projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge tag 'drm-fixes-for-v4.17-rc4' of git://people.freedesktop.org/~airlied/linux
[linux-2.6-block.git] / drivers / infiniband / ulp / ipoib / ipoib_main.c
CommitLineData
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
b63b70d8
SP		 49#include <linux/jhash.h>
50#include <net/arp.h>
ddde896e
GS		 51#include <net/addrconf.h>
52#include <linux/inetdevice.h>
53#include <rdma/ib_cache.h>
14c85021 54
4b48680b
YB		 55#define DRV_VERSION "1.0.0"
56
57const char ipoib_driver_version[] = DRV_VERSION;
58
1da177e4
LT		 59MODULE_AUTHOR("Roland Dreier");
60MODULE_DESCRIPTION("IP-over-InfiniBand net driver");
61MODULE_LICENSE("Dual BSD/GPL");
62
0f485251
SM		 63int ipoib_sendq_size __read_mostly = IPOIB_TX_RING_SIZE;
64int ipoib_recvq_size __read_mostly = IPOIB_RX_RING_SIZE;
65
66module_param_named(send_queue_size, ipoib_sendq_size, int, 0444);
67MODULE_PARM_DESC(send_queue_size, "Number of descriptors in send queue");
68module_param_named(recv_queue_size, ipoib_recvq_size, int, 0444);
69MODULE_PARM_DESC(recv_queue_size, "Number of descriptors in receive queue");
70
1da177e4
LT		 71#ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
72int ipoib_debug_level;
73
74module_param_named(debug_level, ipoib_debug_level, int, 0644);
75MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0");
76#endif
77
1732b0ef
RD		 78struct ipoib_path_iter {
79 struct net_device *dev;
80 struct ipoib_path path;
81};
82
1da177e4
LT		 83static const u8 ipv4_bcast_addr[] = {
84 0x00, 0xff, 0xff, 0xff,
85 0xff, 0x12, 0x40, 0x1b, 0x00, 0x00, 0x00, 0x00,
86 0x00, 0x00, 0x00, 0x00, 0xff, 0xff, 0xff, 0xff
87};
88
89struct workqueue_struct *ipoib_workqueue;
90
c1a0b23b
MT		 91struct ib_sa_client ipoib_sa_client;
92
1da177e4 93static void ipoib_add_one(struct ib_device *device);
7c1eb45a 94static void ipoib_remove_one(struct ib_device *device, void *client_data);
b63b70d8 95static void ipoib_neigh_reclaim(struct rcu_head *rp);
ddde896e
GS		 96static struct net_device *ipoib_get_net_dev_by_params(
97 struct ib_device *dev, u8 port, u16 pkey,
98 const union ib_gid *gid, const struct sockaddr *addr,
99 void *client_data);
492a7e67 100static int ipoib_set_mac(struct net_device *dev, void *addr);
31a82362
FD		 101static int ipoib_ioctl(struct net_device *dev, struct ifreq *ifr,
102 int cmd);
1da177e4
LT		 103
104static struct ib_client ipoib_client = {
105 .name = "ipoib",
106 .add = ipoib_add_one,
ddde896e
GS		 107 .remove = ipoib_remove_one,
108 .get_net_dev_by_params = ipoib_get_net_dev_by_params,
1da177e4
LT		 109};
110
771a5258
SR		 111#ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
112static int ipoib_netdev_event(struct notifier_block *this,
113 unsigned long event, void *ptr)
114{
115 struct netdev_notifier_info *ni = ptr;
116 struct net_device *dev = ni->dev;
117
118 if (dev->netdev_ops->ndo_open != ipoib_open)
119 return NOTIFY_DONE;
120
121 switch (event) {
122 case NETDEV_REGISTER:
123 ipoib_create_debug_files(dev);
124 break;
125 case NETDEV_CHANGENAME:
126 ipoib_delete_debug_files(dev);
127 ipoib_create_debug_files(dev);
128 break;
129 case NETDEV_UNREGISTER:
130 ipoib_delete_debug_files(dev);
131 break;
132 }
133
134 return NOTIFY_DONE;
135}
136#endif
137
1da177e4
LT		 138int ipoib_open(struct net_device *dev)
139{
c1048aff 140 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1da177e4
LT		 141
142 ipoib_dbg(priv, "bringing up interface\n");
143
437708c4
MS		 144 netif_carrier_off(dev);
145
e028cc55 146 set_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags);
1da177e4 147
3b561130
ES		 148 priv->sm_fullmember_sendonly_support = false;
149
efc82eee 150 if (ipoib_ib_dev_open(dev)) {
dd57c930
AE		 151 if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags))
152 return 0;
b8a1b1ce 153 goto err_disable;
dd57c930 154 }
fe25c561 155
5c37077f 156 ipoib_ib_dev_up(dev);
1da177e4
LT		 157
158 if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) {
159 struct ipoib_dev_priv *cpriv;
160
161 /* Bring up any child interfaces too */
f47944cc 162 down_read(&priv->vlan_rwsem);
1da177e4
LT		 163 list_for_each_entry(cpriv, &priv->child_intfs, list) {
164 int flags;
165
166 flags = cpriv->dev->flags;
167 if (flags & IFF_UP)
168 continue;
169
170 dev_change_flags(cpriv->dev, flags | IFF_UP);
171 }
f47944cc 172 up_read(&priv->vlan_rwsem);
1da177e4
LT		 173 }
174
175 netif_start_queue(dev);
176
177 return 0;
b8a1b1ce 178
b8a1b1ce 179err_disable:
b8a1b1ce
RD		 180 clear_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags);
181
182 return -EINVAL;
1da177e4
LT		 183}
184
185static int ipoib_stop(struct net_device *dev)
186{
c1048aff 187 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1da177e4
LT		 188
189 ipoib_dbg(priv, "stopping interface\n");
190
191 clear_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags);
192
193 netif_stop_queue(dev);
194
efc82eee 195 ipoib_ib_dev_down(dev);
cd565b4b 196 ipoib_ib_dev_stop(dev);
1da177e4
LT		 197
198 if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) {
199 struct ipoib_dev_priv *cpriv;
200
201 /* Bring down any child interfaces too */
f47944cc 202 down_read(&priv->vlan_rwsem);
1da177e4
LT		 203 list_for_each_entry(cpriv, &priv->child_intfs, list) {
204 int flags;
205
206 flags = cpriv->dev->flags;
207 if (!(flags & IFF_UP))
208 continue;
209
210 dev_change_flags(cpriv->dev, flags & ~IFF_UP);
211 }
f47944cc 212 up_read(&priv->vlan_rwsem);
1da177e4
LT		 213 }
214
215 return 0;
216}
217
9baa0b03
OG		 218static void ipoib_uninit(struct net_device *dev)
219{
220 ipoib_dev_cleanup(dev);
221}
222
9ca36f7d 223static netdev_features_t ipoib_fix_features(struct net_device *dev, netdev_features_t features)
3d96c74d 224{
c1048aff 225 struct ipoib_dev_priv *priv = ipoib_priv(dev);
3d96c74d
MM		 226
227 if (test_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags))
c4268778 228 features &= ~(NETIF_F_IP_CSUM | NETIF_F_TSO);
3d96c74d
MM		 229
230 return features;
231}
232
1da177e4
LT		 233static int ipoib_change_mtu(struct net_device *dev, int new_mtu)
234{
c1048aff 235 struct ipoib_dev_priv *priv = ipoib_priv(dev);
ed7b521d 236 int ret = 0;
1da177e4 237
839fcaba 238 /* dev->mtu > 2K ==> connected mode */
586a6934
PS		 239 if (ipoib_cm_admin_enabled(dev)) {
240 if (new_mtu > ipoib_cm_max_mtu(dev))
241 return -EINVAL;
242
839fcaba
MT		 243 if (new_mtu > priv->mcast_mtu)
244 ipoib_warn(priv, "mtu > %d will cause multicast packet drops.\n",
245 priv->mcast_mtu);
586a6934 246
839fcaba
MT		 247 dev->mtu = new_mtu;
248 return 0;
249 }
250
bc7b3a36 251 if (new_mtu > IPOIB_UD_MTU(priv->max_ib_mtu))
1da177e4
LT		 252 return -EINVAL;
253
254 priv->admin_mtu = new_mtu;
255
29da686d
FD		 256 if (priv->mcast_mtu < priv->admin_mtu)
257 ipoib_dbg(priv, "MTU must be smaller than the underlying "
258 "link layer MTU - 4 (%u)\n", priv->mcast_mtu);
259
ed7b521d 260 new_mtu = min(priv->mcast_mtu, priv->admin_mtu);
1da177e4 261
ed7b521d
ES		 262 if (priv->rn_ops->ndo_change_mtu) {
263 bool carrier_status = netif_carrier_ok(dev);
264
265 netif_carrier_off(dev);
266
267 /* notify lower level on the real mtu */
268 ret = priv->rn_ops->ndo_change_mtu(dev, new_mtu);
269
270 if (carrier_status)
271 netif_carrier_on(dev);
272 } else {
273 dev->mtu = new_mtu;
274 }
275
276 return ret;
1da177e4
LT		 277}
278
b6c871e5
ES		 279static void ipoib_get_stats(struct net_device *dev,
280 struct rtnl_link_stats64 *stats)
281{
282 struct ipoib_dev_priv *priv = ipoib_priv(dev);
283
284 if (priv->rn_ops->ndo_get_stats64)
285 priv->rn_ops->ndo_get_stats64(dev, stats);
286 else
287 netdev_stats_to_stats64(stats, &dev->stats);
288}
289
ddde896e
GS		 290/* Called with an RCU read lock taken */
291static bool ipoib_is_dev_match_addr_rcu(const struct sockaddr *addr,
292 struct net_device *dev)
293{
294 struct net *net = dev_net(dev);
295 struct in_device *in_dev;
296 struct sockaddr_in *addr_in = (struct sockaddr_in *)addr;
297 struct sockaddr_in6 *addr_in6 = (struct sockaddr_in6 *)addr;
298 __be32 ret_addr;
299
300 switch (addr->sa_family) {
301 case AF_INET:
302 in_dev = in_dev_get(dev);
303 if (!in_dev)
304 return false;
305
306 ret_addr = inet_confirm_addr(net, in_dev, 0,
307 addr_in->sin_addr.s_addr,
308 RT_SCOPE_HOST);
309 in_dev_put(in_dev);
310 if (ret_addr)
311 return true;
312
313 break;
314 case AF_INET6:
315 if (IS_ENABLED(CONFIG_IPV6) &&
316 ipv6_chk_addr(net, &addr_in6->sin6_addr, dev, 1))
317 return true;
318
319 break;
320 }
321 return false;
322}
323
324/**
325 * Find the master net_device on top of the given net_device.
326 * @dev: base IPoIB net_device
327 *
328 * Returns the master net_device with a reference held, or the same net_device
329 * if no master exists.
330 */
331static struct net_device *ipoib_get_master_net_dev(struct net_device *dev)
332{
333 struct net_device *master;
334
335 rcu_read_lock();
336 master = netdev_master_upper_dev_get_rcu(dev);
337 if (master)
338 dev_hold(master);
339 rcu_read_unlock();
340
341 if (master)
342 return master;
343
344 dev_hold(dev);
345 return dev;
346}
347
e0e79c8e
DA		 348struct ipoib_walk_data {
349 const struct sockaddr *addr;
350 struct net_device *result;
351};
352
353static int ipoib_upper_walk(struct net_device *upper, void *_data)
354{
355 struct ipoib_walk_data *data = _data;
356 int ret = 0;
357
358 if (ipoib_is_dev_match_addr_rcu(data->addr, upper)) {
359 dev_hold(upper);
360 data->result = upper;
361 ret = 1;
362 }
363
364 return ret;
365}
366
ddde896e
GS		 367/**
368 * Find a net_device matching the given address, which is an upper device of
369 * the given net_device.
370 * @addr: IP address to look for.
371 * @dev: base IPoIB net_device
372 *
373 * If found, returns the net_device with a reference held. Otherwise return
374 * NULL.
375 */
376static struct net_device *ipoib_get_net_dev_match_addr(
377 const struct sockaddr *addr, struct net_device *dev)
378{
e0e79c8e
DA		 379 struct ipoib_walk_data data = {
380 .addr = addr,
381 };
ddde896e
GS		 382
383 rcu_read_lock();
384 if (ipoib_is_dev_match_addr_rcu(addr, dev)) {
385 dev_hold(dev);
e0e79c8e 386 data.result = dev;
ddde896e
GS		 387 goto out;
388 }
389
e0e79c8e 390 netdev_walk_all_upper_dev_rcu(dev, ipoib_upper_walk, &data);
ddde896e
GS		 391out:
392 rcu_read_unlock();
e0e79c8e 393 return data.result;
ddde896e
GS		 394}
395
396/* returns the number of IPoIB netdevs on top a given ipoib device matching a
397 * pkey_index and address, if one exists.
398 *
399 * @found_net_dev: contains a matching net_device if the return value >= 1,
400 * with a reference held. */
401static int ipoib_match_gid_pkey_addr(struct ipoib_dev_priv *priv,
402 const union ib_gid *gid,
403 u16 pkey_index,
404 const struct sockaddr *addr,
405 int nesting,
406 struct net_device **found_net_dev)
407{
408 struct ipoib_dev_priv *child_priv;
409 struct net_device *net_dev = NULL;
410 int matches = 0;
411
412 if (priv->pkey_index == pkey_index &&
413 (!gid || !memcmp(gid, &priv->local_gid, sizeof(*gid)))) {
414 if (!addr) {
415 net_dev = ipoib_get_master_net_dev(priv->dev);
416 } else {
417 /* Verify the net_device matches the IP address, as
418 * IPoIB child devices currently share a GID. */
419 net_dev = ipoib_get_net_dev_match_addr(addr, priv->dev);
420 }
421 if (net_dev) {
422 if (!*found_net_dev)
423 *found_net_dev = net_dev;
424 else
425 dev_put(net_dev);
426 ++matches;
427 }
428 }
429
430 /* Check child interfaces */
431 down_read_nested(&priv->vlan_rwsem, nesting);
432 list_for_each_entry(child_priv, &priv->child_intfs, list) {
433 matches += ipoib_match_gid_pkey_addr(child_priv, gid,
434 pkey_index, addr,
435 nesting + 1,
436 found_net_dev);
437 if (matches > 1)
438 break;
439 }
440 up_read(&priv->vlan_rwsem);
441
442 return matches;
443}
444
445/* Returns the number of matching net_devs found (between 0 and 2). Also
446 * return the matching net_device in the @net_dev parameter, holding a
447 * reference to the net_device, if the number of matches >= 1 */
448static int __ipoib_get_net_dev_by_params(struct list_head *dev_list, u8 port,
449 u16 pkey_index,
450 const union ib_gid *gid,
451 const struct sockaddr *addr,
452 struct net_device **net_dev)
453{
454 struct ipoib_dev_priv *priv;
455 int matches = 0;
456
457 *net_dev = NULL;
458
459 list_for_each_entry(priv, dev_list, list) {
460 if (priv->port != port)
461 continue;
462
463 matches += ipoib_match_gid_pkey_addr(priv, gid, pkey_index,
464 addr, 0, net_dev);
465 if (matches > 1)
466 break;
467 }
468
469 return matches;
470}
471
472static struct net_device *ipoib_get_net_dev_by_params(
473 struct ib_device *dev, u8 port, u16 pkey,
474 const union ib_gid *gid, const struct sockaddr *addr,
475 void *client_data)
476{
477 struct net_device *net_dev;
478 struct list_head *dev_list = client_data;
479 u16 pkey_index;
480 int matches;
481 int ret;
482
483 if (!rdma_protocol_ib(dev, port))
484 return NULL;
485
486 ret = ib_find_cached_pkey(dev, port, pkey, &pkey_index);
487 if (ret)
488 return NULL;
489
490 if (!dev_list)
491 return NULL;
492
493 /* See if we can find a unique device matching the L2 parameters */
494 matches = __ipoib_get_net_dev_by_params(dev_list, port, pkey_index,
495 gid, NULL, &net_dev);
496
497 switch (matches) {
498 case 0:
499 return NULL;
500 case 1:
501 return net_dev;
502 }
503
504 dev_put(net_dev);
505
506 /* Couldn't find a unique device with L2 parameters only. Use L3
507 * address to uniquely match the net device */
508 matches = __ipoib_get_net_dev_by_params(dev_list, port, pkey_index,
509 gid, addr, &net_dev);
510 switch (matches) {
511 case 0:
512 return NULL;
513 default:
514 dev_warn_ratelimited(&dev->dev,
515 "duplicate IP address detected\n");
516 /* Fall through */
517 case 1:
518 return net_dev;
519 }
520}
521
71d9c5f9
RD		 522int ipoib_set_mode(struct net_device *dev, const char *buf)
523{
c1048aff 524 struct ipoib_dev_priv *priv = ipoib_priv(dev);
71d9c5f9 525
80b5b35a
FD		 526 if ((test_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags) &&
527 !strcmp(buf, "connected\n")) ||
528 (!test_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags) &&
529 !strcmp(buf, "datagram\n"))) {
530 return 0;
531 }
532
71d9c5f9
RD		 533 /* flush paths if we switch modes so that connections are restarted */
534 if (IPOIB_CM_SUPPORTED(dev->dev_addr) && !strcmp(buf, "connected\n")) {
535 set_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags);
536 ipoib_warn(priv, "enabling connected mode "
537 "will cause multicast packet drops\n");
538 netdev_update_features(dev);
edcd2a74 539 dev_set_mtu(dev, ipoib_cm_max_mtu(dev));
71d9c5f9 540 rtnl_unlock();
e622f2f4 541 priv->tx_wr.wr.send_flags &= ~IB_SEND_IP_CSUM;
71d9c5f9
RD		 542
543 ipoib_flush_paths(dev);
0a0007f2 544 return (!rtnl_trylock()) ? -EBUSY : 0;
71d9c5f9
RD		 545 }
546
547 if (!strcmp(buf, "datagram\n")) {
548 clear_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags);
549 netdev_update_features(dev);
550 dev_set_mtu(dev, min(priv->mcast_mtu, dev->mtu));
551 rtnl_unlock();
552 ipoib_flush_paths(dev);
0a0007f2 553 return (!rtnl_trylock()) ? -EBUSY : 0;
71d9c5f9
RD		 554 }
555
556 return -EINVAL;
557}
558
546481c2 559struct ipoib_path *__path_find(struct net_device *dev, void *gid)
1da177e4 560{
c1048aff 561 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1da177e4
LT		 562 struct rb_node *n = priv->path_tree.rb_node;
563 struct ipoib_path *path;
564 int ret;
565
566 while (n) {
567 path = rb_entry(n, struct ipoib_path, rb_node);
568
37c22a77 569 ret = memcmp(gid, path->pathrec.dgid.raw,
1da177e4
LT		 570 sizeof (union ib_gid));
571
572 if (ret < 0)
573 n = n->rb_left;
574 else if (ret > 0)
575 n = n->rb_right;
576 else
577 return path;
578 }
579
580 return NULL;
581}
582
583static int __path_add(struct net_device *dev, struct ipoib_path *path)
584{
c1048aff 585 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1da177e4
LT		 586 struct rb_node **n = &priv->path_tree.rb_node;
587 struct rb_node *pn = NULL;
588 struct ipoib_path *tpath;
589 int ret;
590
591 while (*n) {
592 pn = *n;
593 tpath = rb_entry(pn, struct ipoib_path, rb_node);
594
595 ret = memcmp(path->pathrec.dgid.raw, tpath->pathrec.dgid.raw,
596 sizeof (union ib_gid));
597 if (ret < 0)
598 n = &pn->rb_left;
599 else if (ret > 0)
600 n = &pn->rb_right;
601 else
602 return -EEXIST;
603 }
604
605 rb_link_node(&path->rb_node, pn, n);
606 rb_insert_color(&path->rb_node, &priv->path_tree);
607
608 list_add_tail(&path->list, &priv->path_list);
609
610 return 0;
611}
612
613static void path_free(struct net_device *dev, struct ipoib_path *path)
614{
1da177e4 615 struct sk_buff *skb;
1da177e4
LT		 616
617 while ((skb = __skb_dequeue(&path->queue)))
618 dev_kfree_skb_irq(skb);
619
c1048aff 620 ipoib_dbg(ipoib_priv(dev), "path_free\n");
1da177e4 621
b63b70d8
SP		 622 /* remove all neigh connected to this path */
623 ipoib_del_neighs_by_gid(dev, path->pathrec.dgid.raw);
1da177e4
LT		 624
625 if (path->ah)
626 ipoib_put_ah(path->ah);
627
628 kfree(path);
629}
630
1732b0ef
RD		 631#ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
632
633struct ipoib_path_iter *ipoib_path_iter_init(struct net_device *dev)
634{
635 struct ipoib_path_iter *iter;
636
637 iter = kmalloc(sizeof *iter, GFP_KERNEL);
638 if (!iter)
639 return NULL;
640
641 iter->dev = dev;
642 memset(iter->path.pathrec.dgid.raw, 0, 16);
643
644 if (ipoib_path_iter_next(iter)) {
645 kfree(iter);
646 return NULL;
647 }
648
649 return iter;
650}
651
652int ipoib_path_iter_next(struct ipoib_path_iter *iter)
653{
c1048aff 654 struct ipoib_dev_priv *priv = ipoib_priv(iter->dev);
1732b0ef
RD		 655 struct rb_node *n;
656 struct ipoib_path *path;
657 int ret = 1;
658
659 spin_lock_irq(&priv->lock);
660
661 n = rb_first(&priv->path_tree);
662
663 while (n) {
664 path = rb_entry(n, struct ipoib_path, rb_node);
665
666 if (memcmp(iter->path.pathrec.dgid.raw, path->pathrec.dgid.raw,
667 sizeof (union ib_gid)) < 0) {
668 iter->path = *path;
669 ret = 0;
670 break;
671 }
672
673 n = rb_next(n);
674 }
675
676 spin_unlock_irq(&priv->lock);
677
678 return ret;
679}
680
681void ipoib_path_iter_read(struct ipoib_path_iter *iter,
682 struct ipoib_path *path)
683{
684 *path = iter->path;
685}
686
687#endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */
688
ee1e2c82
MS		 689void ipoib_mark_paths_invalid(struct net_device *dev)
690{
c1048aff 691 struct ipoib_dev_priv *priv = ipoib_priv(dev);
ee1e2c82
MS		 692 struct ipoib_path *path, *tp;
693
694 spin_lock_irq(&priv->lock);
695
696 list_for_each_entry_safe(path, tp, &priv->path_list, list) {
57520751
DC		 697 ipoib_dbg(priv, "mark path LID 0x%08x GID %pI6 invalid\n",
698 be32_to_cpu(sa_path_get_dlid(&path->pathrec)),
699 path->pathrec.dgid.raw);
ee1e2c82
MS		 700 path->valid = 0;
701 }
702
703 spin_unlock_irq(&priv->lock);
704}
705
2b084176
ES		 706static void push_pseudo_header(struct sk_buff *skb, const char *daddr)
707{
708 struct ipoib_pseudo_header *phdr;
709
d58ff351 710 phdr = skb_push(skb, sizeof(*phdr));
2b084176
ES		 711 memcpy(phdr->hwaddr, daddr, INFINIBAND_ALEN);
712}
713
1da177e4
LT		 714void ipoib_flush_paths(struct net_device *dev)
715{
c1048aff 716 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1da177e4
LT		 717 struct ipoib_path *path, *tp;
718 LIST_HEAD(remove_list);
943c246e 719 unsigned long flags;
1da177e4 720
943c246e
RD		 721 netif_tx_lock_bh(dev);
722 spin_lock_irqsave(&priv->lock, flags);
1da177e4 723
157de229 724 list_splice_init(&priv->path_list, &remove_list);
1da177e4
LT		 725
726 list_for_each_entry(path, &remove_list, list)
727 rb_erase(&path->rb_node, &priv->path_tree);
728
1da177e4
LT		 729 list_for_each_entry_safe(path, tp, &remove_list, list) {
730 if (path->query)
731 ib_sa_cancel_query(path->query_id, path->query);
943c246e
RD		 732 spin_unlock_irqrestore(&priv->lock, flags);
733 netif_tx_unlock_bh(dev);
1da177e4
LT		 734 wait_for_completion(&path->done);
735 path_free(dev, path);
943c246e
RD		 736 netif_tx_lock_bh(dev);
737 spin_lock_irqsave(&priv->lock, flags);
1da177e4 738 }
943c246e
RD		 739
740 spin_unlock_irqrestore(&priv->lock, flags);
741 netif_tx_unlock_bh(dev);
1da177e4
LT		 742}
743
744static void path_rec_completion(int status,
c2f8fc4e 745 struct sa_path_rec *pathrec,
1da177e4
LT		 746 void *path_ptr)
747{
748 struct ipoib_path *path = path_ptr;
749 struct net_device *dev = path->dev;
c1048aff 750 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1da177e4 751 struct ipoib_ah *ah = NULL;
c9da4bad 752 struct ipoib_ah *old_ah = NULL;
d04d01b1 753 struct ipoib_neigh *neigh, *tn;
1da177e4
LT		 754 struct sk_buff_head skqueue;
755 struct sk_buff *skb;
756 unsigned long flags;
757
843613b0 758 if (!status)
5b095d98 759 ipoib_dbg(priv, "PathRec LID 0x%04x for GID %pI6\n",
57520751 760 be32_to_cpu(sa_path_get_dlid(pathrec)),
9fdca4da 761 pathrec->dgid.raw);
1da177e4 762 else
5b095d98 763 ipoib_dbg(priv, "PathRec status %d for GID %pI6\n",
fcace2fe 764 status, path->pathrec.dgid.raw);
1da177e4
LT		 765
766 skb_queue_head_init(&skqueue);
767
768 if (!status) {
90898850 769 struct rdma_ah_attr av;
46f1b3d7 770
4ad6a024
PP		 771 if (!ib_init_ah_attr_from_path(priv->ca, priv->port,
772 pathrec, &av))
46f1b3d7 773 ah = ipoib_create_ah(dev, priv->pd, &av);
1da177e4
LT		 774 }
775
776 spin_lock_irqsave(&priv->lock, flags);
777
3874397c 778 if (!IS_ERR_OR_NULL(ah)) {
43900089
ES		 779 /*
780 * pathrec.dgid is used as the database key from the LLADDR,
781 * it must remain unchanged even if the SA returns a different
782 * GID to use in the AH.
783 */
784 if (memcmp(pathrec->dgid.raw, path->pathrec.dgid.raw,
785 sizeof(union ib_gid))) {
786 ipoib_dbg(
787 priv,
788 "%s got PathRec for gid %pI6 while asked for %pI6\n",
789 dev->name, pathrec->dgid.raw,
790 path->pathrec.dgid.raw);
791 memcpy(pathrec->dgid.raw, path->pathrec.dgid.raw,
792 sizeof(union ib_gid));
793 }
794
1da177e4
LT		 795 path->pathrec = *pathrec;
796
c9da4bad
RD		 797 old_ah = path->ah;
798 path->ah = ah;
799
1da177e4 800 ipoib_dbg(priv, "created address handle %p for LID 0x%04x, SL %d\n",
57520751 801 ah, be32_to_cpu(sa_path_get_dlid(pathrec)),
9fdca4da 802 pathrec->sl);
1da177e4
LT		 803
804 while ((skb = __skb_dequeue(&path->queue)))
805 __skb_queue_tail(&skqueue, skb);
806
d04d01b1 807 list_for_each_entry_safe(neigh, tn, &path->neigh_list, list) {
ee1e2c82
MS		 808 if (neigh->ah) {
809 WARN_ON(neigh->ah != old_ah);
810 /*
811 * Dropping the ah reference inside
812 * priv->lock is safe here, because we
813 * will hold one more reference from
814 * the original value of path->ah (ie
815 * old_ah).
816 */
817 ipoib_put_ah(neigh->ah);
818 }
1da177e4
LT		 819 kref_get(&path->ah->ref);
820 neigh->ah = path->ah;
821
b63b70d8 822 if (ipoib_cm_enabled(dev, neigh->daddr)) {
839fcaba
MT		 823 if (!ipoib_cm_get(neigh))
824 ipoib_cm_set(neigh, ipoib_cm_create_tx(dev,
825 path,
826 neigh));
827 if (!ipoib_cm_get(neigh)) {
b63b70d8 828 ipoib_neigh_free(neigh);
839fcaba
MT		 829 continue;
830 }
831 }
832
1da177e4
LT		 833 while ((skb = __skb_dequeue(&neigh->queue)))
834 __skb_queue_tail(&skqueue, skb);
835 }
ee1e2c82 836 path->valid = 1;
5872a9fc 837 }
1da177e4 838
5872a9fc 839 path->query = NULL;
1da177e4
LT		 840 complete(&path->done);
841
842 spin_unlock_irqrestore(&priv->lock, flags);
843
f72dd566
RD		 844 if (IS_ERR_OR_NULL(ah))
845 ipoib_del_neighs_by_gid(dev, path->pathrec.dgid.raw);
846
ee1e2c82
MS		 847 if (old_ah)
848 ipoib_put_ah(old_ah);
849
1da177e4 850 while ((skb = __skb_dequeue(&skqueue))) {
d32b9a81 851 int ret;
1da177e4 852 skb->dev = dev;
d32b9a81
FD		 853 ret = dev_queue_xmit(skb);
854 if (ret)
855 ipoib_warn(priv, "%s: dev_queue_xmit failed to re-queue packet, ret:%d\n",
856 __func__, ret);
1da177e4
LT		 857 }
858}
859
98aebc55
ES		 860static void init_path_rec(struct ipoib_dev_priv *priv, struct ipoib_path *path,
861 void *gid)
862{
863 path->dev = priv->dev;
864
865 if (rdma_cap_opa_ah(priv->ca, priv->port))
866 path->pathrec.rec_type = SA_PATH_REC_TYPE_OPA;
867 else
868 path->pathrec.rec_type = SA_PATH_REC_TYPE_IB;
869
870 memcpy(path->pathrec.dgid.raw, gid, sizeof(union ib_gid));
871 path->pathrec.sgid = priv->local_gid;
872 path->pathrec.pkey = cpu_to_be16(priv->pkey);
873 path->pathrec.numb_path = 1;
874 path->pathrec.traffic_class = priv->broadcast->mcmember.traffic_class;
875}
876
37c22a77 877static struct ipoib_path *path_rec_create(struct net_device *dev, void *gid)
1da177e4 878{
c1048aff 879 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1da177e4
LT		 880 struct ipoib_path *path;
881
1401b53a
JM		 882 if (!priv->broadcast)
883 return NULL;
884
21a38489 885 path = kzalloc(sizeof *path, GFP_ATOMIC);
1da177e4
LT		 886 if (!path)
887 return NULL;
888
1da177e4
LT		 889 skb_queue_head_init(&path->queue);
890
891 INIT_LIST_HEAD(&path->neigh_list);
1da177e4 892
98aebc55 893 init_path_rec(priv, path, gid);
1da177e4
LT		 894
895 return path;
896}
897
898static int path_rec_start(struct net_device *dev,
899 struct ipoib_path *path)
900{
c1048aff 901 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1da177e4 902
5b095d98 903 ipoib_dbg(priv, "Start path record lookup for %pI6\n",
fcace2fe 904 path->pathrec.dgid.raw);
1da177e4 905
65c7edda
RD		 906 init_completion(&path->done);
907
1da177e4 908 path->query_id =
c1a0b23b 909 ib_sa_path_rec_get(&ipoib_sa_client, priv->ca, priv->port,
1da177e4
LT		 910 &path->pathrec,
911 IB_SA_PATH_REC_DGID |
912 IB_SA_PATH_REC_SGID |
913 IB_SA_PATH_REC_NUMB_PATH |
81668838 914 IB_SA_PATH_REC_TRAFFIC_CLASS |
1da177e4
LT		 915 IB_SA_PATH_REC_PKEY,
916 1000, GFP_ATOMIC,
917 path_rec_completion,
918 path, &path->query);
919 if (path->query_id < 0) {
01b3fc8b 920 ipoib_warn(priv, "ib_sa_path_rec_get failed: %d\n", path->query_id);
1da177e4 921 path->query = NULL;
93a3ab93 922 complete(&path->done);
1da177e4
LT		 923 return path->query_id;
924 }
925
926 return 0;
927}
928
16ba3def
ES		 929static struct ipoib_neigh *neigh_add_path(struct sk_buff *skb, u8 *daddr,
930 struct net_device *dev)
1da177e4 931{
c1048aff 932 struct ipoib_dev_priv *priv = ipoib_priv(dev);
cd565b4b 933 struct rdma_netdev *rn = netdev_priv(dev);
1da177e4
LT		 934 struct ipoib_path *path;
935 struct ipoib_neigh *neigh;
943c246e 936 unsigned long flags;
1da177e4 937
b5120a6e 938 spin_lock_irqsave(&priv->lock, flags);
b63b70d8 939 neigh = ipoib_neigh_alloc(daddr, dev);
1da177e4 940 if (!neigh) {
b5120a6e 941 spin_unlock_irqrestore(&priv->lock, flags);
de903512 942 ++dev->stats.tx_dropped;
1da177e4 943 dev_kfree_skb_any(skb);
16ba3def
ES		 944 return NULL;
945 }
946
947 /* To avoid race condition, make sure that the
948 * neigh will be added only once.
949 */
950 if (unlikely(!list_empty(&neigh->list))) {
951 spin_unlock_irqrestore(&priv->lock, flags);
952 return neigh;
1da177e4
LT		 953 }
954
b63b70d8 955 path = __path_find(dev, daddr + 4);
1da177e4 956 if (!path) {
b63b70d8 957 path = path_rec_create(dev, daddr + 4);
1da177e4 958 if (!path)
d2e0655e 959 goto err_path;
1da177e4
LT		 960
961 __path_add(dev, path);
962 }
963
964 list_add_tail(&neigh->list, &path->neigh_list);
965
47f7a071 966 if (path->ah) {
1da177e4
LT		 967 kref_get(&path->ah->ref);
968 neigh->ah = path->ah;
969
b63b70d8 970 if (ipoib_cm_enabled(dev, neigh->daddr)) {
839fcaba
MT		 971 if (!ipoib_cm_get(neigh))
972 ipoib_cm_set(neigh, ipoib_cm_create_tx(dev, path, neigh));
973 if (!ipoib_cm_get(neigh)) {
b63b70d8 974 ipoib_neigh_free(neigh);
839fcaba
MT		 975 goto err_drop;
976 }
fc791b63
PA		 977 if (skb_queue_len(&neigh->queue) <
978 IPOIB_MAX_PATH_REC_QUEUE) {
2b084176 979 push_pseudo_header(skb, neigh->daddr);
839fcaba 980 __skb_queue_tail(&neigh->queue, skb);
fc791b63 981 } else {
839fcaba
MT		 982 ipoib_warn(priv, "queue length limit %d. Packet drop.\n",
983 skb_queue_len(&neigh->queue));
984 goto err_drop;
985 }
721d67cd
RD		 986 } else {
987 spin_unlock_irqrestore(&priv->lock, flags);
cd565b4b
ES		 988 path->ah->last_send = rn->send(dev, skb, path->ah->ah,
989 IPOIB_QPN(daddr));
b63b70d8 990 ipoib_neigh_put(neigh);
16ba3def 991 return NULL;
721d67cd 992 }
1da177e4
LT		 993 } else {
994 neigh->ah = NULL;
1da177e4
LT		 995
996 if (!path->query && path_rec_start(dev, path))
49b8e744 997 goto err_path;
2b084176
ES		 998 if (skb_queue_len(&neigh->queue) < IPOIB_MAX_PATH_REC_QUEUE) {
999 push_pseudo_header(skb, neigh->daddr);
1e85b806 1000 __skb_queue_tail(&neigh->queue, skb);
2b084176 1001 } else {
1e85b806 1002 goto err_drop;
2b084176 1003 }
1da177e4
LT		 1004 }
1005
943c246e 1006 spin_unlock_irqrestore(&priv->lock, flags);
b63b70d8 1007 ipoib_neigh_put(neigh);
16ba3def 1008 return NULL;
1da177e4 1009
d2e0655e 1010err_path:
b63b70d8 1011 ipoib_neigh_free(neigh);
839fcaba 1012err_drop:
de903512 1013 ++dev->stats.tx_dropped;
1da177e4
LT		 1014 dev_kfree_skb_any(skb);
1015
943c246e 1016 spin_unlock_irqrestore(&priv->lock, flags);
b63b70d8 1017 ipoib_neigh_put(neigh);
16ba3def
ES		 1018
1019 return NULL;
1da177e4
LT		 1020}
1021
1022static void unicast_arp_send(struct sk_buff *skb, struct net_device *dev,
fc791b63 1023 struct ipoib_pseudo_header *phdr)
1da177e4 1024{
c1048aff 1025 struct ipoib_dev_priv *priv = ipoib_priv(dev);
cd565b4b 1026 struct rdma_netdev *rn = netdev_priv(dev);
1da177e4 1027 struct ipoib_path *path;
943c246e 1028 unsigned long flags;
1da177e4 1029
943c246e 1030 spin_lock_irqsave(&priv->lock, flags);
1da177e4 1031
98aebc55
ES		 1032 /* no broadcast means that all paths are (going to be) not valid */
1033 if (!priv->broadcast)
1034 goto drop_and_unlock;
1035
fc791b63 1036 path = __path_find(dev, phdr->hwaddr + 4);
ee1e2c82 1037 if (!path || !path->valid) {
71d98b46
JM		 1038 int new_path = 0;
1039
1040 if (!path) {
fc791b63 1041 path = path_rec_create(dev, phdr->hwaddr + 4);
71d98b46
JM		 1042 new_path = 1;
1043 }
1da177e4 1044 if (path) {
98aebc55
ES		 1045 if (!new_path)
1046 /* make sure there is no changes in the existing path record */
1047 init_path_rec(priv, path, phdr->hwaddr + 4);
1048
1e85b806 1049 if (skb_queue_len(&path->queue) < IPOIB_MAX_PATH_REC_QUEUE) {
2b084176 1050 push_pseudo_header(skb, phdr->hwaddr);
1e85b806
ES		 1051 __skb_queue_tail(&path->queue, skb);
1052 } else {
1053 ++dev->stats.tx_dropped;
1054 dev_kfree_skb_any(skb);
1055 }
1da177e4 1056
ff79ae80 1057 if (!path->query && path_rec_start(dev, path)) {
943c246e 1058 spin_unlock_irqrestore(&priv->lock, flags);
71d98b46
JM		 1059 if (new_path)
1060 path_free(dev, path);
1da177e4
LT		 1061 return;
1062 } else
1063 __path_add(dev, path);
1064 } else {
98aebc55 1065 goto drop_and_unlock;
1da177e4
LT		 1066 }
1067
943c246e 1068 spin_unlock_irqrestore(&priv->lock, flags);
1da177e4
LT		 1069 return;
1070 }
1071
47f7a071 1072 if (path->ah) {
57520751
DC		 1073 ipoib_dbg(priv, "Send unicast ARP to %08x\n",
1074 be32_to_cpu(sa_path_get_dlid(&path->pathrec)));
1da177e4 1075
721d67cd 1076 spin_unlock_irqrestore(&priv->lock, flags);
cd565b4b
ES		 1077 path->ah->last_send = rn->send(dev, skb, path->ah->ah,
1078 IPOIB_QPN(phdr->hwaddr));
721d67cd 1079 return;
1da177e4
LT		 1080 } else if ((path->query || !path_rec_start(dev, path)) &&
1081 skb_queue_len(&path->queue) < IPOIB_MAX_PATH_REC_QUEUE) {
2b084176 1082 push_pseudo_header(skb, phdr->hwaddr);
1da177e4
LT		 1083 __skb_queue_tail(&path->queue, skb);
1084 } else {
98aebc55 1085 goto drop_and_unlock;
1da177e4
LT		 1086 }
1087
943c246e 1088 spin_unlock_irqrestore(&priv->lock, flags);
98aebc55
ES		 1089 return;
1090
1091drop_and_unlock:
1092 ++dev->stats.tx_dropped;
1093 dev_kfree_skb_any(skb);
1094 spin_unlock_irqrestore(&priv->lock, flags);
1da177e4
LT		 1095}
1096
47a3968a 1097static netdev_tx_t ipoib_start_xmit(struct sk_buff *skb, struct net_device *dev)
1da177e4 1098{
c1048aff 1099 struct ipoib_dev_priv *priv = ipoib_priv(dev);
cd565b4b 1100 struct rdma_netdev *rn = netdev_priv(dev);
1da177e4 1101 struct ipoib_neigh *neigh;
fc791b63 1102 struct ipoib_pseudo_header *phdr;
b63b70d8 1103 struct ipoib_header *header;
1da177e4
LT		 1104 unsigned long flags;
1105
fc791b63
PA		 1106 phdr = (struct ipoib_pseudo_header *) skb->data;
1107 skb_pull(skb, sizeof(*phdr));
b63b70d8
SP		 1108 header = (struct ipoib_header *) skb->data;
1109
fc791b63 1110 if (unlikely(phdr->hwaddr[4] == 0xff)) {
b63b70d8
SP		 1111 /* multicast, arrange "if" according to probability */
1112 if ((header->proto != htons(ETH_P_IP)) &&
1113 (header->proto != htons(ETH_P_IPV6)) &&
1114 (header->proto != htons(ETH_P_ARP)) &&
dc850b0e
PM		 1115 (header->proto != htons(ETH_P_RARP)) &&
1116 (header->proto != htons(ETH_P_TIPC))) {
b63b70d8 1117 /* ethertype not supported by IPoIB */
17e6abee
DM		 1118 ++dev->stats.tx_dropped;
1119 dev_kfree_skb_any(skb);
b63b70d8 1120 return NETDEV_TX_OK;
17e6abee 1121 }
b63b70d8 1122 /* Add in the P_Key for multicast*/
fc791b63
PA		 1123 phdr->hwaddr[8] = (priv->pkey >> 8) & 0xff;
1124 phdr->hwaddr[9] = priv->pkey & 0xff;
b63b70d8 1125
fc791b63 1126 neigh = ipoib_neigh_get(dev, phdr->hwaddr);
b63b70d8
SP		 1127 if (likely(neigh))
1128 goto send_using_neigh;
fc791b63 1129 ipoib_mcast_send(dev, phdr->hwaddr, skb);
b63b70d8 1130 return NETDEV_TX_OK;
17e6abee 1131 }
1da177e4 1132
b63b70d8
SP		 1133 /* unicast, arrange "switch" according to probability */
1134 switch (header->proto) {
1135 case htons(ETH_P_IP):
1136 case htons(ETH_P_IPV6):
dc850b0e 1137 case htons(ETH_P_TIPC):
fc791b63 1138 neigh = ipoib_neigh_get(dev, phdr->hwaddr);
b63b70d8 1139 if (unlikely(!neigh)) {
16ba3def
ES		 1140 neigh = neigh_add_path(skb, phdr->hwaddr, dev);
1141 if (likely(!neigh))
1142 return NETDEV_TX_OK;
a50df398 1143 }
b63b70d8
SP		 1144 break;
1145 case htons(ETH_P_ARP):
1146 case htons(ETH_P_RARP):
1147 /* for unicast ARP and RARP should always perform path find */
fc791b63 1148 unicast_arp_send(skb, dev, phdr);
b63b70d8
SP		 1149 return NETDEV_TX_OK;
1150 default:
1151 /* ethertype not supported by IPoIB */
1152 ++dev->stats.tx_dropped;
1153 dev_kfree_skb_any(skb);
1154 return NETDEV_TX_OK;
1155 }
8a7f7521 1156
b63b70d8
SP		 1157send_using_neigh:
1158 /* note we now hold a ref to neigh */
1159 if (ipoib_cm_get(neigh)) {
1160 if (ipoib_cm_up(neigh)) {
1161 ipoib_cm_send(dev, skb, ipoib_cm_get(neigh));
1162 goto unref;
1da177e4 1163 }
b63b70d8 1164 } else if (neigh->ah) {
cd565b4b
ES		 1165 neigh->ah->last_send = rn->send(dev, skb, neigh->ah->ah,
1166 IPOIB_QPN(phdr->hwaddr));
b63b70d8
SP		 1167 goto unref;
1168 }
1da177e4 1169
b63b70d8 1170 if (skb_queue_len(&neigh->queue) < IPOIB_MAX_PATH_REC_QUEUE) {
2b084176 1171 push_pseudo_header(skb, phdr->hwaddr);
b63b70d8
SP		 1172 spin_lock_irqsave(&priv->lock, flags);
1173 __skb_queue_tail(&neigh->queue, skb);
1174 spin_unlock_irqrestore(&priv->lock, flags);
1da177e4 1175 } else {
b63b70d8
SP		 1176 ++dev->stats.tx_dropped;
1177 dev_kfree_skb_any(skb);
1178 }
1da177e4 1179
b63b70d8
SP		 1180unref:
1181 ipoib_neigh_put(neigh);
1da177e4 1182
1da177e4
LT		 1183 return NETDEV_TX_OK;
1184}
1185
1da177e4
LT		 1186static void ipoib_timeout(struct net_device *dev)
1187{
c1048aff 1188 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1da177e4 1189
4b2d319b 1190 ipoib_warn(priv, "transmit timeout: latency %d msecs\n",
4d0e9657 1191 jiffies_to_msecs(jiffies - dev_trans_start(dev)));
4b2d319b
RD		 1192 ipoib_warn(priv, "queue stopped %d, tx_head %u, tx_tail %u\n",
1193 netif_queue_stopped(dev),
1194 priv->tx_head, priv->tx_tail);
1da177e4
LT		 1195 /* XXX reset QP, etc. */
1196}
1197
1198static int ipoib_hard_header(struct sk_buff *skb,
1199 struct net_device *dev,
1200 unsigned short type,
3b04ddde 1201 const void *daddr, const void *saddr, unsigned len)
1da177e4
LT		 1202{
1203 struct ipoib_header *header;
1204
d58ff351 1205 header = skb_push(skb, sizeof *header);
1da177e4
LT		 1206
1207 header->proto = htons(type);
1208 header->reserved = 0;
1209
1210 /*
b63b70d8 1211 * we don't rely on dst_entry structure, always stuff the
fc791b63 1212 * destination address into skb hard header so we can figure out where
936d7de3 1213 * to send the packet later.
1da177e4 1214 */
2b084176 1215 push_pseudo_header(skb, daddr);
1da177e4 1216
fc791b63 1217 return IPOIB_HARD_LEN;
1da177e4
LT		 1218}
1219
1220static void ipoib_set_mcast_list(struct net_device *dev)
1221{
c1048aff 1222 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1da177e4 1223
7a343d4c
LA		 1224 if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags)) {
1225 ipoib_dbg(priv, "IPOIB_FLAG_OPER_UP not set");
1226 return;
1227 }
1228
0b39578b 1229 queue_work(priv->wq, &priv->restart_task);
1da177e4
LT		 1230}
1231
5aa7add8
ND		 1232static int ipoib_get_iflink(const struct net_device *dev)
1233{
c1048aff 1234 struct ipoib_dev_priv *priv = ipoib_priv(dev);
5aa7add8 1235
2c153959
ES		 1236 /* parent interface */
1237 if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags))
1238 return dev->ifindex;
1239
1240 /* child/vlan interface */
5aa7add8
ND		 1241 return priv->parent->ifindex;
1242}
1243
b63b70d8 1244static u32 ipoib_addr_hash(struct ipoib_neigh_hash *htbl, u8 *daddr)
1da177e4 1245{
b63b70d8
SP		 1246 /*
1247 * Use only the address parts that contributes to spreading
1248 * The subnet prefix is not used as one can not connect to
1249 * same remote port (GUID) using the same remote QPN via two
1250 * different subnets.
1251 */
1252 /* qpn octets[1:4) & port GUID octets[12:20) */
9d1ad66e 1253 u32 *d32 = (u32 *) daddr;
b63b70d8
SP		 1254 u32 hv;
1255
9d1ad66e 1256 hv = jhash_3words(d32[3], d32[4], IPOIB_QPN_MASK & d32[0], 0);
b63b70d8
SP		 1257 return hv & htbl->mask;
1258}
1259
1260struct ipoib_neigh *ipoib_neigh_get(struct net_device *dev, u8 *daddr)
1261{
c1048aff 1262 struct ipoib_dev_priv *priv = ipoib_priv(dev);
b63b70d8
SP		 1263 struct ipoib_neigh_table *ntbl = &priv->ntbl;
1264 struct ipoib_neigh_hash *htbl;
1265 struct ipoib_neigh *neigh = NULL;
1266 u32 hash_val;
1267
1268 rcu_read_lock_bh();
1269
1270 htbl = rcu_dereference_bh(ntbl->htbl);
1271
1272 if (!htbl)
1273 goto out_unlock;
1274
1275 hash_val = ipoib_addr_hash(htbl, daddr);
1276 for (neigh = rcu_dereference_bh(htbl->buckets[hash_val]);
1277 neigh != NULL;
1278 neigh = rcu_dereference_bh(neigh->hnext)) {
1279 if (memcmp(daddr, neigh->daddr, INFINIBAND_ALEN) == 0) {
1280 /* found, take one ref on behalf of the caller */
1281 if (!atomic_inc_not_zero(&neigh->refcnt)) {
1282 /* deleted */
1283 neigh = NULL;
1284 goto out_unlock;
1285 }
61c78eea
ES		 1286
1287 if (likely(skb_queue_len(&neigh->queue) < IPOIB_MAX_PATH_REC_QUEUE))
1288 neigh->alive = jiffies;
b63b70d8
SP		 1289 goto out_unlock;
1290 }
1291 }
1292
1293out_unlock:
1294 rcu_read_unlock_bh();
1295 return neigh;
1296}
1297
1298static void __ipoib_reap_neigh(struct ipoib_dev_priv *priv)
1299{
1300 struct ipoib_neigh_table *ntbl = &priv->ntbl;
1301 struct ipoib_neigh_hash *htbl;
1302 unsigned long neigh_obsolete;
1303 unsigned long dt;
1da177e4 1304 unsigned long flags;
b63b70d8 1305 int i;
bd99b2e0 1306 LIST_HEAD(remove_list);
1da177e4 1307
b63b70d8 1308 if (test_bit(IPOIB_STOP_NEIGH_GC, &priv->flags))
732a2170 1309 return;
1da177e4 1310
b5120a6e 1311 spin_lock_irqsave(&priv->lock, flags);
b63b70d8
SP		 1312
1313 htbl = rcu_dereference_protected(ntbl->htbl,
b5120a6e 1314 lockdep_is_held(&priv->lock));
b63b70d8
SP		 1315
1316 if (!htbl)
1317 goto out_unlock;
1318
1319 /* neigh is obsolete if it was idle for two GC periods */
1320 dt = 2 * arp_tbl.gc_interval;
1321 neigh_obsolete = jiffies - dt;
1322 /* handle possible race condition */
1323 if (test_bit(IPOIB_STOP_NEIGH_GC, &priv->flags))
1324 goto out_unlock;
1325
1326 for (i = 0; i < htbl->size; i++) {
1327 struct ipoib_neigh *neigh;
1328 struct ipoib_neigh __rcu **np = &htbl->buckets[i];
1329
1330 while ((neigh = rcu_dereference_protected(*np,
b5120a6e 1331 lockdep_is_held(&priv->lock))) != NULL) {
b63b70d8
SP		 1332 /* was the neigh idle for two GC periods */
1333 if (time_after(neigh_obsolete, neigh->alive)) {
bd99b2e0 1334
432c55ff 1335 ipoib_check_and_add_mcast_sendonly(priv, neigh->daddr + 4, &remove_list);
bd99b2e0 1336
b63b70d8
SP		 1337 rcu_assign_pointer(*np,
1338 rcu_dereference_protected(neigh->hnext,
b5120a6e 1339 lockdep_is_held(&priv->lock)));
b63b70d8 1340 /* remove from path/mc list */
c586071d 1341 list_del_init(&neigh->list);
b63b70d8
SP		 1342 call_rcu(&neigh->rcu, ipoib_neigh_reclaim);
1343 } else {
1344 np = &neigh->hnext;
1345 }
1da177e4 1346
b63b70d8
SP		 1347 }
1348 }
1da177e4 1349
b63b70d8 1350out_unlock:
b5120a6e 1351 spin_unlock_irqrestore(&priv->lock, flags);
50be28de 1352 ipoib_mcast_remove_list(&remove_list);
b63b70d8 1353}
1da177e4 1354
b63b70d8
SP		 1355static void ipoib_reap_neigh(struct work_struct *work)
1356{
1357 struct ipoib_dev_priv *priv =
1358 container_of(work, struct ipoib_dev_priv, neigh_reap_task.work);
1359
1360 __ipoib_reap_neigh(priv);
1361
1362 if (!test_bit(IPOIB_STOP_NEIGH_GC, &priv->flags))
0b39578b 1363 queue_delayed_work(priv->wq, &priv->neigh_reap_task,
b63b70d8 1364 arp_tbl.gc_interval);
1da177e4
LT		 1365}
1366
b63b70d8
SP		 1367
1368static struct ipoib_neigh *ipoib_neigh_ctor(u8 *daddr,
732a2170 1369 struct net_device *dev)
d2e0655e
MT		 1370{
1371 struct ipoib_neigh *neigh;
1372
b63b70d8 1373 neigh = kzalloc(sizeof *neigh, GFP_ATOMIC);
d2e0655e
MT		 1374 if (!neigh)
1375 return NULL;
1376
732a2170 1377 neigh->dev = dev;
b63b70d8 1378 memcpy(&neigh->daddr, daddr, sizeof(neigh->daddr));
82b39913 1379 skb_queue_head_init(&neigh->queue);
b63b70d8 1380 INIT_LIST_HEAD(&neigh->list);
839fcaba 1381 ipoib_cm_set(neigh, NULL);
b63b70d8
SP		 1382 /* one ref on behalf of the caller */
1383 atomic_set(&neigh->refcnt, 1);
1384
1385 return neigh;
1386}
1387
1388struct ipoib_neigh *ipoib_neigh_alloc(u8 *daddr,
1389 struct net_device *dev)
1390{
c1048aff 1391 struct ipoib_dev_priv *priv = ipoib_priv(dev);
b63b70d8
SP		 1392 struct ipoib_neigh_table *ntbl = &priv->ntbl;
1393 struct ipoib_neigh_hash *htbl;
1394 struct ipoib_neigh *neigh;
1395 u32 hash_val;
1396
b63b70d8 1397 htbl = rcu_dereference_protected(ntbl->htbl,
b5120a6e 1398 lockdep_is_held(&priv->lock));
b63b70d8
SP		 1399 if (!htbl) {
1400 neigh = NULL;
1401 goto out_unlock;
1402 }
1403
1404 /* need to add a new neigh, but maybe some other thread succeeded?
1405 * recalc hash, maybe hash resize took place so we do a search
1406 */
1407 hash_val = ipoib_addr_hash(htbl, daddr);
1408 for (neigh = rcu_dereference_protected(htbl->buckets[hash_val],
b5120a6e 1409 lockdep_is_held(&priv->lock));
b63b70d8
SP		 1410 neigh != NULL;
1411 neigh = rcu_dereference_protected(neigh->hnext,
b5120a6e 1412 lockdep_is_held(&priv->lock))) {
b63b70d8
SP		 1413 if (memcmp(daddr, neigh->daddr, INFINIBAND_ALEN) == 0) {
1414 /* found, take one ref on behalf of the caller */
1415 if (!atomic_inc_not_zero(&neigh->refcnt)) {
1416 /* deleted */
1417 neigh = NULL;
1418 break;
1419 }
1420 neigh->alive = jiffies;
1421 goto out_unlock;
1422 }
1423 }
1424
1425 neigh = ipoib_neigh_ctor(daddr, dev);
1426 if (!neigh)
1427 goto out_unlock;
1428
1429 /* one ref on behalf of the hash table */
1430 atomic_inc(&neigh->refcnt);
1431 neigh->alive = jiffies;
1432 /* put in hash */
1433 rcu_assign_pointer(neigh->hnext,
1434 rcu_dereference_protected(htbl->buckets[hash_val],
b5120a6e 1435 lockdep_is_held(&priv->lock)));
b63b70d8
SP		 1436 rcu_assign_pointer(htbl->buckets[hash_val], neigh);
1437 atomic_inc(&ntbl->entries);
1438
1439out_unlock:
d2e0655e
MT		 1440
1441 return neigh;
1442}
1443
b63b70d8 1444void ipoib_neigh_dtor(struct ipoib_neigh *neigh)
d2e0655e 1445{
b63b70d8
SP		 1446 /* neigh reference count was dropprd to zero */
1447 struct net_device *dev = neigh->dev;
c1048aff 1448 struct ipoib_dev_priv *priv = ipoib_priv(dev);
2745b5b7 1449 struct sk_buff *skb;
b63b70d8
SP		 1450 if (neigh->ah)
1451 ipoib_put_ah(neigh->ah);
2745b5b7 1452 while ((skb = __skb_dequeue(&neigh->queue))) {
de903512 1453 ++dev->stats.tx_dropped;
2745b5b7
MT		 1454 dev_kfree_skb_any(skb);
1455 }
839fcaba
MT		 1456 if (ipoib_cm_get(neigh))
1457 ipoib_cm_destroy_tx(ipoib_cm_get(neigh));
c1048aff 1458 ipoib_dbg(ipoib_priv(dev),
b63b70d8
SP		 1459 "neigh free for %06x %pI6\n",
1460 IPOIB_QPN(neigh->daddr),
1461 neigh->daddr + 4);
d2e0655e 1462 kfree(neigh);
b63b70d8
SP		 1463 if (atomic_dec_and_test(&priv->ntbl.entries)) {
1464 if (test_bit(IPOIB_NEIGH_TBL_FLUSH, &priv->flags))
1465 complete(&priv->ntbl.flushed);
1466 }
1467}
1468
1469static void ipoib_neigh_reclaim(struct rcu_head *rp)
1470{
1471 /* Called as a result of removal from hash table */
1472 struct ipoib_neigh *neigh = container_of(rp, struct ipoib_neigh, rcu);
1473 /* note TX context may hold another ref */
1474 ipoib_neigh_put(neigh);
d2e0655e
MT		 1475}
1476
b63b70d8 1477void ipoib_neigh_free(struct ipoib_neigh *neigh)
1da177e4 1478{
b63b70d8 1479 struct net_device *dev = neigh->dev;
c1048aff 1480 struct ipoib_dev_priv *priv = ipoib_priv(dev);
b63b70d8
SP		 1481 struct ipoib_neigh_table *ntbl = &priv->ntbl;
1482 struct ipoib_neigh_hash *htbl;
1483 struct ipoib_neigh __rcu **np;
1484 struct ipoib_neigh *n;
1485 u32 hash_val;
1486
b63b70d8 1487 htbl = rcu_dereference_protected(ntbl->htbl,
b5120a6e 1488 lockdep_is_held(&priv->lock));
b63b70d8 1489 if (!htbl)
b5120a6e 1490 return;
b63b70d8
SP		 1491
1492 hash_val = ipoib_addr_hash(htbl, neigh->daddr);
1493 np = &htbl->buckets[hash_val];
1494 for (n = rcu_dereference_protected(*np,
b5120a6e 1495 lockdep_is_held(&priv->lock));
b63b70d8 1496 n != NULL;
6c723a68 1497 n = rcu_dereference_protected(*np,
b5120a6e 1498 lockdep_is_held(&priv->lock))) {
b63b70d8
SP		 1499 if (n == neigh) {
1500 /* found */
1501 rcu_assign_pointer(*np,
1502 rcu_dereference_protected(neigh->hnext,
b5120a6e 1503 lockdep_is_held(&priv->lock)));
49b8e744 1504 /* remove from parent list */
c586071d 1505 list_del_init(&neigh->list);
b63b70d8 1506 call_rcu(&neigh->rcu, ipoib_neigh_reclaim);
b5120a6e 1507 return;
b63b70d8
SP		 1508 } else {
1509 np = &n->hnext;
1510 }
1511 }
b63b70d8
SP		 1512}
1513
1514static int ipoib_neigh_hash_init(struct ipoib_dev_priv *priv)
1515{
1516 struct ipoib_neigh_table *ntbl = &priv->ntbl;
1517 struct ipoib_neigh_hash *htbl;
52374967 1518 struct ipoib_neigh __rcu **buckets;
b63b70d8
SP		 1519 u32 size;
1520
1521 clear_bit(IPOIB_NEIGH_TBL_FLUSH, &priv->flags);
1522 ntbl->htbl = NULL;
b63b70d8
SP		 1523 htbl = kzalloc(sizeof(*htbl), GFP_KERNEL);
1524 if (!htbl)
1525 return -ENOMEM;
1526 set_bit(IPOIB_STOP_NEIGH_GC, &priv->flags);
1527 size = roundup_pow_of_two(arp_tbl.gc_thresh3);
1528 buckets = kzalloc(size * sizeof(*buckets), GFP_KERNEL);
1529 if (!buckets) {
1530 kfree(htbl);
1531 return -ENOMEM;
1532 }
1533 htbl->size = size;
1534 htbl->mask = (size - 1);
1535 htbl->buckets = buckets;
52374967 1536 RCU_INIT_POINTER(ntbl->htbl, htbl);
66172c09 1537 htbl->ntbl = ntbl;
b63b70d8
SP		 1538 atomic_set(&ntbl->entries, 0);
1539
1540 /* start garbage collection */
1541 clear_bit(IPOIB_STOP_NEIGH_GC, &priv->flags);
0b39578b 1542 queue_delayed_work(priv->wq, &priv->neigh_reap_task,
b63b70d8 1543 arp_tbl.gc_interval);
1da177e4
LT		 1544
1545 return 0;
1546}
1547
b63b70d8
SP		 1548static void neigh_hash_free_rcu(struct rcu_head *head)
1549{
1550 struct ipoib_neigh_hash *htbl = container_of(head,
1551 struct ipoib_neigh_hash,
1552 rcu);
1553 struct ipoib_neigh __rcu **buckets = htbl->buckets;
66172c09 1554 struct ipoib_neigh_table *ntbl = htbl->ntbl;
b63b70d8
SP		 1555
1556 kfree(buckets);
1557 kfree(htbl);
66172c09 1558 complete(&ntbl->deleted);
b63b70d8
SP		 1559}
1560
1561void ipoib_del_neighs_by_gid(struct net_device *dev, u8 *gid)
1562{
c1048aff 1563 struct ipoib_dev_priv *priv = ipoib_priv(dev);
b63b70d8
SP		 1564 struct ipoib_neigh_table *ntbl = &priv->ntbl;
1565 struct ipoib_neigh_hash *htbl;
1566 unsigned long flags;
1567 int i;
1568
1569 /* remove all neigh connected to a given path or mcast */
b5120a6e 1570 spin_lock_irqsave(&priv->lock, flags);
b63b70d8
SP		 1571
1572 htbl = rcu_dereference_protected(ntbl->htbl,
b5120a6e 1573 lockdep_is_held(&priv->lock));
b63b70d8
SP		 1574
1575 if (!htbl)
1576 goto out_unlock;
1577
1578 for (i = 0; i < htbl->size; i++) {
1579 struct ipoib_neigh *neigh;
1580 struct ipoib_neigh __rcu **np = &htbl->buckets[i];
1581
1582 while ((neigh = rcu_dereference_protected(*np,
b5120a6e 1583 lockdep_is_held(&priv->lock))) != NULL) {
b63b70d8
SP		 1584 /* delete neighs belong to this parent */
1585 if (!memcmp(gid, neigh->daddr + 4, sizeof (union ib_gid))) {
1586 rcu_assign_pointer(*np,
1587 rcu_dereference_protected(neigh->hnext,
b5120a6e 1588 lockdep_is_held(&priv->lock)));
b63b70d8 1589 /* remove from parent list */
c586071d 1590 list_del_init(&neigh->list);
b63b70d8
SP		 1591 call_rcu(&neigh->rcu, ipoib_neigh_reclaim);
1592 } else {
1593 np = &neigh->hnext;
1594 }
1595
1596 }
1597 }
1598out_unlock:
b5120a6e 1599 spin_unlock_irqrestore(&priv->lock, flags);
b63b70d8
SP		 1600}
1601
1602static void ipoib_flush_neighs(struct ipoib_dev_priv *priv)
1603{
1604 struct ipoib_neigh_table *ntbl = &priv->ntbl;
1605 struct ipoib_neigh_hash *htbl;
1606 unsigned long flags;
66172c09 1607 int i, wait_flushed = 0;
b63b70d8 1608
66172c09 1609 init_completion(&priv->ntbl.flushed);
d2e46fcc 1610 set_bit(IPOIB_NEIGH_TBL_FLUSH, &priv->flags);
b63b70d8 1611
b5120a6e 1612 spin_lock_irqsave(&priv->lock, flags);
b63b70d8
SP		 1613
1614 htbl = rcu_dereference_protected(ntbl->htbl,
b5120a6e 1615 lockdep_is_held(&priv->lock));
b63b70d8
SP		 1616 if (!htbl)
1617 goto out_unlock;
1618
66172c09
SP		 1619 wait_flushed = atomic_read(&priv->ntbl.entries);
1620 if (!wait_flushed)
1621 goto free_htbl;
1622
b63b70d8
SP		 1623 for (i = 0; i < htbl->size; i++) {
1624 struct ipoib_neigh *neigh;
1625 struct ipoib_neigh __rcu **np = &htbl->buckets[i];
1626
1627 while ((neigh = rcu_dereference_protected(*np,
b5120a6e 1628 lockdep_is_held(&priv->lock))) != NULL) {
b63b70d8
SP		 1629 rcu_assign_pointer(*np,
1630 rcu_dereference_protected(neigh->hnext,
b5120a6e 1631 lockdep_is_held(&priv->lock)));
b63b70d8 1632 /* remove from path/mc list */
c586071d 1633 list_del_init(&neigh->list);
b63b70d8
SP		 1634 call_rcu(&neigh->rcu, ipoib_neigh_reclaim);
1635 }
1636 }
1637
66172c09 1638free_htbl:
b63b70d8
SP		 1639 rcu_assign_pointer(ntbl->htbl, NULL);
1640 call_rcu(&htbl->rcu, neigh_hash_free_rcu);
1641
1642out_unlock:
b5120a6e 1643 spin_unlock_irqrestore(&priv->lock, flags);
66172c09
SP		 1644 if (wait_flushed)
1645 wait_for_completion(&priv->ntbl.flushed);
b63b70d8
SP		 1646}
1647
1648static void ipoib_neigh_hash_uninit(struct net_device *dev)
1649{
c1048aff 1650 struct ipoib_dev_priv *priv = ipoib_priv(dev);
b63b70d8
SP		 1651 int stopped;
1652
1653 ipoib_dbg(priv, "ipoib_neigh_hash_uninit\n");
66172c09 1654 init_completion(&priv->ntbl.deleted);
b63b70d8
SP		 1655
1656 /* Stop GC if called at init fail need to cancel work */
1657 stopped = test_and_set_bit(IPOIB_STOP_NEIGH_GC, &priv->flags);
1658 if (!stopped)
1659 cancel_delayed_work(&priv->neigh_reap_task);
1660
66172c09
SP		 1661 ipoib_flush_neighs(priv);
1662
1663 wait_for_completion(&priv->ntbl.deleted);
b63b70d8
SP		 1664}
1665
8966e28d
ES		 1666static void ipoib_napi_add(struct net_device *dev)
1667{
1668 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1669
1670 netif_napi_add(dev, &priv->recv_napi, ipoib_rx_poll, IPOIB_NUM_WC);
1671 netif_napi_add(dev, &priv->send_napi, ipoib_tx_poll, MAX_SEND_CQE);
1672}
1673
1674static void ipoib_napi_del(struct net_device *dev)
1675{
1676 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1677
1678 netif_napi_del(&priv->recv_napi);
1679 netif_napi_del(&priv->send_napi);
1680}
1681
0a1a9726 1682static void ipoib_dev_uninit_default(struct net_device *dev)
515ed4f3 1683{
c1048aff 1684 struct ipoib_dev_priv *priv = ipoib_priv(dev);
b63b70d8 1685
515ed4f3
ES		 1686 ipoib_transport_dev_cleanup(dev);
1687
8966e28d 1688 ipoib_napi_del(dev);
b53d4566 1689
515ed4f3
ES		 1690 ipoib_cm_dev_cleanup(dev);
1691
1692 kfree(priv->rx_ring);
1693 vfree(priv->tx_ring);
1694
1695 priv->rx_ring = NULL;
1696 priv->tx_ring = NULL;
1697}
1698
cd565b4b 1699static int ipoib_dev_init_default(struct net_device *dev)
1da177e4 1700{
c1048aff 1701 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1da177e4 1702
8966e28d 1703 ipoib_napi_add(dev);
cd565b4b 1704
1da177e4 1705 /* Allocate RX/TX "rings" to hold queued skbs */
0f485251 1706 priv->rx_ring = kzalloc(ipoib_recvq_size * sizeof *priv->rx_ring,
1da177e4 1707 GFP_KERNEL);
74226649 1708 if (!priv->rx_ring)
be7aa663 1709 goto out;
1da177e4 1710
948579cd 1711 priv->tx_ring = vzalloc(ipoib_sendq_size * sizeof *priv->tx_ring);
1da177e4 1712 if (!priv->tx_ring) {
c55359a2
YS		 1713 pr_warn("%s: failed to allocate TX ring (%d entries)\n",
1714 priv->ca->name, ipoib_sendq_size);
1da177e4
LT		 1715 goto out_rx_ring_cleanup;
1716 }
1da177e4 1717
1b524963 1718 /* priv->tx_head, tx_tail & tx_outstanding are already 0 */
1da177e4 1719
cd565b4b
ES		 1720 if (ipoib_transport_dev_init(dev, priv->ca)) {
1721 pr_warn("%s: ipoib_transport_dev_init failed\n",
1722 priv->ca->name);
1da177e4 1723 goto out_tx_ring_cleanup;
515ed4f3
ES		 1724 }
1725
cd565b4b
ES		 1726 /* after qp created set dev address */
1727 priv->dev->dev_addr[1] = (priv->qp->qp_num >> 16) & 0xff;
1728 priv->dev->dev_addr[2] = (priv->qp->qp_num >> 8) & 0xff;
1729 priv->dev->dev_addr[3] = (priv->qp->qp_num) & 0xff;
1730
515ed4f3
ES		 1731 return 0;
1732
1733out_tx_ring_cleanup:
1734 vfree(priv->tx_ring);
1735
1736out_rx_ring_cleanup:
1737 kfree(priv->rx_ring);
1738
1739out:
8966e28d 1740 ipoib_napi_del(dev);
515ed4f3
ES		 1741 return -ENOMEM;
1742}
1743
31a82362
FD		 1744static int ipoib_ioctl(struct net_device *dev, struct ifreq *ifr,
1745 int cmd)
1746{
1747 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1748
1749 if (!priv->rn_ops->ndo_do_ioctl)
1750 return -EOPNOTSUPP;
1751
1752 return priv->rn_ops->ndo_do_ioctl(dev, ifr, cmd);
1753}
1754
515ed4f3
ES		 1755int ipoib_dev_init(struct net_device *dev, struct ib_device *ca, int port)
1756{
c1048aff 1757 struct ipoib_dev_priv *priv = ipoib_priv(dev);
515ed4f3
ES		 1758 int ret = -ENOMEM;
1759
1760 priv->ca = ca;
1761 priv->port = port;
1762 priv->qp = NULL;
1da177e4 1763
be7aa663 1764 /*
515ed4f3
ES		 1765 * the various IPoIB tasks assume they will never race against
1766 * themselves, so always use a single thread workqueue
be7aa663 1767 */
515ed4f3
ES		 1768 priv->wq = alloc_ordered_workqueue("ipoib_wq", WQ_MEM_RECLAIM);
1769 if (!priv->wq) {
1770 pr_warn("%s: failed to allocate device WQ\n", dev->name);
1771 goto out;
1772 }
1773
1774 /* create pd, which used both for control and datapath*/
1775 priv->pd = ib_alloc_pd(priv->ca, 0);
1776 if (IS_ERR(priv->pd)) {
1777 pr_warn("%s: failed to allocate PD\n", ca->name);
1778 goto clean_wq;
1779 }
1780
cd565b4b 1781 ret = priv->rn_ops->ndo_init(dev);
515ed4f3
ES		 1782 if (ret) {
1783 pr_warn("%s failed to init HW resource\n", dev->name);
1784 goto out_free_pd;
1785 }
1786
515ed4f3
ES		 1787 if (ipoib_neigh_hash_init(priv) < 0) {
1788 pr_warn("%s failed to init neigh hash\n", dev->name);
be7aa663 1789 goto out_dev_uninit;
515ed4f3
ES		 1790 }
1791
1792 if (dev->flags & IFF_UP) {
1793 if (ipoib_ib_dev_open(dev)) {
1794 pr_warn("%s failed to open device\n", dev->name);
1795 ret = -ENODEV;
1796 goto out_dev_uninit;
1797 }
1798 }
be7aa663 1799
1da177e4
LT		 1800 return 0;
1801
be7aa663
DL		 1802out_dev_uninit:
1803 ipoib_ib_dev_cleanup(dev);
1804
515ed4f3
ES		 1805out_free_pd:
1806 if (priv->pd) {
1807 ib_dealloc_pd(priv->pd);
1808 priv->pd = NULL;
1809 }
1da177e4 1810
515ed4f3
ES		 1811clean_wq:
1812 if (priv->wq) {
1813 destroy_workqueue(priv->wq);
1814 priv->wq = NULL;
1815 }
1da177e4
LT		 1816
1817out:
515ed4f3 1818 return ret;
1da177e4
LT		 1819}
1820
1821void ipoib_dev_cleanup(struct net_device *dev)
1822{
c1048aff 1823 struct ipoib_dev_priv *priv = ipoib_priv(dev), *cpriv, *tcpriv;
9baa0b03
OG		 1824 LIST_HEAD(head);
1825
1826 ASSERT_RTNL();
1da177e4 1827
1da177e4
LT		 1828 /* Delete any child interfaces first */
1829 list_for_each_entry_safe(cpriv, tcpriv, &priv->child_intfs, list) {
b63b70d8
SP		 1830 /* Stop GC on child */
1831 set_bit(IPOIB_STOP_NEIGH_GC, &cpriv->flags);
1832 cancel_delayed_work(&cpriv->neigh_reap_task);
9baa0b03 1833 unregister_netdevice_queue(cpriv->dev, &head);
1da177e4 1834 }
9baa0b03 1835 unregister_netdevice_many(&head);
1da177e4 1836
be7aa663
DL		 1837 ipoib_neigh_hash_uninit(dev);
1838
1da177e4
LT		 1839 ipoib_ib_dev_cleanup(dev);
1840
515ed4f3
ES		 1841 /* no more works over the priv->wq */
1842 if (priv->wq) {
1843 flush_workqueue(priv->wq);
1844 destroy_workqueue(priv->wq);
1845 priv->wq = NULL;
1846 }
1da177e4
LT		 1847}
1848
9c3c5f8e
EC		 1849static int ipoib_set_vf_link_state(struct net_device *dev, int vf, int link_state)
1850{
c1048aff 1851 struct ipoib_dev_priv *priv = ipoib_priv(dev);
9c3c5f8e
EC		 1852
1853 return ib_set_vf_link_state(priv->ca, vf, priv->port, link_state);
1854}
1855
1856static int ipoib_get_vf_config(struct net_device *dev, int vf,
1857 struct ifla_vf_info *ivf)
1858{
c1048aff 1859 struct ipoib_dev_priv *priv = ipoib_priv(dev);
9c3c5f8e
EC		 1860 int err;
1861
1862 err = ib_get_vf_config(priv->ca, vf, priv->port, ivf);
1863 if (err)
1864 return err;
1865
1866 ivf->vf = vf;
1867
1868 return 0;
1869}
1870
1871static int ipoib_set_vf_guid(struct net_device *dev, int vf, u64 guid, int type)
1872{
c1048aff 1873 struct ipoib_dev_priv *priv = ipoib_priv(dev);
9c3c5f8e
EC		 1874
1875 if (type != IFLA_VF_IB_NODE_GUID && type != IFLA_VF_IB_PORT_GUID)
1876 return -EINVAL;
1877
1878 return ib_set_vf_guid(priv->ca, vf, priv->port, guid, type);
1879}
1880
1881static int ipoib_get_vf_stats(struct net_device *dev, int vf,
1882 struct ifla_vf_stats *vf_stats)
1883{
c1048aff 1884 struct ipoib_dev_priv *priv = ipoib_priv(dev);
9c3c5f8e
EC		 1885
1886 return ib_get_vf_stats(priv->ca, vf, priv->port, vf_stats);
1887}
1888
3b04ddde
SH		 1889static const struct header_ops ipoib_header_ops = {
1890 .create = ipoib_hard_header,
1891};
1892
9c3c5f8e
EC		 1893static const struct net_device_ops ipoib_netdev_ops_pf = {
1894 .ndo_uninit = ipoib_uninit,
1895 .ndo_open = ipoib_open,
1896 .ndo_stop = ipoib_stop,
1897 .ndo_change_mtu = ipoib_change_mtu,
1898 .ndo_fix_features = ipoib_fix_features,
1899 .ndo_start_xmit = ipoib_start_xmit,
1900 .ndo_tx_timeout = ipoib_timeout,
1901 .ndo_set_rx_mode = ipoib_set_mcast_list,
1902 .ndo_get_iflink = ipoib_get_iflink,
1903 .ndo_set_vf_link_state = ipoib_set_vf_link_state,
1904 .ndo_get_vf_config = ipoib_get_vf_config,
1905 .ndo_get_vf_stats = ipoib_get_vf_stats,
1906 .ndo_set_vf_guid = ipoib_set_vf_guid,
492a7e67 1907 .ndo_set_mac_address = ipoib_set_mac,
b6c871e5 1908 .ndo_get_stats64 = ipoib_get_stats,
31a82362 1909 .ndo_do_ioctl = ipoib_ioctl,
9c3c5f8e
EC		 1910};
1911
1912static const struct net_device_ops ipoib_netdev_ops_vf = {
9baa0b03 1913 .ndo_uninit = ipoib_uninit,
fe8114e8
SH		 1914 .ndo_open = ipoib_open,
1915 .ndo_stop = ipoib_stop,
1916 .ndo_change_mtu = ipoib_change_mtu,
3d96c74d 1917 .ndo_fix_features = ipoib_fix_features,
fe8114e8
SH		 1918 .ndo_start_xmit = ipoib_start_xmit,
1919 .ndo_tx_timeout = ipoib_timeout,
afc4b13d 1920 .ndo_set_rx_mode = ipoib_set_mcast_list,
5aa7add8 1921 .ndo_get_iflink = ipoib_get_iflink,
eb54714d 1922 .ndo_get_stats64 = ipoib_get_stats,
31a82362 1923 .ndo_do_ioctl = ipoib_ioctl,
fe8114e8
SH		 1924};
1925
cd565b4b 1926void ipoib_setup_common(struct net_device *dev)
1da177e4 1927{
2337f809 1928 dev->header_ops = &ipoib_header_ops;
bea3348e 1929
82c24c18
EC		 1930 ipoib_set_ethtool_ops(dev);
1931
2337f809 1932 dev->watchdog_timeo = HZ;
1da177e4 1933
2337f809 1934 dev->flags |= IFF_BROADCAST | IFF_MULTICAST;
1da177e4 1935
fc791b63 1936 dev->hard_header_len = IPOIB_HARD_LEN;
2337f809
RD		 1937 dev->addr_len = INFINIBAND_ALEN;
1938 dev->type = ARPHRD_INFINIBAND;
1939 dev->tx_queue_len = ipoib_sendq_size * 2;
eb14032f 1940 dev->features = (NETIF_F_VLAN_CHALLENGED |
eb14032f 1941 NETIF_F_HIGHDMA);
02875878 1942 netif_keep_dst(dev);
1da177e4 1943
1da177e4 1944 memcpy(dev->broadcast, ipv4_bcast_addr, INFINIBAND_ALEN);
cd565b4b 1945}
1da177e4 1946
cd565b4b
ES		 1947static void ipoib_build_priv(struct net_device *dev)
1948{
1949 struct ipoib_dev_priv *priv = ipoib_priv(dev);
1da177e4 1950
cd565b4b 1951 priv->dev = dev;
1da177e4 1952 spin_lock_init(&priv->lock);
f47944cc 1953 init_rwsem(&priv->vlan_rwsem);
edf3f301 1954 mutex_init(&priv->mcast_mutex);
69956d83 1955 mutex_init(&priv->sysfs_mutex);
1da177e4
LT		 1956
1957 INIT_LIST_HEAD(&priv->path_list);
1958 INIT_LIST_HEAD(&priv->child_intfs);
1959 INIT_LIST_HEAD(&priv->dead_ahs);
1960 INIT_LIST_HEAD(&priv->multicast_list);
1961
c4028958 1962 INIT_DELAYED_WORK(&priv->mcast_task, ipoib_mcast_join_task);
e8224e4b 1963 INIT_WORK(&priv->carrier_on_task, ipoib_mcast_carrier_on_task);
ee1e2c82
MS		 1964 INIT_WORK(&priv->flush_light, ipoib_ib_dev_flush_light);
1965 INIT_WORK(&priv->flush_normal, ipoib_ib_dev_flush_normal);
1966 INIT_WORK(&priv->flush_heavy, ipoib_ib_dev_flush_heavy);
c4028958
DH		 1967 INIT_WORK(&priv->restart_task, ipoib_mcast_restart_task);
1968 INIT_DELAYED_WORK(&priv->ah_reap_task, ipoib_reap_ah);
b63b70d8 1969 INIT_DELAYED_WORK(&priv->neigh_reap_task, ipoib_reap_neigh);
1da177e4
LT		 1970}
1971
cd565b4b
ES		 1972static const struct net_device_ops ipoib_netdev_default_pf = {
1973 .ndo_init = ipoib_dev_init_default,
1974 .ndo_uninit = ipoib_dev_uninit_default,
1975 .ndo_open = ipoib_ib_dev_open_default,
1976 .ndo_stop = ipoib_ib_dev_stop_default,
1977};
1978
1979static struct net_device
1980*ipoib_create_netdev_default(struct ib_device *hca,
1981 const char *name,
1982 unsigned char name_assign_type,
1983 void (*setup)(struct net_device *))
1da177e4
LT		 1984{
1985 struct net_device *dev;
cd565b4b 1986 struct rdma_netdev *rn;
1da177e4 1987
cd565b4b
ES		 1988 dev = alloc_netdev((int)sizeof(struct rdma_netdev),
1989 name,
1990 name_assign_type, setup);
1da177e4
LT		 1991 if (!dev)
1992 return NULL;
1993
cd565b4b
ES		 1994 rn = netdev_priv(dev);
1995
1996 rn->send = ipoib_send;
1997 rn->attach_mcast = ipoib_mcast_attach;
1998 rn->detach_mcast = ipoib_mcast_detach;
8e959601 1999 rn->free_rdma_netdev = free_netdev;
cd565b4b
ES		 2000 rn->hca = hca;
2001
2002 dev->netdev_ops = &ipoib_netdev_default_pf;
2003
2004 return dev;
2005}
2006
2007static struct net_device *ipoib_get_netdev(struct ib_device *hca, u8 port,
2008 const char *name)
2009{
2010 struct net_device *dev;
2011
2012 if (hca->alloc_rdma_netdev) {
2013 dev = hca->alloc_rdma_netdev(hca, port,
2014 RDMA_NETDEV_IPOIB, name,
2015 NET_NAME_UNKNOWN,
2016 ipoib_setup_common);
2017 if (IS_ERR_OR_NULL(dev) && PTR_ERR(dev) != -EOPNOTSUPP)
2018 return NULL;
2019 }
2020
2021 if (!hca->alloc_rdma_netdev || PTR_ERR(dev) == -EOPNOTSUPP)
2022 dev = ipoib_create_netdev_default(hca, name, NET_NAME_UNKNOWN,
2023 ipoib_setup_common);
2024
2025 return dev;
2026}
2027
2028struct ipoib_dev_priv *ipoib_intf_alloc(struct ib_device *hca, u8 port,
2029 const char *name)
2030{
2031 struct net_device *dev;
2032 struct ipoib_dev_priv *priv;
2033 struct rdma_netdev *rn;
2034
2035 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
2036 if (!priv)
2037 return NULL;
2038
2039 dev = ipoib_get_netdev(hca, port, name);
2040 if (!dev)
2041 goto free_priv;
2042
2043 priv->rn_ops = dev->netdev_ops;
2044
2045 /* fixme : should be after the query_cap */
2046 if (priv->hca_caps & IB_DEVICE_VIRTUAL_FUNCTION)
2047 dev->netdev_ops = &ipoib_netdev_ops_vf;
2048 else
2049 dev->netdev_ops = &ipoib_netdev_ops_pf;
2050
2051 rn = netdev_priv(dev);
2052 rn->clnt_priv = priv;
2053 ipoib_build_priv(dev);
2054
2055 return priv;
2056free_priv:
2057 kfree(priv);
2058 return NULL;
1da177e4
LT		 2059}
2060
43cb76d9
GKH		 2061static ssize_t show_pkey(struct device *dev,
2062 struct device_attribute *attr, char *buf)
1da177e4 2063{
c1048aff
ES		 2064 struct net_device *ndev = to_net_dev(dev);
2065 struct ipoib_dev_priv *priv = ipoib_priv(ndev);
1da177e4
LT		 2066
2067 return sprintf(buf, "0x%04x\n", priv->pkey);
2068}
43cb76d9 2069static DEVICE_ATTR(pkey, S_IRUGO, show_pkey, NULL);
1da177e4 2070
335a64a5
OG		 2071static ssize_t show_umcast(struct device *dev,
2072 struct device_attribute *attr, char *buf)
2073{
c1048aff
ES		 2074 struct net_device *ndev = to_net_dev(dev);
2075 struct ipoib_dev_priv *priv = ipoib_priv(ndev);
335a64a5
OG		 2076
2077 return sprintf(buf, "%d\n", test_bit(IPOIB_FLAG_UMCAST, &priv->flags));
2078}
2079
862096a8 2080void ipoib_set_umcast(struct net_device *ndev, int umcast_val)
335a64a5 2081{
c1048aff 2082 struct ipoib_dev_priv *priv = ipoib_priv(ndev);
335a64a5
OG		 2083
2084 if (umcast_val > 0) {
2085 set_bit(IPOIB_FLAG_UMCAST, &priv->flags);
2086 ipoib_warn(priv, "ignoring multicast groups joined directly "
2087 "by userspace\n");
2088 } else
2089 clear_bit(IPOIB_FLAG_UMCAST, &priv->flags);
862096a8
OG		 2090}
2091
2092static ssize_t set_umcast(struct device *dev,
2093 struct device_attribute *attr,
2094 const char *buf, size_t count)
2095{
2096 unsigned long umcast_val = simple_strtoul(buf, NULL, 0);
2097
2098 ipoib_set_umcast(to_net_dev(dev), umcast_val);
335a64a5
OG		 2099
2100 return count;
2101}
2102static DEVICE_ATTR(umcast, S_IWUSR | S_IRUGO, show_umcast, set_umcast);
2103
2104int ipoib_add_umcast_attr(struct net_device *dev)
2105{
2106 return device_create_file(&dev->dev, &dev_attr_umcast);
2107}
2108
492a7e67
MB		 2109static void set_base_guid(struct ipoib_dev_priv *priv, union ib_gid *gid)
2110{
2111 struct ipoib_dev_priv *child_priv;
2112 struct net_device *netdev = priv->dev;
2113
9b29953b 2114 netif_addr_lock_bh(netdev);
492a7e67
MB		 2115
2116 memcpy(&priv->local_gid.global.interface_id,
2117 &gid->global.interface_id,
2118 sizeof(gid->global.interface_id));
2119 memcpy(netdev->dev_addr + 4, &priv->local_gid, sizeof(priv->local_gid));
2120 clear_bit(IPOIB_FLAG_DEV_ADDR_SET, &priv->flags);
2121
9b29953b 2122 netif_addr_unlock_bh(netdev);
492a7e67
MB		 2123
2124 if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) {
2125 down_read(&priv->vlan_rwsem);
2126 list_for_each_entry(child_priv, &priv->child_intfs, list)
2127 set_base_guid(child_priv, gid);
2128 up_read(&priv->vlan_rwsem);
2129 }
2130}
2131
2132static int ipoib_check_lladdr(struct net_device *dev,
2133 struct sockaddr_storage *ss)
2134{
2135 union ib_gid *gid = (union ib_gid *)(ss->__data + 4);
2136 int ret = 0;
2137
9b29953b 2138 netif_addr_lock_bh(dev);
492a7e67
MB		 2139
2140 /* Make sure the QPN, reserved and subnet prefix match the current
2141 * lladdr, it also makes sure the lladdr is unicast.
2142 */
2143 if (memcmp(dev->dev_addr, ss->__data,
2144 4 + sizeof(gid->global.subnet_prefix)) ||
2145 gid->global.interface_id == 0)
2146 ret = -EINVAL;
2147
9b29953b 2148 netif_addr_unlock_bh(dev);
492a7e67
MB		 2149
2150 return ret;
2151}
2152
2153static int ipoib_set_mac(struct net_device *dev, void *addr)
2154{
c1048aff 2155 struct ipoib_dev_priv *priv = ipoib_priv(dev);
492a7e67
MB		 2156 struct sockaddr_storage *ss = addr;
2157 int ret;
2158
2159 if (!(dev->priv_flags & IFF_LIVE_ADDR_CHANGE) && netif_running(dev))
2160 return -EBUSY;
2161
2162 ret = ipoib_check_lladdr(dev, ss);
2163 if (ret)
2164 return ret;
2165
2166 set_base_guid(priv, (union ib_gid *)(ss->__data + 4));
2167
2168 queue_work(ipoib_workqueue, &priv->flush_light);
2169
2170 return 0;
2171}
2172
43cb76d9
GKH		 2173static ssize_t create_child(struct device *dev,
2174 struct device_attribute *attr,
1da177e4
LT		 2175 const char *buf, size_t count)
2176{
2177 int pkey;
2178 int ret;
2179
2180 if (sscanf(buf, "%i", &pkey) != 1)
2181 return -EINVAL;
2182
3d790a4c 2183 if (pkey <= 0 || pkey > 0xffff || pkey == 0x8000)
1da177e4
LT		 2184 return -EINVAL;
2185
4ce05937
RD		 2186 /*
2187 * Set the full membership bit, so that we join the right
2188 * broadcast group, etc.
2189 */
2190 pkey |= 0x8000;
2191
43cb76d9 2192 ret = ipoib_vlan_add(to_net_dev(dev), pkey);
1da177e4
LT		 2193
2194 return ret ? ret : count;
2195}
7a52b34b 2196static DEVICE_ATTR(create_child, S_IWUSR, NULL, create_child);
1da177e4 2197
43cb76d9
GKH		 2198static ssize_t delete_child(struct device *dev,
2199 struct device_attribute *attr,
1da177e4
LT		 2200 const char *buf, size_t count)
2201{
2202 int pkey;
2203 int ret;
2204
2205 if (sscanf(buf, "%i", &pkey) != 1)
2206 return -EINVAL;
2207
2208 if (pkey < 0 || pkey > 0xffff)
2209 return -EINVAL;
2210
43cb76d9 2211 ret = ipoib_vlan_delete(to_net_dev(dev), pkey);
1da177e4
LT		 2212
2213 return ret ? ret : count;
2214
2215}
7a52b34b 2216static DEVICE_ATTR(delete_child, S_IWUSR, NULL, delete_child);
1da177e4
LT		 2217
2218int ipoib_add_pkey_attr(struct net_device *dev)
2219{
43cb76d9 2220 return device_create_file(&dev->dev, &dev_attr_pkey);
1da177e4
LT		 2221}
2222
f7534f45 2223void ipoib_set_dev_features(struct ipoib_dev_priv *priv, struct ib_device *hca)
83bb63f6 2224{
4a061b28 2225 priv->hca_caps = hca->attrs.device_cap_flags;
83bb63f6
OG		 2226
2227 if (priv->hca_caps & IB_DEVICE_UD_IP_CSUM) {
cd565b4b 2228 priv->dev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_RXCSUM;
83bb63f6 2229
3d96c74d
MM		 2230 if (priv->hca_caps & IB_DEVICE_UD_TSO)
2231 priv->dev->hw_features |= NETIF_F_TSO;
8ae31e5b 2232
3d96c74d
MM		 2233 priv->dev->features |= priv->dev->hw_features;
2234 }
83bb63f6
OG		 2235}
2236
1da177e4
LT		 2237static struct net_device *ipoib_add_port(const char *format,
2238 struct ib_device *hca, u8 port)
2239{
2240 struct ipoib_dev_priv *priv;
bc7b3a36 2241 struct ib_port_attr attr;
7c9d9662 2242 struct rdma_netdev *rn;
1da177e4
LT		 2243 int result = -ENOMEM;
2244
cd565b4b 2245 priv = ipoib_intf_alloc(hca, port, format);
ac6dbf7f
YS		 2246 if (!priv) {
2247 pr_warn("%s, %d: ipoib_intf_alloc failed\n", hca->name, port);
1da177e4 2248 goto alloc_mem_failed;
ac6dbf7f 2249 }
1da177e4 2250
db97ed0a 2251 SET_NETDEV_DEV(priv->dev, hca->dev.parent);
c3aa9b18 2252 priv->dev->dev_id = port - 1;
1da177e4 2253
58e9cc90 2254 result = ib_query_port(hca, port, &attr);
dc892e17 2255 if (result) {
c55359a2 2256 pr_warn("%s: ib_query_port %d failed\n", hca->name, port);
bc7b3a36
SM		 2257 goto device_init_failed;
2258 }
2259
dc892e17
LR		 2260 priv->max_ib_mtu = ib_mtu_enum_to_int(attr.max_mtu);
2261
bc7b3a36
SM		 2262 /* MTU will be reset when mcast join happens */
2263 priv->dev->mtu = IPOIB_UD_MTU(priv->max_ib_mtu);
2264 priv->mcast_mtu = priv->admin_mtu = priv->dev->mtu;
b3e3893e 2265 priv->dev->max_mtu = IPOIB_CM_MTU;
bc7b3a36 2266
596b9b68
DM		 2267 priv->dev->neigh_priv_len = sizeof(struct ipoib_neigh);
2268
1da177e4
LT		 2269 result = ib_query_pkey(hca, port, 0, &priv->pkey);
2270 if (result) {
c55359a2
YS		 2271 pr_warn("%s: ib_query_pkey port %d failed (ret = %d)\n",
2272 hca->name, port, result);
ca6de177 2273 goto device_init_failed;
1da177e4
LT		 2274 }
2275
f7534f45 2276 ipoib_set_dev_features(priv, hca);
af40da89 2277
4ce05937
RD		 2278 /*
2279 * Set the full membership bit, so that we join the right
2280 * broadcast group, etc.
2281 */
2282 priv->pkey |= 0x8000;
2283
1da177e4
LT		 2284 priv->dev->broadcast[8] = priv->pkey >> 8;
2285 priv->dev->broadcast[9] = priv->pkey & 0xff;
2286
55ee3ab2 2287 result = ib_query_gid(hca, port, 0, &priv->local_gid, NULL);
1da177e4 2288 if (result) {
c55359a2
YS		 2289 pr_warn("%s: ib_query_gid port %d failed (ret = %d)\n",
2290 hca->name, port, result);
ca6de177 2291 goto device_init_failed;
dc892e17
LR		 2292 }
2293
2294 memcpy(priv->dev->dev_addr + 4, priv->local_gid.raw,
2295 sizeof(union ib_gid));
492a7e67 2296 set_bit(IPOIB_FLAG_DEV_ADDR_SET, &priv->flags);
1da177e4 2297
1da177e4 2298 result = ipoib_dev_init(priv->dev, hca, port);
dc892e17 2299 if (result) {
c55359a2
YS		 2300 pr_warn("%s: failed to initialize port %d (ret = %d)\n",
2301 hca->name, port, result);
1da177e4
LT		 2302 goto device_init_failed;
2303 }
2304
2305 INIT_IB_EVENT_HANDLER(&priv->event_handler,
2306 priv->ca, ipoib_event);
dcc9881e 2307 ib_register_event_handler(&priv->event_handler);
1da177e4 2308
10293610
AE		 2309 /* call event handler to ensure pkey in sync */
2310 queue_work(ipoib_workqueue, &priv->flush_heavy);
2311
1da177e4
LT		 2312 result = register_netdev(priv->dev);
2313 if (result) {
c55359a2
YS		 2314 pr_warn("%s: couldn't register ipoib port %d; error %d\n",
2315 hca->name, port, result);
1da177e4
LT		 2316 goto register_failed;
2317 }
2318
5c8857b6 2319 result = -ENOMEM;
839fcaba
MT		 2320 if (ipoib_cm_add_mode_attr(priv->dev))
2321 goto sysfs_failed;
1da177e4
LT		 2322 if (ipoib_add_pkey_attr(priv->dev))
2323 goto sysfs_failed;
335a64a5
OG		 2324 if (ipoib_add_umcast_attr(priv->dev))
2325 goto sysfs_failed;
43cb76d9 2326 if (device_create_file(&priv->dev->dev, &dev_attr_create_child))
1da177e4 2327 goto sysfs_failed;
43cb76d9 2328 if (device_create_file(&priv->dev->dev, &dev_attr_delete_child))
1da177e4
LT		 2329 goto sysfs_failed;
2330
2331 return priv->dev;
2332
2333sysfs_failed:
1da177e4
LT		 2334 unregister_netdev(priv->dev);
2335
2336register_failed:
2337 ib_unregister_event_handler(&priv->event_handler);
0b39578b 2338 flush_workqueue(ipoib_workqueue);
b63b70d8
SP		 2339 /* Stop GC if started before flush */
2340 set_bit(IPOIB_STOP_NEIGH_GC, &priv->flags);
2341 cancel_delayed_work(&priv->neigh_reap_task);
0b39578b 2342 flush_workqueue(priv->wq);
1da177e4
LT		 2343 ipoib_dev_cleanup(priv->dev);
2344
2345device_init_failed:
7c9d9662
AV		 2346 rn = netdev_priv(priv->dev);
2347 rn->free_rdma_netdev(priv->dev);
ab156afd 2348 kfree(priv);
1da177e4
LT		 2349
2350alloc_mem_failed:
2351 return ERR_PTR(result);
2352}
2353
2354static void ipoib_add_one(struct ib_device *device)
2355{
2356 struct list_head *dev_list;
2357 struct net_device *dev;
2358 struct ipoib_dev_priv *priv;
4139032b 2359 int p;
8e37ab68 2360 int count = 0;
07ebafba 2361
1da177e4
LT		 2362 dev_list = kmalloc(sizeof *dev_list, GFP_KERNEL);
2363 if (!dev_list)
2364 return;
2365
2366 INIT_LIST_HEAD(dev_list);
2367
4139032b 2368 for (p = rdma_start_port(device); p <= rdma_end_port(device); ++p) {
8e37ab68 2369 if (!rdma_protocol_ib(device, p))
7b4c8769 2370 continue;
1da177e4
LT		 2371 dev = ipoib_add_port("ib%d", device, p);
2372 if (!IS_ERR(dev)) {
c1048aff 2373 priv = ipoib_priv(dev);
1da177e4 2374 list_add_tail(&priv->list, dev_list);
8e37ab68 2375 count++;
1da177e4
LT		 2376 }
2377 }
2378
8e37ab68 2379 if (!count) {
ac6dbf7f 2380 kfree(dev_list);
8e37ab68
MW		 2381 return;
2382 }
2383
1da177e4
LT		 2384 ib_set_client_data(device, &ipoib_client, dev_list);
2385}
2386
7c1eb45a 2387static void ipoib_remove_one(struct ib_device *device, void *client_data)
1da177e4 2388{
ab156afd 2389 struct ipoib_dev_priv *priv, *tmp, *cpriv, *tcpriv;
7c1eb45a 2390 struct list_head *dev_list = client_data;
1da177e4 2391
5a2815f0
IG		 2392 if (!dev_list)
2393 return;
1da177e4
LT		 2394
2395 list_for_each_entry_safe(priv, tmp, dev_list, list) {
7c9d9662 2396 struct rdma_netdev *parent_rn = netdev_priv(priv->dev);
8e959601 2397
1da177e4 2398 ib_unregister_event_handler(&priv->event_handler);
0b39578b 2399 flush_workqueue(ipoib_workqueue);
a77a57a1 2400
198b12f7
ES		 2401 /* mark interface in the middle of destruction */
2402 set_bit(IPOIB_FLAG_GOING_DOWN, &priv->flags);
2403
a77a57a1
RD		 2404 rtnl_lock();
2405 dev_change_flags(priv->dev, priv->dev->flags & ~IFF_UP);
2406 rtnl_unlock();
2407
b63b70d8
SP		 2408 /* Stop GC */
2409 set_bit(IPOIB_STOP_NEIGH_GC, &priv->flags);
2410 cancel_delayed_work(&priv->neigh_reap_task);
0b39578b 2411 flush_workqueue(priv->wq);
1da177e4 2412
69956d83
ES		 2413 /* Wrap rtnl_lock/unlock with mutex to protect sysfs calls */
2414 mutex_lock(&priv->sysfs_mutex);
1da177e4 2415 unregister_netdev(priv->dev);
69956d83
ES		 2416 mutex_unlock(&priv->sysfs_mutex);
2417
7c9d9662
AV		 2418 parent_rn->free_rdma_netdev(priv->dev);
2419
2420 list_for_each_entry_safe(cpriv, tcpriv, &priv->child_intfs, list) {
2421 struct rdma_netdev *child_rn;
ab156afd 2422
7c9d9662
AV		 2423 child_rn = netdev_priv(cpriv->dev);
2424 child_rn->free_rdma_netdev(cpriv->dev);
ab156afd 2425 kfree(cpriv);
7c9d9662 2426 }
ab156afd 2427
cd565b4b 2428 kfree(priv);
1da177e4 2429 }
06c56e44
MT		 2430
2431 kfree(dev_list);
1da177e4
LT		 2432}
2433
771a5258
SR		 2434#ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
2435static struct notifier_block ipoib_netdev_notifier = {
2436 .notifier_call = ipoib_netdev_event,
2437};
2438#endif
2439
1da177e4
LT		 2440static int __init ipoib_init_module(void)
2441{
2442 int ret;
2443
0f485251
SM		 2444 ipoib_recvq_size = roundup_pow_of_two(ipoib_recvq_size);
2445 ipoib_recvq_size = min(ipoib_recvq_size, IPOIB_MAX_QUEUE_SIZE);
2446 ipoib_recvq_size = max(ipoib_recvq_size, IPOIB_MIN_QUEUE_SIZE);
2447
2448 ipoib_sendq_size = roundup_pow_of_two(ipoib_sendq_size);
2449 ipoib_sendq_size = min(ipoib_sendq_size, IPOIB_MAX_QUEUE_SIZE);
732eacc0 2450 ipoib_sendq_size = max3(ipoib_sendq_size, 2 * MAX_SEND_CQE, IPOIB_MIN_QUEUE_SIZE);
68e995a2
PS		 2451#ifdef CONFIG_INFINIBAND_IPOIB_CM
2452 ipoib_max_conn_qp = min(ipoib_max_conn_qp, IPOIB_CM_MAX_CONN_QP);
11f74b40 2453 ipoib_max_conn_qp = max(ipoib_max_conn_qp, 0);
68e995a2 2454#endif
0f485251 2455
f89271da
EC		 2456 /*
2457 * When copying small received packets, we only copy from the
2458 * linear data part of the SKB, so we rely on this condition.
2459 */
2460 BUILD_BUG_ON(IPOIB_CM_COPYBREAK > IPOIB_CM_HEAD_SIZE);
2461
1da177e4
LT		 2462 ret = ipoib_register_debugfs();
2463 if (ret)
2464 return ret;
2465
2466 /*
0b39578b
DL		 2467 * We create a global workqueue here that is used for all flush
2468 * operations. However, if you attempt to flush a workqueue
2469 * from a task on that same workqueue, it deadlocks the system.
2470 * We want to be able to flush the tasks associated with a
2471 * specific net device, so we also create a workqueue for each
2472 * netdevice. We queue up the tasks for that device only on
2473 * its private workqueue, and we only queue up flush events
2474 * on our global flush workqueue. This avoids the deadlocks.
1da177e4 2475 */
855cda68
BS		 2476 ipoib_workqueue = alloc_ordered_workqueue("ipoib_flush",
2477 WQ_MEM_RECLAIM);
1da177e4
LT		 2478 if (!ipoib_workqueue) {
2479 ret = -ENOMEM;
2480 goto err_fs;
2481 }
2482
c1a0b23b
MT		 2483 ib_sa_register_client(&ipoib_sa_client);
2484
1da177e4
LT		 2485 ret = ib_register_client(&ipoib_client);
2486 if (ret)
c1a0b23b 2487 goto err_sa;
1da177e4 2488
9baa0b03
OG		 2489 ret = ipoib_netlink_init();
2490 if (ret)
2491 goto err_client;
2492
771a5258
SR		 2493#ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
2494 register_netdevice_notifier(&ipoib_netdev_notifier);
2495#endif
1da177e4
LT		 2496 return 0;
2497
9baa0b03
OG		 2498err_client:
2499 ib_unregister_client(&ipoib_client);
2500
c1a0b23b
MT		 2501err_sa:
2502 ib_sa_unregister_client(&ipoib_sa_client);
1da177e4
LT		 2503 destroy_workqueue(ipoib_workqueue);
2504
9adec1a8
RD		 2505err_fs:
2506 ipoib_unregister_debugfs();
2507
1da177e4
LT		 2508 return ret;
2509}
2510
2511static void __exit ipoib_cleanup_module(void)
2512{
771a5258
SR		 2513#ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
2514 unregister_netdevice_notifier(&ipoib_netdev_notifier);
2515#endif
9baa0b03 2516 ipoib_netlink_fini();
1da177e4 2517 ib_unregister_client(&ipoib_client);
c1a0b23b 2518 ib_sa_unregister_client(&ipoib_sa_client);
9adec1a8 2519 ipoib_unregister_debugfs();
1da177e4
LT		 2520 destroy_workqueue(ipoib_workqueue);
2521}
2522
2523module_init(ipoib_init_module);
2524module_exit(ipoib_cleanup_module);
Linux 6.x block layer and io_uring tree(s)