1 /*-
2 * SPDX-License-Identifier: BSD-2-Clause OR GPL-2.0
3 *
4 * Copyright (c) 2004 Topspin Communications. All rights reserved.
5 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
6 * Copyright (c) 2004 Voltaire, Inc. All rights reserved.
7 *
8 * This software is available to you under a choice of one of two
9 * licenses. You may choose to be licensed under the terms of the GNU
10 * General Public License (GPL) Version 2, available from the file
11 * COPYING in the main directory of this source tree, or the
12 * OpenIB.org BSD license below:
13 *
14 * Redistribution and use in source and binary forms, with or
15 * without modification, are permitted provided that the following
16 * conditions are met:
17 *
18 * - Redistributions of source code must retain the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer.
21 *
22 * - Redistributions in binary form must reproduce the above
23 * copyright notice, this list of conditions and the following
24 * disclaimer in the documentation and/or other materials
25 * provided with the distribution.
26 *
27 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
28 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
29 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
30 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
31 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
32 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
33 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
34 * SOFTWARE.
35 */
36
37 #include <sys/cdefs.h>
38 #include "ipoib.h"
39 #include <sys/eventhandler.h>
40
41 #include <linux/module.h>
42
43 #include <linux/slab.h>
44 #include <linux/kernel.h>
45 #include <linux/vmalloc.h>
46
47 #include <linux/if_vlan.h>
48
49 #include <net/infiniband.h>
50
51 #include <rdma/ib_addr.h>
52 #include <rdma/ib_cache.h>
53
54 MODULE_AUTHOR("Roland Dreier");
55 MODULE_DESCRIPTION("IP-over-InfiniBand net driver");
56 MODULE_LICENSE("Dual BSD/GPL");
57
58 int ipoib_sendq_size = IPOIB_TX_RING_SIZE;
59 int ipoib_recvq_size = IPOIB_RX_RING_SIZE;
60
61 module_param_named(send_queue_size, ipoib_sendq_size, int, 0444);
62 MODULE_PARM_DESC(send_queue_size, "Number of descriptors in send queue");
63 module_param_named(recv_queue_size, ipoib_recvq_size, int, 0444);
64 MODULE_PARM_DESC(recv_queue_size, "Number of descriptors in receive queue");
65
66 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
67 int ipoib_debug_level = 1;
68
69 module_param_named(debug_level, ipoib_debug_level, int, 0644);
70 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0");
71 #endif
72
73 struct ipoib_path_iter {
74 struct ipoib_dev_priv *priv;
75 struct ipoib_path path;
76 };
77
78 static const u8 ipv4_bcast_addr[] = {
79 0x00, 0xff, 0xff, 0xff,
80 0xff, 0x12, 0x40, 0x1b, 0x00, 0x00, 0x00, 0x00,
81 0x00, 0x00, 0x00, 0x00, 0xff, 0xff, 0xff, 0xff
82 };
83
84 struct workqueue_struct *ipoib_workqueue;
85
86 struct ib_sa_client ipoib_sa_client;
87
88 static void ipoib_add_one(struct ib_device *device);
89 static void ipoib_remove_one(struct ib_device *device, void *client_data);
90 static if_t ipoib_get_net_dev_by_params(
91 struct ib_device *dev, u8 port, u16 pkey,
92 const union ib_gid *gid, const struct sockaddr *addr,
93 void *client_data);
94 static void ipoib_start(if_t dev);
95 static int ipoib_ioctl(if_t ifp, u_long command, caddr_t data);
96
97 static struct unrhdr *ipoib_unrhdr;
98
99 static void
ipoib_unrhdr_init(void * arg)100 ipoib_unrhdr_init(void *arg)
101 {
102
103 ipoib_unrhdr = new_unrhdr(0, 65535, NULL);
104 }
105 SYSINIT(ipoib_unrhdr_init, SI_SUB_KLD - 1, SI_ORDER_ANY, ipoib_unrhdr_init, NULL);
106
107 static void
ipoib_unrhdr_uninit(void * arg)108 ipoib_unrhdr_uninit(void *arg)
109 {
110
111 if (ipoib_unrhdr != NULL) {
112 struct unrhdr *hdr;
113
114 hdr = ipoib_unrhdr;
115 ipoib_unrhdr = NULL;
116
117 delete_unrhdr(hdr);
118 }
119 }
120 SYSUNINIT(ipoib_unrhdr_uninit, SI_SUB_KLD - 1, SI_ORDER_ANY, ipoib_unrhdr_uninit, NULL);
121
122 static struct ib_client ipoib_client = {
123 .name = "ipoib",
124 .add = ipoib_add_one,
125 .remove = ipoib_remove_one,
126 .get_net_dev_by_params = ipoib_get_net_dev_by_params,
127 };
128
129 int
ipoib_open(struct ipoib_dev_priv * priv)130 ipoib_open(struct ipoib_dev_priv *priv)
131 {
132 if_t dev = priv->dev;
133
134 ipoib_dbg(priv, "bringing up interface\n");
135
136 set_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags);
137
138 if (ipoib_pkey_dev_delay_open(priv))
139 return 0;
140
141 if (ipoib_ib_dev_open(priv))
142 goto err_disable;
143
144 if (ipoib_ib_dev_up(priv))
145 goto err_stop;
146
147 if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) {
148 struct ipoib_dev_priv *cpriv;
149
150 /* Bring up any child interfaces too */
151 mutex_lock(&priv->vlan_mutex);
152 list_for_each_entry(cpriv, &priv->child_intfs, list)
153 if ((if_getdrvflags(cpriv->dev) & IFF_DRV_RUNNING) == 0)
154 ipoib_open(cpriv);
155 mutex_unlock(&priv->vlan_mutex);
156 }
157 if_setdrvflagbits(dev, IFF_DRV_RUNNING, IFF_DRV_OACTIVE);
158
159 return 0;
160
161 err_stop:
162 ipoib_ib_dev_stop(priv, 1);
163
164 err_disable:
165 clear_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags);
166
167 return -EINVAL;
168 }
169
170 static void
ipoib_init(void * arg)171 ipoib_init(void *arg)
172 {
173 if_t dev;
174 struct ipoib_dev_priv *priv;
175
176 priv = arg;
177 dev = priv->dev;
178 if ((if_getdrvflags(dev) & IFF_DRV_RUNNING) == 0)
179 ipoib_open(priv);
180 queue_work(ipoib_workqueue, &priv->flush_light);
181 }
182
183
184 static int
ipoib_stop(struct ipoib_dev_priv * priv)185 ipoib_stop(struct ipoib_dev_priv *priv)
186 {
187 if_t dev = priv->dev;
188
189 ipoib_dbg(priv, "stopping interface\n");
190
191 clear_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags);
192
193 if_setdrvflagbits(dev, 0, IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
194
195 ipoib_ib_dev_down(priv, 0);
196 ipoib_ib_dev_stop(priv, 0);
197
198 if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) {
199 struct ipoib_dev_priv *cpriv;
200
201 /* Bring down any child interfaces too */
202 mutex_lock(&priv->vlan_mutex);
203 list_for_each_entry(cpriv, &priv->child_intfs, list)
204 if ((if_getdrvflags(cpriv->dev) & IFF_DRV_RUNNING) != 0)
205 ipoib_stop(cpriv);
206 mutex_unlock(&priv->vlan_mutex);
207 }
208
209 return 0;
210 }
211
212 static int
ipoib_propagate_ifnet_mtu(struct ipoib_dev_priv * priv,int new_mtu,bool propagate)213 ipoib_propagate_ifnet_mtu(struct ipoib_dev_priv *priv, int new_mtu,
214 bool propagate)
215 {
216 if_t ifp;
217 struct ifreq ifr;
218 int error;
219
220 ifp = priv->dev;
221 if (if_getmtu(ifp) == new_mtu)
222 return (0);
223 if (propagate) {
224 strlcpy(ifr.ifr_name, if_name(ifp), IFNAMSIZ);
225 ifr.ifr_mtu = new_mtu;
226 CURVNET_SET(if_getvnet(ifp));
227 error = ifhwioctl(SIOCSIFMTU, ifp, (caddr_t)&ifr, curthread);
228 CURVNET_RESTORE();
229 } else {
230 if_setmtu(ifp, new_mtu);
231 error = 0;
232 }
233 return (error);
234 }
235
236 int
ipoib_change_mtu(struct ipoib_dev_priv * priv,int new_mtu,bool propagate)237 ipoib_change_mtu(struct ipoib_dev_priv *priv, int new_mtu, bool propagate)
238 {
239 int error, prev_admin_mtu;
240
241 /* dev->if_mtu > 2K ==> connected mode */
242 if (ipoib_cm_admin_enabled(priv)) {
243 if (new_mtu > IPOIB_CM_MTU(ipoib_cm_max_mtu(priv)))
244 return -EINVAL;
245
246 if (new_mtu > priv->mcast_mtu)
247 ipoib_warn(priv, "mtu > %d will cause multicast packet drops.\n",
248 priv->mcast_mtu);
249
250 return (ipoib_propagate_ifnet_mtu(priv, new_mtu, propagate));
251 }
252
253 if (new_mtu > IPOIB_UD_MTU(priv->max_ib_mtu))
254 return -EINVAL;
255
256 prev_admin_mtu = priv->admin_mtu;
257 priv->admin_mtu = new_mtu;
258 error = ipoib_propagate_ifnet_mtu(priv, min(priv->mcast_mtu,
259 priv->admin_mtu), propagate);
260 if (error == 0) {
261 /* check for MTU change to avoid infinite loop */
262 if (prev_admin_mtu != new_mtu)
263 queue_work(ipoib_workqueue, &priv->flush_light);
264 } else
265 priv->admin_mtu = prev_admin_mtu;
266 return (error);
267 }
268
269 static int
ipoib_ioctl(if_t ifp,u_long command,caddr_t data)270 ipoib_ioctl(if_t ifp, u_long command, caddr_t data)
271 {
272 struct ipoib_dev_priv *priv = if_getsoftc(ifp);
273 struct ifaddr *ifa = (struct ifaddr *) data;
274 struct ifreq *ifr = (struct ifreq *) data;
275 int error = 0;
276
277 /* check if detaching */
278 if (priv == NULL)
279 return (ENXIO);
280 /* wait for device to become ready, if any */
281 while (priv->gone == 2)
282 pause("W", 1);
283 /* check for device gone */
284 if (priv->gone != 0)
285 return (ENXIO);
286
287 switch (command) {
288 case SIOCSIFFLAGS:
289 if (if_getflags(ifp) & IFF_UP) {
290 if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0)
291 error = -ipoib_open(priv);
292 } else
293 if (if_getdrvflags(ifp) & IFF_DRV_RUNNING)
294 ipoib_stop(priv);
295 break;
296 case SIOCADDMULTI:
297 case SIOCDELMULTI:
298 if (if_getdrvflags(ifp) & IFF_DRV_RUNNING)
299 queue_work(ipoib_workqueue, &priv->restart_task);
300 break;
301 case SIOCSIFADDR:
302 if_setflagbits(ifp, IFF_UP, 0);
303
304 switch (ifa->ifa_addr->sa_family) {
305 #ifdef INET
306 case AF_INET:
307 if_init(ifp, if_getsoftc(ifp)); /* before arpwhohas */
308 arp_ifinit(ifp, ifa);
309 break;
310 #endif
311 default:
312 if_init(ifp, if_getsoftc(ifp));
313 break;
314 }
315 break;
316
317 case SIOCGIFADDR:
318 bcopy(if_getlladdr(ifp), &ifr->ifr_addr.sa_data[0],
319 INFINIBAND_ALEN);
320 break;
321
322 case SIOCSIFMTU:
323 /*
324 * Set the interface MTU.
325 */
326 error = -ipoib_change_mtu(priv, ifr->ifr_mtu, false);
327 break;
328 default:
329 error = EINVAL;
330 break;
331 }
332 return (error);
333 }
334
335
336 static struct ipoib_path *
__path_find(struct ipoib_dev_priv * priv,void * gid)337 __path_find(struct ipoib_dev_priv *priv, void *gid)
338 {
339 struct rb_node *n = priv->path_tree.rb_node;
340 struct ipoib_path *path;
341 int ret;
342
343 while (n) {
344 path = rb_entry(n, struct ipoib_path, rb_node);
345
346 ret = memcmp(gid, path->pathrec.dgid.raw,
347 sizeof (union ib_gid));
348
349 if (ret < 0)
350 n = n->rb_left;
351 else if (ret > 0)
352 n = n->rb_right;
353 else
354 return path;
355 }
356
357 return NULL;
358 }
359
360 static int
__path_add(struct ipoib_dev_priv * priv,struct ipoib_path * path)361 __path_add(struct ipoib_dev_priv *priv, struct ipoib_path *path)
362 {
363 struct rb_node **n = &priv->path_tree.rb_node;
364 struct rb_node *pn = NULL;
365 struct ipoib_path *tpath;
366 int ret;
367
368 while (*n) {
369 pn = *n;
370 tpath = rb_entry(pn, struct ipoib_path, rb_node);
371
372 ret = memcmp(path->pathrec.dgid.raw, tpath->pathrec.dgid.raw,
373 sizeof (union ib_gid));
374 if (ret < 0)
375 n = &pn->rb_left;
376 else if (ret > 0)
377 n = &pn->rb_right;
378 else
379 return -EEXIST;
380 }
381
382 rb_link_node(&path->rb_node, pn, n);
383 rb_insert_color(&path->rb_node, &priv->path_tree);
384
385 list_add_tail(&path->list, &priv->path_list);
386
387 return 0;
388 }
389
390 void
ipoib_path_free(struct ipoib_dev_priv * priv,struct ipoib_path * path)391 ipoib_path_free(struct ipoib_dev_priv *priv, struct ipoib_path *path)
392 {
393
394 _IF_DRAIN(&path->queue);
395
396 if (path->ah)
397 ipoib_put_ah(path->ah);
398 if (ipoib_cm_get(path))
399 ipoib_cm_destroy_tx(ipoib_cm_get(path));
400
401 kfree(path);
402 }
403
404 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
405
406 struct ipoib_path_iter *
ipoib_path_iter_init(struct ipoib_dev_priv * priv)407 ipoib_path_iter_init(struct ipoib_dev_priv *priv)
408 {
409 struct ipoib_path_iter *iter;
410
411 iter = kmalloc(sizeof *iter, GFP_KERNEL);
412 if (!iter)
413 return NULL;
414
415 iter->priv = priv;
416 memset(iter->path.pathrec.dgid.raw, 0, 16);
417
418 if (ipoib_path_iter_next(iter)) {
419 kfree(iter);
420 return NULL;
421 }
422
423 return iter;
424 }
425
426 int
ipoib_path_iter_next(struct ipoib_path_iter * iter)427 ipoib_path_iter_next(struct ipoib_path_iter *iter)
428 {
429 struct ipoib_dev_priv *priv = iter->priv;
430 struct rb_node *n;
431 struct ipoib_path *path;
432 int ret = 1;
433
434 spin_lock_irq(&priv->lock);
435
436 n = rb_first(&priv->path_tree);
437
438 while (n) {
439 path = rb_entry(n, struct ipoib_path, rb_node);
440
441 if (memcmp(iter->path.pathrec.dgid.raw, path->pathrec.dgid.raw,
442 sizeof (union ib_gid)) < 0) {
443 iter->path = *path;
444 ret = 0;
445 break;
446 }
447
448 n = rb_next(n);
449 }
450
451 spin_unlock_irq(&priv->lock);
452
453 return ret;
454 }
455
456 void
ipoib_path_iter_read(struct ipoib_path_iter * iter,struct ipoib_path * path)457 ipoib_path_iter_read(struct ipoib_path_iter *iter, struct ipoib_path *path)
458 {
459 *path = iter->path;
460 }
461
462 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */
463
464 void
ipoib_mark_paths_invalid(struct ipoib_dev_priv * priv)465 ipoib_mark_paths_invalid(struct ipoib_dev_priv *priv)
466 {
467 struct ipoib_path *path, *tp;
468
469 spin_lock_irq(&priv->lock);
470
471 list_for_each_entry_safe(path, tp, &priv->path_list, list) {
472 ipoib_dbg(priv, "mark path LID 0x%04x GID %16D invalid\n",
473 be16_to_cpu(path->pathrec.dlid),
474 path->pathrec.dgid.raw, ":");
475 path->valid = 0;
476 }
477
478 spin_unlock_irq(&priv->lock);
479 }
480
481 void
ipoib_flush_paths(struct ipoib_dev_priv * priv)482 ipoib_flush_paths(struct ipoib_dev_priv *priv)
483 {
484 struct ipoib_path *path, *tp;
485 LIST_HEAD(remove_list);
486 unsigned long flags;
487
488 spin_lock_irqsave(&priv->lock, flags);
489
490 list_splice_init(&priv->path_list, &remove_list);
491
492 list_for_each_entry(path, &remove_list, list)
493 rb_erase(&path->rb_node, &priv->path_tree);
494
495 list_for_each_entry_safe(path, tp, &remove_list, list) {
496 if (path->query)
497 ib_sa_cancel_query(path->query_id, path->query);
498 spin_unlock_irqrestore(&priv->lock, flags);
499 wait_for_completion(&path->done);
500 ipoib_path_free(priv, path);
501 spin_lock_irqsave(&priv->lock, flags);
502 }
503
504 spin_unlock_irqrestore(&priv->lock, flags);
505 }
506
507 static void
path_rec_completion(int status,struct ib_sa_path_rec * pathrec,void * path_ptr)508 path_rec_completion(int status, struct ib_sa_path_rec *pathrec, void *path_ptr)
509 {
510 struct ipoib_path *path = path_ptr;
511 struct ipoib_dev_priv *priv = path->priv;
512 if_t dev = priv->dev;
513 struct ipoib_ah *ah = NULL;
514 struct ipoib_ah *old_ah = NULL;
515 struct epoch_tracker et;
516 struct ifqueue mbqueue;
517 struct mbuf *mb;
518 unsigned long flags;
519
520 if (!status)
521 ipoib_dbg(priv, "PathRec LID 0x%04x for GID %16D\n",
522 be16_to_cpu(pathrec->dlid), pathrec->dgid.raw, ":");
523 else
524 ipoib_dbg(priv, "PathRec status %d for GID %16D\n",
525 status, path->pathrec.dgid.raw, ":");
526
527 bzero(&mbqueue, sizeof(mbqueue));
528
529 if (!status) {
530 struct ib_ah_attr av;
531
532 if (!ib_init_ah_from_path(priv->ca, priv->port, pathrec, &av))
533 ah = ipoib_create_ah(priv, priv->pd, &av);
534 }
535
536 spin_lock_irqsave(&priv->lock, flags);
537
538 if (ah) {
539 path->pathrec = *pathrec;
540
541 old_ah = path->ah;
542 path->ah = ah;
543
544 ipoib_dbg(priv, "created address handle %p for LID 0x%04x, SL %d\n",
545 ah, be16_to_cpu(pathrec->dlid), pathrec->sl);
546
547 for (;;) {
548 _IF_DEQUEUE(&path->queue, mb);
549 if (mb == NULL)
550 break;
551 _IF_ENQUEUE(&mbqueue, mb);
552 }
553
554 #ifdef CONFIG_INFINIBAND_IPOIB_CM
555 if (ipoib_cm_enabled(priv, path->hwaddr) && !ipoib_cm_get(path))
556 ipoib_cm_set(path, ipoib_cm_create_tx(priv, path));
557 #endif
558
559 path->valid = 1;
560 }
561
562 path->query = NULL;
563 complete(&path->done);
564
565 spin_unlock_irqrestore(&priv->lock, flags);
566
567 if (old_ah)
568 ipoib_put_ah(old_ah);
569
570 NET_EPOCH_ENTER(et);
571 for (;;) {
572 _IF_DEQUEUE(&mbqueue, mb);
573 if (mb == NULL)
574 break;
575 mb->m_pkthdr.rcvif = dev;
576 if (if_transmit(dev, mb))
577 ipoib_warn(priv, "dev_queue_xmit failed "
578 "to requeue packet\n");
579 }
580 NET_EPOCH_EXIT(et);
581 }
582
583 static struct ipoib_path *
path_rec_create(struct ipoib_dev_priv * priv,uint8_t * hwaddr)584 path_rec_create(struct ipoib_dev_priv *priv, uint8_t *hwaddr)
585 {
586 struct ipoib_path *path;
587
588 if (!priv->broadcast)
589 return NULL;
590
591 path = kzalloc(sizeof *path, GFP_ATOMIC);
592 if (!path)
593 return NULL;
594
595 path->priv = priv;
596
597 bzero(&path->queue, sizeof(path->queue));
598
599 #ifdef CONFIG_INFINIBAND_IPOIB_CM
600 memcpy(&path->hwaddr, hwaddr, INFINIBAND_ALEN);
601 #endif
602 memcpy(path->pathrec.dgid.raw, &hwaddr[4], sizeof (union ib_gid));
603 path->pathrec.sgid = priv->local_gid;
604 path->pathrec.pkey = cpu_to_be16(priv->pkey);
605 path->pathrec.numb_path = 1;
606 path->pathrec.traffic_class = priv->broadcast->mcmember.traffic_class;
607
608 return path;
609 }
610
611 static int
path_rec_start(struct ipoib_dev_priv * priv,struct ipoib_path * path)612 path_rec_start(struct ipoib_dev_priv *priv, struct ipoib_path *path)
613 {
614 if_t dev = priv->dev;
615
616 ib_sa_comp_mask comp_mask = IB_SA_PATH_REC_MTU_SELECTOR | IB_SA_PATH_REC_MTU;
617 struct ib_sa_path_rec p_rec;
618
619 p_rec = path->pathrec;
620 p_rec.mtu_selector = IB_SA_GT;
621
622 switch (roundup_pow_of_two(if_getmtu(dev) + IPOIB_ENCAP_LEN)) {
623 case 512:
624 p_rec.mtu = IB_MTU_256;
625 break;
626 case 1024:
627 p_rec.mtu = IB_MTU_512;
628 break;
629 case 2048:
630 p_rec.mtu = IB_MTU_1024;
631 break;
632 case 4096:
633 p_rec.mtu = IB_MTU_2048;
634 break;
635 default:
636 /* Wildcard everything */
637 comp_mask = 0;
638 p_rec.mtu = 0;
639 p_rec.mtu_selector = 0;
640 }
641
642 ipoib_dbg(priv, "Start path record lookup for %16D MTU > %d\n",
643 p_rec.dgid.raw, ":",
644 comp_mask ? ib_mtu_enum_to_int(p_rec.mtu) : 0);
645
646 init_completion(&path->done);
647
648 path->query_id =
649 ib_sa_path_rec_get(&ipoib_sa_client, priv->ca, priv->port,
650 &p_rec, comp_mask |
651 IB_SA_PATH_REC_DGID |
652 IB_SA_PATH_REC_SGID |
653 IB_SA_PATH_REC_NUMB_PATH |
654 IB_SA_PATH_REC_TRAFFIC_CLASS |
655 IB_SA_PATH_REC_PKEY,
656 1000, GFP_ATOMIC,
657 path_rec_completion,
658 path, &path->query);
659 if (path->query_id < 0) {
660 ipoib_warn(priv, "ib_sa_path_rec_get failed: %d\n", path->query_id);
661 path->query = NULL;
662 complete(&path->done);
663 return path->query_id;
664 }
665
666 return 0;
667 }
668
669 static void
ipoib_unicast_send(struct mbuf * mb,struct ipoib_dev_priv * priv,struct ipoib_header * eh)670 ipoib_unicast_send(struct mbuf *mb, struct ipoib_dev_priv *priv, struct ipoib_header *eh)
671 {
672 struct ipoib_path *path;
673
674 path = __path_find(priv, eh->hwaddr + 4);
675 if (!path || !path->valid) {
676 int new_path = 0;
677
678 if (!path) {
679 path = path_rec_create(priv, eh->hwaddr);
680 new_path = 1;
681 }
682 if (path) {
683 if (_IF_QLEN(&path->queue) < IPOIB_MAX_PATH_REC_QUEUE)
684 _IF_ENQUEUE(&path->queue, mb);
685 else {
686 if_inc_counter(priv->dev, IFCOUNTER_OERRORS, 1);
687 m_freem(mb);
688 }
689
690 if (!path->query && path_rec_start(priv, path)) {
691 if (new_path)
692 ipoib_path_free(priv, path);
693 return;
694 } else
695 __path_add(priv, path);
696 } else {
697 if_inc_counter(priv->dev, IFCOUNTER_OERRORS, 1);
698 m_freem(mb);
699 }
700
701 return;
702 }
703
704 if (ipoib_cm_get(path) && ipoib_cm_up(path)) {
705 ipoib_cm_send(priv, mb, ipoib_cm_get(path));
706 } else if (path->ah) {
707 ipoib_send(priv, mb, path->ah, IPOIB_QPN(eh->hwaddr));
708 } else if ((path->query || !path_rec_start(priv, path)) &&
709 path->queue.ifq_len < IPOIB_MAX_PATH_REC_QUEUE) {
710 _IF_ENQUEUE(&path->queue, mb);
711 } else {
712 if_inc_counter(priv->dev, IFCOUNTER_OERRORS, 1);
713 m_freem(mb);
714 }
715 }
716
717 static int
ipoib_send_one(struct ipoib_dev_priv * priv,struct mbuf * mb)718 ipoib_send_one(struct ipoib_dev_priv *priv, struct mbuf *mb)
719 {
720 struct ipoib_header *eh;
721
722 eh = mtod(mb, struct ipoib_header *);
723 if (IPOIB_IS_MULTICAST(eh->hwaddr)) {
724 /* Add in the P_Key for multicast*/
725 eh->hwaddr[8] = (priv->pkey >> 8) & 0xff;
726 eh->hwaddr[9] = priv->pkey & 0xff;
727
728 ipoib_mcast_send(priv, eh->hwaddr + 4, mb);
729 } else
730 ipoib_unicast_send(mb, priv, eh);
731
732 return 0;
733 }
734
735 void
ipoib_start_locked(if_t dev,struct ipoib_dev_priv * priv)736 ipoib_start_locked(if_t dev, struct ipoib_dev_priv *priv)
737 {
738 struct mbuf *mb;
739
740 assert_spin_locked(&priv->lock);
741
742 while (!if_sendq_empty(dev) &&
743 (if_getdrvflags(dev) & IFF_DRV_OACTIVE) == 0) {
744 mb = if_dequeue(dev);
745 if (mb == NULL)
746 break;
747 infiniband_bpf_mtap(dev, mb);
748 ipoib_send_one(priv, mb);
749 }
750 }
751
752 static void
_ipoib_start(if_t dev,struct ipoib_dev_priv * priv)753 _ipoib_start(if_t dev, struct ipoib_dev_priv *priv)
754 {
755
756 if ((if_getdrvflags(dev) & (IFF_DRV_RUNNING|IFF_DRV_OACTIVE)) !=
757 IFF_DRV_RUNNING)
758 return;
759
760 spin_lock(&priv->lock);
761 ipoib_start_locked(dev, priv);
762 spin_unlock(&priv->lock);
763 }
764
765 static void
ipoib_start(if_t dev)766 ipoib_start(if_t dev)
767 {
768 _ipoib_start(dev, if_getsoftc(dev));
769 }
770
771 static void
ipoib_vlan_start(if_t dev)772 ipoib_vlan_start(if_t dev)
773 {
774 struct ipoib_dev_priv *priv;
775 struct mbuf *mb;
776
777 priv = VLAN_COOKIE(dev);
778 if (priv != NULL)
779 return _ipoib_start(dev, priv);
780 while (!if_sendq_empty(dev)) {
781 mb = if_dequeue(dev);
782 if (mb == NULL)
783 break;
784 m_freem(mb);
785 if_inc_counter(dev, IFCOUNTER_OERRORS, 1);
786 }
787 }
788
789 int
ipoib_dev_init(struct ipoib_dev_priv * priv,struct ib_device * ca,int port)790 ipoib_dev_init(struct ipoib_dev_priv *priv, struct ib_device *ca, int port)
791 {
792
793 /* Allocate RX/TX "rings" to hold queued mbs */
794 priv->rx_ring = kzalloc(ipoib_recvq_size * sizeof *priv->rx_ring,
795 GFP_KERNEL);
796 if (!priv->rx_ring) {
797 printk(KERN_WARNING "%s: failed to allocate RX ring (%d entries)\n",
798 ca->name, ipoib_recvq_size);
799 goto out;
800 }
801
802 priv->tx_ring = kzalloc(ipoib_sendq_size * sizeof *priv->tx_ring, GFP_KERNEL);
803 if (!priv->tx_ring) {
804 printk(KERN_WARNING "%s: failed to allocate TX ring (%d entries)\n",
805 ca->name, ipoib_sendq_size);
806 goto out_rx_ring_cleanup;
807 }
808 memset(priv->tx_ring, 0, ipoib_sendq_size * sizeof *priv->tx_ring);
809
810 /* priv->tx_head, tx_tail & tx_outstanding are already 0 */
811
812 if (ipoib_ib_dev_init(priv, ca, port))
813 goto out_tx_ring_cleanup;
814
815 return 0;
816
817 out_tx_ring_cleanup:
818 kfree(priv->tx_ring);
819
820 out_rx_ring_cleanup:
821 kfree(priv->rx_ring);
822
823 out:
824 return -ENOMEM;
825 }
826
827 static void
ipoib_ifdetach(struct ipoib_dev_priv * priv)828 ipoib_ifdetach(struct ipoib_dev_priv *priv)
829 {
830 if_t dev;
831
832 dev = priv->dev;
833 if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) {
834 priv->gone = 1;
835 infiniband_ifdetach(dev);
836 }
837 }
838
839 static void
ipoib_detach(struct ipoib_dev_priv * priv)840 ipoib_detach(struct ipoib_dev_priv *priv)
841 {
842 if_t dev;
843
844 dev = priv->dev;
845 if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) {
846 if_free(dev);
847 free_unr(ipoib_unrhdr, priv->unit);
848 } else
849 VLAN_SETCOOKIE(priv->dev, NULL);
850
851 free(priv, M_TEMP);
852 }
853
854 void
ipoib_dev_cleanup(struct ipoib_dev_priv * priv)855 ipoib_dev_cleanup(struct ipoib_dev_priv *priv)
856 {
857 struct ipoib_dev_priv *cpriv, *tcpriv;
858
859 /* Delete any child interfaces first */
860 list_for_each_entry_safe(cpriv, tcpriv, &priv->child_intfs, list) {
861 ipoib_ifdetach(cpriv);
862 ipoib_dev_cleanup(cpriv);
863 ipoib_detach(cpriv);
864 }
865
866 ipoib_ib_dev_cleanup(priv);
867
868 kfree(priv->rx_ring);
869 kfree(priv->tx_ring);
870
871 priv->rx_ring = NULL;
872 priv->tx_ring = NULL;
873 }
874
875 static struct ipoib_dev_priv *
ipoib_priv_alloc(void)876 ipoib_priv_alloc(void)
877 {
878 struct ipoib_dev_priv *priv;
879
880 priv = malloc(sizeof(struct ipoib_dev_priv), M_TEMP, M_ZERO|M_WAITOK);
881 spin_lock_init(&priv->lock);
882 spin_lock_init(&priv->drain_lock);
883 mutex_init(&priv->vlan_mutex);
884 INIT_LIST_HEAD(&priv->path_list);
885 INIT_LIST_HEAD(&priv->child_intfs);
886 INIT_LIST_HEAD(&priv->dead_ahs);
887 INIT_LIST_HEAD(&priv->multicast_list);
888 INIT_DELAYED_WORK(&priv->pkey_poll_task, ipoib_pkey_poll);
889 INIT_DELAYED_WORK(&priv->mcast_task, ipoib_mcast_join_task);
890 INIT_WORK(&priv->carrier_on_task, ipoib_mcast_carrier_on_task);
891 INIT_WORK(&priv->flush_light, ipoib_ib_dev_flush_light);
892 INIT_WORK(&priv->flush_normal, ipoib_ib_dev_flush_normal);
893 INIT_WORK(&priv->flush_heavy, ipoib_ib_dev_flush_heavy);
894 INIT_WORK(&priv->restart_task, ipoib_mcast_restart_task);
895 INIT_DELAYED_WORK(&priv->ah_reap_task, ipoib_reap_ah);
896 memcpy(priv->broadcastaddr, ipv4_bcast_addr, INFINIBAND_ALEN);
897
898 return (priv);
899 }
900
901 struct ipoib_dev_priv *
ipoib_intf_alloc(const char * name,struct ib_device * hca)902 ipoib_intf_alloc(const char *name, struct ib_device *hca)
903 {
904 struct ipoib_dev_priv *priv;
905 if_t dev;
906
907 priv = ipoib_priv_alloc();
908 dev = priv->dev = if_alloc(IFT_INFINIBAND);
909 if_setsoftc(dev, priv);
910 priv->gone = 2; /* initializing */
911 priv->unit = alloc_unr(ipoib_unrhdr);
912 if (priv->unit == -1) {
913 if_free(dev);
914 free(priv, M_TEMP);
915 return NULL;
916 }
917 if_initname(dev, name, priv->unit);
918 if_setflags(dev, IFF_BROADCAST | IFF_MULTICAST);
919 if ((hca->attrs.device_cap_flags & IB_DEVICE_KNOWSEPOCH) == 0)
920 if_setflagbits(dev, IFF_NEEDSEPOCH, 0);
921
922 infiniband_ifattach(priv->dev, NULL, priv->broadcastaddr);
923
924 if_setinitfn(dev, ipoib_init);
925 if_setioctlfn(dev, ipoib_ioctl);
926 if_setstartfn(dev, ipoib_start);
927
928 if_setsendqlen(dev, ipoib_sendq_size * 2);
929
930 priv->dev = dev;
931 if_link_state_change(priv->dev, LINK_STATE_DOWN);
932
933 return if_getsoftc(dev);
934 }
935
936 int
ipoib_set_dev_features(struct ipoib_dev_priv * priv,struct ib_device * hca)937 ipoib_set_dev_features(struct ipoib_dev_priv *priv, struct ib_device *hca)
938 {
939 struct ib_device_attr *device_attr = &hca->attrs;
940
941 priv->hca_caps = device_attr->device_cap_flags;
942
943 if_sethwassist(priv->dev, 0);
944 if_setcapabilities(priv->dev, 0);
945
946 #ifndef CONFIG_INFINIBAND_IPOIB_CM
947 if (priv->hca_caps & IB_DEVICE_UD_IP_CSUM) {
948 set_bit(IPOIB_FLAG_CSUM, &priv->flags);
949 if_sethwassist(priv->dev, CSUM_IP | CSUM_TCP | CSUM_UDP);
950 if_setcapabilities(priv->dev, IFCAP_HWCSUM | IFCAP_VLAN_HWCSUM);
951 }
952
953 #if 0
954 if (priv->dev->features & NETIF_F_SG && priv->hca_caps & IB_DEVICE_UD_TSO) {
955 priv->dev->if_capabilities |= IFCAP_TSO4;
956 priv->dev->if_hwassist |= CSUM_TSO;
957 }
958 #endif
959 #endif
960 if_setcapabilitiesbit(priv->dev,
961 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_LINKSTATE, 0);
962 if_setcapenable(priv->dev, if_getcapabilities(priv->dev));
963
964 return 0;
965 }
966
967
968 static if_t
ipoib_add_port(const char * format,struct ib_device * hca,u8 port)969 ipoib_add_port(const char *format, struct ib_device *hca, u8 port)
970 {
971 struct ipoib_dev_priv *priv;
972 struct ib_port_attr attr;
973 int result = -ENOMEM;
974
975 priv = ipoib_intf_alloc(format, hca);
976 if (!priv)
977 goto alloc_mem_failed;
978
979 if (!ib_query_port(hca, port, &attr))
980 priv->max_ib_mtu = ib_mtu_enum_to_int(attr.max_mtu);
981 else {
982 printk(KERN_WARNING "%s: ib_query_port %d failed\n",
983 hca->name, port);
984 goto device_init_failed;
985 }
986
987 /* MTU will be reset when mcast join happens */
988 if_setmtu(priv->dev, IPOIB_UD_MTU(priv->max_ib_mtu));
989 priv->mcast_mtu = priv->admin_mtu = if_getmtu(priv->dev);
990
991 result = ib_query_pkey(hca, port, 0, &priv->pkey);
992 if (result) {
993 printk(KERN_WARNING "%s: ib_query_pkey port %d failed (ret = %d)\n",
994 hca->name, port, result);
995 goto device_init_failed;
996 }
997
998 if (ipoib_set_dev_features(priv, hca))
999 goto device_init_failed;
1000
1001 /*
1002 * Set the full membership bit, so that we join the right
1003 * broadcast group, etc.
1004 */
1005 priv->pkey |= 0x8000;
1006
1007 priv->broadcastaddr[8] = priv->pkey >> 8;
1008 priv->broadcastaddr[9] = priv->pkey & 0xff;
1009
1010 result = ib_query_gid(hca, port, 0, &priv->local_gid, NULL);
1011 if (result) {
1012 printk(KERN_WARNING "%s: ib_query_gid port %d failed (ret = %d)\n",
1013 hca->name, port, result);
1014 goto device_init_failed;
1015 }
1016 memcpy(if_getlladdr(priv->dev) + 4, priv->local_gid.raw, sizeof(union ib_gid));
1017
1018 result = ipoib_dev_init(priv, hca, port);
1019 if (result < 0) {
1020 printk(KERN_WARNING "%s: failed to initialize port %d (ret = %d)\n",
1021 hca->name, port, result);
1022 goto device_init_failed;
1023 }
1024 if (ipoib_cm_admin_enabled(priv))
1025 if_setmtu(priv->dev, IPOIB_CM_MTU(ipoib_cm_max_mtu(priv)));
1026
1027 INIT_IB_EVENT_HANDLER(&priv->event_handler,
1028 priv->ca, ipoib_event);
1029 result = ib_register_event_handler(&priv->event_handler);
1030 if (result < 0) {
1031 printk(KERN_WARNING "%s: ib_register_event_handler failed for "
1032 "port %d (ret = %d)\n",
1033 hca->name, port, result);
1034 goto event_failed;
1035 }
1036 if_printf(priv->dev, "Attached to %s port %d\n", hca->name, port);
1037
1038 priv->gone = 0; /* ready */
1039
1040 return priv->dev;
1041
1042 event_failed:
1043 ipoib_dev_cleanup(priv);
1044
1045 device_init_failed:
1046 ipoib_ifdetach(priv);
1047 ipoib_detach(priv);
1048
1049 alloc_mem_failed:
1050 return ERR_PTR(result);
1051 }
1052
1053 static void
ipoib_add_one(struct ib_device * device)1054 ipoib_add_one(struct ib_device *device)
1055 {
1056 struct list_head *dev_list;
1057 if_t dev;
1058 struct ipoib_dev_priv *priv;
1059 int s, e, p;
1060
1061 if (rdma_node_get_transport(device->node_type) != RDMA_TRANSPORT_IB)
1062 return;
1063
1064 dev_list = kmalloc(sizeof *dev_list, GFP_KERNEL);
1065 if (!dev_list)
1066 return;
1067
1068 INIT_LIST_HEAD(dev_list);
1069
1070 if (device->node_type == RDMA_NODE_IB_SWITCH) {
1071 s = 0;
1072 e = 0;
1073 } else {
1074 s = 1;
1075 e = device->phys_port_cnt;
1076 }
1077
1078 for (p = s; p <= e; ++p) {
1079 if (rdma_port_get_link_layer(device, p) != IB_LINK_LAYER_INFINIBAND)
1080 continue;
1081 dev = ipoib_add_port("ib", device, p);
1082 if (!IS_ERR(dev)) {
1083 priv = if_getsoftc(dev);
1084 list_add_tail(&priv->list, dev_list);
1085 }
1086 }
1087
1088 ib_set_client_data(device, &ipoib_client, dev_list);
1089 }
1090
1091 static void
ipoib_remove_one(struct ib_device * device,void * client_data)1092 ipoib_remove_one(struct ib_device *device, void *client_data)
1093 {
1094 struct ipoib_dev_priv *priv, *tmp;
1095 struct list_head *dev_list = client_data;
1096
1097 if (!dev_list)
1098 return;
1099
1100 if (rdma_node_get_transport(device->node_type) != RDMA_TRANSPORT_IB)
1101 return;
1102
1103 list_for_each_entry_safe(priv, tmp, dev_list, list) {
1104 if (rdma_port_get_link_layer(device, priv->port) != IB_LINK_LAYER_INFINIBAND)
1105 continue;
1106
1107 ipoib_ifdetach(priv);
1108 ipoib_stop(priv);
1109
1110 ib_unregister_event_handler(&priv->event_handler);
1111
1112 flush_workqueue(ipoib_workqueue);
1113
1114 ipoib_dev_cleanup(priv);
1115 ipoib_detach(priv);
1116 }
1117
1118 kfree(dev_list);
1119 }
1120
1121 static u_int
ipoib_match_dev_addr_cb(void * arg,struct ifaddr * ifa,u_int count)1122 ipoib_match_dev_addr_cb(void *arg, struct ifaddr *ifa, u_int count)
1123 {
1124 struct sockaddr *addr = arg;
1125
1126 /* If a match is already found, skip this. */
1127 if (count > 0)
1128 return (0);
1129
1130 if (ifa->ifa_addr->sa_len != addr->sa_len)
1131 return (0);
1132
1133 if (memcmp(ifa->ifa_addr, addr, addr->sa_len) == 0)
1134 return (1);
1135
1136 return (0);
1137 }
1138
1139 static int
ipoib_match_dev_addr(const struct sockaddr * addr,if_t dev)1140 ipoib_match_dev_addr(const struct sockaddr *addr, if_t dev)
1141 {
1142 struct epoch_tracker et;
1143 int retval = 0;
1144
1145 NET_EPOCH_ENTER(et);
1146 retval = if_foreach_addr_type(dev, addr->sa_family,
1147 ipoib_match_dev_addr_cb, __DECONST(void *, addr));
1148 NET_EPOCH_EXIT(et);
1149
1150 return (retval);
1151 }
1152
1153 /*
1154 * ipoib_match_gid_pkey_addr - returns the number of IPoIB netdevs on
1155 * top a given ipoib device matching a pkey_index and address, if one
1156 * exists.
1157 *
1158 * @found_net_dev: contains a matching net_device if the return value
1159 * >= 1, with a reference held.
1160 */
1161 static int
ipoib_match_gid_pkey_addr(struct ipoib_dev_priv * priv,const union ib_gid * gid,u16 pkey_index,const struct sockaddr * addr,if_t * found_net_dev)1162 ipoib_match_gid_pkey_addr(struct ipoib_dev_priv *priv,
1163 const union ib_gid *gid, u16 pkey_index, const struct sockaddr *addr,
1164 if_t *found_net_dev)
1165 {
1166 struct ipoib_dev_priv *child_priv;
1167 int matches = 0;
1168
1169 if (priv->pkey_index == pkey_index &&
1170 (!gid || !memcmp(gid, &priv->local_gid, sizeof(*gid)))) {
1171 if (addr == NULL || ipoib_match_dev_addr(addr, priv->dev) != 0) {
1172 if (*found_net_dev == NULL) {
1173 if_t net_dev;
1174
1175 if (priv->parent != NULL)
1176 net_dev = priv->parent;
1177 else
1178 net_dev = priv->dev;
1179 *found_net_dev = net_dev;
1180 dev_hold(net_dev);
1181 }
1182 matches++;
1183 }
1184 }
1185
1186 /* Check child interfaces */
1187 mutex_lock(&priv->vlan_mutex);
1188 list_for_each_entry(child_priv, &priv->child_intfs, list) {
1189 matches += ipoib_match_gid_pkey_addr(child_priv, gid,
1190 pkey_index, addr, found_net_dev);
1191 if (matches > 1)
1192 break;
1193 }
1194 mutex_unlock(&priv->vlan_mutex);
1195
1196 return matches;
1197 }
1198
1199 /*
1200 * __ipoib_get_net_dev_by_params - returns the number of matching
1201 * net_devs found (between 0 and 2). Also return the matching
1202 * net_device in the @net_dev parameter, holding a reference to the
1203 * net_device, if the number of matches >= 1
1204 */
1205 static int
__ipoib_get_net_dev_by_params(struct list_head * dev_list,u8 port,u16 pkey_index,const union ib_gid * gid,const struct sockaddr * addr,if_t * net_dev)1206 __ipoib_get_net_dev_by_params(struct list_head *dev_list, u8 port,
1207 u16 pkey_index, const union ib_gid *gid,
1208 const struct sockaddr *addr, if_t *net_dev)
1209 {
1210 struct ipoib_dev_priv *priv;
1211 int matches = 0;
1212
1213 *net_dev = NULL;
1214
1215 list_for_each_entry(priv, dev_list, list) {
1216 if (priv->port != port)
1217 continue;
1218
1219 matches += ipoib_match_gid_pkey_addr(priv, gid, pkey_index,
1220 addr, net_dev);
1221
1222 if (matches > 1)
1223 break;
1224 }
1225
1226 return matches;
1227 }
1228
1229 static if_t
ipoib_get_net_dev_by_params(struct ib_device * dev,u8 port,u16 pkey,const union ib_gid * gid,const struct sockaddr * addr,void * client_data)1230 ipoib_get_net_dev_by_params(struct ib_device *dev, u8 port, u16 pkey,
1231 const union ib_gid *gid, const struct sockaddr *addr, void *client_data)
1232 {
1233 if_t net_dev;
1234 struct list_head *dev_list = client_data;
1235 u16 pkey_index;
1236 int matches;
1237 int ret;
1238
1239 if (!rdma_protocol_ib(dev, port))
1240 return NULL;
1241
1242 ret = ib_find_cached_pkey(dev, port, pkey, &pkey_index);
1243 if (ret)
1244 return NULL;
1245
1246 if (!dev_list)
1247 return NULL;
1248
1249 /* See if we can find a unique device matching the L2 parameters */
1250 matches = __ipoib_get_net_dev_by_params(dev_list, port, pkey_index,
1251 gid, NULL, &net_dev);
1252
1253 switch (matches) {
1254 case 0:
1255 return NULL;
1256 case 1:
1257 return net_dev;
1258 }
1259
1260 dev_put(net_dev);
1261
1262 /* Couldn't find a unique device with L2 parameters only. Use L3
1263 * address to uniquely match the net device */
1264 matches = __ipoib_get_net_dev_by_params(dev_list, port, pkey_index,
1265 gid, addr, &net_dev);
1266 switch (matches) {
1267 case 0:
1268 return NULL;
1269 default:
1270 dev_warn_ratelimited(&dev->dev,
1271 "duplicate IP address detected\n");
1272 /* Fall through */
1273 case 1:
1274 return net_dev;
1275 }
1276 }
1277
1278 static void
ipoib_config_vlan(void * arg,if_t ifp,uint16_t vtag)1279 ipoib_config_vlan(void *arg, if_t ifp, uint16_t vtag)
1280 {
1281 struct ipoib_dev_priv *parent;
1282 struct ipoib_dev_priv *priv;
1283 struct epoch_tracker et;
1284 if_t dev;
1285 uint16_t pkey;
1286 int error;
1287
1288 if (if_gettype(ifp) != IFT_INFINIBAND)
1289 return;
1290 NET_EPOCH_ENTER(et);
1291 dev = VLAN_DEVAT(ifp, vtag);
1292 NET_EPOCH_EXIT(et);
1293 if (dev == NULL)
1294 return;
1295 priv = NULL;
1296 error = 0;
1297 parent = if_getsoftc(ifp);
1298 /* We only support 15 bits of pkey. */
1299 if (vtag & 0x8000)
1300 return;
1301 pkey = vtag | 0x8000; /* Set full membership bit. */
1302 if (pkey == parent->pkey)
1303 return;
1304 /* Check for dups */
1305 mutex_lock(&parent->vlan_mutex);
1306 list_for_each_entry(priv, &parent->child_intfs, list) {
1307 if (priv->pkey == pkey) {
1308 priv = NULL;
1309 error = EBUSY;
1310 goto out;
1311 }
1312 }
1313 priv = ipoib_priv_alloc();
1314 priv->dev = dev;
1315 priv->max_ib_mtu = parent->max_ib_mtu;
1316 priv->mcast_mtu = priv->admin_mtu = if_getmtu(parent->dev);
1317 set_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags);
1318 error = ipoib_set_dev_features(priv, parent->ca);
1319 if (error)
1320 goto out;
1321 priv->pkey = pkey;
1322 priv->broadcastaddr[8] = pkey >> 8;
1323 priv->broadcastaddr[9] = pkey & 0xff;
1324 if_setbroadcastaddr(dev, priv->broadcastaddr);
1325 error = ipoib_dev_init(priv, parent->ca, parent->port);
1326 if (error)
1327 goto out;
1328 priv->parent = parent->dev;
1329 list_add_tail(&priv->list, &parent->child_intfs);
1330 VLAN_SETCOOKIE(dev, priv);
1331 if_setstartfn(dev, ipoib_vlan_start);
1332 if_setdrvflagbits(dev, 0, IFF_DRV_RUNNING);
1333 if_setifheaderlen(dev, IPOIB_HEADER_LEN);
1334 if (if_getdrvflags(ifp) & IFF_DRV_RUNNING)
1335 ipoib_open(priv);
1336 mutex_unlock(&parent->vlan_mutex);
1337 return;
1338 out:
1339 mutex_unlock(&parent->vlan_mutex);
1340 if (priv)
1341 free(priv, M_TEMP);
1342 if (error)
1343 ipoib_warn(parent,
1344 "failed to initialize subinterface: device %s, port %d vtag 0x%X",
1345 parent->ca->name, parent->port, vtag);
1346 return;
1347 }
1348
1349 static void
ipoib_unconfig_vlan(void * arg,if_t ifp,uint16_t vtag)1350 ipoib_unconfig_vlan(void *arg, if_t ifp, uint16_t vtag)
1351 {
1352 struct ipoib_dev_priv *parent;
1353 struct ipoib_dev_priv *priv;
1354 struct epoch_tracker et;
1355 if_t dev;
1356 uint16_t pkey;
1357
1358 if (if_gettype(ifp) != IFT_INFINIBAND)
1359 return;
1360
1361 NET_EPOCH_ENTER(et);
1362 dev = VLAN_DEVAT(ifp, vtag);
1363 NET_EPOCH_EXIT(et);
1364 if (dev)
1365 VLAN_SETCOOKIE(dev, NULL);
1366 pkey = vtag | 0x8000;
1367 parent = if_getsoftc(ifp);
1368 mutex_lock(&parent->vlan_mutex);
1369 list_for_each_entry(priv, &parent->child_intfs, list) {
1370 if (priv->pkey == pkey) {
1371 ipoib_dev_cleanup(priv);
1372 list_del(&priv->list);
1373 break;
1374 }
1375 }
1376 mutex_unlock(&parent->vlan_mutex);
1377 }
1378
1379 eventhandler_tag ipoib_vlan_attach;
1380 eventhandler_tag ipoib_vlan_detach;
1381
1382 static int __init
ipoib_init_module(void)1383 ipoib_init_module(void)
1384 {
1385 int ret;
1386
1387 ipoib_recvq_size = roundup_pow_of_two(ipoib_recvq_size);
1388 ipoib_recvq_size = min(ipoib_recvq_size, IPOIB_MAX_QUEUE_SIZE);
1389 ipoib_recvq_size = max(ipoib_recvq_size, IPOIB_MIN_QUEUE_SIZE);
1390
1391 ipoib_sendq_size = roundup_pow_of_two(ipoib_sendq_size);
1392 ipoib_sendq_size = min(ipoib_sendq_size, IPOIB_MAX_QUEUE_SIZE);
1393 ipoib_sendq_size = max(ipoib_sendq_size, max(2 * MAX_SEND_CQE,
1394 IPOIB_MIN_QUEUE_SIZE));
1395 #ifdef CONFIG_INFINIBAND_IPOIB_CM
1396 ipoib_max_conn_qp = min(ipoib_max_conn_qp, IPOIB_CM_MAX_CONN_QP);
1397 #endif
1398
1399 ipoib_vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
1400 ipoib_config_vlan, NULL, EVENTHANDLER_PRI_FIRST);
1401 ipoib_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
1402 ipoib_unconfig_vlan, NULL, EVENTHANDLER_PRI_FIRST);
1403
1404 /*
1405 * We create our own workqueue mainly because we want to be
1406 * able to flush it when devices are being removed. We can't
1407 * use schedule_work()/flush_scheduled_work() because both
1408 * unregister_netdev() and linkwatch_event take the rtnl lock,
1409 * so flush_scheduled_work() can deadlock during device
1410 * removal.
1411 */
1412 ipoib_workqueue = create_singlethread_workqueue("ipoib");
1413 if (!ipoib_workqueue) {
1414 ret = -ENOMEM;
1415 goto err_fs;
1416 }
1417
1418 ib_sa_register_client(&ipoib_sa_client);
1419
1420 ret = ib_register_client(&ipoib_client);
1421 if (ret)
1422 goto err_sa;
1423
1424 return 0;
1425
1426 err_sa:
1427 ib_sa_unregister_client(&ipoib_sa_client);
1428 destroy_workqueue(ipoib_workqueue);
1429
1430 err_fs:
1431 return ret;
1432 }
1433
1434 static void __exit
ipoib_cleanup_module(void)1435 ipoib_cleanup_module(void)
1436 {
1437
1438 EVENTHANDLER_DEREGISTER(vlan_config, ipoib_vlan_attach);
1439 EVENTHANDLER_DEREGISTER(vlan_unconfig, ipoib_vlan_detach);
1440 ib_unregister_client(&ipoib_client);
1441 ib_sa_unregister_client(&ipoib_sa_client);
1442 destroy_workqueue(ipoib_workqueue);
1443 }
1444 module_init_order(ipoib_init_module, SI_ORDER_FIFTH);
1445 module_exit_order(ipoib_cleanup_module, SI_ORDER_FIFTH);
1446
1447 static int
ipoib_evhand(module_t mod,int event,void * arg)1448 ipoib_evhand(module_t mod, int event, void *arg)
1449 {
1450 return (0);
1451 }
1452
1453 static moduledata_t ipoib_mod = {
1454 .name = "ipoib",
1455 .evhand = ipoib_evhand,
1456 };
1457
1458 DECLARE_MODULE(ipoib, ipoib_mod, SI_SUB_LAST, SI_ORDER_ANY);
1459 MODULE_DEPEND(ipoib, ibcore, 1, 1, 1);
1460 MODULE_DEPEND(ipoib, if_infiniband, 1, 1, 1);
1461 MODULE_DEPEND(ipoib, linuxkpi, 1, 1, 1);
1462