projects / deliverable / linux.git / blob
? search:


36536ce5a3e2f9d51278be970d51bb51ec07232c
[deliverable/linux.git] / drivers / infiniband / ulp / ipoib / ipoib_main.c
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 #include <net/addrconf.h>
52 #include <linux/inetdevice.h>
53 #include <rdma/ib_cache.h>
54
55 #define DRV_VERSION "1.0.0"
56
57 const char ipoib_driver_version[] = DRV_VERSION;
58
59 MODULE_AUTHOR("Roland Dreier");
60 MODULE_DESCRIPTION("IP-over-InfiniBand net driver");
61 MODULE_LICENSE("Dual BSD/GPL");
62 MODULE_VERSION(DRV_VERSION);
63
64 int ipoib_sendq_size __read_mostly = IPOIB_TX_RING_SIZE;
65 int ipoib_recvq_size __read_mostly = IPOIB_RX_RING_SIZE;
66
67 module_param_named(send_queue_size, ipoib_sendq_size, int, 0444);
68 MODULE_PARM_DESC(send_queue_size, "Number of descriptors in send queue");
69 module_param_named(recv_queue_size, ipoib_recvq_size, int, 0444);
70 MODULE_PARM_DESC(recv_queue_size, "Number of descriptors in receive queue");
71
72 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
73 int ipoib_debug_level;
74
75 module_param_named(debug_level, ipoib_debug_level, int, 0644);
76 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0");
77 #endif
78
79 struct ipoib_path_iter {
80 struct net_device *dev;
81 struct ipoib_path path;
82 };
83
84 static const u8 ipv4_bcast_addr[] = {
85 0x00, 0xff, 0xff, 0xff,
86 0xff, 0x12, 0x40, 0x1b, 0x00, 0x00, 0x00, 0x00,
87 0x00, 0x00, 0x00, 0x00, 0xff, 0xff, 0xff, 0xff
88 };
89
90 struct workqueue_struct *ipoib_workqueue;
91
92 struct ib_sa_client ipoib_sa_client;
93
94 static void ipoib_add_one(struct ib_device *device);
95 static void ipoib_remove_one(struct ib_device *device, void *client_data);
96 static void ipoib_neigh_reclaim(struct rcu_head *rp);
97 static struct net_device *ipoib_get_net_dev_by_params(
98 struct ib_device *dev, u8 port, u16 pkey,
99 const union ib_gid *gid, const struct sockaddr *addr,
100 void *client_data);
101
102 static struct ib_client ipoib_client = {
103 .name = "ipoib",
104 .add = ipoib_add_one,
105 .remove = ipoib_remove_one,
106 .get_net_dev_by_params = ipoib_get_net_dev_by_params,
107 };
108
109 int ipoib_open(struct net_device *dev)
110 {
111 struct ipoib_dev_priv *priv = netdev_priv(dev);
112
113 ipoib_dbg(priv, "bringing up interface\n");
114
115 netif_carrier_off(dev);
116
117 set_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags);
118
119 if (ipoib_ib_dev_open(dev)) {
120 if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags))
121 return 0;
122 goto err_disable;
123 }
124
125 if (ipoib_ib_dev_up(dev))
126 goto err_stop;
127
128 if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) {
129 struct ipoib_dev_priv *cpriv;
130
131 /* Bring up any child interfaces too */
132 down_read(&priv->vlan_rwsem);
133 list_for_each_entry(cpriv, &priv->child_intfs, list) {
134 int flags;
135
136 flags = cpriv->dev->flags;
137 if (flags & IFF_UP)
138 continue;
139
140 dev_change_flags(cpriv->dev, flags | IFF_UP);
141 }
142 up_read(&priv->vlan_rwsem);
143 }
144
145 netif_start_queue(dev);
146
147 return 0;
148
149 err_stop:
150 ipoib_ib_dev_stop(dev);
151
152 err_disable:
153 clear_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags);
154
155 return -EINVAL;
156 }
157
158 static int ipoib_stop(struct net_device *dev)
159 {
160 struct ipoib_dev_priv *priv = netdev_priv(dev);
161
162 ipoib_dbg(priv, "stopping interface\n");
163
164 clear_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags);
165
166 netif_stop_queue(dev);
167
168 ipoib_ib_dev_down(dev);
169 ipoib_ib_dev_stop(dev);
170
171 if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) {
172 struct ipoib_dev_priv *cpriv;
173
174 /* Bring down any child interfaces too */
175 down_read(&priv->vlan_rwsem);
176 list_for_each_entry(cpriv, &priv->child_intfs, list) {
177 int flags;
178
179 flags = cpriv->dev->flags;
180 if (!(flags & IFF_UP))
181 continue;
182
183 dev_change_flags(cpriv->dev, flags & ~IFF_UP);
184 }
185 up_read(&priv->vlan_rwsem);
186 }
187
188 return 0;
189 }
190
191 static void ipoib_uninit(struct net_device *dev)
192 {
193 ipoib_dev_cleanup(dev);
194 }
195
196 static netdev_features_t ipoib_fix_features(struct net_device *dev, netdev_features_t features)
197 {
198 struct ipoib_dev_priv *priv = netdev_priv(dev);
199
200 if (test_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags))
201 features &= ~(NETIF_F_IP_CSUM | NETIF_F_TSO);
202
203 return features;
204 }
205
206 static int ipoib_change_mtu(struct net_device *dev, int new_mtu)
207 {
208 struct ipoib_dev_priv *priv = netdev_priv(dev);
209
210 /* dev->mtu > 2K ==> connected mode */
211 if (ipoib_cm_admin_enabled(dev)) {
212 if (new_mtu > ipoib_cm_max_mtu(dev))
213 return -EINVAL;
214
215 if (new_mtu > priv->mcast_mtu)
216 ipoib_warn(priv, "mtu > %d will cause multicast packet drops.\n",
217 priv->mcast_mtu);
218
219 dev->mtu = new_mtu;
220 return 0;
221 }
222
223 if (new_mtu > IPOIB_UD_MTU(priv->max_ib_mtu))
224 return -EINVAL;
225
226 priv->admin_mtu = new_mtu;
227
228 dev->mtu = min(priv->mcast_mtu, priv->admin_mtu);
229
230 return 0;
231 }
232
233 /* Called with an RCU read lock taken */
234 static bool ipoib_is_dev_match_addr_rcu(const struct sockaddr *addr,
235 struct net_device *dev)
236 {
237 struct net *net = dev_net(dev);
238 struct in_device *in_dev;
239 struct sockaddr_in *addr_in = (struct sockaddr_in *)addr;
240 struct sockaddr_in6 *addr_in6 = (struct sockaddr_in6 *)addr;
241 __be32 ret_addr;
242
243 switch (addr->sa_family) {
244 case AF_INET:
245 in_dev = in_dev_get(dev);
246 if (!in_dev)
247 return false;
248
249 ret_addr = inet_confirm_addr(net, in_dev, 0,
250 addr_in->sin_addr.s_addr,
251 RT_SCOPE_HOST);
252 in_dev_put(in_dev);
253 if (ret_addr)
254 return true;
255
256 break;
257 case AF_INET6:
258 if (IS_ENABLED(CONFIG_IPV6) &&
259 ipv6_chk_addr(net, &addr_in6->sin6_addr, dev, 1))
260 return true;
261
262 break;
263 }
264 return false;
265 }
266
267 /**
268 * Find the master net_device on top of the given net_device.
269 * @dev: base IPoIB net_device
270 *
271 * Returns the master net_device with a reference held, or the same net_device
272 * if no master exists.
273 */
274 static struct net_device *ipoib_get_master_net_dev(struct net_device *dev)
275 {
276 struct net_device *master;
277
278 rcu_read_lock();
279 master = netdev_master_upper_dev_get_rcu(dev);
280 if (master)
281 dev_hold(master);
282 rcu_read_unlock();
283
284 if (master)
285 return master;
286
287 dev_hold(dev);
288 return dev;
289 }
290
291 /**
292 * Find a net_device matching the given address, which is an upper device of
293 * the given net_device.
294 * @addr: IP address to look for.
295 * @dev: base IPoIB net_device
296 *
297 * If found, returns the net_device with a reference held. Otherwise return
298 * NULL.
299 */
300 static struct net_device *ipoib_get_net_dev_match_addr(
301 const struct sockaddr *addr, struct net_device *dev)
302 {
303 struct net_device *upper,
304 *result = NULL;
305 struct list_head *iter;
306
307 rcu_read_lock();
308 if (ipoib_is_dev_match_addr_rcu(addr, dev)) {
309 dev_hold(dev);
310 result = dev;
311 goto out;
312 }
313
314 netdev_for_each_all_upper_dev_rcu(dev, upper, iter) {
315 if (ipoib_is_dev_match_addr_rcu(addr, upper)) {
316 dev_hold(upper);
317 result = upper;
318 break;
319 }
320 }
321 out:
322 rcu_read_unlock();
323 return result;
324 }
325
326 /* returns the number of IPoIB netdevs on top a given ipoib device matching a
327 * pkey_index and address, if one exists.
328 *
329 * @found_net_dev: contains a matching net_device if the return value >= 1,
330 * with a reference held. */
331 static int ipoib_match_gid_pkey_addr(struct ipoib_dev_priv *priv,
332 const union ib_gid *gid,
333 u16 pkey_index,
334 const struct sockaddr *addr,
335 int nesting,
336 struct net_device **found_net_dev)
337 {
338 struct ipoib_dev_priv *child_priv;
339 struct net_device *net_dev = NULL;
340 int matches = 0;
341
342 if (priv->pkey_index == pkey_index &&
343 (!gid || !memcmp(gid, &priv->local_gid, sizeof(*gid)))) {
344 if (!addr) {
345 net_dev = ipoib_get_master_net_dev(priv->dev);
346 } else {
347 /* Verify the net_device matches the IP address, as
348 * IPoIB child devices currently share a GID. */
349 net_dev = ipoib_get_net_dev_match_addr(addr, priv->dev);
350 }
351 if (net_dev) {
352 if (!*found_net_dev)
353 *found_net_dev = net_dev;
354 else
355 dev_put(net_dev);
356 ++matches;
357 }
358 }
359
360 /* Check child interfaces */
361 down_read_nested(&priv->vlan_rwsem, nesting);
362 list_for_each_entry(child_priv, &priv->child_intfs, list) {
363 matches += ipoib_match_gid_pkey_addr(child_priv, gid,
364 pkey_index, addr,
365 nesting + 1,
366 found_net_dev);
367 if (matches > 1)
368 break;
369 }
370 up_read(&priv->vlan_rwsem);
371
372 return matches;
373 }
374
375 /* Returns the number of matching net_devs found (between 0 and 2). Also
376 * return the matching net_device in the @net_dev parameter, holding a
377 * reference to the net_device, if the number of matches >= 1 */
378 static int __ipoib_get_net_dev_by_params(struct list_head *dev_list, u8 port,
379 u16 pkey_index,
380 const union ib_gid *gid,
381 const struct sockaddr *addr,
382 struct net_device **net_dev)
383 {
384 struct ipoib_dev_priv *priv;
385 int matches = 0;
386
387 *net_dev = NULL;
388
389 list_for_each_entry(priv, dev_list, list) {
390 if (priv->port != port)
391 continue;
392
393 matches += ipoib_match_gid_pkey_addr(priv, gid, pkey_index,
394 addr, 0, net_dev);
395 if (matches > 1)
396 break;
397 }
398
399 return matches;
400 }
401
402 static struct net_device *ipoib_get_net_dev_by_params(
403 struct ib_device *dev, u8 port, u16 pkey,
404 const union ib_gid *gid, const struct sockaddr *addr,
405 void *client_data)
406 {
407 struct net_device *net_dev;
408 struct list_head *dev_list = client_data;
409 u16 pkey_index;
410 int matches;
411 int ret;
412
413 if (!rdma_protocol_ib(dev, port))
414 return NULL;
415
416 ret = ib_find_cached_pkey(dev, port, pkey, &pkey_index);
417 if (ret)
418 return NULL;
419
420 if (!dev_list)
421 return NULL;
422
423 /* See if we can find a unique device matching the L2 parameters */
424 matches = __ipoib_get_net_dev_by_params(dev_list, port, pkey_index,
425 gid, NULL, &net_dev);
426
427 switch (matches) {
428 case 0:
429 return NULL;
430 case 1:
431 return net_dev;
432 }
433
434 dev_put(net_dev);
435
436 /* Couldn't find a unique device with L2 parameters only. Use L3
437 * address to uniquely match the net device */
438 matches = __ipoib_get_net_dev_by_params(dev_list, port, pkey_index,
439 gid, addr, &net_dev);
440 switch (matches) {
441 case 0:
442 return NULL;
443 default:
444 dev_warn_ratelimited(&dev->dev,
445 "duplicate IP address detected\n");
446 /* Fall through */
447 case 1:
448 return net_dev;
449 }
450 }
451
452 int ipoib_set_mode(struct net_device *dev, const char *buf)
453 {
454 struct ipoib_dev_priv *priv = netdev_priv(dev);
455
456 /* flush paths if we switch modes so that connections are restarted */
457 if (IPOIB_CM_SUPPORTED(dev->dev_addr) && !strcmp(buf, "connected\n")) {
458 set_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags);
459 ipoib_warn(priv, "enabling connected mode "
460 "will cause multicast packet drops\n");
461 netdev_update_features(dev);
462 dev_set_mtu(dev, ipoib_cm_max_mtu(dev));
463 rtnl_unlock();
464 priv->tx_wr.send_flags &= ~IB_SEND_IP_CSUM;
465
466 ipoib_flush_paths(dev);
467 rtnl_lock();
468 return 0;
469 }
470
471 if (!strcmp(buf, "datagram\n")) {
472 clear_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags);
473 netdev_update_features(dev);
474 dev_set_mtu(dev, min(priv->mcast_mtu, dev->mtu));
475 rtnl_unlock();
476 ipoib_flush_paths(dev);
477 rtnl_lock();
478 return 0;
479 }
480
481 return -EINVAL;
482 }
483
484 static struct ipoib_path *__path_find(struct net_device *dev, void *gid)
485 {
486 struct ipoib_dev_priv *priv = netdev_priv(dev);
487 struct rb_node *n = priv->path_tree.rb_node;
488 struct ipoib_path *path;
489 int ret;
490
491 while (n) {
492 path = rb_entry(n, struct ipoib_path, rb_node);
493
494 ret = memcmp(gid, path->pathrec.dgid.raw,
495 sizeof (union ib_gid));
496
497 if (ret < 0)
498 n = n->rb_left;
499 else if (ret > 0)
500 n = n->rb_right;
501 else
502 return path;
503 }
504
505 return NULL;
506 }
507
508 static int __path_add(struct net_device *dev, struct ipoib_path *path)
509 {
510 struct ipoib_dev_priv *priv = netdev_priv(dev);
511 struct rb_node **n = &priv->path_tree.rb_node;
512 struct rb_node *pn = NULL;
513 struct ipoib_path *tpath;
514 int ret;
515
516 while (*n) {
517 pn = *n;
518 tpath = rb_entry(pn, struct ipoib_path, rb_node);
519
520 ret = memcmp(path->pathrec.dgid.raw, tpath->pathrec.dgid.raw,
521 sizeof (union ib_gid));
522 if (ret < 0)
523 n = &pn->rb_left;
524 else if (ret > 0)
525 n = &pn->rb_right;
526 else
527 return -EEXIST;
528 }
529
530 rb_link_node(&path->rb_node, pn, n);
531 rb_insert_color(&path->rb_node, &priv->path_tree);
532
533 list_add_tail(&path->list, &priv->path_list);
534
535 return 0;
536 }
537
538 static void path_free(struct net_device *dev, struct ipoib_path *path)
539 {
540 struct sk_buff *skb;
541
542 while ((skb = __skb_dequeue(&path->queue)))
543 dev_kfree_skb_irq(skb);
544
545 ipoib_dbg(netdev_priv(dev), "path_free\n");
546
547 /* remove all neigh connected to this path */
548 ipoib_del_neighs_by_gid(dev, path->pathrec.dgid.raw);
549
550 if (path->ah)
551 ipoib_put_ah(path->ah);
552
553 kfree(path);
554 }
555
556 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
557
558 struct ipoib_path_iter *ipoib_path_iter_init(struct net_device *dev)
559 {
560 struct ipoib_path_iter *iter;
561
562 iter = kmalloc(sizeof *iter, GFP_KERNEL);
563 if (!iter)
564 return NULL;
565
566 iter->dev = dev;
567 memset(iter->path.pathrec.dgid.raw, 0, 16);
568
569 if (ipoib_path_iter_next(iter)) {
570 kfree(iter);
571 return NULL;
572 }
573
574 return iter;
575 }
576
577 int ipoib_path_iter_next(struct ipoib_path_iter *iter)
578 {
579 struct ipoib_dev_priv *priv = netdev_priv(iter->dev);
580 struct rb_node *n;
581 struct ipoib_path *path;
582 int ret = 1;
583
584 spin_lock_irq(&priv->lock);
585
586 n = rb_first(&priv->path_tree);
587
588 while (n) {
589 path = rb_entry(n, struct ipoib_path, rb_node);
590
591 if (memcmp(iter->path.pathrec.dgid.raw, path->pathrec.dgid.raw,
592 sizeof (union ib_gid)) < 0) {
593 iter->path = *path;
594 ret = 0;
595 break;
596 }
597
598 n = rb_next(n);
599 }
600
601 spin_unlock_irq(&priv->lock);
602
603 return ret;
604 }
605
606 void ipoib_path_iter_read(struct ipoib_path_iter *iter,
607 struct ipoib_path *path)
608 {
609 *path = iter->path;
610 }
611
612 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */
613
614 void ipoib_mark_paths_invalid(struct net_device *dev)
615 {
616 struct ipoib_dev_priv *priv = netdev_priv(dev);
617 struct ipoib_path *path, *tp;
618
619 spin_lock_irq(&priv->lock);
620
621 list_for_each_entry_safe(path, tp, &priv->path_list, list) {
622 ipoib_dbg(priv, "mark path LID 0x%04x GID %pI6 invalid\n",
623 be16_to_cpu(path->pathrec.dlid),
624 path->pathrec.dgid.raw);
625 path->valid = 0;
626 }
627
628 spin_unlock_irq(&priv->lock);
629 }
630
631 void ipoib_flush_paths(struct net_device *dev)
632 {
633 struct ipoib_dev_priv *priv = netdev_priv(dev);
634 struct ipoib_path *path, *tp;
635 LIST_HEAD(remove_list);
636 unsigned long flags;
637
638 netif_tx_lock_bh(dev);
639 spin_lock_irqsave(&priv->lock, flags);
640
641 list_splice_init(&priv->path_list, &remove_list);
642
643 list_for_each_entry(path, &remove_list, list)
644 rb_erase(&path->rb_node, &priv->path_tree);
645
646 list_for_each_entry_safe(path, tp, &remove_list, list) {
647 if (path->query)
648 ib_sa_cancel_query(path->query_id, path->query);
649 spin_unlock_irqrestore(&priv->lock, flags);
650 netif_tx_unlock_bh(dev);
651 wait_for_completion(&path->done);
652 path_free(dev, path);
653 netif_tx_lock_bh(dev);
654 spin_lock_irqsave(&priv->lock, flags);
655 }
656
657 spin_unlock_irqrestore(&priv->lock, flags);
658 netif_tx_unlock_bh(dev);
659 }
660
661 static void path_rec_completion(int status,
662 struct ib_sa_path_rec *pathrec,
663 void *path_ptr)
664 {
665 struct ipoib_path *path = path_ptr;
666 struct net_device *dev = path->dev;
667 struct ipoib_dev_priv *priv = netdev_priv(dev);
668 struct ipoib_ah *ah = NULL;
669 struct ipoib_ah *old_ah = NULL;
670 struct ipoib_neigh *neigh, *tn;
671 struct sk_buff_head skqueue;
672 struct sk_buff *skb;
673 unsigned long flags;
674
675 if (!status)
676 ipoib_dbg(priv, "PathRec LID 0x%04x for GID %pI6\n",
677 be16_to_cpu(pathrec->dlid), pathrec->dgid.raw);
678 else
679 ipoib_dbg(priv, "PathRec status %d for GID %pI6\n",
680 status, path->pathrec.dgid.raw);
681
682 skb_queue_head_init(&skqueue);
683
684 if (!status) {
685 struct ib_ah_attr av;
686
687 if (!ib_init_ah_from_path(priv->ca, priv->port, pathrec, &av))
688 ah = ipoib_create_ah(dev, priv->pd, &av);
689 }
690
691 spin_lock_irqsave(&priv->lock, flags);
692
693 if (!IS_ERR_OR_NULL(ah)) {
694 path->pathrec = *pathrec;
695
696 old_ah = path->ah;
697 path->ah = ah;
698
699 ipoib_dbg(priv, "created address handle %p for LID 0x%04x, SL %d\n",
700 ah, be16_to_cpu(pathrec->dlid), pathrec->sl);
701
702 while ((skb = __skb_dequeue(&path->queue)))
703 __skb_queue_tail(&skqueue, skb);
704
705 list_for_each_entry_safe(neigh, tn, &path->neigh_list, list) {
706 if (neigh->ah) {
707 WARN_ON(neigh->ah != old_ah);
708 /*
709 * Dropping the ah reference inside
710 * priv->lock is safe here, because we
711 * will hold one more reference from
712 * the original value of path->ah (ie
713 * old_ah).
714 */
715 ipoib_put_ah(neigh->ah);
716 }
717 kref_get(&path->ah->ref);
718 neigh->ah = path->ah;
719
720 if (ipoib_cm_enabled(dev, neigh->daddr)) {
721 if (!ipoib_cm_get(neigh))
722 ipoib_cm_set(neigh, ipoib_cm_create_tx(dev,
723 path,
724 neigh));
725 if (!ipoib_cm_get(neigh)) {
726 ipoib_neigh_free(neigh);
727 continue;
728 }
729 }
730
731 while ((skb = __skb_dequeue(&neigh->queue)))
732 __skb_queue_tail(&skqueue, skb);
733 }
734 path->valid = 1;
735 }
736
737 path->query = NULL;
738 complete(&path->done);
739
740 spin_unlock_irqrestore(&priv->lock, flags);
741
742 if (IS_ERR_OR_NULL(ah))
743 ipoib_del_neighs_by_gid(dev, path->pathrec.dgid.raw);
744
745 if (old_ah)
746 ipoib_put_ah(old_ah);
747
748 while ((skb = __skb_dequeue(&skqueue))) {
749 skb->dev = dev;
750 if (dev_queue_xmit(skb))
751 ipoib_warn(priv, "dev_queue_xmit failed "
752 "to requeue packet\n");
753 }
754 }
755
756 static struct ipoib_path *path_rec_create(struct net_device *dev, void *gid)
757 {
758 struct ipoib_dev_priv *priv = netdev_priv(dev);
759 struct ipoib_path *path;
760
761 if (!priv->broadcast)
762 return NULL;
763
764 path = kzalloc(sizeof *path, GFP_ATOMIC);
765 if (!path)
766 return NULL;
767
768 path->dev = dev;
769
770 skb_queue_head_init(&path->queue);
771
772 INIT_LIST_HEAD(&path->neigh_list);
773
774 memcpy(path->pathrec.dgid.raw, gid, sizeof (union ib_gid));
775 path->pathrec.sgid = priv->local_gid;
776 path->pathrec.pkey = cpu_to_be16(priv->pkey);
777 path->pathrec.numb_path = 1;
778 path->pathrec.traffic_class = priv->broadcast->mcmember.traffic_class;
779
780 return path;
781 }
782
783 static int path_rec_start(struct net_device *dev,
784 struct ipoib_path *path)
785 {
786 struct ipoib_dev_priv *priv = netdev_priv(dev);
787
788 ipoib_dbg(priv, "Start path record lookup for %pI6\n",
789 path->pathrec.dgid.raw);
790
791 init_completion(&path->done);
792
793 path->query_id =
794 ib_sa_path_rec_get(&ipoib_sa_client, priv->ca, priv->port,
795 &path->pathrec,
796 IB_SA_PATH_REC_DGID |
797 IB_SA_PATH_REC_SGID |
798 IB_SA_PATH_REC_NUMB_PATH |
799 IB_SA_PATH_REC_TRAFFIC_CLASS |
800 IB_SA_PATH_REC_PKEY,
801 1000, GFP_ATOMIC,
802 path_rec_completion,
803 path, &path->query);
804 if (path->query_id < 0) {
805 ipoib_warn(priv, "ib_sa_path_rec_get failed: %d\n", path->query_id);
806 path->query = NULL;
807 complete(&path->done);
808 return path->query_id;
809 }
810
811 return 0;
812 }
813
814 static void neigh_add_path(struct sk_buff *skb, u8 *daddr,
815 struct net_device *dev)
816 {
817 struct ipoib_dev_priv *priv = netdev_priv(dev);
818 struct ipoib_path *path;
819 struct ipoib_neigh *neigh;
820 unsigned long flags;
821
822 spin_lock_irqsave(&priv->lock, flags);
823 neigh = ipoib_neigh_alloc(daddr, dev);
824 if (!neigh) {
825 spin_unlock_irqrestore(&priv->lock, flags);
826 ++dev->stats.tx_dropped;
827 dev_kfree_skb_any(skb);
828 return;
829 }
830
831 path = __path_find(dev, daddr + 4);
832 if (!path) {
833 path = path_rec_create(dev, daddr + 4);
834 if (!path)
835 goto err_path;
836
837 __path_add(dev, path);
838 }
839
840 list_add_tail(&neigh->list, &path->neigh_list);
841
842 if (path->ah) {
843 kref_get(&path->ah->ref);
844 neigh->ah = path->ah;
845
846 if (ipoib_cm_enabled(dev, neigh->daddr)) {
847 if (!ipoib_cm_get(neigh))
848 ipoib_cm_set(neigh, ipoib_cm_create_tx(dev, path, neigh));
849 if (!ipoib_cm_get(neigh)) {
850 ipoib_neigh_free(neigh);
851 goto err_drop;
852 }
853 if (skb_queue_len(&neigh->queue) < IPOIB_MAX_PATH_REC_QUEUE)
854 __skb_queue_tail(&neigh->queue, skb);
855 else {
856 ipoib_warn(priv, "queue length limit %d. Packet drop.\n",
857 skb_queue_len(&neigh->queue));
858 goto err_drop;
859 }
860 } else {
861 spin_unlock_irqrestore(&priv->lock, flags);
862 ipoib_send(dev, skb, path->ah, IPOIB_QPN(daddr));
863 ipoib_neigh_put(neigh);
864 return;
865 }
866 } else {
867 neigh->ah = NULL;
868
869 if (!path->query && path_rec_start(dev, path))
870 goto err_path;
871 if (skb_queue_len(&neigh->queue) < IPOIB_MAX_PATH_REC_QUEUE)
872 __skb_queue_tail(&neigh->queue, skb);
873 else
874 goto err_drop;
875 }
876
877 spin_unlock_irqrestore(&priv->lock, flags);
878 ipoib_neigh_put(neigh);
879 return;
880
881 err_path:
882 ipoib_neigh_free(neigh);
883 err_drop:
884 ++dev->stats.tx_dropped;
885 dev_kfree_skb_any(skb);
886
887 spin_unlock_irqrestore(&priv->lock, flags);
888 ipoib_neigh_put(neigh);
889 }
890
891 static void unicast_arp_send(struct sk_buff *skb, struct net_device *dev,
892 struct ipoib_cb *cb)
893 {
894 struct ipoib_dev_priv *priv = netdev_priv(dev);
895 struct ipoib_path *path;
896 unsigned long flags;
897
898 spin_lock_irqsave(&priv->lock, flags);
899
900 path = __path_find(dev, cb->hwaddr + 4);
901 if (!path || !path->valid) {
902 int new_path = 0;
903
904 if (!path) {
905 path = path_rec_create(dev, cb->hwaddr + 4);
906 new_path = 1;
907 }
908 if (path) {
909 if (skb_queue_len(&path->queue) < IPOIB_MAX_PATH_REC_QUEUE) {
910 __skb_queue_tail(&path->queue, skb);
911 } else {
912 ++dev->stats.tx_dropped;
913 dev_kfree_skb_any(skb);
914 }
915
916 if (!path->query && path_rec_start(dev, path)) {
917 spin_unlock_irqrestore(&priv->lock, flags);
918 if (new_path)
919 path_free(dev, path);
920 return;
921 } else
922 __path_add(dev, path);
923 } else {
924 ++dev->stats.tx_dropped;
925 dev_kfree_skb_any(skb);
926 }
927
928 spin_unlock_irqrestore(&priv->lock, flags);
929 return;
930 }
931
932 if (path->ah) {
933 ipoib_dbg(priv, "Send unicast ARP to %04x\n",
934 be16_to_cpu(path->pathrec.dlid));
935
936 spin_unlock_irqrestore(&priv->lock, flags);
937 ipoib_send(dev, skb, path->ah, IPOIB_QPN(cb->hwaddr));
938 return;
939 } else if ((path->query || !path_rec_start(dev, path)) &&
940 skb_queue_len(&path->queue) < IPOIB_MAX_PATH_REC_QUEUE) {
941 __skb_queue_tail(&path->queue, skb);
942 } else {
943 ++dev->stats.tx_dropped;
944 dev_kfree_skb_any(skb);
945 }
946
947 spin_unlock_irqrestore(&priv->lock, flags);
948 }
949
950 static int ipoib_start_xmit(struct sk_buff *skb, struct net_device *dev)
951 {
952 struct ipoib_dev_priv *priv = netdev_priv(dev);
953 struct ipoib_neigh *neigh;
954 struct ipoib_cb *cb = ipoib_skb_cb(skb);
955 struct ipoib_header *header;
956 unsigned long flags;
957
958 header = (struct ipoib_header *) skb->data;
959
960 if (unlikely(cb->hwaddr[4] == 0xff)) {
961 /* multicast, arrange "if" according to probability */
962 if ((header->proto != htons(ETH_P_IP)) &&
963 (header->proto != htons(ETH_P_IPV6)) &&
964 (header->proto != htons(ETH_P_ARP)) &&
965 (header->proto != htons(ETH_P_RARP)) &&
966 (header->proto != htons(ETH_P_TIPC))) {
967 /* ethertype not supported by IPoIB */
968 ++dev->stats.tx_dropped;
969 dev_kfree_skb_any(skb);
970 return NETDEV_TX_OK;
971 }
972 /* Add in the P_Key for multicast*/
973 cb->hwaddr[8] = (priv->pkey >> 8) & 0xff;
974 cb->hwaddr[9] = priv->pkey & 0xff;
975
976 neigh = ipoib_neigh_get(dev, cb->hwaddr);
977 if (likely(neigh))
978 goto send_using_neigh;
979 ipoib_mcast_send(dev, cb->hwaddr, skb);
980 return NETDEV_TX_OK;
981 }
982
983 /* unicast, arrange "switch" according to probability */
984 switch (header->proto) {
985 case htons(ETH_P_IP):
986 case htons(ETH_P_IPV6):
987 case htons(ETH_P_TIPC):
988 neigh = ipoib_neigh_get(dev, cb->hwaddr);
989 if (unlikely(!neigh)) {
990 neigh_add_path(skb, cb->hwaddr, dev);
991 return NETDEV_TX_OK;
992 }
993 break;
994 case htons(ETH_P_ARP):
995 case htons(ETH_P_RARP):
996 /* for unicast ARP and RARP should always perform path find */
997 unicast_arp_send(skb, dev, cb);
998 return NETDEV_TX_OK;
999 default:
1000 /* ethertype not supported by IPoIB */
1001 ++dev->stats.tx_dropped;
1002 dev_kfree_skb_any(skb);
1003 return NETDEV_TX_OK;
1004 }
1005
1006 send_using_neigh:
1007 /* note we now hold a ref to neigh */
1008 if (ipoib_cm_get(neigh)) {
1009 if (ipoib_cm_up(neigh)) {
1010 ipoib_cm_send(dev, skb, ipoib_cm_get(neigh));
1011 goto unref;
1012 }
1013 } else if (neigh->ah) {
1014 ipoib_send(dev, skb, neigh->ah, IPOIB_QPN(cb->hwaddr));
1015 goto unref;
1016 }
1017
1018 if (skb_queue_len(&neigh->queue) < IPOIB_MAX_PATH_REC_QUEUE) {
1019 spin_lock_irqsave(&priv->lock, flags);
1020 __skb_queue_tail(&neigh->queue, skb);
1021 spin_unlock_irqrestore(&priv->lock, flags);
1022 } else {
1023 ++dev->stats.tx_dropped;
1024 dev_kfree_skb_any(skb);
1025 }
1026
1027 unref:
1028 ipoib_neigh_put(neigh);
1029
1030 return NETDEV_TX_OK;
1031 }
1032
1033 static void ipoib_timeout(struct net_device *dev)
1034 {
1035 struct ipoib_dev_priv *priv = netdev_priv(dev);
1036
1037 ipoib_warn(priv, "transmit timeout: latency %d msecs\n",
1038 jiffies_to_msecs(jiffies - dev->trans_start));
1039 ipoib_warn(priv, "queue stopped %d, tx_head %u, tx_tail %u\n",
1040 netif_queue_stopped(dev),
1041 priv->tx_head, priv->tx_tail);
1042 /* XXX reset QP, etc. */
1043 }
1044
1045 static int ipoib_hard_header(struct sk_buff *skb,
1046 struct net_device *dev,
1047 unsigned short type,
1048 const void *daddr, const void *saddr, unsigned len)
1049 {
1050 struct ipoib_header *header;
1051 struct ipoib_cb *cb = ipoib_skb_cb(skb);
1052
1053 header = (struct ipoib_header *) skb_push(skb, sizeof *header);
1054
1055 header->proto = htons(type);
1056 header->reserved = 0;
1057
1058 /*
1059 * we don't rely on dst_entry structure, always stuff the
1060 * destination address into skb->cb so we can figure out where
1061 * to send the packet later.
1062 */
1063 memcpy(cb->hwaddr, daddr, INFINIBAND_ALEN);
1064
1065 return sizeof *header;
1066 }
1067
1068 static void ipoib_set_mcast_list(struct net_device *dev)
1069 {
1070 struct ipoib_dev_priv *priv = netdev_priv(dev);
1071
1072 if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags)) {
1073 ipoib_dbg(priv, "IPOIB_FLAG_OPER_UP not set");
1074 return;
1075 }
1076
1077 queue_work(priv->wq, &priv->restart_task);
1078 }
1079
1080 static int ipoib_get_iflink(const struct net_device *dev)
1081 {
1082 struct ipoib_dev_priv *priv = netdev_priv(dev);
1083
1084 /* parent interface */
1085 if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags))
1086 return dev->ifindex;
1087
1088 /* child/vlan interface */
1089 return priv->parent->ifindex;
1090 }
1091
1092 static u32 ipoib_addr_hash(struct ipoib_neigh_hash *htbl, u8 *daddr)
1093 {
1094 /*
1095 * Use only the address parts that contributes to spreading
1096 * The subnet prefix is not used as one can not connect to
1097 * same remote port (GUID) using the same remote QPN via two
1098 * different subnets.
1099 */
1100 /* qpn octets[1:4) & port GUID octets[12:20) */
1101 u32 *d32 = (u32 *) daddr;
1102 u32 hv;
1103
1104 hv = jhash_3words(d32[3], d32[4], IPOIB_QPN_MASK & d32[0], 0);
1105 return hv & htbl->mask;
1106 }
1107
1108 struct ipoib_neigh *ipoib_neigh_get(struct net_device *dev, u8 *daddr)
1109 {
1110 struct ipoib_dev_priv *priv = netdev_priv(dev);
1111 struct ipoib_neigh_table *ntbl = &priv->ntbl;
1112 struct ipoib_neigh_hash *htbl;
1113 struct ipoib_neigh *neigh = NULL;
1114 u32 hash_val;
1115
1116 rcu_read_lock_bh();
1117
1118 htbl = rcu_dereference_bh(ntbl->htbl);
1119
1120 if (!htbl)
1121 goto out_unlock;
1122
1123 hash_val = ipoib_addr_hash(htbl, daddr);
1124 for (neigh = rcu_dereference_bh(htbl->buckets[hash_val]);
1125 neigh != NULL;
1126 neigh = rcu_dereference_bh(neigh->hnext)) {
1127 if (memcmp(daddr, neigh->daddr, INFINIBAND_ALEN) == 0) {
1128 /* found, take one ref on behalf of the caller */
1129 if (!atomic_inc_not_zero(&neigh->refcnt)) {
1130 /* deleted */
1131 neigh = NULL;
1132 goto out_unlock;
1133 }
1134 neigh->alive = jiffies;
1135 goto out_unlock;
1136 }
1137 }
1138
1139 out_unlock:
1140 rcu_read_unlock_bh();
1141 return neigh;
1142 }
1143
1144 static void __ipoib_reap_neigh(struct ipoib_dev_priv *priv)
1145 {
1146 struct ipoib_neigh_table *ntbl = &priv->ntbl;
1147 struct ipoib_neigh_hash *htbl;
1148 unsigned long neigh_obsolete;
1149 unsigned long dt;
1150 unsigned long flags;
1151 int i;
1152
1153 if (test_bit(IPOIB_STOP_NEIGH_GC, &priv->flags))
1154 return;
1155
1156 spin_lock_irqsave(&priv->lock, flags);
1157
1158 htbl = rcu_dereference_protected(ntbl->htbl,
1159 lockdep_is_held(&priv->lock));
1160
1161 if (!htbl)
1162 goto out_unlock;
1163
1164 /* neigh is obsolete if it was idle for two GC periods */
1165 dt = 2 * arp_tbl.gc_interval;
1166 neigh_obsolete = jiffies - dt;
1167 /* handle possible race condition */
1168 if (test_bit(IPOIB_STOP_NEIGH_GC, &priv->flags))
1169 goto out_unlock;
1170
1171 for (i = 0; i < htbl->size; i++) {
1172 struct ipoib_neigh *neigh;
1173 struct ipoib_neigh __rcu **np = &htbl->buckets[i];
1174
1175 while ((neigh = rcu_dereference_protected(*np,
1176 lockdep_is_held(&priv->lock))) != NULL) {
1177 /* was the neigh idle for two GC periods */
1178 if (time_after(neigh_obsolete, neigh->alive)) {
1179 rcu_assign_pointer(*np,
1180 rcu_dereference_protected(neigh->hnext,
1181 lockdep_is_held(&priv->lock)));
1182 /* remove from path/mc list */
1183 list_del(&neigh->list);
1184 call_rcu(&neigh->rcu, ipoib_neigh_reclaim);
1185 } else {
1186 np = &neigh->hnext;
1187 }
1188
1189 }
1190 }
1191
1192 out_unlock:
1193 spin_unlock_irqrestore(&priv->lock, flags);
1194 }
1195
1196 static void ipoib_reap_neigh(struct work_struct *work)
1197 {
1198 struct ipoib_dev_priv *priv =
1199 container_of(work, struct ipoib_dev_priv, neigh_reap_task.work);
1200
1201 __ipoib_reap_neigh(priv);
1202
1203 if (!test_bit(IPOIB_STOP_NEIGH_GC, &priv->flags))
1204 queue_delayed_work(priv->wq, &priv->neigh_reap_task,
1205 arp_tbl.gc_interval);
1206 }
1207
1208
1209 static struct ipoib_neigh *ipoib_neigh_ctor(u8 *daddr,
1210 struct net_device *dev)
1211 {
1212 struct ipoib_neigh *neigh;
1213
1214 neigh = kzalloc(sizeof *neigh, GFP_ATOMIC);
1215 if (!neigh)
1216 return NULL;
1217
1218 neigh->dev = dev;
1219 memcpy(&neigh->daddr, daddr, sizeof(neigh->daddr));
1220 skb_queue_head_init(&neigh->queue);
1221 INIT_LIST_HEAD(&neigh->list);
1222 ipoib_cm_set(neigh, NULL);
1223 /* one ref on behalf of the caller */
1224 atomic_set(&neigh->refcnt, 1);
1225
1226 return neigh;
1227 }
1228
1229 struct ipoib_neigh *ipoib_neigh_alloc(u8 *daddr,
1230 struct net_device *dev)
1231 {
1232 struct ipoib_dev_priv *priv = netdev_priv(dev);
1233 struct ipoib_neigh_table *ntbl = &priv->ntbl;
1234 struct ipoib_neigh_hash *htbl;
1235 struct ipoib_neigh *neigh;
1236 u32 hash_val;
1237
1238 htbl = rcu_dereference_protected(ntbl->htbl,
1239 lockdep_is_held(&priv->lock));
1240 if (!htbl) {
1241 neigh = NULL;
1242 goto out_unlock;
1243 }
1244
1245 /* need to add a new neigh, but maybe some other thread succeeded?
1246 * recalc hash, maybe hash resize took place so we do a search
1247 */
1248 hash_val = ipoib_addr_hash(htbl, daddr);
1249 for (neigh = rcu_dereference_protected(htbl->buckets[hash_val],
1250 lockdep_is_held(&priv->lock));
1251 neigh != NULL;
1252 neigh = rcu_dereference_protected(neigh->hnext,
1253 lockdep_is_held(&priv->lock))) {
1254 if (memcmp(daddr, neigh->daddr, INFINIBAND_ALEN) == 0) {
1255 /* found, take one ref on behalf of the caller */
1256 if (!atomic_inc_not_zero(&neigh->refcnt)) {
1257 /* deleted */
1258 neigh = NULL;
1259 break;
1260 }
1261 neigh->alive = jiffies;
1262 goto out_unlock;
1263 }
1264 }
1265
1266 neigh = ipoib_neigh_ctor(daddr, dev);
1267 if (!neigh)
1268 goto out_unlock;
1269
1270 /* one ref on behalf of the hash table */
1271 atomic_inc(&neigh->refcnt);
1272 neigh->alive = jiffies;
1273 /* put in hash */
1274 rcu_assign_pointer(neigh->hnext,
1275 rcu_dereference_protected(htbl->buckets[hash_val],
1276 lockdep_is_held(&priv->lock)));
1277 rcu_assign_pointer(htbl->buckets[hash_val], neigh);
1278 atomic_inc(&ntbl->entries);
1279
1280 out_unlock:
1281
1282 return neigh;
1283 }
1284
1285 void ipoib_neigh_dtor(struct ipoib_neigh *neigh)
1286 {
1287 /* neigh reference count was dropprd to zero */
1288 struct net_device *dev = neigh->dev;
1289 struct ipoib_dev_priv *priv = netdev_priv(dev);
1290 struct sk_buff *skb;
1291 if (neigh->ah)
1292 ipoib_put_ah(neigh->ah);
1293 while ((skb = __skb_dequeue(&neigh->queue))) {
1294 ++dev->stats.tx_dropped;
1295 dev_kfree_skb_any(skb);
1296 }
1297 if (ipoib_cm_get(neigh))
1298 ipoib_cm_destroy_tx(ipoib_cm_get(neigh));
1299 ipoib_dbg(netdev_priv(dev),
1300 "neigh free for %06x %pI6\n",
1301 IPOIB_QPN(neigh->daddr),
1302 neigh->daddr + 4);
1303 kfree(neigh);
1304 if (atomic_dec_and_test(&priv->ntbl.entries)) {
1305 if (test_bit(IPOIB_NEIGH_TBL_FLUSH, &priv->flags))
1306 complete(&priv->ntbl.flushed);
1307 }
1308 }
1309
1310 static void ipoib_neigh_reclaim(struct rcu_head *rp)
1311 {
1312 /* Called as a result of removal from hash table */
1313 struct ipoib_neigh *neigh = container_of(rp, struct ipoib_neigh, rcu);
1314 /* note TX context may hold another ref */
1315 ipoib_neigh_put(neigh);
1316 }
1317
1318 void ipoib_neigh_free(struct ipoib_neigh *neigh)
1319 {
1320 struct net_device *dev = neigh->dev;
1321 struct ipoib_dev_priv *priv = netdev_priv(dev);
1322 struct ipoib_neigh_table *ntbl = &priv->ntbl;
1323 struct ipoib_neigh_hash *htbl;
1324 struct ipoib_neigh __rcu **np;
1325 struct ipoib_neigh *n;
1326 u32 hash_val;
1327
1328 htbl = rcu_dereference_protected(ntbl->htbl,
1329 lockdep_is_held(&priv->lock));
1330 if (!htbl)
1331 return;
1332
1333 hash_val = ipoib_addr_hash(htbl, neigh->daddr);
1334 np = &htbl->buckets[hash_val];
1335 for (n = rcu_dereference_protected(*np,
1336 lockdep_is_held(&priv->lock));
1337 n != NULL;
1338 n = rcu_dereference_protected(*np,
1339 lockdep_is_held(&priv->lock))) {
1340 if (n == neigh) {
1341 /* found */
1342 rcu_assign_pointer(*np,
1343 rcu_dereference_protected(neigh->hnext,
1344 lockdep_is_held(&priv->lock)));
1345 /* remove from parent list */
1346 list_del(&neigh->list);
1347 call_rcu(&neigh->rcu, ipoib_neigh_reclaim);
1348 return;
1349 } else {
1350 np = &n->hnext;
1351 }
1352 }
1353 }
1354
1355 static int ipoib_neigh_hash_init(struct ipoib_dev_priv *priv)
1356 {
1357 struct ipoib_neigh_table *ntbl = &priv->ntbl;
1358 struct ipoib_neigh_hash *htbl;
1359 struct ipoib_neigh __rcu **buckets;
1360 u32 size;
1361
1362 clear_bit(IPOIB_NEIGH_TBL_FLUSH, &priv->flags);
1363 ntbl->htbl = NULL;
1364 htbl = kzalloc(sizeof(*htbl), GFP_KERNEL);
1365 if (!htbl)
1366 return -ENOMEM;
1367 set_bit(IPOIB_STOP_NEIGH_GC, &priv->flags);
1368 size = roundup_pow_of_two(arp_tbl.gc_thresh3);
1369 buckets = kzalloc(size * sizeof(*buckets), GFP_KERNEL);
1370 if (!buckets) {
1371 kfree(htbl);
1372 return -ENOMEM;
1373 }
1374 htbl->size = size;
1375 htbl->mask = (size - 1);
1376 htbl->buckets = buckets;
1377 RCU_INIT_POINTER(ntbl->htbl, htbl);
1378 htbl->ntbl = ntbl;
1379 atomic_set(&ntbl->entries, 0);
1380
1381 /* start garbage collection */
1382 clear_bit(IPOIB_STOP_NEIGH_GC, &priv->flags);
1383 queue_delayed_work(priv->wq, &priv->neigh_reap_task,
1384 arp_tbl.gc_interval);
1385
1386 return 0;
1387 }
1388
1389 static void neigh_hash_free_rcu(struct rcu_head *head)
1390 {
1391 struct ipoib_neigh_hash *htbl = container_of(head,
1392 struct ipoib_neigh_hash,
1393 rcu);
1394 struct ipoib_neigh __rcu **buckets = htbl->buckets;
1395 struct ipoib_neigh_table *ntbl = htbl->ntbl;
1396
1397 kfree(buckets);
1398 kfree(htbl);
1399 complete(&ntbl->deleted);
1400 }
1401
1402 void ipoib_del_neighs_by_gid(struct net_device *dev, u8 *gid)
1403 {
1404 struct ipoib_dev_priv *priv = netdev_priv(dev);
1405 struct ipoib_neigh_table *ntbl = &priv->ntbl;
1406 struct ipoib_neigh_hash *htbl;
1407 unsigned long flags;
1408 int i;
1409
1410 /* remove all neigh connected to a given path or mcast */
1411 spin_lock_irqsave(&priv->lock, flags);
1412
1413 htbl = rcu_dereference_protected(ntbl->htbl,
1414 lockdep_is_held(&priv->lock));
1415
1416 if (!htbl)
1417 goto out_unlock;
1418
1419 for (i = 0; i < htbl->size; i++) {
1420 struct ipoib_neigh *neigh;
1421 struct ipoib_neigh __rcu **np = &htbl->buckets[i];
1422
1423 while ((neigh = rcu_dereference_protected(*np,
1424 lockdep_is_held(&priv->lock))) != NULL) {
1425 /* delete neighs belong to this parent */
1426 if (!memcmp(gid, neigh->daddr + 4, sizeof (union ib_gid))) {
1427 rcu_assign_pointer(*np,
1428 rcu_dereference_protected(neigh->hnext,
1429 lockdep_is_held(&priv->lock)));
1430 /* remove from parent list */
1431 list_del(&neigh->list);
1432 call_rcu(&neigh->rcu, ipoib_neigh_reclaim);
1433 } else {
1434 np = &neigh->hnext;
1435 }
1436
1437 }
1438 }
1439 out_unlock:
1440 spin_unlock_irqrestore(&priv->lock, flags);
1441 }
1442
1443 static void ipoib_flush_neighs(struct ipoib_dev_priv *priv)
1444 {
1445 struct ipoib_neigh_table *ntbl = &priv->ntbl;
1446 struct ipoib_neigh_hash *htbl;
1447 unsigned long flags;
1448 int i, wait_flushed = 0;
1449
1450 init_completion(&priv->ntbl.flushed);
1451
1452 spin_lock_irqsave(&priv->lock, flags);
1453
1454 htbl = rcu_dereference_protected(ntbl->htbl,
1455 lockdep_is_held(&priv->lock));
1456 if (!htbl)
1457 goto out_unlock;
1458
1459 wait_flushed = atomic_read(&priv->ntbl.entries);
1460 if (!wait_flushed)
1461 goto free_htbl;
1462
1463 for (i = 0; i < htbl->size; i++) {
1464 struct ipoib_neigh *neigh;
1465 struct ipoib_neigh __rcu **np = &htbl->buckets[i];
1466
1467 while ((neigh = rcu_dereference_protected(*np,
1468 lockdep_is_held(&priv->lock))) != NULL) {
1469 rcu_assign_pointer(*np,
1470 rcu_dereference_protected(neigh->hnext,
1471 lockdep_is_held(&priv->lock)));
1472 /* remove from path/mc list */
1473 list_del(&neigh->list);
1474 call_rcu(&neigh->rcu, ipoib_neigh_reclaim);
1475 }
1476 }
1477
1478 free_htbl:
1479 rcu_assign_pointer(ntbl->htbl, NULL);
1480 call_rcu(&htbl->rcu, neigh_hash_free_rcu);
1481
1482 out_unlock:
1483 spin_unlock_irqrestore(&priv->lock, flags);
1484 if (wait_flushed)
1485 wait_for_completion(&priv->ntbl.flushed);
1486 }
1487
1488 static void ipoib_neigh_hash_uninit(struct net_device *dev)
1489 {
1490 struct ipoib_dev_priv *priv = netdev_priv(dev);
1491 int stopped;
1492
1493 ipoib_dbg(priv, "ipoib_neigh_hash_uninit\n");
1494 init_completion(&priv->ntbl.deleted);
1495 set_bit(IPOIB_NEIGH_TBL_FLUSH, &priv->flags);
1496
1497 /* Stop GC if called at init fail need to cancel work */
1498 stopped = test_and_set_bit(IPOIB_STOP_NEIGH_GC, &priv->flags);
1499 if (!stopped)
1500 cancel_delayed_work(&priv->neigh_reap_task);
1501
1502 ipoib_flush_neighs(priv);
1503
1504 wait_for_completion(&priv->ntbl.deleted);
1505 }
1506
1507
1508 int ipoib_dev_init(struct net_device *dev, struct ib_device *ca, int port)
1509 {
1510 struct ipoib_dev_priv *priv = netdev_priv(dev);
1511
1512 /* Allocate RX/TX "rings" to hold queued skbs */
1513 priv->rx_ring = kzalloc(ipoib_recvq_size * sizeof *priv->rx_ring,
1514 GFP_KERNEL);
1515 if (!priv->rx_ring) {
1516 printk(KERN_WARNING "%s: failed to allocate RX ring (%d entries)\n",
1517 ca->name, ipoib_recvq_size);
1518 goto out;
1519 }
1520
1521 priv->tx_ring = vzalloc(ipoib_sendq_size * sizeof *priv->tx_ring);
1522 if (!priv->tx_ring) {
1523 printk(KERN_WARNING "%s: failed to allocate TX ring (%d entries)\n",
1524 ca->name, ipoib_sendq_size);
1525 goto out_rx_ring_cleanup;
1526 }
1527
1528 /* priv->tx_head, tx_tail & tx_outstanding are already 0 */
1529
1530 if (ipoib_ib_dev_init(dev, ca, port))
1531 goto out_tx_ring_cleanup;
1532
1533 /*
1534 * Must be after ipoib_ib_dev_init so we can allocate a per
1535 * device wq there and use it here
1536 */
1537 if (ipoib_neigh_hash_init(priv) < 0)
1538 goto out_dev_uninit;
1539
1540 return 0;
1541
1542 out_dev_uninit:
1543 ipoib_ib_dev_cleanup(dev);
1544
1545 out_tx_ring_cleanup:
1546 vfree(priv->tx_ring);
1547
1548 out_rx_ring_cleanup:
1549 kfree(priv->rx_ring);
1550
1551 out:
1552 return -ENOMEM;
1553 }
1554
1555 void ipoib_dev_cleanup(struct net_device *dev)
1556 {
1557 struct ipoib_dev_priv *priv = netdev_priv(dev), *cpriv, *tcpriv;
1558 LIST_HEAD(head);
1559
1560 ASSERT_RTNL();
1561
1562 ipoib_delete_debug_files(dev);
1563
1564 /* Delete any child interfaces first */
1565 list_for_each_entry_safe(cpriv, tcpriv, &priv->child_intfs, list) {
1566 /* Stop GC on child */
1567 set_bit(IPOIB_STOP_NEIGH_GC, &cpriv->flags);
1568 cancel_delayed_work(&cpriv->neigh_reap_task);
1569 unregister_netdevice_queue(cpriv->dev, &head);
1570 }
1571 unregister_netdevice_many(&head);
1572
1573 /*
1574 * Must be before ipoib_ib_dev_cleanup or we delete an in use
1575 * work queue
1576 */
1577 ipoib_neigh_hash_uninit(dev);
1578
1579 ipoib_ib_dev_cleanup(dev);
1580
1581 kfree(priv->rx_ring);
1582 vfree(priv->tx_ring);
1583
1584 priv->rx_ring = NULL;
1585 priv->tx_ring = NULL;
1586 }
1587
1588 static const struct header_ops ipoib_header_ops = {
1589 .create = ipoib_hard_header,
1590 };
1591
1592 static const struct net_device_ops ipoib_netdev_ops = {
1593 .ndo_uninit = ipoib_uninit,
1594 .ndo_open = ipoib_open,
1595 .ndo_stop = ipoib_stop,
1596 .ndo_change_mtu = ipoib_change_mtu,
1597 .ndo_fix_features = ipoib_fix_features,
1598 .ndo_start_xmit = ipoib_start_xmit,
1599 .ndo_tx_timeout = ipoib_timeout,
1600 .ndo_set_rx_mode = ipoib_set_mcast_list,
1601 .ndo_get_iflink = ipoib_get_iflink,
1602 };
1603
1604 void ipoib_setup(struct net_device *dev)
1605 {
1606 struct ipoib_dev_priv *priv = netdev_priv(dev);
1607
1608 dev->netdev_ops = &ipoib_netdev_ops;
1609 dev->header_ops = &ipoib_header_ops;
1610
1611 ipoib_set_ethtool_ops(dev);
1612
1613 netif_napi_add(dev, &priv->napi, ipoib_poll, NAPI_POLL_WEIGHT);
1614
1615 dev->watchdog_timeo = HZ;
1616
1617 dev->flags |= IFF_BROADCAST | IFF_MULTICAST;
1618
1619 dev->hard_header_len = IPOIB_ENCAP_LEN;
1620 dev->addr_len = INFINIBAND_ALEN;
1621 dev->type = ARPHRD_INFINIBAND;
1622 dev->tx_queue_len = ipoib_sendq_size * 2;
1623 dev->features = (NETIF_F_VLAN_CHALLENGED |
1624 NETIF_F_HIGHDMA);
1625 netif_keep_dst(dev);
1626
1627 memcpy(dev->broadcast, ipv4_bcast_addr, INFINIBAND_ALEN);
1628
1629 priv->dev = dev;
1630
1631 spin_lock_init(&priv->lock);
1632
1633 init_rwsem(&priv->vlan_rwsem);
1634
1635 INIT_LIST_HEAD(&priv->path_list);
1636 INIT_LIST_HEAD(&priv->child_intfs);
1637 INIT_LIST_HEAD(&priv->dead_ahs);
1638 INIT_LIST_HEAD(&priv->multicast_list);
1639
1640 INIT_DELAYED_WORK(&priv->mcast_task, ipoib_mcast_join_task);
1641 INIT_WORK(&priv->carrier_on_task, ipoib_mcast_carrier_on_task);
1642 INIT_WORK(&priv->flush_light, ipoib_ib_dev_flush_light);
1643 INIT_WORK(&priv->flush_normal, ipoib_ib_dev_flush_normal);
1644 INIT_WORK(&priv->flush_heavy, ipoib_ib_dev_flush_heavy);
1645 INIT_WORK(&priv->restart_task, ipoib_mcast_restart_task);
1646 INIT_DELAYED_WORK(&priv->ah_reap_task, ipoib_reap_ah);
1647 INIT_DELAYED_WORK(&priv->neigh_reap_task, ipoib_reap_neigh);
1648 }
1649
1650 struct ipoib_dev_priv *ipoib_intf_alloc(const char *name)
1651 {
1652 struct net_device *dev;
1653
1654 dev = alloc_netdev((int)sizeof(struct ipoib_dev_priv), name,
1655 NET_NAME_UNKNOWN, ipoib_setup);
1656 if (!dev)
1657 return NULL;
1658
1659 return netdev_priv(dev);
1660 }
1661
1662 static ssize_t show_pkey(struct device *dev,
1663 struct device_attribute *attr, char *buf)
1664 {
1665 struct ipoib_dev_priv *priv = netdev_priv(to_net_dev(dev));
1666
1667 return sprintf(buf, "0x%04x\n", priv->pkey);
1668 }
1669 static DEVICE_ATTR(pkey, S_IRUGO, show_pkey, NULL);
1670
1671 static ssize_t show_umcast(struct device *dev,
1672 struct device_attribute *attr, char *buf)
1673 {
1674 struct ipoib_dev_priv *priv = netdev_priv(to_net_dev(dev));
1675
1676 return sprintf(buf, "%d\n", test_bit(IPOIB_FLAG_UMCAST, &priv->flags));
1677 }
1678
1679 void ipoib_set_umcast(struct net_device *ndev, int umcast_val)
1680 {
1681 struct ipoib_dev_priv *priv = netdev_priv(ndev);
1682
1683 if (umcast_val > 0) {
1684 set_bit(IPOIB_FLAG_UMCAST, &priv->flags);
1685 ipoib_warn(priv, "ignoring multicast groups joined directly "
1686 "by userspace\n");
1687 } else
1688 clear_bit(IPOIB_FLAG_UMCAST, &priv->flags);
1689 }
1690
1691 static ssize_t set_umcast(struct device *dev,
1692 struct device_attribute *attr,
1693 const char *buf, size_t count)
1694 {
1695 unsigned long umcast_val = simple_strtoul(buf, NULL, 0);
1696
1697 ipoib_set_umcast(to_net_dev(dev), umcast_val);
1698
1699 return count;
1700 }
1701 static DEVICE_ATTR(umcast, S_IWUSR | S_IRUGO, show_umcast, set_umcast);
1702
1703 int ipoib_add_umcast_attr(struct net_device *dev)
1704 {
1705 return device_create_file(&dev->dev, &dev_attr_umcast);
1706 }
1707
1708 static ssize_t create_child(struct device *dev,
1709 struct device_attribute *attr,
1710 const char *buf, size_t count)
1711 {
1712 int pkey;
1713 int ret;
1714
1715 if (sscanf(buf, "%i", &pkey) != 1)
1716 return -EINVAL;
1717
1718 if (pkey <= 0 || pkey > 0xffff || pkey == 0x8000)
1719 return -EINVAL;
1720
1721 /*
1722 * Set the full membership bit, so that we join the right
1723 * broadcast group, etc.
1724 */
1725 pkey |= 0x8000;
1726
1727 ret = ipoib_vlan_add(to_net_dev(dev), pkey);
1728
1729 return ret ? ret : count;
1730 }
1731 static DEVICE_ATTR(create_child, S_IWUSR, NULL, create_child);
1732
1733 static ssize_t delete_child(struct device *dev,
1734 struct device_attribute *attr,
1735 const char *buf, size_t count)
1736 {
1737 int pkey;
1738 int ret;
1739
1740 if (sscanf(buf, "%i", &pkey) != 1)
1741 return -EINVAL;
1742
1743 if (pkey < 0 || pkey > 0xffff)
1744 return -EINVAL;
1745
1746 ret = ipoib_vlan_delete(to_net_dev(dev), pkey);
1747
1748 return ret ? ret : count;
1749
1750 }
1751 static DEVICE_ATTR(delete_child, S_IWUSR, NULL, delete_child);
1752
1753 int ipoib_add_pkey_attr(struct net_device *dev)
1754 {
1755 return device_create_file(&dev->dev, &dev_attr_pkey);
1756 }
1757
1758 int ipoib_set_dev_features(struct ipoib_dev_priv *priv, struct ib_device *hca)
1759 {
1760 struct ib_device_attr *device_attr;
1761 int result = -ENOMEM;
1762
1763 device_attr = kmalloc(sizeof *device_attr, GFP_KERNEL);
1764 if (!device_attr) {
1765 printk(KERN_WARNING "%s: allocation of %zu bytes failed\n",
1766 hca->name, sizeof *device_attr);
1767 return result;
1768 }
1769
1770 result = ib_query_device(hca, device_attr);
1771 if (result) {
1772 printk(KERN_WARNING "%s: ib_query_device failed (ret = %d)\n",
1773 hca->name, result);
1774 kfree(device_attr);
1775 return result;
1776 }
1777 priv->hca_caps = device_attr->device_cap_flags;
1778
1779 kfree(device_attr);
1780
1781 if (priv->hca_caps & IB_DEVICE_UD_IP_CSUM) {
1782 priv->dev->hw_features = NETIF_F_SG |
1783 NETIF_F_IP_CSUM | NETIF_F_RXCSUM;
1784
1785 if (priv->hca_caps & IB_DEVICE_UD_TSO)
1786 priv->dev->hw_features |= NETIF_F_TSO;
1787
1788 priv->dev->features |= priv->dev->hw_features;
1789 }
1790
1791 return 0;
1792 }
1793
1794 static struct net_device *ipoib_add_port(const char *format,
1795 struct ib_device *hca, u8 port)
1796 {
1797 struct ipoib_dev_priv *priv;
1798 struct ib_port_attr attr;
1799 int result = -ENOMEM;
1800
1801 priv = ipoib_intf_alloc(format);
1802 if (!priv)
1803 goto alloc_mem_failed;
1804
1805 SET_NETDEV_DEV(priv->dev, hca->dma_device);
1806 priv->dev->dev_id = port - 1;
1807
1808 result = ib_query_port(hca, port, &attr);
1809 if (!result)
1810 priv->max_ib_mtu = ib_mtu_enum_to_int(attr.max_mtu);
1811 else {
1812 printk(KERN_WARNING "%s: ib_query_port %d failed\n",
1813 hca->name, port);
1814 goto device_init_failed;
1815 }
1816
1817 /* MTU will be reset when mcast join happens */
1818 priv->dev->mtu = IPOIB_UD_MTU(priv->max_ib_mtu);
1819 priv->mcast_mtu = priv->admin_mtu = priv->dev->mtu;
1820
1821 priv->dev->neigh_priv_len = sizeof(struct ipoib_neigh);
1822
1823 result = ib_query_pkey(hca, port, 0, &priv->pkey);
1824 if (result) {
1825 printk(KERN_WARNING "%s: ib_query_pkey port %d failed (ret = %d)\n",
1826 hca->name, port, result);
1827 goto device_init_failed;
1828 }
1829
1830 result = ipoib_set_dev_features(priv, hca);
1831 if (result)
1832 goto device_init_failed;
1833
1834 /*
1835 * Set the full membership bit, so that we join the right
1836 * broadcast group, etc.
1837 */
1838 priv->pkey |= 0x8000;
1839
1840 priv->dev->broadcast[8] = priv->pkey >> 8;
1841 priv->dev->broadcast[9] = priv->pkey & 0xff;
1842
1843 result = ib_query_gid(hca, port, 0, &priv->local_gid);
1844 if (result) {
1845 printk(KERN_WARNING "%s: ib_query_gid port %d failed (ret = %d)\n",
1846 hca->name, port, result);
1847 goto device_init_failed;
1848 } else
1849 memcpy(priv->dev->dev_addr + 4, priv->local_gid.raw, sizeof (union ib_gid));
1850
1851 result = ipoib_dev_init(priv->dev, hca, port);
1852 if (result < 0) {
1853 printk(KERN_WARNING "%s: failed to initialize port %d (ret = %d)\n",
1854 hca->name, port, result);
1855 goto device_init_failed;
1856 }
1857
1858 INIT_IB_EVENT_HANDLER(&priv->event_handler,
1859 priv->ca, ipoib_event);
1860 result = ib_register_event_handler(&priv->event_handler);
1861 if (result < 0) {
1862 printk(KERN_WARNING "%s: ib_register_event_handler failed for "
1863 "port %d (ret = %d)\n",
1864 hca->name, port, result);
1865 goto event_failed;
1866 }
1867
1868 result = register_netdev(priv->dev);
1869 if (result) {
1870 printk(KERN_WARNING "%s: couldn't register ipoib port %d; error %d\n",
1871 hca->name, port, result);
1872 goto register_failed;
1873 }
1874
1875 ipoib_create_debug_files(priv->dev);
1876
1877 if (ipoib_cm_add_mode_attr(priv->dev))
1878 goto sysfs_failed;
1879 if (ipoib_add_pkey_attr(priv->dev))
1880 goto sysfs_failed;
1881 if (ipoib_add_umcast_attr(priv->dev))
1882 goto sysfs_failed;
1883 if (device_create_file(&priv->dev->dev, &dev_attr_create_child))
1884 goto sysfs_failed;
1885 if (device_create_file(&priv->dev->dev, &dev_attr_delete_child))
1886 goto sysfs_failed;
1887
1888 return priv->dev;
1889
1890 sysfs_failed:
1891 ipoib_delete_debug_files(priv->dev);
1892 unregister_netdev(priv->dev);
1893
1894 register_failed:
1895 ib_unregister_event_handler(&priv->event_handler);
1896 flush_workqueue(ipoib_workqueue);
1897 /* Stop GC if started before flush */
1898 set_bit(IPOIB_STOP_NEIGH_GC, &priv->flags);
1899 cancel_delayed_work(&priv->neigh_reap_task);
1900 flush_workqueue(priv->wq);
1901
1902 event_failed:
1903 ipoib_dev_cleanup(priv->dev);
1904
1905 device_init_failed:
1906 free_netdev(priv->dev);
1907
1908 alloc_mem_failed:
1909 return ERR_PTR(result);
1910 }
1911
1912 static void ipoib_add_one(struct ib_device *device)
1913 {
1914 struct list_head *dev_list;
1915 struct net_device *dev;
1916 struct ipoib_dev_priv *priv;
1917 int p;
1918 int count = 0;
1919
1920 dev_list = kmalloc(sizeof *dev_list, GFP_KERNEL);
1921 if (!dev_list)
1922 return;
1923
1924 INIT_LIST_HEAD(dev_list);
1925
1926 for (p = rdma_start_port(device); p <= rdma_end_port(device); ++p) {
1927 if (!rdma_protocol_ib(device, p))
1928 continue;
1929 dev = ipoib_add_port("ib%d", device, p);
1930 if (!IS_ERR(dev)) {
1931 priv = netdev_priv(dev);
1932 list_add_tail(&priv->list, dev_list);
1933 count++;
1934 }
1935 }
1936
1937 if (!count) {
1938 kfree(dev_list);
1939 return;
1940 }
1941
1942 ib_set_client_data(device, &ipoib_client, dev_list);
1943 }
1944
1945 static void ipoib_remove_one(struct ib_device *device, void *client_data)
1946 {
1947 struct ipoib_dev_priv *priv, *tmp;
1948 struct list_head *dev_list = client_data;
1949
1950 if (!dev_list)
1951 return;
1952
1953 list_for_each_entry_safe(priv, tmp, dev_list, list) {
1954 ib_unregister_event_handler(&priv->event_handler);
1955 flush_workqueue(ipoib_workqueue);
1956
1957 rtnl_lock();
1958 dev_change_flags(priv->dev, priv->dev->flags & ~IFF_UP);
1959 rtnl_unlock();
1960
1961 /* Stop GC */
1962 set_bit(IPOIB_STOP_NEIGH_GC, &priv->flags);
1963 cancel_delayed_work(&priv->neigh_reap_task);
1964 flush_workqueue(priv->wq);
1965
1966 unregister_netdev(priv->dev);
1967 free_netdev(priv->dev);
1968 }
1969
1970 kfree(dev_list);
1971 }
1972
1973 static int __init ipoib_init_module(void)
1974 {
1975 int ret;
1976
1977 ipoib_recvq_size = roundup_pow_of_two(ipoib_recvq_size);
1978 ipoib_recvq_size = min(ipoib_recvq_size, IPOIB_MAX_QUEUE_SIZE);
1979 ipoib_recvq_size = max(ipoib_recvq_size, IPOIB_MIN_QUEUE_SIZE);
1980
1981 ipoib_sendq_size = roundup_pow_of_two(ipoib_sendq_size);
1982 ipoib_sendq_size = min(ipoib_sendq_size, IPOIB_MAX_QUEUE_SIZE);
1983 ipoib_sendq_size = max3(ipoib_sendq_size, 2 * MAX_SEND_CQE, IPOIB_MIN_QUEUE_SIZE);
1984 #ifdef CONFIG_INFINIBAND_IPOIB_CM
1985 ipoib_max_conn_qp = min(ipoib_max_conn_qp, IPOIB_CM_MAX_CONN_QP);
1986 #endif
1987
1988 /*
1989 * When copying small received packets, we only copy from the
1990 * linear data part of the SKB, so we rely on this condition.
1991 */
1992 BUILD_BUG_ON(IPOIB_CM_COPYBREAK > IPOIB_CM_HEAD_SIZE);
1993
1994 ret = ipoib_register_debugfs();
1995 if (ret)
1996 return ret;
1997
1998 /*
1999 * We create a global workqueue here that is used for all flush
2000 * operations. However, if you attempt to flush a workqueue
2001 * from a task on that same workqueue, it deadlocks the system.
2002 * We want to be able to flush the tasks associated with a
2003 * specific net device, so we also create a workqueue for each
2004 * netdevice. We queue up the tasks for that device only on
2005 * its private workqueue, and we only queue up flush events
2006 * on our global flush workqueue. This avoids the deadlocks.
2007 */
2008 ipoib_workqueue = create_singlethread_workqueue("ipoib_flush");
2009 if (!ipoib_workqueue) {
2010 ret = -ENOMEM;
2011 goto err_fs;
2012 }
2013
2014 ib_sa_register_client(&ipoib_sa_client);
2015
2016 ret = ib_register_client(&ipoib_client);
2017 if (ret)
2018 goto err_sa;
2019
2020 ret = ipoib_netlink_init();
2021 if (ret)
2022 goto err_client;
2023
2024 return 0;
2025
2026 err_client:
2027 ib_unregister_client(&ipoib_client);
2028
2029 err_sa:
2030 ib_sa_unregister_client(&ipoib_sa_client);
2031 destroy_workqueue(ipoib_workqueue);
2032
2033 err_fs:
2034 ipoib_unregister_debugfs();
2035
2036 return ret;
2037 }
2038
2039 static void __exit ipoib_cleanup_module(void)
2040 {
2041 ipoib_netlink_fini();
2042 ib_unregister_client(&ipoib_client);
2043 ib_sa_unregister_client(&ipoib_sa_client);
2044 ipoib_unregister_debugfs();
2045 destroy_workqueue(ipoib_workqueue);
2046 }
2047
2048 module_init(ipoib_init_module);
2049 module_exit(ipoib_cleanup_module);
Linux kernel - Deliverables
This page took 0.069925 seconds and 4 git commands to generate.