet_stat {
1770 char name[ETH_GSTRING_LEN];
1771 u16 offset;
1772 } xennet_stats[] = {
1773 {
1774 "rx_gso_checksum_fixup",
1775 offsetof(struct netfront_info, rx_gso_checksum_fixup)
1776 },
1777 };
1778
1779 static int xennet_get_sset_count(struct net_device *dev, int string_set)
1780 {
1781 switch (string_set) {
1782 case ETH_SS_STATS:
1783 return ARRAY_SIZE(xennet_stats);
1784 default:
1785 return -EINVAL;
1786 }
1787 }
1788
1789 static void xennet_get_ethtool_stats(struct net_device *dev,
1790 struct ethtool_stats *stats, u64 * data)
1791 {
1792 void *np = netdev_priv(dev);
1793 int i;
1794
1795 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
1796 data[i] = *(unsigned long *)(np + xennet_stats[i].offset);
1797 }
1798
1799 static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
1800 {
1801 int i;
1802
1803 switch (stringset) {
1804 case ETH_SS_STATS:
1805 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
1806 memcpy(data + i * ETH_GSTRING_LEN,
1807 xennet_stats[i].name, ETH_GSTRING_LEN);
1808 break;
1809 }
1810 }
1811
1812 static const struct ethtool_ops xennet_ethtool_ops =
1813 {
1814 .get_link = ethtool_op_get_link,
1815
1816 .get_sset_count = xennet_get_sset_count,
1817 .get_ethtool_stats = xennet_get_ethtool_stats,
1818 .get_strings = xennet_get_strings,
1819 };
1820
1821 #ifdef CONFIG_SYSFS
1822 static ssize_t show_rxbuf_min(struct device *dev,
1823 struct device_attribute *attr, char *buf)
1824 {
1825 struct net_device *netdev = to_net_dev(dev);
1826 struct netfront_info *info = netdev_priv(netdev);
1827
1828 return sprintf(buf, "%u\n", info->rx_min_target);
1829 }
1830
1831 static ssize_t store_rxbuf_min(struct device *dev,
1832 struct device_attribute *attr,
1833 const char *buf, size_t len)
1834 {
1835 struct net_device *netdev = to_net_dev(dev);
1836 struct netfront_info *np = netdev_priv(netdev);
1837 char *endp;
1838 unsigned long target;
1839
1840 if (!capable(CAP_NET_ADMIN))
1841 return -EPERM;
1842
1843 target = simple_strtoul(buf, &endp, 0);
1844 if (endp == buf)
1845 return -EBADMSG;
1846
1847 if (target < RX_MIN_TARGET)
1848 target = RX_MIN_TARGET;
1849 if (target > RX_MAX_TARGET)
1850 target = RX_MAX_TARGET;
1851
1852 spin_lock_bh(&np->rx_lock);
1853 if (target > np->rx_max_target)
1854 np->rx_max_target = target;
1855 np->rx_min_target = target;
1856 if (target > np->rx_target)
1857 np->rx_target = target;
1858
1859 xennet_alloc_rx_buffers(netdev);
1860
1861 spin_unlock_bh(&np->rx_lock);
1862 return len;
1863 }
1864
1865 static ssize_t show_rxbuf_max(struct device *dev,
1866 struct device_attribute *attr, char *buf)
1867 {
1868 struct net_device *netdev = to_net_dev(dev);
1869 struct netfront_info *info = netdev_priv(netdev);
1870
1871 return sprintf(buf, "%u\n", info->rx_max_target);
1872 }
1873
1874 static ssize_t store_rxbuf_max(struct device *dev,
1875 struct device_attribute *attr,
1876 const char *buf, size_t len)
1877 {
1878 struct net_device *netdev = to_net_dev(dev);
1879 struct netfront_info *np = netdev_priv(netdev);
1880 char *endp;
1881 unsigned long target;
1882
1883 if (!capable(CAP_NET_ADMIN))
1884 return -EPERM;
1885
1886 target = simple_strtoul(buf, &endp, 0);
1887 if (endp == buf)
1888 return -EBADMSG;
1889
1890 if (target < RX_MIN_TARGET)
1891 target = RX_MIN_TARGET;
1892 if (target > RX_MAX_TARGET)
1893 target = RX_MAX_TARGET;
1894
1895 spin_lock_bh(&np->rx_lock);
1896 if (target < np->rx_min_target)
1897 np->rx_min_target = target;
1898 np->rx_max_target = target;
1899 if (target < np->rx_target)
1900 np->rx_target = target;
1901
1902 xennet_alloc_rx_buffers(netdev);
1903
1904 spin_unlock_bh(&np->rx_lock);
1905 return len;
1906 }
1907
1908 static ssize_t show_rxbuf_cur(struct device *dev,
1909 struct device_attribute *attr, char *buf)
1910 {
1911 struct net_device *netdev = to_net_dev(dev);
1912 struct netfront_info *info = netdev_priv(netdev);
1913
1914 return sprintf(buf, "%u\n", info->rx_target);
1915 }
1916
1917 static struct device_attribute xennet_attrs[] = {
1918 __ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf_min, store_rxbuf_min),
1919 __ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf_max, store_rxbuf_max),
1920 __ATTR(rxbuf_cur, S_IRUGO, show_rxbuf_cur, NULL),
1921 };
1922
1923 static int xennet_sysfs_addif(struct net_device *netdev)
1924 {
1925 int i;
1926 int err;
1927
1928 for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++) {
1929 err = device_create_file(&netdev->dev,
1930 &xennet_attrs[i]);
1931 if (err)
1932 goto fail;
1933 }
1934 return 0;
1935
1936 fail:
1937 while (--i >= 0)
1938 device_remove_file(&netdev->dev, &xennet_attrs[i]);
1939 return err;
1940 }
1941
1942 static void xennet_sysfs_delif(struct net_device *netdev)
1943 {
1944 int i;
1945
1946 for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++)
1947 device_remove_file(&netdev->dev, &xennet_attrs[i]);
1948 }
1949
1950 #endif /* CONFIG_SYSFS */
1951
1952 static const struct xenbus_device_id netfront_ids[] = {
1953 { "vif" },
1954 { "" }
1955 };
1956
1957
1958 static int xennet_remove(struct xenbus_device *dev)
1959 {
1960 struct netfront_info *info = dev_get_drvdata(&dev->dev);
1961
1962 dev_dbg(&dev->dev, "%s\n", dev->nodename);
1963
1964 xennet_disconnect_backend(info);
1965
1966 xennet_sysfs_delif(info->netdev);
1967
1968 unregister_netdev(info->netdev);
1969
1970 del_timer_sync(&info->rx_refill_timer);
1971
1972 free_percpu(info->stats);
1973
1974 free_netdev(info->netdev);
1975
1976 return 0;
1977 }
1978
1979 static DEFINE_XENBUS_DRIVER(netfront, ,
1980 .probe = netfront_probe,
1981 .remove = xennet_remove,
1982 .resume = netfront_resume,
1983 .otherend_changed = netback_changed,
1984 );
1985
1986 static int __init netif_init(void)
1987 {
1988 if (!xen_domain())
1989 return -ENODEV;
1990
1991 if (xen_hvm_domain() && !xen_platform_pci_unplug)
1992 return -ENODEV;
1993
1994 printk(KERN_INFO "Initialising Xen virtual ethernet driver.\n");
1995
1996 return xenbus_register_frontend(&netfront_driver);
1997 }
1998 module_init(netif_init);
1999
2000
2001 static void __exit netif_exit(void)
2002 {
2003 xenbus_unregister_driver(&netfront_driver);
2004 }
2005 module_exit(netif_exit);
2006
2007 MODULE_DESCRIPTION("Xen virtual network device frontend");
2008 MODULE_LICENSE("GPL");
2009 MODULE_ALIAS("xen:vif");
2010 MODULE_ALIAS("xennet");
Linux with added FPC-III support