blob: 5f948b9a9943644a4bb6eafbb068705af7212b69 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Copyright (c) 2004 Topspin Communications. All rights reserved.
Roland Dreier2a1d9b72005-08-10 23:03:10 -07003 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
4 * Copyright (c) 2004 Voltaire, Inc. All rights reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07005 *
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 * $Id: ipoib_main.c 1377 2004-12-23 19:57:12Z roland $
35 */
36
37#include "ipoib.h"
38
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include <linux/module.h>
40
41#include <linux/init.h>
42#include <linux/slab.h>
Shirley Ma0f485252006-04-10 09:43:58 -070043#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070044
45#include <linux/if_arp.h> /* For ARPHRD_xxx */
46
47#include <linux/ip.h>
48#include <linux/in.h>
49
Arnaldo Carvalho de Melo14c85022005-12-27 02:43:12 -020050#include <net/dst.h>
51
Linus Torvalds1da177e2005-04-16 15:20:36 -070052MODULE_AUTHOR("Roland Dreier");
53MODULE_DESCRIPTION("IP-over-InfiniBand net driver");
54MODULE_LICENSE("Dual BSD/GPL");
55
Shirley Ma0f485252006-04-10 09:43:58 -070056int ipoib_sendq_size __read_mostly = IPOIB_TX_RING_SIZE;
57int ipoib_recvq_size __read_mostly = IPOIB_RX_RING_SIZE;
58
59module_param_named(send_queue_size, ipoib_sendq_size, int, 0444);
60MODULE_PARM_DESC(send_queue_size, "Number of descriptors in send queue");
61module_param_named(recv_queue_size, ipoib_recvq_size, int, 0444);
62MODULE_PARM_DESC(recv_queue_size, "Number of descriptors in receive queue");
63
Linus Torvalds1da177e2005-04-16 15:20:36 -070064#ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
65int ipoib_debug_level;
66
67module_param_named(debug_level, ipoib_debug_level, int, 0644);
68MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0");
69#endif
70
Roland Dreier1732b0e2005-11-07 10:33:11 -080071struct ipoib_path_iter {
72 struct net_device *dev;
73 struct ipoib_path path;
74};
75
Linus Torvalds1da177e2005-04-16 15:20:36 -070076static const u8 ipv4_bcast_addr[] = {
77 0x00, 0xff, 0xff, 0xff,
78 0xff, 0x12, 0x40, 0x1b, 0x00, 0x00, 0x00, 0x00,
79 0x00, 0x00, 0x00, 0x00, 0xff, 0xff, 0xff, 0xff
80};
81
82struct workqueue_struct *ipoib_workqueue;
83
Michael S. Tsirkinc1a0b232006-08-21 16:40:12 -070084struct ib_sa_client ipoib_sa_client;
85
Linus Torvalds1da177e2005-04-16 15:20:36 -070086static void ipoib_add_one(struct ib_device *device);
87static void ipoib_remove_one(struct ib_device *device);
88
89static struct ib_client ipoib_client = {
90 .name = "ipoib",
91 .add = ipoib_add_one,
92 .remove = ipoib_remove_one
93};
94
95int ipoib_open(struct net_device *dev)
96{
97 struct ipoib_dev_priv *priv = netdev_priv(dev);
98
99 ipoib_dbg(priv, "bringing up interface\n");
100
101 set_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags);
102
103 if (ipoib_pkey_dev_delay_open(dev))
104 return 0;
105
106 if (ipoib_ib_dev_open(dev))
107 return -EINVAL;
108
Roland Dreier267ee882005-11-29 10:55:58 -0800109 if (ipoib_ib_dev_up(dev)) {
Yosef Etigin26bbf132007-05-19 08:51:54 -0700110 ipoib_ib_dev_stop(dev, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 return -EINVAL;
Roland Dreier267ee882005-11-29 10:55:58 -0800112 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113
114 if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) {
115 struct ipoib_dev_priv *cpriv;
116
117 /* Bring up any child interfaces too */
Ingo Molnar95ed6442006-01-13 14:51:39 -0800118 mutex_lock(&priv->vlan_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 list_for_each_entry(cpriv, &priv->child_intfs, list) {
120 int flags;
121
122 flags = cpriv->dev->flags;
123 if (flags & IFF_UP)
124 continue;
125
126 dev_change_flags(cpriv->dev, flags | IFF_UP);
127 }
Ingo Molnar95ed6442006-01-13 14:51:39 -0800128 mutex_unlock(&priv->vlan_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 }
130
131 netif_start_queue(dev);
132
133 return 0;
134}
135
136static int ipoib_stop(struct net_device *dev)
137{
138 struct ipoib_dev_priv *priv = netdev_priv(dev);
139
140 ipoib_dbg(priv, "stopping interface\n");
141
142 clear_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags);
143
144 netif_stop_queue(dev);
145
Michael S. Tsirkin839fcab2007-02-05 22:12:23 +0200146 clear_bit(IPOIB_FLAG_NETIF_STOPPED, &priv->flags);
147
Jack Morgenstein0b3ea082006-03-20 10:08:24 -0800148 /*
149 * Now flush workqueue to make sure a scheduled task doesn't
150 * bring our internal state back up.
151 */
152 flush_workqueue(ipoib_workqueue);
153
154 ipoib_ib_dev_down(dev, 1);
Yosef Etigin26bbf132007-05-19 08:51:54 -0700155 ipoib_ib_dev_stop(dev, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156
157 if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) {
158 struct ipoib_dev_priv *cpriv;
159
160 /* Bring down any child interfaces too */
Ingo Molnar95ed6442006-01-13 14:51:39 -0800161 mutex_lock(&priv->vlan_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 list_for_each_entry(cpriv, &priv->child_intfs, list) {
163 int flags;
164
165 flags = cpriv->dev->flags;
166 if (!(flags & IFF_UP))
167 continue;
168
169 dev_change_flags(cpriv->dev, flags & ~IFF_UP);
170 }
Ingo Molnar95ed6442006-01-13 14:51:39 -0800171 mutex_unlock(&priv->vlan_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 }
173
174 return 0;
175}
176
177static int ipoib_change_mtu(struct net_device *dev, int new_mtu)
178{
179 struct ipoib_dev_priv *priv = netdev_priv(dev);
180
Michael S. Tsirkin839fcab2007-02-05 22:12:23 +0200181 /* dev->mtu > 2K ==> connected mode */
182 if (ipoib_cm_admin_enabled(dev) && new_mtu <= IPOIB_CM_MTU) {
183 if (new_mtu > priv->mcast_mtu)
184 ipoib_warn(priv, "mtu > %d will cause multicast packet drops.\n",
185 priv->mcast_mtu);
186 dev->mtu = new_mtu;
187 return 0;
188 }
189
190 if (new_mtu > IPOIB_PACKET_SIZE - IPOIB_ENCAP_LEN) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 return -EINVAL;
Michael S. Tsirkin839fcab2007-02-05 22:12:23 +0200192 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193
194 priv->admin_mtu = new_mtu;
195
196 dev->mtu = min(priv->mcast_mtu, priv->admin_mtu);
197
198 return 0;
199}
200
Jack Morgenstein37c22a72006-05-29 19:14:05 +0300201static struct ipoib_path *__path_find(struct net_device *dev, void *gid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202{
203 struct ipoib_dev_priv *priv = netdev_priv(dev);
204 struct rb_node *n = priv->path_tree.rb_node;
205 struct ipoib_path *path;
206 int ret;
207
208 while (n) {
209 path = rb_entry(n, struct ipoib_path, rb_node);
210
Jack Morgenstein37c22a72006-05-29 19:14:05 +0300211 ret = memcmp(gid, path->pathrec.dgid.raw,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 sizeof (union ib_gid));
213
214 if (ret < 0)
215 n = n->rb_left;
216 else if (ret > 0)
217 n = n->rb_right;
218 else
219 return path;
220 }
221
222 return NULL;
223}
224
225static int __path_add(struct net_device *dev, struct ipoib_path *path)
226{
227 struct ipoib_dev_priv *priv = netdev_priv(dev);
228 struct rb_node **n = &priv->path_tree.rb_node;
229 struct rb_node *pn = NULL;
230 struct ipoib_path *tpath;
231 int ret;
232
233 while (*n) {
234 pn = *n;
235 tpath = rb_entry(pn, struct ipoib_path, rb_node);
236
237 ret = memcmp(path->pathrec.dgid.raw, tpath->pathrec.dgid.raw,
238 sizeof (union ib_gid));
239 if (ret < 0)
240 n = &pn->rb_left;
241 else if (ret > 0)
242 n = &pn->rb_right;
243 else
244 return -EEXIST;
245 }
246
247 rb_link_node(&path->rb_node, pn, n);
248 rb_insert_color(&path->rb_node, &priv->path_tree);
249
250 list_add_tail(&path->list, &priv->path_list);
251
252 return 0;
253}
254
255static void path_free(struct net_device *dev, struct ipoib_path *path)
256{
257 struct ipoib_dev_priv *priv = netdev_priv(dev);
258 struct ipoib_neigh *neigh, *tn;
259 struct sk_buff *skb;
260 unsigned long flags;
261
262 while ((skb = __skb_dequeue(&path->queue)))
263 dev_kfree_skb_irq(skb);
264
265 spin_lock_irqsave(&priv->lock, flags);
266
267 list_for_each_entry_safe(neigh, tn, &path->neigh_list, list) {
268 /*
269 * It's safe to call ipoib_put_ah() inside priv->lock
270 * here, because we know that path->ah will always
271 * hold one more reference, so ipoib_put_ah() will
272 * never do more than decrement the ref count.
273 */
274 if (neigh->ah)
275 ipoib_put_ah(neigh->ah);
Michael S. Tsirkind2e06552006-04-04 19:59:40 +0300276
Michael S. Tsirkin2745b5b2006-11-16 14:16:47 +0200277 ipoib_neigh_free(dev, neigh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 }
279
280 spin_unlock_irqrestore(&priv->lock, flags);
281
282 if (path->ah)
283 ipoib_put_ah(path->ah);
284
285 kfree(path);
286}
287
Roland Dreier1732b0e2005-11-07 10:33:11 -0800288#ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
289
290struct ipoib_path_iter *ipoib_path_iter_init(struct net_device *dev)
291{
292 struct ipoib_path_iter *iter;
293
294 iter = kmalloc(sizeof *iter, GFP_KERNEL);
295 if (!iter)
296 return NULL;
297
298 iter->dev = dev;
299 memset(iter->path.pathrec.dgid.raw, 0, 16);
300
301 if (ipoib_path_iter_next(iter)) {
302 kfree(iter);
303 return NULL;
304 }
305
306 return iter;
307}
308
309int ipoib_path_iter_next(struct ipoib_path_iter *iter)
310{
311 struct ipoib_dev_priv *priv = netdev_priv(iter->dev);
312 struct rb_node *n;
313 struct ipoib_path *path;
314 int ret = 1;
315
316 spin_lock_irq(&priv->lock);
317
318 n = rb_first(&priv->path_tree);
319
320 while (n) {
321 path = rb_entry(n, struct ipoib_path, rb_node);
322
323 if (memcmp(iter->path.pathrec.dgid.raw, path->pathrec.dgid.raw,
324 sizeof (union ib_gid)) < 0) {
325 iter->path = *path;
326 ret = 0;
327 break;
328 }
329
330 n = rb_next(n);
331 }
332
333 spin_unlock_irq(&priv->lock);
334
335 return ret;
336}
337
338void ipoib_path_iter_read(struct ipoib_path_iter *iter,
339 struct ipoib_path *path)
340{
341 *path = iter->path;
342}
343
344#endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */
345
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346void ipoib_flush_paths(struct net_device *dev)
347{
348 struct ipoib_dev_priv *priv = netdev_priv(dev);
349 struct ipoib_path *path, *tp;
350 LIST_HEAD(remove_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351
Michael S. Tsirkin9217b272006-08-03 22:16:06 +0300352 spin_lock_irq(&priv->tx_lock);
353 spin_lock(&priv->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354
355 list_splice(&priv->path_list, &remove_list);
356 INIT_LIST_HEAD(&priv->path_list);
357
358 list_for_each_entry(path, &remove_list, list)
359 rb_erase(&path->rb_node, &priv->path_tree);
360
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 list_for_each_entry_safe(path, tp, &remove_list, list) {
362 if (path->query)
363 ib_sa_cancel_query(path->query_id, path->query);
Michael S. Tsirkin9217b272006-08-03 22:16:06 +0300364 spin_unlock(&priv->lock);
365 spin_unlock_irq(&priv->tx_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 wait_for_completion(&path->done);
367 path_free(dev, path);
Michael S. Tsirkin9217b272006-08-03 22:16:06 +0300368 spin_lock_irq(&priv->tx_lock);
369 spin_lock(&priv->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 }
Michael S. Tsirkin9217b272006-08-03 22:16:06 +0300371 spin_unlock(&priv->lock);
372 spin_unlock_irq(&priv->tx_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373}
374
375static void path_rec_completion(int status,
376 struct ib_sa_path_rec *pathrec,
377 void *path_ptr)
378{
379 struct ipoib_path *path = path_ptr;
380 struct net_device *dev = path->dev;
381 struct ipoib_dev_priv *priv = netdev_priv(dev);
382 struct ipoib_ah *ah = NULL;
Michael S. Tsirkind04d01b2007-03-22 14:40:16 -0700383 struct ipoib_neigh *neigh, *tn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 struct sk_buff_head skqueue;
385 struct sk_buff *skb;
386 unsigned long flags;
387
Roland Dreier843613b2007-02-26 12:57:08 -0800388 if (!status)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 ipoib_dbg(priv, "PathRec LID 0x%04x for GID " IPOIB_GID_FMT "\n",
390 be16_to_cpu(pathrec->dlid), IPOIB_GID_ARG(pathrec->dgid));
391 else
392 ipoib_dbg(priv, "PathRec status %d for GID " IPOIB_GID_FMT "\n",
393 status, IPOIB_GID_ARG(path->pathrec.dgid));
394
395 skb_queue_head_init(&skqueue);
396
397 if (!status) {
Sean Hefty46f1b3d2007-04-05 11:50:11 -0700398 struct ib_ah_attr av;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399
Sean Hefty46f1b3d2007-04-05 11:50:11 -0700400 if (!ib_init_ah_from_path(priv->ca, priv->port, pathrec, &av))
401 ah = ipoib_create_ah(dev, priv->pd, &av);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 }
403
404 spin_lock_irqsave(&priv->lock, flags);
405
406 path->ah = ah;
407
408 if (ah) {
409 path->pathrec = *pathrec;
410
411 ipoib_dbg(priv, "created address handle %p for LID 0x%04x, SL %d\n",
412 ah, be16_to_cpu(pathrec->dlid), pathrec->sl);
413
414 while ((skb = __skb_dequeue(&path->queue)))
415 __skb_queue_tail(&skqueue, skb);
416
Michael S. Tsirkind04d01b2007-03-22 14:40:16 -0700417 list_for_each_entry_safe(neigh, tn, &path->neigh_list, list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 kref_get(&path->ah->ref);
419 neigh->ah = path->ah;
Michael S. Tsirkin8a7f7522006-07-19 17:44:37 +0300420 memcpy(&neigh->dgid.raw, &path->pathrec.dgid.raw,
421 sizeof(union ib_gid));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422
Michael S. Tsirkin839fcab2007-02-05 22:12:23 +0200423 if (ipoib_cm_enabled(dev, neigh->neighbour)) {
424 if (!ipoib_cm_get(neigh))
425 ipoib_cm_set(neigh, ipoib_cm_create_tx(dev,
426 path,
427 neigh));
428 if (!ipoib_cm_get(neigh)) {
429 list_del(&neigh->list);
430 if (neigh->ah)
431 ipoib_put_ah(neigh->ah);
432 ipoib_neigh_free(dev, neigh);
433 continue;
434 }
435 }
436
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 while ((skb = __skb_dequeue(&neigh->queue)))
438 __skb_queue_tail(&skqueue, skb);
439 }
Roland Dreier5872a9f2005-11-29 10:13:54 -0800440 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441
Roland Dreier5872a9f2005-11-29 10:13:54 -0800442 path->query = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 complete(&path->done);
444
445 spin_unlock_irqrestore(&priv->lock, flags);
446
447 while ((skb = __skb_dequeue(&skqueue))) {
448 skb->dev = dev;
449 if (dev_queue_xmit(skb))
450 ipoib_warn(priv, "dev_queue_xmit failed "
451 "to requeue packet\n");
452 }
453}
454
Jack Morgenstein37c22a72006-05-29 19:14:05 +0300455static struct ipoib_path *path_rec_create(struct net_device *dev, void *gid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456{
457 struct ipoib_dev_priv *priv = netdev_priv(dev);
458 struct ipoib_path *path;
459
Roland Dreier21a38482005-11-02 10:07:59 -0800460 path = kzalloc(sizeof *path, GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 if (!path)
462 return NULL;
463
Roland Dreier21a38482005-11-02 10:07:59 -0800464 path->dev = dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465
466 skb_queue_head_init(&path->queue);
467
468 INIT_LIST_HEAD(&path->neigh_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469
Jack Morgenstein37c22a72006-05-29 19:14:05 +0300470 memcpy(path->pathrec.dgid.raw, gid, sizeof (union ib_gid));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 path->pathrec.sgid = priv->local_gid;
472 path->pathrec.pkey = cpu_to_be16(priv->pkey);
473 path->pathrec.numb_path = 1;
474
475 return path;
476}
477
478static int path_rec_start(struct net_device *dev,
479 struct ipoib_path *path)
480{
481 struct ipoib_dev_priv *priv = netdev_priv(dev);
482
483 ipoib_dbg(priv, "Start path record lookup for " IPOIB_GID_FMT "\n",
484 IPOIB_GID_ARG(path->pathrec.dgid));
485
Roland Dreier65c7edd2005-11-28 21:20:34 -0800486 init_completion(&path->done);
487
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 path->query_id =
Michael S. Tsirkinc1a0b232006-08-21 16:40:12 -0700489 ib_sa_path_rec_get(&ipoib_sa_client, priv->ca, priv->port,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 &path->pathrec,
491 IB_SA_PATH_REC_DGID |
492 IB_SA_PATH_REC_SGID |
493 IB_SA_PATH_REC_NUMB_PATH |
494 IB_SA_PATH_REC_PKEY,
495 1000, GFP_ATOMIC,
496 path_rec_completion,
497 path, &path->query);
498 if (path->query_id < 0) {
499 ipoib_warn(priv, "ib_sa_path_rec_get failed\n");
500 path->query = NULL;
501 return path->query_id;
502 }
503
504 return 0;
505}
506
507static void neigh_add_path(struct sk_buff *skb, struct net_device *dev)
508{
509 struct ipoib_dev_priv *priv = netdev_priv(dev);
510 struct ipoib_path *path;
511 struct ipoib_neigh *neigh;
512
Michael S. Tsirkind2e06552006-04-04 19:59:40 +0300513 neigh = ipoib_neigh_alloc(skb->dst->neighbour);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 if (!neigh) {
515 ++priv->stats.tx_dropped;
516 dev_kfree_skb_any(skb);
517 return;
518 }
519
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 /*
521 * We can only be called from ipoib_start_xmit, so we're
522 * inside tx_lock -- no need to save/restore flags.
523 */
524 spin_lock(&priv->lock);
525
Jack Morgenstein37c22a72006-05-29 19:14:05 +0300526 path = __path_find(dev, skb->dst->neighbour->ha + 4);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 if (!path) {
Jack Morgenstein37c22a72006-05-29 19:14:05 +0300528 path = path_rec_create(dev, skb->dst->neighbour->ha + 4);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 if (!path)
Michael S. Tsirkind2e06552006-04-04 19:59:40 +0300530 goto err_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531
532 __path_add(dev, path);
533 }
534
535 list_add_tail(&neigh->list, &path->neigh_list);
536
Michael S. Tsirkin47f7a072006-01-17 09:22:05 -0800537 if (path->ah) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 kref_get(&path->ah->ref);
539 neigh->ah = path->ah;
Michael S. Tsirkin8a7f7522006-07-19 17:44:37 +0300540 memcpy(&neigh->dgid.raw, &path->pathrec.dgid.raw,
541 sizeof(union ib_gid));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542
Michael S. Tsirkin839fcab2007-02-05 22:12:23 +0200543 if (ipoib_cm_enabled(dev, neigh->neighbour)) {
544 if (!ipoib_cm_get(neigh))
545 ipoib_cm_set(neigh, ipoib_cm_create_tx(dev, path, neigh));
546 if (!ipoib_cm_get(neigh)) {
547 list_del(&neigh->list);
548 if (neigh->ah)
549 ipoib_put_ah(neigh->ah);
550 ipoib_neigh_free(dev, neigh);
551 goto err_drop;
552 }
553 if (skb_queue_len(&neigh->queue) < IPOIB_MAX_PATH_REC_QUEUE)
554 __skb_queue_tail(&neigh->queue, skb);
555 else {
556 ipoib_warn(priv, "queue length limit %d. Packet drop.\n",
557 skb_queue_len(&neigh->queue));
558 goto err_drop;
559 }
560 } else
561 ipoib_send(dev, skb, path->ah, IPOIB_QPN(skb->dst->neighbour->ha));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 } else {
563 neigh->ah = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564
565 if (!path->query && path_rec_start(dev, path))
Michael S. Tsirkind2e06552006-04-04 19:59:40 +0300566 goto err_list;
Michael S. Tsirkin2745b5b2006-11-16 14:16:47 +0200567
568 __skb_queue_tail(&neigh->queue, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 }
570
571 spin_unlock(&priv->lock);
572 return;
573
Michael S. Tsirkind2e06552006-04-04 19:59:40 +0300574err_list:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 list_del(&neigh->list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576
Michael S. Tsirkind2e06552006-04-04 19:59:40 +0300577err_path:
Michael S. Tsirkin2745b5b2006-11-16 14:16:47 +0200578 ipoib_neigh_free(dev, neigh);
Michael S. Tsirkin839fcab2007-02-05 22:12:23 +0200579err_drop:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 ++priv->stats.tx_dropped;
581 dev_kfree_skb_any(skb);
582
583 spin_unlock(&priv->lock);
584}
585
Roland Dreierd70ed602005-09-28 19:56:57 -0700586static void ipoib_path_lookup(struct sk_buff *skb, struct net_device *dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587{
588 struct ipoib_dev_priv *priv = netdev_priv(skb->dev);
589
590 /* Look up path record for unicasts */
591 if (skb->dst->neighbour->ha[4] != 0xff) {
592 neigh_add_path(skb, dev);
593 return;
594 }
595
596 /* Add in the P_Key for multicasts */
597 skb->dst->neighbour->ha[8] = (priv->pkey >> 8) & 0xff;
598 skb->dst->neighbour->ha[9] = priv->pkey & 0xff;
Jack Morgenstein37c22a72006-05-29 19:14:05 +0300599 ipoib_mcast_send(dev, skb->dst->neighbour->ha + 4, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600}
601
602static void unicast_arp_send(struct sk_buff *skb, struct net_device *dev,
603 struct ipoib_pseudoheader *phdr)
604{
605 struct ipoib_dev_priv *priv = netdev_priv(dev);
606 struct ipoib_path *path;
607
608 /*
609 * We can only be called from ipoib_start_xmit, so we're
610 * inside tx_lock -- no need to save/restore flags.
611 */
612 spin_lock(&priv->lock);
613
Jack Morgenstein37c22a72006-05-29 19:14:05 +0300614 path = __path_find(dev, phdr->hwaddr + 4);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 if (!path) {
Jack Morgenstein37c22a72006-05-29 19:14:05 +0300616 path = path_rec_create(dev, phdr->hwaddr + 4);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 if (path) {
618 /* put pseudoheader back on for next time */
619 skb_push(skb, sizeof *phdr);
620 __skb_queue_tail(&path->queue, skb);
621
622 if (path_rec_start(dev, path)) {
623 spin_unlock(&priv->lock);
624 path_free(dev, path);
625 return;
626 } else
627 __path_add(dev, path);
628 } else {
629 ++priv->stats.tx_dropped;
630 dev_kfree_skb_any(skb);
631 }
632
633 spin_unlock(&priv->lock);
634 return;
635 }
636
Michael S. Tsirkin47f7a072006-01-17 09:22:05 -0800637 if (path->ah) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 ipoib_dbg(priv, "Send unicast ARP to %04x\n",
639 be16_to_cpu(path->pathrec.dlid));
640
Michael S. Tsirkin073ae842006-11-16 10:59:12 +0200641 ipoib_send(dev, skb, path->ah, IPOIB_QPN(phdr->hwaddr));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 } else if ((path->query || !path_rec_start(dev, path)) &&
643 skb_queue_len(&path->queue) < IPOIB_MAX_PATH_REC_QUEUE) {
644 /* put pseudoheader back on for next time */
645 skb_push(skb, sizeof *phdr);
646 __skb_queue_tail(&path->queue, skb);
647 } else {
648 ++priv->stats.tx_dropped;
649 dev_kfree_skb_any(skb);
650 }
651
652 spin_unlock(&priv->lock);
653}
654
655static int ipoib_start_xmit(struct sk_buff *skb, struct net_device *dev)
656{
657 struct ipoib_dev_priv *priv = netdev_priv(dev);
658 struct ipoib_neigh *neigh;
659 unsigned long flags;
660
Eli Cohena8bfca02006-09-22 15:22:58 -0700661 if (unlikely(!spin_trylock_irqsave(&priv->tx_lock, flags)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 return NETDEV_TX_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663
664 /*
665 * Check if our queue is stopped. Since we have the LLTX bit
666 * set, we can't rely on netif_stop_queue() preventing our
667 * xmit function from being called with a full queue.
668 */
669 if (unlikely(netif_queue_stopped(dev))) {
670 spin_unlock_irqrestore(&priv->tx_lock, flags);
671 return NETDEV_TX_BUSY;
672 }
673
Eli Cohena8bfca02006-09-22 15:22:58 -0700674 if (likely(skb->dst && skb->dst->neighbour)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 if (unlikely(!*to_ipoib_neigh(skb->dst->neighbour))) {
Roland Dreierd70ed602005-09-28 19:56:57 -0700676 ipoib_path_lookup(skb, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 goto out;
678 }
679
680 neigh = *to_ipoib_neigh(skb->dst->neighbour);
681
Michael S. Tsirkin839fcab2007-02-05 22:12:23 +0200682 if (ipoib_cm_get(neigh)) {
683 if (ipoib_cm_up(neigh)) {
684 ipoib_cm_send(dev, skb, ipoib_cm_get(neigh));
685 goto out;
686 }
687 } else if (neigh->ah) {
Michael S. Tsirkin8a7f7522006-07-19 17:44:37 +0300688 if (unlikely(memcmp(&neigh->dgid.raw,
689 skb->dst->neighbour->ha + 4,
690 sizeof(union ib_gid)))) {
691 spin_lock(&priv->lock);
692 /*
693 * It's safe to call ipoib_put_ah() inside
694 * priv->lock here, because we know that
695 * path->ah will always hold one more reference,
696 * so ipoib_put_ah() will never do more than
697 * decrement the ref count.
698 */
699 ipoib_put_ah(neigh->ah);
700 list_del(&neigh->list);
Michael S. Tsirkin2745b5b2006-11-16 14:16:47 +0200701 ipoib_neigh_free(dev, neigh);
Michael S. Tsirkin8a7f7522006-07-19 17:44:37 +0300702 spin_unlock(&priv->lock);
703 ipoib_path_lookup(skb, dev);
704 goto out;
705 }
706
Michael S. Tsirkin073ae842006-11-16 10:59:12 +0200707 ipoib_send(dev, skb, neigh->ah, IPOIB_QPN(skb->dst->neighbour->ha));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 goto out;
709 }
710
711 if (skb_queue_len(&neigh->queue) < IPOIB_MAX_PATH_REC_QUEUE) {
712 spin_lock(&priv->lock);
713 __skb_queue_tail(&neigh->queue, skb);
714 spin_unlock(&priv->lock);
715 } else {
716 ++priv->stats.tx_dropped;
717 dev_kfree_skb_any(skb);
718 }
719 } else {
720 struct ipoib_pseudoheader *phdr =
721 (struct ipoib_pseudoheader *) skb->data;
722 skb_pull(skb, sizeof *phdr);
723
724 if (phdr->hwaddr[4] == 0xff) {
725 /* Add in the P_Key for multicast*/
726 phdr->hwaddr[8] = (priv->pkey >> 8) & 0xff;
727 phdr->hwaddr[9] = priv->pkey & 0xff;
728
Jack Morgenstein37c22a72006-05-29 19:14:05 +0300729 ipoib_mcast_send(dev, phdr->hwaddr + 4, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 } else {
Hal Rosenstock0dca0f72005-07-28 13:17:26 -0700731 /* unicast GID -- should be ARP or RARP reply */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732
Hal Rosenstock0dca0f72005-07-28 13:17:26 -0700733 if ((be16_to_cpup((__be16 *) skb->data) != ETH_P_ARP) &&
734 (be16_to_cpup((__be16 *) skb->data) != ETH_P_RARP)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 ipoib_warn(priv, "Unicast, no %s: type %04x, QPN %06x "
736 IPOIB_GID_FMT "\n",
737 skb->dst ? "neigh" : "dst",
Sean Hefty97f52eb2005-08-13 21:05:57 -0700738 be16_to_cpup((__be16 *) skb->data),
Michael S. Tsirkin073ae842006-11-16 10:59:12 +0200739 IPOIB_QPN(phdr->hwaddr),
Jack Morgenstein37c22a72006-05-29 19:14:05 +0300740 IPOIB_GID_RAW_ARG(phdr->hwaddr + 4));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 dev_kfree_skb_any(skb);
742 ++priv->stats.tx_dropped;
743 goto out;
744 }
745
746 unicast_arp_send(skb, dev, phdr);
747 }
748 }
749
750out:
751 spin_unlock_irqrestore(&priv->tx_lock, flags);
752
753 return NETDEV_TX_OK;
754}
755
756static struct net_device_stats *ipoib_get_stats(struct net_device *dev)
757{
758 struct ipoib_dev_priv *priv = netdev_priv(dev);
759
760 return &priv->stats;
761}
762
763static void ipoib_timeout(struct net_device *dev)
764{
765 struct ipoib_dev_priv *priv = netdev_priv(dev);
766
Roland Dreier4b2d3192005-10-18 12:20:06 -0700767 ipoib_warn(priv, "transmit timeout: latency %d msecs\n",
768 jiffies_to_msecs(jiffies - dev->trans_start));
769 ipoib_warn(priv, "queue stopped %d, tx_head %u, tx_tail %u\n",
770 netif_queue_stopped(dev),
771 priv->tx_head, priv->tx_tail);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 /* XXX reset QP, etc. */
773}
774
775static int ipoib_hard_header(struct sk_buff *skb,
776 struct net_device *dev,
777 unsigned short type,
778 void *daddr, void *saddr, unsigned len)
779{
780 struct ipoib_header *header;
781
782 header = (struct ipoib_header *) skb_push(skb, sizeof *header);
783
784 header->proto = htons(type);
785 header->reserved = 0;
786
787 /*
788 * If we don't have a neighbour structure, stuff the
789 * destination address onto the front of the skb so we can
790 * figure out where to send the packet later.
791 */
Roland Dreieref12d452006-03-29 09:36:46 -0800792 if ((!skb->dst || !skb->dst->neighbour) && daddr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 struct ipoib_pseudoheader *phdr =
794 (struct ipoib_pseudoheader *) skb_push(skb, sizeof *phdr);
795 memcpy(phdr->hwaddr, daddr, INFINIBAND_ALEN);
796 }
797
798 return 0;
799}
800
801static void ipoib_set_mcast_list(struct net_device *dev)
802{
803 struct ipoib_dev_priv *priv = netdev_priv(dev);
804
Leonid Arsh7a343d42006-03-23 19:52:51 +0200805 if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags)) {
806 ipoib_dbg(priv, "IPOIB_FLAG_OPER_UP not set");
807 return;
808 }
809
Michael S. Tsirkin1ad62a12005-08-24 14:41:51 -0700810 queue_work(ipoib_workqueue, &priv->restart_task);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811}
812
Alexey Kuznetsovecbb4162007-03-24 12:52:16 -0700813static void ipoib_neigh_cleanup(struct neighbour *n)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814{
815 struct ipoib_neigh *neigh;
816 struct ipoib_dev_priv *priv = netdev_priv(n->dev);
817 unsigned long flags;
818 struct ipoib_ah *ah = NULL;
819
820 ipoib_dbg(priv,
Alexey Kuznetsovecbb4162007-03-24 12:52:16 -0700821 "neigh_cleanup for %06x " IPOIB_GID_FMT "\n",
Michael S. Tsirkin073ae842006-11-16 10:59:12 +0200822 IPOIB_QPN(n->ha),
Jack Morgenstein37c22a72006-05-29 19:14:05 +0300823 IPOIB_GID_RAW_ARG(n->ha + 4));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824
825 spin_lock_irqsave(&priv->lock, flags);
826
827 neigh = *to_ipoib_neigh(n);
828 if (neigh) {
829 if (neigh->ah)
830 ah = neigh->ah;
831 list_del(&neigh->list);
Michael S. Tsirkin2745b5b2006-11-16 14:16:47 +0200832 ipoib_neigh_free(n->dev, neigh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 }
834
835 spin_unlock_irqrestore(&priv->lock, flags);
836
837 if (ah)
838 ipoib_put_ah(ah);
839}
840
Michael S. Tsirkind2e06552006-04-04 19:59:40 +0300841struct ipoib_neigh *ipoib_neigh_alloc(struct neighbour *neighbour)
842{
843 struct ipoib_neigh *neigh;
844
845 neigh = kmalloc(sizeof *neigh, GFP_ATOMIC);
846 if (!neigh)
847 return NULL;
848
849 neigh->neighbour = neighbour;
850 *to_ipoib_neigh(neighbour) = neigh;
Roland Dreier82b39912006-12-12 14:48:18 -0800851 skb_queue_head_init(&neigh->queue);
Michael S. Tsirkin839fcab2007-02-05 22:12:23 +0200852 ipoib_cm_set(neigh, NULL);
Michael S. Tsirkind2e06552006-04-04 19:59:40 +0300853
854 return neigh;
855}
856
Michael S. Tsirkin2745b5b2006-11-16 14:16:47 +0200857void ipoib_neigh_free(struct net_device *dev, struct ipoib_neigh *neigh)
Michael S. Tsirkind2e06552006-04-04 19:59:40 +0300858{
Michael S. Tsirkin2745b5b2006-11-16 14:16:47 +0200859 struct ipoib_dev_priv *priv = netdev_priv(dev);
860 struct sk_buff *skb;
Michael S. Tsirkind2e06552006-04-04 19:59:40 +0300861 *to_ipoib_neigh(neigh->neighbour) = NULL;
Michael S. Tsirkin2745b5b2006-11-16 14:16:47 +0200862 while ((skb = __skb_dequeue(&neigh->queue))) {
863 ++priv->stats.tx_dropped;
864 dev_kfree_skb_any(skb);
865 }
Michael S. Tsirkin839fcab2007-02-05 22:12:23 +0200866 if (ipoib_cm_get(neigh))
867 ipoib_cm_destroy_tx(ipoib_cm_get(neigh));
Michael S. Tsirkind2e06552006-04-04 19:59:40 +0300868 kfree(neigh);
869}
870
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871static int ipoib_neigh_setup_dev(struct net_device *dev, struct neigh_parms *parms)
872{
Alexey Kuznetsovecbb4162007-03-24 12:52:16 -0700873 parms->neigh_cleanup = ipoib_neigh_cleanup;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874
875 return 0;
876}
877
878int ipoib_dev_init(struct net_device *dev, struct ib_device *ca, int port)
879{
880 struct ipoib_dev_priv *priv = netdev_priv(dev);
881
882 /* Allocate RX/TX "rings" to hold queued skbs */
Shirley Ma0f485252006-04-10 09:43:58 -0700883 priv->rx_ring = kzalloc(ipoib_recvq_size * sizeof *priv->rx_ring,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 GFP_KERNEL);
885 if (!priv->rx_ring) {
886 printk(KERN_WARNING "%s: failed to allocate RX ring (%d entries)\n",
Shirley Ma0f485252006-04-10 09:43:58 -0700887 ca->name, ipoib_recvq_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 goto out;
889 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890
Shirley Ma0f485252006-04-10 09:43:58 -0700891 priv->tx_ring = kzalloc(ipoib_sendq_size * sizeof *priv->tx_ring,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 GFP_KERNEL);
893 if (!priv->tx_ring) {
894 printk(KERN_WARNING "%s: failed to allocate TX ring (%d entries)\n",
Shirley Ma0f485252006-04-10 09:43:58 -0700895 ca->name, ipoib_sendq_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 goto out_rx_ring_cleanup;
897 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898
899 /* priv->tx_head & tx_tail are already 0 */
900
901 if (ipoib_ib_dev_init(dev, ca, port))
902 goto out_tx_ring_cleanup;
903
904 return 0;
905
906out_tx_ring_cleanup:
907 kfree(priv->tx_ring);
908
909out_rx_ring_cleanup:
910 kfree(priv->rx_ring);
911
912out:
913 return -ENOMEM;
914}
915
916void ipoib_dev_cleanup(struct net_device *dev)
917{
918 struct ipoib_dev_priv *priv = netdev_priv(dev), *cpriv, *tcpriv;
919
Roland Dreier1732b0e2005-11-07 10:33:11 -0800920 ipoib_delete_debug_files(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921
922 /* Delete any child interfaces first */
923 list_for_each_entry_safe(cpriv, tcpriv, &priv->child_intfs, list) {
924 unregister_netdev(cpriv->dev);
925 ipoib_dev_cleanup(cpriv->dev);
926 free_netdev(cpriv->dev);
927 }
928
929 ipoib_ib_dev_cleanup(dev);
930
Hal Rosenstock92a6b342005-08-13 20:50:27 -0700931 kfree(priv->rx_ring);
932 kfree(priv->tx_ring);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933
Hal Rosenstock92a6b342005-08-13 20:50:27 -0700934 priv->rx_ring = NULL;
935 priv->tx_ring = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936}
937
938static void ipoib_setup(struct net_device *dev)
939{
940 struct ipoib_dev_priv *priv = netdev_priv(dev);
941
942 dev->open = ipoib_open;
943 dev->stop = ipoib_stop;
944 dev->change_mtu = ipoib_change_mtu;
945 dev->hard_start_xmit = ipoib_start_xmit;
946 dev->get_stats = ipoib_get_stats;
947 dev->tx_timeout = ipoib_timeout;
948 dev->hard_header = ipoib_hard_header;
949 dev->set_multicast_list = ipoib_set_mcast_list;
950 dev->neigh_setup = ipoib_neigh_setup_dev;
Roland Dreier8d1cc862007-05-06 21:05:32 -0700951 dev->poll = ipoib_poll;
952 dev->weight = 100;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953
954 dev->watchdog_timeo = HZ;
955
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 dev->flags |= IFF_BROADCAST | IFF_MULTICAST;
957
958 /*
959 * We add in INFINIBAND_ALEN to allow for the destination
960 * address "pseudoheader" for skbs without neighbour struct.
961 */
962 dev->hard_header_len = IPOIB_ENCAP_LEN + INFINIBAND_ALEN;
963 dev->addr_len = INFINIBAND_ALEN;
964 dev->type = ARPHRD_INFINIBAND;
Shirley Ma0f485252006-04-10 09:43:58 -0700965 dev->tx_queue_len = ipoib_sendq_size * 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 dev->features = NETIF_F_VLAN_CHALLENGED | NETIF_F_LLTX;
967
968 /* MTU will be reset when mcast join happens */
969 dev->mtu = IPOIB_PACKET_SIZE - IPOIB_ENCAP_LEN;
970 priv->mcast_mtu = priv->admin_mtu = dev->mtu;
971
972 memcpy(dev->broadcast, ipv4_bcast_addr, INFINIBAND_ALEN);
973
974 netif_carrier_off(dev);
975
976 SET_MODULE_OWNER(dev);
977
978 priv->dev = dev;
979
980 spin_lock_init(&priv->lock);
981 spin_lock_init(&priv->tx_lock);
982
Ingo Molnar95ed6442006-01-13 14:51:39 -0800983 mutex_init(&priv->mcast_mutex);
984 mutex_init(&priv->vlan_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985
986 INIT_LIST_HEAD(&priv->path_list);
987 INIT_LIST_HEAD(&priv->child_intfs);
988 INIT_LIST_HEAD(&priv->dead_ahs);
989 INIT_LIST_HEAD(&priv->multicast_list);
990
Yosef Etigin26bbf132007-05-19 08:51:54 -0700991 INIT_DELAYED_WORK(&priv->pkey_poll_task, ipoib_pkey_poll);
992 INIT_WORK(&priv->pkey_event_task, ipoib_pkey_event);
David Howellsc4028952006-11-22 14:57:56 +0000993 INIT_DELAYED_WORK(&priv->mcast_task, ipoib_mcast_join_task);
994 INIT_WORK(&priv->flush_task, ipoib_ib_dev_flush);
995 INIT_WORK(&priv->restart_task, ipoib_mcast_restart_task);
996 INIT_DELAYED_WORK(&priv->ah_reap_task, ipoib_reap_ah);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997}
998
999struct ipoib_dev_priv *ipoib_intf_alloc(const char *name)
1000{
1001 struct net_device *dev;
1002
1003 dev = alloc_netdev((int) sizeof (struct ipoib_dev_priv), name,
1004 ipoib_setup);
1005 if (!dev)
1006 return NULL;
1007
1008 return netdev_priv(dev);
1009}
1010
Greg Kroah-Hartman43cb76d2002-04-09 12:14:34 -07001011static ssize_t show_pkey(struct device *dev,
1012 struct device_attribute *attr, char *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013{
Greg Kroah-Hartman43cb76d2002-04-09 12:14:34 -07001014 struct ipoib_dev_priv *priv = netdev_priv(to_net_dev(dev));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015
1016 return sprintf(buf, "0x%04x\n", priv->pkey);
1017}
Greg Kroah-Hartman43cb76d2002-04-09 12:14:34 -07001018static DEVICE_ATTR(pkey, S_IRUGO, show_pkey, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019
Greg Kroah-Hartman43cb76d2002-04-09 12:14:34 -07001020static ssize_t create_child(struct device *dev,
1021 struct device_attribute *attr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 const char *buf, size_t count)
1023{
1024 int pkey;
1025 int ret;
1026
1027 if (sscanf(buf, "%i", &pkey) != 1)
1028 return -EINVAL;
1029
1030 if (pkey < 0 || pkey > 0xffff)
1031 return -EINVAL;
1032
Roland Dreier4ce05932005-08-19 12:03:17 -07001033 /*
1034 * Set the full membership bit, so that we join the right
1035 * broadcast group, etc.
1036 */
1037 pkey |= 0x8000;
1038
Greg Kroah-Hartman43cb76d2002-04-09 12:14:34 -07001039 ret = ipoib_vlan_add(to_net_dev(dev), pkey);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040
1041 return ret ? ret : count;
1042}
Greg Kroah-Hartman43cb76d2002-04-09 12:14:34 -07001043static DEVICE_ATTR(create_child, S_IWUGO, NULL, create_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044
Greg Kroah-Hartman43cb76d2002-04-09 12:14:34 -07001045static ssize_t delete_child(struct device *dev,
1046 struct device_attribute *attr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 const char *buf, size_t count)
1048{
1049 int pkey;
1050 int ret;
1051
1052 if (sscanf(buf, "%i", &pkey) != 1)
1053 return -EINVAL;
1054
1055 if (pkey < 0 || pkey > 0xffff)
1056 return -EINVAL;
1057
Greg Kroah-Hartman43cb76d2002-04-09 12:14:34 -07001058 ret = ipoib_vlan_delete(to_net_dev(dev), pkey);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059
1060 return ret ? ret : count;
1061
1062}
Greg Kroah-Hartman43cb76d2002-04-09 12:14:34 -07001063static DEVICE_ATTR(delete_child, S_IWUGO, NULL, delete_child);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064
1065int ipoib_add_pkey_attr(struct net_device *dev)
1066{
Greg Kroah-Hartman43cb76d2002-04-09 12:14:34 -07001067 return device_create_file(&dev->dev, &dev_attr_pkey);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068}
1069
1070static struct net_device *ipoib_add_port(const char *format,
1071 struct ib_device *hca, u8 port)
1072{
1073 struct ipoib_dev_priv *priv;
1074 int result = -ENOMEM;
1075
1076 priv = ipoib_intf_alloc(format);
1077 if (!priv)
1078 goto alloc_mem_failed;
1079
1080 SET_NETDEV_DEV(priv->dev, hca->dma_device);
1081
1082 result = ib_query_pkey(hca, port, 0, &priv->pkey);
1083 if (result) {
1084 printk(KERN_WARNING "%s: ib_query_pkey port %d failed (ret = %d)\n",
1085 hca->name, port, result);
Eli Cohenca6de172007-08-21 18:46:10 +03001086 goto device_init_failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 }
1088
Roland Dreier4ce05932005-08-19 12:03:17 -07001089 /*
1090 * Set the full membership bit, so that we join the right
1091 * broadcast group, etc.
1092 */
1093 priv->pkey |= 0x8000;
1094
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095 priv->dev->broadcast[8] = priv->pkey >> 8;
1096 priv->dev->broadcast[9] = priv->pkey & 0xff;
1097
1098 result = ib_query_gid(hca, port, 0, &priv->local_gid);
1099 if (result) {
1100 printk(KERN_WARNING "%s: ib_query_gid port %d failed (ret = %d)\n",
1101 hca->name, port, result);
Eli Cohenca6de172007-08-21 18:46:10 +03001102 goto device_init_failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 } else
1104 memcpy(priv->dev->dev_addr + 4, priv->local_gid.raw, sizeof (union ib_gid));
1105
1106
1107 result = ipoib_dev_init(priv->dev, hca, port);
1108 if (result < 0) {
1109 printk(KERN_WARNING "%s: failed to initialize port %d (ret = %d)\n",
1110 hca->name, port, result);
1111 goto device_init_failed;
1112 }
1113
1114 INIT_IB_EVENT_HANDLER(&priv->event_handler,
1115 priv->ca, ipoib_event);
1116 result = ib_register_event_handler(&priv->event_handler);
1117 if (result < 0) {
1118 printk(KERN_WARNING "%s: ib_register_event_handler failed for "
1119 "port %d (ret = %d)\n",
1120 hca->name, port, result);
1121 goto event_failed;
1122 }
1123
1124 result = register_netdev(priv->dev);
1125 if (result) {
1126 printk(KERN_WARNING "%s: couldn't register ipoib port %d; error %d\n",
1127 hca->name, port, result);
1128 goto register_failed;
1129 }
1130
Roland Dreier1732b0e2005-11-07 10:33:11 -08001131 ipoib_create_debug_files(priv->dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132
Michael S. Tsirkin839fcab2007-02-05 22:12:23 +02001133 if (ipoib_cm_add_mode_attr(priv->dev))
1134 goto sysfs_failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 if (ipoib_add_pkey_attr(priv->dev))
1136 goto sysfs_failed;
Greg Kroah-Hartman43cb76d2002-04-09 12:14:34 -07001137 if (device_create_file(&priv->dev->dev, &dev_attr_create_child))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 goto sysfs_failed;
Greg Kroah-Hartman43cb76d2002-04-09 12:14:34 -07001139 if (device_create_file(&priv->dev->dev, &dev_attr_delete_child))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 goto sysfs_failed;
1141
1142 return priv->dev;
1143
1144sysfs_failed:
Roland Dreier1732b0e2005-11-07 10:33:11 -08001145 ipoib_delete_debug_files(priv->dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 unregister_netdev(priv->dev);
1147
1148register_failed:
1149 ib_unregister_event_handler(&priv->event_handler);
Michael S. Tsirkin51574e02005-09-12 09:52:28 -07001150 flush_scheduled_work();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151
1152event_failed:
1153 ipoib_dev_cleanup(priv->dev);
1154
1155device_init_failed:
1156 free_netdev(priv->dev);
1157
1158alloc_mem_failed:
1159 return ERR_PTR(result);
1160}
1161
1162static void ipoib_add_one(struct ib_device *device)
1163{
1164 struct list_head *dev_list;
1165 struct net_device *dev;
1166 struct ipoib_dev_priv *priv;
1167 int s, e, p;
1168
Tom Tucker07ebafb2006-08-03 16:02:42 -05001169 if (rdma_node_get_transport(device->node_type) != RDMA_TRANSPORT_IB)
1170 return;
1171
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 dev_list = kmalloc(sizeof *dev_list, GFP_KERNEL);
1173 if (!dev_list)
1174 return;
1175
1176 INIT_LIST_HEAD(dev_list);
1177
Tom Tucker07ebafb2006-08-03 16:02:42 -05001178 if (device->node_type == RDMA_NODE_IB_SWITCH) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 s = 0;
1180 e = 0;
1181 } else {
1182 s = 1;
1183 e = device->phys_port_cnt;
1184 }
1185
1186 for (p = s; p <= e; ++p) {
1187 dev = ipoib_add_port("ib%d", device, p);
1188 if (!IS_ERR(dev)) {
1189 priv = netdev_priv(dev);
1190 list_add_tail(&priv->list, dev_list);
1191 }
1192 }
1193
1194 ib_set_client_data(device, &ipoib_client, dev_list);
1195}
1196
1197static void ipoib_remove_one(struct ib_device *device)
1198{
1199 struct ipoib_dev_priv *priv, *tmp;
1200 struct list_head *dev_list;
1201
Tom Tucker07ebafb2006-08-03 16:02:42 -05001202 if (rdma_node_get_transport(device->node_type) != RDMA_TRANSPORT_IB)
1203 return;
1204
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 dev_list = ib_get_client_data(device, &ipoib_client);
1206
1207 list_for_each_entry_safe(priv, tmp, dev_list, list) {
1208 ib_unregister_event_handler(&priv->event_handler);
Michael S. Tsirkin51574e02005-09-12 09:52:28 -07001209 flush_scheduled_work();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210
1211 unregister_netdev(priv->dev);
1212 ipoib_dev_cleanup(priv->dev);
1213 free_netdev(priv->dev);
1214 }
Michael S. Tsirkin06c56e42005-09-01 09:19:02 -07001215
1216 kfree(dev_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217}
1218
1219static int __init ipoib_init_module(void)
1220{
1221 int ret;
1222
Shirley Ma0f485252006-04-10 09:43:58 -07001223 ipoib_recvq_size = roundup_pow_of_two(ipoib_recvq_size);
1224 ipoib_recvq_size = min(ipoib_recvq_size, IPOIB_MAX_QUEUE_SIZE);
1225 ipoib_recvq_size = max(ipoib_recvq_size, IPOIB_MIN_QUEUE_SIZE);
1226
1227 ipoib_sendq_size = roundup_pow_of_two(ipoib_sendq_size);
1228 ipoib_sendq_size = min(ipoib_sendq_size, IPOIB_MAX_QUEUE_SIZE);
1229 ipoib_sendq_size = max(ipoib_sendq_size, IPOIB_MIN_QUEUE_SIZE);
1230
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 ret = ipoib_register_debugfs();
1232 if (ret)
1233 return ret;
1234
1235 /*
1236 * We create our own workqueue mainly because we want to be
1237 * able to flush it when devices are being removed. We can't
1238 * use schedule_work()/flush_scheduled_work() because both
1239 * unregister_netdev() and linkwatch_event take the rtnl lock,
1240 * so flush_scheduled_work() can deadlock during device
1241 * removal.
1242 */
1243 ipoib_workqueue = create_singlethread_workqueue("ipoib");
1244 if (!ipoib_workqueue) {
1245 ret = -ENOMEM;
1246 goto err_fs;
1247 }
1248
Michael S. Tsirkinc1a0b232006-08-21 16:40:12 -07001249 ib_sa_register_client(&ipoib_sa_client);
1250
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 ret = ib_register_client(&ipoib_client);
1252 if (ret)
Michael S. Tsirkinc1a0b232006-08-21 16:40:12 -07001253 goto err_sa;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254
1255 return 0;
1256
Michael S. Tsirkinc1a0b232006-08-21 16:40:12 -07001257err_sa:
1258 ib_sa_unregister_client(&ipoib_sa_client);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 destroy_workqueue(ipoib_workqueue);
1260
Roland Dreier9adec1a2005-04-16 15:26:07 -07001261err_fs:
1262 ipoib_unregister_debugfs();
1263
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 return ret;
1265}
1266
1267static void __exit ipoib_cleanup_module(void)
1268{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 ib_unregister_client(&ipoib_client);
Michael S. Tsirkinc1a0b232006-08-21 16:40:12 -07001270 ib_sa_unregister_client(&ipoib_sa_client);
Roland Dreier9adec1a2005-04-16 15:26:07 -07001271 ipoib_unregister_debugfs();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 destroy_workqueue(ipoib_workqueue);
1273}
1274
1275module_init(ipoib_init_module);
1276module_exit(ipoib_cleanup_module);