]> bbs.cooldavid.org Git - net-next-2.6.git/blame - drivers/net/xen-netfront.c
git://bbs.cooldavid.org / net-next-2.6.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
[NET]: Make NAPI polling independent of struct net_device objects.
[net-next-2.6.git] / drivers / net / xen-netfront.c
CommitLineData
0d160211
JF		 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
54static struct ethtool_ops xennet_ethtool_ops;
55
56struct 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
71struct netfront_info {
72 struct list_head list;
73 struct net_device *netdev;
74
bea3348e 75 struct napi_struct napi;
0d160211
JF		 76 struct net_device_stats stats;
77
78 struct xen_netif_tx_front_ring tx;
79 struct xen_netif_rx_front_ring rx;
80
81 spinlock_t tx_lock;
82 spinlock_t rx_lock;
83
84 unsigned int evtchn;
85
86 /* Receive-ring batched refills. */
87#define RX_MIN_TARGET 8
88#define RX_DFL_MIN_TARGET 64
89#define RX_MAX_TARGET min_t(int, NET_RX_RING_SIZE, 256)
90 unsigned rx_min_target, rx_max_target, rx_target;
91 struct sk_buff_head rx_batch;
92
93 struct timer_list rx_refill_timer;
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 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 struct sk_buff *rx_skbs[NET_RX_RING_SIZE];
113 grant_ref_t gref_rx_head;
114 grant_ref_t grant_rx_ref[NET_RX_RING_SIZE];
115
116 struct xenbus_device *xbdev;
117 int tx_ring_ref;
118 int rx_ring_ref;
119
120 unsigned long rx_pfn_array[NET_RX_RING_SIZE];
121 struct multicall_entry rx_mcl[NET_RX_RING_SIZE+1];
122 struct mmu_update rx_mmu[NET_RX_RING_SIZE];
123};
124
125struct netfront_rx_info {
126 struct xen_netif_rx_response rx;
127 struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
128};
129
130/*
131 * Access macros for acquiring freeing slots in tx_skbs[].
132 */
133
134static void add_id_to_freelist(unsigned *head, union skb_entry *list,
135 unsigned short id)
136{
137 list[id].link = *head;
138 *head = id;
139}
140
141static unsigned short get_id_from_freelist(unsigned *head,
142 union skb_entry *list)
143{
144 unsigned int id = *head;
145 *head = list[id].link;
146 return id;
147}
148
149static int xennet_rxidx(RING_IDX idx)
150{
151 return idx & (NET_RX_RING_SIZE - 1);
152}
153
154static struct sk_buff *xennet_get_rx_skb(struct netfront_info *np,
155 RING_IDX ri)
156{
157 int i = xennet_rxidx(ri);
158 struct sk_buff *skb = np->rx_skbs[i];
159 np->rx_skbs[i] = NULL;
160 return skb;
161}
162
163static grant_ref_t xennet_get_rx_ref(struct netfront_info *np,
164 RING_IDX ri)
165{
166 int i = xennet_rxidx(ri);
167 grant_ref_t ref = np->grant_rx_ref[i];
168 np->grant_rx_ref[i] = GRANT_INVALID_REF;
169 return ref;
170}
171
172#ifdef CONFIG_SYSFS
173static int xennet_sysfs_addif(struct net_device *netdev);
174static void xennet_sysfs_delif(struct net_device *netdev);
175#else /* !CONFIG_SYSFS */
176#define xennet_sysfs_addif(dev) (0)
177#define xennet_sysfs_delif(dev) do { } while (0)
178#endif
179
180static int xennet_can_sg(struct net_device *dev)
181{
182 return dev->features & NETIF_F_SG;
183}
184
185
186static void rx_refill_timeout(unsigned long data)
187{
188 struct net_device *dev = (struct net_device *)data;
bea3348e
SH		 189 struct netfront_info *np = netdev_priv(dev);
190 netif_rx_schedule(dev, &np->napi);
0d160211
JF		 191}
192
193static int netfront_tx_slot_available(struct netfront_info *np)
194{
195 return ((np->tx.req_prod_pvt - np->tx.rsp_cons) <
196 (TX_MAX_TARGET - MAX_SKB_FRAGS - 2));
197}
198
199static void xennet_maybe_wake_tx(struct net_device *dev)
200{
201 struct netfront_info *np = netdev_priv(dev);
202
203 if (unlikely(netif_queue_stopped(dev)) &&
204 netfront_tx_slot_available(np) &&
205 likely(netif_running(dev)))
206 netif_wake_queue(dev);
207}
208
209static void xennet_alloc_rx_buffers(struct net_device *dev)
210{
211 unsigned short id;
212 struct netfront_info *np = netdev_priv(dev);
213 struct sk_buff *skb;
214 struct page *page;
215 int i, batch_target, notify;
216 RING_IDX req_prod = np->rx.req_prod_pvt;
217 struct xen_memory_reservation reservation;
218 grant_ref_t ref;
219 unsigned long pfn;
220 void *vaddr;
221 int nr_flips;
222 struct xen_netif_rx_request *req;
223
224 if (unlikely(!netif_carrier_ok(dev)))
225 return;
226
227 /*
228 * Allocate skbuffs greedily, even though we batch updates to the
229 * receive ring. This creates a less bursty demand on the memory
230 * allocator, so should reduce the chance of failed allocation requests
231 * both for ourself and for other kernel subsystems.
232 */
233 batch_target = np->rx_target - (req_prod - np->rx.rsp_cons);
234 for (i = skb_queue_len(&np->rx_batch); i < batch_target; i++) {
235 skb = __netdev_alloc_skb(dev, RX_COPY_THRESHOLD,
236 GFP_ATOMIC | __GFP_NOWARN);
237 if (unlikely(!skb))
238 goto no_skb;
239
240 page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
241 if (!page) {
242 kfree_skb(skb);
243no_skb:
244 /* Any skbuffs queued for refill? Force them out. */
245 if (i != 0)
246 goto refill;
247 /* Could not allocate any skbuffs. Try again later. */
248 mod_timer(&np->rx_refill_timer,
249 jiffies + (HZ/10));
250 break;
251 }
252
253 skb_shinfo(skb)->frags[0].page = page;
254 skb_shinfo(skb)->nr_frags = 1;
255 __skb_queue_tail(&np->rx_batch, skb);
256 }
257
258 /* Is the batch large enough to be worthwhile? */
259 if (i < (np->rx_target/2)) {
260 if (req_prod > np->rx.sring->req_prod)
261 goto push;
262 return;
263 }
264
265 /* Adjust our fill target if we risked running out of buffers. */
266 if (((req_prod - np->rx.sring->rsp_prod) < (np->rx_target / 4)) &&
267 ((np->rx_target *= 2) > np->rx_max_target))
268 np->rx_target = np->rx_max_target;
269
270 refill:
271 for (nr_flips = i = 0; ; i++) {
272 skb = __skb_dequeue(&np->rx_batch);
273 if (skb == NULL)
274 break;
275
276 skb->dev = dev;
277
278 id = xennet_rxidx(req_prod + i);
279
280 BUG_ON(np->rx_skbs[id]);
281 np->rx_skbs[id] = skb;
282
283 ref = gnttab_claim_grant_reference(&np->gref_rx_head);
284 BUG_ON((signed short)ref < 0);
285 np->grant_rx_ref[id] = ref;
286
287 pfn = page_to_pfn(skb_shinfo(skb)->frags[0].page);
288 vaddr = page_address(skb_shinfo(skb)->frags[0].page);
289
290 req = RING_GET_REQUEST(&np->rx, req_prod + i);
291 gnttab_grant_foreign_access_ref(ref,
292 np->xbdev->otherend_id,
293 pfn_to_mfn(pfn),
294 0);
295
296 req->id = id;
297 req->gref = ref;
298 }
299
300 if (nr_flips != 0) {
301 reservation.extent_start = np->rx_pfn_array;
302 reservation.nr_extents = nr_flips;
303 reservation.extent_order = 0;
304 reservation.address_bits = 0;
305 reservation.domid = DOMID_SELF;
306
307 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
308 /* After all PTEs have been zapped, flush the TLB. */
309 np->rx_mcl[i-1].args[MULTI_UVMFLAGS_INDEX] =
310 UVMF_TLB_FLUSH|UVMF_ALL;
311
312 /* Give away a batch of pages. */
313 np->rx_mcl[i].op = __HYPERVISOR_memory_op;
314 np->rx_mcl[i].args[0] = XENMEM_decrease_reservation;
315 np->rx_mcl[i].args[1] = (unsigned long)&reservation;
316
317 /* Zap PTEs and give away pages in one big
318 * multicall. */
319 (void)HYPERVISOR_multicall(np->rx_mcl, i+1);
320
321 /* Check return status of HYPERVISOR_memory_op(). */
322 if (unlikely(np->rx_mcl[i].result != i))
323 panic("Unable to reduce memory reservation\n");
324 } else {
325 if (HYPERVISOR_memory_op(XENMEM_decrease_reservation,
326 &reservation) != i)
327 panic("Unable to reduce memory reservation\n");
328 }
329 } else {
330 wmb(); /* barrier so backend seens requests */
331 }
332
333 /* Above is a suitable barrier to ensure backend will see requests. */
334 np->rx.req_prod_pvt = req_prod + i;
335 push:
336 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->rx, notify);
337 if (notify)
338 notify_remote_via_irq(np->netdev->irq);
339}
340
341static int xennet_open(struct net_device *dev)
342{
343 struct netfront_info *np = netdev_priv(dev);
344
345 memset(&np->stats, 0, sizeof(np->stats));
346
bea3348e
SH		 347 napi_enable(&np->napi);
348
0d160211
JF		 349 spin_lock_bh(&np->rx_lock);
350 if (netif_carrier_ok(dev)) {
351 xennet_alloc_rx_buffers(dev);
352 np->rx.sring->rsp_event = np->rx.rsp_cons + 1;
353 if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx))
bea3348e 354 netif_rx_schedule(dev, &np->napi);
0d160211
JF		 355 }
356 spin_unlock_bh(&np->rx_lock);
357
358 xennet_maybe_wake_tx(dev);
359
360 return 0;
361}
362
363static void xennet_tx_buf_gc(struct net_device *dev)
364{
365 RING_IDX cons, prod;
366 unsigned short id;
367 struct netfront_info *np = netdev_priv(dev);
368 struct sk_buff *skb;
369
370 BUG_ON(!netif_carrier_ok(dev));
371
372 do {
373 prod = np->tx.sring->rsp_prod;
374 rmb(); /* Ensure we see responses up to 'rp'. */
375
376 for (cons = np->tx.rsp_cons; cons != prod; cons++) {
377 struct xen_netif_tx_response *txrsp;
378
379 txrsp = RING_GET_RESPONSE(&np->tx, cons);
380 if (txrsp->status == NETIF_RSP_NULL)
381 continue;
382
383 id = txrsp->id;
384 skb = np->tx_skbs[id].skb;
385 if (unlikely(gnttab_query_foreign_access(
386 np->grant_tx_ref[id]) != 0)) {
387 printk(KERN_ALERT "xennet_tx_buf_gc: warning "
388 "-- grant still in use by backend "
389 "domain.\n");
390 BUG();
391 }
392 gnttab_end_foreign_access_ref(
393 np->grant_tx_ref[id], GNTMAP_readonly);
394 gnttab_release_grant_reference(
395 &np->gref_tx_head, np->grant_tx_ref[id]);
396 np->grant_tx_ref[id] = GRANT_INVALID_REF;
397 add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, id);
398 dev_kfree_skb_irq(skb);
399 }
400
401 np->tx.rsp_cons = prod;
402
403 /*
404 * Set a new event, then check for race with update of tx_cons.
405 * Note that it is essential to schedule a callback, no matter
406 * how few buffers are pending. Even if there is space in the
407 * transmit ring, higher layers may be blocked because too much
408 * data is outstanding: in such cases notification from Xen is
409 * likely to be the only kick that we'll get.
410 */
411 np->tx.sring->rsp_event =
412 prod + ((np->tx.sring->req_prod - prod) >> 1) + 1;
413 mb(); /* update shared area */
414 } while ((cons == prod) && (prod != np->tx.sring->rsp_prod));
415
416 xennet_maybe_wake_tx(dev);
417}
418
419static void xennet_make_frags(struct sk_buff *skb, struct net_device *dev,
420 struct xen_netif_tx_request *tx)
421{
422 struct netfront_info *np = netdev_priv(dev);
423 char *data = skb->data;
424 unsigned long mfn;
425 RING_IDX prod = np->tx.req_prod_pvt;
426 int frags = skb_shinfo(skb)->nr_frags;
427 unsigned int offset = offset_in_page(data);
428 unsigned int len = skb_headlen(skb);
429 unsigned int id;
430 grant_ref_t ref;
431 int i;
432
433 /* While the header overlaps a page boundary (including being
434 larger than a page), split it it into page-sized chunks. */
435 while (len > PAGE_SIZE - offset) {
436 tx->size = PAGE_SIZE - offset;
437 tx->flags |= NETTXF_more_data;
438 len -= tx->size;
439 data += tx->size;
440 offset = 0;
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 = virt_to_mfn(data);
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 = offset;
455 tx->size = len;
456 tx->flags = 0;
457 }
458
459 /* Grant backend access to each skb fragment page. */
460 for (i = 0; i < frags; i++) {
461 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
462
463 tx->flags |= NETTXF_more_data;
464
465 id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
466 np->tx_skbs[id].skb = skb_get(skb);
467 tx = RING_GET_REQUEST(&np->tx, prod++);
468 tx->id = id;
469 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
470 BUG_ON((signed short)ref < 0);
471
472 mfn = pfn_to_mfn(page_to_pfn(frag->page));
473 gnttab_grant_foreign_access_ref(ref, np->xbdev->otherend_id,
474 mfn, GNTMAP_readonly);
475
476 tx->gref = np->grant_tx_ref[id] = ref;
477 tx->offset = frag->page_offset;
478 tx->size = frag->size;
479 tx->flags = 0;
480 }
481
482 np->tx.req_prod_pvt = prod;
483}
484
485static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
486{
487 unsigned short id;
488 struct netfront_info *np = netdev_priv(dev);
489 struct xen_netif_tx_request *tx;
490 struct xen_netif_extra_info *extra;
491 char *data = skb->data;
492 RING_IDX i;
493 grant_ref_t ref;
494 unsigned long mfn;
495 int notify;
496 int frags = skb_shinfo(skb)->nr_frags;
497 unsigned int offset = offset_in_page(data);
498 unsigned int len = skb_headlen(skb);
499
500 frags += (offset + len + PAGE_SIZE - 1) / PAGE_SIZE;
501 if (unlikely(frags > MAX_SKB_FRAGS + 1)) {
502 printk(KERN_ALERT "xennet: skb rides the rocket: %d frags\n",
503 frags);
504 dump_stack();
505 goto drop;
506 }
507
508 spin_lock_irq(&np->tx_lock);
509
510 if (unlikely(!netif_carrier_ok(dev) ||
511 (frags > 1 && !xennet_can_sg(dev)) ||
512 netif_needs_gso(dev, skb))) {
513 spin_unlock_irq(&np->tx_lock);
514 goto drop;
515 }
516
517 i = np->tx.req_prod_pvt;
518
519 id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
520 np->tx_skbs[id].skb = skb;
521
522 tx = RING_GET_REQUEST(&np->tx, i);
523
524 tx->id = id;
525 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
526 BUG_ON((signed short)ref < 0);
527 mfn = virt_to_mfn(data);
528 gnttab_grant_foreign_access_ref(
529 ref, np->xbdev->otherend_id, mfn, GNTMAP_readonly);
530 tx->gref = np->grant_tx_ref[id] = ref;
531 tx->offset = offset;
532 tx->size = len;
533 extra = NULL;
534
535 tx->flags = 0;
536 if (skb->ip_summed == CHECKSUM_PARTIAL)
537 /* local packet? */
538 tx->flags |= NETTXF_csum_blank | NETTXF_data_validated;
539 else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
540 /* remote but checksummed. */
541 tx->flags |= NETTXF_data_validated;
542
543 if (skb_shinfo(skb)->gso_size) {
544 struct xen_netif_extra_info *gso;
545
546 gso = (struct xen_netif_extra_info *)
547 RING_GET_REQUEST(&np->tx, ++i);
548
549 if (extra)
550 extra->flags |= XEN_NETIF_EXTRA_FLAG_MORE;
551 else
552 tx->flags |= NETTXF_extra_info;
553
554 gso->u.gso.size = skb_shinfo(skb)->gso_size;
555 gso->u.gso.type = XEN_NETIF_GSO_TYPE_TCPV4;
556 gso->u.gso.pad = 0;
557 gso->u.gso.features = 0;
558
559 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
560 gso->flags = 0;
561 extra = gso;
562 }
563
564 np->tx.req_prod_pvt = i + 1;
565
566 xennet_make_frags(skb, dev, tx);
567 tx->size = skb->len;
568
569 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->tx, notify);
570 if (notify)
571 notify_remote_via_irq(np->netdev->irq);
572
10a273a6
JF		 573 np->stats.tx_bytes += skb->len;
574 np->stats.tx_packets++;
575
576 /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
0d160211
JF		 577 xennet_tx_buf_gc(dev);
578
579 if (!netfront_tx_slot_available(np))
580 netif_stop_queue(dev);
581
582 spin_unlock_irq(&np->tx_lock);
583
0d160211
JF		 584 return 0;
585
586 drop:
587 np->stats.tx_dropped++;
588 dev_kfree_skb(skb);
589 return 0;
590}
591
592static int xennet_close(struct net_device *dev)
593{
594 struct netfront_info *np = netdev_priv(dev);
595 netif_stop_queue(np->netdev);
bea3348e 596 napi_disable(&np->napi);
0d160211
JF		 597 return 0;
598}
599
600static struct net_device_stats *xennet_get_stats(struct net_device *dev)
601{
602 struct netfront_info *np = netdev_priv(dev);
603 return &np->stats;
604}
605
606static void xennet_move_rx_slot(struct netfront_info *np, struct sk_buff *skb,
607 grant_ref_t ref)
608{
609 int new = xennet_rxidx(np->rx.req_prod_pvt);
610
611 BUG_ON(np->rx_skbs[new]);
612 np->rx_skbs[new] = skb;
613 np->grant_rx_ref[new] = ref;
614 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->id = new;
615 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->gref = ref;
616 np->rx.req_prod_pvt++;
617}
618
619static int xennet_get_extras(struct netfront_info *np,
620 struct xen_netif_extra_info *extras,
621 RING_IDX rp)
622
623{
624 struct xen_netif_extra_info *extra;
625 struct device *dev = &np->netdev->dev;
626 RING_IDX cons = np->rx.rsp_cons;
627 int err = 0;
628
629 do {
630 struct sk_buff *skb;
631 grant_ref_t ref;
632
633 if (unlikely(cons + 1 == rp)) {
634 if (net_ratelimit())
635 dev_warn(dev, "Missing extra info\n");
636 err = -EBADR;
637 break;
638 }
639
640 extra = (struct xen_netif_extra_info *)
641 RING_GET_RESPONSE(&np->rx, ++cons);
642
643 if (unlikely(!extra->type ||
644 extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
645 if (net_ratelimit())
646 dev_warn(dev, "Invalid extra type: %d\n",
647 extra->type);
648 err = -EINVAL;
649 } else {
650 memcpy(&extras[extra->type - 1], extra,
651 sizeof(*extra));
652 }
653
654 skb = xennet_get_rx_skb(np, cons);
655 ref = xennet_get_rx_ref(np, cons);
656 xennet_move_rx_slot(np, skb, ref);
657 } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
658
659 np->rx.rsp_cons = cons;
660 return err;
661}
662
663static int xennet_get_responses(struct netfront_info *np,
664 struct netfront_rx_info *rinfo, RING_IDX rp,
665 struct sk_buff_head *list)
666{
667 struct xen_netif_rx_response *rx = &rinfo->rx;
668 struct xen_netif_extra_info *extras = rinfo->extras;
669 struct device *dev = &np->netdev->dev;
670 RING_IDX cons = np->rx.rsp_cons;
671 struct sk_buff *skb = xennet_get_rx_skb(np, cons);
672 grant_ref_t ref = xennet_get_rx_ref(np, cons);
673 int max = MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD);
674 int frags = 1;
675 int err = 0;
676 unsigned long ret;
677
678 if (rx->flags & NETRXF_extra_info) {
679 err = xennet_get_extras(np, extras, rp);
680 cons = np->rx.rsp_cons;
681 }
682
683 for (;;) {
684 if (unlikely(rx->status < 0 ||
685 rx->offset + rx->status > PAGE_SIZE)) {
686 if (net_ratelimit())
687 dev_warn(dev, "rx->offset: %x, size: %u\n",
688 rx->offset, rx->status);
689 xennet_move_rx_slot(np, skb, ref);
690 err = -EINVAL;
691 goto next;
692 }
693
694 /*
695 * This definitely indicates a bug, either in this driver or in
696 * the backend driver. In future this should flag the bad
697 * situation to the system controller to reboot the backed.
698 */
699 if (ref == GRANT_INVALID_REF) {
700 if (net_ratelimit())
701 dev_warn(dev, "Bad rx response id %d.\n",
702 rx->id);
703 err = -EINVAL;
704 goto next;
705 }
706
707 ret = gnttab_end_foreign_access_ref(ref, 0);
708 BUG_ON(!ret);
709
710 gnttab_release_grant_reference(&np->gref_rx_head, ref);
711
712 __skb_queue_tail(list, skb);
713
714next:
715 if (!(rx->flags & NETRXF_more_data))
716 break;
717
718 if (cons + frags == rp) {
719 if (net_ratelimit())
720 dev_warn(dev, "Need more frags\n");
721 err = -ENOENT;
722 break;
723 }
724
725 rx = RING_GET_RESPONSE(&np->rx, cons + frags);
726 skb = xennet_get_rx_skb(np, cons + frags);
727 ref = xennet_get_rx_ref(np, cons + frags);
728 frags++;
729 }
730
731 if (unlikely(frags > max)) {
732 if (net_ratelimit())
733 dev_warn(dev, "Too many frags\n");
734 err = -E2BIG;
735 }
736
737 if (unlikely(err))
738 np->rx.rsp_cons = cons + frags;
739
740 return err;
741}
742
743static int xennet_set_skb_gso(struct sk_buff *skb,
744 struct xen_netif_extra_info *gso)
745{
746 if (!gso->u.gso.size) {
747 if (net_ratelimit())
748 printk(KERN_WARNING "GSO size must not be zero.\n");
749 return -EINVAL;
750 }
751
752 /* Currently only TCPv4 S.O. is supported. */
753 if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4) {
754 if (net_ratelimit())
755 printk(KERN_WARNING "Bad GSO type %d.\n", gso->u.gso.type);
756 return -EINVAL;
757 }
758
759 skb_shinfo(skb)->gso_size = gso->u.gso.size;
760 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
761
762 /* Header must be checked, and gso_segs computed. */
763 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
764 skb_shinfo(skb)->gso_segs = 0;
765
766 return 0;
767}
768
769static RING_IDX xennet_fill_frags(struct netfront_info *np,
770 struct sk_buff *skb,
771 struct sk_buff_head *list)
772{
773 struct skb_shared_info *shinfo = skb_shinfo(skb);
774 int nr_frags = shinfo->nr_frags;
775 RING_IDX cons = np->rx.rsp_cons;
776 skb_frag_t *frag = shinfo->frags + nr_frags;
777 struct sk_buff *nskb;
778
779 while ((nskb = __skb_dequeue(list))) {
780 struct xen_netif_rx_response *rx =
781 RING_GET_RESPONSE(&np->rx, ++cons);
782
783 frag->page = skb_shinfo(nskb)->frags[0].page;
784 frag->page_offset = rx->offset;
785 frag->size = rx->status;
786
787 skb->data_len += rx->status;
788
789 skb_shinfo(nskb)->nr_frags = 0;
790 kfree_skb(nskb);
791
792 frag++;
793 nr_frags++;
794 }
795
796 shinfo->nr_frags = nr_frags;
797 return cons;
798}
799
800static int skb_checksum_setup(struct sk_buff *skb)
801{
802 struct iphdr *iph;
803 unsigned char *th;
804 int err = -EPROTO;
805
806 if (skb->protocol != htons(ETH_P_IP))
807 goto out;
808
809 iph = (void *)skb->data;
810 th = skb->data + 4 * iph->ihl;
811 if (th >= skb_tail_pointer(skb))
812 goto out;
813
814 skb->csum_start = th - skb->head;
815 switch (iph->protocol) {
816 case IPPROTO_TCP:
817 skb->csum_offset = offsetof(struct tcphdr, check);
818 break;
819 case IPPROTO_UDP:
820 skb->csum_offset = offsetof(struct udphdr, check);
821 break;
822 default:
823 if (net_ratelimit())
824 printk(KERN_ERR "Attempting to checksum a non-"
825 "TCP/UDP packet, dropping a protocol"
826 " %d packet", iph->protocol);
827 goto out;
828 }
829
830 if ((th + skb->csum_offset + 2) > skb_tail_pointer(skb))
831 goto out;
832
833 err = 0;
834
835out:
836 return err;
837}
838
839static int handle_incoming_queue(struct net_device *dev,
840 struct sk_buff_head *rxq)
841{
842 struct netfront_info *np = netdev_priv(dev);
843 int packets_dropped = 0;
844 struct sk_buff *skb;
845
846 while ((skb = __skb_dequeue(rxq)) != NULL) {
847 struct page *page = NETFRONT_SKB_CB(skb)->page;
848 void *vaddr = page_address(page);
849 unsigned offset = NETFRONT_SKB_CB(skb)->offset;
850
851 memcpy(skb->data, vaddr + offset,
852 skb_headlen(skb));
853
854 if (page != skb_shinfo(skb)->frags[0].page)
855 __free_page(page);
856
857 /* Ethernet work: Delayed to here as it peeks the header. */
858 skb->protocol = eth_type_trans(skb, dev);
859
860 if (skb->ip_summed == CHECKSUM_PARTIAL) {
861 if (skb_checksum_setup(skb)) {
862 kfree_skb(skb);
863 packets_dropped++;
864 np->stats.rx_errors++;
865 continue;
866 }
867 }
868
869 np->stats.rx_packets++;
870 np->stats.rx_bytes += skb->len;
871
872 /* Pass it up. */
873 netif_receive_skb(skb);
874 dev->last_rx = jiffies;
875 }
876
877 return packets_dropped;
878}
879
bea3348e 880static int xennet_poll(struct napi_struct *napi, int budget)
0d160211 881{
bea3348e
SH		 882 struct netfront_info *np = container_of(napi, struct netfront_info, napi);
883 struct net_device *dev = np->netdev;
0d160211
JF		 884 struct sk_buff *skb;
885 struct netfront_rx_info rinfo;
886 struct xen_netif_rx_response *rx = &rinfo.rx;
887 struct xen_netif_extra_info *extras = rinfo.extras;
888 RING_IDX i, rp;
bea3348e 889 int work_done;
0d160211
JF		 890 struct sk_buff_head rxq;
891 struct sk_buff_head errq;
892 struct sk_buff_head tmpq;
893 unsigned long flags;
894 unsigned int len;
895 int err;
896
897 spin_lock(&np->rx_lock);
898
899 if (unlikely(!netif_carrier_ok(dev))) {
900 spin_unlock(&np->rx_lock);
901 return 0;
902 }
903
904 skb_queue_head_init(&rxq);
905 skb_queue_head_init(&errq);
906 skb_queue_head_init(&tmpq);
907
0d160211
JF		 908 rp = np->rx.sring->rsp_prod;
909 rmb(); /* Ensure we see queued responses up to 'rp'. */
910
911 i = np->rx.rsp_cons;
912 work_done = 0;
913 while ((i != rp) && (work_done < budget)) {
914 memcpy(rx, RING_GET_RESPONSE(&np->rx, i), sizeof(*rx));
915 memset(extras, 0, sizeof(rinfo.extras));
916
917 err = xennet_get_responses(np, &rinfo, rp, &tmpq);
918
919 if (unlikely(err)) {
920err:
921 while ((skb = __skb_dequeue(&tmpq)))
922 __skb_queue_tail(&errq, skb);
923 np->stats.rx_errors++;
924 i = np->rx.rsp_cons;
925 continue;
926 }
927
928 skb = __skb_dequeue(&tmpq);
929
930 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
931 struct xen_netif_extra_info *gso;
932 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
933
934 if (unlikely(xennet_set_skb_gso(skb, gso))) {
935 __skb_queue_head(&tmpq, skb);
936 np->rx.rsp_cons += skb_queue_len(&tmpq);
937 goto err;
938 }
939 }
940
941 NETFRONT_SKB_CB(skb)->page = skb_shinfo(skb)->frags[0].page;
942 NETFRONT_SKB_CB(skb)->offset = rx->offset;
943
944 len = rx->status;
945 if (len > RX_COPY_THRESHOLD)
946 len = RX_COPY_THRESHOLD;
947 skb_put(skb, len);
948
949 if (rx->status > len) {
950 skb_shinfo(skb)->frags[0].page_offset =
951 rx->offset + len;
952 skb_shinfo(skb)->frags[0].size = rx->status - len;
953 skb->data_len = rx->status - len;
954 } else {
955 skb_shinfo(skb)->frags[0].page = NULL;
956 skb_shinfo(skb)->nr_frags = 0;
957 }
958
959 i = xennet_fill_frags(np, skb, &tmpq);
960
961 /*
962 * Truesize approximates the size of true data plus
963 * any supervisor overheads. Adding hypervisor
964 * overheads has been shown to significantly reduce
965 * achievable bandwidth with the default receive
966 * buffer size. It is therefore not wise to account
967 * for it here.
968 *
969 * After alloc_skb(RX_COPY_THRESHOLD), truesize is set
970 * to RX_COPY_THRESHOLD + the supervisor
971 * overheads. Here, we add the size of the data pulled
972 * in xennet_fill_frags().
973 *
974 * We also adjust for any unused space in the main
975 * data area by subtracting (RX_COPY_THRESHOLD -
976 * len). This is especially important with drivers
977 * which split incoming packets into header and data,
978 * using only 66 bytes of the main data area (see the
979 * e1000 driver for example.) On such systems,
980 * without this last adjustement, our achievable
981 * receive throughout using the standard receive
982 * buffer size was cut by 25%(!!!).
983 */
984 skb->truesize += skb->data_len - (RX_COPY_THRESHOLD - len);
985 skb->len += skb->data_len;
986
987 if (rx->flags & NETRXF_csum_blank)
988 skb->ip_summed = CHECKSUM_PARTIAL;
989 else if (rx->flags & NETRXF_data_validated)
990 skb->ip_summed = CHECKSUM_UNNECESSARY;
991
992 __skb_queue_tail(&rxq, skb);
993
994 np->rx.rsp_cons = ++i;
995 work_done++;
996 }
997
998 while ((skb = __skb_dequeue(&errq)))
999 kfree_skb(skb);
1000
1001 work_done -= handle_incoming_queue(dev, &rxq);
1002
1003 /* If we get a callback with very few responses, reduce fill target. */
1004 /* NB. Note exponential increase, linear decrease. */
1005 if (((np->rx.req_prod_pvt - np->rx.sring->rsp_prod) >
1006 ((3*np->rx_target) / 4)) &&
1007 (--np->rx_target < np->rx_min_target))
1008 np->rx_target = np->rx_min_target;
1009
1010 xennet_alloc_rx_buffers(dev);
1011
0d160211 1012 if (work_done < budget) {
bea3348e
SH		 1013 int more_to_do = 0;
1014
0d160211
JF		 1015 local_irq_save(flags);
1016
1017 RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, more_to_do);
1018 if (!more_to_do)
bea3348e 1019 __netif_rx_complete(dev, napi);
0d160211
JF		 1020
1021 local_irq_restore(flags);
1022 }
1023
1024 spin_unlock(&np->rx_lock);
1025
bea3348e 1026 return work_done;
0d160211
JF		 1027}
1028
1029static int xennet_change_mtu(struct net_device *dev, int mtu)
1030{
1031 int max = xennet_can_sg(dev) ? 65535 - ETH_HLEN : ETH_DATA_LEN;
1032
1033 if (mtu > max)
1034 return -EINVAL;
1035 dev->mtu = mtu;
1036 return 0;
1037}
1038
1039static void xennet_release_tx_bufs(struct netfront_info *np)
1040{
1041 struct sk_buff *skb;
1042 int i;
1043
1044 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1045 /* Skip over entries which are actually freelist references */
1046 if ((unsigned long)np->tx_skbs[i].skb < PAGE_OFFSET)
1047 continue;
1048
1049 skb = np->tx_skbs[i].skb;
1050 gnttab_end_foreign_access_ref(np->grant_tx_ref[i],
1051 GNTMAP_readonly);
1052 gnttab_release_grant_reference(&np->gref_tx_head,
1053 np->grant_tx_ref[i]);
1054 np->grant_tx_ref[i] = GRANT_INVALID_REF;
1055 add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, i);
1056 dev_kfree_skb_irq(skb);
1057 }
1058}
1059
1060static void xennet_release_rx_bufs(struct netfront_info *np)
1061{
1062 struct mmu_update *mmu = np->rx_mmu;
1063 struct multicall_entry *mcl = np->rx_mcl;
1064 struct sk_buff_head free_list;
1065 struct sk_buff *skb;
1066 unsigned long mfn;
1067 int xfer = 0, noxfer = 0, unused = 0;
1068 int id, ref;
1069
1070 dev_warn(&np->netdev->dev, "%s: fix me for copying receiver.\n",
1071 __func__);
1072 return;
1073
1074 skb_queue_head_init(&free_list);
1075
1076 spin_lock_bh(&np->rx_lock);
1077
1078 for (id = 0; id < NET_RX_RING_SIZE; id++) {
1079 ref = np->grant_rx_ref[id];
1080 if (ref == GRANT_INVALID_REF) {
1081 unused++;
1082 continue;
1083 }
1084
1085 skb = np->rx_skbs[id];
1086 mfn = gnttab_end_foreign_transfer_ref(ref);
1087 gnttab_release_grant_reference(&np->gref_rx_head, ref);
1088 np->grant_rx_ref[id] = GRANT_INVALID_REF;
1089
1090 if (0 == mfn) {
1091 skb_shinfo(skb)->nr_frags = 0;
1092 dev_kfree_skb(skb);
1093 noxfer++;
1094 continue;
1095 }
1096
1097 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1098 /* Remap the page. */
1099 struct page *page = skb_shinfo(skb)->frags[0].page;
1100 unsigned long pfn = page_to_pfn(page);
1101 void *vaddr = page_address(page);
1102
1103 MULTI_update_va_mapping(mcl, (unsigned long)vaddr,
1104 mfn_pte(mfn, PAGE_KERNEL),
1105 0);
1106 mcl++;
1107 mmu->ptr = ((u64)mfn << PAGE_SHIFT)
1108 | MMU_MACHPHYS_UPDATE;
1109 mmu->val = pfn;
1110 mmu++;
1111
1112 set_phys_to_machine(pfn, mfn);
1113 }
1114 __skb_queue_tail(&free_list, skb);
1115 xfer++;
1116 }
1117
1118 dev_info(&np->netdev->dev, "%s: %d xfer, %d noxfer, %d unused\n",
1119 __func__, xfer, noxfer, unused);
1120
1121 if (xfer) {
1122 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1123 /* Do all the remapping work and M2P updates. */
1124 MULTI_mmu_update(mcl, np->rx_mmu, mmu - np->rx_mmu,
1125 0, DOMID_SELF);
1126 mcl++;
1127 HYPERVISOR_multicall(np->rx_mcl, mcl - np->rx_mcl);
1128 }
1129 }
1130
1131 while ((skb = __skb_dequeue(&free_list)) != NULL)
1132 dev_kfree_skb(skb);
1133
1134 spin_unlock_bh(&np->rx_lock);
1135}
1136
1137static void xennet_uninit(struct net_device *dev)
1138{
1139 struct netfront_info *np = netdev_priv(dev);
1140 xennet_release_tx_bufs(np);
1141 xennet_release_rx_bufs(np);
1142 gnttab_free_grant_references(np->gref_tx_head);
1143 gnttab_free_grant_references(np->gref_rx_head);
1144}
1145
1146static struct net_device * __devinit xennet_create_dev(struct xenbus_device *dev)
1147{
1148 int i, err;
1149 struct net_device *netdev;
1150 struct netfront_info *np;
1151
1152 netdev = alloc_etherdev(sizeof(struct netfront_info));
1153 if (!netdev) {
1154 printk(KERN_WARNING "%s> alloc_etherdev failed.\n",
1155 __func__);
1156 return ERR_PTR(-ENOMEM);
1157 }
1158
1159 np = netdev_priv(netdev);
1160 np->xbdev = dev;
1161
1162 spin_lock_init(&np->tx_lock);
1163 spin_lock_init(&np->rx_lock);
1164
1165 skb_queue_head_init(&np->rx_batch);
1166 np->rx_target = RX_DFL_MIN_TARGET;
1167 np->rx_min_target = RX_DFL_MIN_TARGET;
1168 np->rx_max_target = RX_MAX_TARGET;
1169
1170 init_timer(&np->rx_refill_timer);
1171 np->rx_refill_timer.data = (unsigned long)netdev;
1172 np->rx_refill_timer.function = rx_refill_timeout;
1173
1174 /* Initialise tx_skbs as a free chain containing every entry. */
1175 np->tx_skb_freelist = 0;
1176 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1177 np->tx_skbs[i].link = i+1;
1178 np->grant_tx_ref[i] = GRANT_INVALID_REF;
1179 }
1180
1181 /* Clear out rx_skbs */
1182 for (i = 0; i < NET_RX_RING_SIZE; i++) {
1183 np->rx_skbs[i] = NULL;
1184 np->grant_rx_ref[i] = GRANT_INVALID_REF;
1185 }
1186
1187 /* A grant for every tx ring slot */
1188 if (gnttab_alloc_grant_references(TX_MAX_TARGET,
1189 &np->gref_tx_head) < 0) {
1190 printk(KERN_ALERT "#### netfront can't alloc tx grant refs\n");
1191 err = -ENOMEM;
1192 goto exit;
1193 }
1194 /* A grant for every rx ring slot */
1195 if (gnttab_alloc_grant_references(RX_MAX_TARGET,
1196 &np->gref_rx_head) < 0) {
1197 printk(KERN_ALERT "#### netfront can't alloc rx grant refs\n");
1198 err = -ENOMEM;
1199 goto exit_free_tx;
1200 }
1201
1202 netdev->open = xennet_open;
1203 netdev->hard_start_xmit = xennet_start_xmit;
1204 netdev->stop = xennet_close;
1205 netdev->get_stats = xennet_get_stats;
bea3348e 1206 netif_napi_add(netdev, &np->napi, xennet_poll, 64);
0d160211
JF		 1207 netdev->uninit = xennet_uninit;
1208 netdev->change_mtu = xennet_change_mtu;
0d160211
JF		 1209 netdev->features = NETIF_F_IP_CSUM;
1210
1211 SET_ETHTOOL_OPS(netdev, &xennet_ethtool_ops);
1212 SET_MODULE_OWNER(netdev);
1213 SET_NETDEV_DEV(netdev, &dev->dev);
1214
1215 np->netdev = netdev;
1216
1217 netif_carrier_off(netdev);
1218
1219 return netdev;
1220
1221 exit_free_tx:
1222 gnttab_free_grant_references(np->gref_tx_head);
1223 exit:
1224 free_netdev(netdev);
1225 return ERR_PTR(err);
1226}
1227
1228/**
1229 * Entry point to this code when a new device is created. Allocate the basic
1230 * structures and the ring buffers for communication with the backend, and
1231 * inform the backend of the appropriate details for those.
1232 */
1233static int __devinit netfront_probe(struct xenbus_device *dev,
1234 const struct xenbus_device_id *id)
1235{
1236 int err;
1237 struct net_device *netdev;
1238 struct netfront_info *info;
1239
1240 netdev = xennet_create_dev(dev);
1241 if (IS_ERR(netdev)) {
1242 err = PTR_ERR(netdev);
1243 xenbus_dev_fatal(dev, err, "creating netdev");
1244 return err;
1245 }
1246
1247 info = netdev_priv(netdev);
1248 dev->dev.driver_data = info;
1249
1250 err = register_netdev(info->netdev);
1251 if (err) {
1252 printk(KERN_WARNING "%s: register_netdev err=%d\n",
1253 __func__, err);
1254 goto fail;
1255 }
1256
1257 err = xennet_sysfs_addif(info->netdev);
1258 if (err) {
1259 unregister_netdev(info->netdev);
1260 printk(KERN_WARNING "%s: add sysfs failed err=%d\n",
1261 __func__, err);
1262 goto fail;
1263 }
1264
1265 return 0;
1266
1267 fail:
1268 free_netdev(netdev);
1269 dev->dev.driver_data = NULL;
1270 return err;
1271}
1272
1273static void xennet_end_access(int ref, void *page)
1274{
1275 /* This frees the page as a side-effect */
1276 if (ref != GRANT_INVALID_REF)
1277 gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1278}
1279
1280static void xennet_disconnect_backend(struct netfront_info *info)
1281{
1282 /* Stop old i/f to prevent errors whilst we rebuild the state. */
1283 spin_lock_bh(&info->rx_lock);
1284 spin_lock_irq(&info->tx_lock);
1285 netif_carrier_off(info->netdev);
1286 spin_unlock_irq(&info->tx_lock);
1287 spin_unlock_bh(&info->rx_lock);
1288
1289 if (info->netdev->irq)
1290 unbind_from_irqhandler(info->netdev->irq, info->netdev);
1291 info->evtchn = info->netdev->irq = 0;
1292
1293 /* End access and free the pages */
1294 xennet_end_access(info->tx_ring_ref, info->tx.sring);
1295 xennet_end_access(info->rx_ring_ref, info->rx.sring);
1296
1297 info->tx_ring_ref = GRANT_INVALID_REF;
1298 info->rx_ring_ref = GRANT_INVALID_REF;
1299 info->tx.sring = NULL;
1300 info->rx.sring = NULL;
1301}
1302
1303/**
1304 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1305 * driver restart. We tear down our netif structure and recreate it, but
1306 * leave the device-layer structures intact so that this is transparent to the
1307 * rest of the kernel.
1308 */
1309static int netfront_resume(struct xenbus_device *dev)
1310{
1311 struct netfront_info *info = dev->dev.driver_data;
1312
1313 dev_dbg(&dev->dev, "%s\n", dev->nodename);
1314
1315 xennet_disconnect_backend(info);
1316 return 0;
1317}
1318
1319static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1320{
1321 char *s, *e, *macstr;
1322 int i;
1323
1324 macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1325 if (IS_ERR(macstr))
1326 return PTR_ERR(macstr);
1327
1328 for (i = 0; i < ETH_ALEN; i++) {
1329 mac[i] = simple_strtoul(s, &e, 16);
1330 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1331 kfree(macstr);
1332 return -ENOENT;
1333 }
1334 s = e+1;
1335 }
1336
1337 kfree(macstr);
1338 return 0;
1339}
1340
1341static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1342{
1343 struct net_device *dev = dev_id;
1344 struct netfront_info *np = netdev_priv(dev);
1345 unsigned long flags;
1346
1347 spin_lock_irqsave(&np->tx_lock, flags);
1348
1349 if (likely(netif_carrier_ok(dev))) {
1350 xennet_tx_buf_gc(dev);
1351 /* Under tx_lock: protects access to rx shared-ring indexes. */
1352 if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx))
bea3348e 1353 netif_rx_schedule(dev, &np->napi);
0d160211
JF		 1354 }
1355
1356 spin_unlock_irqrestore(&np->tx_lock, flags);
1357
1358 return IRQ_HANDLED;
1359}
1360
1361static int setup_netfront(struct xenbus_device *dev, struct netfront_info *info)
1362{
1363 struct xen_netif_tx_sring *txs;
1364 struct xen_netif_rx_sring *rxs;
1365 int err;
1366 struct net_device *netdev = info->netdev;
1367
1368 info->tx_ring_ref = GRANT_INVALID_REF;
1369 info->rx_ring_ref = GRANT_INVALID_REF;
1370 info->rx.sring = NULL;
1371 info->tx.sring = NULL;
1372 netdev->irq = 0;
1373
1374 err = xen_net_read_mac(dev, netdev->dev_addr);
1375 if (err) {
1376 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
1377 goto fail;
1378 }
1379
1380 txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_KERNEL);
1381 if (!txs) {
1382 err = -ENOMEM;
1383 xenbus_dev_fatal(dev, err, "allocating tx ring page");
1384 goto fail;
1385 }
1386 SHARED_RING_INIT(txs);
1387 FRONT_RING_INIT(&info->tx, txs, PAGE_SIZE);
1388
1389 err = xenbus_grant_ring(dev, virt_to_mfn(txs));
1390 if (err < 0) {
1391 free_page((unsigned long)txs);
1392 goto fail;
1393 }
1394
1395 info->tx_ring_ref = err;
1396 rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_KERNEL);
1397 if (!rxs) {
1398 err = -ENOMEM;
1399 xenbus_dev_fatal(dev, err, "allocating rx ring page");
1400 goto fail;
1401 }
1402 SHARED_RING_INIT(rxs);
1403 FRONT_RING_INIT(&info->rx, rxs, PAGE_SIZE);
1404
1405 err = xenbus_grant_ring(dev, virt_to_mfn(rxs));
1406 if (err < 0) {
1407 free_page((unsigned long)rxs);
1408 goto fail;
1409 }
1410 info->rx_ring_ref = err;
1411
1412 err = xenbus_alloc_evtchn(dev, &info->evtchn);
1413 if (err)
1414 goto fail;
1415
1416 err = bind_evtchn_to_irqhandler(info->evtchn, xennet_interrupt,
1417 IRQF_SAMPLE_RANDOM, netdev->name,
1418 netdev);
1419 if (err < 0)
1420 goto fail;
1421 netdev->irq = err;
1422 return 0;
1423
1424 fail:
1425 return err;
1426}
1427
1428/* Common code used when first setting up, and when resuming. */
1429static int talk_to_backend(struct xenbus_device *dev,
1430 struct netfront_info *info)
1431{
1432 const char *message;
1433 struct xenbus_transaction xbt;
1434 int err;
1435
1436 /* Create shared ring, alloc event channel. */
1437 err = setup_netfront(dev, info);
1438 if (err)
1439 goto out;
1440
1441again:
1442 err = xenbus_transaction_start(&xbt);
1443 if (err) {
1444 xenbus_dev_fatal(dev, err, "starting transaction");
1445 goto destroy_ring;
1446 }
1447
1448 err = xenbus_printf(xbt, dev->nodename, "tx-ring-ref", "%u",
1449 info->tx_ring_ref);
1450 if (err) {
1451 message = "writing tx ring-ref";
1452 goto abort_transaction;
1453 }
1454 err = xenbus_printf(xbt, dev->nodename, "rx-ring-ref", "%u",
1455 info->rx_ring_ref);
1456 if (err) {
1457 message = "writing rx ring-ref";
1458 goto abort_transaction;
1459 }
1460 err = xenbus_printf(xbt, dev->nodename,
1461 "event-channel", "%u", info->evtchn);
1462 if (err) {
1463 message = "writing event-channel";
1464 goto abort_transaction;
1465 }
1466
1467 err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
1468 1);
1469 if (err) {
1470 message = "writing request-rx-copy";
1471 goto abort_transaction;
1472 }
1473
1474 err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
1475 if (err) {
1476 message = "writing feature-rx-notify";
1477 goto abort_transaction;
1478 }
1479
1480 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
1481 if (err) {
1482 message = "writing feature-sg";
1483 goto abort_transaction;
1484 }
1485
1486 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
1487 if (err) {
1488 message = "writing feature-gso-tcpv4";
1489 goto abort_transaction;
1490 }
1491
1492 err = xenbus_transaction_end(xbt, 0);
1493 if (err) {
1494 if (err == -EAGAIN)
1495 goto again;
1496 xenbus_dev_fatal(dev, err, "completing transaction");
1497 goto destroy_ring;
1498 }
1499
1500 return 0;
1501
1502 abort_transaction:
1503 xenbus_transaction_end(xbt, 1);
1504 xenbus_dev_fatal(dev, err, "%s", message);
1505 destroy_ring:
1506 xennet_disconnect_backend(info);
1507 out:
1508 return err;
1509}
1510
1511static int xennet_set_sg(struct net_device *dev, u32 data)
1512{
1513 if (data) {
1514 struct netfront_info *np = netdev_priv(dev);
1515 int val;
1516
1517 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, "feature-sg",
1518 "%d", &val) < 0)
1519 val = 0;
1520 if (!val)
1521 return -ENOSYS;
1522 } else if (dev->mtu > ETH_DATA_LEN)
1523 dev->mtu = ETH_DATA_LEN;
1524
1525 return ethtool_op_set_sg(dev, data);
1526}
1527
1528static int xennet_set_tso(struct net_device *dev, u32 data)
1529{
1530 if (data) {
1531 struct netfront_info *np = netdev_priv(dev);
1532 int val;
1533
1534 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1535 "feature-gso-tcpv4", "%d", &val) < 0)
1536 val = 0;
1537 if (!val)
1538 return -ENOSYS;
1539 }
1540
1541 return ethtool_op_set_tso(dev, data);
1542}
1543
1544static void xennet_set_features(struct net_device *dev)
1545{
1546 /* Turn off all GSO bits except ROBUST. */
1547 dev->features &= (1 << NETIF_F_GSO_SHIFT) - 1;
1548 dev->features |= NETIF_F_GSO_ROBUST;
1549 xennet_set_sg(dev, 0);
1550
1551 /* We need checksum offload to enable scatter/gather and TSO. */
1552 if (!(dev->features & NETIF_F_IP_CSUM))
1553 return;
1554
1555 if (!xennet_set_sg(dev, 1))
1556 xennet_set_tso(dev, 1);
1557}
1558
1559static int xennet_connect(struct net_device *dev)
1560{
1561 struct netfront_info *np = netdev_priv(dev);
1562 int i, requeue_idx, err;
1563 struct sk_buff *skb;
1564 grant_ref_t ref;
1565 struct xen_netif_rx_request *req;
1566 unsigned int feature_rx_copy;
1567
1568 err = xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1569 "feature-rx-copy", "%u", &feature_rx_copy);
1570 if (err != 1)
1571 feature_rx_copy = 0;
1572
1573 if (!feature_rx_copy) {
1574 dev_info(&dev->dev,
1575 "backend does not support copying recieve path");
1576 return -ENODEV;
1577 }
1578
1579 err = talk_to_backend(np->xbdev, np);
1580 if (err)
1581 return err;
1582
1583 xennet_set_features(dev);
1584
1585 spin_lock_bh(&np->rx_lock);
1586 spin_lock_irq(&np->tx_lock);
1587
1588 /* Step 1: Discard all pending TX packet fragments. */
1589 xennet_release_tx_bufs(np);
1590
1591 /* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */
1592 for (requeue_idx = 0, i = 0; i < NET_RX_RING_SIZE; i++) {
1593 if (!np->rx_skbs[i])
1594 continue;
1595
1596 skb = np->rx_skbs[requeue_idx] = xennet_get_rx_skb(np, i);
1597 ref = np->grant_rx_ref[requeue_idx] = xennet_get_rx_ref(np, i);
1598 req = RING_GET_REQUEST(&np->rx, requeue_idx);
1599
1600 gnttab_grant_foreign_access_ref(
1601 ref, np->xbdev->otherend_id,
1602 pfn_to_mfn(page_to_pfn(skb_shinfo(skb)->
1603 frags->page)),
1604 0);
1605 req->gref = ref;
1606 req->id = requeue_idx;
1607
1608 requeue_idx++;
1609 }
1610
1611 np->rx.req_prod_pvt = requeue_idx;
1612
1613 /*
1614 * Step 3: All public and private state should now be sane. Get
1615 * ready to start sending and receiving packets and give the driver
1616 * domain a kick because we've probably just requeued some
1617 * packets.
1618 */
1619 netif_carrier_on(np->netdev);
1620 notify_remote_via_irq(np->netdev->irq);
1621 xennet_tx_buf_gc(dev);
1622 xennet_alloc_rx_buffers(dev);
1623
1624 spin_unlock_irq(&np->tx_lock);
1625 spin_unlock_bh(&np->rx_lock);
1626
1627 return 0;
1628}
1629
1630/**
1631 * Callback received when the backend's state changes.
1632 */
1633static void backend_changed(struct xenbus_device *dev,
1634 enum xenbus_state backend_state)
1635{
1636 struct netfront_info *np = dev->dev.driver_data;
1637 struct net_device *netdev = np->netdev;
1638
1639 dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
1640
1641 switch (backend_state) {
1642 case XenbusStateInitialising:
1643 case XenbusStateInitialised:
1644 case XenbusStateConnected:
1645 case XenbusStateUnknown:
1646 case XenbusStateClosed:
1647 break;
1648
1649 case XenbusStateInitWait:
1650 if (dev->state != XenbusStateInitialising)
1651 break;
1652 if (xennet_connect(netdev) != 0)
1653 break;
1654 xenbus_switch_state(dev, XenbusStateConnected);
1655 break;
1656
1657 case XenbusStateClosing:
1658 xenbus_frontend_closed(dev);
1659 break;
1660 }
1661}
1662
1663static struct ethtool_ops xennet_ethtool_ops =
1664{
1665 .get_tx_csum = ethtool_op_get_tx_csum,
1666 .set_tx_csum = ethtool_op_set_tx_csum,
1667 .get_sg = ethtool_op_get_sg,
1668 .set_sg = xennet_set_sg,
1669 .get_tso = ethtool_op_get_tso,
1670 .set_tso = xennet_set_tso,
1671 .get_link = ethtool_op_get_link,
1672};
1673
1674#ifdef CONFIG_SYSFS
1675static ssize_t show_rxbuf_min(struct device *dev,
1676 struct device_attribute *attr, char *buf)
1677{
1678 struct net_device *netdev = to_net_dev(dev);
1679 struct netfront_info *info = netdev_priv(netdev);
1680
1681 return sprintf(buf, "%u\n", info->rx_min_target);
1682}
1683
1684static ssize_t store_rxbuf_min(struct device *dev,
1685 struct device_attribute *attr,
1686 const char *buf, size_t len)
1687{
1688 struct net_device *netdev = to_net_dev(dev);
1689 struct netfront_info *np = netdev_priv(netdev);
1690 char *endp;
1691 unsigned long target;
1692
1693 if (!capable(CAP_NET_ADMIN))
1694 return -EPERM;
1695
1696 target = simple_strtoul(buf, &endp, 0);
1697 if (endp == buf)
1698 return -EBADMSG;
1699
1700 if (target < RX_MIN_TARGET)
1701 target = RX_MIN_TARGET;
1702 if (target > RX_MAX_TARGET)
1703 target = RX_MAX_TARGET;
1704
1705 spin_lock_bh(&np->rx_lock);
1706 if (target > np->rx_max_target)
1707 np->rx_max_target = target;
1708 np->rx_min_target = target;
1709 if (target > np->rx_target)
1710 np->rx_target = target;
1711
1712 xennet_alloc_rx_buffers(netdev);
1713
1714 spin_unlock_bh(&np->rx_lock);
1715 return len;
1716}
1717
1718static ssize_t show_rxbuf_max(struct device *dev,
1719 struct device_attribute *attr, char *buf)
1720{
1721 struct net_device *netdev = to_net_dev(dev);
1722 struct netfront_info *info = netdev_priv(netdev);
1723
1724 return sprintf(buf, "%u\n", info->rx_max_target);
1725}
1726
1727static ssize_t store_rxbuf_max(struct device *dev,
1728 struct device_attribute *attr,
1729 const char *buf, size_t len)
1730{
1731 struct net_device *netdev = to_net_dev(dev);
1732 struct netfront_info *np = netdev_priv(netdev);
1733 char *endp;
1734 unsigned long target;
1735
1736 if (!capable(CAP_NET_ADMIN))
1737 return -EPERM;
1738
1739 target = simple_strtoul(buf, &endp, 0);
1740 if (endp == buf)
1741 return -EBADMSG;
1742
1743 if (target < RX_MIN_TARGET)
1744 target = RX_MIN_TARGET;
1745 if (target > RX_MAX_TARGET)
1746 target = RX_MAX_TARGET;
1747
1748 spin_lock_bh(&np->rx_lock);
1749 if (target < np->rx_min_target)
1750 np->rx_min_target = target;
1751 np->rx_max_target = target;
1752 if (target < np->rx_target)
1753 np->rx_target = target;
1754
1755 xennet_alloc_rx_buffers(netdev);
1756
1757 spin_unlock_bh(&np->rx_lock);
1758 return len;
1759}
1760
1761static ssize_t show_rxbuf_cur(struct device *dev,
1762 struct device_attribute *attr, char *buf)
1763{
1764 struct net_device *netdev = to_net_dev(dev);
1765 struct netfront_info *info = netdev_priv(netdev);
1766
1767 return sprintf(buf, "%u\n", info->rx_target);
1768}
1769
1770static struct device_attribute xennet_attrs[] = {
1771 __ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf_min, store_rxbuf_min),
1772 __ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf_max, store_rxbuf_max),
1773 __ATTR(rxbuf_cur, S_IRUGO, show_rxbuf_cur, NULL),
1774};
1775
1776static int xennet_sysfs_addif(struct net_device *netdev)
1777{
1778 int i;
1779 int err;
1780
1781 for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++) {
1782 err = device_create_file(&netdev->dev,
1783 &xennet_attrs[i]);
1784 if (err)
1785 goto fail;
1786 }
1787 return 0;
1788
1789 fail:
1790 while (--i >= 0)
1791 device_remove_file(&netdev->dev, &xennet_attrs[i]);
1792 return err;
1793}
1794
1795static void xennet_sysfs_delif(struct net_device *netdev)
1796{
1797 int i;
1798
1799 for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++)
1800 device_remove_file(&netdev->dev, &xennet_attrs[i]);
1801}
1802
1803#endif /* CONFIG_SYSFS */
1804
1805static struct xenbus_device_id netfront_ids[] = {
1806 { "vif" },
1807 { "" }
1808};
1809
1810
1811static int __devexit xennet_remove(struct xenbus_device *dev)
1812{
1813 struct netfront_info *info = dev->dev.driver_data;
1814
1815 dev_dbg(&dev->dev, "%s\n", dev->nodename);
1816
1817 unregister_netdev(info->netdev);
1818
1819 xennet_disconnect_backend(info);
1820
1821 del_timer_sync(&info->rx_refill_timer);
1822
1823 xennet_sysfs_delif(info->netdev);
1824
1825 free_netdev(info->netdev);
1826
1827 return 0;
1828}
1829
1830static struct xenbus_driver netfront = {
1831 .name = "vif",
1832 .owner = THIS_MODULE,
1833 .ids = netfront_ids,
1834 .probe = netfront_probe,
1835 .remove = __devexit_p(xennet_remove),
1836 .resume = netfront_resume,
1837 .otherend_changed = backend_changed,
1838};
1839
1840static int __init netif_init(void)
1841{
1842 if (!is_running_on_xen())
1843 return -ENODEV;
1844
1845 if (is_initial_xendomain())
1846 return 0;
1847
1848 printk(KERN_INFO "Initialising Xen virtual ethernet driver.\n");
1849
1850 return xenbus_register_frontend(&netfront);
1851}
1852module_init(netif_init);
1853
1854
1855static void __exit netif_exit(void)
1856{
1857 if (is_initial_xendomain())
1858 return;
1859
1860 return xenbus_unregister_driver(&netfront);
1861}
1862module_exit(netif_exit);
1863
1864MODULE_DESCRIPTION("Xen virtual network device frontend");
1865MODULE_LICENSE("GPL");
Clone of davem's net-next-2.6 tree