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 */
50 MODULE_AUTHOR("Roland Dreier");
51 MODULE_DESCRIPTION("IP-over-InfiniBand net driver");
52 MODULE_LICENSE("Dual BSD/GPL");
54 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
55 int ipoib_debug_level
;
57 module_param_named(debug_level
, ipoib_debug_level
, int, 0644);
58 MODULE_PARM_DESC(debug_level
, "Enable debug tracing if > 0");
61 static const u8 ipv4_bcast_addr
[] = {
62 0x00, 0xff, 0xff, 0xff,
63 0xff, 0x12, 0x40, 0x1b, 0x00, 0x00, 0x00, 0x00,
64 0x00, 0x00, 0x00, 0x00, 0xff, 0xff, 0xff, 0xff
67 struct workqueue_struct
*ipoib_workqueue
;
69 static void ipoib_add_one(struct ib_device
*device
);
70 static void ipoib_remove_one(struct ib_device
*device
);
72 static struct ib_client ipoib_client
= {
75 .remove
= ipoib_remove_one
78 int ipoib_open(struct net_device
*dev
)
80 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
82 ipoib_dbg(priv
, "bringing up interface\n");
84 set_bit(IPOIB_FLAG_ADMIN_UP
, &priv
->flags
);
86 if (ipoib_pkey_dev_delay_open(dev
))
89 if (ipoib_ib_dev_open(dev
))
92 if (ipoib_ib_dev_up(dev
))
95 if (!test_bit(IPOIB_FLAG_SUBINTERFACE
, &priv
->flags
)) {
96 struct ipoib_dev_priv
*cpriv
;
98 /* Bring up any child interfaces too */
99 down(&priv
->vlan_mutex
);
100 list_for_each_entry(cpriv
, &priv
->child_intfs
, list
) {
103 flags
= cpriv
->dev
->flags
;
107 dev_change_flags(cpriv
->dev
, flags
| IFF_UP
);
109 up(&priv
->vlan_mutex
);
112 netif_start_queue(dev
);
117 static int ipoib_stop(struct net_device
*dev
)
119 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
121 ipoib_dbg(priv
, "stopping interface\n");
123 clear_bit(IPOIB_FLAG_ADMIN_UP
, &priv
->flags
);
125 netif_stop_queue(dev
);
127 ipoib_ib_dev_down(dev
);
128 ipoib_ib_dev_stop(dev
);
130 if (!test_bit(IPOIB_FLAG_SUBINTERFACE
, &priv
->flags
)) {
131 struct ipoib_dev_priv
*cpriv
;
133 /* Bring down any child interfaces too */
134 down(&priv
->vlan_mutex
);
135 list_for_each_entry(cpriv
, &priv
->child_intfs
, list
) {
138 flags
= cpriv
->dev
->flags
;
139 if (!(flags
& IFF_UP
))
142 dev_change_flags(cpriv
->dev
, flags
& ~IFF_UP
);
144 up(&priv
->vlan_mutex
);
150 static int ipoib_change_mtu(struct net_device
*dev
, int new_mtu
)
152 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
154 if (new_mtu
> IPOIB_PACKET_SIZE
- IPOIB_ENCAP_LEN
)
157 priv
->admin_mtu
= new_mtu
;
159 dev
->mtu
= min(priv
->mcast_mtu
, priv
->admin_mtu
);
164 static struct ipoib_path
*__path_find(struct net_device
*dev
,
167 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
168 struct rb_node
*n
= priv
->path_tree
.rb_node
;
169 struct ipoib_path
*path
;
173 path
= rb_entry(n
, struct ipoib_path
, rb_node
);
175 ret
= memcmp(gid
->raw
, path
->pathrec
.dgid
.raw
,
176 sizeof (union ib_gid
));
189 static int __path_add(struct net_device
*dev
, struct ipoib_path
*path
)
191 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
192 struct rb_node
**n
= &priv
->path_tree
.rb_node
;
193 struct rb_node
*pn
= NULL
;
194 struct ipoib_path
*tpath
;
199 tpath
= rb_entry(pn
, struct ipoib_path
, rb_node
);
201 ret
= memcmp(path
->pathrec
.dgid
.raw
, tpath
->pathrec
.dgid
.raw
,
202 sizeof (union ib_gid
));
211 rb_link_node(&path
->rb_node
, pn
, n
);
212 rb_insert_color(&path
->rb_node
, &priv
->path_tree
);
214 list_add_tail(&path
->list
, &priv
->path_list
);
219 static void path_free(struct net_device
*dev
, struct ipoib_path
*path
)
221 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
222 struct ipoib_neigh
*neigh
, *tn
;
226 while ((skb
= __skb_dequeue(&path
->queue
)))
227 dev_kfree_skb_irq(skb
);
229 spin_lock_irqsave(&priv
->lock
, flags
);
231 list_for_each_entry_safe(neigh
, tn
, &path
->neigh_list
, list
) {
233 * It's safe to call ipoib_put_ah() inside priv->lock
234 * here, because we know that path->ah will always
235 * hold one more reference, so ipoib_put_ah() will
236 * never do more than decrement the ref count.
239 ipoib_put_ah(neigh
->ah
);
240 *to_ipoib_neigh(neigh
->neighbour
) = NULL
;
241 neigh
->neighbour
->ops
->destructor
= NULL
;
245 spin_unlock_irqrestore(&priv
->lock
, flags
);
248 ipoib_put_ah(path
->ah
);
253 void ipoib_flush_paths(struct net_device
*dev
)
255 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
256 struct ipoib_path
*path
, *tp
;
257 LIST_HEAD(remove_list
);
260 spin_lock_irqsave(&priv
->lock
, flags
);
262 list_splice(&priv
->path_list
, &remove_list
);
263 INIT_LIST_HEAD(&priv
->path_list
);
265 list_for_each_entry(path
, &remove_list
, list
)
266 rb_erase(&path
->rb_node
, &priv
->path_tree
);
268 spin_unlock_irqrestore(&priv
->lock
, flags
);
270 list_for_each_entry_safe(path
, tp
, &remove_list
, list
) {
272 ib_sa_cancel_query(path
->query_id
, path
->query
);
273 wait_for_completion(&path
->done
);
274 path_free(dev
, path
);
278 static void path_rec_completion(int status
,
279 struct ib_sa_path_rec
*pathrec
,
282 struct ipoib_path
*path
= path_ptr
;
283 struct net_device
*dev
= path
->dev
;
284 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
285 struct ipoib_ah
*ah
= NULL
;
286 struct ipoib_neigh
*neigh
;
287 struct sk_buff_head skqueue
;
292 ipoib_dbg(priv
, "PathRec LID 0x%04x for GID " IPOIB_GID_FMT
"\n",
293 be16_to_cpu(pathrec
->dlid
), IPOIB_GID_ARG(pathrec
->dgid
));
295 ipoib_dbg(priv
, "PathRec status %d for GID " IPOIB_GID_FMT
"\n",
296 status
, IPOIB_GID_ARG(path
->pathrec
.dgid
));
298 skb_queue_head_init(&skqueue
);
301 struct ib_ah_attr av
= {
302 .dlid
= be16_to_cpu(pathrec
->dlid
),
304 .port_num
= priv
->port
306 int path_rate
= ib_sa_rate_enum_to_int(pathrec
->rate
);
308 if (path_rate
> 0 && priv
->local_rate
> path_rate
)
309 av
.static_rate
= (priv
->local_rate
- 1) / path_rate
;
311 ipoib_dbg(priv
, "static_rate %d for local port %dX, path %dX\n",
312 av
.static_rate
, priv
->local_rate
,
313 ib_sa_rate_enum_to_int(pathrec
->rate
));
315 ah
= ipoib_create_ah(dev
, priv
->pd
, &av
);
318 spin_lock_irqsave(&priv
->lock
, flags
);
323 path
->pathrec
= *pathrec
;
325 ipoib_dbg(priv
, "created address handle %p for LID 0x%04x, SL %d\n",
326 ah
, be16_to_cpu(pathrec
->dlid
), pathrec
->sl
);
328 while ((skb
= __skb_dequeue(&path
->queue
)))
329 __skb_queue_tail(&skqueue
, skb
);
331 list_for_each_entry(neigh
, &path
->neigh_list
, list
) {
332 kref_get(&path
->ah
->ref
);
333 neigh
->ah
= path
->ah
;
335 while ((skb
= __skb_dequeue(&neigh
->queue
)))
336 __skb_queue_tail(&skqueue
, skb
);
341 complete(&path
->done
);
343 spin_unlock_irqrestore(&priv
->lock
, flags
);
345 while ((skb
= __skb_dequeue(&skqueue
))) {
347 if (dev_queue_xmit(skb
))
348 ipoib_warn(priv
, "dev_queue_xmit failed "
349 "to requeue packet\n");
353 static struct ipoib_path
*path_rec_create(struct net_device
*dev
,
356 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
357 struct ipoib_path
*path
;
359 path
= kmalloc(sizeof *path
, GFP_ATOMIC
);
364 path
->pathrec
.dlid
= 0;
367 skb_queue_head_init(&path
->queue
);
369 INIT_LIST_HEAD(&path
->neigh_list
);
371 init_completion(&path
->done
);
373 memcpy(path
->pathrec
.dgid
.raw
, gid
->raw
, sizeof (union ib_gid
));
374 path
->pathrec
.sgid
= priv
->local_gid
;
375 path
->pathrec
.pkey
= cpu_to_be16(priv
->pkey
);
376 path
->pathrec
.numb_path
= 1;
381 static int path_rec_start(struct net_device
*dev
,
382 struct ipoib_path
*path
)
384 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
386 ipoib_dbg(priv
, "Start path record lookup for " IPOIB_GID_FMT
"\n",
387 IPOIB_GID_ARG(path
->pathrec
.dgid
));
390 ib_sa_path_rec_get(priv
->ca
, priv
->port
,
392 IB_SA_PATH_REC_DGID
|
393 IB_SA_PATH_REC_SGID
|
394 IB_SA_PATH_REC_NUMB_PATH
|
399 if (path
->query_id
< 0) {
400 ipoib_warn(priv
, "ib_sa_path_rec_get failed\n");
402 return path
->query_id
;
408 static void neigh_add_path(struct sk_buff
*skb
, struct net_device
*dev
)
410 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
411 struct ipoib_path
*path
;
412 struct ipoib_neigh
*neigh
;
414 neigh
= kmalloc(sizeof *neigh
, GFP_ATOMIC
);
416 ++priv
->stats
.tx_dropped
;
417 dev_kfree_skb_any(skb
);
421 skb_queue_head_init(&neigh
->queue
);
422 neigh
->neighbour
= skb
->dst
->neighbour
;
423 *to_ipoib_neigh(skb
->dst
->neighbour
) = neigh
;
426 * We can only be called from ipoib_start_xmit, so we're
427 * inside tx_lock -- no need to save/restore flags.
429 spin_lock(&priv
->lock
);
431 path
= __path_find(dev
, (union ib_gid
*) (skb
->dst
->neighbour
->ha
+ 4));
433 path
= path_rec_create(dev
,
434 (union ib_gid
*) (skb
->dst
->neighbour
->ha
+ 4));
438 __path_add(dev
, path
);
441 list_add_tail(&neigh
->list
, &path
->neigh_list
);
443 if (path
->pathrec
.dlid
) {
444 kref_get(&path
->ah
->ref
);
445 neigh
->ah
= path
->ah
;
447 ipoib_send(dev
, skb
, path
->ah
,
448 be32_to_cpup((__be32
*) skb
->dst
->neighbour
->ha
));
451 if (skb_queue_len(&neigh
->queue
) < IPOIB_MAX_PATH_REC_QUEUE
) {
452 __skb_queue_tail(&neigh
->queue
, skb
);
454 ++priv
->stats
.tx_dropped
;
455 dev_kfree_skb_any(skb
);
458 if (!path
->query
&& path_rec_start(dev
, path
))
462 spin_unlock(&priv
->lock
);
466 *to_ipoib_neigh(skb
->dst
->neighbour
) = NULL
;
467 list_del(&neigh
->list
);
468 neigh
->neighbour
->ops
->destructor
= NULL
;
471 ++priv
->stats
.tx_dropped
;
472 dev_kfree_skb_any(skb
);
474 spin_unlock(&priv
->lock
);
477 static void path_lookup(struct sk_buff
*skb
, struct net_device
*dev
)
479 struct ipoib_dev_priv
*priv
= netdev_priv(skb
->dev
);
481 /* Look up path record for unicasts */
482 if (skb
->dst
->neighbour
->ha
[4] != 0xff) {
483 neigh_add_path(skb
, dev
);
487 /* Add in the P_Key for multicasts */
488 skb
->dst
->neighbour
->ha
[8] = (priv
->pkey
>> 8) & 0xff;
489 skb
->dst
->neighbour
->ha
[9] = priv
->pkey
& 0xff;
490 ipoib_mcast_send(dev
, (union ib_gid
*) (skb
->dst
->neighbour
->ha
+ 4), skb
);
493 static void unicast_arp_send(struct sk_buff
*skb
, struct net_device
*dev
,
494 struct ipoib_pseudoheader
*phdr
)
496 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
497 struct ipoib_path
*path
;
500 * We can only be called from ipoib_start_xmit, so we're
501 * inside tx_lock -- no need to save/restore flags.
503 spin_lock(&priv
->lock
);
505 path
= __path_find(dev
, (union ib_gid
*) (phdr
->hwaddr
+ 4));
507 path
= path_rec_create(dev
,
508 (union ib_gid
*) (phdr
->hwaddr
+ 4));
510 /* put pseudoheader back on for next time */
511 skb_push(skb
, sizeof *phdr
);
512 __skb_queue_tail(&path
->queue
, skb
);
514 if (path_rec_start(dev
, path
)) {
515 spin_unlock(&priv
->lock
);
516 path_free(dev
, path
);
519 __path_add(dev
, path
);
521 ++priv
->stats
.tx_dropped
;
522 dev_kfree_skb_any(skb
);
525 spin_unlock(&priv
->lock
);
529 if (path
->pathrec
.dlid
) {
530 ipoib_dbg(priv
, "Send unicast ARP to %04x\n",
531 be16_to_cpu(path
->pathrec
.dlid
));
533 ipoib_send(dev
, skb
, path
->ah
,
534 be32_to_cpup((__be32
*) phdr
->hwaddr
));
535 } else if ((path
->query
|| !path_rec_start(dev
, path
)) &&
536 skb_queue_len(&path
->queue
) < IPOIB_MAX_PATH_REC_QUEUE
) {
537 /* put pseudoheader back on for next time */
538 skb_push(skb
, sizeof *phdr
);
539 __skb_queue_tail(&path
->queue
, skb
);
541 ++priv
->stats
.tx_dropped
;
542 dev_kfree_skb_any(skb
);
545 spin_unlock(&priv
->lock
);
548 static int ipoib_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
550 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
551 struct ipoib_neigh
*neigh
;
554 local_irq_save(flags
);
555 if (!spin_trylock(&priv
->tx_lock
)) {
556 local_irq_restore(flags
);
557 return NETDEV_TX_LOCKED
;
561 * Check if our queue is stopped. Since we have the LLTX bit
562 * set, we can't rely on netif_stop_queue() preventing our
563 * xmit function from being called with a full queue.
565 if (unlikely(netif_queue_stopped(dev
))) {
566 spin_unlock_irqrestore(&priv
->tx_lock
, flags
);
567 return NETDEV_TX_BUSY
;
570 if (skb
->dst
&& skb
->dst
->neighbour
) {
571 if (unlikely(!*to_ipoib_neigh(skb
->dst
->neighbour
))) {
572 path_lookup(skb
, dev
);
576 neigh
= *to_ipoib_neigh(skb
->dst
->neighbour
);
578 if (likely(neigh
->ah
)) {
579 ipoib_send(dev
, skb
, neigh
->ah
,
580 be32_to_cpup((__be32
*) skb
->dst
->neighbour
->ha
));
584 if (skb_queue_len(&neigh
->queue
) < IPOIB_MAX_PATH_REC_QUEUE
) {
585 spin_lock(&priv
->lock
);
586 __skb_queue_tail(&neigh
->queue
, skb
);
587 spin_unlock(&priv
->lock
);
589 ++priv
->stats
.tx_dropped
;
590 dev_kfree_skb_any(skb
);
593 struct ipoib_pseudoheader
*phdr
=
594 (struct ipoib_pseudoheader
*) skb
->data
;
595 skb_pull(skb
, sizeof *phdr
);
597 if (phdr
->hwaddr
[4] == 0xff) {
598 /* Add in the P_Key for multicast*/
599 phdr
->hwaddr
[8] = (priv
->pkey
>> 8) & 0xff;
600 phdr
->hwaddr
[9] = priv
->pkey
& 0xff;
602 ipoib_mcast_send(dev
, (union ib_gid
*) (phdr
->hwaddr
+ 4), skb
);
604 /* unicast GID -- should be ARP or RARP reply */
606 if ((be16_to_cpup((__be16
*) skb
->data
) != ETH_P_ARP
) &&
607 (be16_to_cpup((__be16
*) skb
->data
) != ETH_P_RARP
)) {
608 ipoib_warn(priv
, "Unicast, no %s: type %04x, QPN %06x "
610 skb
->dst
? "neigh" : "dst",
611 be16_to_cpup((__be16
*) skb
->data
),
612 be32_to_cpup((__be32
*) phdr
->hwaddr
),
613 IPOIB_GID_ARG(*(union ib_gid
*) (phdr
->hwaddr
+ 4)));
614 dev_kfree_skb_any(skb
);
615 ++priv
->stats
.tx_dropped
;
619 unicast_arp_send(skb
, dev
, phdr
);
624 spin_unlock_irqrestore(&priv
->tx_lock
, flags
);
629 static struct net_device_stats
*ipoib_get_stats(struct net_device
*dev
)
631 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
636 static void ipoib_timeout(struct net_device
*dev
)
638 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
640 ipoib_warn(priv
, "transmit timeout: latency %ld\n",
641 jiffies
- dev
->trans_start
);
642 /* XXX reset QP, etc. */
645 static int ipoib_hard_header(struct sk_buff
*skb
,
646 struct net_device
*dev
,
648 void *daddr
, void *saddr
, unsigned len
)
650 struct ipoib_header
*header
;
652 header
= (struct ipoib_header
*) skb_push(skb
, sizeof *header
);
654 header
->proto
= htons(type
);
655 header
->reserved
= 0;
658 * If we don't have a neighbour structure, stuff the
659 * destination address onto the front of the skb so we can
660 * figure out where to send the packet later.
662 if (!skb
->dst
|| !skb
->dst
->neighbour
) {
663 struct ipoib_pseudoheader
*phdr
=
664 (struct ipoib_pseudoheader
*) skb_push(skb
, sizeof *phdr
);
665 memcpy(phdr
->hwaddr
, daddr
, INFINIBAND_ALEN
);
671 static void ipoib_set_mcast_list(struct net_device
*dev
)
673 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
675 schedule_work(&priv
->restart_task
);
678 static void ipoib_neigh_destructor(struct neighbour
*n
)
680 struct ipoib_neigh
*neigh
;
681 struct ipoib_dev_priv
*priv
= netdev_priv(n
->dev
);
683 struct ipoib_ah
*ah
= NULL
;
686 "neigh_destructor for %06x " IPOIB_GID_FMT
"\n",
687 be32_to_cpup((__be32
*) n
->ha
),
688 IPOIB_GID_ARG(*((union ib_gid
*) (n
->ha
+ 4))));
690 spin_lock_irqsave(&priv
->lock
, flags
);
692 neigh
= *to_ipoib_neigh(n
);
696 list_del(&neigh
->list
);
697 *to_ipoib_neigh(n
) = NULL
;
701 spin_unlock_irqrestore(&priv
->lock
, flags
);
707 static int ipoib_neigh_setup(struct neighbour
*neigh
)
710 * Is this kosher? I can't find anybody in the kernel that
711 * sets neigh->destructor, so we should be able to set it here
714 neigh
->ops
->destructor
= ipoib_neigh_destructor
;
719 static int ipoib_neigh_setup_dev(struct net_device
*dev
, struct neigh_parms
*parms
)
721 parms
->neigh_setup
= ipoib_neigh_setup
;
726 int ipoib_dev_init(struct net_device
*dev
, struct ib_device
*ca
, int port
)
728 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
730 /* Allocate RX/TX "rings" to hold queued skbs */
732 priv
->rx_ring
= kmalloc(IPOIB_RX_RING_SIZE
* sizeof (struct ipoib_buf
),
734 if (!priv
->rx_ring
) {
735 printk(KERN_WARNING
"%s: failed to allocate RX ring (%d entries)\n",
736 ca
->name
, IPOIB_RX_RING_SIZE
);
739 memset(priv
->rx_ring
, 0,
740 IPOIB_RX_RING_SIZE
* sizeof (struct ipoib_buf
));
742 priv
->tx_ring
= kmalloc(IPOIB_TX_RING_SIZE
* sizeof (struct ipoib_buf
),
744 if (!priv
->tx_ring
) {
745 printk(KERN_WARNING
"%s: failed to allocate TX ring (%d entries)\n",
746 ca
->name
, IPOIB_TX_RING_SIZE
);
747 goto out_rx_ring_cleanup
;
749 memset(priv
->tx_ring
, 0,
750 IPOIB_TX_RING_SIZE
* sizeof (struct ipoib_buf
));
752 /* priv->tx_head & tx_tail are already 0 */
754 if (ipoib_ib_dev_init(dev
, ca
, port
))
755 goto out_tx_ring_cleanup
;
760 kfree(priv
->tx_ring
);
763 kfree(priv
->rx_ring
);
769 void ipoib_dev_cleanup(struct net_device
*dev
)
771 struct ipoib_dev_priv
*priv
= netdev_priv(dev
), *cpriv
, *tcpriv
;
773 ipoib_delete_debug_file(dev
);
775 /* Delete any child interfaces first */
776 list_for_each_entry_safe(cpriv
, tcpriv
, &priv
->child_intfs
, list
) {
777 unregister_netdev(cpriv
->dev
);
778 ipoib_dev_cleanup(cpriv
->dev
);
779 free_netdev(cpriv
->dev
);
782 ipoib_ib_dev_cleanup(dev
);
784 kfree(priv
->rx_ring
);
785 kfree(priv
->tx_ring
);
787 priv
->rx_ring
= NULL
;
788 priv
->tx_ring
= NULL
;
791 static void ipoib_setup(struct net_device
*dev
)
793 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
795 dev
->open
= ipoib_open
;
796 dev
->stop
= ipoib_stop
;
797 dev
->change_mtu
= ipoib_change_mtu
;
798 dev
->hard_start_xmit
= ipoib_start_xmit
;
799 dev
->get_stats
= ipoib_get_stats
;
800 dev
->tx_timeout
= ipoib_timeout
;
801 dev
->hard_header
= ipoib_hard_header
;
802 dev
->set_multicast_list
= ipoib_set_mcast_list
;
803 dev
->neigh_setup
= ipoib_neigh_setup_dev
;
805 dev
->watchdog_timeo
= HZ
;
807 dev
->rebuild_header
= NULL
;
808 dev
->set_mac_address
= NULL
;
809 dev
->header_cache_update
= NULL
;
811 dev
->flags
|= IFF_BROADCAST
| IFF_MULTICAST
;
814 * We add in INFINIBAND_ALEN to allow for the destination
815 * address "pseudoheader" for skbs without neighbour struct.
817 dev
->hard_header_len
= IPOIB_ENCAP_LEN
+ INFINIBAND_ALEN
;
818 dev
->addr_len
= INFINIBAND_ALEN
;
819 dev
->type
= ARPHRD_INFINIBAND
;
820 dev
->tx_queue_len
= IPOIB_TX_RING_SIZE
* 2;
821 dev
->features
= NETIF_F_VLAN_CHALLENGED
| NETIF_F_LLTX
;
823 /* MTU will be reset when mcast join happens */
824 dev
->mtu
= IPOIB_PACKET_SIZE
- IPOIB_ENCAP_LEN
;
825 priv
->mcast_mtu
= priv
->admin_mtu
= dev
->mtu
;
827 memcpy(dev
->broadcast
, ipv4_bcast_addr
, INFINIBAND_ALEN
);
829 netif_carrier_off(dev
);
831 SET_MODULE_OWNER(dev
);
835 spin_lock_init(&priv
->lock
);
836 spin_lock_init(&priv
->tx_lock
);
838 init_MUTEX(&priv
->mcast_mutex
);
839 init_MUTEX(&priv
->vlan_mutex
);
841 INIT_LIST_HEAD(&priv
->path_list
);
842 INIT_LIST_HEAD(&priv
->child_intfs
);
843 INIT_LIST_HEAD(&priv
->dead_ahs
);
844 INIT_LIST_HEAD(&priv
->multicast_list
);
846 INIT_WORK(&priv
->pkey_task
, ipoib_pkey_poll
, priv
->dev
);
847 INIT_WORK(&priv
->mcast_task
, ipoib_mcast_join_task
, priv
->dev
);
848 INIT_WORK(&priv
->flush_task
, ipoib_ib_dev_flush
, priv
->dev
);
849 INIT_WORK(&priv
->restart_task
, ipoib_mcast_restart_task
, priv
->dev
);
850 INIT_WORK(&priv
->ah_reap_task
, ipoib_reap_ah
, priv
->dev
);
853 struct ipoib_dev_priv
*ipoib_intf_alloc(const char *name
)
855 struct net_device
*dev
;
857 dev
= alloc_netdev((int) sizeof (struct ipoib_dev_priv
), name
,
862 return netdev_priv(dev
);
865 static ssize_t
show_pkey(struct class_device
*cdev
, char *buf
)
867 struct ipoib_dev_priv
*priv
=
868 netdev_priv(container_of(cdev
, struct net_device
, class_dev
));
870 return sprintf(buf
, "0x%04x\n", priv
->pkey
);
872 static CLASS_DEVICE_ATTR(pkey
, S_IRUGO
, show_pkey
, NULL
);
874 static ssize_t
create_child(struct class_device
*cdev
,
875 const char *buf
, size_t count
)
880 if (sscanf(buf
, "%i", &pkey
) != 1)
883 if (pkey
< 0 || pkey
> 0xffff)
887 * Set the full membership bit, so that we join the right
888 * broadcast group, etc.
892 ret
= ipoib_vlan_add(container_of(cdev
, struct net_device
, class_dev
),
895 return ret
? ret
: count
;
897 static CLASS_DEVICE_ATTR(create_child
, S_IWUGO
, NULL
, create_child
);
899 static ssize_t
delete_child(struct class_device
*cdev
,
900 const char *buf
, size_t count
)
905 if (sscanf(buf
, "%i", &pkey
) != 1)
908 if (pkey
< 0 || pkey
> 0xffff)
911 ret
= ipoib_vlan_delete(container_of(cdev
, struct net_device
, class_dev
),
914 return ret
? ret
: count
;
917 static CLASS_DEVICE_ATTR(delete_child
, S_IWUGO
, NULL
, delete_child
);
919 int ipoib_add_pkey_attr(struct net_device
*dev
)
921 return class_device_create_file(&dev
->class_dev
,
922 &class_device_attr_pkey
);
925 static struct net_device
*ipoib_add_port(const char *format
,
926 struct ib_device
*hca
, u8 port
)
928 struct ipoib_dev_priv
*priv
;
929 int result
= -ENOMEM
;
931 priv
= ipoib_intf_alloc(format
);
933 goto alloc_mem_failed
;
935 SET_NETDEV_DEV(priv
->dev
, hca
->dma_device
);
937 result
= ib_query_pkey(hca
, port
, 0, &priv
->pkey
);
939 printk(KERN_WARNING
"%s: ib_query_pkey port %d failed (ret = %d)\n",
940 hca
->name
, port
, result
);
941 goto alloc_mem_failed
;
945 * Set the full membership bit, so that we join the right
946 * broadcast group, etc.
948 priv
->pkey
|= 0x8000;
950 priv
->dev
->broadcast
[8] = priv
->pkey
>> 8;
951 priv
->dev
->broadcast
[9] = priv
->pkey
& 0xff;
953 result
= ib_query_gid(hca
, port
, 0, &priv
->local_gid
);
955 printk(KERN_WARNING
"%s: ib_query_gid port %d failed (ret = %d)\n",
956 hca
->name
, port
, result
);
957 goto alloc_mem_failed
;
959 memcpy(priv
->dev
->dev_addr
+ 4, priv
->local_gid
.raw
, sizeof (union ib_gid
));
962 result
= ipoib_dev_init(priv
->dev
, hca
, port
);
964 printk(KERN_WARNING
"%s: failed to initialize port %d (ret = %d)\n",
965 hca
->name
, port
, result
);
966 goto device_init_failed
;
969 INIT_IB_EVENT_HANDLER(&priv
->event_handler
,
970 priv
->ca
, ipoib_event
);
971 result
= ib_register_event_handler(&priv
->event_handler
);
973 printk(KERN_WARNING
"%s: ib_register_event_handler failed for "
974 "port %d (ret = %d)\n",
975 hca
->name
, port
, result
);
979 result
= register_netdev(priv
->dev
);
981 printk(KERN_WARNING
"%s: couldn't register ipoib port %d; error %d\n",
982 hca
->name
, port
, result
);
983 goto register_failed
;
986 if (ipoib_create_debug_file(priv
->dev
))
989 if (ipoib_add_pkey_attr(priv
->dev
))
991 if (class_device_create_file(&priv
->dev
->class_dev
,
992 &class_device_attr_create_child
))
994 if (class_device_create_file(&priv
->dev
->class_dev
,
995 &class_device_attr_delete_child
))
1001 ipoib_delete_debug_file(priv
->dev
);
1004 unregister_netdev(priv
->dev
);
1007 ib_unregister_event_handler(&priv
->event_handler
);
1010 ipoib_dev_cleanup(priv
->dev
);
1013 free_netdev(priv
->dev
);
1016 return ERR_PTR(result
);
1019 static void ipoib_add_one(struct ib_device
*device
)
1021 struct list_head
*dev_list
;
1022 struct net_device
*dev
;
1023 struct ipoib_dev_priv
*priv
;
1026 dev_list
= kmalloc(sizeof *dev_list
, GFP_KERNEL
);
1030 INIT_LIST_HEAD(dev_list
);
1032 if (device
->node_type
== IB_NODE_SWITCH
) {
1037 e
= device
->phys_port_cnt
;
1040 for (p
= s
; p
<= e
; ++p
) {
1041 dev
= ipoib_add_port("ib%d", device
, p
);
1043 priv
= netdev_priv(dev
);
1044 list_add_tail(&priv
->list
, dev_list
);
1048 ib_set_client_data(device
, &ipoib_client
, dev_list
);
1051 static void ipoib_remove_one(struct ib_device
*device
)
1053 struct ipoib_dev_priv
*priv
, *tmp
;
1054 struct list_head
*dev_list
;
1056 dev_list
= ib_get_client_data(device
, &ipoib_client
);
1058 list_for_each_entry_safe(priv
, tmp
, dev_list
, list
) {
1059 ib_unregister_event_handler(&priv
->event_handler
);
1061 unregister_netdev(priv
->dev
);
1062 ipoib_dev_cleanup(priv
->dev
);
1063 free_netdev(priv
->dev
);
1067 static int __init
ipoib_init_module(void)
1071 ret
= ipoib_register_debugfs();
1076 * We create our own workqueue mainly because we want to be
1077 * able to flush it when devices are being removed. We can't
1078 * use schedule_work()/flush_scheduled_work() because both
1079 * unregister_netdev() and linkwatch_event take the rtnl lock,
1080 * so flush_scheduled_work() can deadlock during device
1083 ipoib_workqueue
= create_singlethread_workqueue("ipoib");
1084 if (!ipoib_workqueue
) {
1089 ret
= ib_register_client(&ipoib_client
);
1096 destroy_workqueue(ipoib_workqueue
);
1099 ipoib_unregister_debugfs();
1104 static void __exit
ipoib_cleanup_module(void)
1106 ib_unregister_client(&ipoib_client
);
1107 ipoib_unregister_debugfs();
1108 destroy_workqueue(ipoib_workqueue
);
1111 module_init(ipoib_init_module
);
1112 module_exit(ipoib_cleanup_module
);