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>
61 IPOIB_UD_HEAD_SIZE
= IB_GRH_BYTES
+ IPOIB_ENCAP_LEN
,
62 IPOIB_UD_RX_SG
= 2, /* max buffer needed for 4K mtu */
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,
72 IPOIB_CM_MAX_CONN_QP
= 4096,
76 IPOIB_MAX_PATH_REC_QUEUE
= 3,
77 IPOIB_MAX_MCAST_QUEUE
= 3,
79 IPOIB_FLAG_OPER_UP
= 0,
80 IPOIB_FLAG_INITIALIZED
= 1,
81 IPOIB_FLAG_ADMIN_UP
= 2,
82 IPOIB_PKEY_ASSIGNED
= 3,
84 IPOIB_FLAG_SUBINTERFACE
= 5,
86 IPOIB_STOP_REAPER
= 7,
87 IPOIB_MCAST_STARTED
= 8,
88 IPOIB_FLAG_ADMIN_CM
= 9,
89 IPOIB_FLAG_UMCAST
= 10,
92 IPOIB_MAX_BACKOFF_SECONDS
= 16,
94 IPOIB_MCAST_FLAG_FOUND
= 0, /* used in set_multicast_list */
95 IPOIB_MCAST_FLAG_SENDONLY
= 1,
96 IPOIB_MCAST_FLAG_BUSY
= 2, /* joining or already joined */
97 IPOIB_MCAST_FLAG_ATTACHED
= 3,
102 #define IPOIB_OP_RECV (1ul << 31)
103 #ifdef CONFIG_INFINIBAND_IPOIB_CM
104 #define IPOIB_OP_CM (1ul << 30)
106 #define IPOIB_OP_CM (0)
111 struct ipoib_header
{
116 struct ipoib_pseudoheader
{
117 u8 hwaddr
[INFINIBAND_ALEN
];
120 /* Used for all multicast joins (broadcast, IPv4 mcast and IPv6 mcast) */
122 struct ib_sa_mcmember_rec mcmember
;
123 struct ib_sa_multicast
*mc
;
126 struct rb_node rb_node
;
127 struct list_head list
;
129 unsigned long created
;
130 unsigned long backoff
;
133 unsigned char logcount
;
135 struct list_head neigh_list
;
137 struct sk_buff_head pkt_queue
;
139 struct net_device
*dev
;
142 struct ipoib_rx_buf
{
144 u64 mapping
[IPOIB_UD_RX_SG
];
147 struct ipoib_tx_buf
{
149 u64 mapping
[MAX_SKB_FRAGS
+ 1];
154 struct ipoib_cm_data
{
155 __be32 qpn
; /* High byte MUST be ignored on receive */
160 * Quoting 10.3.1 Queue Pair and EE Context States:
162 * Note, for QPs that are associated with an SRQ, the Consumer should take the
163 * QP through the Error State before invoking a Destroy QP or a Modify QP to the
164 * Reset State. The Consumer may invoke the Destroy QP without first performing
165 * a Modify QP to the Error State and waiting for the Affiliated Asynchronous
166 * Last WQE Reached Event. However, if the Consumer does not wait for the
167 * Affiliated Asynchronous Last WQE Reached Event, then WQE and Data Segment
168 * leakage may occur. Therefore, it is good programming practice to tear down a
169 * QP that is associated with an SRQ by using the following process:
171 * - Put the QP in the Error State
172 * - Wait for the Affiliated Asynchronous Last WQE Reached Event;
174 * drain the CQ by invoking the Poll CQ verb and either wait for CQ
175 * to be empty or the number of Poll CQ operations has exceeded
178 * post another WR that completes on the same CQ and wait for this
179 * WR to return as a WC;
180 * - and then invoke a Destroy QP or Reset QP.
182 * We use the second option and wait for a completion on the
183 * same CQ before destroying QPs attached to our SRQ.
186 enum ipoib_cm_state
{
188 IPOIB_CM_RX_ERROR
, /* Ignored by stale task */
189 IPOIB_CM_RX_FLUSH
/* Last WQE Reached event observed */
195 struct ipoib_cm_rx_buf
*rx_ring
;
196 struct list_head list
;
197 struct net_device
*dev
;
198 unsigned long jiffies
;
199 enum ipoib_cm_state state
;
206 struct list_head list
;
207 struct net_device
*dev
;
208 struct ipoib_neigh
*neigh
;
209 struct ipoib_path
*path
;
210 struct ipoib_tx_buf
*tx_ring
;
217 struct ipoib_cm_rx_buf
{
219 u64 mapping
[IPOIB_CM_RX_SG
];
222 struct ipoib_cm_dev_priv
{
224 struct ipoib_cm_rx_buf
*srq_ring
;
226 struct list_head passive_ids
; /* state: LIVE */
227 struct list_head rx_error_list
; /* state: ERROR */
228 struct list_head rx_flush_list
; /* state: FLUSH, drain not started */
229 struct list_head rx_drain_list
; /* state: FLUSH, drain started */
230 struct list_head rx_reap_list
; /* state: FLUSH, drain done */
231 struct work_struct start_task
;
232 struct work_struct reap_task
;
233 struct work_struct skb_task
;
234 struct work_struct rx_reap_task
;
235 struct delayed_work stale_task
;
236 struct sk_buff_head skb_queue
;
237 struct list_head start_list
;
238 struct list_head reap_list
;
239 struct ib_wc ibwc
[IPOIB_NUM_WC
];
240 struct ib_sge rx_sge
[IPOIB_CM_RX_SG
];
241 struct ib_recv_wr rx_wr
;
247 struct ipoib_ethtool_st
{
249 u16 max_coalesced_frames
;
253 * Device private locking: tx_lock protects members used in TX fast
254 * path (and we use LLTX so upper layers don't do extra locking).
255 * lock protects everything else. lock nests inside of tx_lock (ie
256 * tx_lock must be acquired first if needed).
258 struct ipoib_dev_priv
{
261 struct net_device
*dev
;
263 struct napi_struct napi
;
267 struct mutex mcast_mutex
;
268 struct mutex vlan_mutex
;
270 struct rb_root path_tree
;
271 struct list_head path_list
;
273 struct ipoib_mcast
*broadcast
;
274 struct list_head multicast_list
;
275 struct rb_root multicast_tree
;
277 struct delayed_work pkey_poll_task
;
278 struct delayed_work mcast_task
;
279 struct work_struct flush_task
;
280 struct work_struct restart_task
;
281 struct delayed_work ah_reap_task
;
282 struct work_struct pkey_event_task
;
284 struct ib_device
*ca
;
290 struct ib_cq
*recv_cq
;
291 struct ib_cq
*send_cq
;
295 union ib_gid local_gid
;
298 unsigned int admin_mtu
;
299 unsigned int mcast_mtu
;
300 unsigned int max_ib_mtu
;
302 struct ipoib_rx_buf
*rx_ring
;
305 struct ipoib_tx_buf
*tx_ring
;
308 struct ib_sge tx_sge
[MAX_SKB_FRAGS
+ 1];
309 struct ib_send_wr tx_wr
;
310 unsigned tx_outstanding
;
311 struct ib_wc send_wc
[MAX_SEND_CQE
];
313 struct ib_recv_wr rx_wr
;
314 struct ib_sge rx_sge
[IPOIB_UD_RX_SG
];
316 struct ib_wc ibwc
[IPOIB_NUM_WC
];
318 struct list_head dead_ahs
;
320 struct ib_event_handler event_handler
;
322 struct net_device
*parent
;
323 struct list_head child_intfs
;
324 struct list_head list
;
326 #ifdef CONFIG_INFINIBAND_IPOIB_CM
327 struct ipoib_cm_dev_priv cm
;
330 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
331 struct list_head fs_list
;
332 struct dentry
*mcg_dentry
;
333 struct dentry
*path_dentry
;
336 struct ipoib_ethtool_st ethtool
;
337 struct timer_list poll_timer
;
341 struct net_device
*dev
;
343 struct list_head list
;
349 struct net_device
*dev
;
350 struct ib_sa_path_rec pathrec
;
352 struct sk_buff_head queue
;
354 struct list_head neigh_list
;
357 struct ib_sa_query
*query
;
358 struct completion done
;
360 struct rb_node rb_node
;
361 struct list_head list
;
366 #ifdef CONFIG_INFINIBAND_IPOIB_CM
367 struct ipoib_cm_tx
*cm
;
370 struct sk_buff_head queue
;
372 struct neighbour
*neighbour
;
373 struct net_device
*dev
;
375 struct list_head list
;
378 #define IPOIB_UD_MTU(ib_mtu) (ib_mtu - IPOIB_ENCAP_LEN)
379 #define IPOIB_UD_BUF_SIZE(ib_mtu) (ib_mtu + IB_GRH_BYTES)
381 static inline int ipoib_ud_need_sg(unsigned int ib_mtu
)
383 return IPOIB_UD_BUF_SIZE(ib_mtu
) > PAGE_SIZE
;
387 * We stash a pointer to our private neighbour information after our
388 * hardware address in neigh->ha. The ALIGN() expression here makes
389 * sure that this pointer is stored aligned so that an unaligned
390 * load is not needed to dereference it.
392 static inline struct ipoib_neigh
**to_ipoib_neigh(struct neighbour
*neigh
)
394 return (void*) neigh
+ ALIGN(offsetof(struct neighbour
, ha
) +
395 INFINIBAND_ALEN
, sizeof(void *));
398 struct ipoib_neigh
*ipoib_neigh_alloc(struct neighbour
*neigh
,
399 struct net_device
*dev
);
400 void ipoib_neigh_free(struct net_device
*dev
, struct ipoib_neigh
*neigh
);
402 extern struct workqueue_struct
*ipoib_workqueue
;
406 int ipoib_poll(struct napi_struct
*napi
, int budget
);
407 void ipoib_ib_completion(struct ib_cq
*cq
, void *dev_ptr
);
408 void ipoib_send_comp_handler(struct ib_cq
*cq
, void *dev_ptr
);
410 struct ipoib_ah
*ipoib_create_ah(struct net_device
*dev
,
411 struct ib_pd
*pd
, struct ib_ah_attr
*attr
);
412 void ipoib_free_ah(struct kref
*kref
);
413 static inline void ipoib_put_ah(struct ipoib_ah
*ah
)
415 kref_put(&ah
->ref
, ipoib_free_ah
);
418 int ipoib_open(struct net_device
*dev
);
419 int ipoib_add_pkey_attr(struct net_device
*dev
);
420 int ipoib_add_umcast_attr(struct net_device
*dev
);
422 void ipoib_send(struct net_device
*dev
, struct sk_buff
*skb
,
423 struct ipoib_ah
*address
, u32 qpn
);
424 void ipoib_reap_ah(struct work_struct
*work
);
426 void ipoib_flush_paths(struct net_device
*dev
);
427 struct ipoib_dev_priv
*ipoib_intf_alloc(const char *format
);
429 int ipoib_ib_dev_init(struct net_device
*dev
, struct ib_device
*ca
, int port
);
430 void ipoib_ib_dev_flush(struct work_struct
*work
);
431 void ipoib_pkey_event(struct work_struct
*work
);
432 void ipoib_ib_dev_cleanup(struct net_device
*dev
);
434 int ipoib_ib_dev_open(struct net_device
*dev
);
435 int ipoib_ib_dev_up(struct net_device
*dev
);
436 int ipoib_ib_dev_down(struct net_device
*dev
, int flush
);
437 int ipoib_ib_dev_stop(struct net_device
*dev
, int flush
);
439 int ipoib_dev_init(struct net_device
*dev
, struct ib_device
*ca
, int port
);
440 void ipoib_dev_cleanup(struct net_device
*dev
);
442 void ipoib_mcast_join_task(struct work_struct
*work
);
443 void ipoib_mcast_send(struct net_device
*dev
, void *mgid
, struct sk_buff
*skb
);
445 void ipoib_mcast_restart_task(struct work_struct
*work
);
446 int ipoib_mcast_start_thread(struct net_device
*dev
);
447 int ipoib_mcast_stop_thread(struct net_device
*dev
, int flush
);
449 void ipoib_mcast_dev_down(struct net_device
*dev
);
450 void ipoib_mcast_dev_flush(struct net_device
*dev
);
452 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
453 struct ipoib_mcast_iter
*ipoib_mcast_iter_init(struct net_device
*dev
);
454 int ipoib_mcast_iter_next(struct ipoib_mcast_iter
*iter
);
455 void ipoib_mcast_iter_read(struct ipoib_mcast_iter
*iter
,
457 unsigned long *created
,
458 unsigned int *queuelen
,
459 unsigned int *complete
,
460 unsigned int *send_only
);
462 struct ipoib_path_iter
*ipoib_path_iter_init(struct net_device
*dev
);
463 int ipoib_path_iter_next(struct ipoib_path_iter
*iter
);
464 void ipoib_path_iter_read(struct ipoib_path_iter
*iter
,
465 struct ipoib_path
*path
);
468 int ipoib_mcast_attach(struct net_device
*dev
, u16 mlid
,
470 int ipoib_mcast_detach(struct net_device
*dev
, u16 mlid
,
473 int ipoib_init_qp(struct net_device
*dev
);
474 int ipoib_transport_dev_init(struct net_device
*dev
, struct ib_device
*ca
);
475 void ipoib_transport_dev_cleanup(struct net_device
*dev
);
477 void ipoib_event(struct ib_event_handler
*handler
,
478 struct ib_event
*record
);
480 int ipoib_vlan_add(struct net_device
*pdev
, unsigned short pkey
);
481 int ipoib_vlan_delete(struct net_device
*pdev
, unsigned short pkey
);
483 void ipoib_pkey_poll(struct work_struct
*work
);
484 int ipoib_pkey_dev_delay_open(struct net_device
*dev
);
485 void ipoib_drain_cq(struct net_device
*dev
);
487 void ipoib_set_ethtool_ops(struct net_device
*dev
);
489 #ifdef CONFIG_INFINIBAND_IPOIB_CM
491 #define IPOIB_FLAGS_RC 0x80
492 #define IPOIB_FLAGS_UC 0x40
494 /* We don't support UC connections at the moment */
495 #define IPOIB_CM_SUPPORTED(ha) (ha[0] & (IPOIB_FLAGS_RC))
497 extern int ipoib_max_conn_qp
;
499 static inline int ipoib_cm_admin_enabled(struct net_device
*dev
)
501 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
502 return IPOIB_CM_SUPPORTED(dev
->dev_addr
) &&
503 test_bit(IPOIB_FLAG_ADMIN_CM
, &priv
->flags
);
506 static inline int ipoib_cm_enabled(struct net_device
*dev
, struct neighbour
*n
)
508 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
509 return IPOIB_CM_SUPPORTED(n
->ha
) &&
510 test_bit(IPOIB_FLAG_ADMIN_CM
, &priv
->flags
);
513 static inline int ipoib_cm_up(struct ipoib_neigh
*neigh
)
516 return test_bit(IPOIB_FLAG_OPER_UP
, &neigh
->cm
->flags
);
519 static inline struct ipoib_cm_tx
*ipoib_cm_get(struct ipoib_neigh
*neigh
)
524 static inline void ipoib_cm_set(struct ipoib_neigh
*neigh
, struct ipoib_cm_tx
*tx
)
529 static inline int ipoib_cm_has_srq(struct net_device
*dev
)
531 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
532 return !!priv
->cm
.srq
;
535 static inline unsigned int ipoib_cm_max_mtu(struct net_device
*dev
)
537 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
538 return priv
->cm
.max_cm_mtu
;
541 void ipoib_cm_send(struct net_device
*dev
, struct sk_buff
*skb
, struct ipoib_cm_tx
*tx
);
542 int ipoib_cm_dev_open(struct net_device
*dev
);
543 void ipoib_cm_dev_stop(struct net_device
*dev
);
544 int ipoib_cm_dev_init(struct net_device
*dev
);
545 int ipoib_cm_add_mode_attr(struct net_device
*dev
);
546 void ipoib_cm_dev_cleanup(struct net_device
*dev
);
547 struct ipoib_cm_tx
*ipoib_cm_create_tx(struct net_device
*dev
, struct ipoib_path
*path
,
548 struct ipoib_neigh
*neigh
);
549 void ipoib_cm_destroy_tx(struct ipoib_cm_tx
*tx
);
550 void ipoib_cm_skb_too_long(struct net_device
*dev
, struct sk_buff
*skb
,
552 void ipoib_cm_handle_rx_wc(struct net_device
*dev
, struct ib_wc
*wc
);
553 void ipoib_cm_handle_tx_wc(struct net_device
*dev
, struct ib_wc
*wc
);
558 #define ipoib_max_conn_qp 0
560 static inline int ipoib_cm_admin_enabled(struct net_device
*dev
)
564 static inline int ipoib_cm_enabled(struct net_device
*dev
, struct neighbour
*n
)
570 static inline int ipoib_cm_up(struct ipoib_neigh
*neigh
)
576 static inline struct ipoib_cm_tx
*ipoib_cm_get(struct ipoib_neigh
*neigh
)
581 static inline void ipoib_cm_set(struct ipoib_neigh
*neigh
, struct ipoib_cm_tx
*tx
)
585 static inline int ipoib_cm_has_srq(struct net_device
*dev
)
590 static inline unsigned int ipoib_cm_max_mtu(struct net_device
*dev
)
596 void ipoib_cm_send(struct net_device
*dev
, struct sk_buff
*skb
, struct ipoib_cm_tx
*tx
)
602 int ipoib_cm_dev_open(struct net_device
*dev
)
608 void ipoib_cm_dev_stop(struct net_device
*dev
)
614 int ipoib_cm_dev_init(struct net_device
*dev
)
620 void ipoib_cm_dev_cleanup(struct net_device
*dev
)
626 struct ipoib_cm_tx
*ipoib_cm_create_tx(struct net_device
*dev
, struct ipoib_path
*path
,
627 struct ipoib_neigh
*neigh
)
633 void ipoib_cm_destroy_tx(struct ipoib_cm_tx
*tx
)
639 int ipoib_cm_add_mode_attr(struct net_device
*dev
)
644 static inline void ipoib_cm_skb_too_long(struct net_device
*dev
, struct sk_buff
*skb
,
647 dev_kfree_skb_any(skb
);
650 static inline void ipoib_cm_handle_rx_wc(struct net_device
*dev
, struct ib_wc
*wc
)
654 static inline void ipoib_cm_handle_tx_wc(struct net_device
*dev
, struct ib_wc
*wc
)
659 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
660 void ipoib_create_debug_files(struct net_device
*dev
);
661 void ipoib_delete_debug_files(struct net_device
*dev
);
662 int ipoib_register_debugfs(void);
663 void ipoib_unregister_debugfs(void);
665 static inline void ipoib_create_debug_files(struct net_device
*dev
) { }
666 static inline void ipoib_delete_debug_files(struct net_device
*dev
) { }
667 static inline int ipoib_register_debugfs(void) { return 0; }
668 static inline void ipoib_unregister_debugfs(void) { }
671 #define ipoib_printk(level, priv, format, arg...) \
672 printk(level "%s: " format, ((struct ipoib_dev_priv *) priv)->dev->name , ## arg)
673 #define ipoib_warn(priv, format, arg...) \
674 ipoib_printk(KERN_WARNING, priv, format , ## arg)
676 extern int ipoib_sendq_size
;
677 extern int ipoib_recvq_size
;
679 extern struct ib_sa_client ipoib_sa_client
;
681 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
682 extern int ipoib_debug_level
;
684 #define ipoib_dbg(priv, format, arg...) \
686 if (ipoib_debug_level > 0) \
687 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
689 #define ipoib_dbg_mcast(priv, format, arg...) \
691 if (mcast_debug_level > 0) \
692 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
694 #else /* CONFIG_INFINIBAND_IPOIB_DEBUG */
695 #define ipoib_dbg(priv, format, arg...) \
696 do { (void) (priv); } while (0)
697 #define ipoib_dbg_mcast(priv, format, arg...) \
698 do { (void) (priv); } while (0)
699 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */
701 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG_DATA
702 #define ipoib_dbg_data(priv, format, arg...) \
704 if (data_debug_level > 0) \
705 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
707 #else /* CONFIG_INFINIBAND_IPOIB_DEBUG_DATA */
708 #define ipoib_dbg_data(priv, format, arg...) \
709 do { (void) (priv); } while (0)
710 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG_DATA */
713 #define IPOIB_GID_FMT "%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x:" \
714 "%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x"
716 #define IPOIB_GID_RAW_ARG(gid) ((u8 *)(gid))[0], \
733 #define IPOIB_GID_ARG(gid) IPOIB_GID_RAW_ARG((gid).raw)
735 #define IPOIB_QPN(ha) (be32_to_cpup((__be32 *) ha) & 0xffffff)
737 #endif /* _IPOIB_H */