Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (c) 2004 Topspin Communications. All rights reserved. | |
2a1d9b7f RD |
3 | * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved. |
4 | * Copyright (c) 2004 Voltaire, Inc. All rights reserved. | |
1da177e4 LT |
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. | |
1da177e4 LT |
33 | */ |
34 | ||
35 | #include "ipoib.h" | |
36 | ||
1da177e4 LT |
37 | #include <linux/module.h> |
38 | ||
39 | #include <linux/init.h> | |
40 | #include <linux/slab.h> | |
0f485251 | 41 | #include <linux/kernel.h> |
10313cbb | 42 | #include <linux/vmalloc.h> |
1da177e4 LT |
43 | |
44 | #include <linux/if_arp.h> /* For ARPHRD_xxx */ | |
45 | ||
46 | #include <linux/ip.h> | |
47 | #include <linux/in.h> | |
48 | ||
14c85021 ACM |
49 | #include <net/dst.h> |
50 | ||
1da177e4 LT |
51 | MODULE_AUTHOR("Roland Dreier"); |
52 | MODULE_DESCRIPTION("IP-over-InfiniBand net driver"); | |
53 | MODULE_LICENSE("Dual BSD/GPL"); | |
54 | ||
0f485251 SM |
55 | int ipoib_sendq_size __read_mostly = IPOIB_TX_RING_SIZE; |
56 | int ipoib_recvq_size __read_mostly = IPOIB_RX_RING_SIZE; | |
57 | ||
58 | module_param_named(send_queue_size, ipoib_sendq_size, int, 0444); | |
59 | MODULE_PARM_DESC(send_queue_size, "Number of descriptors in send queue"); | |
60 | module_param_named(recv_queue_size, ipoib_recvq_size, int, 0444); | |
61 | MODULE_PARM_DESC(recv_queue_size, "Number of descriptors in receive queue"); | |
62 | ||
af40da89 VS |
63 | static int lro; |
64 | module_param(lro, bool, 0444); | |
65 | MODULE_PARM_DESC(lro, "Enable LRO (Large Receive Offload)"); | |
66 | ||
67 | static int lro_max_aggr = IPOIB_LRO_MAX_AGGR; | |
68 | module_param(lro_max_aggr, int, 0644); | |
69 | MODULE_PARM_DESC(lro_max_aggr, "LRO: Max packets to be aggregated " | |
70 | "(default = 64)"); | |
71 | ||
1da177e4 LT |
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 | ||
1732b0ef RD |
79 | struct ipoib_path_iter { |
80 | struct net_device *dev; | |
81 | struct ipoib_path path; | |
82 | }; | |
83 | ||
1da177e4 LT |
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 | ||
c1a0b23b MT |
92 | struct ib_sa_client ipoib_sa_client; |
93 | ||
1da177e4 LT |
94 | static void ipoib_add_one(struct ib_device *device); |
95 | static void ipoib_remove_one(struct ib_device *device); | |
96 | ||
97 | static struct ib_client ipoib_client = { | |
98 | .name = "ipoib", | |
99 | .add = ipoib_add_one, | |
100 | .remove = ipoib_remove_one | |
101 | }; | |
102 | ||
103 | int ipoib_open(struct net_device *dev) | |
104 | { | |
105 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
106 | ||
107 | ipoib_dbg(priv, "bringing up interface\n"); | |
108 | ||
bea3348e | 109 | napi_enable(&priv->napi); |
1da177e4 LT |
110 | set_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags); |
111 | ||
112 | if (ipoib_pkey_dev_delay_open(dev)) | |
113 | return 0; | |
114 | ||
bea3348e SH |
115 | if (ipoib_ib_dev_open(dev)) { |
116 | napi_disable(&priv->napi); | |
1da177e4 | 117 | return -EINVAL; |
bea3348e | 118 | } |
1da177e4 | 119 | |
267ee88e | 120 | if (ipoib_ib_dev_up(dev)) { |
26bbf13c | 121 | ipoib_ib_dev_stop(dev, 1); |
bea3348e | 122 | napi_disable(&priv->napi); |
1da177e4 | 123 | return -EINVAL; |
267ee88e | 124 | } |
1da177e4 LT |
125 | |
126 | if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) { | |
127 | struct ipoib_dev_priv *cpriv; | |
128 | ||
129 | /* Bring up any child interfaces too */ | |
95ed644f | 130 | mutex_lock(&priv->vlan_mutex); |
1da177e4 LT |
131 | list_for_each_entry(cpriv, &priv->child_intfs, list) { |
132 | int flags; | |
133 | ||
134 | flags = cpriv->dev->flags; | |
135 | if (flags & IFF_UP) | |
136 | continue; | |
137 | ||
138 | dev_change_flags(cpriv->dev, flags | IFF_UP); | |
139 | } | |
95ed644f | 140 | mutex_unlock(&priv->vlan_mutex); |
1da177e4 LT |
141 | } |
142 | ||
143 | netif_start_queue(dev); | |
144 | ||
145 | return 0; | |
146 | } | |
147 | ||
148 | static int ipoib_stop(struct net_device *dev) | |
149 | { | |
150 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
151 | ||
152 | ipoib_dbg(priv, "stopping interface\n"); | |
153 | ||
154 | clear_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags); | |
bea3348e | 155 | napi_disable(&priv->napi); |
1da177e4 LT |
156 | |
157 | netif_stop_queue(dev); | |
158 | ||
a77a57a1 RD |
159 | ipoib_ib_dev_down(dev, 0); |
160 | ipoib_ib_dev_stop(dev, 0); | |
1da177e4 LT |
161 | |
162 | if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) { | |
163 | struct ipoib_dev_priv *cpriv; | |
164 | ||
165 | /* Bring down any child interfaces too */ | |
95ed644f | 166 | mutex_lock(&priv->vlan_mutex); |
1da177e4 LT |
167 | list_for_each_entry(cpriv, &priv->child_intfs, list) { |
168 | int flags; | |
169 | ||
170 | flags = cpriv->dev->flags; | |
171 | if (!(flags & IFF_UP)) | |
172 | continue; | |
173 | ||
174 | dev_change_flags(cpriv->dev, flags & ~IFF_UP); | |
175 | } | |
95ed644f | 176 | mutex_unlock(&priv->vlan_mutex); |
1da177e4 LT |
177 | } |
178 | ||
179 | return 0; | |
180 | } | |
181 | ||
182 | static int ipoib_change_mtu(struct net_device *dev, int new_mtu) | |
183 | { | |
184 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
185 | ||
839fcaba | 186 | /* dev->mtu > 2K ==> connected mode */ |
586a6934 PS |
187 | if (ipoib_cm_admin_enabled(dev)) { |
188 | if (new_mtu > ipoib_cm_max_mtu(dev)) | |
189 | return -EINVAL; | |
190 | ||
839fcaba MT |
191 | if (new_mtu > priv->mcast_mtu) |
192 | ipoib_warn(priv, "mtu > %d will cause multicast packet drops.\n", | |
193 | priv->mcast_mtu); | |
586a6934 | 194 | |
839fcaba MT |
195 | dev->mtu = new_mtu; |
196 | return 0; | |
197 | } | |
198 | ||
bc7b3a36 | 199 | if (new_mtu > IPOIB_UD_MTU(priv->max_ib_mtu)) |
1da177e4 LT |
200 | return -EINVAL; |
201 | ||
202 | priv->admin_mtu = new_mtu; | |
203 | ||
204 | dev->mtu = min(priv->mcast_mtu, priv->admin_mtu); | |
205 | ||
206 | return 0; | |
207 | } | |
208 | ||
37c22a77 | 209 | static struct ipoib_path *__path_find(struct net_device *dev, void *gid) |
1da177e4 LT |
210 | { |
211 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
212 | struct rb_node *n = priv->path_tree.rb_node; | |
213 | struct ipoib_path *path; | |
214 | int ret; | |
215 | ||
216 | while (n) { | |
217 | path = rb_entry(n, struct ipoib_path, rb_node); | |
218 | ||
37c22a77 | 219 | ret = memcmp(gid, path->pathrec.dgid.raw, |
1da177e4 LT |
220 | sizeof (union ib_gid)); |
221 | ||
222 | if (ret < 0) | |
223 | n = n->rb_left; | |
224 | else if (ret > 0) | |
225 | n = n->rb_right; | |
226 | else | |
227 | return path; | |
228 | } | |
229 | ||
230 | return NULL; | |
231 | } | |
232 | ||
233 | static int __path_add(struct net_device *dev, struct ipoib_path *path) | |
234 | { | |
235 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
236 | struct rb_node **n = &priv->path_tree.rb_node; | |
237 | struct rb_node *pn = NULL; | |
238 | struct ipoib_path *tpath; | |
239 | int ret; | |
240 | ||
241 | while (*n) { | |
242 | pn = *n; | |
243 | tpath = rb_entry(pn, struct ipoib_path, rb_node); | |
244 | ||
245 | ret = memcmp(path->pathrec.dgid.raw, tpath->pathrec.dgid.raw, | |
246 | sizeof (union ib_gid)); | |
247 | if (ret < 0) | |
248 | n = &pn->rb_left; | |
249 | else if (ret > 0) | |
250 | n = &pn->rb_right; | |
251 | else | |
252 | return -EEXIST; | |
253 | } | |
254 | ||
255 | rb_link_node(&path->rb_node, pn, n); | |
256 | rb_insert_color(&path->rb_node, &priv->path_tree); | |
257 | ||
258 | list_add_tail(&path->list, &priv->path_list); | |
259 | ||
260 | return 0; | |
261 | } | |
262 | ||
263 | static void path_free(struct net_device *dev, struct ipoib_path *path) | |
264 | { | |
265 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
266 | struct ipoib_neigh *neigh, *tn; | |
267 | struct sk_buff *skb; | |
268 | unsigned long flags; | |
269 | ||
270 | while ((skb = __skb_dequeue(&path->queue))) | |
271 | dev_kfree_skb_irq(skb); | |
272 | ||
273 | spin_lock_irqsave(&priv->lock, flags); | |
274 | ||
275 | list_for_each_entry_safe(neigh, tn, &path->neigh_list, list) { | |
276 | /* | |
277 | * It's safe to call ipoib_put_ah() inside priv->lock | |
278 | * here, because we know that path->ah will always | |
279 | * hold one more reference, so ipoib_put_ah() will | |
280 | * never do more than decrement the ref count. | |
281 | */ | |
282 | if (neigh->ah) | |
283 | ipoib_put_ah(neigh->ah); | |
d2e0655e | 284 | |
2745b5b7 | 285 | ipoib_neigh_free(dev, neigh); |
1da177e4 LT |
286 | } |
287 | ||
288 | spin_unlock_irqrestore(&priv->lock, flags); | |
289 | ||
290 | if (path->ah) | |
291 | ipoib_put_ah(path->ah); | |
292 | ||
293 | kfree(path); | |
294 | } | |
295 | ||
1732b0ef RD |
296 | #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG |
297 | ||
298 | struct ipoib_path_iter *ipoib_path_iter_init(struct net_device *dev) | |
299 | { | |
300 | struct ipoib_path_iter *iter; | |
301 | ||
302 | iter = kmalloc(sizeof *iter, GFP_KERNEL); | |
303 | if (!iter) | |
304 | return NULL; | |
305 | ||
306 | iter->dev = dev; | |
307 | memset(iter->path.pathrec.dgid.raw, 0, 16); | |
308 | ||
309 | if (ipoib_path_iter_next(iter)) { | |
310 | kfree(iter); | |
311 | return NULL; | |
312 | } | |
313 | ||
314 | return iter; | |
315 | } | |
316 | ||
317 | int ipoib_path_iter_next(struct ipoib_path_iter *iter) | |
318 | { | |
319 | struct ipoib_dev_priv *priv = netdev_priv(iter->dev); | |
320 | struct rb_node *n; | |
321 | struct ipoib_path *path; | |
322 | int ret = 1; | |
323 | ||
324 | spin_lock_irq(&priv->lock); | |
325 | ||
326 | n = rb_first(&priv->path_tree); | |
327 | ||
328 | while (n) { | |
329 | path = rb_entry(n, struct ipoib_path, rb_node); | |
330 | ||
331 | if (memcmp(iter->path.pathrec.dgid.raw, path->pathrec.dgid.raw, | |
332 | sizeof (union ib_gid)) < 0) { | |
333 | iter->path = *path; | |
334 | ret = 0; | |
335 | break; | |
336 | } | |
337 | ||
338 | n = rb_next(n); | |
339 | } | |
340 | ||
341 | spin_unlock_irq(&priv->lock); | |
342 | ||
343 | return ret; | |
344 | } | |
345 | ||
346 | void ipoib_path_iter_read(struct ipoib_path_iter *iter, | |
347 | struct ipoib_path *path) | |
348 | { | |
349 | *path = iter->path; | |
350 | } | |
351 | ||
352 | #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */ | |
353 | ||
ee1e2c82 MS |
354 | void ipoib_mark_paths_invalid(struct net_device *dev) |
355 | { | |
356 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
357 | struct ipoib_path *path, *tp; | |
358 | ||
359 | spin_lock_irq(&priv->lock); | |
360 | ||
361 | list_for_each_entry_safe(path, tp, &priv->path_list, list) { | |
fcace2fe | 362 | ipoib_dbg(priv, "mark path LID 0x%04x GID %p6 invalid\n", |
ee1e2c82 | 363 | be16_to_cpu(path->pathrec.dlid), |
fcace2fe | 364 | path->pathrec.dgid.raw); |
ee1e2c82 MS |
365 | path->valid = 0; |
366 | } | |
367 | ||
368 | spin_unlock_irq(&priv->lock); | |
369 | } | |
370 | ||
1da177e4 LT |
371 | void ipoib_flush_paths(struct net_device *dev) |
372 | { | |
373 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
374 | struct ipoib_path *path, *tp; | |
375 | LIST_HEAD(remove_list); | |
943c246e | 376 | unsigned long flags; |
1da177e4 | 377 | |
943c246e RD |
378 | netif_tx_lock_bh(dev); |
379 | spin_lock_irqsave(&priv->lock, flags); | |
1da177e4 | 380 | |
157de229 | 381 | list_splice_init(&priv->path_list, &remove_list); |
1da177e4 LT |
382 | |
383 | list_for_each_entry(path, &remove_list, list) | |
384 | rb_erase(&path->rb_node, &priv->path_tree); | |
385 | ||
1da177e4 LT |
386 | list_for_each_entry_safe(path, tp, &remove_list, list) { |
387 | if (path->query) | |
388 | ib_sa_cancel_query(path->query_id, path->query); | |
943c246e RD |
389 | spin_unlock_irqrestore(&priv->lock, flags); |
390 | netif_tx_unlock_bh(dev); | |
1da177e4 LT |
391 | wait_for_completion(&path->done); |
392 | path_free(dev, path); | |
943c246e RD |
393 | netif_tx_lock_bh(dev); |
394 | spin_lock_irqsave(&priv->lock, flags); | |
1da177e4 | 395 | } |
943c246e RD |
396 | |
397 | spin_unlock_irqrestore(&priv->lock, flags); | |
398 | netif_tx_unlock_bh(dev); | |
1da177e4 LT |
399 | } |
400 | ||
401 | static void path_rec_completion(int status, | |
402 | struct ib_sa_path_rec *pathrec, | |
403 | void *path_ptr) | |
404 | { | |
405 | struct ipoib_path *path = path_ptr; | |
406 | struct net_device *dev = path->dev; | |
407 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
408 | struct ipoib_ah *ah = NULL; | |
c9da4bad | 409 | struct ipoib_ah *old_ah = NULL; |
d04d01b1 | 410 | struct ipoib_neigh *neigh, *tn; |
1da177e4 LT |
411 | struct sk_buff_head skqueue; |
412 | struct sk_buff *skb; | |
413 | unsigned long flags; | |
414 | ||
843613b0 | 415 | if (!status) |
fcace2fe HH |
416 | ipoib_dbg(priv, "PathRec LID 0x%04x for GID %p6\n", |
417 | be16_to_cpu(pathrec->dlid), pathrec->dgid.raw); | |
1da177e4 | 418 | else |
fcace2fe HH |
419 | ipoib_dbg(priv, "PathRec status %d for GID %p6\n", |
420 | status, path->pathrec.dgid.raw); | |
1da177e4 LT |
421 | |
422 | skb_queue_head_init(&skqueue); | |
423 | ||
424 | if (!status) { | |
46f1b3d7 SH |
425 | struct ib_ah_attr av; |
426 | ||
427 | if (!ib_init_ah_from_path(priv->ca, priv->port, pathrec, &av)) | |
428 | ah = ipoib_create_ah(dev, priv->pd, &av); | |
1da177e4 LT |
429 | } |
430 | ||
431 | spin_lock_irqsave(&priv->lock, flags); | |
432 | ||
1da177e4 LT |
433 | if (ah) { |
434 | path->pathrec = *pathrec; | |
435 | ||
c9da4bad RD |
436 | old_ah = path->ah; |
437 | path->ah = ah; | |
438 | ||
1da177e4 LT |
439 | ipoib_dbg(priv, "created address handle %p for LID 0x%04x, SL %d\n", |
440 | ah, be16_to_cpu(pathrec->dlid), pathrec->sl); | |
441 | ||
442 | while ((skb = __skb_dequeue(&path->queue))) | |
443 | __skb_queue_tail(&skqueue, skb); | |
444 | ||
d04d01b1 | 445 | list_for_each_entry_safe(neigh, tn, &path->neigh_list, list) { |
ee1e2c82 MS |
446 | if (neigh->ah) { |
447 | WARN_ON(neigh->ah != old_ah); | |
448 | /* | |
449 | * Dropping the ah reference inside | |
450 | * priv->lock is safe here, because we | |
451 | * will hold one more reference from | |
452 | * the original value of path->ah (ie | |
453 | * old_ah). | |
454 | */ | |
455 | ipoib_put_ah(neigh->ah); | |
456 | } | |
1da177e4 LT |
457 | kref_get(&path->ah->ref); |
458 | neigh->ah = path->ah; | |
8a7f7521 MT |
459 | memcpy(&neigh->dgid.raw, &path->pathrec.dgid.raw, |
460 | sizeof(union ib_gid)); | |
1da177e4 | 461 | |
839fcaba MT |
462 | if (ipoib_cm_enabled(dev, neigh->neighbour)) { |
463 | if (!ipoib_cm_get(neigh)) | |
464 | ipoib_cm_set(neigh, ipoib_cm_create_tx(dev, | |
465 | path, | |
466 | neigh)); | |
467 | if (!ipoib_cm_get(neigh)) { | |
468 | list_del(&neigh->list); | |
469 | if (neigh->ah) | |
470 | ipoib_put_ah(neigh->ah); | |
471 | ipoib_neigh_free(dev, neigh); | |
472 | continue; | |
473 | } | |
474 | } | |
475 | ||
1da177e4 LT |
476 | while ((skb = __skb_dequeue(&neigh->queue))) |
477 | __skb_queue_tail(&skqueue, skb); | |
478 | } | |
ee1e2c82 | 479 | path->valid = 1; |
5872a9fc | 480 | } |
1da177e4 | 481 | |
5872a9fc | 482 | path->query = NULL; |
1da177e4 LT |
483 | complete(&path->done); |
484 | ||
485 | spin_unlock_irqrestore(&priv->lock, flags); | |
486 | ||
ee1e2c82 MS |
487 | if (old_ah) |
488 | ipoib_put_ah(old_ah); | |
489 | ||
1da177e4 LT |
490 | while ((skb = __skb_dequeue(&skqueue))) { |
491 | skb->dev = dev; | |
492 | if (dev_queue_xmit(skb)) | |
493 | ipoib_warn(priv, "dev_queue_xmit failed " | |
494 | "to requeue packet\n"); | |
495 | } | |
496 | } | |
497 | ||
37c22a77 | 498 | static struct ipoib_path *path_rec_create(struct net_device *dev, void *gid) |
1da177e4 LT |
499 | { |
500 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
501 | struct ipoib_path *path; | |
502 | ||
1401b53a JM |
503 | if (!priv->broadcast) |
504 | return NULL; | |
505 | ||
21a38489 | 506 | path = kzalloc(sizeof *path, GFP_ATOMIC); |
1da177e4 LT |
507 | if (!path) |
508 | return NULL; | |
509 | ||
21a38489 | 510 | path->dev = dev; |
1da177e4 LT |
511 | |
512 | skb_queue_head_init(&path->queue); | |
513 | ||
514 | INIT_LIST_HEAD(&path->neigh_list); | |
1da177e4 | 515 | |
37c22a77 | 516 | memcpy(path->pathrec.dgid.raw, gid, sizeof (union ib_gid)); |
2337f809 RD |
517 | path->pathrec.sgid = priv->local_gid; |
518 | path->pathrec.pkey = cpu_to_be16(priv->pkey); | |
81668838 SH |
519 | path->pathrec.numb_path = 1; |
520 | path->pathrec.traffic_class = priv->broadcast->mcmember.traffic_class; | |
1da177e4 LT |
521 | |
522 | return path; | |
523 | } | |
524 | ||
525 | static int path_rec_start(struct net_device *dev, | |
526 | struct ipoib_path *path) | |
527 | { | |
528 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
529 | ||
fcace2fe HH |
530 | ipoib_dbg(priv, "Start path record lookup for %p6\n", |
531 | path->pathrec.dgid.raw); | |
1da177e4 | 532 | |
65c7edda RD |
533 | init_completion(&path->done); |
534 | ||
1da177e4 | 535 | path->query_id = |
c1a0b23b | 536 | ib_sa_path_rec_get(&ipoib_sa_client, priv->ca, priv->port, |
1da177e4 LT |
537 | &path->pathrec, |
538 | IB_SA_PATH_REC_DGID | | |
539 | IB_SA_PATH_REC_SGID | | |
540 | IB_SA_PATH_REC_NUMB_PATH | | |
81668838 | 541 | IB_SA_PATH_REC_TRAFFIC_CLASS | |
1da177e4 LT |
542 | IB_SA_PATH_REC_PKEY, |
543 | 1000, GFP_ATOMIC, | |
544 | path_rec_completion, | |
545 | path, &path->query); | |
546 | if (path->query_id < 0) { | |
01b3fc8b | 547 | ipoib_warn(priv, "ib_sa_path_rec_get failed: %d\n", path->query_id); |
1da177e4 LT |
548 | path->query = NULL; |
549 | return path->query_id; | |
550 | } | |
551 | ||
552 | return 0; | |
553 | } | |
554 | ||
555 | static void neigh_add_path(struct sk_buff *skb, struct net_device *dev) | |
556 | { | |
557 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
558 | struct ipoib_path *path; | |
559 | struct ipoib_neigh *neigh; | |
943c246e | 560 | unsigned long flags; |
1da177e4 | 561 | |
732a2170 | 562 | neigh = ipoib_neigh_alloc(skb->dst->neighbour, skb->dev); |
1da177e4 | 563 | if (!neigh) { |
de903512 | 564 | ++dev->stats.tx_dropped; |
1da177e4 LT |
565 | dev_kfree_skb_any(skb); |
566 | return; | |
567 | } | |
568 | ||
943c246e | 569 | spin_lock_irqsave(&priv->lock, flags); |
1da177e4 | 570 | |
37c22a77 | 571 | path = __path_find(dev, skb->dst->neighbour->ha + 4); |
1da177e4 | 572 | if (!path) { |
37c22a77 | 573 | path = path_rec_create(dev, skb->dst->neighbour->ha + 4); |
1da177e4 | 574 | if (!path) |
d2e0655e | 575 | goto err_path; |
1da177e4 LT |
576 | |
577 | __path_add(dev, path); | |
578 | } | |
579 | ||
580 | list_add_tail(&neigh->list, &path->neigh_list); | |
581 | ||
47f7a071 | 582 | if (path->ah) { |
1da177e4 LT |
583 | kref_get(&path->ah->ref); |
584 | neigh->ah = path->ah; | |
8a7f7521 MT |
585 | memcpy(&neigh->dgid.raw, &path->pathrec.dgid.raw, |
586 | sizeof(union ib_gid)); | |
1da177e4 | 587 | |
839fcaba MT |
588 | if (ipoib_cm_enabled(dev, neigh->neighbour)) { |
589 | if (!ipoib_cm_get(neigh)) | |
590 | ipoib_cm_set(neigh, ipoib_cm_create_tx(dev, path, neigh)); | |
591 | if (!ipoib_cm_get(neigh)) { | |
592 | list_del(&neigh->list); | |
593 | if (neigh->ah) | |
594 | ipoib_put_ah(neigh->ah); | |
595 | ipoib_neigh_free(dev, neigh); | |
596 | goto err_drop; | |
597 | } | |
598 | if (skb_queue_len(&neigh->queue) < IPOIB_MAX_PATH_REC_QUEUE) | |
599 | __skb_queue_tail(&neigh->queue, skb); | |
600 | else { | |
601 | ipoib_warn(priv, "queue length limit %d. Packet drop.\n", | |
602 | skb_queue_len(&neigh->queue)); | |
603 | goto err_drop; | |
604 | } | |
605 | } else | |
606 | ipoib_send(dev, skb, path->ah, IPOIB_QPN(skb->dst->neighbour->ha)); | |
1da177e4 LT |
607 | } else { |
608 | neigh->ah = NULL; | |
1da177e4 LT |
609 | |
610 | if (!path->query && path_rec_start(dev, path)) | |
d2e0655e | 611 | goto err_list; |
2745b5b7 MT |
612 | |
613 | __skb_queue_tail(&neigh->queue, skb); | |
1da177e4 LT |
614 | } |
615 | ||
943c246e | 616 | spin_unlock_irqrestore(&priv->lock, flags); |
1da177e4 LT |
617 | return; |
618 | ||
d2e0655e | 619 | err_list: |
1da177e4 | 620 | list_del(&neigh->list); |
1da177e4 | 621 | |
d2e0655e | 622 | err_path: |
2745b5b7 | 623 | ipoib_neigh_free(dev, neigh); |
839fcaba | 624 | err_drop: |
de903512 | 625 | ++dev->stats.tx_dropped; |
1da177e4 LT |
626 | dev_kfree_skb_any(skb); |
627 | ||
943c246e | 628 | spin_unlock_irqrestore(&priv->lock, flags); |
1da177e4 LT |
629 | } |
630 | ||
d70ed607 | 631 | static void ipoib_path_lookup(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
632 | { |
633 | struct ipoib_dev_priv *priv = netdev_priv(skb->dev); | |
634 | ||
635 | /* Look up path record for unicasts */ | |
636 | if (skb->dst->neighbour->ha[4] != 0xff) { | |
637 | neigh_add_path(skb, dev); | |
638 | return; | |
639 | } | |
640 | ||
641 | /* Add in the P_Key for multicasts */ | |
642 | skb->dst->neighbour->ha[8] = (priv->pkey >> 8) & 0xff; | |
643 | skb->dst->neighbour->ha[9] = priv->pkey & 0xff; | |
37c22a77 | 644 | ipoib_mcast_send(dev, skb->dst->neighbour->ha + 4, skb); |
1da177e4 LT |
645 | } |
646 | ||
647 | static void unicast_arp_send(struct sk_buff *skb, struct net_device *dev, | |
648 | struct ipoib_pseudoheader *phdr) | |
649 | { | |
650 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
651 | struct ipoib_path *path; | |
943c246e | 652 | unsigned long flags; |
1da177e4 | 653 | |
943c246e | 654 | spin_lock_irqsave(&priv->lock, flags); |
1da177e4 | 655 | |
37c22a77 | 656 | path = __path_find(dev, phdr->hwaddr + 4); |
ee1e2c82 MS |
657 | if (!path || !path->valid) { |
658 | if (!path) | |
659 | path = path_rec_create(dev, phdr->hwaddr + 4); | |
1da177e4 LT |
660 | if (path) { |
661 | /* put pseudoheader back on for next time */ | |
662 | skb_push(skb, sizeof *phdr); | |
663 | __skb_queue_tail(&path->queue, skb); | |
664 | ||
665 | if (path_rec_start(dev, path)) { | |
943c246e | 666 | spin_unlock_irqrestore(&priv->lock, flags); |
1da177e4 LT |
667 | path_free(dev, path); |
668 | return; | |
669 | } else | |
670 | __path_add(dev, path); | |
671 | } else { | |
de903512 | 672 | ++dev->stats.tx_dropped; |
1da177e4 LT |
673 | dev_kfree_skb_any(skb); |
674 | } | |
675 | ||
943c246e | 676 | spin_unlock_irqrestore(&priv->lock, flags); |
1da177e4 LT |
677 | return; |
678 | } | |
679 | ||
47f7a071 | 680 | if (path->ah) { |
1da177e4 LT |
681 | ipoib_dbg(priv, "Send unicast ARP to %04x\n", |
682 | be16_to_cpu(path->pathrec.dlid)); | |
683 | ||
073ae841 | 684 | ipoib_send(dev, skb, path->ah, IPOIB_QPN(phdr->hwaddr)); |
1da177e4 LT |
685 | } else if ((path->query || !path_rec_start(dev, path)) && |
686 | skb_queue_len(&path->queue) < IPOIB_MAX_PATH_REC_QUEUE) { | |
687 | /* put pseudoheader back on for next time */ | |
688 | skb_push(skb, sizeof *phdr); | |
689 | __skb_queue_tail(&path->queue, skb); | |
690 | } else { | |
de903512 | 691 | ++dev->stats.tx_dropped; |
1da177e4 LT |
692 | dev_kfree_skb_any(skb); |
693 | } | |
694 | ||
943c246e | 695 | spin_unlock_irqrestore(&priv->lock, flags); |
1da177e4 LT |
696 | } |
697 | ||
698 | static int ipoib_start_xmit(struct sk_buff *skb, struct net_device *dev) | |
699 | { | |
700 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
701 | struct ipoib_neigh *neigh; | |
702 | unsigned long flags; | |
703 | ||
a8bfca02 | 704 | if (likely(skb->dst && skb->dst->neighbour)) { |
1da177e4 | 705 | if (unlikely(!*to_ipoib_neigh(skb->dst->neighbour))) { |
d70ed607 | 706 | ipoib_path_lookup(skb, dev); |
943c246e | 707 | return NETDEV_TX_OK; |
1da177e4 LT |
708 | } |
709 | ||
710 | neigh = *to_ipoib_neigh(skb->dst->neighbour); | |
711 | ||
bafff974 | 712 | if (neigh->ah) |
200d1713 | 713 | if (unlikely((memcmp(&neigh->dgid.raw, |
8a7f7521 | 714 | skb->dst->neighbour->ha + 4, |
200d1713 MS |
715 | sizeof(union ib_gid))) || |
716 | (neigh->dev != dev))) { | |
943c246e | 717 | spin_lock_irqsave(&priv->lock, flags); |
8a7f7521 MT |
718 | /* |
719 | * It's safe to call ipoib_put_ah() inside | |
720 | * priv->lock here, because we know that | |
721 | * path->ah will always hold one more reference, | |
722 | * so ipoib_put_ah() will never do more than | |
723 | * decrement the ref count. | |
724 | */ | |
725 | ipoib_put_ah(neigh->ah); | |
726 | list_del(&neigh->list); | |
2745b5b7 | 727 | ipoib_neigh_free(dev, neigh); |
943c246e | 728 | spin_unlock_irqrestore(&priv->lock, flags); |
8a7f7521 | 729 | ipoib_path_lookup(skb, dev); |
943c246e | 730 | return NETDEV_TX_OK; |
8a7f7521 MT |
731 | } |
732 | ||
bafff974 OG |
733 | if (ipoib_cm_get(neigh)) { |
734 | if (ipoib_cm_up(neigh)) { | |
735 | ipoib_cm_send(dev, skb, ipoib_cm_get(neigh)); | |
943c246e | 736 | return NETDEV_TX_OK; |
bafff974 OG |
737 | } |
738 | } else if (neigh->ah) { | |
073ae841 | 739 | ipoib_send(dev, skb, neigh->ah, IPOIB_QPN(skb->dst->neighbour->ha)); |
943c246e | 740 | return NETDEV_TX_OK; |
1da177e4 LT |
741 | } |
742 | ||
743 | if (skb_queue_len(&neigh->queue) < IPOIB_MAX_PATH_REC_QUEUE) { | |
943c246e | 744 | spin_lock_irqsave(&priv->lock, flags); |
1da177e4 | 745 | __skb_queue_tail(&neigh->queue, skb); |
943c246e | 746 | spin_unlock_irqrestore(&priv->lock, flags); |
1da177e4 | 747 | } else { |
de903512 | 748 | ++dev->stats.tx_dropped; |
1da177e4 LT |
749 | dev_kfree_skb_any(skb); |
750 | } | |
751 | } else { | |
752 | struct ipoib_pseudoheader *phdr = | |
753 | (struct ipoib_pseudoheader *) skb->data; | |
754 | skb_pull(skb, sizeof *phdr); | |
755 | ||
756 | if (phdr->hwaddr[4] == 0xff) { | |
757 | /* Add in the P_Key for multicast*/ | |
758 | phdr->hwaddr[8] = (priv->pkey >> 8) & 0xff; | |
759 | phdr->hwaddr[9] = priv->pkey & 0xff; | |
760 | ||
37c22a77 | 761 | ipoib_mcast_send(dev, phdr->hwaddr + 4, skb); |
1da177e4 | 762 | } else { |
0dca0f7b | 763 | /* unicast GID -- should be ARP or RARP reply */ |
1da177e4 | 764 | |
0dca0f7b HR |
765 | if ((be16_to_cpup((__be16 *) skb->data) != ETH_P_ARP) && |
766 | (be16_to_cpup((__be16 *) skb->data) != ETH_P_RARP)) { | |
fcace2fe | 767 | ipoib_warn(priv, "Unicast, no %s: type %04x, QPN %06x %p6\n", |
1da177e4 | 768 | skb->dst ? "neigh" : "dst", |
97f52eb4 | 769 | be16_to_cpup((__be16 *) skb->data), |
073ae841 | 770 | IPOIB_QPN(phdr->hwaddr), |
fcace2fe | 771 | phdr->hwaddr + 4); |
1da177e4 | 772 | dev_kfree_skb_any(skb); |
de903512 | 773 | ++dev->stats.tx_dropped; |
943c246e | 774 | return NETDEV_TX_OK; |
1da177e4 LT |
775 | } |
776 | ||
777 | unicast_arp_send(skb, dev, phdr); | |
778 | } | |
779 | } | |
780 | ||
1da177e4 LT |
781 | return NETDEV_TX_OK; |
782 | } | |
783 | ||
1da177e4 LT |
784 | static void ipoib_timeout(struct net_device *dev) |
785 | { | |
786 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
787 | ||
4b2d319b RD |
788 | ipoib_warn(priv, "transmit timeout: latency %d msecs\n", |
789 | jiffies_to_msecs(jiffies - dev->trans_start)); | |
790 | ipoib_warn(priv, "queue stopped %d, tx_head %u, tx_tail %u\n", | |
791 | netif_queue_stopped(dev), | |
792 | priv->tx_head, priv->tx_tail); | |
1da177e4 LT |
793 | /* XXX reset QP, etc. */ |
794 | } | |
795 | ||
796 | static int ipoib_hard_header(struct sk_buff *skb, | |
797 | struct net_device *dev, | |
798 | unsigned short type, | |
3b04ddde | 799 | const void *daddr, const void *saddr, unsigned len) |
1da177e4 LT |
800 | { |
801 | struct ipoib_header *header; | |
802 | ||
803 | header = (struct ipoib_header *) skb_push(skb, sizeof *header); | |
804 | ||
805 | header->proto = htons(type); | |
806 | header->reserved = 0; | |
807 | ||
808 | /* | |
809 | * If we don't have a neighbour structure, stuff the | |
810 | * destination address onto the front of the skb so we can | |
811 | * figure out where to send the packet later. | |
812 | */ | |
ef12d456 | 813 | if ((!skb->dst || !skb->dst->neighbour) && daddr) { |
1da177e4 LT |
814 | struct ipoib_pseudoheader *phdr = |
815 | (struct ipoib_pseudoheader *) skb_push(skb, sizeof *phdr); | |
816 | memcpy(phdr->hwaddr, daddr, INFINIBAND_ALEN); | |
817 | } | |
818 | ||
819 | return 0; | |
820 | } | |
821 | ||
822 | static void ipoib_set_mcast_list(struct net_device *dev) | |
823 | { | |
824 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
825 | ||
7a343d4c LA |
826 | if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags)) { |
827 | ipoib_dbg(priv, "IPOIB_FLAG_OPER_UP not set"); | |
828 | return; | |
829 | } | |
830 | ||
1ad62a19 | 831 | queue_work(ipoib_workqueue, &priv->restart_task); |
1da177e4 LT |
832 | } |
833 | ||
ecbb4169 | 834 | static void ipoib_neigh_cleanup(struct neighbour *n) |
1da177e4 LT |
835 | { |
836 | struct ipoib_neigh *neigh; | |
837 | struct ipoib_dev_priv *priv = netdev_priv(n->dev); | |
838 | unsigned long flags; | |
839 | struct ipoib_ah *ah = NULL; | |
840 | ||
732a2170 | 841 | neigh = *to_ipoib_neigh(n); |
7bc531dd | 842 | if (neigh) |
732a2170 | 843 | priv = netdev_priv(neigh->dev); |
7bc531dd | 844 | else |
732a2170 | 845 | return; |
1da177e4 | 846 | ipoib_dbg(priv, |
fcace2fe | 847 | "neigh_cleanup for %06x %p6\n", |
073ae841 | 848 | IPOIB_QPN(n->ha), |
fcace2fe | 849 | n->ha + 4); |
1da177e4 LT |
850 | |
851 | spin_lock_irqsave(&priv->lock, flags); | |
852 | ||
732a2170 MS |
853 | if (neigh->ah) |
854 | ah = neigh->ah; | |
855 | list_del(&neigh->list); | |
856 | ipoib_neigh_free(n->dev, neigh); | |
1da177e4 LT |
857 | |
858 | spin_unlock_irqrestore(&priv->lock, flags); | |
859 | ||
860 | if (ah) | |
861 | ipoib_put_ah(ah); | |
862 | } | |
863 | ||
732a2170 MS |
864 | struct ipoib_neigh *ipoib_neigh_alloc(struct neighbour *neighbour, |
865 | struct net_device *dev) | |
d2e0655e MT |
866 | { |
867 | struct ipoib_neigh *neigh; | |
868 | ||
869 | neigh = kmalloc(sizeof *neigh, GFP_ATOMIC); | |
870 | if (!neigh) | |
871 | return NULL; | |
872 | ||
873 | neigh->neighbour = neighbour; | |
732a2170 | 874 | neigh->dev = dev; |
d2e0655e | 875 | *to_ipoib_neigh(neighbour) = neigh; |
82b39913 | 876 | skb_queue_head_init(&neigh->queue); |
839fcaba | 877 | ipoib_cm_set(neigh, NULL); |
d2e0655e MT |
878 | |
879 | return neigh; | |
880 | } | |
881 | ||
2745b5b7 | 882 | void ipoib_neigh_free(struct net_device *dev, struct ipoib_neigh *neigh) |
d2e0655e | 883 | { |
2745b5b7 | 884 | struct sk_buff *skb; |
d2e0655e | 885 | *to_ipoib_neigh(neigh->neighbour) = NULL; |
2745b5b7 | 886 | while ((skb = __skb_dequeue(&neigh->queue))) { |
de903512 | 887 | ++dev->stats.tx_dropped; |
2745b5b7 MT |
888 | dev_kfree_skb_any(skb); |
889 | } | |
839fcaba MT |
890 | if (ipoib_cm_get(neigh)) |
891 | ipoib_cm_destroy_tx(ipoib_cm_get(neigh)); | |
d2e0655e MT |
892 | kfree(neigh); |
893 | } | |
894 | ||
1da177e4 LT |
895 | static int ipoib_neigh_setup_dev(struct net_device *dev, struct neigh_parms *parms) |
896 | { | |
ecbb4169 | 897 | parms->neigh_cleanup = ipoib_neigh_cleanup; |
1da177e4 LT |
898 | |
899 | return 0; | |
900 | } | |
901 | ||
902 | int ipoib_dev_init(struct net_device *dev, struct ib_device *ca, int port) | |
903 | { | |
904 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
905 | ||
906 | /* Allocate RX/TX "rings" to hold queued skbs */ | |
0f485251 | 907 | priv->rx_ring = kzalloc(ipoib_recvq_size * sizeof *priv->rx_ring, |
1da177e4 LT |
908 | GFP_KERNEL); |
909 | if (!priv->rx_ring) { | |
910 | printk(KERN_WARNING "%s: failed to allocate RX ring (%d entries)\n", | |
0f485251 | 911 | ca->name, ipoib_recvq_size); |
1da177e4 LT |
912 | goto out; |
913 | } | |
1da177e4 | 914 | |
10313cbb | 915 | priv->tx_ring = vmalloc(ipoib_sendq_size * sizeof *priv->tx_ring); |
1da177e4 LT |
916 | if (!priv->tx_ring) { |
917 | printk(KERN_WARNING "%s: failed to allocate TX ring (%d entries)\n", | |
0f485251 | 918 | ca->name, ipoib_sendq_size); |
1da177e4 LT |
919 | goto out_rx_ring_cleanup; |
920 | } | |
10313cbb | 921 | memset(priv->tx_ring, 0, ipoib_sendq_size * sizeof *priv->tx_ring); |
1da177e4 | 922 | |
1b524963 | 923 | /* priv->tx_head, tx_tail & tx_outstanding are already 0 */ |
1da177e4 LT |
924 | |
925 | if (ipoib_ib_dev_init(dev, ca, port)) | |
926 | goto out_tx_ring_cleanup; | |
927 | ||
928 | return 0; | |
929 | ||
930 | out_tx_ring_cleanup: | |
10313cbb | 931 | vfree(priv->tx_ring); |
1da177e4 LT |
932 | |
933 | out_rx_ring_cleanup: | |
934 | kfree(priv->rx_ring); | |
935 | ||
936 | out: | |
937 | return -ENOMEM; | |
938 | } | |
939 | ||
940 | void ipoib_dev_cleanup(struct net_device *dev) | |
941 | { | |
942 | struct ipoib_dev_priv *priv = netdev_priv(dev), *cpriv, *tcpriv; | |
943 | ||
1732b0ef | 944 | ipoib_delete_debug_files(dev); |
1da177e4 LT |
945 | |
946 | /* Delete any child interfaces first */ | |
947 | list_for_each_entry_safe(cpriv, tcpriv, &priv->child_intfs, list) { | |
948 | unregister_netdev(cpriv->dev); | |
949 | ipoib_dev_cleanup(cpriv->dev); | |
950 | free_netdev(cpriv->dev); | |
951 | } | |
952 | ||
953 | ipoib_ib_dev_cleanup(dev); | |
954 | ||
92a6b34b | 955 | kfree(priv->rx_ring); |
10313cbb | 956 | vfree(priv->tx_ring); |
1da177e4 | 957 | |
92a6b34b HR |
958 | priv->rx_ring = NULL; |
959 | priv->tx_ring = NULL; | |
1da177e4 LT |
960 | } |
961 | ||
3b04ddde SH |
962 | static const struct header_ops ipoib_header_ops = { |
963 | .create = ipoib_hard_header, | |
964 | }; | |
965 | ||
af40da89 VS |
966 | static int get_skb_hdr(struct sk_buff *skb, void **iphdr, |
967 | void **tcph, u64 *hdr_flags, void *priv) | |
968 | { | |
969 | unsigned int ip_len; | |
970 | struct iphdr *iph; | |
971 | ||
972 | if (unlikely(skb->protocol != htons(ETH_P_IP))) | |
973 | return -1; | |
974 | ||
975 | /* | |
976 | * In the future we may add an else clause that verifies the | |
977 | * checksum and allows devices which do not calculate checksum | |
978 | * to use LRO. | |
979 | */ | |
980 | if (unlikely(skb->ip_summed != CHECKSUM_UNNECESSARY)) | |
981 | return -1; | |
982 | ||
983 | /* Check for non-TCP packet */ | |
984 | skb_reset_network_header(skb); | |
985 | iph = ip_hdr(skb); | |
986 | if (iph->protocol != IPPROTO_TCP) | |
987 | return -1; | |
988 | ||
989 | ip_len = ip_hdrlen(skb); | |
990 | skb_set_transport_header(skb, ip_len); | |
991 | *tcph = tcp_hdr(skb); | |
992 | ||
993 | /* check if IP header and TCP header are complete */ | |
994 | if (ntohs(iph->tot_len) < ip_len + tcp_hdrlen(skb)) | |
995 | return -1; | |
996 | ||
997 | *hdr_flags = LRO_IPV4 | LRO_TCP; | |
998 | *iphdr = iph; | |
999 | ||
1000 | return 0; | |
1001 | } | |
1002 | ||
1003 | static void ipoib_lro_setup(struct ipoib_dev_priv *priv) | |
1004 | { | |
1005 | priv->lro.lro_mgr.max_aggr = lro_max_aggr; | |
1006 | priv->lro.lro_mgr.max_desc = IPOIB_MAX_LRO_DESCRIPTORS; | |
1007 | priv->lro.lro_mgr.lro_arr = priv->lro.lro_desc; | |
1008 | priv->lro.lro_mgr.get_skb_header = get_skb_hdr; | |
1009 | priv->lro.lro_mgr.features = LRO_F_NAPI; | |
1010 | priv->lro.lro_mgr.dev = priv->dev; | |
1011 | priv->lro.lro_mgr.ip_summed_aggr = CHECKSUM_UNNECESSARY; | |
1012 | } | |
1013 | ||
1da177e4 LT |
1014 | static void ipoib_setup(struct net_device *dev) |
1015 | { | |
1016 | struct ipoib_dev_priv *priv = netdev_priv(dev); | |
1017 | ||
2337f809 RD |
1018 | dev->open = ipoib_open; |
1019 | dev->stop = ipoib_stop; | |
1020 | dev->change_mtu = ipoib_change_mtu; | |
1021 | dev->hard_start_xmit = ipoib_start_xmit; | |
1022 | dev->tx_timeout = ipoib_timeout; | |
1023 | dev->header_ops = &ipoib_header_ops; | |
1024 | dev->set_multicast_list = ipoib_set_mcast_list; | |
1025 | dev->neigh_setup = ipoib_neigh_setup_dev; | |
bea3348e | 1026 | |
82c24c18 EC |
1027 | ipoib_set_ethtool_ops(dev); |
1028 | ||
bea3348e | 1029 | netif_napi_add(dev, &priv->napi, ipoib_poll, 100); |
1da177e4 | 1030 | |
2337f809 | 1031 | dev->watchdog_timeo = HZ; |
1da177e4 | 1032 | |
2337f809 | 1033 | dev->flags |= IFF_BROADCAST | IFF_MULTICAST; |
1da177e4 LT |
1034 | |
1035 | /* | |
1036 | * We add in INFINIBAND_ALEN to allow for the destination | |
1037 | * address "pseudoheader" for skbs without neighbour struct. | |
1038 | */ | |
2337f809 RD |
1039 | dev->hard_header_len = IPOIB_ENCAP_LEN + INFINIBAND_ALEN; |
1040 | dev->addr_len = INFINIBAND_ALEN; | |
1041 | dev->type = ARPHRD_INFINIBAND; | |
1042 | dev->tx_queue_len = ipoib_sendq_size * 2; | |
eb14032f | 1043 | dev->features = (NETIF_F_VLAN_CHALLENGED | |
eb14032f | 1044 | NETIF_F_HIGHDMA); |
1da177e4 | 1045 | |
1da177e4 LT |
1046 | memcpy(dev->broadcast, ipv4_bcast_addr, INFINIBAND_ALEN); |
1047 | ||
1048 | netif_carrier_off(dev); | |
1049 | ||
1da177e4 LT |
1050 | priv->dev = dev; |
1051 | ||
af40da89 VS |
1052 | ipoib_lro_setup(priv); |
1053 | ||
1da177e4 | 1054 | spin_lock_init(&priv->lock); |
1da177e4 | 1055 | |
95ed644f | 1056 | mutex_init(&priv->vlan_mutex); |
1da177e4 LT |
1057 | |
1058 | INIT_LIST_HEAD(&priv->path_list); | |
1059 | INIT_LIST_HEAD(&priv->child_intfs); | |
1060 | INIT_LIST_HEAD(&priv->dead_ahs); | |
1061 | INIT_LIST_HEAD(&priv->multicast_list); | |
1062 | ||
26bbf13c | 1063 | INIT_DELAYED_WORK(&priv->pkey_poll_task, ipoib_pkey_poll); |
c4028958 | 1064 | INIT_DELAYED_WORK(&priv->mcast_task, ipoib_mcast_join_task); |
e8224e4b | 1065 | INIT_WORK(&priv->carrier_on_task, ipoib_mcast_carrier_on_task); |
ee1e2c82 MS |
1066 | INIT_WORK(&priv->flush_light, ipoib_ib_dev_flush_light); |
1067 | INIT_WORK(&priv->flush_normal, ipoib_ib_dev_flush_normal); | |
1068 | INIT_WORK(&priv->flush_heavy, ipoib_ib_dev_flush_heavy); | |
c4028958 DH |
1069 | INIT_WORK(&priv->restart_task, ipoib_mcast_restart_task); |
1070 | INIT_DELAYED_WORK(&priv->ah_reap_task, ipoib_reap_ah); | |
1da177e4 LT |
1071 | } |
1072 | ||
1073 | struct ipoib_dev_priv *ipoib_intf_alloc(const char *name) | |
1074 | { | |
1075 | struct net_device *dev; | |
1076 | ||
1077 | dev = alloc_netdev((int) sizeof (struct ipoib_dev_priv), name, | |
1078 | ipoib_setup); | |
1079 | if (!dev) | |
1080 | return NULL; | |
1081 | ||
1082 | return netdev_priv(dev); | |
1083 | } | |
1084 | ||
43cb76d9 GKH |
1085 | static ssize_t show_pkey(struct device *dev, |
1086 | struct device_attribute *attr, char *buf) | |
1da177e4 | 1087 | { |
43cb76d9 | 1088 | struct ipoib_dev_priv *priv = netdev_priv(to_net_dev(dev)); |
1da177e4 LT |
1089 | |
1090 | return sprintf(buf, "0x%04x\n", priv->pkey); | |
1091 | } | |
43cb76d9 | 1092 | static DEVICE_ATTR(pkey, S_IRUGO, show_pkey, NULL); |
1da177e4 | 1093 | |
335a64a5 OG |
1094 | static ssize_t show_umcast(struct device *dev, |
1095 | struct device_attribute *attr, char *buf) | |
1096 | { | |
1097 | struct ipoib_dev_priv *priv = netdev_priv(to_net_dev(dev)); | |
1098 | ||
1099 | return sprintf(buf, "%d\n", test_bit(IPOIB_FLAG_UMCAST, &priv->flags)); | |
1100 | } | |
1101 | ||
1102 | static ssize_t set_umcast(struct device *dev, | |
1103 | struct device_attribute *attr, | |
1104 | const char *buf, size_t count) | |
1105 | { | |
1106 | struct ipoib_dev_priv *priv = netdev_priv(to_net_dev(dev)); | |
1107 | unsigned long umcast_val = simple_strtoul(buf, NULL, 0); | |
1108 | ||
1109 | if (umcast_val > 0) { | |
1110 | set_bit(IPOIB_FLAG_UMCAST, &priv->flags); | |
1111 | ipoib_warn(priv, "ignoring multicast groups joined directly " | |
1112 | "by userspace\n"); | |
1113 | } else | |
1114 | clear_bit(IPOIB_FLAG_UMCAST, &priv->flags); | |
1115 | ||
1116 | return count; | |
1117 | } | |
1118 | static DEVICE_ATTR(umcast, S_IWUSR | S_IRUGO, show_umcast, set_umcast); | |
1119 | ||
1120 | int ipoib_add_umcast_attr(struct net_device *dev) | |
1121 | { | |
1122 | return device_create_file(&dev->dev, &dev_attr_umcast); | |
1123 | } | |
1124 | ||
43cb76d9 GKH |
1125 | static ssize_t create_child(struct device *dev, |
1126 | struct device_attribute *attr, | |
1da177e4 LT |
1127 | const char *buf, size_t count) |
1128 | { | |
1129 | int pkey; | |
1130 | int ret; | |
1131 | ||
1132 | if (sscanf(buf, "%i", &pkey) != 1) | |
1133 | return -EINVAL; | |
1134 | ||
1135 | if (pkey < 0 || pkey > 0xffff) | |
1136 | return -EINVAL; | |
1137 | ||
4ce05937 RD |
1138 | /* |
1139 | * Set the full membership bit, so that we join the right | |
1140 | * broadcast group, etc. | |
1141 | */ | |
1142 | pkey |= 0x8000; | |
1143 | ||
43cb76d9 | 1144 | ret = ipoib_vlan_add(to_net_dev(dev), pkey); |
1da177e4 LT |
1145 | |
1146 | return ret ? ret : count; | |
1147 | } | |
43cb76d9 | 1148 | static DEVICE_ATTR(create_child, S_IWUGO, NULL, create_child); |
1da177e4 | 1149 | |
43cb76d9 GKH |
1150 | static ssize_t delete_child(struct device *dev, |
1151 | struct device_attribute *attr, | |
1da177e4 LT |
1152 | const char *buf, size_t count) |
1153 | { | |
1154 | int pkey; | |
1155 | int ret; | |
1156 | ||
1157 | if (sscanf(buf, "%i", &pkey) != 1) | |
1158 | return -EINVAL; | |
1159 | ||
1160 | if (pkey < 0 || pkey > 0xffff) | |
1161 | return -EINVAL; | |
1162 | ||
43cb76d9 | 1163 | ret = ipoib_vlan_delete(to_net_dev(dev), pkey); |
1da177e4 LT |
1164 | |
1165 | return ret ? ret : count; | |
1166 | ||
1167 | } | |
43cb76d9 | 1168 | static DEVICE_ATTR(delete_child, S_IWUGO, NULL, delete_child); |
1da177e4 LT |
1169 | |
1170 | int ipoib_add_pkey_attr(struct net_device *dev) | |
1171 | { | |
43cb76d9 | 1172 | return device_create_file(&dev->dev, &dev_attr_pkey); |
1da177e4 LT |
1173 | } |
1174 | ||
83bb63f6 OG |
1175 | int ipoib_set_dev_features(struct ipoib_dev_priv *priv, struct ib_device *hca) |
1176 | { | |
1177 | struct ib_device_attr *device_attr; | |
1178 | int result = -ENOMEM; | |
1179 | ||
1180 | device_attr = kmalloc(sizeof *device_attr, GFP_KERNEL); | |
1181 | if (!device_attr) { | |
1182 | printk(KERN_WARNING "%s: allocation of %zu bytes failed\n", | |
1183 | hca->name, sizeof *device_attr); | |
1184 | return result; | |
1185 | } | |
1186 | ||
1187 | result = ib_query_device(hca, device_attr); | |
1188 | if (result) { | |
1189 | printk(KERN_WARNING "%s: ib_query_device failed (ret = %d)\n", | |
1190 | hca->name, result); | |
1191 | kfree(device_attr); | |
1192 | return result; | |
1193 | } | |
1194 | priv->hca_caps = device_attr->device_cap_flags; | |
1195 | ||
1196 | kfree(device_attr); | |
1197 | ||
1198 | if (priv->hca_caps & IB_DEVICE_UD_IP_CSUM) { | |
1199 | set_bit(IPOIB_FLAG_CSUM, &priv->flags); | |
1200 | priv->dev->features |= NETIF_F_SG | NETIF_F_IP_CSUM; | |
1201 | } | |
1202 | ||
1203 | if (lro) | |
1204 | priv->dev->features |= NETIF_F_LRO; | |
1205 | ||
1206 | if (priv->dev->features & NETIF_F_SG && priv->hca_caps & IB_DEVICE_UD_TSO) | |
1207 | priv->dev->features |= NETIF_F_TSO; | |
1208 | ||
1209 | return 0; | |
1210 | } | |
1211 | ||
1212 | ||
1da177e4 LT |
1213 | static struct net_device *ipoib_add_port(const char *format, |
1214 | struct ib_device *hca, u8 port) | |
1215 | { | |
1216 | struct ipoib_dev_priv *priv; | |
bc7b3a36 | 1217 | struct ib_port_attr attr; |
1da177e4 LT |
1218 | int result = -ENOMEM; |
1219 | ||
1220 | priv = ipoib_intf_alloc(format); | |
1221 | if (!priv) | |
1222 | goto alloc_mem_failed; | |
1223 | ||
1224 | SET_NETDEV_DEV(priv->dev, hca->dma_device); | |
1225 | ||
bc7b3a36 SM |
1226 | if (!ib_query_port(hca, port, &attr)) |
1227 | priv->max_ib_mtu = ib_mtu_enum_to_int(attr.max_mtu); | |
1228 | else { | |
1229 | printk(KERN_WARNING "%s: ib_query_port %d failed\n", | |
1230 | hca->name, port); | |
1231 | goto device_init_failed; | |
1232 | } | |
1233 | ||
1234 | /* MTU will be reset when mcast join happens */ | |
1235 | priv->dev->mtu = IPOIB_UD_MTU(priv->max_ib_mtu); | |
1236 | priv->mcast_mtu = priv->admin_mtu = priv->dev->mtu; | |
1237 | ||
1da177e4 LT |
1238 | result = ib_query_pkey(hca, port, 0, &priv->pkey); |
1239 | if (result) { | |
1240 | printk(KERN_WARNING "%s: ib_query_pkey port %d failed (ret = %d)\n", | |
1241 | hca->name, port, result); | |
ca6de177 | 1242 | goto device_init_failed; |
1da177e4 LT |
1243 | } |
1244 | ||
83bb63f6 | 1245 | if (ipoib_set_dev_features(priv, hca)) |
6046136c | 1246 | goto device_init_failed; |
af40da89 | 1247 | |
4ce05937 RD |
1248 | /* |
1249 | * Set the full membership bit, so that we join the right | |
1250 | * broadcast group, etc. | |
1251 | */ | |
1252 | priv->pkey |= 0x8000; | |
1253 | ||
1da177e4 LT |
1254 | priv->dev->broadcast[8] = priv->pkey >> 8; |
1255 | priv->dev->broadcast[9] = priv->pkey & 0xff; | |
1256 | ||
1257 | result = ib_query_gid(hca, port, 0, &priv->local_gid); | |
1258 | if (result) { | |
1259 | printk(KERN_WARNING "%s: ib_query_gid port %d failed (ret = %d)\n", | |
1260 | hca->name, port, result); | |
ca6de177 | 1261 | goto device_init_failed; |
1da177e4 LT |
1262 | } else |
1263 | memcpy(priv->dev->dev_addr + 4, priv->local_gid.raw, sizeof (union ib_gid)); | |
1264 | ||
1da177e4 LT |
1265 | result = ipoib_dev_init(priv->dev, hca, port); |
1266 | if (result < 0) { | |
1267 | printk(KERN_WARNING "%s: failed to initialize port %d (ret = %d)\n", | |
1268 | hca->name, port, result); | |
1269 | goto device_init_failed; | |
1270 | } | |
1271 | ||
1272 | INIT_IB_EVENT_HANDLER(&priv->event_handler, | |
1273 | priv->ca, ipoib_event); | |
1274 | result = ib_register_event_handler(&priv->event_handler); | |
1275 | if (result < 0) { | |
1276 | printk(KERN_WARNING "%s: ib_register_event_handler failed for " | |
1277 | "port %d (ret = %d)\n", | |
1278 | hca->name, port, result); | |
1279 | goto event_failed; | |
1280 | } | |
1281 | ||
1282 | result = register_netdev(priv->dev); | |
1283 | if (result) { | |
1284 | printk(KERN_WARNING "%s: couldn't register ipoib port %d; error %d\n", | |
1285 | hca->name, port, result); | |
1286 | goto register_failed; | |
1287 | } | |
1288 | ||
1732b0ef | 1289 | ipoib_create_debug_files(priv->dev); |
1da177e4 | 1290 | |
839fcaba MT |
1291 | if (ipoib_cm_add_mode_attr(priv->dev)) |
1292 | goto sysfs_failed; | |
1da177e4 LT |
1293 | if (ipoib_add_pkey_attr(priv->dev)) |
1294 | goto sysfs_failed; | |
335a64a5 OG |
1295 | if (ipoib_add_umcast_attr(priv->dev)) |
1296 | goto sysfs_failed; | |
43cb76d9 | 1297 | if (device_create_file(&priv->dev->dev, &dev_attr_create_child)) |
1da177e4 | 1298 | goto sysfs_failed; |
43cb76d9 | 1299 | if (device_create_file(&priv->dev->dev, &dev_attr_delete_child)) |
1da177e4 LT |
1300 | goto sysfs_failed; |
1301 | ||
1302 | return priv->dev; | |
1303 | ||
1304 | sysfs_failed: | |
1732b0ef | 1305 | ipoib_delete_debug_files(priv->dev); |
1da177e4 LT |
1306 | unregister_netdev(priv->dev); |
1307 | ||
1308 | register_failed: | |
1309 | ib_unregister_event_handler(&priv->event_handler); | |
a77a57a1 | 1310 | flush_workqueue(ipoib_workqueue); |
1da177e4 LT |
1311 | |
1312 | event_failed: | |
1313 | ipoib_dev_cleanup(priv->dev); | |
1314 | ||
1315 | device_init_failed: | |
1316 | free_netdev(priv->dev); | |
1317 | ||
1318 | alloc_mem_failed: | |
1319 | return ERR_PTR(result); | |
1320 | } | |
1321 | ||
1322 | static void ipoib_add_one(struct ib_device *device) | |
1323 | { | |
1324 | struct list_head *dev_list; | |
1325 | struct net_device *dev; | |
1326 | struct ipoib_dev_priv *priv; | |
1327 | int s, e, p; | |
1328 | ||
07ebafba TT |
1329 | if (rdma_node_get_transport(device->node_type) != RDMA_TRANSPORT_IB) |
1330 | return; | |
1331 | ||
1da177e4 LT |
1332 | dev_list = kmalloc(sizeof *dev_list, GFP_KERNEL); |
1333 | if (!dev_list) | |
1334 | return; | |
1335 | ||
1336 | INIT_LIST_HEAD(dev_list); | |
1337 | ||
07ebafba | 1338 | if (device->node_type == RDMA_NODE_IB_SWITCH) { |
1da177e4 LT |
1339 | s = 0; |
1340 | e = 0; | |
1341 | } else { | |
1342 | s = 1; | |
1343 | e = device->phys_port_cnt; | |
1344 | } | |
1345 | ||
1346 | for (p = s; p <= e; ++p) { | |
1347 | dev = ipoib_add_port("ib%d", device, p); | |
1348 | if (!IS_ERR(dev)) { | |
1349 | priv = netdev_priv(dev); | |
1350 | list_add_tail(&priv->list, dev_list); | |
1351 | } | |
1352 | } | |
1353 | ||
1354 | ib_set_client_data(device, &ipoib_client, dev_list); | |
1355 | } | |
1356 | ||
1357 | static void ipoib_remove_one(struct ib_device *device) | |
1358 | { | |
1359 | struct ipoib_dev_priv *priv, *tmp; | |
1360 | struct list_head *dev_list; | |
1361 | ||
07ebafba TT |
1362 | if (rdma_node_get_transport(device->node_type) != RDMA_TRANSPORT_IB) |
1363 | return; | |
1364 | ||
1da177e4 LT |
1365 | dev_list = ib_get_client_data(device, &ipoib_client); |
1366 | ||
1367 | list_for_each_entry_safe(priv, tmp, dev_list, list) { | |
1368 | ib_unregister_event_handler(&priv->event_handler); | |
a77a57a1 RD |
1369 | |
1370 | rtnl_lock(); | |
1371 | dev_change_flags(priv->dev, priv->dev->flags & ~IFF_UP); | |
1372 | rtnl_unlock(); | |
1373 | ||
1374 | flush_workqueue(ipoib_workqueue); | |
1da177e4 LT |
1375 | |
1376 | unregister_netdev(priv->dev); | |
1377 | ipoib_dev_cleanup(priv->dev); | |
1378 | free_netdev(priv->dev); | |
1379 | } | |
06c56e44 MT |
1380 | |
1381 | kfree(dev_list); | |
1da177e4 LT |
1382 | } |
1383 | ||
1384 | static int __init ipoib_init_module(void) | |
1385 | { | |
1386 | int ret; | |
1387 | ||
0f485251 SM |
1388 | ipoib_recvq_size = roundup_pow_of_two(ipoib_recvq_size); |
1389 | ipoib_recvq_size = min(ipoib_recvq_size, IPOIB_MAX_QUEUE_SIZE); | |
1390 | ipoib_recvq_size = max(ipoib_recvq_size, IPOIB_MIN_QUEUE_SIZE); | |
1391 | ||
1392 | ipoib_sendq_size = roundup_pow_of_two(ipoib_sendq_size); | |
1393 | ipoib_sendq_size = min(ipoib_sendq_size, IPOIB_MAX_QUEUE_SIZE); | |
f56bcd80 EC |
1394 | ipoib_sendq_size = max(ipoib_sendq_size, max(2 * MAX_SEND_CQE, |
1395 | IPOIB_MIN_QUEUE_SIZE)); | |
68e995a2 PS |
1396 | #ifdef CONFIG_INFINIBAND_IPOIB_CM |
1397 | ipoib_max_conn_qp = min(ipoib_max_conn_qp, IPOIB_CM_MAX_CONN_QP); | |
1398 | #endif | |
0f485251 | 1399 | |
f89271da EC |
1400 | /* |
1401 | * When copying small received packets, we only copy from the | |
1402 | * linear data part of the SKB, so we rely on this condition. | |
1403 | */ | |
1404 | BUILD_BUG_ON(IPOIB_CM_COPYBREAK > IPOIB_CM_HEAD_SIZE); | |
1405 | ||
1da177e4 LT |
1406 | ret = ipoib_register_debugfs(); |
1407 | if (ret) | |
1408 | return ret; | |
1409 | ||
1410 | /* | |
1411 | * We create our own workqueue mainly because we want to be | |
1412 | * able to flush it when devices are being removed. We can't | |
1413 | * use schedule_work()/flush_scheduled_work() because both | |
1414 | * unregister_netdev() and linkwatch_event take the rtnl lock, | |
1415 | * so flush_scheduled_work() can deadlock during device | |
1416 | * removal. | |
1417 | */ | |
1418 | ipoib_workqueue = create_singlethread_workqueue("ipoib"); | |
1419 | if (!ipoib_workqueue) { | |
1420 | ret = -ENOMEM; | |
1421 | goto err_fs; | |
1422 | } | |
1423 | ||
c1a0b23b MT |
1424 | ib_sa_register_client(&ipoib_sa_client); |
1425 | ||
1da177e4 LT |
1426 | ret = ib_register_client(&ipoib_client); |
1427 | if (ret) | |
c1a0b23b | 1428 | goto err_sa; |
1da177e4 LT |
1429 | |
1430 | return 0; | |
1431 | ||
c1a0b23b MT |
1432 | err_sa: |
1433 | ib_sa_unregister_client(&ipoib_sa_client); | |
1da177e4 LT |
1434 | destroy_workqueue(ipoib_workqueue); |
1435 | ||
9adec1a8 RD |
1436 | err_fs: |
1437 | ipoib_unregister_debugfs(); | |
1438 | ||
1da177e4 LT |
1439 | return ret; |
1440 | } | |
1441 | ||
1442 | static void __exit ipoib_cleanup_module(void) | |
1443 | { | |
1da177e4 | 1444 | ib_unregister_client(&ipoib_client); |
c1a0b23b | 1445 | ib_sa_unregister_client(&ipoib_sa_client); |
9adec1a8 | 1446 | ipoib_unregister_debugfs(); |
1da177e4 LT |
1447 | destroy_workqueue(ipoib_workqueue); |
1448 | } | |
1449 | ||
1450 | module_init(ipoib_init_module); | |
1451 | module_exit(ipoib_cleanup_module); |