2 * Copyright (c) 2004, 2005 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.h 1358 2004-12-17 22:00:11Z roland $
40 #include <linux/list.h>
41 #include <linux/skbuff.h>
42 #include <linux/netdevice.h>
43 #include <linux/workqueue.h>
44 #include <linux/kref.h>
45 #include <linux/if_infiniband.h>
46 #include <linux/mutex.h>
48 #include <net/neighbour.h>
50 #include <asm/atomic.h>
52 #include <rdma/ib_verbs.h>
53 #include <rdma/ib_pack.h>
54 #include <rdma/ib_sa.h>
59 IPOIB_PACKET_SIZE
= 2048,
60 IPOIB_BUF_SIZE
= IPOIB_PACKET_SIZE
+ IB_GRH_BYTES
,
64 IPOIB_CM_MTU
= 0x10000 - 0x10, /* padding to align header to 16 */
65 IPOIB_CM_BUF_SIZE
= IPOIB_CM_MTU
+ IPOIB_ENCAP_LEN
,
66 IPOIB_CM_HEAD_SIZE
= IPOIB_CM_BUF_SIZE
% PAGE_SIZE
,
67 IPOIB_CM_RX_SG
= ALIGN(IPOIB_CM_BUF_SIZE
, PAGE_SIZE
) / PAGE_SIZE
,
68 IPOIB_RX_RING_SIZE
= 128,
69 IPOIB_TX_RING_SIZE
= 64,
70 IPOIB_MAX_QUEUE_SIZE
= 8192,
71 IPOIB_MIN_QUEUE_SIZE
= 2,
75 IPOIB_MAX_PATH_REC_QUEUE
= 3,
76 IPOIB_MAX_MCAST_QUEUE
= 3,
78 IPOIB_FLAG_OPER_UP
= 0,
79 IPOIB_FLAG_INITIALIZED
= 1,
80 IPOIB_FLAG_ADMIN_UP
= 2,
81 IPOIB_PKEY_ASSIGNED
= 3,
83 IPOIB_FLAG_SUBINTERFACE
= 5,
85 IPOIB_STOP_REAPER
= 7,
86 IPOIB_MCAST_STARTED
= 8,
87 IPOIB_FLAG_NETIF_STOPPED
= 9,
88 IPOIB_FLAG_ADMIN_CM
= 10,
90 IPOIB_MAX_BACKOFF_SECONDS
= 16,
92 IPOIB_MCAST_FLAG_FOUND
= 0, /* used in set_multicast_list */
93 IPOIB_MCAST_FLAG_SENDONLY
= 1,
94 IPOIB_MCAST_FLAG_BUSY
= 2, /* joining or already joined */
95 IPOIB_MCAST_FLAG_ATTACHED
= 3,
98 #define IPOIB_OP_RECV (1ul << 31)
99 #ifdef CONFIG_INFINIBAND_IPOIB_CM
100 #define IPOIB_CM_OP_SRQ (1ul << 30)
102 #define IPOIB_CM_OP_SRQ (0)
107 struct ipoib_header
{
112 struct ipoib_pseudoheader
{
113 u8 hwaddr
[INFINIBAND_ALEN
];
118 struct ipoib_rx_buf
{
123 struct ipoib_tx_buf
{
130 struct ipoib_cm_data
{
131 __be32 qpn
; /* High byte MUST be ignored on receive */
138 struct list_head list
;
139 struct net_device
*dev
;
140 unsigned long jiffies
;
147 struct list_head list
;
148 struct net_device
*dev
;
149 struct ipoib_neigh
*neigh
;
150 struct ipoib_path
*path
;
151 struct ipoib_tx_buf
*tx_ring
;
156 struct ib_wc ibwc
[IPOIB_NUM_WC
];
159 struct ipoib_cm_rx_buf
{
161 u64 mapping
[IPOIB_CM_RX_SG
];
164 struct ipoib_cm_dev_priv
{
166 struct ipoib_cm_rx_buf
*srq_ring
;
168 struct list_head passive_ids
;
169 struct work_struct start_task
;
170 struct work_struct reap_task
;
171 struct work_struct skb_task
;
172 struct delayed_work stale_task
;
173 struct sk_buff_head skb_queue
;
174 struct list_head start_list
;
175 struct list_head reap_list
;
176 struct ib_wc ibwc
[IPOIB_NUM_WC
];
177 struct ib_sge rx_sge
[IPOIB_CM_RX_SG
];
178 struct ib_recv_wr rx_wr
;
182 * Device private locking: tx_lock protects members used in TX fast
183 * path (and we use LLTX so upper layers don't do extra locking).
184 * lock protects everything else. lock nests inside of tx_lock (ie
185 * tx_lock must be acquired first if needed).
187 struct ipoib_dev_priv
{
190 struct net_device
*dev
;
194 struct mutex mcast_mutex
;
195 struct mutex vlan_mutex
;
197 struct rb_root path_tree
;
198 struct list_head path_list
;
200 struct ipoib_mcast
*broadcast
;
201 struct list_head multicast_list
;
202 struct rb_root multicast_tree
;
204 struct delayed_work pkey_poll_task
;
205 struct delayed_work mcast_task
;
206 struct work_struct flush_task
;
207 struct work_struct restart_task
;
208 struct delayed_work ah_reap_task
;
209 struct work_struct pkey_event_task
;
211 struct ib_device
*ca
;
221 union ib_gid local_gid
;
224 unsigned int admin_mtu
;
225 unsigned int mcast_mtu
;
227 struct ipoib_rx_buf
*rx_ring
;
230 struct ipoib_tx_buf
*tx_ring
;
233 struct ib_sge tx_sge
;
234 struct ib_send_wr tx_wr
;
236 struct ib_wc ibwc
[IPOIB_NUM_WC
];
238 struct list_head dead_ahs
;
240 struct ib_event_handler event_handler
;
242 struct net_device_stats stats
;
244 struct net_device
*parent
;
245 struct list_head child_intfs
;
246 struct list_head list
;
248 #ifdef CONFIG_INFINIBAND_IPOIB_CM
249 struct ipoib_cm_dev_priv cm
;
252 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
253 struct list_head fs_list
;
254 struct dentry
*mcg_dentry
;
255 struct dentry
*path_dentry
;
260 struct net_device
*dev
;
262 struct list_head list
;
268 struct net_device
*dev
;
269 struct ib_sa_path_rec pathrec
;
271 struct sk_buff_head queue
;
273 struct list_head neigh_list
;
276 struct ib_sa_query
*query
;
277 struct completion done
;
279 struct rb_node rb_node
;
280 struct list_head list
;
285 #ifdef CONFIG_INFINIBAND_IPOIB_CM
286 struct ipoib_cm_tx
*cm
;
289 struct sk_buff_head queue
;
291 struct neighbour
*neighbour
;
293 struct list_head list
;
297 * We stash a pointer to our private neighbour information after our
298 * hardware address in neigh->ha. The ALIGN() expression here makes
299 * sure that this pointer is stored aligned so that an unaligned
300 * load is not needed to dereference it.
302 static inline struct ipoib_neigh
**to_ipoib_neigh(struct neighbour
*neigh
)
304 return (void*) neigh
+ ALIGN(offsetof(struct neighbour
, ha
) +
305 INFINIBAND_ALEN
, sizeof(void *));
308 struct ipoib_neigh
*ipoib_neigh_alloc(struct neighbour
*neigh
);
309 void ipoib_neigh_free(struct net_device
*dev
, struct ipoib_neigh
*neigh
);
311 extern struct workqueue_struct
*ipoib_workqueue
;
315 int ipoib_poll(struct net_device
*dev
, int *budget
);
316 void ipoib_ib_completion(struct ib_cq
*cq
, void *dev_ptr
);
318 struct ipoib_ah
*ipoib_create_ah(struct net_device
*dev
,
319 struct ib_pd
*pd
, struct ib_ah_attr
*attr
);
320 void ipoib_free_ah(struct kref
*kref
);
321 static inline void ipoib_put_ah(struct ipoib_ah
*ah
)
323 kref_put(&ah
->ref
, ipoib_free_ah
);
326 int ipoib_open(struct net_device
*dev
);
327 int ipoib_add_pkey_attr(struct net_device
*dev
);
329 void ipoib_send(struct net_device
*dev
, struct sk_buff
*skb
,
330 struct ipoib_ah
*address
, u32 qpn
);
331 void ipoib_reap_ah(struct work_struct
*work
);
333 void ipoib_flush_paths(struct net_device
*dev
);
334 struct ipoib_dev_priv
*ipoib_intf_alloc(const char *format
);
336 int ipoib_ib_dev_init(struct net_device
*dev
, struct ib_device
*ca
, int port
);
337 void ipoib_ib_dev_flush(struct work_struct
*work
);
338 void ipoib_pkey_event(struct work_struct
*work
);
339 void ipoib_ib_dev_cleanup(struct net_device
*dev
);
341 int ipoib_ib_dev_open(struct net_device
*dev
);
342 int ipoib_ib_dev_up(struct net_device
*dev
);
343 int ipoib_ib_dev_down(struct net_device
*dev
, int flush
);
344 int ipoib_ib_dev_stop(struct net_device
*dev
, int flush
);
346 int ipoib_dev_init(struct net_device
*dev
, struct ib_device
*ca
, int port
);
347 void ipoib_dev_cleanup(struct net_device
*dev
);
349 void ipoib_mcast_join_task(struct work_struct
*work
);
350 void ipoib_mcast_send(struct net_device
*dev
, void *mgid
, struct sk_buff
*skb
);
352 void ipoib_mcast_restart_task(struct work_struct
*work
);
353 int ipoib_mcast_start_thread(struct net_device
*dev
);
354 int ipoib_mcast_stop_thread(struct net_device
*dev
, int flush
);
356 void ipoib_mcast_dev_down(struct net_device
*dev
);
357 void ipoib_mcast_dev_flush(struct net_device
*dev
);
359 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
360 struct ipoib_mcast_iter
*ipoib_mcast_iter_init(struct net_device
*dev
);
361 int ipoib_mcast_iter_next(struct ipoib_mcast_iter
*iter
);
362 void ipoib_mcast_iter_read(struct ipoib_mcast_iter
*iter
,
364 unsigned long *created
,
365 unsigned int *queuelen
,
366 unsigned int *complete
,
367 unsigned int *send_only
);
369 struct ipoib_path_iter
*ipoib_path_iter_init(struct net_device
*dev
);
370 int ipoib_path_iter_next(struct ipoib_path_iter
*iter
);
371 void ipoib_path_iter_read(struct ipoib_path_iter
*iter
,
372 struct ipoib_path
*path
);
375 int ipoib_mcast_attach(struct net_device
*dev
, u16 mlid
,
377 int ipoib_mcast_detach(struct net_device
*dev
, u16 mlid
,
380 int ipoib_init_qp(struct net_device
*dev
);
381 int ipoib_transport_dev_init(struct net_device
*dev
, struct ib_device
*ca
);
382 void ipoib_transport_dev_cleanup(struct net_device
*dev
);
384 void ipoib_event(struct ib_event_handler
*handler
,
385 struct ib_event
*record
);
387 int ipoib_vlan_add(struct net_device
*pdev
, unsigned short pkey
);
388 int ipoib_vlan_delete(struct net_device
*pdev
, unsigned short pkey
);
390 void ipoib_pkey_poll(struct work_struct
*work
);
391 int ipoib_pkey_dev_delay_open(struct net_device
*dev
);
393 #ifdef CONFIG_INFINIBAND_IPOIB_CM
395 #define IPOIB_FLAGS_RC 0x80
396 #define IPOIB_FLAGS_UC 0x40
398 /* We don't support UC connections at the moment */
399 #define IPOIB_CM_SUPPORTED(ha) (ha[0] & (IPOIB_FLAGS_RC))
401 static inline int ipoib_cm_admin_enabled(struct net_device
*dev
)
403 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
404 return IPOIB_CM_SUPPORTED(dev
->dev_addr
) &&
405 test_bit(IPOIB_FLAG_ADMIN_CM
, &priv
->flags
);
408 static inline int ipoib_cm_enabled(struct net_device
*dev
, struct neighbour
*n
)
410 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
411 return IPOIB_CM_SUPPORTED(n
->ha
) &&
412 test_bit(IPOIB_FLAG_ADMIN_CM
, &priv
->flags
);
415 static inline int ipoib_cm_up(struct ipoib_neigh
*neigh
)
418 return test_bit(IPOIB_FLAG_OPER_UP
, &neigh
->cm
->flags
);
421 static inline struct ipoib_cm_tx
*ipoib_cm_get(struct ipoib_neigh
*neigh
)
426 static inline void ipoib_cm_set(struct ipoib_neigh
*neigh
, struct ipoib_cm_tx
*tx
)
431 void ipoib_cm_send(struct net_device
*dev
, struct sk_buff
*skb
, struct ipoib_cm_tx
*tx
);
432 int ipoib_cm_dev_open(struct net_device
*dev
);
433 void ipoib_cm_dev_stop(struct net_device
*dev
);
434 int ipoib_cm_dev_init(struct net_device
*dev
);
435 int ipoib_cm_add_mode_attr(struct net_device
*dev
);
436 void ipoib_cm_dev_cleanup(struct net_device
*dev
);
437 struct ipoib_cm_tx
*ipoib_cm_create_tx(struct net_device
*dev
, struct ipoib_path
*path
,
438 struct ipoib_neigh
*neigh
);
439 void ipoib_cm_destroy_tx(struct ipoib_cm_tx
*tx
);
440 void ipoib_cm_skb_too_long(struct net_device
* dev
, struct sk_buff
*skb
,
442 void ipoib_cm_handle_rx_wc(struct net_device
*dev
, struct ib_wc
*wc
);
447 static inline int ipoib_cm_admin_enabled(struct net_device
*dev
)
451 static inline int ipoib_cm_enabled(struct net_device
*dev
, struct neighbour
*n
)
457 static inline int ipoib_cm_up(struct ipoib_neigh
*neigh
)
463 static inline struct ipoib_cm_tx
*ipoib_cm_get(struct ipoib_neigh
*neigh
)
468 static inline void ipoib_cm_set(struct ipoib_neigh
*neigh
, struct ipoib_cm_tx
*tx
)
473 void ipoib_cm_send(struct net_device
*dev
, struct sk_buff
*skb
, struct ipoib_cm_tx
*tx
)
479 int ipoib_cm_dev_open(struct net_device
*dev
)
485 void ipoib_cm_dev_stop(struct net_device
*dev
)
491 int ipoib_cm_dev_init(struct net_device
*dev
)
497 void ipoib_cm_dev_cleanup(struct net_device
*dev
)
503 struct ipoib_cm_tx
*ipoib_cm_create_tx(struct net_device
*dev
, struct ipoib_path
*path
,
504 struct ipoib_neigh
*neigh
)
510 void ipoib_cm_destroy_tx(struct ipoib_cm_tx
*tx
)
516 int ipoib_cm_add_mode_attr(struct net_device
*dev
)
521 static inline void ipoib_cm_skb_too_long(struct net_device
* dev
, struct sk_buff
*skb
,
524 dev_kfree_skb_any(skb
);
527 static inline void ipoib_cm_handle_rx_wc(struct net_device
*dev
, struct ib_wc
*wc
)
533 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
534 void ipoib_create_debug_files(struct net_device
*dev
);
535 void ipoib_delete_debug_files(struct net_device
*dev
);
536 int ipoib_register_debugfs(void);
537 void ipoib_unregister_debugfs(void);
539 static inline void ipoib_create_debug_files(struct net_device
*dev
) { }
540 static inline void ipoib_delete_debug_files(struct net_device
*dev
) { }
541 static inline int ipoib_register_debugfs(void) { return 0; }
542 static inline void ipoib_unregister_debugfs(void) { }
546 #define ipoib_printk(level, priv, format, arg...) \
547 printk(level "%s: " format, ((struct ipoib_dev_priv *) priv)->dev->name , ## arg)
548 #define ipoib_warn(priv, format, arg...) \
549 ipoib_printk(KERN_WARNING, priv, format , ## arg)
551 extern int ipoib_sendq_size
;
552 extern int ipoib_recvq_size
;
554 extern struct ib_sa_client ipoib_sa_client
;
556 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
557 extern int ipoib_debug_level
;
559 #define ipoib_dbg(priv, format, arg...) \
561 if (ipoib_debug_level > 0) \
562 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
564 #define ipoib_dbg_mcast(priv, format, arg...) \
566 if (mcast_debug_level > 0) \
567 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
569 #else /* CONFIG_INFINIBAND_IPOIB_DEBUG */
570 #define ipoib_dbg(priv, format, arg...) \
571 do { (void) (priv); } while (0)
572 #define ipoib_dbg_mcast(priv, format, arg...) \
573 do { (void) (priv); } while (0)
574 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */
576 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG_DATA
577 #define ipoib_dbg_data(priv, format, arg...) \
579 if (data_debug_level > 0) \
580 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
582 #else /* CONFIG_INFINIBAND_IPOIB_DEBUG_DATA */
583 #define ipoib_dbg_data(priv, format, arg...) \
584 do { (void) (priv); } while (0)
585 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG_DATA */
588 #define IPOIB_GID_FMT "%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x:" \
589 "%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x"
591 #define IPOIB_GID_RAW_ARG(gid) ((u8 *)(gid))[0], \
608 #define IPOIB_GID_ARG(gid) IPOIB_GID_RAW_ARG((gid).raw)
610 #define IPOIB_QPN(ha) (be32_to_cpup((__be32 *) ha) & 0xffffff)
612 #endif /* _IPOIB_H */