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
37 #include <linux/module.h>
39 #include <linux/init.h>
40 #include <linux/slab.h>
41 #include <linux/kernel.h>
42 #include <linux/vmalloc.h>
44 #include <linux/if_arp.h> /* For ARPHRD_xxx */
49 #include <linux/jhash.h>
51 #include <net/addrconf.h>
52 #include <linux/inetdevice.h>
53 #include <rdma/ib_cache.h>
54 #include <linux/pci.h>
56 #define DRV_VERSION "1.0.0"
58 const char ipoib_driver_version
[] = DRV_VERSION
;
60 MODULE_AUTHOR("Roland Dreier");
61 MODULE_DESCRIPTION("IP-over-InfiniBand net driver");
62 MODULE_LICENSE("Dual BSD/GPL");
63 MODULE_VERSION(DRV_VERSION
);
65 int ipoib_sendq_size __read_mostly
= IPOIB_TX_RING_SIZE
;
66 int ipoib_recvq_size __read_mostly
= IPOIB_RX_RING_SIZE
;
68 module_param_named(send_queue_size
, ipoib_sendq_size
, int, 0444);
69 MODULE_PARM_DESC(send_queue_size
, "Number of descriptors in send queue");
70 module_param_named(recv_queue_size
, ipoib_recvq_size
, int, 0444);
71 MODULE_PARM_DESC(recv_queue_size
, "Number of descriptors in receive queue");
73 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
74 int ipoib_debug_level
;
76 module_param_named(debug_level
, ipoib_debug_level
, int, 0644);
77 MODULE_PARM_DESC(debug_level
, "Enable debug tracing if > 0");
80 struct ipoib_path_iter
{
81 struct net_device
*dev
;
82 struct ipoib_path path
;
85 static const u8 ipv4_bcast_addr
[] = {
86 0x00, 0xff, 0xff, 0xff,
87 0xff, 0x12, 0x40, 0x1b, 0x00, 0x00, 0x00, 0x00,
88 0x00, 0x00, 0x00, 0x00, 0xff, 0xff, 0xff, 0xff
91 struct workqueue_struct
*ipoib_workqueue
;
93 struct ib_sa_client ipoib_sa_client
;
95 static void ipoib_add_one(struct ib_device
*device
);
96 static void ipoib_remove_one(struct ib_device
*device
, void *client_data
);
97 static void ipoib_neigh_reclaim(struct rcu_head
*rp
);
98 static struct net_device
*ipoib_get_net_dev_by_params(
99 struct ib_device
*dev
, u8 port
, u16 pkey
,
100 const union ib_gid
*gid
, const struct sockaddr
*addr
,
102 static int ipoib_set_mac(struct net_device
*dev
, void *addr
);
104 static struct ib_client ipoib_client
= {
106 .add
= ipoib_add_one
,
107 .remove
= ipoib_remove_one
,
108 .get_net_dev_by_params
= ipoib_get_net_dev_by_params
,
111 int ipoib_open(struct net_device
*dev
)
113 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
115 ipoib_dbg(priv
, "bringing up interface\n");
117 netif_carrier_off(dev
);
119 set_bit(IPOIB_FLAG_ADMIN_UP
, &priv
->flags
);
121 priv
->sm_fullmember_sendonly_support
= false;
123 if (ipoib_ib_dev_open(dev
)) {
124 if (!test_bit(IPOIB_PKEY_ASSIGNED
, &priv
->flags
))
129 if (ipoib_ib_dev_up(dev
))
132 if (!test_bit(IPOIB_FLAG_SUBINTERFACE
, &priv
->flags
)) {
133 struct ipoib_dev_priv
*cpriv
;
135 /* Bring up any child interfaces too */
136 down_read(&priv
->vlan_rwsem
);
137 list_for_each_entry(cpriv
, &priv
->child_intfs
, list
) {
140 flags
= cpriv
->dev
->flags
;
144 dev_change_flags(cpriv
->dev
, flags
| IFF_UP
);
146 up_read(&priv
->vlan_rwsem
);
149 netif_start_queue(dev
);
154 ipoib_ib_dev_stop(dev
);
157 clear_bit(IPOIB_FLAG_ADMIN_UP
, &priv
->flags
);
162 static int ipoib_stop(struct net_device
*dev
)
164 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
166 ipoib_dbg(priv
, "stopping interface\n");
168 clear_bit(IPOIB_FLAG_ADMIN_UP
, &priv
->flags
);
170 netif_stop_queue(dev
);
172 ipoib_ib_dev_down(dev
);
173 ipoib_ib_dev_stop(dev
);
175 if (!test_bit(IPOIB_FLAG_SUBINTERFACE
, &priv
->flags
)) {
176 struct ipoib_dev_priv
*cpriv
;
178 /* Bring down any child interfaces too */
179 down_read(&priv
->vlan_rwsem
);
180 list_for_each_entry(cpriv
, &priv
->child_intfs
, list
) {
183 flags
= cpriv
->dev
->flags
;
184 if (!(flags
& IFF_UP
))
187 dev_change_flags(cpriv
->dev
, flags
& ~IFF_UP
);
189 up_read(&priv
->vlan_rwsem
);
195 static void ipoib_uninit(struct net_device
*dev
)
197 ipoib_dev_cleanup(dev
);
200 static netdev_features_t
ipoib_fix_features(struct net_device
*dev
, netdev_features_t features
)
202 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
204 if (test_bit(IPOIB_FLAG_ADMIN_CM
, &priv
->flags
))
205 features
&= ~(NETIF_F_IP_CSUM
| NETIF_F_TSO
);
210 static int ipoib_change_mtu(struct net_device
*dev
, int new_mtu
)
212 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
214 /* dev->mtu > 2K ==> connected mode */
215 if (ipoib_cm_admin_enabled(dev
)) {
216 if (new_mtu
> ipoib_cm_max_mtu(dev
))
219 if (new_mtu
> priv
->mcast_mtu
)
220 ipoib_warn(priv
, "mtu > %d will cause multicast packet drops.\n",
227 if (new_mtu
> IPOIB_UD_MTU(priv
->max_ib_mtu
))
230 priv
->admin_mtu
= new_mtu
;
232 dev
->mtu
= min(priv
->mcast_mtu
, priv
->admin_mtu
);
237 /* Called with an RCU read lock taken */
238 static bool ipoib_is_dev_match_addr_rcu(const struct sockaddr
*addr
,
239 struct net_device
*dev
)
241 struct net
*net
= dev_net(dev
);
242 struct in_device
*in_dev
;
243 struct sockaddr_in
*addr_in
= (struct sockaddr_in
*)addr
;
244 struct sockaddr_in6
*addr_in6
= (struct sockaddr_in6
*)addr
;
247 switch (addr
->sa_family
) {
249 in_dev
= in_dev_get(dev
);
253 ret_addr
= inet_confirm_addr(net
, in_dev
, 0,
254 addr_in
->sin_addr
.s_addr
,
262 if (IS_ENABLED(CONFIG_IPV6
) &&
263 ipv6_chk_addr(net
, &addr_in6
->sin6_addr
, dev
, 1))
272 * Find the master net_device on top of the given net_device.
273 * @dev: base IPoIB net_device
275 * Returns the master net_device with a reference held, or the same net_device
276 * if no master exists.
278 static struct net_device
*ipoib_get_master_net_dev(struct net_device
*dev
)
280 struct net_device
*master
;
283 master
= netdev_master_upper_dev_get_rcu(dev
);
296 * Find a net_device matching the given address, which is an upper device of
297 * the given net_device.
298 * @addr: IP address to look for.
299 * @dev: base IPoIB net_device
301 * If found, returns the net_device with a reference held. Otherwise return
304 static struct net_device
*ipoib_get_net_dev_match_addr(
305 const struct sockaddr
*addr
, struct net_device
*dev
)
307 struct net_device
*upper
,
309 struct list_head
*iter
;
312 if (ipoib_is_dev_match_addr_rcu(addr
, dev
)) {
318 netdev_for_each_all_upper_dev_rcu(dev
, upper
, iter
) {
319 if (ipoib_is_dev_match_addr_rcu(addr
, upper
)) {
330 /* returns the number of IPoIB netdevs on top a given ipoib device matching a
331 * pkey_index and address, if one exists.
333 * @found_net_dev: contains a matching net_device if the return value >= 1,
334 * with a reference held. */
335 static int ipoib_match_gid_pkey_addr(struct ipoib_dev_priv
*priv
,
336 const union ib_gid
*gid
,
338 const struct sockaddr
*addr
,
340 struct net_device
**found_net_dev
)
342 struct ipoib_dev_priv
*child_priv
;
343 struct net_device
*net_dev
= NULL
;
346 if (priv
->pkey_index
== pkey_index
&&
347 (!gid
|| !memcmp(gid
, &priv
->local_gid
, sizeof(*gid
)))) {
349 net_dev
= ipoib_get_master_net_dev(priv
->dev
);
351 /* Verify the net_device matches the IP address, as
352 * IPoIB child devices currently share a GID. */
353 net_dev
= ipoib_get_net_dev_match_addr(addr
, priv
->dev
);
357 *found_net_dev
= net_dev
;
364 /* Check child interfaces */
365 down_read_nested(&priv
->vlan_rwsem
, nesting
);
366 list_for_each_entry(child_priv
, &priv
->child_intfs
, list
) {
367 matches
+= ipoib_match_gid_pkey_addr(child_priv
, gid
,
374 up_read(&priv
->vlan_rwsem
);
379 /* Returns the number of matching net_devs found (between 0 and 2). Also
380 * return the matching net_device in the @net_dev parameter, holding a
381 * reference to the net_device, if the number of matches >= 1 */
382 static int __ipoib_get_net_dev_by_params(struct list_head
*dev_list
, u8 port
,
384 const union ib_gid
*gid
,
385 const struct sockaddr
*addr
,
386 struct net_device
**net_dev
)
388 struct ipoib_dev_priv
*priv
;
393 list_for_each_entry(priv
, dev_list
, list
) {
394 if (priv
->port
!= port
)
397 matches
+= ipoib_match_gid_pkey_addr(priv
, gid
, pkey_index
,
406 static struct net_device
*ipoib_get_net_dev_by_params(
407 struct ib_device
*dev
, u8 port
, u16 pkey
,
408 const union ib_gid
*gid
, const struct sockaddr
*addr
,
411 struct net_device
*net_dev
;
412 struct list_head
*dev_list
= client_data
;
417 if (!rdma_protocol_ib(dev
, port
))
420 ret
= ib_find_cached_pkey(dev
, port
, pkey
, &pkey_index
);
427 /* See if we can find a unique device matching the L2 parameters */
428 matches
= __ipoib_get_net_dev_by_params(dev_list
, port
, pkey_index
,
429 gid
, NULL
, &net_dev
);
440 /* Couldn't find a unique device with L2 parameters only. Use L3
441 * address to uniquely match the net device */
442 matches
= __ipoib_get_net_dev_by_params(dev_list
, port
, pkey_index
,
443 gid
, addr
, &net_dev
);
448 dev_warn_ratelimited(&dev
->dev
,
449 "duplicate IP address detected\n");
456 int ipoib_set_mode(struct net_device
*dev
, const char *buf
)
458 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
460 /* flush paths if we switch modes so that connections are restarted */
461 if (IPOIB_CM_SUPPORTED(dev
->dev_addr
) && !strcmp(buf
, "connected\n")) {
462 set_bit(IPOIB_FLAG_ADMIN_CM
, &priv
->flags
);
463 ipoib_warn(priv
, "enabling connected mode "
464 "will cause multicast packet drops\n");
465 netdev_update_features(dev
);
466 dev_set_mtu(dev
, ipoib_cm_max_mtu(dev
));
468 priv
->tx_wr
.wr
.send_flags
&= ~IB_SEND_IP_CSUM
;
470 ipoib_flush_paths(dev
);
475 if (!strcmp(buf
, "datagram\n")) {
476 clear_bit(IPOIB_FLAG_ADMIN_CM
, &priv
->flags
);
477 netdev_update_features(dev
);
478 dev_set_mtu(dev
, min(priv
->mcast_mtu
, dev
->mtu
));
480 ipoib_flush_paths(dev
);
488 struct ipoib_path
*__path_find(struct net_device
*dev
, void *gid
)
490 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
491 struct rb_node
*n
= priv
->path_tree
.rb_node
;
492 struct ipoib_path
*path
;
496 path
= rb_entry(n
, struct ipoib_path
, rb_node
);
498 ret
= memcmp(gid
, path
->pathrec
.dgid
.raw
,
499 sizeof (union ib_gid
));
512 static int __path_add(struct net_device
*dev
, struct ipoib_path
*path
)
514 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
515 struct rb_node
**n
= &priv
->path_tree
.rb_node
;
516 struct rb_node
*pn
= NULL
;
517 struct ipoib_path
*tpath
;
522 tpath
= rb_entry(pn
, struct ipoib_path
, rb_node
);
524 ret
= memcmp(path
->pathrec
.dgid
.raw
, tpath
->pathrec
.dgid
.raw
,
525 sizeof (union ib_gid
));
534 rb_link_node(&path
->rb_node
, pn
, n
);
535 rb_insert_color(&path
->rb_node
, &priv
->path_tree
);
537 list_add_tail(&path
->list
, &priv
->path_list
);
542 static void path_free(struct net_device
*dev
, struct ipoib_path
*path
)
546 while ((skb
= __skb_dequeue(&path
->queue
)))
547 dev_kfree_skb_irq(skb
);
549 ipoib_dbg(netdev_priv(dev
), "path_free\n");
551 /* remove all neigh connected to this path */
552 ipoib_del_neighs_by_gid(dev
, path
->pathrec
.dgid
.raw
);
555 ipoib_put_ah(path
->ah
);
560 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
562 struct ipoib_path_iter
*ipoib_path_iter_init(struct net_device
*dev
)
564 struct ipoib_path_iter
*iter
;
566 iter
= kmalloc(sizeof *iter
, GFP_KERNEL
);
571 memset(iter
->path
.pathrec
.dgid
.raw
, 0, 16);
573 if (ipoib_path_iter_next(iter
)) {
581 int ipoib_path_iter_next(struct ipoib_path_iter
*iter
)
583 struct ipoib_dev_priv
*priv
= netdev_priv(iter
->dev
);
585 struct ipoib_path
*path
;
588 spin_lock_irq(&priv
->lock
);
590 n
= rb_first(&priv
->path_tree
);
593 path
= rb_entry(n
, struct ipoib_path
, rb_node
);
595 if (memcmp(iter
->path
.pathrec
.dgid
.raw
, path
->pathrec
.dgid
.raw
,
596 sizeof (union ib_gid
)) < 0) {
605 spin_unlock_irq(&priv
->lock
);
610 void ipoib_path_iter_read(struct ipoib_path_iter
*iter
,
611 struct ipoib_path
*path
)
616 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */
618 void ipoib_mark_paths_invalid(struct net_device
*dev
)
620 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
621 struct ipoib_path
*path
, *tp
;
623 spin_lock_irq(&priv
->lock
);
625 list_for_each_entry_safe(path
, tp
, &priv
->path_list
, list
) {
626 ipoib_dbg(priv
, "mark path LID 0x%04x GID %pI6 invalid\n",
627 be16_to_cpu(path
->pathrec
.dlid
),
628 path
->pathrec
.dgid
.raw
);
632 spin_unlock_irq(&priv
->lock
);
635 struct classport_info_context
{
636 struct ipoib_dev_priv
*priv
;
637 struct completion done
;
638 struct ib_sa_query
*sa_query
;
641 static void classport_info_query_cb(int status
, struct ib_class_port_info
*rec
,
644 struct classport_info_context
*cb_ctx
= context
;
645 struct ipoib_dev_priv
*priv
;
651 if (status
|| !rec
) {
652 pr_debug("device: %s failed query classport_info status: %d\n",
653 priv
->dev
->name
, status
);
654 /* keeps the default, will try next mcast_restart */
655 priv
->sm_fullmember_sendonly_support
= false;
659 if (ib_get_cpi_capmask2(rec
) &
660 IB_SA_CAP_MASK2_SENDONLY_FULL_MEM_SUPPORT
) {
661 pr_debug("device: %s enabled fullmember-sendonly for sendonly MCG\n",
663 priv
->sm_fullmember_sendonly_support
= true;
665 pr_debug("device: %s disabled fullmember-sendonly for sendonly MCG\n",
667 priv
->sm_fullmember_sendonly_support
= false;
671 complete(&cb_ctx
->done
);
674 int ipoib_check_sm_sendonly_fullmember_support(struct ipoib_dev_priv
*priv
)
676 struct classport_info_context
*callback_context
;
679 callback_context
= kmalloc(sizeof(*callback_context
), GFP_KERNEL
);
680 if (!callback_context
)
683 callback_context
->priv
= priv
;
684 init_completion(&callback_context
->done
);
686 ret
= ib_sa_classport_info_rec_query(&ipoib_sa_client
,
687 priv
->ca
, priv
->port
, 3000,
689 classport_info_query_cb
,
691 &callback_context
->sa_query
);
693 pr_info("%s failed to send ib_sa_classport_info query, ret: %d\n",
694 priv
->dev
->name
, ret
);
695 kfree(callback_context
);
699 /* waiting for the callback to finish before returnning */
700 wait_for_completion(&callback_context
->done
);
701 kfree(callback_context
);
706 void ipoib_flush_paths(struct net_device
*dev
)
708 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
709 struct ipoib_path
*path
, *tp
;
710 LIST_HEAD(remove_list
);
713 netif_tx_lock_bh(dev
);
714 spin_lock_irqsave(&priv
->lock
, flags
);
716 list_splice_init(&priv
->path_list
, &remove_list
);
718 list_for_each_entry(path
, &remove_list
, list
)
719 rb_erase(&path
->rb_node
, &priv
->path_tree
);
721 list_for_each_entry_safe(path
, tp
, &remove_list
, list
) {
723 ib_sa_cancel_query(path
->query_id
, path
->query
);
724 spin_unlock_irqrestore(&priv
->lock
, flags
);
725 netif_tx_unlock_bh(dev
);
726 wait_for_completion(&path
->done
);
727 path_free(dev
, path
);
728 netif_tx_lock_bh(dev
);
729 spin_lock_irqsave(&priv
->lock
, flags
);
732 spin_unlock_irqrestore(&priv
->lock
, flags
);
733 netif_tx_unlock_bh(dev
);
736 static void path_rec_completion(int status
,
737 struct ib_sa_path_rec
*pathrec
,
740 struct ipoib_path
*path
= path_ptr
;
741 struct net_device
*dev
= path
->dev
;
742 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
743 struct ipoib_ah
*ah
= NULL
;
744 struct ipoib_ah
*old_ah
= NULL
;
745 struct ipoib_neigh
*neigh
, *tn
;
746 struct sk_buff_head skqueue
;
751 ipoib_dbg(priv
, "PathRec LID 0x%04x for GID %pI6\n",
752 be16_to_cpu(pathrec
->dlid
), pathrec
->dgid
.raw
);
754 ipoib_dbg(priv
, "PathRec status %d for GID %pI6\n",
755 status
, path
->pathrec
.dgid
.raw
);
757 skb_queue_head_init(&skqueue
);
760 struct ib_ah_attr av
;
762 if (!ib_init_ah_from_path(priv
->ca
, priv
->port
, pathrec
, &av
))
763 ah
= ipoib_create_ah(dev
, priv
->pd
, &av
);
766 spin_lock_irqsave(&priv
->lock
, flags
);
768 if (!IS_ERR_OR_NULL(ah
)) {
769 path
->pathrec
= *pathrec
;
774 ipoib_dbg(priv
, "created address handle %p for LID 0x%04x, SL %d\n",
775 ah
, be16_to_cpu(pathrec
->dlid
), pathrec
->sl
);
777 while ((skb
= __skb_dequeue(&path
->queue
)))
778 __skb_queue_tail(&skqueue
, skb
);
780 list_for_each_entry_safe(neigh
, tn
, &path
->neigh_list
, list
) {
782 WARN_ON(neigh
->ah
!= old_ah
);
784 * Dropping the ah reference inside
785 * priv->lock is safe here, because we
786 * will hold one more reference from
787 * the original value of path->ah (ie
790 ipoib_put_ah(neigh
->ah
);
792 kref_get(&path
->ah
->ref
);
793 neigh
->ah
= path
->ah
;
795 if (ipoib_cm_enabled(dev
, neigh
->daddr
)) {
796 if (!ipoib_cm_get(neigh
))
797 ipoib_cm_set(neigh
, ipoib_cm_create_tx(dev
,
800 if (!ipoib_cm_get(neigh
)) {
801 ipoib_neigh_free(neigh
);
806 while ((skb
= __skb_dequeue(&neigh
->queue
)))
807 __skb_queue_tail(&skqueue
, skb
);
813 complete(&path
->done
);
815 spin_unlock_irqrestore(&priv
->lock
, flags
);
817 if (IS_ERR_OR_NULL(ah
))
818 ipoib_del_neighs_by_gid(dev
, path
->pathrec
.dgid
.raw
);
821 ipoib_put_ah(old_ah
);
823 while ((skb
= __skb_dequeue(&skqueue
))) {
825 if (dev_queue_xmit(skb
))
826 ipoib_warn(priv
, "dev_queue_xmit failed "
827 "to requeue packet\n");
831 static struct ipoib_path
*path_rec_create(struct net_device
*dev
, void *gid
)
833 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
834 struct ipoib_path
*path
;
836 if (!priv
->broadcast
)
839 path
= kzalloc(sizeof *path
, GFP_ATOMIC
);
845 skb_queue_head_init(&path
->queue
);
847 INIT_LIST_HEAD(&path
->neigh_list
);
849 memcpy(path
->pathrec
.dgid
.raw
, gid
, sizeof (union ib_gid
));
850 path
->pathrec
.sgid
= priv
->local_gid
;
851 path
->pathrec
.pkey
= cpu_to_be16(priv
->pkey
);
852 path
->pathrec
.numb_path
= 1;
853 path
->pathrec
.traffic_class
= priv
->broadcast
->mcmember
.traffic_class
;
858 static int path_rec_start(struct net_device
*dev
,
859 struct ipoib_path
*path
)
861 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
863 ipoib_dbg(priv
, "Start path record lookup for %pI6\n",
864 path
->pathrec
.dgid
.raw
);
866 init_completion(&path
->done
);
869 ib_sa_path_rec_get(&ipoib_sa_client
, priv
->ca
, priv
->port
,
871 IB_SA_PATH_REC_DGID
|
872 IB_SA_PATH_REC_SGID
|
873 IB_SA_PATH_REC_NUMB_PATH
|
874 IB_SA_PATH_REC_TRAFFIC_CLASS
|
879 if (path
->query_id
< 0) {
880 ipoib_warn(priv
, "ib_sa_path_rec_get failed: %d\n", path
->query_id
);
882 complete(&path
->done
);
883 return path
->query_id
;
889 static void neigh_add_path(struct sk_buff
*skb
, u8
*daddr
,
890 struct net_device
*dev
)
892 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
893 struct ipoib_path
*path
;
894 struct ipoib_neigh
*neigh
;
897 spin_lock_irqsave(&priv
->lock
, flags
);
898 neigh
= ipoib_neigh_alloc(daddr
, dev
);
900 spin_unlock_irqrestore(&priv
->lock
, flags
);
901 ++dev
->stats
.tx_dropped
;
902 dev_kfree_skb_any(skb
);
906 path
= __path_find(dev
, daddr
+ 4);
908 path
= path_rec_create(dev
, daddr
+ 4);
912 __path_add(dev
, path
);
915 list_add_tail(&neigh
->list
, &path
->neigh_list
);
918 kref_get(&path
->ah
->ref
);
919 neigh
->ah
= path
->ah
;
921 if (ipoib_cm_enabled(dev
, neigh
->daddr
)) {
922 if (!ipoib_cm_get(neigh
))
923 ipoib_cm_set(neigh
, ipoib_cm_create_tx(dev
, path
, neigh
));
924 if (!ipoib_cm_get(neigh
)) {
925 ipoib_neigh_free(neigh
);
928 if (skb_queue_len(&neigh
->queue
) < IPOIB_MAX_PATH_REC_QUEUE
)
929 __skb_queue_tail(&neigh
->queue
, skb
);
931 ipoib_warn(priv
, "queue length limit %d. Packet drop.\n",
932 skb_queue_len(&neigh
->queue
));
936 spin_unlock_irqrestore(&priv
->lock
, flags
);
937 ipoib_send(dev
, skb
, path
->ah
, IPOIB_QPN(daddr
));
938 ipoib_neigh_put(neigh
);
944 if (!path
->query
&& path_rec_start(dev
, path
))
946 if (skb_queue_len(&neigh
->queue
) < IPOIB_MAX_PATH_REC_QUEUE
)
947 __skb_queue_tail(&neigh
->queue
, skb
);
952 spin_unlock_irqrestore(&priv
->lock
, flags
);
953 ipoib_neigh_put(neigh
);
957 ipoib_neigh_free(neigh
);
959 ++dev
->stats
.tx_dropped
;
960 dev_kfree_skb_any(skb
);
962 spin_unlock_irqrestore(&priv
->lock
, flags
);
963 ipoib_neigh_put(neigh
);
966 static void unicast_arp_send(struct sk_buff
*skb
, struct net_device
*dev
,
969 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
970 struct ipoib_path
*path
;
973 spin_lock_irqsave(&priv
->lock
, flags
);
975 path
= __path_find(dev
, cb
->hwaddr
+ 4);
976 if (!path
|| !path
->valid
) {
980 path
= path_rec_create(dev
, cb
->hwaddr
+ 4);
984 if (skb_queue_len(&path
->queue
) < IPOIB_MAX_PATH_REC_QUEUE
) {
985 __skb_queue_tail(&path
->queue
, skb
);
987 ++dev
->stats
.tx_dropped
;
988 dev_kfree_skb_any(skb
);
991 if (!path
->query
&& path_rec_start(dev
, path
)) {
992 spin_unlock_irqrestore(&priv
->lock
, flags
);
994 path_free(dev
, path
);
997 __path_add(dev
, path
);
999 ++dev
->stats
.tx_dropped
;
1000 dev_kfree_skb_any(skb
);
1003 spin_unlock_irqrestore(&priv
->lock
, flags
);
1008 ipoib_dbg(priv
, "Send unicast ARP to %04x\n",
1009 be16_to_cpu(path
->pathrec
.dlid
));
1011 spin_unlock_irqrestore(&priv
->lock
, flags
);
1012 ipoib_send(dev
, skb
, path
->ah
, IPOIB_QPN(cb
->hwaddr
));
1014 } else if ((path
->query
|| !path_rec_start(dev
, path
)) &&
1015 skb_queue_len(&path
->queue
) < IPOIB_MAX_PATH_REC_QUEUE
) {
1016 __skb_queue_tail(&path
->queue
, skb
);
1018 ++dev
->stats
.tx_dropped
;
1019 dev_kfree_skb_any(skb
);
1022 spin_unlock_irqrestore(&priv
->lock
, flags
);
1025 static int ipoib_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
1027 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
1028 struct ipoib_neigh
*neigh
;
1029 struct ipoib_cb
*cb
= ipoib_skb_cb(skb
);
1030 struct ipoib_header
*header
;
1031 unsigned long flags
;
1033 header
= (struct ipoib_header
*) skb
->data
;
1035 if (unlikely(cb
->hwaddr
[4] == 0xff)) {
1036 /* multicast, arrange "if" according to probability */
1037 if ((header
->proto
!= htons(ETH_P_IP
)) &&
1038 (header
->proto
!= htons(ETH_P_IPV6
)) &&
1039 (header
->proto
!= htons(ETH_P_ARP
)) &&
1040 (header
->proto
!= htons(ETH_P_RARP
)) &&
1041 (header
->proto
!= htons(ETH_P_TIPC
))) {
1042 /* ethertype not supported by IPoIB */
1043 ++dev
->stats
.tx_dropped
;
1044 dev_kfree_skb_any(skb
);
1045 return NETDEV_TX_OK
;
1047 /* Add in the P_Key for multicast*/
1048 cb
->hwaddr
[8] = (priv
->pkey
>> 8) & 0xff;
1049 cb
->hwaddr
[9] = priv
->pkey
& 0xff;
1051 neigh
= ipoib_neigh_get(dev
, cb
->hwaddr
);
1053 goto send_using_neigh
;
1054 ipoib_mcast_send(dev
, cb
->hwaddr
, skb
);
1055 return NETDEV_TX_OK
;
1058 /* unicast, arrange "switch" according to probability */
1059 switch (header
->proto
) {
1060 case htons(ETH_P_IP
):
1061 case htons(ETH_P_IPV6
):
1062 case htons(ETH_P_TIPC
):
1063 neigh
= ipoib_neigh_get(dev
, cb
->hwaddr
);
1064 if (unlikely(!neigh
)) {
1065 neigh_add_path(skb
, cb
->hwaddr
, dev
);
1066 return NETDEV_TX_OK
;
1069 case htons(ETH_P_ARP
):
1070 case htons(ETH_P_RARP
):
1071 /* for unicast ARP and RARP should always perform path find */
1072 unicast_arp_send(skb
, dev
, cb
);
1073 return NETDEV_TX_OK
;
1075 /* ethertype not supported by IPoIB */
1076 ++dev
->stats
.tx_dropped
;
1077 dev_kfree_skb_any(skb
);
1078 return NETDEV_TX_OK
;
1082 /* note we now hold a ref to neigh */
1083 if (ipoib_cm_get(neigh
)) {
1084 if (ipoib_cm_up(neigh
)) {
1085 ipoib_cm_send(dev
, skb
, ipoib_cm_get(neigh
));
1088 } else if (neigh
->ah
) {
1089 ipoib_send(dev
, skb
, neigh
->ah
, IPOIB_QPN(cb
->hwaddr
));
1093 if (skb_queue_len(&neigh
->queue
) < IPOIB_MAX_PATH_REC_QUEUE
) {
1094 spin_lock_irqsave(&priv
->lock
, flags
);
1095 __skb_queue_tail(&neigh
->queue
, skb
);
1096 spin_unlock_irqrestore(&priv
->lock
, flags
);
1098 ++dev
->stats
.tx_dropped
;
1099 dev_kfree_skb_any(skb
);
1103 ipoib_neigh_put(neigh
);
1105 return NETDEV_TX_OK
;
1108 static void ipoib_timeout(struct net_device
*dev
)
1110 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
1112 ipoib_warn(priv
, "transmit timeout: latency %d msecs\n",
1113 jiffies_to_msecs(jiffies
- dev_trans_start(dev
)));
1114 ipoib_warn(priv
, "queue stopped %d, tx_head %u, tx_tail %u\n",
1115 netif_queue_stopped(dev
),
1116 priv
->tx_head
, priv
->tx_tail
);
1117 /* XXX reset QP, etc. */
1120 static int ipoib_hard_header(struct sk_buff
*skb
,
1121 struct net_device
*dev
,
1122 unsigned short type
,
1123 const void *daddr
, const void *saddr
, unsigned len
)
1125 struct ipoib_header
*header
;
1126 struct ipoib_cb
*cb
= ipoib_skb_cb(skb
);
1128 header
= (struct ipoib_header
*) skb_push(skb
, sizeof *header
);
1130 header
->proto
= htons(type
);
1131 header
->reserved
= 0;
1134 * we don't rely on dst_entry structure, always stuff the
1135 * destination address into skb->cb so we can figure out where
1136 * to send the packet later.
1138 memcpy(cb
->hwaddr
, daddr
, INFINIBAND_ALEN
);
1140 return sizeof *header
;
1143 static void ipoib_set_mcast_list(struct net_device
*dev
)
1145 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
1147 if (!test_bit(IPOIB_FLAG_OPER_UP
, &priv
->flags
)) {
1148 ipoib_dbg(priv
, "IPOIB_FLAG_OPER_UP not set");
1152 queue_work(priv
->wq
, &priv
->restart_task
);
1155 static int ipoib_get_iflink(const struct net_device
*dev
)
1157 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
1159 /* parent interface */
1160 if (!test_bit(IPOIB_FLAG_SUBINTERFACE
, &priv
->flags
))
1161 return dev
->ifindex
;
1163 /* child/vlan interface */
1164 return priv
->parent
->ifindex
;
1167 static u32
ipoib_addr_hash(struct ipoib_neigh_hash
*htbl
, u8
*daddr
)
1170 * Use only the address parts that contributes to spreading
1171 * The subnet prefix is not used as one can not connect to
1172 * same remote port (GUID) using the same remote QPN via two
1173 * different subnets.
1175 /* qpn octets[1:4) & port GUID octets[12:20) */
1176 u32
*d32
= (u32
*) daddr
;
1179 hv
= jhash_3words(d32
[3], d32
[4], IPOIB_QPN_MASK
& d32
[0], 0);
1180 return hv
& htbl
->mask
;
1183 struct ipoib_neigh
*ipoib_neigh_get(struct net_device
*dev
, u8
*daddr
)
1185 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
1186 struct ipoib_neigh_table
*ntbl
= &priv
->ntbl
;
1187 struct ipoib_neigh_hash
*htbl
;
1188 struct ipoib_neigh
*neigh
= NULL
;
1193 htbl
= rcu_dereference_bh(ntbl
->htbl
);
1198 hash_val
= ipoib_addr_hash(htbl
, daddr
);
1199 for (neigh
= rcu_dereference_bh(htbl
->buckets
[hash_val
]);
1201 neigh
= rcu_dereference_bh(neigh
->hnext
)) {
1202 if (memcmp(daddr
, neigh
->daddr
, INFINIBAND_ALEN
) == 0) {
1203 /* found, take one ref on behalf of the caller */
1204 if (!atomic_inc_not_zero(&neigh
->refcnt
)) {
1210 if (likely(skb_queue_len(&neigh
->queue
) < IPOIB_MAX_PATH_REC_QUEUE
))
1211 neigh
->alive
= jiffies
;
1217 rcu_read_unlock_bh();
1221 static void __ipoib_reap_neigh(struct ipoib_dev_priv
*priv
)
1223 struct ipoib_neigh_table
*ntbl
= &priv
->ntbl
;
1224 struct ipoib_neigh_hash
*htbl
;
1225 unsigned long neigh_obsolete
;
1227 unsigned long flags
;
1229 LIST_HEAD(remove_list
);
1231 if (test_bit(IPOIB_STOP_NEIGH_GC
, &priv
->flags
))
1234 spin_lock_irqsave(&priv
->lock
, flags
);
1236 htbl
= rcu_dereference_protected(ntbl
->htbl
,
1237 lockdep_is_held(&priv
->lock
));
1242 /* neigh is obsolete if it was idle for two GC periods */
1243 dt
= 2 * arp_tbl
.gc_interval
;
1244 neigh_obsolete
= jiffies
- dt
;
1245 /* handle possible race condition */
1246 if (test_bit(IPOIB_STOP_NEIGH_GC
, &priv
->flags
))
1249 for (i
= 0; i
< htbl
->size
; i
++) {
1250 struct ipoib_neigh
*neigh
;
1251 struct ipoib_neigh __rcu
**np
= &htbl
->buckets
[i
];
1253 while ((neigh
= rcu_dereference_protected(*np
,
1254 lockdep_is_held(&priv
->lock
))) != NULL
) {
1255 /* was the neigh idle for two GC periods */
1256 if (time_after(neigh_obsolete
, neigh
->alive
)) {
1258 ipoib_check_and_add_mcast_sendonly(priv
, neigh
->daddr
+ 4, &remove_list
);
1260 rcu_assign_pointer(*np
,
1261 rcu_dereference_protected(neigh
->hnext
,
1262 lockdep_is_held(&priv
->lock
)));
1263 /* remove from path/mc list */
1264 list_del(&neigh
->list
);
1265 call_rcu(&neigh
->rcu
, ipoib_neigh_reclaim
);
1274 spin_unlock_irqrestore(&priv
->lock
, flags
);
1275 ipoib_mcast_remove_list(&remove_list
);
1278 static void ipoib_reap_neigh(struct work_struct
*work
)
1280 struct ipoib_dev_priv
*priv
=
1281 container_of(work
, struct ipoib_dev_priv
, neigh_reap_task
.work
);
1283 __ipoib_reap_neigh(priv
);
1285 if (!test_bit(IPOIB_STOP_NEIGH_GC
, &priv
->flags
))
1286 queue_delayed_work(priv
->wq
, &priv
->neigh_reap_task
,
1287 arp_tbl
.gc_interval
);
1291 static struct ipoib_neigh
*ipoib_neigh_ctor(u8
*daddr
,
1292 struct net_device
*dev
)
1294 struct ipoib_neigh
*neigh
;
1296 neigh
= kzalloc(sizeof *neigh
, GFP_ATOMIC
);
1301 memcpy(&neigh
->daddr
, daddr
, sizeof(neigh
->daddr
));
1302 skb_queue_head_init(&neigh
->queue
);
1303 INIT_LIST_HEAD(&neigh
->list
);
1304 ipoib_cm_set(neigh
, NULL
);
1305 /* one ref on behalf of the caller */
1306 atomic_set(&neigh
->refcnt
, 1);
1311 struct ipoib_neigh
*ipoib_neigh_alloc(u8
*daddr
,
1312 struct net_device
*dev
)
1314 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
1315 struct ipoib_neigh_table
*ntbl
= &priv
->ntbl
;
1316 struct ipoib_neigh_hash
*htbl
;
1317 struct ipoib_neigh
*neigh
;
1320 htbl
= rcu_dereference_protected(ntbl
->htbl
,
1321 lockdep_is_held(&priv
->lock
));
1327 /* need to add a new neigh, but maybe some other thread succeeded?
1328 * recalc hash, maybe hash resize took place so we do a search
1330 hash_val
= ipoib_addr_hash(htbl
, daddr
);
1331 for (neigh
= rcu_dereference_protected(htbl
->buckets
[hash_val
],
1332 lockdep_is_held(&priv
->lock
));
1334 neigh
= rcu_dereference_protected(neigh
->hnext
,
1335 lockdep_is_held(&priv
->lock
))) {
1336 if (memcmp(daddr
, neigh
->daddr
, INFINIBAND_ALEN
) == 0) {
1337 /* found, take one ref on behalf of the caller */
1338 if (!atomic_inc_not_zero(&neigh
->refcnt
)) {
1343 neigh
->alive
= jiffies
;
1348 neigh
= ipoib_neigh_ctor(daddr
, dev
);
1352 /* one ref on behalf of the hash table */
1353 atomic_inc(&neigh
->refcnt
);
1354 neigh
->alive
= jiffies
;
1356 rcu_assign_pointer(neigh
->hnext
,
1357 rcu_dereference_protected(htbl
->buckets
[hash_val
],
1358 lockdep_is_held(&priv
->lock
)));
1359 rcu_assign_pointer(htbl
->buckets
[hash_val
], neigh
);
1360 atomic_inc(&ntbl
->entries
);
1367 void ipoib_neigh_dtor(struct ipoib_neigh
*neigh
)
1369 /* neigh reference count was dropprd to zero */
1370 struct net_device
*dev
= neigh
->dev
;
1371 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
1372 struct sk_buff
*skb
;
1374 ipoib_put_ah(neigh
->ah
);
1375 while ((skb
= __skb_dequeue(&neigh
->queue
))) {
1376 ++dev
->stats
.tx_dropped
;
1377 dev_kfree_skb_any(skb
);
1379 if (ipoib_cm_get(neigh
))
1380 ipoib_cm_destroy_tx(ipoib_cm_get(neigh
));
1381 ipoib_dbg(netdev_priv(dev
),
1382 "neigh free for %06x %pI6\n",
1383 IPOIB_QPN(neigh
->daddr
),
1386 if (atomic_dec_and_test(&priv
->ntbl
.entries
)) {
1387 if (test_bit(IPOIB_NEIGH_TBL_FLUSH
, &priv
->flags
))
1388 complete(&priv
->ntbl
.flushed
);
1392 static void ipoib_neigh_reclaim(struct rcu_head
*rp
)
1394 /* Called as a result of removal from hash table */
1395 struct ipoib_neigh
*neigh
= container_of(rp
, struct ipoib_neigh
, rcu
);
1396 /* note TX context may hold another ref */
1397 ipoib_neigh_put(neigh
);
1400 void ipoib_neigh_free(struct ipoib_neigh
*neigh
)
1402 struct net_device
*dev
= neigh
->dev
;
1403 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
1404 struct ipoib_neigh_table
*ntbl
= &priv
->ntbl
;
1405 struct ipoib_neigh_hash
*htbl
;
1406 struct ipoib_neigh __rcu
**np
;
1407 struct ipoib_neigh
*n
;
1410 htbl
= rcu_dereference_protected(ntbl
->htbl
,
1411 lockdep_is_held(&priv
->lock
));
1415 hash_val
= ipoib_addr_hash(htbl
, neigh
->daddr
);
1416 np
= &htbl
->buckets
[hash_val
];
1417 for (n
= rcu_dereference_protected(*np
,
1418 lockdep_is_held(&priv
->lock
));
1420 n
= rcu_dereference_protected(*np
,
1421 lockdep_is_held(&priv
->lock
))) {
1424 rcu_assign_pointer(*np
,
1425 rcu_dereference_protected(neigh
->hnext
,
1426 lockdep_is_held(&priv
->lock
)));
1427 /* remove from parent list */
1428 list_del(&neigh
->list
);
1429 call_rcu(&neigh
->rcu
, ipoib_neigh_reclaim
);
1437 static int ipoib_neigh_hash_init(struct ipoib_dev_priv
*priv
)
1439 struct ipoib_neigh_table
*ntbl
= &priv
->ntbl
;
1440 struct ipoib_neigh_hash
*htbl
;
1441 struct ipoib_neigh __rcu
**buckets
;
1444 clear_bit(IPOIB_NEIGH_TBL_FLUSH
, &priv
->flags
);
1446 htbl
= kzalloc(sizeof(*htbl
), GFP_KERNEL
);
1449 set_bit(IPOIB_STOP_NEIGH_GC
, &priv
->flags
);
1450 size
= roundup_pow_of_two(arp_tbl
.gc_thresh3
);
1451 buckets
= kzalloc(size
* sizeof(*buckets
), GFP_KERNEL
);
1457 htbl
->mask
= (size
- 1);
1458 htbl
->buckets
= buckets
;
1459 RCU_INIT_POINTER(ntbl
->htbl
, htbl
);
1461 atomic_set(&ntbl
->entries
, 0);
1463 /* start garbage collection */
1464 clear_bit(IPOIB_STOP_NEIGH_GC
, &priv
->flags
);
1465 queue_delayed_work(priv
->wq
, &priv
->neigh_reap_task
,
1466 arp_tbl
.gc_interval
);
1471 static void neigh_hash_free_rcu(struct rcu_head
*head
)
1473 struct ipoib_neigh_hash
*htbl
= container_of(head
,
1474 struct ipoib_neigh_hash
,
1476 struct ipoib_neigh __rcu
**buckets
= htbl
->buckets
;
1477 struct ipoib_neigh_table
*ntbl
= htbl
->ntbl
;
1481 complete(&ntbl
->deleted
);
1484 void ipoib_del_neighs_by_gid(struct net_device
*dev
, u8
*gid
)
1486 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
1487 struct ipoib_neigh_table
*ntbl
= &priv
->ntbl
;
1488 struct ipoib_neigh_hash
*htbl
;
1489 unsigned long flags
;
1492 /* remove all neigh connected to a given path or mcast */
1493 spin_lock_irqsave(&priv
->lock
, flags
);
1495 htbl
= rcu_dereference_protected(ntbl
->htbl
,
1496 lockdep_is_held(&priv
->lock
));
1501 for (i
= 0; i
< htbl
->size
; i
++) {
1502 struct ipoib_neigh
*neigh
;
1503 struct ipoib_neigh __rcu
**np
= &htbl
->buckets
[i
];
1505 while ((neigh
= rcu_dereference_protected(*np
,
1506 lockdep_is_held(&priv
->lock
))) != NULL
) {
1507 /* delete neighs belong to this parent */
1508 if (!memcmp(gid
, neigh
->daddr
+ 4, sizeof (union ib_gid
))) {
1509 rcu_assign_pointer(*np
,
1510 rcu_dereference_protected(neigh
->hnext
,
1511 lockdep_is_held(&priv
->lock
)));
1512 /* remove from parent list */
1513 list_del(&neigh
->list
);
1514 call_rcu(&neigh
->rcu
, ipoib_neigh_reclaim
);
1522 spin_unlock_irqrestore(&priv
->lock
, flags
);
1525 static void ipoib_flush_neighs(struct ipoib_dev_priv
*priv
)
1527 struct ipoib_neigh_table
*ntbl
= &priv
->ntbl
;
1528 struct ipoib_neigh_hash
*htbl
;
1529 unsigned long flags
;
1530 int i
, wait_flushed
= 0;
1532 init_completion(&priv
->ntbl
.flushed
);
1534 spin_lock_irqsave(&priv
->lock
, flags
);
1536 htbl
= rcu_dereference_protected(ntbl
->htbl
,
1537 lockdep_is_held(&priv
->lock
));
1541 wait_flushed
= atomic_read(&priv
->ntbl
.entries
);
1545 for (i
= 0; i
< htbl
->size
; i
++) {
1546 struct ipoib_neigh
*neigh
;
1547 struct ipoib_neigh __rcu
**np
= &htbl
->buckets
[i
];
1549 while ((neigh
= rcu_dereference_protected(*np
,
1550 lockdep_is_held(&priv
->lock
))) != NULL
) {
1551 rcu_assign_pointer(*np
,
1552 rcu_dereference_protected(neigh
->hnext
,
1553 lockdep_is_held(&priv
->lock
)));
1554 /* remove from path/mc list */
1555 list_del(&neigh
->list
);
1556 call_rcu(&neigh
->rcu
, ipoib_neigh_reclaim
);
1561 rcu_assign_pointer(ntbl
->htbl
, NULL
);
1562 call_rcu(&htbl
->rcu
, neigh_hash_free_rcu
);
1565 spin_unlock_irqrestore(&priv
->lock
, flags
);
1567 wait_for_completion(&priv
->ntbl
.flushed
);
1570 static void ipoib_neigh_hash_uninit(struct net_device
*dev
)
1572 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
1575 ipoib_dbg(priv
, "ipoib_neigh_hash_uninit\n");
1576 init_completion(&priv
->ntbl
.deleted
);
1577 set_bit(IPOIB_NEIGH_TBL_FLUSH
, &priv
->flags
);
1579 /* Stop GC if called at init fail need to cancel work */
1580 stopped
= test_and_set_bit(IPOIB_STOP_NEIGH_GC
, &priv
->flags
);
1582 cancel_delayed_work(&priv
->neigh_reap_task
);
1584 ipoib_flush_neighs(priv
);
1586 wait_for_completion(&priv
->ntbl
.deleted
);
1590 int ipoib_dev_init(struct net_device
*dev
, struct ib_device
*ca
, int port
)
1592 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
1594 /* Allocate RX/TX "rings" to hold queued skbs */
1595 priv
->rx_ring
= kzalloc(ipoib_recvq_size
* sizeof *priv
->rx_ring
,
1597 if (!priv
->rx_ring
) {
1598 printk(KERN_WARNING
"%s: failed to allocate RX ring (%d entries)\n",
1599 ca
->name
, ipoib_recvq_size
);
1603 priv
->tx_ring
= vzalloc(ipoib_sendq_size
* sizeof *priv
->tx_ring
);
1604 if (!priv
->tx_ring
) {
1605 printk(KERN_WARNING
"%s: failed to allocate TX ring (%d entries)\n",
1606 ca
->name
, ipoib_sendq_size
);
1607 goto out_rx_ring_cleanup
;
1610 /* priv->tx_head, tx_tail & tx_outstanding are already 0 */
1612 if (ipoib_ib_dev_init(dev
, ca
, port
))
1613 goto out_tx_ring_cleanup
;
1616 * Must be after ipoib_ib_dev_init so we can allocate a per
1617 * device wq there and use it here
1619 if (ipoib_neigh_hash_init(priv
) < 0)
1620 goto out_dev_uninit
;
1625 ipoib_ib_dev_cleanup(dev
);
1627 out_tx_ring_cleanup
:
1628 vfree(priv
->tx_ring
);
1630 out_rx_ring_cleanup
:
1631 kfree(priv
->rx_ring
);
1637 void ipoib_dev_cleanup(struct net_device
*dev
)
1639 struct ipoib_dev_priv
*priv
= netdev_priv(dev
), *cpriv
, *tcpriv
;
1644 ipoib_delete_debug_files(dev
);
1646 /* Delete any child interfaces first */
1647 list_for_each_entry_safe(cpriv
, tcpriv
, &priv
->child_intfs
, list
) {
1648 /* Stop GC on child */
1649 set_bit(IPOIB_STOP_NEIGH_GC
, &cpriv
->flags
);
1650 cancel_delayed_work(&cpriv
->neigh_reap_task
);
1651 unregister_netdevice_queue(cpriv
->dev
, &head
);
1653 unregister_netdevice_many(&head
);
1656 * Must be before ipoib_ib_dev_cleanup or we delete an in use
1659 ipoib_neigh_hash_uninit(dev
);
1661 ipoib_ib_dev_cleanup(dev
);
1663 kfree(priv
->rx_ring
);
1664 vfree(priv
->tx_ring
);
1666 priv
->rx_ring
= NULL
;
1667 priv
->tx_ring
= NULL
;
1670 static int ipoib_set_vf_link_state(struct net_device
*dev
, int vf
, int link_state
)
1672 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
1674 return ib_set_vf_link_state(priv
->ca
, vf
, priv
->port
, link_state
);
1677 static int ipoib_get_vf_config(struct net_device
*dev
, int vf
,
1678 struct ifla_vf_info
*ivf
)
1680 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
1683 err
= ib_get_vf_config(priv
->ca
, vf
, priv
->port
, ivf
);
1692 static int ipoib_set_vf_guid(struct net_device
*dev
, int vf
, u64 guid
, int type
)
1694 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
1696 if (type
!= IFLA_VF_IB_NODE_GUID
&& type
!= IFLA_VF_IB_PORT_GUID
)
1699 return ib_set_vf_guid(priv
->ca
, vf
, priv
->port
, guid
, type
);
1702 static int ipoib_get_vf_stats(struct net_device
*dev
, int vf
,
1703 struct ifla_vf_stats
*vf_stats
)
1705 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
1707 return ib_get_vf_stats(priv
->ca
, vf
, priv
->port
, vf_stats
);
1710 static const struct header_ops ipoib_header_ops
= {
1711 .create
= ipoib_hard_header
,
1714 static const struct net_device_ops ipoib_netdev_ops_pf
= {
1715 .ndo_uninit
= ipoib_uninit
,
1716 .ndo_open
= ipoib_open
,
1717 .ndo_stop
= ipoib_stop
,
1718 .ndo_change_mtu
= ipoib_change_mtu
,
1719 .ndo_fix_features
= ipoib_fix_features
,
1720 .ndo_start_xmit
= ipoib_start_xmit
,
1721 .ndo_tx_timeout
= ipoib_timeout
,
1722 .ndo_set_rx_mode
= ipoib_set_mcast_list
,
1723 .ndo_get_iflink
= ipoib_get_iflink
,
1724 .ndo_set_vf_link_state
= ipoib_set_vf_link_state
,
1725 .ndo_get_vf_config
= ipoib_get_vf_config
,
1726 .ndo_get_vf_stats
= ipoib_get_vf_stats
,
1727 .ndo_set_vf_guid
= ipoib_set_vf_guid
,
1728 .ndo_set_mac_address
= ipoib_set_mac
,
1731 static const struct net_device_ops ipoib_netdev_ops_vf
= {
1732 .ndo_uninit
= ipoib_uninit
,
1733 .ndo_open
= ipoib_open
,
1734 .ndo_stop
= ipoib_stop
,
1735 .ndo_change_mtu
= ipoib_change_mtu
,
1736 .ndo_fix_features
= ipoib_fix_features
,
1737 .ndo_start_xmit
= ipoib_start_xmit
,
1738 .ndo_tx_timeout
= ipoib_timeout
,
1739 .ndo_set_rx_mode
= ipoib_set_mcast_list
,
1740 .ndo_get_iflink
= ipoib_get_iflink
,
1743 void ipoib_setup(struct net_device
*dev
)
1745 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
1747 if (priv
->hca_caps
& IB_DEVICE_VIRTUAL_FUNCTION
)
1748 dev
->netdev_ops
= &ipoib_netdev_ops_vf
;
1750 dev
->netdev_ops
= &ipoib_netdev_ops_pf
;
1752 dev
->header_ops
= &ipoib_header_ops
;
1754 ipoib_set_ethtool_ops(dev
);
1756 netif_napi_add(dev
, &priv
->napi
, ipoib_poll
, NAPI_POLL_WEIGHT
);
1758 dev
->watchdog_timeo
= HZ
;
1760 dev
->flags
|= IFF_BROADCAST
| IFF_MULTICAST
;
1762 dev
->hard_header_len
= IPOIB_ENCAP_LEN
;
1763 dev
->addr_len
= INFINIBAND_ALEN
;
1764 dev
->type
= ARPHRD_INFINIBAND
;
1765 dev
->tx_queue_len
= ipoib_sendq_size
* 2;
1766 dev
->features
= (NETIF_F_VLAN_CHALLENGED
|
1768 netif_keep_dst(dev
);
1770 memcpy(dev
->broadcast
, ipv4_bcast_addr
, INFINIBAND_ALEN
);
1774 spin_lock_init(&priv
->lock
);
1776 init_rwsem(&priv
->vlan_rwsem
);
1778 INIT_LIST_HEAD(&priv
->path_list
);
1779 INIT_LIST_HEAD(&priv
->child_intfs
);
1780 INIT_LIST_HEAD(&priv
->dead_ahs
);
1781 INIT_LIST_HEAD(&priv
->multicast_list
);
1783 INIT_DELAYED_WORK(&priv
->mcast_task
, ipoib_mcast_join_task
);
1784 INIT_WORK(&priv
->carrier_on_task
, ipoib_mcast_carrier_on_task
);
1785 INIT_WORK(&priv
->flush_light
, ipoib_ib_dev_flush_light
);
1786 INIT_WORK(&priv
->flush_normal
, ipoib_ib_dev_flush_normal
);
1787 INIT_WORK(&priv
->flush_heavy
, ipoib_ib_dev_flush_heavy
);
1788 INIT_WORK(&priv
->restart_task
, ipoib_mcast_restart_task
);
1789 INIT_DELAYED_WORK(&priv
->ah_reap_task
, ipoib_reap_ah
);
1790 INIT_DELAYED_WORK(&priv
->neigh_reap_task
, ipoib_reap_neigh
);
1793 struct ipoib_dev_priv
*ipoib_intf_alloc(const char *name
)
1795 struct net_device
*dev
;
1797 dev
= alloc_netdev((int)sizeof(struct ipoib_dev_priv
), name
,
1798 NET_NAME_UNKNOWN
, ipoib_setup
);
1802 return netdev_priv(dev
);
1805 static ssize_t
show_pkey(struct device
*dev
,
1806 struct device_attribute
*attr
, char *buf
)
1808 struct ipoib_dev_priv
*priv
= netdev_priv(to_net_dev(dev
));
1810 return sprintf(buf
, "0x%04x\n", priv
->pkey
);
1812 static DEVICE_ATTR(pkey
, S_IRUGO
, show_pkey
, NULL
);
1814 static ssize_t
show_umcast(struct device
*dev
,
1815 struct device_attribute
*attr
, char *buf
)
1817 struct ipoib_dev_priv
*priv
= netdev_priv(to_net_dev(dev
));
1819 return sprintf(buf
, "%d\n", test_bit(IPOIB_FLAG_UMCAST
, &priv
->flags
));
1822 void ipoib_set_umcast(struct net_device
*ndev
, int umcast_val
)
1824 struct ipoib_dev_priv
*priv
= netdev_priv(ndev
);
1826 if (umcast_val
> 0) {
1827 set_bit(IPOIB_FLAG_UMCAST
, &priv
->flags
);
1828 ipoib_warn(priv
, "ignoring multicast groups joined directly "
1831 clear_bit(IPOIB_FLAG_UMCAST
, &priv
->flags
);
1834 static ssize_t
set_umcast(struct device
*dev
,
1835 struct device_attribute
*attr
,
1836 const char *buf
, size_t count
)
1838 unsigned long umcast_val
= simple_strtoul(buf
, NULL
, 0);
1840 ipoib_set_umcast(to_net_dev(dev
), umcast_val
);
1844 static DEVICE_ATTR(umcast
, S_IWUSR
| S_IRUGO
, show_umcast
, set_umcast
);
1846 int ipoib_add_umcast_attr(struct net_device
*dev
)
1848 return device_create_file(&dev
->dev
, &dev_attr_umcast
);
1851 static void set_base_guid(struct ipoib_dev_priv
*priv
, union ib_gid
*gid
)
1853 struct ipoib_dev_priv
*child_priv
;
1854 struct net_device
*netdev
= priv
->dev
;
1856 netif_addr_lock_bh(netdev
);
1858 memcpy(&priv
->local_gid
.global
.interface_id
,
1859 &gid
->global
.interface_id
,
1860 sizeof(gid
->global
.interface_id
));
1861 memcpy(netdev
->dev_addr
+ 4, &priv
->local_gid
, sizeof(priv
->local_gid
));
1862 clear_bit(IPOIB_FLAG_DEV_ADDR_SET
, &priv
->flags
);
1864 netif_addr_unlock_bh(netdev
);
1866 if (!test_bit(IPOIB_FLAG_SUBINTERFACE
, &priv
->flags
)) {
1867 down_read(&priv
->vlan_rwsem
);
1868 list_for_each_entry(child_priv
, &priv
->child_intfs
, list
)
1869 set_base_guid(child_priv
, gid
);
1870 up_read(&priv
->vlan_rwsem
);
1874 static int ipoib_check_lladdr(struct net_device
*dev
,
1875 struct sockaddr_storage
*ss
)
1877 union ib_gid
*gid
= (union ib_gid
*)(ss
->__data
+ 4);
1880 netif_addr_lock_bh(dev
);
1882 /* Make sure the QPN, reserved and subnet prefix match the current
1883 * lladdr, it also makes sure the lladdr is unicast.
1885 if (memcmp(dev
->dev_addr
, ss
->__data
,
1886 4 + sizeof(gid
->global
.subnet_prefix
)) ||
1887 gid
->global
.interface_id
== 0)
1890 netif_addr_unlock_bh(dev
);
1895 static int ipoib_set_mac(struct net_device
*dev
, void *addr
)
1897 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
1898 struct sockaddr_storage
*ss
= addr
;
1901 if (!(dev
->priv_flags
& IFF_LIVE_ADDR_CHANGE
) && netif_running(dev
))
1904 ret
= ipoib_check_lladdr(dev
, ss
);
1908 set_base_guid(priv
, (union ib_gid
*)(ss
->__data
+ 4));
1910 queue_work(ipoib_workqueue
, &priv
->flush_light
);
1915 static ssize_t
create_child(struct device
*dev
,
1916 struct device_attribute
*attr
,
1917 const char *buf
, size_t count
)
1922 if (sscanf(buf
, "%i", &pkey
) != 1)
1925 if (pkey
<= 0 || pkey
> 0xffff || pkey
== 0x8000)
1929 * Set the full membership bit, so that we join the right
1930 * broadcast group, etc.
1934 ret
= ipoib_vlan_add(to_net_dev(dev
), pkey
);
1936 return ret
? ret
: count
;
1938 static DEVICE_ATTR(create_child
, S_IWUSR
, NULL
, create_child
);
1940 static ssize_t
delete_child(struct device
*dev
,
1941 struct device_attribute
*attr
,
1942 const char *buf
, size_t count
)
1947 if (sscanf(buf
, "%i", &pkey
) != 1)
1950 if (pkey
< 0 || pkey
> 0xffff)
1953 ret
= ipoib_vlan_delete(to_net_dev(dev
), pkey
);
1955 return ret
? ret
: count
;
1958 static DEVICE_ATTR(delete_child
, S_IWUSR
, NULL
, delete_child
);
1960 int ipoib_add_pkey_attr(struct net_device
*dev
)
1962 return device_create_file(&dev
->dev
, &dev_attr_pkey
);
1965 int ipoib_set_dev_features(struct ipoib_dev_priv
*priv
, struct ib_device
*hca
)
1967 priv
->hca_caps
= hca
->attrs
.device_cap_flags
;
1969 if (priv
->hca_caps
& IB_DEVICE_UD_IP_CSUM
) {
1970 priv
->dev
->hw_features
= NETIF_F_IP_CSUM
| NETIF_F_RXCSUM
;
1972 if (priv
->hca_caps
& IB_DEVICE_UD_TSO
)
1973 priv
->dev
->hw_features
|= NETIF_F_TSO
;
1975 priv
->dev
->features
|= priv
->dev
->hw_features
;
1981 static struct net_device
*ipoib_add_port(const char *format
,
1982 struct ib_device
*hca
, u8 port
)
1984 struct ipoib_dev_priv
*priv
;
1985 struct ib_port_attr attr
;
1986 int result
= -ENOMEM
;
1988 priv
= ipoib_intf_alloc(format
);
1990 goto alloc_mem_failed
;
1992 SET_NETDEV_DEV(priv
->dev
, hca
->dma_device
);
1993 priv
->dev
->dev_id
= port
- 1;
1995 result
= ib_query_port(hca
, port
, &attr
);
1997 priv
->max_ib_mtu
= ib_mtu_enum_to_int(attr
.max_mtu
);
1999 printk(KERN_WARNING
"%s: ib_query_port %d failed\n",
2001 goto device_init_failed
;
2004 /* MTU will be reset when mcast join happens */
2005 priv
->dev
->mtu
= IPOIB_UD_MTU(priv
->max_ib_mtu
);
2006 priv
->mcast_mtu
= priv
->admin_mtu
= priv
->dev
->mtu
;
2008 priv
->dev
->neigh_priv_len
= sizeof(struct ipoib_neigh
);
2010 result
= ib_query_pkey(hca
, port
, 0, &priv
->pkey
);
2012 printk(KERN_WARNING
"%s: ib_query_pkey port %d failed (ret = %d)\n",
2013 hca
->name
, port
, result
);
2014 goto device_init_failed
;
2017 result
= ipoib_set_dev_features(priv
, hca
);
2019 goto device_init_failed
;
2022 * Set the full membership bit, so that we join the right
2023 * broadcast group, etc.
2025 priv
->pkey
|= 0x8000;
2027 priv
->dev
->broadcast
[8] = priv
->pkey
>> 8;
2028 priv
->dev
->broadcast
[9] = priv
->pkey
& 0xff;
2030 result
= ib_query_gid(hca
, port
, 0, &priv
->local_gid
, NULL
);
2032 printk(KERN_WARNING
"%s: ib_query_gid port %d failed (ret = %d)\n",
2033 hca
->name
, port
, result
);
2034 goto device_init_failed
;
2036 memcpy(priv
->dev
->dev_addr
+ 4, priv
->local_gid
.raw
, sizeof (union ib_gid
));
2037 set_bit(IPOIB_FLAG_DEV_ADDR_SET
, &priv
->flags
);
2039 result
= ipoib_dev_init(priv
->dev
, hca
, port
);
2041 printk(KERN_WARNING
"%s: failed to initialize port %d (ret = %d)\n",
2042 hca
->name
, port
, result
);
2043 goto device_init_failed
;
2046 INIT_IB_EVENT_HANDLER(&priv
->event_handler
,
2047 priv
->ca
, ipoib_event
);
2048 result
= ib_register_event_handler(&priv
->event_handler
);
2050 printk(KERN_WARNING
"%s: ib_register_event_handler failed for "
2051 "port %d (ret = %d)\n",
2052 hca
->name
, port
, result
);
2056 result
= register_netdev(priv
->dev
);
2058 printk(KERN_WARNING
"%s: couldn't register ipoib port %d; error %d\n",
2059 hca
->name
, port
, result
);
2060 goto register_failed
;
2063 ipoib_create_debug_files(priv
->dev
);
2065 if (ipoib_cm_add_mode_attr(priv
->dev
))
2067 if (ipoib_add_pkey_attr(priv
->dev
))
2069 if (ipoib_add_umcast_attr(priv
->dev
))
2071 if (device_create_file(&priv
->dev
->dev
, &dev_attr_create_child
))
2073 if (device_create_file(&priv
->dev
->dev
, &dev_attr_delete_child
))
2079 ipoib_delete_debug_files(priv
->dev
);
2080 unregister_netdev(priv
->dev
);
2083 ib_unregister_event_handler(&priv
->event_handler
);
2084 flush_workqueue(ipoib_workqueue
);
2085 /* Stop GC if started before flush */
2086 set_bit(IPOIB_STOP_NEIGH_GC
, &priv
->flags
);
2087 cancel_delayed_work(&priv
->neigh_reap_task
);
2088 flush_workqueue(priv
->wq
);
2091 ipoib_dev_cleanup(priv
->dev
);
2094 free_netdev(priv
->dev
);
2097 return ERR_PTR(result
);
2100 static void ipoib_add_one(struct ib_device
*device
)
2102 struct list_head
*dev_list
;
2103 struct net_device
*dev
;
2104 struct ipoib_dev_priv
*priv
;
2108 dev_list
= kmalloc(sizeof *dev_list
, GFP_KERNEL
);
2112 INIT_LIST_HEAD(dev_list
);
2114 for (p
= rdma_start_port(device
); p
<= rdma_end_port(device
); ++p
) {
2115 if (!rdma_protocol_ib(device
, p
))
2117 dev
= ipoib_add_port("ib%d", device
, p
);
2119 priv
= netdev_priv(dev
);
2120 list_add_tail(&priv
->list
, dev_list
);
2130 ib_set_client_data(device
, &ipoib_client
, dev_list
);
2133 static void ipoib_remove_one(struct ib_device
*device
, void *client_data
)
2135 struct ipoib_dev_priv
*priv
, *tmp
;
2136 struct list_head
*dev_list
= client_data
;
2141 list_for_each_entry_safe(priv
, tmp
, dev_list
, list
) {
2142 ib_unregister_event_handler(&priv
->event_handler
);
2143 flush_workqueue(ipoib_workqueue
);
2145 /* mark interface in the middle of destruction */
2146 set_bit(IPOIB_FLAG_GOING_DOWN
, &priv
->flags
);
2149 dev_change_flags(priv
->dev
, priv
->dev
->flags
& ~IFF_UP
);
2153 set_bit(IPOIB_STOP_NEIGH_GC
, &priv
->flags
);
2154 cancel_delayed_work(&priv
->neigh_reap_task
);
2155 flush_workqueue(priv
->wq
);
2157 unregister_netdev(priv
->dev
);
2158 free_netdev(priv
->dev
);
2164 static int __init
ipoib_init_module(void)
2168 ipoib_recvq_size
= roundup_pow_of_two(ipoib_recvq_size
);
2169 ipoib_recvq_size
= min(ipoib_recvq_size
, IPOIB_MAX_QUEUE_SIZE
);
2170 ipoib_recvq_size
= max(ipoib_recvq_size
, IPOIB_MIN_QUEUE_SIZE
);
2172 ipoib_sendq_size
= roundup_pow_of_two(ipoib_sendq_size
);
2173 ipoib_sendq_size
= min(ipoib_sendq_size
, IPOIB_MAX_QUEUE_SIZE
);
2174 ipoib_sendq_size
= max3(ipoib_sendq_size
, 2 * MAX_SEND_CQE
, IPOIB_MIN_QUEUE_SIZE
);
2175 #ifdef CONFIG_INFINIBAND_IPOIB_CM
2176 ipoib_max_conn_qp
= min(ipoib_max_conn_qp
, IPOIB_CM_MAX_CONN_QP
);
2180 * When copying small received packets, we only copy from the
2181 * linear data part of the SKB, so we rely on this condition.
2183 BUILD_BUG_ON(IPOIB_CM_COPYBREAK
> IPOIB_CM_HEAD_SIZE
);
2185 ret
= ipoib_register_debugfs();
2190 * We create a global workqueue here that is used for all flush
2191 * operations. However, if you attempt to flush a workqueue
2192 * from a task on that same workqueue, it deadlocks the system.
2193 * We want to be able to flush the tasks associated with a
2194 * specific net device, so we also create a workqueue for each
2195 * netdevice. We queue up the tasks for that device only on
2196 * its private workqueue, and we only queue up flush events
2197 * on our global flush workqueue. This avoids the deadlocks.
2199 ipoib_workqueue
= create_singlethread_workqueue("ipoib_flush");
2200 if (!ipoib_workqueue
) {
2205 ib_sa_register_client(&ipoib_sa_client
);
2207 ret
= ib_register_client(&ipoib_client
);
2211 ret
= ipoib_netlink_init();
2218 ib_unregister_client(&ipoib_client
);
2221 ib_sa_unregister_client(&ipoib_sa_client
);
2222 destroy_workqueue(ipoib_workqueue
);
2225 ipoib_unregister_debugfs();
2230 static void __exit
ipoib_cleanup_module(void)
2232 ipoib_netlink_fini();
2233 ib_unregister_client(&ipoib_client
);
2234 ib_sa_unregister_client(&ipoib_sa_client
);
2235 ipoib_unregister_debugfs();
2236 destroy_workqueue(ipoib_workqueue
);
2239 module_init(ipoib_init_module
);
2240 module_exit(ipoib_cleanup_module
);