2 * Copyright (c) 2004 Topspin Communications. All rights reserved.
3 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
4 * Copyright (c) 2004 Voltaire, Inc. All rights reserved.
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:
12 * Redistribution and use in source and binary forms, with or
13 * without modification, are permitted provided that the following
16 * - Redistributions of source code must retain the above
17 * copyright notice, this list of conditions and the following
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.
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
34 * $Id: ipoib_main.c 1377 2004-12-23 19:57:12Z roland $
39 #include <linux/module.h>
41 #include <linux/init.h>
42 #include <linux/slab.h>
43 #include <linux/vmalloc.h>
45 #include <linux/if_arp.h> /* For ARPHRD_xxx */
52 MODULE_AUTHOR("Roland Dreier");
53 MODULE_DESCRIPTION("IP-over-InfiniBand net driver");
54 MODULE_LICENSE("Dual BSD/GPL");
56 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
57 int ipoib_debug_level;
59 module_param_named(debug_level, ipoib_debug_level, int, 0644);
60 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0");
63 struct ipoib_path_iter {
64 struct net_device *dev;
65 struct ipoib_path path;
68 static const u8 ipv4_bcast_addr[] = {
69 0x00, 0xff, 0xff, 0xff,
70 0xff, 0x12, 0x40, 0x1b, 0x00, 0x00, 0x00, 0x00,
71 0x00, 0x00, 0x00, 0x00, 0xff, 0xff, 0xff, 0xff
74 struct workqueue_struct *ipoib_workqueue;
76 static void ipoib_add_one(struct ib_device *device);
77 static void ipoib_remove_one(struct ib_device *device);
79 static struct ib_client ipoib_client = {
82 .remove = ipoib_remove_one
85 int ipoib_open(struct net_device *dev)
87 struct ipoib_dev_priv *priv = netdev_priv(dev);
89 ipoib_dbg(priv, "bringing up interface\n");
91 set_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags);
93 if (ipoib_pkey_dev_delay_open(dev))
96 if (ipoib_ib_dev_open(dev))
99 if (ipoib_ib_dev_up(dev)) {
100 ipoib_ib_dev_stop(dev);
104 if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) {
105 struct ipoib_dev_priv *cpriv;
107 /* Bring up any child interfaces too */
108 mutex_lock(&priv->vlan_mutex);
109 list_for_each_entry(cpriv, &priv->child_intfs, list) {
112 flags = cpriv->dev->flags;
116 dev_change_flags(cpriv->dev, flags | IFF_UP);
118 mutex_unlock(&priv->vlan_mutex);
121 netif_start_queue(dev);
126 static int ipoib_stop(struct net_device *dev)
128 struct ipoib_dev_priv *priv = netdev_priv(dev);
130 ipoib_dbg(priv, "stopping interface\n");
132 clear_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags);
134 netif_stop_queue(dev);
137 * Now flush workqueue to make sure a scheduled task doesn't
138 * bring our internal state back up.
140 flush_workqueue(ipoib_workqueue);
142 ipoib_ib_dev_down(dev, 1);
143 ipoib_ib_dev_stop(dev);
145 if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) {
146 struct ipoib_dev_priv *cpriv;
148 /* Bring down any child interfaces too */
149 mutex_lock(&priv->vlan_mutex);
150 list_for_each_entry(cpriv, &priv->child_intfs, list) {
153 flags = cpriv->dev->flags;
154 if (!(flags & IFF_UP))
157 dev_change_flags(cpriv->dev, flags & ~IFF_UP);
159 mutex_unlock(&priv->vlan_mutex);
165 static int ipoib_change_mtu(struct net_device *dev, int new_mtu)
167 struct ipoib_dev_priv *priv = netdev_priv(dev);
169 if (new_mtu > IPOIB_PACKET_SIZE - IPOIB_ENCAP_LEN)
172 priv->admin_mtu = new_mtu;
174 dev->mtu = min(priv->mcast_mtu, priv->admin_mtu);
179 static struct ipoib_path *__path_find(struct net_device *dev,
182 struct ipoib_dev_priv *priv = netdev_priv(dev);
183 struct rb_node *n = priv->path_tree.rb_node;
184 struct ipoib_path *path;
188 path = rb_entry(n, struct ipoib_path, rb_node);
190 ret = memcmp(gid->raw, path->pathrec.dgid.raw,
191 sizeof (union ib_gid));
204 static int __path_add(struct net_device *dev, struct ipoib_path *path)
206 struct ipoib_dev_priv *priv = netdev_priv(dev);
207 struct rb_node **n = &priv->path_tree.rb_node;
208 struct rb_node *pn = NULL;
209 struct ipoib_path *tpath;
214 tpath = rb_entry(pn, struct ipoib_path, rb_node);
216 ret = memcmp(path->pathrec.dgid.raw, tpath->pathrec.dgid.raw,
217 sizeof (union ib_gid));
226 rb_link_node(&path->rb_node, pn, n);
227 rb_insert_color(&path->rb_node, &priv->path_tree);
229 list_add_tail(&path->list, &priv->path_list);
234 static void path_free(struct net_device *dev, struct ipoib_path *path)
236 struct ipoib_dev_priv *priv = netdev_priv(dev);
237 struct ipoib_neigh *neigh, *tn;
241 while ((skb = __skb_dequeue(&path->queue)))
242 dev_kfree_skb_irq(skb);
244 spin_lock_irqsave(&priv->lock, flags);
246 list_for_each_entry_safe(neigh, tn, &path->neigh_list, list) {
248 * It's safe to call ipoib_put_ah() inside priv->lock
249 * here, because we know that path->ah will always
250 * hold one more reference, so ipoib_put_ah() will
251 * never do more than decrement the ref count.
254 ipoib_put_ah(neigh->ah);
256 ipoib_neigh_free(neigh);
259 spin_unlock_irqrestore(&priv->lock, flags);
262 ipoib_put_ah(path->ah);
267 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
269 struct ipoib_path_iter *ipoib_path_iter_init(struct net_device *dev)
271 struct ipoib_path_iter *iter;
273 iter = kmalloc(sizeof *iter, GFP_KERNEL);
278 memset(iter->path.pathrec.dgid.raw, 0, 16);
280 if (ipoib_path_iter_next(iter)) {
288 int ipoib_path_iter_next(struct ipoib_path_iter *iter)
290 struct ipoib_dev_priv *priv = netdev_priv(iter->dev);
292 struct ipoib_path *path;
295 spin_lock_irq(&priv->lock);
297 n = rb_first(&priv->path_tree);
300 path = rb_entry(n, struct ipoib_path, rb_node);
302 if (memcmp(iter->path.pathrec.dgid.raw, path->pathrec.dgid.raw,
303 sizeof (union ib_gid)) < 0) {
312 spin_unlock_irq(&priv->lock);
317 void ipoib_path_iter_read(struct ipoib_path_iter *iter,
318 struct ipoib_path *path)
323 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */
325 void ipoib_flush_paths(struct net_device *dev)
327 struct ipoib_dev_priv *priv = netdev_priv(dev);
328 struct ipoib_path *path, *tp;
329 LIST_HEAD(remove_list);
332 spin_lock_irqsave(&priv->lock, flags);
334 list_splice(&priv->path_list, &remove_list);
335 INIT_LIST_HEAD(&priv->path_list);
337 list_for_each_entry(path, &remove_list, list)
338 rb_erase(&path->rb_node, &priv->path_tree);
340 spin_unlock_irqrestore(&priv->lock, flags);
342 list_for_each_entry_safe(path, tp, &remove_list, list) {
344 ib_sa_cancel_query(path->query_id, path->query);
345 wait_for_completion(&path->done);
346 path_free(dev, path);
350 static void path_rec_completion(int status,
351 struct ib_sa_path_rec *pathrec,
354 struct ipoib_path *path = path_ptr;
355 struct net_device *dev = path->dev;
356 struct ipoib_dev_priv *priv = netdev_priv(dev);
357 struct ipoib_ah *ah = NULL;
358 struct ipoib_neigh *neigh;
359 struct sk_buff_head skqueue;
364 ipoib_dbg(priv, "PathRec LID 0x%04x for GID " IPOIB_GID_FMT "\n",
365 be16_to_cpu(pathrec->dlid), IPOIB_GID_ARG(pathrec->dgid));
367 ipoib_dbg(priv, "PathRec status %d for GID " IPOIB_GID_FMT "\n",
368 status, IPOIB_GID_ARG(path->pathrec.dgid));
370 skb_queue_head_init(&skqueue);
373 struct ib_ah_attr av = {
374 .dlid = be16_to_cpu(pathrec->dlid),
376 .port_num = priv->port,
377 .static_rate = pathrec->rate
380 ah = ipoib_create_ah(dev, priv->pd, &av);
383 spin_lock_irqsave(&priv->lock, flags);
388 path->pathrec = *pathrec;
390 ipoib_dbg(priv, "created address handle %p for LID 0x%04x, SL %d\n",
391 ah, be16_to_cpu(pathrec->dlid), pathrec->sl);
393 while ((skb = __skb_dequeue(&path->queue)))
394 __skb_queue_tail(&skqueue, skb);
396 list_for_each_entry(neigh, &path->neigh_list, list) {
397 kref_get(&path->ah->ref);
398 neigh->ah = path->ah;
400 while ((skb = __skb_dequeue(&neigh->queue)))
401 __skb_queue_tail(&skqueue, skb);
406 complete(&path->done);
408 spin_unlock_irqrestore(&priv->lock, flags);
410 while ((skb = __skb_dequeue(&skqueue))) {
412 if (dev_queue_xmit(skb))
413 ipoib_warn(priv, "dev_queue_xmit failed "
414 "to requeue packet\n");
418 static struct ipoib_path *path_rec_create(struct net_device *dev,
421 struct ipoib_dev_priv *priv = netdev_priv(dev);
422 struct ipoib_path *path;
424 path = kzalloc(sizeof *path, GFP_ATOMIC);
430 skb_queue_head_init(&path->queue);
432 INIT_LIST_HEAD(&path->neigh_list);
434 memcpy(path->pathrec.dgid.raw, gid->raw, sizeof (union ib_gid));
435 path->pathrec.sgid = priv->local_gid;
436 path->pathrec.pkey = cpu_to_be16(priv->pkey);
437 path->pathrec.numb_path = 1;
442 static int path_rec_start(struct net_device *dev,
443 struct ipoib_path *path)
445 struct ipoib_dev_priv *priv = netdev_priv(dev);
447 ipoib_dbg(priv, "Start path record lookup for " IPOIB_GID_FMT "\n",
448 IPOIB_GID_ARG(path->pathrec.dgid));
450 init_completion(&path->done);
453 ib_sa_path_rec_get(priv->ca, priv->port,
455 IB_SA_PATH_REC_DGID |
456 IB_SA_PATH_REC_SGID |
457 IB_SA_PATH_REC_NUMB_PATH |
462 if (path->query_id < 0) {
463 ipoib_warn(priv, "ib_sa_path_rec_get failed\n");
465 return path->query_id;
471 static void neigh_add_path(struct sk_buff *skb, struct net_device *dev)
473 struct ipoib_dev_priv *priv = netdev_priv(dev);
474 struct ipoib_path *path;
475 struct ipoib_neigh *neigh;
477 neigh = ipoib_neigh_alloc(skb->dst->neighbour);
479 ++priv->stats.tx_dropped;
480 dev_kfree_skb_any(skb);
484 skb_queue_head_init(&neigh->queue);
487 * We can only be called from ipoib_start_xmit, so we're
488 * inside tx_lock -- no need to save/restore flags.
490 spin_lock(&priv->lock);
492 path = __path_find(dev, (union ib_gid *) (skb->dst->neighbour->ha + 4));
494 path = path_rec_create(dev,
495 (union ib_gid *) (skb->dst->neighbour->ha + 4));
499 __path_add(dev, path);
502 list_add_tail(&neigh->list, &path->neigh_list);
505 kref_get(&path->ah->ref);
506 neigh->ah = path->ah;
508 ipoib_send(dev, skb, path->ah,
509 be32_to_cpup((__be32 *) skb->dst->neighbour->ha));
512 __skb_queue_tail(&neigh->queue, skb);
514 if (!path->query && path_rec_start(dev, path))
518 spin_unlock(&priv->lock);
522 list_del(&neigh->list);
525 ipoib_neigh_free(neigh);
526 ++priv->stats.tx_dropped;
527 dev_kfree_skb_any(skb);
529 spin_unlock(&priv->lock);
532 static void ipoib_path_lookup(struct sk_buff *skb, struct net_device *dev)
534 struct ipoib_dev_priv *priv = netdev_priv(skb->dev);
536 /* Look up path record for unicasts */
537 if (skb->dst->neighbour->ha[4] != 0xff) {
538 neigh_add_path(skb, dev);
542 /* Add in the P_Key for multicasts */
543 skb->dst->neighbour->ha[8] = (priv->pkey >> 8) & 0xff;
544 skb->dst->neighbour->ha[9] = priv->pkey & 0xff;
545 ipoib_mcast_send(dev, (union ib_gid *) (skb->dst->neighbour->ha + 4), skb);
548 static void unicast_arp_send(struct sk_buff *skb, struct net_device *dev,
549 struct ipoib_pseudoheader *phdr)
551 struct ipoib_dev_priv *priv = netdev_priv(dev);
552 struct ipoib_path *path;
555 * We can only be called from ipoib_start_xmit, so we're
556 * inside tx_lock -- no need to save/restore flags.
558 spin_lock(&priv->lock);
560 path = __path_find(dev, (union ib_gid *) (phdr->hwaddr + 4));
562 path = path_rec_create(dev,
563 (union ib_gid *) (phdr->hwaddr + 4));
565 /* put pseudoheader back on for next time */
566 skb_push(skb, sizeof *phdr);
567 __skb_queue_tail(&path->queue, skb);
569 if (path_rec_start(dev, path)) {
570 spin_unlock(&priv->lock);
571 path_free(dev, path);
574 __path_add(dev, path);
576 ++priv->stats.tx_dropped;
577 dev_kfree_skb_any(skb);
580 spin_unlock(&priv->lock);
585 ipoib_dbg(priv, "Send unicast ARP to %04x\n",
586 be16_to_cpu(path->pathrec.dlid));
588 ipoib_send(dev, skb, path->ah,
589 be32_to_cpup((__be32 *) phdr->hwaddr));
590 } else if ((path->query || !path_rec_start(dev, path)) &&
591 skb_queue_len(&path->queue) < IPOIB_MAX_PATH_REC_QUEUE) {
592 /* put pseudoheader back on for next time */
593 skb_push(skb, sizeof *phdr);
594 __skb_queue_tail(&path->queue, skb);
596 ++priv->stats.tx_dropped;
597 dev_kfree_skb_any(skb);
600 spin_unlock(&priv->lock);
603 static int ipoib_start_xmit(struct sk_buff *skb, struct net_device *dev)
605 struct ipoib_dev_priv *priv = netdev_priv(dev);
606 struct ipoib_neigh *neigh;
609 if (!spin_trylock_irqsave(&priv->tx_lock, flags))
610 return NETDEV_TX_LOCKED;
613 * Check if our queue is stopped. Since we have the LLTX bit
614 * set, we can't rely on netif_stop_queue() preventing our
615 * xmit function from being called with a full queue.
617 if (unlikely(netif_queue_stopped(dev))) {
618 spin_unlock_irqrestore(&priv->tx_lock, flags);
619 return NETDEV_TX_BUSY;
622 if (skb->dst && skb->dst->neighbour) {
623 if (unlikely(!*to_ipoib_neigh(skb->dst->neighbour))) {
624 ipoib_path_lookup(skb, dev);
628 neigh = *to_ipoib_neigh(skb->dst->neighbour);
630 if (likely(neigh->ah)) {
631 ipoib_send(dev, skb, neigh->ah,
632 be32_to_cpup((__be32 *) skb->dst->neighbour->ha));
636 if (skb_queue_len(&neigh->queue) < IPOIB_MAX_PATH_REC_QUEUE) {
637 spin_lock(&priv->lock);
638 __skb_queue_tail(&neigh->queue, skb);
639 spin_unlock(&priv->lock);
641 ++priv->stats.tx_dropped;
642 dev_kfree_skb_any(skb);
645 struct ipoib_pseudoheader *phdr =
646 (struct ipoib_pseudoheader *) skb->data;
647 skb_pull(skb, sizeof *phdr);
649 if (phdr->hwaddr[4] == 0xff) {
650 /* Add in the P_Key for multicast*/
651 phdr->hwaddr[8] = (priv->pkey >> 8) & 0xff;
652 phdr->hwaddr[9] = priv->pkey & 0xff;
654 ipoib_mcast_send(dev, (union ib_gid *) (phdr->hwaddr + 4), skb);
656 /* unicast GID -- should be ARP or RARP reply */
658 if ((be16_to_cpup((__be16 *) skb->data) != ETH_P_ARP) &&
659 (be16_to_cpup((__be16 *) skb->data) != ETH_P_RARP)) {
660 ipoib_warn(priv, "Unicast, no %s: type %04x, QPN %06x "
662 skb->dst ? "neigh" : "dst",
663 be16_to_cpup((__be16 *) skb->data),
664 be32_to_cpup((__be32 *) phdr->hwaddr),
665 IPOIB_GID_ARG(*(union ib_gid *) (phdr->hwaddr + 4)));
666 dev_kfree_skb_any(skb);
667 ++priv->stats.tx_dropped;
671 unicast_arp_send(skb, dev, phdr);
676 spin_unlock_irqrestore(&priv->tx_lock, flags);
681 static struct net_device_stats *ipoib_get_stats(struct net_device *dev)
683 struct ipoib_dev_priv *priv = netdev_priv(dev);
688 static void ipoib_timeout(struct net_device *dev)
690 struct ipoib_dev_priv *priv = netdev_priv(dev);
692 ipoib_warn(priv, "transmit timeout: latency %d msecs\n",
693 jiffies_to_msecs(jiffies - dev->trans_start));
694 ipoib_warn(priv, "queue stopped %d, tx_head %u, tx_tail %u\n",
695 netif_queue_stopped(dev),
696 priv->tx_head, priv->tx_tail);
697 /* XXX reset QP, etc. */
700 static int ipoib_hard_header(struct sk_buff *skb,
701 struct net_device *dev,
703 void *daddr, void *saddr, unsigned len)
705 struct ipoib_header *header;
707 header = (struct ipoib_header *) skb_push(skb, sizeof *header);
709 header->proto = htons(type);
710 header->reserved = 0;
713 * If we don't have a neighbour structure, stuff the
714 * destination address onto the front of the skb so we can
715 * figure out where to send the packet later.
717 if ((!skb->dst || !skb->dst->neighbour) && daddr) {
718 struct ipoib_pseudoheader *phdr =
719 (struct ipoib_pseudoheader *) skb_push(skb, sizeof *phdr);
720 memcpy(phdr->hwaddr, daddr, INFINIBAND_ALEN);
726 static void ipoib_set_mcast_list(struct net_device *dev)
728 struct ipoib_dev_priv *priv = netdev_priv(dev);
730 if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags)) {
731 ipoib_dbg(priv, "IPOIB_FLAG_OPER_UP not set");
735 queue_work(ipoib_workqueue, &priv->restart_task);
738 static void ipoib_neigh_destructor(struct neighbour *n)
740 struct ipoib_neigh *neigh;
741 struct ipoib_dev_priv *priv = netdev_priv(n->dev);
743 struct ipoib_ah *ah = NULL;
746 "neigh_destructor for %06x " IPOIB_GID_FMT "\n",
747 be32_to_cpup((__be32 *) n->ha),
748 IPOIB_GID_ARG(*((union ib_gid *) (n->ha + 4))));
750 spin_lock_irqsave(&priv->lock, flags);
752 neigh = *to_ipoib_neigh(n);
756 list_del(&neigh->list);
757 ipoib_neigh_free(neigh);
760 spin_unlock_irqrestore(&priv->lock, flags);
766 struct ipoib_neigh *ipoib_neigh_alloc(struct neighbour *neighbour)
768 struct ipoib_neigh *neigh;
770 neigh = kmalloc(sizeof *neigh, GFP_ATOMIC);
774 neigh->neighbour = neighbour;
775 *to_ipoib_neigh(neighbour) = neigh;
780 void ipoib_neigh_free(struct ipoib_neigh *neigh)
782 *to_ipoib_neigh(neigh->neighbour) = NULL;
786 static int ipoib_neigh_setup_dev(struct net_device *dev, struct neigh_parms *parms)
788 parms->neigh_destructor = ipoib_neigh_destructor;
793 int ipoib_dev_init(struct net_device *dev, struct ib_device *ca, int port)
795 struct ipoib_dev_priv *priv = netdev_priv(dev);
797 /* Allocate RX/TX "rings" to hold queued skbs */
799 priv->rx_ring = kzalloc(IPOIB_RX_RING_SIZE * sizeof (struct ipoib_rx_buf),
801 if (!priv->rx_ring) {
802 printk(KERN_WARNING "%s: failed to allocate RX ring (%d entries)\n",
803 ca->name, IPOIB_RX_RING_SIZE);
807 priv->tx_ring = kzalloc(IPOIB_TX_RING_SIZE * sizeof (struct ipoib_tx_buf),
809 if (!priv->tx_ring) {
810 printk(KERN_WARNING "%s: failed to allocate TX ring (%d entries)\n",
811 ca->name, IPOIB_TX_RING_SIZE);
812 goto out_rx_ring_cleanup;
815 /* priv->tx_head & tx_tail are already 0 */
817 if (ipoib_ib_dev_init(dev, ca, port))
818 goto out_tx_ring_cleanup;
823 kfree(priv->tx_ring);
826 kfree(priv->rx_ring);
832 void ipoib_dev_cleanup(struct net_device *dev)
834 struct ipoib_dev_priv *priv = netdev_priv(dev), *cpriv, *tcpriv;
836 ipoib_delete_debug_files(dev);
838 /* Delete any child interfaces first */
839 list_for_each_entry_safe(cpriv, tcpriv, &priv->child_intfs, list) {
840 unregister_netdev(cpriv->dev);
841 ipoib_dev_cleanup(cpriv->dev);
842 free_netdev(cpriv->dev);
845 ipoib_ib_dev_cleanup(dev);
847 kfree(priv->rx_ring);
848 kfree(priv->tx_ring);
850 priv->rx_ring = NULL;
851 priv->tx_ring = NULL;
854 static void ipoib_setup(struct net_device *dev)
856 struct ipoib_dev_priv *priv = netdev_priv(dev);
858 dev->open = ipoib_open;
859 dev->stop = ipoib_stop;
860 dev->change_mtu = ipoib_change_mtu;
861 dev->hard_start_xmit = ipoib_start_xmit;
862 dev->get_stats = ipoib_get_stats;
863 dev->tx_timeout = ipoib_timeout;
864 dev->hard_header = ipoib_hard_header;
865 dev->set_multicast_list = ipoib_set_mcast_list;
866 dev->neigh_setup = ipoib_neigh_setup_dev;
868 dev->watchdog_timeo = HZ;
870 dev->flags |= IFF_BROADCAST | IFF_MULTICAST;
873 * We add in INFINIBAND_ALEN to allow for the destination
874 * address "pseudoheader" for skbs without neighbour struct.
876 dev->hard_header_len = IPOIB_ENCAP_LEN + INFINIBAND_ALEN;
877 dev->addr_len = INFINIBAND_ALEN;
878 dev->type = ARPHRD_INFINIBAND;
879 dev->tx_queue_len = IPOIB_TX_RING_SIZE * 2;
880 dev->features = NETIF_F_VLAN_CHALLENGED | NETIF_F_LLTX;
882 /* MTU will be reset when mcast join happens */
883 dev->mtu = IPOIB_PACKET_SIZE - IPOIB_ENCAP_LEN;
884 priv->mcast_mtu = priv->admin_mtu = dev->mtu;
886 memcpy(dev->broadcast, ipv4_bcast_addr, INFINIBAND_ALEN);
888 netif_carrier_off(dev);
890 SET_MODULE_OWNER(dev);
894 spin_lock_init(&priv->lock);
895 spin_lock_init(&priv->tx_lock);
897 mutex_init(&priv->mcast_mutex);
898 mutex_init(&priv->vlan_mutex);
900 INIT_LIST_HEAD(&priv->path_list);
901 INIT_LIST_HEAD(&priv->child_intfs);
902 INIT_LIST_HEAD(&priv->dead_ahs);
903 INIT_LIST_HEAD(&priv->multicast_list);
905 INIT_WORK(&priv->pkey_task, ipoib_pkey_poll, priv->dev);
906 INIT_WORK(&priv->mcast_task, ipoib_mcast_join_task, priv->dev);
907 INIT_WORK(&priv->flush_task, ipoib_ib_dev_flush, priv->dev);
908 INIT_WORK(&priv->restart_task, ipoib_mcast_restart_task, priv->dev);
909 INIT_WORK(&priv->ah_reap_task, ipoib_reap_ah, priv->dev);
912 struct ipoib_dev_priv *ipoib_intf_alloc(const char *name)
914 struct net_device *dev;
916 dev = alloc_netdev((int) sizeof (struct ipoib_dev_priv), name,
921 return netdev_priv(dev);
924 static ssize_t show_pkey(struct class_device *cdev, char *buf)
926 struct ipoib_dev_priv *priv =
927 netdev_priv(container_of(cdev, struct net_device, class_dev));
929 return sprintf(buf, "0x%04x\n", priv->pkey);
931 static CLASS_DEVICE_ATTR(pkey, S_IRUGO, show_pkey, NULL);
933 static ssize_t create_child(struct class_device *cdev,
934 const char *buf, size_t count)
939 if (sscanf(buf, "%i", &pkey) != 1)
942 if (pkey < 0 || pkey > 0xffff)
946 * Set the full membership bit, so that we join the right
947 * broadcast group, etc.
951 ret = ipoib_vlan_add(container_of(cdev, struct net_device, class_dev),
954 return ret ? ret : count;
956 static CLASS_DEVICE_ATTR(create_child, S_IWUGO, NULL, create_child);
958 static ssize_t delete_child(struct class_device *cdev,
959 const char *buf, size_t count)
964 if (sscanf(buf, "%i", &pkey) != 1)
967 if (pkey < 0 || pkey > 0xffff)
970 ret = ipoib_vlan_delete(container_of(cdev, struct net_device, class_dev),
973 return ret ? ret : count;
976 static CLASS_DEVICE_ATTR(delete_child, S_IWUGO, NULL, delete_child);
978 int ipoib_add_pkey_attr(struct net_device *dev)
980 return class_device_create_file(&dev->class_dev,
981 &class_device_attr_pkey);
984 static struct net_device *ipoib_add_port(const char *format,
985 struct ib_device *hca, u8 port)
987 struct ipoib_dev_priv *priv;
988 int result = -ENOMEM;
990 priv = ipoib_intf_alloc(format);
992 goto alloc_mem_failed;
994 SET_NETDEV_DEV(priv->dev, hca->dma_device);
996 result = ib_query_pkey(hca, port, 0, &priv->pkey);
998 printk(KERN_WARNING "%s: ib_query_pkey port %d failed (ret = %d)\n",
999 hca->name, port, result);
1000 goto alloc_mem_failed;
1004 * Set the full membership bit, so that we join the right
1005 * broadcast group, etc.
1007 priv->pkey |= 0x8000;
1009 priv->dev->broadcast[8] = priv->pkey >> 8;
1010 priv->dev->broadcast[9] = priv->pkey & 0xff;
1012 result = ib_query_gid(hca, port, 0, &priv->local_gid);
1014 printk(KERN_WARNING "%s: ib_query_gid port %d failed (ret = %d)\n",
1015 hca->name, port, result);
1016 goto alloc_mem_failed;
1018 memcpy(priv->dev->dev_addr + 4, priv->local_gid.raw, sizeof (union ib_gid));
1021 result = ipoib_dev_init(priv->dev, hca, port);
1023 printk(KERN_WARNING "%s: failed to initialize port %d (ret = %d)\n",
1024 hca->name, port, result);
1025 goto device_init_failed;
1028 INIT_IB_EVENT_HANDLER(&priv->event_handler,
1029 priv->ca, ipoib_event);
1030 result = ib_register_event_handler(&priv->event_handler);
1032 printk(KERN_WARNING "%s: ib_register_event_handler failed for "
1033 "port %d (ret = %d)\n",
1034 hca->name, port, result);
1038 result = register_netdev(priv->dev);
1040 printk(KERN_WARNING "%s: couldn't register ipoib port %d; error %d\n",
1041 hca->name, port, result);
1042 goto register_failed;
1045 ipoib_create_debug_files(priv->dev);
1047 if (ipoib_add_pkey_attr(priv->dev))
1049 if (class_device_create_file(&priv->dev->class_dev,
1050 &class_device_attr_create_child))
1052 if (class_device_create_file(&priv->dev->class_dev,
1053 &class_device_attr_delete_child))
1059 ipoib_delete_debug_files(priv->dev);
1060 unregister_netdev(priv->dev);
1063 ib_unregister_event_handler(&priv->event_handler);
1064 flush_scheduled_work();
1067 ipoib_dev_cleanup(priv->dev);
1070 free_netdev(priv->dev);
1073 return ERR_PTR(result);
1076 static void ipoib_add_one(struct ib_device *device)
1078 struct list_head *dev_list;
1079 struct net_device *dev;
1080 struct ipoib_dev_priv *priv;
1083 dev_list = kmalloc(sizeof *dev_list, GFP_KERNEL);
1087 INIT_LIST_HEAD(dev_list);
1089 if (device->node_type == IB_NODE_SWITCH) {
1094 e = device->phys_port_cnt;
1097 for (p = s; p <= e; ++p) {
1098 dev = ipoib_add_port("ib%d", device, p);
1100 priv = netdev_priv(dev);
1101 list_add_tail(&priv->list, dev_list);
1105 ib_set_client_data(device, &ipoib_client, dev_list);
1108 static void ipoib_remove_one(struct ib_device *device)
1110 struct ipoib_dev_priv *priv, *tmp;
1111 struct list_head *dev_list;
1113 dev_list = ib_get_client_data(device, &ipoib_client);
1115 list_for_each_entry_safe(priv, tmp, dev_list, list) {
1116 ib_unregister_event_handler(&priv->event_handler);
1117 flush_scheduled_work();
1119 unregister_netdev(priv->dev);
1120 ipoib_dev_cleanup(priv->dev);
1121 free_netdev(priv->dev);
1127 static int __init ipoib_init_module(void)
1131 ret = ipoib_register_debugfs();
1136 * We create our own workqueue mainly because we want to be
1137 * able to flush it when devices are being removed. We can't
1138 * use schedule_work()/flush_scheduled_work() because both
1139 * unregister_netdev() and linkwatch_event take the rtnl lock,
1140 * so flush_scheduled_work() can deadlock during device
1143 ipoib_workqueue = create_singlethread_workqueue("ipoib");
1144 if (!ipoib_workqueue) {
1149 ret = ib_register_client(&ipoib_client);
1156 destroy_workqueue(ipoib_workqueue);
1159 ipoib_unregister_debugfs();
1164 static void __exit ipoib_cleanup_module(void)
1166 ib_unregister_client(&ipoib_client);
1167 ipoib_unregister_debugfs();
1168 destroy_workqueue(ipoib_workqueue);
1171 module_init(ipoib_init_module);
1172 module_exit(ipoib_cleanup_module);