?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
mfd: wm8350-i2c: Make sure the i2c regmap functions are compiled
[linux/fpc-iii.git] / drivers / infiniband / ulp / ipoib / ipoib_main.c
blob469f98156b284f66b752fc6779d13fec34ccf7fe
1 /*
2 * Copyright (c) 2004 Topspin Communications. All rights reserved.
3 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
4 * Copyright (c) 2004 Voltaire, Inc. All rights reserved.
5 *
6 * This software is available to you under a choice of one of two
7 * licenses. You may choose to be licensed under the terms of the GNU
8 * General Public License (GPL) Version 2, available from the file
9 * COPYING in the main directory of this source tree, or the
10 * OpenIB.org BSD license below:
11 *
12 * Redistribution and use in source and binary forms, with or
13 * without modification, are permitted provided that the following
14 * conditions are met:
15 *
16 * - Redistributions of source code must retain the above
17 * copyright notice, this list of conditions and the following
18 * disclaimer.
19 *
20 * - Redistributions in binary form must reproduce the above
21 * copyright notice, this list of conditions and the following
22 * disclaimer in the documentation and/or other materials
23 * provided with the distribution.
24 *
25 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
26 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
27 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
28 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
29 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
30 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
31 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
32 * SOFTWARE.
33 */
34
35 #include "ipoib.h"
36
37 #include <linux/module.h>
38
39 #include <linux/init.h>
40 #include <linux/slab.h>
41 #include <linux/kernel.h>
42 #include <linux/vmalloc.h>
43
44 #include <linux/if_arp.h> /* For ARPHRD_xxx */
45
46 #include <linux/ip.h>
47 #include <linux/in.h>
48
49 #include <linux/jhash.h>
50 #include <net/arp.h>
51
52 #define DRV_VERSION "1.0.0"
53
54 const char ipoib_driver_version[] = DRV_VERSION;
55
56 MODULE_AUTHOR("Roland Dreier");
57 MODULE_DESCRIPTION("IP-over-InfiniBand net driver");
58 MODULE_LICENSE("Dual BSD/GPL");
59 MODULE_VERSION(DRV_VERSION);
60
61 int ipoib_sendq_size __read_mostly = IPOIB_TX_RING_SIZE;
62 int ipoib_recvq_size __read_mostly = IPOIB_RX_RING_SIZE;
63
64 module_param_named(send_queue_size, ipoib_sendq_size, int, 0444);
65 MODULE_PARM_DESC(send_queue_size, "Number of descriptors in send queue");
66 module_param_named(recv_queue_size, ipoib_recvq_size, int, 0444);
67 MODULE_PARM_DESC(recv_queue_size, "Number of descriptors in receive queue");
68
69 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
70 int ipoib_debug_level;
71
72 module_param_named(debug_level, ipoib_debug_level, int, 0644);
73 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0");
74 #endif
75
76 struct ipoib_path_iter {
77 struct net_device *dev;
78 struct ipoib_path path;
79 };
80
81 static const u8 ipv4_bcast_addr[] = {
82 0x00, 0xff, 0xff, 0xff,
83 0xff, 0x12, 0x40, 0x1b, 0x00, 0x00, 0x00, 0x00,
84 0x00, 0x00, 0x00, 0x00, 0xff, 0xff, 0xff, 0xff
85 };
86
87 struct workqueue_struct *ipoib_workqueue;
88
89 struct ib_sa_client ipoib_sa_client;
90
91 static void ipoib_add_one(struct ib_device *device);
92 static void ipoib_remove_one(struct ib_device *device);
93 static void ipoib_neigh_reclaim(struct rcu_head *rp);
94
95 static struct ib_client ipoib_client = {
96 .name = "ipoib",
97 .add = ipoib_add_one,
98 .remove = ipoib_remove_one
99 };
100
101 int ipoib_open(struct net_device *dev)
102 {
103 struct ipoib_dev_priv *priv = netdev_priv(dev);
104
105 ipoib_dbg(priv, "bringing up interface\n");
106
107 set_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags);
108
109 if (ipoib_pkey_dev_delay_open(dev))
110 return 0;
111
112 if (ipoib_ib_dev_open(dev))
113 goto err_disable;
114
115 if (ipoib_ib_dev_up(dev))
116 goto err_stop;
117
118 if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) {
119 struct ipoib_dev_priv *cpriv;
120
121 /* Bring up any child interfaces too */
122 mutex_lock(&priv->vlan_mutex);
123 list_for_each_entry(cpriv, &priv->child_intfs, list) {
124 int flags;
125
126 flags = cpriv->dev->flags;
127 if (flags & IFF_UP)
128 continue;
129
130 dev_change_flags(cpriv->dev, flags | IFF_UP);
131 }
132 mutex_unlock(&priv->vlan_mutex);
133 }
134
135 netif_start_queue(dev);
136
137 return 0;
138
139 err_stop:
140 ipoib_ib_dev_stop(dev, 1);
141
142 err_disable:
143 clear_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags);
144
145 return -EINVAL;
146 }
147
148 static int ipoib_stop(struct net_device *dev)
149 {
150 struct ipoib_dev_priv *priv = netdev_priv(dev);
151
152 ipoib_dbg(priv, "stopping interface\n");
153
154 clear_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags);
155
156 netif_stop_queue(dev);
157
158 ipoib_ib_dev_down(dev, 1);
159 ipoib_ib_dev_stop(dev, 0);
160
161 if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) {
162 struct ipoib_dev_priv *cpriv;
163
164 /* Bring down any child interfaces too */
165 mutex_lock(&priv->vlan_mutex);
166 list_for_each_entry(cpriv, &priv->child_intfs, list) {
167 int flags;
168
169 flags = cpriv->dev->flags;
170 if (!(flags & IFF_UP))
171 continue;
172
173 dev_change_flags(cpriv->dev, flags & ~IFF_UP);
174 }
175 mutex_unlock(&priv->vlan_mutex);
176 }
177
178 return 0;
179 }
180
181 static void ipoib_uninit(struct net_device *dev)
182 {
183 ipoib_dev_cleanup(dev);
184 }
185
186 static netdev_features_t ipoib_fix_features(struct net_device *dev, netdev_features_t features)
187 {
188 struct ipoib_dev_priv *priv = netdev_priv(dev);
189
190 if (test_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags))
191 features &= ~(NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_TSO);
192
193 return features;
194 }
195
196 static int ipoib_change_mtu(struct net_device *dev, int new_mtu)
197 {
198 struct ipoib_dev_priv *priv = netdev_priv(dev);
199
200 /* dev->mtu > 2K ==> connected mode */
201 if (ipoib_cm_admin_enabled(dev)) {
202 if (new_mtu > ipoib_cm_max_mtu(dev))
203 return -EINVAL;
204
205 if (new_mtu > priv->mcast_mtu)
206 ipoib_warn(priv, "mtu > %d will cause multicast packet drops.\n",
207 priv->mcast_mtu);
208
209 dev->mtu = new_mtu;
210 return 0;
211 }
212
213 if (new_mtu > IPOIB_UD_MTU(priv->max_ib_mtu))
214 return -EINVAL;
215
216 priv->admin_mtu = new_mtu;
217
218 dev->mtu = min(priv->mcast_mtu, priv->admin_mtu);
219
220 return 0;
221 }
222
223 int ipoib_set_mode(struct net_device *dev, const char *buf)
224 {
225 struct ipoib_dev_priv *priv = netdev_priv(dev);
226
227 /* flush paths if we switch modes so that connections are restarted */
228 if (IPOIB_CM_SUPPORTED(dev->dev_addr) && !strcmp(buf, "connected\n")) {
229 set_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags);
230 ipoib_warn(priv, "enabling connected mode "
231 "will cause multicast packet drops\n");
232 netdev_update_features(dev);
233 rtnl_unlock();
234 priv->tx_wr.send_flags &= ~IB_SEND_IP_CSUM;
235
236 ipoib_flush_paths(dev);
237 rtnl_lock();
238 return 0;
239 }
240
241 if (!strcmp(buf, "datagram\n")) {
242 clear_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags);
243 netdev_update_features(dev);
244 dev_set_mtu(dev, min(priv->mcast_mtu, dev->mtu));
245 rtnl_unlock();
246 ipoib_flush_paths(dev);
247 rtnl_lock();
248 return 0;
249 }
250
251 return -EINVAL;
252 }
253
254 struct ipoib_path *__path_find(struct net_device *dev, void *gid)
255 {
256 struct ipoib_dev_priv *priv = netdev_priv(dev);
257 struct rb_node *n = priv->path_tree.rb_node;
258 struct ipoib_path *path;
259 int ret;
260
261 while (n) {
262 path = rb_entry(n, struct ipoib_path, rb_node);
263
264 ret = memcmp(gid, path->pathrec.dgid.raw,
265 sizeof (union ib_gid));
266
267 if (ret < 0)
268 n = n->rb_left;
269 else if (ret > 0)
270 n = n->rb_right;
271 else
272 return path;
273 }
274
275 return NULL;
276 }
277
278 static int __path_add(struct net_device *dev, struct ipoib_path *path)
279 {
280 struct ipoib_dev_priv *priv = netdev_priv(dev);
281 struct rb_node **n = &priv->path_tree.rb_node;
282 struct rb_node *pn = NULL;
283 struct ipoib_path *tpath;
284 int ret;
285
286 while (*n) {
287 pn = *n;
288 tpath = rb_entry(pn, struct ipoib_path, rb_node);
289
290 ret = memcmp(path->pathrec.dgid.raw, tpath->pathrec.dgid.raw,
291 sizeof (union ib_gid));
292 if (ret < 0)
293 n = &pn->rb_left;
294 else if (ret > 0)
295 n = &pn->rb_right;
296 else
297 return -EEXIST;
298 }
299
300 rb_link_node(&path->rb_node, pn, n);
301 rb_insert_color(&path->rb_node, &priv->path_tree);
302
303 list_add_tail(&path->list, &priv->path_list);
304
305 return 0;
306 }
307
308 static void path_free(struct net_device *dev, struct ipoib_path *path)
309 {
310 struct sk_buff *skb;
311
312 while ((skb = __skb_dequeue(&path->queue)))
313 dev_kfree_skb_irq(skb);
314
315 ipoib_dbg(netdev_priv(dev), "path_free\n");
316
317 /* remove all neigh connected to this path */
318 ipoib_del_neighs_by_gid(dev, path->pathrec.dgid.raw);
319
320 if (path->ah)
321 ipoib_put_ah(path->ah);
322
323 kfree(path);
324 }
325
326 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
327
328 struct ipoib_path_iter *ipoib_path_iter_init(struct net_device *dev)
329 {
330 struct ipoib_path_iter *iter;
331
332 iter = kmalloc(sizeof *iter, GFP_KERNEL);
333 if (!iter)
334 return NULL;
335
336 iter->dev = dev;
337 memset(iter->path.pathrec.dgid.raw, 0, 16);
338
339 if (ipoib_path_iter_next(iter)) {
340 kfree(iter);
341 return NULL;
342 }
343
344 return iter;
345 }
346
347 int ipoib_path_iter_next(struct ipoib_path_iter *iter)
348 {
349 struct ipoib_dev_priv *priv = netdev_priv(iter->dev);
350 struct rb_node *n;
351 struct ipoib_path *path;
352 int ret = 1;
353
354 spin_lock_irq(&priv->lock);
355
356 n = rb_first(&priv->path_tree);
357
358 while (n) {
359 path = rb_entry(n, struct ipoib_path, rb_node);
360
361 if (memcmp(iter->path.pathrec.dgid.raw, path->pathrec.dgid.raw,
362 sizeof (union ib_gid)) < 0) {
363 iter->path = *path;
364 ret = 0;
365 break;
366 }
367
368 n = rb_next(n);
369 }
370
371 spin_unlock_irq(&priv->lock);
372
373 return ret;
374 }
375
376 void ipoib_path_iter_read(struct ipoib_path_iter *iter,
377 struct ipoib_path *path)
378 {
379 *path = iter->path;
380 }
381
382 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */
383
384 void ipoib_mark_paths_invalid(struct net_device *dev)
385 {
386 struct ipoib_dev_priv *priv = netdev_priv(dev);
387 struct ipoib_path *path, *tp;
388
389 spin_lock_irq(&priv->lock);
390
391 list_for_each_entry_safe(path, tp, &priv->path_list, list) {
392 ipoib_dbg(priv, "mark path LID 0x%04x GID %pI6 invalid\n",
393 be16_to_cpu(path->pathrec.dlid),
394 path->pathrec.dgid.raw);
395 path->valid = 0;
396 }
397
398 spin_unlock_irq(&priv->lock);
399 }
400
401 void ipoib_flush_paths(struct net_device *dev)
402 {
403 struct ipoib_dev_priv *priv = netdev_priv(dev);
404 struct ipoib_path *path, *tp;
405 LIST_HEAD(remove_list);
406 unsigned long flags;
407
408 netif_tx_lock_bh(dev);
409 spin_lock_irqsave(&priv->lock, flags);
410
411 list_splice_init(&priv->path_list, &remove_list);
412
413 list_for_each_entry(path, &remove_list, list)
414 rb_erase(&path->rb_node, &priv->path_tree);
415
416 list_for_each_entry_safe(path, tp, &remove_list, list) {
417 if (path->query)
418 ib_sa_cancel_query(path->query_id, path->query);
419 spin_unlock_irqrestore(&priv->lock, flags);
420 netif_tx_unlock_bh(dev);
421 wait_for_completion(&path->done);
422 path_free(dev, path);
423 netif_tx_lock_bh(dev);
424 spin_lock_irqsave(&priv->lock, flags);
425 }
426
427 spin_unlock_irqrestore(&priv->lock, flags);
428 netif_tx_unlock_bh(dev);
429 }
430
431 static void path_rec_completion(int status,
432 struct ib_sa_path_rec *pathrec,
433 void *path_ptr)
434 {
435 struct ipoib_path *path = path_ptr;
436 struct net_device *dev = path->dev;
437 struct ipoib_dev_priv *priv = netdev_priv(dev);
438 struct ipoib_ah *ah = NULL;
439 struct ipoib_ah *old_ah = NULL;
440 struct ipoib_neigh *neigh, *tn;
441 struct sk_buff_head skqueue;
442 struct sk_buff *skb;
443 unsigned long flags;
444
445 if (!status)
446 ipoib_dbg(priv, "PathRec LID 0x%04x for GID %pI6\n",
447 be16_to_cpu(pathrec->dlid), pathrec->dgid.raw);
448 else
449 ipoib_dbg(priv, "PathRec status %d for GID %pI6\n",
450 status, path->pathrec.dgid.raw);
451
452 skb_queue_head_init(&skqueue);
453
454 if (!status) {
455 struct ib_ah_attr av;
456
457 if (!ib_init_ah_from_path(priv->ca, priv->port, pathrec, &av))
458 ah = ipoib_create_ah(dev, priv->pd, &av);
459 }
460
461 spin_lock_irqsave(&priv->lock, flags);
462
463 if (!IS_ERR_OR_NULL(ah)) {
464 path->pathrec = *pathrec;
465
466 old_ah = path->ah;
467 path->ah = ah;
468
469 ipoib_dbg(priv, "created address handle %p for LID 0x%04x, SL %d\n",
470 ah, be16_to_cpu(pathrec->dlid), pathrec->sl);
471
472 while ((skb = __skb_dequeue(&path->queue)))
473 __skb_queue_tail(&skqueue, skb);
474
475 list_for_each_entry_safe(neigh, tn, &path->neigh_list, list) {
476 if (neigh->ah) {
477 WARN_ON(neigh->ah != old_ah);
478 /*
479 * Dropping the ah reference inside
480 * priv->lock is safe here, because we
481 * will hold one more reference from
482 * the original value of path->ah (ie
483 * old_ah).
484 */
485 ipoib_put_ah(neigh->ah);
486 }
487 kref_get(&path->ah->ref);
488 neigh->ah = path->ah;
489
490 if (ipoib_cm_enabled(dev, neigh->daddr)) {
491 if (!ipoib_cm_get(neigh))
492 ipoib_cm_set(neigh, ipoib_cm_create_tx(dev,
493 path,
494 neigh));
495 if (!ipoib_cm_get(neigh)) {
496 ipoib_neigh_free(neigh);
497 continue;
498 }
499 }
500
501 while ((skb = __skb_dequeue(&neigh->queue)))
502 __skb_queue_tail(&skqueue, skb);
503 }
504 path->valid = 1;
505 }
506
507 path->query = NULL;
508 complete(&path->done);
509
510 spin_unlock_irqrestore(&priv->lock, flags);
511
512 if (IS_ERR_OR_NULL(ah))
513 ipoib_del_neighs_by_gid(dev, path->pathrec.dgid.raw);
514
515 if (old_ah)
516 ipoib_put_ah(old_ah);
517
518 while ((skb = __skb_dequeue(&skqueue))) {
519 skb->dev = dev;
520 if (dev_queue_xmit(skb))
521 ipoib_warn(priv, "dev_queue_xmit failed "
522 "to requeue packet\n");
523 }
524 }
525
526 static struct ipoib_path *path_rec_create(struct net_device *dev, void *gid)
527 {
528 struct ipoib_dev_priv *priv = netdev_priv(dev);
529 struct ipoib_path *path;
530
531 if (!priv->broadcast)
532 return NULL;
533
534 path = kzalloc(sizeof *path, GFP_ATOMIC);
535 if (!path)
536 return NULL;
537
538 path->dev = dev;
539
540 skb_queue_head_init(&path->queue);
541
542 INIT_LIST_HEAD(&path->neigh_list);
543
544 memcpy(path->pathrec.dgid.raw, gid, sizeof (union ib_gid));
545 path->pathrec.sgid = priv->local_gid;
546 path->pathrec.pkey = cpu_to_be16(priv->pkey);
547 path->pathrec.numb_path = 1;
548 path->pathrec.traffic_class = priv->broadcast->mcmember.traffic_class;
549
550 return path;
551 }
552
553 static int path_rec_start(struct net_device *dev,
554 struct ipoib_path *path)
555 {
556 struct ipoib_dev_priv *priv = netdev_priv(dev);
557
558 ipoib_dbg(priv, "Start path record lookup for %pI6\n",
559 path->pathrec.dgid.raw);
560
561 init_completion(&path->done);
562
563 path->query_id =
564 ib_sa_path_rec_get(&ipoib_sa_client, priv->ca, priv->port,
565 &path->pathrec,
566 IB_SA_PATH_REC_DGID |
567 IB_SA_PATH_REC_SGID |
568 IB_SA_PATH_REC_NUMB_PATH |
569 IB_SA_PATH_REC_TRAFFIC_CLASS |
570 IB_SA_PATH_REC_PKEY,
571 1000, GFP_ATOMIC,
572 path_rec_completion,
573 path, &path->query);
574 if (path->query_id < 0) {
575 ipoib_warn(priv, "ib_sa_path_rec_get failed: %d\n", path->query_id);
576 path->query = NULL;
577 complete(&path->done);
578 return path->query_id;
579 }
580
581 return 0;
582 }
583
584 static void neigh_add_path(struct sk_buff *skb, u8 *daddr,
585 struct net_device *dev)
586 {
587 struct ipoib_dev_priv *priv = netdev_priv(dev);
588 struct ipoib_path *path;
589 struct ipoib_neigh *neigh;
590 unsigned long flags;
591
592 spin_lock_irqsave(&priv->lock, flags);
593 neigh = ipoib_neigh_alloc(daddr, dev);
594 if (!neigh) {
595 spin_unlock_irqrestore(&priv->lock, flags);
596 ++dev->stats.tx_dropped;
597 dev_kfree_skb_any(skb);
598 return;
599 }
600
601 path = __path_find(dev, daddr + 4);
602 if (!path) {
603 path = path_rec_create(dev, daddr + 4);
604 if (!path)
605 goto err_path;
606
607 __path_add(dev, path);
608 }
609
610 list_add_tail(&neigh->list, &path->neigh_list);
611
612 if (path->ah) {
613 kref_get(&path->ah->ref);
614 neigh->ah = path->ah;
615
616 if (ipoib_cm_enabled(dev, neigh->daddr)) {
617 if (!ipoib_cm_get(neigh))
618 ipoib_cm_set(neigh, ipoib_cm_create_tx(dev, path, neigh));
619 if (!ipoib_cm_get(neigh)) {
620 ipoib_neigh_free(neigh);
621 goto err_drop;
622 }
623 if (skb_queue_len(&neigh->queue) < IPOIB_MAX_PATH_REC_QUEUE)
624 __skb_queue_tail(&neigh->queue, skb);
625 else {
626 ipoib_warn(priv, "queue length limit %d. Packet drop.\n",
627 skb_queue_len(&neigh->queue));
628 goto err_drop;
629 }
630 } else {
631 spin_unlock_irqrestore(&priv->lock, flags);
632 ipoib_send(dev, skb, path->ah, IPOIB_QPN(daddr));
633 ipoib_neigh_put(neigh);
634 return;
635 }
636 } else {
637 neigh->ah = NULL;
638
639 if (!path->query && path_rec_start(dev, path))
640 goto err_path;
641
642 __skb_queue_tail(&neigh->queue, skb);
643 }
644
645 spin_unlock_irqrestore(&priv->lock, flags);
646 ipoib_neigh_put(neigh);
647 return;
648
649 err_path:
650 ipoib_neigh_free(neigh);
651 err_drop:
652 ++dev->stats.tx_dropped;
653 dev_kfree_skb_any(skb);
654
655 spin_unlock_irqrestore(&priv->lock, flags);
656 ipoib_neigh_put(neigh);
657 }
658
659 static void unicast_arp_send(struct sk_buff *skb, struct net_device *dev,
660 struct ipoib_cb *cb)
661 {
662 struct ipoib_dev_priv *priv = netdev_priv(dev);
663 struct ipoib_path *path;
664 unsigned long flags;
665
666 spin_lock_irqsave(&priv->lock, flags);
667
668 path = __path_find(dev, cb->hwaddr + 4);
669 if (!path || !path->valid) {
670 int new_path = 0;
671
672 if (!path) {
673 path = path_rec_create(dev, cb->hwaddr + 4);
674 new_path = 1;
675 }
676 if (path) {
677 __skb_queue_tail(&path->queue, skb);
678
679 if (!path->query && path_rec_start(dev, path)) {
680 spin_unlock_irqrestore(&priv->lock, flags);
681 if (new_path)
682 path_free(dev, path);
683 return;
684 } else
685 __path_add(dev, path);
686 } else {
687 ++dev->stats.tx_dropped;
688 dev_kfree_skb_any(skb);
689 }
690
691 spin_unlock_irqrestore(&priv->lock, flags);
692 return;
693 }
694
695 if (path->ah) {
696 ipoib_dbg(priv, "Send unicast ARP to %04x\n",
697 be16_to_cpu(path->pathrec.dlid));
698
699 spin_unlock_irqrestore(&priv->lock, flags);
700 ipoib_send(dev, skb, path->ah, IPOIB_QPN(cb->hwaddr));
701 return;
702 } else if ((path->query || !path_rec_start(dev, path)) &&
703 skb_queue_len(&path->queue) < IPOIB_MAX_PATH_REC_QUEUE) {
704 __skb_queue_tail(&path->queue, skb);
705 } else {
706 ++dev->stats.tx_dropped;
707 dev_kfree_skb_any(skb);
708 }
709
710 spin_unlock_irqrestore(&priv->lock, flags);
711 }
712
713 static int ipoib_start_xmit(struct sk_buff *skb, struct net_device *dev)
714 {
715 struct ipoib_dev_priv *priv = netdev_priv(dev);
716 struct ipoib_neigh *neigh;
717 struct ipoib_cb *cb = (struct ipoib_cb *) skb->cb;
718 struct ipoib_header *header;
719 unsigned long flags;
720
721 header = (struct ipoib_header *) skb->data;
722
723 if (unlikely(cb->hwaddr[4] == 0xff)) {
724 /* multicast, arrange "if" according to probability */
725 if ((header->proto != htons(ETH_P_IP)) &&
726 (header->proto != htons(ETH_P_IPV6)) &&
727 (header->proto != htons(ETH_P_ARP)) &&
728 (header->proto != htons(ETH_P_RARP)) &&
729 (header->proto != htons(ETH_P_TIPC))) {
730 /* ethertype not supported by IPoIB */
731 ++dev->stats.tx_dropped;
732 dev_kfree_skb_any(skb);
733 return NETDEV_TX_OK;
734 }
735 /* Add in the P_Key for multicast*/
736 cb->hwaddr[8] = (priv->pkey >> 8) & 0xff;
737 cb->hwaddr[9] = priv->pkey & 0xff;
738
739 neigh = ipoib_neigh_get(dev, cb->hwaddr);
740 if (likely(neigh))
741 goto send_using_neigh;
742 ipoib_mcast_send(dev, cb->hwaddr, skb);
743 return NETDEV_TX_OK;
744 }
745
746 /* unicast, arrange "switch" according to probability */
747 switch (header->proto) {
748 case htons(ETH_P_IP):
749 case htons(ETH_P_IPV6):
750 case htons(ETH_P_TIPC):
751 neigh = ipoib_neigh_get(dev, cb->hwaddr);
752 if (unlikely(!neigh)) {
753 neigh_add_path(skb, cb->hwaddr, dev);
754 return NETDEV_TX_OK;
755 }
756 break;
757 case htons(ETH_P_ARP):
758 case htons(ETH_P_RARP):
759 /* for unicast ARP and RARP should always perform path find */
760 unicast_arp_send(skb, dev, cb);
761 return NETDEV_TX_OK;
762 default:
763 /* ethertype not supported by IPoIB */
764 ++dev->stats.tx_dropped;
765 dev_kfree_skb_any(skb);
766 return NETDEV_TX_OK;
767 }
768
769 send_using_neigh:
770 /* note we now hold a ref to neigh */
771 if (ipoib_cm_get(neigh)) {
772 if (ipoib_cm_up(neigh)) {
773 ipoib_cm_send(dev, skb, ipoib_cm_get(neigh));
774 goto unref;
775 }
776 } else if (neigh->ah) {
777 ipoib_send(dev, skb, neigh->ah, IPOIB_QPN(cb->hwaddr));
778 goto unref;
779 }
780
781 if (skb_queue_len(&neigh->queue) < IPOIB_MAX_PATH_REC_QUEUE) {
782 spin_lock_irqsave(&priv->lock, flags);
783 __skb_queue_tail(&neigh->queue, skb);
784 spin_unlock_irqrestore(&priv->lock, flags);
785 } else {
786 ++dev->stats.tx_dropped;
787 dev_kfree_skb_any(skb);
788 }
789
790 unref:
791 ipoib_neigh_put(neigh);
792
793 return NETDEV_TX_OK;
794 }
795
796 static void ipoib_timeout(struct net_device *dev)
797 {
798 struct ipoib_dev_priv *priv = netdev_priv(dev);
799
800 ipoib_warn(priv, "transmit timeout: latency %d msecs\n",
801 jiffies_to_msecs(jiffies - dev->trans_start));
802 ipoib_warn(priv, "queue stopped %d, tx_head %u, tx_tail %u\n",
803 netif_queue_stopped(dev),
804 priv->tx_head, priv->tx_tail);
805 /* XXX reset QP, etc. */
806 }
807
808 static int ipoib_hard_header(struct sk_buff *skb,
809 struct net_device *dev,
810 unsigned short type,
811 const void *daddr, const void *saddr, unsigned len)
812 {
813 struct ipoib_header *header;
814 struct ipoib_cb *cb = (struct ipoib_cb *) skb->cb;
815
816 header = (struct ipoib_header *) skb_push(skb, sizeof *header);
817
818 header->proto = htons(type);
819 header->reserved = 0;
820
821 /*
822 * we don't rely on dst_entry structure, always stuff the
823 * destination address into skb->cb so we can figure out where
824 * to send the packet later.
825 */
826 memcpy(cb->hwaddr, daddr, INFINIBAND_ALEN);
827
828 return sizeof *header;
829 }
830
831 static void ipoib_set_mcast_list(struct net_device *dev)
832 {
833 struct ipoib_dev_priv *priv = netdev_priv(dev);
834
835 if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags)) {
836 ipoib_dbg(priv, "IPOIB_FLAG_OPER_UP not set");
837 return;
838 }
839
840 queue_work(ipoib_workqueue, &priv->restart_task);
841 }
842
843 static u32 ipoib_addr_hash(struct ipoib_neigh_hash *htbl, u8 *daddr)
844 {
845 /*
846 * Use only the address parts that contributes to spreading
847 * The subnet prefix is not used as one can not connect to
848 * same remote port (GUID) using the same remote QPN via two
849 * different subnets.
850 */
851 /* qpn octets[1:4) & port GUID octets[12:20) */
852 u32 *d32 = (u32 *) daddr;
853 u32 hv;
854
855 hv = jhash_3words(d32[3], d32[4], IPOIB_QPN_MASK & d32[0], 0);
856 return hv & htbl->mask;
857 }
858
859 struct ipoib_neigh *ipoib_neigh_get(struct net_device *dev, u8 *daddr)
860 {
861 struct ipoib_dev_priv *priv = netdev_priv(dev);
862 struct ipoib_neigh_table *ntbl = &priv->ntbl;
863 struct ipoib_neigh_hash *htbl;
864 struct ipoib_neigh *neigh = NULL;
865 u32 hash_val;
866
867 rcu_read_lock_bh();
868
869 htbl = rcu_dereference_bh(ntbl->htbl);
870
871 if (!htbl)
872 goto out_unlock;
873
874 hash_val = ipoib_addr_hash(htbl, daddr);
875 for (neigh = rcu_dereference_bh(htbl->buckets[hash_val]);
876 neigh != NULL;
877 neigh = rcu_dereference_bh(neigh->hnext)) {
878 if (memcmp(daddr, neigh->daddr, INFINIBAND_ALEN) == 0) {
879 /* found, take one ref on behalf of the caller */
880 if (!atomic_inc_not_zero(&neigh->refcnt)) {
881 /* deleted */
882 neigh = NULL;
883 goto out_unlock;
884 }
885
886 if (likely(skb_queue_len(&neigh->queue) < IPOIB_MAX_PATH_REC_QUEUE))
887 neigh->alive = jiffies;
888 goto out_unlock;
889 }
890 }
891
892 out_unlock:
893 rcu_read_unlock_bh();
894 return neigh;
895 }
896
897 static void __ipoib_reap_neigh(struct ipoib_dev_priv *priv)
898 {
899 struct ipoib_neigh_table *ntbl = &priv->ntbl;
900 struct ipoib_neigh_hash *htbl;
901 unsigned long neigh_obsolete;
902 unsigned long dt;
903 unsigned long flags;
904 int i;
905
906 if (test_bit(IPOIB_STOP_NEIGH_GC, &priv->flags))
907 return;
908
909 spin_lock_irqsave(&priv->lock, flags);
910
911 htbl = rcu_dereference_protected(ntbl->htbl,
912 lockdep_is_held(&priv->lock));
913
914 if (!htbl)
915 goto out_unlock;
916
917 /* neigh is obsolete if it was idle for two GC periods */
918 dt = 2 * arp_tbl.gc_interval;
919 neigh_obsolete = jiffies - dt;
920 /* handle possible race condition */
921 if (test_bit(IPOIB_STOP_NEIGH_GC, &priv->flags))
922 goto out_unlock;
923
924 for (i = 0; i < htbl->size; i++) {
925 struct ipoib_neigh *neigh;
926 struct ipoib_neigh __rcu **np = &htbl->buckets[i];
927
928 while ((neigh = rcu_dereference_protected(*np,
929 lockdep_is_held(&priv->lock))) != NULL) {
930 /* was the neigh idle for two GC periods */
931 if (time_after(neigh_obsolete, neigh->alive)) {
932 rcu_assign_pointer(*np,
933 rcu_dereference_protected(neigh->hnext,
934 lockdep_is_held(&priv->lock)));
935 /* remove from path/mc list */
936 list_del(&neigh->list);
937 call_rcu(&neigh->rcu, ipoib_neigh_reclaim);
938 } else {
939 np = &neigh->hnext;
940 }
941
942 }
943 }
944
945 out_unlock:
946 spin_unlock_irqrestore(&priv->lock, flags);
947 }
948
949 static void ipoib_reap_neigh(struct work_struct *work)
950 {
951 struct ipoib_dev_priv *priv =
952 container_of(work, struct ipoib_dev_priv, neigh_reap_task.work);
953
954 __ipoib_reap_neigh(priv);
955
956 if (!test_bit(IPOIB_STOP_NEIGH_GC, &priv->flags))
957 queue_delayed_work(ipoib_workqueue, &priv->neigh_reap_task,
958 arp_tbl.gc_interval);
959 }
960
961
962 static struct ipoib_neigh *ipoib_neigh_ctor(u8 *daddr,
963 struct net_device *dev)
964 {
965 struct ipoib_neigh *neigh;
966
967 neigh = kzalloc(sizeof *neigh, GFP_ATOMIC);
968 if (!neigh)
969 return NULL;
970
971 neigh->dev = dev;
972 memcpy(&neigh->daddr, daddr, sizeof(neigh->daddr));
973 skb_queue_head_init(&neigh->queue);
974 INIT_LIST_HEAD(&neigh->list);
975 ipoib_cm_set(neigh, NULL);
976 /* one ref on behalf of the caller */
977 atomic_set(&neigh->refcnt, 1);
978
979 return neigh;
980 }
981
982 struct ipoib_neigh *ipoib_neigh_alloc(u8 *daddr,
983 struct net_device *dev)
984 {
985 struct ipoib_dev_priv *priv = netdev_priv(dev);
986 struct ipoib_neigh_table *ntbl = &priv->ntbl;
987 struct ipoib_neigh_hash *htbl;
988 struct ipoib_neigh *neigh;
989 u32 hash_val;
990
991 htbl = rcu_dereference_protected(ntbl->htbl,
992 lockdep_is_held(&priv->lock));
993 if (!htbl) {
994 neigh = NULL;
995 goto out_unlock;
996 }
997
998 /* need to add a new neigh, but maybe some other thread succeeded?
999 * recalc hash, maybe hash resize took place so we do a search
1000 */
1001 hash_val = ipoib_addr_hash(htbl, daddr);
1002 for (neigh = rcu_dereference_protected(htbl->buckets[hash_val],
1003 lockdep_is_held(&priv->lock));
1004 neigh != NULL;
1005 neigh = rcu_dereference_protected(neigh->hnext,
1006 lockdep_is_held(&priv->lock))) {
1007 if (memcmp(daddr, neigh->daddr, INFINIBAND_ALEN) == 0) {
1008 /* found, take one ref on behalf of the caller */
1009 if (!atomic_inc_not_zero(&neigh->refcnt)) {
1010 /* deleted */
1011 neigh = NULL;
1012 break;
1013 }
1014 neigh->alive = jiffies;
1015 goto out_unlock;
1016 }
1017 }
1018
1019 neigh = ipoib_neigh_ctor(daddr, dev);
1020 if (!neigh)
1021 goto out_unlock;
1022
1023 /* one ref on behalf of the hash table */
1024 atomic_inc(&neigh->refcnt);
1025 neigh->alive = jiffies;
1026 /* put in hash */
1027 rcu_assign_pointer(neigh->hnext,
1028 rcu_dereference_protected(htbl->buckets[hash_val],
1029 lockdep_is_held(&priv->lock)));
1030 rcu_assign_pointer(htbl->buckets[hash_val], neigh);
1031 atomic_inc(&ntbl->entries);
1032
1033 out_unlock:
1034
1035 return neigh;
1036 }
1037
1038 void ipoib_neigh_dtor(struct ipoib_neigh *neigh)
1039 {
1040 /* neigh reference count was dropprd to zero */
1041 struct net_device *dev = neigh->dev;
1042 struct ipoib_dev_priv *priv = netdev_priv(dev);
1043 struct sk_buff *skb;
1044 if (neigh->ah)
1045 ipoib_put_ah(neigh->ah);
1046 while ((skb = __skb_dequeue(&neigh->queue))) {
1047 ++dev->stats.tx_dropped;
1048 dev_kfree_skb_any(skb);
1049 }
1050 if (ipoib_cm_get(neigh))
1051 ipoib_cm_destroy_tx(ipoib_cm_get(neigh));
1052 ipoib_dbg(netdev_priv(dev),
1053 "neigh free for %06x %pI6\n",
1054 IPOIB_QPN(neigh->daddr),
1055 neigh->daddr + 4);
1056 kfree(neigh);
1057 if (atomic_dec_and_test(&priv->ntbl.entries)) {
1058 if (test_bit(IPOIB_NEIGH_TBL_FLUSH, &priv->flags))
1059 complete(&priv->ntbl.flushed);
1060 }
1061 }
1062
1063 static void ipoib_neigh_reclaim(struct rcu_head *rp)
1064 {
1065 /* Called as a result of removal from hash table */
1066 struct ipoib_neigh *neigh = container_of(rp, struct ipoib_neigh, rcu);
1067 /* note TX context may hold another ref */
1068 ipoib_neigh_put(neigh);
1069 }
1070
1071 void ipoib_neigh_free(struct ipoib_neigh *neigh)
1072 {
1073 struct net_device *dev = neigh->dev;
1074 struct ipoib_dev_priv *priv = netdev_priv(dev);
1075 struct ipoib_neigh_table *ntbl = &priv->ntbl;
1076 struct ipoib_neigh_hash *htbl;
1077 struct ipoib_neigh __rcu **np;
1078 struct ipoib_neigh *n;
1079 u32 hash_val;
1080
1081 htbl = rcu_dereference_protected(ntbl->htbl,
1082 lockdep_is_held(&priv->lock));
1083 if (!htbl)
1084 return;
1085
1086 hash_val = ipoib_addr_hash(htbl, neigh->daddr);
1087 np = &htbl->buckets[hash_val];
1088 for (n = rcu_dereference_protected(*np,
1089 lockdep_is_held(&priv->lock));
1090 n != NULL;
1091 n = rcu_dereference_protected(*np,
1092 lockdep_is_held(&priv->lock))) {
1093 if (n == neigh) {
1094 /* found */
1095 rcu_assign_pointer(*np,
1096 rcu_dereference_protected(neigh->hnext,
1097 lockdep_is_held(&priv->lock)));
1098 /* remove from parent list */
1099 list_del(&neigh->list);
1100 call_rcu(&neigh->rcu, ipoib_neigh_reclaim);
1101 return;
1102 } else {
1103 np = &n->hnext;
1104 }
1105 }
1106 }
1107
1108 static int ipoib_neigh_hash_init(struct ipoib_dev_priv *priv)
1109 {
1110 struct ipoib_neigh_table *ntbl = &priv->ntbl;
1111 struct ipoib_neigh_hash *htbl;
1112 struct ipoib_neigh **buckets;
1113 u32 size;
1114
1115 clear_bit(IPOIB_NEIGH_TBL_FLUSH, &priv->flags);
1116 ntbl->htbl = NULL;
1117 htbl = kzalloc(sizeof(*htbl), GFP_KERNEL);
1118 if (!htbl)
1119 return -ENOMEM;
1120 set_bit(IPOIB_STOP_NEIGH_GC, &priv->flags);
1121 size = roundup_pow_of_two(arp_tbl.gc_thresh3);
1122 buckets = kzalloc(size * sizeof(*buckets), GFP_KERNEL);
1123 if (!buckets) {
1124 kfree(htbl);
1125 return -ENOMEM;
1126 }
1127 htbl->size = size;
1128 htbl->mask = (size - 1);
1129 htbl->buckets = buckets;
1130 ntbl->htbl = htbl;
1131 htbl->ntbl = ntbl;
1132 atomic_set(&ntbl->entries, 0);
1133
1134 /* start garbage collection */
1135 clear_bit(IPOIB_STOP_NEIGH_GC, &priv->flags);
1136 queue_delayed_work(ipoib_workqueue, &priv->neigh_reap_task,
1137 arp_tbl.gc_interval);
1138
1139 return 0;
1140 }
1141
1142 static void neigh_hash_free_rcu(struct rcu_head *head)
1143 {
1144 struct ipoib_neigh_hash *htbl = container_of(head,
1145 struct ipoib_neigh_hash,
1146 rcu);
1147 struct ipoib_neigh __rcu **buckets = htbl->buckets;
1148 struct ipoib_neigh_table *ntbl = htbl->ntbl;
1149
1150 kfree(buckets);
1151 kfree(htbl);
1152 complete(&ntbl->deleted);
1153 }
1154
1155 void ipoib_del_neighs_by_gid(struct net_device *dev, u8 *gid)
1156 {
1157 struct ipoib_dev_priv *priv = netdev_priv(dev);
1158 struct ipoib_neigh_table *ntbl = &priv->ntbl;
1159 struct ipoib_neigh_hash *htbl;
1160 unsigned long flags;
1161 int i;
1162
1163 /* remove all neigh connected to a given path or mcast */
1164 spin_lock_irqsave(&priv->lock, flags);
1165
1166 htbl = rcu_dereference_protected(ntbl->htbl,
1167 lockdep_is_held(&priv->lock));
1168
1169 if (!htbl)
1170 goto out_unlock;
1171
1172 for (i = 0; i < htbl->size; i++) {
1173 struct ipoib_neigh *neigh;
1174 struct ipoib_neigh __rcu **np = &htbl->buckets[i];
1175
1176 while ((neigh = rcu_dereference_protected(*np,
1177 lockdep_is_held(&priv->lock))) != NULL) {
1178 /* delete neighs belong to this parent */
1179 if (!memcmp(gid, neigh->daddr + 4, sizeof (union ib_gid))) {
1180 rcu_assign_pointer(*np,
1181 rcu_dereference_protected(neigh->hnext,
1182 lockdep_is_held(&priv->lock)));
1183 /* remove from parent list */
1184 list_del(&neigh->list);
1185 call_rcu(&neigh->rcu, ipoib_neigh_reclaim);
1186 } else {
1187 np = &neigh->hnext;
1188 }
1189
1190 }
1191 }
1192 out_unlock:
1193 spin_unlock_irqrestore(&priv->lock, flags);
1194 }
1195
1196 static void ipoib_flush_neighs(struct ipoib_dev_priv *priv)
1197 {
1198 struct ipoib_neigh_table *ntbl = &priv->ntbl;
1199 struct ipoib_neigh_hash *htbl;
1200 unsigned long flags;
1201 int i, wait_flushed = 0;
1202
1203 init_completion(&priv->ntbl.flushed);
1204
1205 spin_lock_irqsave(&priv->lock, flags);
1206
1207 htbl = rcu_dereference_protected(ntbl->htbl,
1208 lockdep_is_held(&priv->lock));
1209 if (!htbl)
1210 goto out_unlock;
1211
1212 wait_flushed = atomic_read(&priv->ntbl.entries);
1213 if (!wait_flushed)
1214 goto free_htbl;
1215
1216 for (i = 0; i < htbl->size; i++) {
1217 struct ipoib_neigh *neigh;
1218 struct ipoib_neigh __rcu **np = &htbl->buckets[i];
1219
1220 while ((neigh = rcu_dereference_protected(*np,
1221 lockdep_is_held(&priv->lock))) != NULL) {
1222 rcu_assign_pointer(*np,
1223 rcu_dereference_protected(neigh->hnext,
1224 lockdep_is_held(&priv->lock)));
1225 /* remove from path/mc list */
1226 list_del(&neigh->list);
1227 call_rcu(&neigh->rcu, ipoib_neigh_reclaim);
1228 }
1229 }
1230
1231 free_htbl:
1232 rcu_assign_pointer(ntbl->htbl, NULL);
1233 call_rcu(&htbl->rcu, neigh_hash_free_rcu);
1234
1235 out_unlock:
1236 spin_unlock_irqrestore(&priv->lock, flags);
1237 if (wait_flushed)
1238 wait_for_completion(&priv->ntbl.flushed);
1239 }
1240
1241 static void ipoib_neigh_hash_uninit(struct net_device *dev)
1242 {
1243 struct ipoib_dev_priv *priv = netdev_priv(dev);
1244 int stopped;
1245
1246 ipoib_dbg(priv, "ipoib_neigh_hash_uninit\n");
1247 init_completion(&priv->ntbl.deleted);
1248 set_bit(IPOIB_NEIGH_TBL_FLUSH, &priv->flags);
1249
1250 /* Stop GC if called at init fail need to cancel work */
1251 stopped = test_and_set_bit(IPOIB_STOP_NEIGH_GC, &priv->flags);
1252 if (!stopped)
1253 cancel_delayed_work(&priv->neigh_reap_task);
1254
1255 ipoib_flush_neighs(priv);
1256
1257 wait_for_completion(&priv->ntbl.deleted);
1258 }
1259
1260
1261 int ipoib_dev_init(struct net_device *dev, struct ib_device *ca, int port)
1262 {
1263 struct ipoib_dev_priv *priv = netdev_priv(dev);
1264
1265 if (ipoib_neigh_hash_init(priv) < 0)
1266 goto out;
1267 /* Allocate RX/TX "rings" to hold queued skbs */
1268 priv->rx_ring = kzalloc(ipoib_recvq_size * sizeof *priv->rx_ring,
1269 GFP_KERNEL);
1270 if (!priv->rx_ring) {
1271 printk(KERN_WARNING "%s: failed to allocate RX ring (%d entries)\n",
1272 ca->name, ipoib_recvq_size);
1273 goto out_neigh_hash_cleanup;
1274 }
1275
1276 priv->tx_ring = vzalloc(ipoib_sendq_size * sizeof *priv->tx_ring);
1277 if (!priv->tx_ring) {
1278 printk(KERN_WARNING "%s: failed to allocate TX ring (%d entries)\n",
1279 ca->name, ipoib_sendq_size);
1280 goto out_rx_ring_cleanup;
1281 }
1282
1283 /* priv->tx_head, tx_tail & tx_outstanding are already 0 */
1284
1285 if (ipoib_ib_dev_init(dev, ca, port))
1286 goto out_tx_ring_cleanup;
1287
1288 return 0;
1289
1290 out_tx_ring_cleanup:
1291 vfree(priv->tx_ring);
1292
1293 out_rx_ring_cleanup:
1294 kfree(priv->rx_ring);
1295
1296 out_neigh_hash_cleanup:
1297 ipoib_neigh_hash_uninit(dev);
1298 out:
1299 return -ENOMEM;
1300 }
1301
1302 void ipoib_dev_cleanup(struct net_device *dev)
1303 {
1304 struct ipoib_dev_priv *priv = netdev_priv(dev), *cpriv, *tcpriv;
1305 LIST_HEAD(head);
1306
1307 ASSERT_RTNL();
1308
1309 ipoib_delete_debug_files(dev);
1310
1311 /* Delete any child interfaces first */
1312 list_for_each_entry_safe(cpriv, tcpriv, &priv->child_intfs, list) {
1313 /* Stop GC on child */
1314 set_bit(IPOIB_STOP_NEIGH_GC, &cpriv->flags);
1315 cancel_delayed_work(&cpriv->neigh_reap_task);
1316 unregister_netdevice_queue(cpriv->dev, &head);
1317 }
1318 unregister_netdevice_many(&head);
1319
1320 ipoib_ib_dev_cleanup(dev);
1321
1322 kfree(priv->rx_ring);
1323 vfree(priv->tx_ring);
1324
1325 priv->rx_ring = NULL;
1326 priv->tx_ring = NULL;
1327
1328 ipoib_neigh_hash_uninit(dev);
1329 }
1330
1331 static const struct header_ops ipoib_header_ops = {
1332 .create = ipoib_hard_header,
1333 };
1334
1335 static const struct net_device_ops ipoib_netdev_ops = {
1336 .ndo_uninit = ipoib_uninit,
1337 .ndo_open = ipoib_open,
1338 .ndo_stop = ipoib_stop,
1339 .ndo_change_mtu = ipoib_change_mtu,
1340 .ndo_fix_features = ipoib_fix_features,
1341 .ndo_start_xmit = ipoib_start_xmit,
1342 .ndo_tx_timeout = ipoib_timeout,
1343 .ndo_set_rx_mode = ipoib_set_mcast_list,
1344 };
1345
1346 void ipoib_setup(struct net_device *dev)
1347 {
1348 struct ipoib_dev_priv *priv = netdev_priv(dev);
1349
1350 dev->netdev_ops = &ipoib_netdev_ops;
1351 dev->header_ops = &ipoib_header_ops;
1352
1353 ipoib_set_ethtool_ops(dev);
1354
1355 netif_napi_add(dev, &priv->napi, ipoib_poll, 100);
1356
1357 dev->watchdog_timeo = HZ;
1358
1359 dev->flags |= IFF_BROADCAST | IFF_MULTICAST;
1360
1361 dev->hard_header_len = IPOIB_ENCAP_LEN;
1362 dev->addr_len = INFINIBAND_ALEN;
1363 dev->type = ARPHRD_INFINIBAND;
1364 dev->tx_queue_len = ipoib_sendq_size * 2;
1365 dev->features = (NETIF_F_VLAN_CHALLENGED |
1366 NETIF_F_HIGHDMA);
1367 dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1368
1369 memcpy(dev->broadcast, ipv4_bcast_addr, INFINIBAND_ALEN);
1370
1371 netif_carrier_off(dev);
1372
1373 priv->dev = dev;
1374
1375 spin_lock_init(&priv->lock);
1376
1377 mutex_init(&priv->vlan_mutex);
1378
1379 INIT_LIST_HEAD(&priv->path_list);
1380 INIT_LIST_HEAD(&priv->child_intfs);
1381 INIT_LIST_HEAD(&priv->dead_ahs);
1382 INIT_LIST_HEAD(&priv->multicast_list);
1383
1384 INIT_DELAYED_WORK(&priv->pkey_poll_task, ipoib_pkey_poll);
1385 INIT_DELAYED_WORK(&priv->mcast_task, ipoib_mcast_join_task);
1386 INIT_WORK(&priv->carrier_on_task, ipoib_mcast_carrier_on_task);
1387 INIT_WORK(&priv->flush_light, ipoib_ib_dev_flush_light);
1388 INIT_WORK(&priv->flush_normal, ipoib_ib_dev_flush_normal);
1389 INIT_WORK(&priv->flush_heavy, ipoib_ib_dev_flush_heavy);
1390 INIT_WORK(&priv->restart_task, ipoib_mcast_restart_task);
1391 INIT_DELAYED_WORK(&priv->ah_reap_task, ipoib_reap_ah);
1392 INIT_DELAYED_WORK(&priv->neigh_reap_task, ipoib_reap_neigh);
1393 }
1394
1395 struct ipoib_dev_priv *ipoib_intf_alloc(const char *name)
1396 {
1397 struct net_device *dev;
1398
1399 dev = alloc_netdev((int) sizeof (struct ipoib_dev_priv), name,
1400 ipoib_setup);
1401 if (!dev)
1402 return NULL;
1403
1404 return netdev_priv(dev);
1405 }
1406
1407 static ssize_t show_pkey(struct device *dev,
1408 struct device_attribute *attr, char *buf)
1409 {
1410 struct ipoib_dev_priv *priv = netdev_priv(to_net_dev(dev));
1411
1412 return sprintf(buf, "0x%04x\n", priv->pkey);
1413 }
1414 static DEVICE_ATTR(pkey, S_IRUGO, show_pkey, NULL);
1415
1416 static ssize_t show_umcast(struct device *dev,
1417 struct device_attribute *attr, char *buf)
1418 {
1419 struct ipoib_dev_priv *priv = netdev_priv(to_net_dev(dev));
1420
1421 return sprintf(buf, "%d\n", test_bit(IPOIB_FLAG_UMCAST, &priv->flags));
1422 }
1423
1424 void ipoib_set_umcast(struct net_device *ndev, int umcast_val)
1425 {
1426 struct ipoib_dev_priv *priv = netdev_priv(ndev);
1427
1428 if (umcast_val > 0) {
1429 set_bit(IPOIB_FLAG_UMCAST, &priv->flags);
1430 ipoib_warn(priv, "ignoring multicast groups joined directly "
1431 "by userspace\n");
1432 } else
1433 clear_bit(IPOIB_FLAG_UMCAST, &priv->flags);
1434 }
1435
1436 static ssize_t set_umcast(struct device *dev,
1437 struct device_attribute *attr,
1438 const char *buf, size_t count)
1439 {
1440 unsigned long umcast_val = simple_strtoul(buf, NULL, 0);
1441
1442 ipoib_set_umcast(to_net_dev(dev), umcast_val);
1443
1444 return count;
1445 }
1446 static DEVICE_ATTR(umcast, S_IWUSR | S_IRUGO, show_umcast, set_umcast);
1447
1448 int ipoib_add_umcast_attr(struct net_device *dev)
1449 {
1450 return device_create_file(&dev->dev, &dev_attr_umcast);
1451 }
1452
1453 static ssize_t create_child(struct device *dev,
1454 struct device_attribute *attr,
1455 const char *buf, size_t count)
1456 {
1457 int pkey;
1458 int ret;
1459
1460 if (sscanf(buf, "%i", &pkey) != 1)
1461 return -EINVAL;
1462
1463 if (pkey <= 0 || pkey > 0xffff || pkey == 0x8000)
1464 return -EINVAL;
1465
1466 /*
1467 * Set the full membership bit, so that we join the right
1468 * broadcast group, etc.
1469 */
1470 pkey |= 0x8000;
1471
1472 ret = ipoib_vlan_add(to_net_dev(dev), pkey);
1473
1474 return ret ? ret : count;
1475 }
1476 static DEVICE_ATTR(create_child, S_IWUSR, NULL, create_child);
1477
1478 static ssize_t delete_child(struct device *dev,
1479 struct device_attribute *attr,
1480 const char *buf, size_t count)
1481 {
1482 int pkey;
1483 int ret;
1484
1485 if (sscanf(buf, "%i", &pkey) != 1)
1486 return -EINVAL;
1487
1488 if (pkey < 0 || pkey > 0xffff)
1489 return -EINVAL;
1490
1491 ret = ipoib_vlan_delete(to_net_dev(dev), pkey);
1492
1493 return ret ? ret : count;
1494
1495 }
1496 static DEVICE_ATTR(delete_child, S_IWUSR, NULL, delete_child);
1497
1498 int ipoib_add_pkey_attr(struct net_device *dev)
1499 {
1500 return device_create_file(&dev->dev, &dev_attr_pkey);
1501 }
1502
1503 int ipoib_set_dev_features(struct ipoib_dev_priv *priv, struct ib_device *hca)
1504 {
1505 struct ib_device_attr *device_attr;
1506 int result = -ENOMEM;
1507
1508 device_attr = kmalloc(sizeof *device_attr, GFP_KERNEL);
1509 if (!device_attr) {
1510 printk(KERN_WARNING "%s: allocation of %zu bytes failed\n",
1511 hca->name, sizeof *device_attr);
1512 return result;
1513 }
1514
1515 result = ib_query_device(hca, device_attr);
1516 if (result) {
1517 printk(KERN_WARNING "%s: ib_query_device failed (ret = %d)\n",
1518 hca->name, result);
1519 kfree(device_attr);
1520 return result;
1521 }
1522 priv->hca_caps = device_attr->device_cap_flags;
1523
1524 kfree(device_attr);
1525
1526 if (priv->hca_caps & IB_DEVICE_UD_IP_CSUM) {
1527 priv->dev->hw_features = NETIF_F_SG |
1528 NETIF_F_IP_CSUM | NETIF_F_RXCSUM;
1529
1530 if (priv->hca_caps & IB_DEVICE_UD_TSO)
1531 priv->dev->hw_features |= NETIF_F_TSO;
1532
1533 priv->dev->features |= priv->dev->hw_features;
1534 }
1535
1536 return 0;
1537 }
1538
1539 static struct net_device *ipoib_add_port(const char *format,
1540 struct ib_device *hca, u8 port)
1541 {
1542 struct ipoib_dev_priv *priv;
1543 struct ib_port_attr attr;
1544 int result = -ENOMEM;
1545
1546 priv = ipoib_intf_alloc(format);
1547 if (!priv)
1548 goto alloc_mem_failed;
1549
1550 SET_NETDEV_DEV(priv->dev, hca->dma_device);
1551 priv->dev->dev_id = port - 1;
1552
1553 if (!ib_query_port(hca, port, &attr))
1554 priv->max_ib_mtu = ib_mtu_enum_to_int(attr.max_mtu);
1555 else {
1556 printk(KERN_WARNING "%s: ib_query_port %d failed\n",
1557 hca->name, port);
1558 goto device_init_failed;
1559 }
1560
1561 /* MTU will be reset when mcast join happens */
1562 priv->dev->mtu = IPOIB_UD_MTU(priv->max_ib_mtu);
1563 priv->mcast_mtu = priv->admin_mtu = priv->dev->mtu;
1564
1565 priv->dev->neigh_priv_len = sizeof(struct ipoib_neigh);
1566
1567 result = ib_query_pkey(hca, port, 0, &priv->pkey);
1568 if (result) {
1569 printk(KERN_WARNING "%s: ib_query_pkey port %d failed (ret = %d)\n",
1570 hca->name, port, result);
1571 goto device_init_failed;
1572 }
1573
1574 if (ipoib_set_dev_features(priv, hca))
1575 goto device_init_failed;
1576
1577 /*
1578 * Set the full membership bit, so that we join the right
1579 * broadcast group, etc.
1580 */
1581 priv->pkey |= 0x8000;
1582
1583 priv->dev->broadcast[8] = priv->pkey >> 8;
1584 priv->dev->broadcast[9] = priv->pkey & 0xff;
1585
1586 result = ib_query_gid(hca, port, 0, &priv->local_gid);
1587 if (result) {
1588 printk(KERN_WARNING "%s: ib_query_gid port %d failed (ret = %d)\n",
1589 hca->name, port, result);
1590 goto device_init_failed;
1591 } else
1592 memcpy(priv->dev->dev_addr + 4, priv->local_gid.raw, sizeof (union ib_gid));
1593
1594 result = ipoib_dev_init(priv->dev, hca, port);
1595 if (result < 0) {
1596 printk(KERN_WARNING "%s: failed to initialize port %d (ret = %d)\n",
1597 hca->name, port, result);
1598 goto device_init_failed;
1599 }
1600
1601 INIT_IB_EVENT_HANDLER(&priv->event_handler,
1602 priv->ca, ipoib_event);
1603 result = ib_register_event_handler(&priv->event_handler);
1604 if (result < 0) {
1605 printk(KERN_WARNING "%s: ib_register_event_handler failed for "
1606 "port %d (ret = %d)\n",
1607 hca->name, port, result);
1608 goto event_failed;
1609 }
1610
1611 result = register_netdev(priv->dev);
1612 if (result) {
1613 printk(KERN_WARNING "%s: couldn't register ipoib port %d; error %d\n",
1614 hca->name, port, result);
1615 goto register_failed;
1616 }
1617
1618 ipoib_create_debug_files(priv->dev);
1619
1620 if (ipoib_cm_add_mode_attr(priv->dev))
1621 goto sysfs_failed;
1622 if (ipoib_add_pkey_attr(priv->dev))
1623 goto sysfs_failed;
1624 if (ipoib_add_umcast_attr(priv->dev))
1625 goto sysfs_failed;
1626 if (device_create_file(&priv->dev->dev, &dev_attr_create_child))
1627 goto sysfs_failed;
1628 if (device_create_file(&priv->dev->dev, &dev_attr_delete_child))
1629 goto sysfs_failed;
1630
1631 return priv->dev;
1632
1633 sysfs_failed:
1634 ipoib_delete_debug_files(priv->dev);
1635 unregister_netdev(priv->dev);
1636
1637 register_failed:
1638 ib_unregister_event_handler(&priv->event_handler);
1639 /* Stop GC if started before flush */
1640 set_bit(IPOIB_STOP_NEIGH_GC, &priv->flags);
1641 cancel_delayed_work(&priv->neigh_reap_task);
1642 flush_workqueue(ipoib_workqueue);
1643
1644 event_failed:
1645 ipoib_dev_cleanup(priv->dev);
1646
1647 device_init_failed:
1648 free_netdev(priv->dev);
1649
1650 alloc_mem_failed:
1651 return ERR_PTR(result);
1652 }
1653
1654 static void ipoib_add_one(struct ib_device *device)
1655 {
1656 struct list_head *dev_list;
1657 struct net_device *dev;
1658 struct ipoib_dev_priv *priv;
1659 int s, e, p;
1660
1661 if (rdma_node_get_transport(device->node_type) != RDMA_TRANSPORT_IB)
1662 return;
1663
1664 dev_list = kmalloc(sizeof *dev_list, GFP_KERNEL);
1665 if (!dev_list)
1666 return;
1667
1668 INIT_LIST_HEAD(dev_list);
1669
1670 if (device->node_type == RDMA_NODE_IB_SWITCH) {
1671 s = 0;
1672 e = 0;
1673 } else {
1674 s = 1;
1675 e = device->phys_port_cnt;
1676 }
1677
1678 for (p = s; p <= e; ++p) {
1679 if (rdma_port_get_link_layer(device, p) != IB_LINK_LAYER_INFINIBAND)
1680 continue;
1681 dev = ipoib_add_port("ib%d", device, p);
1682 if (!IS_ERR(dev)) {
1683 priv = netdev_priv(dev);
1684 list_add_tail(&priv->list, dev_list);
1685 }
1686 }
1687
1688 ib_set_client_data(device, &ipoib_client, dev_list);
1689 }
1690
1691 static void ipoib_remove_one(struct ib_device *device)
1692 {
1693 struct ipoib_dev_priv *priv, *tmp;
1694 struct list_head *dev_list;
1695
1696 if (rdma_node_get_transport(device->node_type) != RDMA_TRANSPORT_IB)
1697 return;
1698
1699 dev_list = ib_get_client_data(device, &ipoib_client);
1700 if (!dev_list)
1701 return;
1702
1703 list_for_each_entry_safe(priv, tmp, dev_list, list) {
1704 ib_unregister_event_handler(&priv->event_handler);
1705
1706 rtnl_lock();
1707 dev_change_flags(priv->dev, priv->dev->flags & ~IFF_UP);
1708 rtnl_unlock();
1709
1710 /* Stop GC */
1711 set_bit(IPOIB_STOP_NEIGH_GC, &priv->flags);
1712 cancel_delayed_work(&priv->neigh_reap_task);
1713 flush_workqueue(ipoib_workqueue);
1714
1715 unregister_netdev(priv->dev);
1716 free_netdev(priv->dev);
1717 }
1718
1719 kfree(dev_list);
1720 }
1721
1722 static int __init ipoib_init_module(void)
1723 {
1724 int ret;
1725
1726 ipoib_recvq_size = roundup_pow_of_two(ipoib_recvq_size);
1727 ipoib_recvq_size = min(ipoib_recvq_size, IPOIB_MAX_QUEUE_SIZE);
1728 ipoib_recvq_size = max(ipoib_recvq_size, IPOIB_MIN_QUEUE_SIZE);
1729
1730 ipoib_sendq_size = roundup_pow_of_two(ipoib_sendq_size);
1731 ipoib_sendq_size = min(ipoib_sendq_size, IPOIB_MAX_QUEUE_SIZE);
1732 ipoib_sendq_size = max3(ipoib_sendq_size, 2 * MAX_SEND_CQE, IPOIB_MIN_QUEUE_SIZE);
1733 #ifdef CONFIG_INFINIBAND_IPOIB_CM
1734 ipoib_max_conn_qp = min(ipoib_max_conn_qp, IPOIB_CM_MAX_CONN_QP);
1735 #endif
1736
1737 /*
1738 * When copying small received packets, we only copy from the
1739 * linear data part of the SKB, so we rely on this condition.
1740 */
1741 BUILD_BUG_ON(IPOIB_CM_COPYBREAK > IPOIB_CM_HEAD_SIZE);
1742
1743 ret = ipoib_register_debugfs();
1744 if (ret)
1745 return ret;
1746
1747 /*
1748 * We create our own workqueue mainly because we want to be
1749 * able to flush it when devices are being removed. We can't
1750 * use schedule_work()/flush_scheduled_work() because both
1751 * unregister_netdev() and linkwatch_event take the rtnl lock,
1752 * so flush_scheduled_work() can deadlock during device
1753 * removal.
1754 */
1755 ipoib_workqueue = create_singlethread_workqueue("ipoib");
1756 if (!ipoib_workqueue) {
1757 ret = -ENOMEM;
1758 goto err_fs;
1759 }
1760
1761 ib_sa_register_client(&ipoib_sa_client);
1762
1763 ret = ib_register_client(&ipoib_client);
1764 if (ret)
1765 goto err_sa;
1766
1767 ret = ipoib_netlink_init();
1768 if (ret)
1769 goto err_client;
1770
1771 return 0;
1772
1773 err_client:
1774 ib_unregister_client(&ipoib_client);
1775
1776 err_sa:
1777 ib_sa_unregister_client(&ipoib_sa_client);
1778 destroy_workqueue(ipoib_workqueue);
1779
1780 err_fs:
1781 ipoib_unregister_debugfs();
1782
1783 return ret;
1784 }
1785
1786 static void __exit ipoib_cleanup_module(void)
1787 {
1788 ipoib_netlink_fini();
1789 ib_unregister_client(&ipoib_client);
1790 ib_sa_unregister_client(&ipoib_sa_client);
1791 ipoib_unregister_debugfs();
1792 destroy_workqueue(ipoib_workqueue);
1793 }
1794
1795 module_init(ipoib_init_module);
1796 module_exit(ipoib_cleanup_module);
Linux with added FPC-III support