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_ADMIN_CM
= 9,
88 IPOIB_FLAG_UMCAST
= 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_OP_CM (1ul << 30)
102 #define IPOIB_OP_CM (0)
107 struct ipoib_header
{
112 struct ipoib_pseudoheader
{
113 u8 hwaddr
[INFINIBAND_ALEN
];
116 /* Used for all multicast joins (broadcast, IPv4 mcast and IPv6 mcast) */
118 struct ib_sa_mcmember_rec mcmember
;
119 struct ib_sa_multicast
*mc
;
122 struct rb_node rb_node
;
123 struct list_head list
;
125 unsigned long created
;
126 unsigned long backoff
;
129 unsigned char logcount
;
131 struct list_head neigh_list
;
133 struct sk_buff_head pkt_queue
;
135 struct net_device
*dev
;
138 struct ipoib_rx_buf
{
143 struct ipoib_tx_buf
{
150 struct ipoib_cm_data
{
151 __be32 qpn
; /* High byte MUST be ignored on receive */
156 * Quoting 10.3.1 Queue Pair and EE Context States:
158 * Note, for QPs that are associated with an SRQ, the Consumer should take the
159 * QP through the Error State before invoking a Destroy QP or a Modify QP to the
160 * Reset State. The Consumer may invoke the Destroy QP without first performing
161 * a Modify QP to the Error State and waiting for the Affiliated Asynchronous
162 * Last WQE Reached Event. However, if the Consumer does not wait for the
163 * Affiliated Asynchronous Last WQE Reached Event, then WQE and Data Segment
164 * leakage may occur. Therefore, it is good programming practice to tear down a
165 * QP that is associated with an SRQ by using the following process:
167 * - Put the QP in the Error State
168 * - Wait for the Affiliated Asynchronous Last WQE Reached Event;
170 * drain the CQ by invoking the Poll CQ verb and either wait for CQ
171 * to be empty or the number of Poll CQ operations has exceeded
174 * post another WR that completes on the same CQ and wait for this
175 * WR to return as a WC;
176 * - and then invoke a Destroy QP or Reset QP.
178 * We use the second option and wait for a completion on the
179 * same CQ before destroying QPs attached to our SRQ.
182 enum ipoib_cm_state
{
184 IPOIB_CM_RX_ERROR
, /* Ignored by stale task */
185 IPOIB_CM_RX_FLUSH
/* Last WQE Reached event observed */
191 struct list_head list
;
192 struct net_device
*dev
;
193 unsigned long jiffies
;
194 enum ipoib_cm_state state
;
200 struct list_head list
;
201 struct net_device
*dev
;
202 struct ipoib_neigh
*neigh
;
203 struct ipoib_path
*path
;
204 struct ipoib_tx_buf
*tx_ring
;
209 struct ib_wc ibwc
[IPOIB_NUM_WC
];
212 struct ipoib_cm_rx_buf
{
214 u64 mapping
[IPOIB_CM_RX_SG
];
217 struct ipoib_cm_dev_priv
{
219 struct ipoib_cm_rx_buf
*srq_ring
;
221 struct list_head passive_ids
; /* state: LIVE */
222 struct list_head rx_error_list
; /* state: ERROR */
223 struct list_head rx_flush_list
; /* state: FLUSH, drain not started */
224 struct list_head rx_drain_list
; /* state: FLUSH, drain started */
225 struct list_head rx_reap_list
; /* state: FLUSH, drain done */
226 struct work_struct start_task
;
227 struct work_struct reap_task
;
228 struct work_struct skb_task
;
229 struct work_struct rx_reap_task
;
230 struct delayed_work stale_task
;
231 struct sk_buff_head skb_queue
;
232 struct list_head start_list
;
233 struct list_head reap_list
;
234 struct ib_wc ibwc
[IPOIB_NUM_WC
];
235 struct ib_sge rx_sge
[IPOIB_CM_RX_SG
];
236 struct ib_recv_wr rx_wr
;
240 * Device private locking: tx_lock protects members used in TX fast
241 * path (and we use LLTX so upper layers don't do extra locking).
242 * lock protects everything else. lock nests inside of tx_lock (ie
243 * tx_lock must be acquired first if needed).
245 struct ipoib_dev_priv
{
248 struct net_device
*dev
;
250 struct napi_struct napi
;
254 struct mutex mcast_mutex
;
255 struct mutex vlan_mutex
;
257 struct rb_root path_tree
;
258 struct list_head path_list
;
260 struct ipoib_mcast
*broadcast
;
261 struct list_head multicast_list
;
262 struct rb_root multicast_tree
;
264 struct delayed_work pkey_poll_task
;
265 struct delayed_work mcast_task
;
266 struct work_struct flush_task
;
267 struct work_struct restart_task
;
268 struct delayed_work ah_reap_task
;
269 struct work_struct pkey_event_task
;
271 struct ib_device
*ca
;
281 union ib_gid local_gid
;
284 unsigned int admin_mtu
;
285 unsigned int mcast_mtu
;
287 struct ipoib_rx_buf
*rx_ring
;
290 struct ipoib_tx_buf
*tx_ring
;
293 struct ib_sge tx_sge
;
294 struct ib_send_wr tx_wr
;
295 unsigned tx_outstanding
;
297 struct ib_wc ibwc
[IPOIB_NUM_WC
];
299 struct list_head dead_ahs
;
301 struct ib_event_handler event_handler
;
303 struct net_device
*parent
;
304 struct list_head child_intfs
;
305 struct list_head list
;
307 #ifdef CONFIG_INFINIBAND_IPOIB_CM
308 struct ipoib_cm_dev_priv cm
;
311 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
312 struct list_head fs_list
;
313 struct dentry
*mcg_dentry
;
314 struct dentry
*path_dentry
;
319 struct net_device
*dev
;
321 struct list_head list
;
327 struct net_device
*dev
;
328 struct ib_sa_path_rec pathrec
;
330 struct sk_buff_head queue
;
332 struct list_head neigh_list
;
335 struct ib_sa_query
*query
;
336 struct completion done
;
338 struct rb_node rb_node
;
339 struct list_head list
;
344 #ifdef CONFIG_INFINIBAND_IPOIB_CM
345 struct ipoib_cm_tx
*cm
;
348 struct sk_buff_head queue
;
350 struct neighbour
*neighbour
;
352 struct list_head list
;
356 * We stash a pointer to our private neighbour information after our
357 * hardware address in neigh->ha. The ALIGN() expression here makes
358 * sure that this pointer is stored aligned so that an unaligned
359 * load is not needed to dereference it.
361 static inline struct ipoib_neigh
**to_ipoib_neigh(struct neighbour
*neigh
)
363 return (void*) neigh
+ ALIGN(offsetof(struct neighbour
, ha
) +
364 INFINIBAND_ALEN
, sizeof(void *));
367 struct ipoib_neigh
*ipoib_neigh_alloc(struct neighbour
*neigh
);
368 void ipoib_neigh_free(struct net_device
*dev
, struct ipoib_neigh
*neigh
);
370 extern struct workqueue_struct
*ipoib_workqueue
;
374 int ipoib_poll(struct napi_struct
*napi
, int budget
);
375 void ipoib_ib_completion(struct ib_cq
*cq
, void *dev_ptr
);
377 struct ipoib_ah
*ipoib_create_ah(struct net_device
*dev
,
378 struct ib_pd
*pd
, struct ib_ah_attr
*attr
);
379 void ipoib_free_ah(struct kref
*kref
);
380 static inline void ipoib_put_ah(struct ipoib_ah
*ah
)
382 kref_put(&ah
->ref
, ipoib_free_ah
);
385 int ipoib_open(struct net_device
*dev
);
386 int ipoib_add_pkey_attr(struct net_device
*dev
);
387 int ipoib_add_umcast_attr(struct net_device
*dev
);
389 void ipoib_send(struct net_device
*dev
, struct sk_buff
*skb
,
390 struct ipoib_ah
*address
, u32 qpn
);
391 void ipoib_reap_ah(struct work_struct
*work
);
393 void ipoib_flush_paths(struct net_device
*dev
);
394 struct ipoib_dev_priv
*ipoib_intf_alloc(const char *format
);
396 int ipoib_ib_dev_init(struct net_device
*dev
, struct ib_device
*ca
, int port
);
397 void ipoib_ib_dev_flush(struct work_struct
*work
);
398 void ipoib_pkey_event(struct work_struct
*work
);
399 void ipoib_ib_dev_cleanup(struct net_device
*dev
);
401 int ipoib_ib_dev_open(struct net_device
*dev
);
402 int ipoib_ib_dev_up(struct net_device
*dev
);
403 int ipoib_ib_dev_down(struct net_device
*dev
, int flush
);
404 int ipoib_ib_dev_stop(struct net_device
*dev
, int flush
);
406 int ipoib_dev_init(struct net_device
*dev
, struct ib_device
*ca
, int port
);
407 void ipoib_dev_cleanup(struct net_device
*dev
);
409 void ipoib_mcast_join_task(struct work_struct
*work
);
410 void ipoib_mcast_send(struct net_device
*dev
, void *mgid
, struct sk_buff
*skb
);
412 void ipoib_mcast_restart_task(struct work_struct
*work
);
413 int ipoib_mcast_start_thread(struct net_device
*dev
);
414 int ipoib_mcast_stop_thread(struct net_device
*dev
, int flush
);
416 void ipoib_mcast_dev_down(struct net_device
*dev
);
417 void ipoib_mcast_dev_flush(struct net_device
*dev
);
419 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
420 struct ipoib_mcast_iter
*ipoib_mcast_iter_init(struct net_device
*dev
);
421 int ipoib_mcast_iter_next(struct ipoib_mcast_iter
*iter
);
422 void ipoib_mcast_iter_read(struct ipoib_mcast_iter
*iter
,
424 unsigned long *created
,
425 unsigned int *queuelen
,
426 unsigned int *complete
,
427 unsigned int *send_only
);
429 struct ipoib_path_iter
*ipoib_path_iter_init(struct net_device
*dev
);
430 int ipoib_path_iter_next(struct ipoib_path_iter
*iter
);
431 void ipoib_path_iter_read(struct ipoib_path_iter
*iter
,
432 struct ipoib_path
*path
);
435 int ipoib_mcast_attach(struct net_device
*dev
, u16 mlid
,
437 int ipoib_mcast_detach(struct net_device
*dev
, u16 mlid
,
440 int ipoib_init_qp(struct net_device
*dev
);
441 int ipoib_transport_dev_init(struct net_device
*dev
, struct ib_device
*ca
);
442 void ipoib_transport_dev_cleanup(struct net_device
*dev
);
444 void ipoib_event(struct ib_event_handler
*handler
,
445 struct ib_event
*record
);
447 int ipoib_vlan_add(struct net_device
*pdev
, unsigned short pkey
);
448 int ipoib_vlan_delete(struct net_device
*pdev
, unsigned short pkey
);
450 void ipoib_pkey_poll(struct work_struct
*work
);
451 int ipoib_pkey_dev_delay_open(struct net_device
*dev
);
452 void ipoib_drain_cq(struct net_device
*dev
);
454 #ifdef CONFIG_INFINIBAND_IPOIB_CM
456 #define IPOIB_FLAGS_RC 0x80
457 #define IPOIB_FLAGS_UC 0x40
459 /* We don't support UC connections at the moment */
460 #define IPOIB_CM_SUPPORTED(ha) (ha[0] & (IPOIB_FLAGS_RC))
462 static inline int ipoib_cm_admin_enabled(struct net_device
*dev
)
464 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
465 return IPOIB_CM_SUPPORTED(dev
->dev_addr
) &&
466 test_bit(IPOIB_FLAG_ADMIN_CM
, &priv
->flags
);
469 static inline int ipoib_cm_enabled(struct net_device
*dev
, struct neighbour
*n
)
471 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
472 return IPOIB_CM_SUPPORTED(n
->ha
) &&
473 test_bit(IPOIB_FLAG_ADMIN_CM
, &priv
->flags
);
476 static inline int ipoib_cm_up(struct ipoib_neigh
*neigh
)
479 return test_bit(IPOIB_FLAG_OPER_UP
, &neigh
->cm
->flags
);
482 static inline struct ipoib_cm_tx
*ipoib_cm_get(struct ipoib_neigh
*neigh
)
487 static inline void ipoib_cm_set(struct ipoib_neigh
*neigh
, struct ipoib_cm_tx
*tx
)
492 void ipoib_cm_send(struct net_device
*dev
, struct sk_buff
*skb
, struct ipoib_cm_tx
*tx
);
493 int ipoib_cm_dev_open(struct net_device
*dev
);
494 void ipoib_cm_dev_stop(struct net_device
*dev
);
495 int ipoib_cm_dev_init(struct net_device
*dev
);
496 int ipoib_cm_add_mode_attr(struct net_device
*dev
);
497 void ipoib_cm_dev_cleanup(struct net_device
*dev
);
498 struct ipoib_cm_tx
*ipoib_cm_create_tx(struct net_device
*dev
, struct ipoib_path
*path
,
499 struct ipoib_neigh
*neigh
);
500 void ipoib_cm_destroy_tx(struct ipoib_cm_tx
*tx
);
501 void ipoib_cm_skb_too_long(struct net_device
* dev
, struct sk_buff
*skb
,
503 void ipoib_cm_handle_rx_wc(struct net_device
*dev
, struct ib_wc
*wc
);
504 void ipoib_cm_handle_tx_wc(struct net_device
*dev
, struct ib_wc
*wc
);
509 static inline int ipoib_cm_admin_enabled(struct net_device
*dev
)
513 static inline int ipoib_cm_enabled(struct net_device
*dev
, struct neighbour
*n
)
519 static inline int ipoib_cm_up(struct ipoib_neigh
*neigh
)
525 static inline struct ipoib_cm_tx
*ipoib_cm_get(struct ipoib_neigh
*neigh
)
530 static inline void ipoib_cm_set(struct ipoib_neigh
*neigh
, struct ipoib_cm_tx
*tx
)
535 void ipoib_cm_send(struct net_device
*dev
, struct sk_buff
*skb
, struct ipoib_cm_tx
*tx
)
541 int ipoib_cm_dev_open(struct net_device
*dev
)
547 void ipoib_cm_dev_stop(struct net_device
*dev
)
553 int ipoib_cm_dev_init(struct net_device
*dev
)
559 void ipoib_cm_dev_cleanup(struct net_device
*dev
)
565 struct ipoib_cm_tx
*ipoib_cm_create_tx(struct net_device
*dev
, struct ipoib_path
*path
,
566 struct ipoib_neigh
*neigh
)
572 void ipoib_cm_destroy_tx(struct ipoib_cm_tx
*tx
)
578 int ipoib_cm_add_mode_attr(struct net_device
*dev
)
583 static inline void ipoib_cm_skb_too_long(struct net_device
* dev
, struct sk_buff
*skb
,
586 dev_kfree_skb_any(skb
);
589 static inline void ipoib_cm_handle_rx_wc(struct net_device
*dev
, struct ib_wc
*wc
)
593 static inline void ipoib_cm_handle_tx_wc(struct net_device
*dev
, struct ib_wc
*wc
)
598 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
599 void ipoib_create_debug_files(struct net_device
*dev
);
600 void ipoib_delete_debug_files(struct net_device
*dev
);
601 int ipoib_register_debugfs(void);
602 void ipoib_unregister_debugfs(void);
604 static inline void ipoib_create_debug_files(struct net_device
*dev
) { }
605 static inline void ipoib_delete_debug_files(struct net_device
*dev
) { }
606 static inline int ipoib_register_debugfs(void) { return 0; }
607 static inline void ipoib_unregister_debugfs(void) { }
611 #define ipoib_printk(level, priv, format, arg...) \
612 printk(level "%s: " format, ((struct ipoib_dev_priv *) priv)->dev->name , ## arg)
613 #define ipoib_warn(priv, format, arg...) \
614 ipoib_printk(KERN_WARNING, priv, format , ## arg)
616 extern int ipoib_sendq_size
;
617 extern int ipoib_recvq_size
;
619 extern struct ib_sa_client ipoib_sa_client
;
621 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
622 extern int ipoib_debug_level
;
624 #define ipoib_dbg(priv, format, arg...) \
626 if (ipoib_debug_level > 0) \
627 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
629 #define ipoib_dbg_mcast(priv, format, arg...) \
631 if (mcast_debug_level > 0) \
632 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
634 #else /* CONFIG_INFINIBAND_IPOIB_DEBUG */
635 #define ipoib_dbg(priv, format, arg...) \
636 do { (void) (priv); } while (0)
637 #define ipoib_dbg_mcast(priv, format, arg...) \
638 do { (void) (priv); } while (0)
639 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */
641 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG_DATA
642 #define ipoib_dbg_data(priv, format, arg...) \
644 if (data_debug_level > 0) \
645 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
647 #else /* CONFIG_INFINIBAND_IPOIB_DEBUG_DATA */
648 #define ipoib_dbg_data(priv, format, arg...) \
649 do { (void) (priv); } while (0)
650 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG_DATA */
653 #define IPOIB_GID_FMT "%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x:" \
654 "%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x"
656 #define IPOIB_GID_RAW_ARG(gid) ((u8 *)(gid))[0], \
673 #define IPOIB_GID_ARG(gid) IPOIB_GID_RAW_ARG((gid).raw)
675 #define IPOIB_QPN(ha) (be32_to_cpup((__be32 *) ha) & 0xffffff)
677 #endif /* _IPOIB_H */