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,
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,
91 IPOIB_MAX_BACKOFF_SECONDS
= 16,
93 IPOIB_MCAST_FLAG_FOUND
= 0, /* used in set_multicast_list */
94 IPOIB_MCAST_FLAG_SENDONLY
= 1,
95 IPOIB_MCAST_FLAG_BUSY
= 2, /* joining or already joined */
96 IPOIB_MCAST_FLAG_ATTACHED
= 3,
99 #define IPOIB_OP_RECV (1ul << 31)
100 #ifdef CONFIG_INFINIBAND_IPOIB_CM
101 #define IPOIB_OP_CM (1ul << 30)
103 #define IPOIB_OP_CM (0)
108 struct ipoib_header
{
113 struct ipoib_pseudoheader
{
114 u8 hwaddr
[INFINIBAND_ALEN
];
117 /* Used for all multicast joins (broadcast, IPv4 mcast and IPv6 mcast) */
119 struct ib_sa_mcmember_rec mcmember
;
120 struct ib_sa_multicast
*mc
;
123 struct rb_node rb_node
;
124 struct list_head list
;
126 unsigned long created
;
127 unsigned long backoff
;
130 unsigned char logcount
;
132 struct list_head neigh_list
;
134 struct sk_buff_head pkt_queue
;
136 struct net_device
*dev
;
139 struct ipoib_rx_buf
{
144 struct ipoib_tx_buf
{
146 u64 mapping
[MAX_SKB_FRAGS
+ 1];
151 struct ipoib_cm_data
{
152 __be32 qpn
; /* High byte MUST be ignored on receive */
157 * Quoting 10.3.1 Queue Pair and EE Context States:
159 * Note, for QPs that are associated with an SRQ, the Consumer should take the
160 * QP through the Error State before invoking a Destroy QP or a Modify QP to the
161 * Reset State. The Consumer may invoke the Destroy QP without first performing
162 * a Modify QP to the Error State and waiting for the Affiliated Asynchronous
163 * Last WQE Reached Event. However, if the Consumer does not wait for the
164 * Affiliated Asynchronous Last WQE Reached Event, then WQE and Data Segment
165 * leakage may occur. Therefore, it is good programming practice to tear down a
166 * QP that is associated with an SRQ by using the following process:
168 * - Put the QP in the Error State
169 * - Wait for the Affiliated Asynchronous Last WQE Reached Event;
171 * drain the CQ by invoking the Poll CQ verb and either wait for CQ
172 * to be empty or the number of Poll CQ operations has exceeded
175 * post another WR that completes on the same CQ and wait for this
176 * WR to return as a WC;
177 * - and then invoke a Destroy QP or Reset QP.
179 * We use the second option and wait for a completion on the
180 * same CQ before destroying QPs attached to our SRQ.
183 enum ipoib_cm_state
{
185 IPOIB_CM_RX_ERROR
, /* Ignored by stale task */
186 IPOIB_CM_RX_FLUSH
/* Last WQE Reached event observed */
192 struct ipoib_cm_rx_buf
*rx_ring
;
193 struct list_head list
;
194 struct net_device
*dev
;
195 unsigned long jiffies
;
196 enum ipoib_cm_state state
;
203 struct list_head list
;
204 struct net_device
*dev
;
205 struct ipoib_neigh
*neigh
;
206 struct ipoib_path
*path
;
207 struct ipoib_tx_buf
*tx_ring
;
212 struct ib_wc ibwc
[IPOIB_NUM_WC
];
215 struct ipoib_cm_rx_buf
{
217 u64 mapping
[IPOIB_CM_RX_SG
];
220 struct ipoib_cm_dev_priv
{
222 struct ipoib_cm_rx_buf
*srq_ring
;
224 struct list_head passive_ids
; /* state: LIVE */
225 struct list_head rx_error_list
; /* state: ERROR */
226 struct list_head rx_flush_list
; /* state: FLUSH, drain not started */
227 struct list_head rx_drain_list
; /* state: FLUSH, drain started */
228 struct list_head rx_reap_list
; /* state: FLUSH, drain done */
229 struct work_struct start_task
;
230 struct work_struct reap_task
;
231 struct work_struct skb_task
;
232 struct work_struct rx_reap_task
;
233 struct delayed_work stale_task
;
234 struct sk_buff_head skb_queue
;
235 struct list_head start_list
;
236 struct list_head reap_list
;
237 struct ib_wc ibwc
[IPOIB_NUM_WC
];
238 struct ib_sge rx_sge
[IPOIB_CM_RX_SG
];
239 struct ib_recv_wr rx_wr
;
246 * Device private locking: tx_lock protects members used in TX fast
247 * path (and we use LLTX so upper layers don't do extra locking).
248 * lock protects everything else. lock nests inside of tx_lock (ie
249 * tx_lock must be acquired first if needed).
251 struct ipoib_dev_priv
{
254 struct net_device
*dev
;
256 struct napi_struct napi
;
260 struct mutex mcast_mutex
;
261 struct mutex vlan_mutex
;
263 struct rb_root path_tree
;
264 struct list_head path_list
;
266 struct ipoib_mcast
*broadcast
;
267 struct list_head multicast_list
;
268 struct rb_root multicast_tree
;
270 struct delayed_work pkey_poll_task
;
271 struct delayed_work mcast_task
;
272 struct work_struct flush_task
;
273 struct work_struct restart_task
;
274 struct delayed_work ah_reap_task
;
275 struct work_struct pkey_event_task
;
277 struct ib_device
*ca
;
287 union ib_gid local_gid
;
290 unsigned int admin_mtu
;
291 unsigned int mcast_mtu
;
293 struct ipoib_rx_buf
*rx_ring
;
296 struct ipoib_tx_buf
*tx_ring
;
299 struct ib_sge tx_sge
[MAX_SKB_FRAGS
+ 1];
300 struct ib_send_wr tx_wr
;
301 unsigned tx_outstanding
;
303 struct ib_wc ibwc
[IPOIB_NUM_WC
];
305 struct list_head dead_ahs
;
307 struct ib_event_handler event_handler
;
309 struct net_device
*parent
;
310 struct list_head child_intfs
;
311 struct list_head list
;
313 #ifdef CONFIG_INFINIBAND_IPOIB_CM
314 struct ipoib_cm_dev_priv cm
;
317 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
318 struct list_head fs_list
;
319 struct dentry
*mcg_dentry
;
320 struct dentry
*path_dentry
;
325 struct net_device
*dev
;
327 struct list_head list
;
333 struct net_device
*dev
;
334 struct ib_sa_path_rec pathrec
;
336 struct sk_buff_head queue
;
338 struct list_head neigh_list
;
341 struct ib_sa_query
*query
;
342 struct completion done
;
344 struct rb_node rb_node
;
345 struct list_head list
;
350 #ifdef CONFIG_INFINIBAND_IPOIB_CM
351 struct ipoib_cm_tx
*cm
;
354 struct sk_buff_head queue
;
356 struct neighbour
*neighbour
;
357 struct net_device
*dev
;
359 struct list_head list
;
363 * We stash a pointer to our private neighbour information after our
364 * hardware address in neigh->ha. The ALIGN() expression here makes
365 * sure that this pointer is stored aligned so that an unaligned
366 * load is not needed to dereference it.
368 static inline struct ipoib_neigh
**to_ipoib_neigh(struct neighbour
*neigh
)
370 return (void*) neigh
+ ALIGN(offsetof(struct neighbour
, ha
) +
371 INFINIBAND_ALEN
, sizeof(void *));
374 struct ipoib_neigh
*ipoib_neigh_alloc(struct neighbour
*neigh
,
375 struct net_device
*dev
);
376 void ipoib_neigh_free(struct net_device
*dev
, struct ipoib_neigh
*neigh
);
378 extern struct workqueue_struct
*ipoib_workqueue
;
382 int ipoib_poll(struct napi_struct
*napi
, int budget
);
383 void ipoib_ib_completion(struct ib_cq
*cq
, void *dev_ptr
);
385 struct ipoib_ah
*ipoib_create_ah(struct net_device
*dev
,
386 struct ib_pd
*pd
, struct ib_ah_attr
*attr
);
387 void ipoib_free_ah(struct kref
*kref
);
388 static inline void ipoib_put_ah(struct ipoib_ah
*ah
)
390 kref_put(&ah
->ref
, ipoib_free_ah
);
393 int ipoib_open(struct net_device
*dev
);
394 int ipoib_add_pkey_attr(struct net_device
*dev
);
395 int ipoib_add_umcast_attr(struct net_device
*dev
);
397 void ipoib_send(struct net_device
*dev
, struct sk_buff
*skb
,
398 struct ipoib_ah
*address
, u32 qpn
);
399 void ipoib_reap_ah(struct work_struct
*work
);
401 void ipoib_flush_paths(struct net_device
*dev
);
402 struct ipoib_dev_priv
*ipoib_intf_alloc(const char *format
);
404 int ipoib_ib_dev_init(struct net_device
*dev
, struct ib_device
*ca
, int port
);
405 void ipoib_ib_dev_flush(struct work_struct
*work
);
406 void ipoib_pkey_event(struct work_struct
*work
);
407 void ipoib_ib_dev_cleanup(struct net_device
*dev
);
409 int ipoib_ib_dev_open(struct net_device
*dev
);
410 int ipoib_ib_dev_up(struct net_device
*dev
);
411 int ipoib_ib_dev_down(struct net_device
*dev
, int flush
);
412 int ipoib_ib_dev_stop(struct net_device
*dev
, int flush
);
414 int ipoib_dev_init(struct net_device
*dev
, struct ib_device
*ca
, int port
);
415 void ipoib_dev_cleanup(struct net_device
*dev
);
417 void ipoib_mcast_join_task(struct work_struct
*work
);
418 void ipoib_mcast_send(struct net_device
*dev
, void *mgid
, struct sk_buff
*skb
);
420 void ipoib_mcast_restart_task(struct work_struct
*work
);
421 int ipoib_mcast_start_thread(struct net_device
*dev
);
422 int ipoib_mcast_stop_thread(struct net_device
*dev
, int flush
);
424 void ipoib_mcast_dev_down(struct net_device
*dev
);
425 void ipoib_mcast_dev_flush(struct net_device
*dev
);
427 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
428 struct ipoib_mcast_iter
*ipoib_mcast_iter_init(struct net_device
*dev
);
429 int ipoib_mcast_iter_next(struct ipoib_mcast_iter
*iter
);
430 void ipoib_mcast_iter_read(struct ipoib_mcast_iter
*iter
,
432 unsigned long *created
,
433 unsigned int *queuelen
,
434 unsigned int *complete
,
435 unsigned int *send_only
);
437 struct ipoib_path_iter
*ipoib_path_iter_init(struct net_device
*dev
);
438 int ipoib_path_iter_next(struct ipoib_path_iter
*iter
);
439 void ipoib_path_iter_read(struct ipoib_path_iter
*iter
,
440 struct ipoib_path
*path
);
443 int ipoib_mcast_attach(struct net_device
*dev
, u16 mlid
,
445 int ipoib_mcast_detach(struct net_device
*dev
, u16 mlid
,
448 int ipoib_init_qp(struct net_device
*dev
);
449 int ipoib_transport_dev_init(struct net_device
*dev
, struct ib_device
*ca
);
450 void ipoib_transport_dev_cleanup(struct net_device
*dev
);
452 void ipoib_event(struct ib_event_handler
*handler
,
453 struct ib_event
*record
);
455 int ipoib_vlan_add(struct net_device
*pdev
, unsigned short pkey
);
456 int ipoib_vlan_delete(struct net_device
*pdev
, unsigned short pkey
);
458 void ipoib_pkey_poll(struct work_struct
*work
);
459 int ipoib_pkey_dev_delay_open(struct net_device
*dev
);
460 void ipoib_drain_cq(struct net_device
*dev
);
462 #ifdef CONFIG_INFINIBAND_IPOIB_CM
464 #define IPOIB_FLAGS_RC 0x80
465 #define IPOIB_FLAGS_UC 0x40
467 /* We don't support UC connections at the moment */
468 #define IPOIB_CM_SUPPORTED(ha) (ha[0] & (IPOIB_FLAGS_RC))
470 extern int ipoib_max_conn_qp
;
472 static inline int ipoib_cm_admin_enabled(struct net_device
*dev
)
474 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
475 return IPOIB_CM_SUPPORTED(dev
->dev_addr
) &&
476 test_bit(IPOIB_FLAG_ADMIN_CM
, &priv
->flags
);
479 static inline int ipoib_cm_enabled(struct net_device
*dev
, struct neighbour
*n
)
481 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
482 return IPOIB_CM_SUPPORTED(n
->ha
) &&
483 test_bit(IPOIB_FLAG_ADMIN_CM
, &priv
->flags
);
486 static inline int ipoib_cm_up(struct ipoib_neigh
*neigh
)
489 return test_bit(IPOIB_FLAG_OPER_UP
, &neigh
->cm
->flags
);
492 static inline struct ipoib_cm_tx
*ipoib_cm_get(struct ipoib_neigh
*neigh
)
497 static inline void ipoib_cm_set(struct ipoib_neigh
*neigh
, struct ipoib_cm_tx
*tx
)
502 static inline int ipoib_cm_has_srq(struct net_device
*dev
)
504 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
505 return !!priv
->cm
.srq
;
508 static inline unsigned int ipoib_cm_max_mtu(struct net_device
*dev
)
510 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
511 return priv
->cm
.max_cm_mtu
;
514 void ipoib_cm_send(struct net_device
*dev
, struct sk_buff
*skb
, struct ipoib_cm_tx
*tx
);
515 int ipoib_cm_dev_open(struct net_device
*dev
);
516 void ipoib_cm_dev_stop(struct net_device
*dev
);
517 int ipoib_cm_dev_init(struct net_device
*dev
);
518 int ipoib_cm_add_mode_attr(struct net_device
*dev
);
519 void ipoib_cm_dev_cleanup(struct net_device
*dev
);
520 struct ipoib_cm_tx
*ipoib_cm_create_tx(struct net_device
*dev
, struct ipoib_path
*path
,
521 struct ipoib_neigh
*neigh
);
522 void ipoib_cm_destroy_tx(struct ipoib_cm_tx
*tx
);
523 void ipoib_cm_skb_too_long(struct net_device
*dev
, struct sk_buff
*skb
,
525 void ipoib_cm_handle_rx_wc(struct net_device
*dev
, struct ib_wc
*wc
);
526 void ipoib_cm_handle_tx_wc(struct net_device
*dev
, struct ib_wc
*wc
);
531 #define ipoib_max_conn_qp 0
533 static inline int ipoib_cm_admin_enabled(struct net_device
*dev
)
537 static inline int ipoib_cm_enabled(struct net_device
*dev
, struct neighbour
*n
)
543 static inline int ipoib_cm_up(struct ipoib_neigh
*neigh
)
549 static inline struct ipoib_cm_tx
*ipoib_cm_get(struct ipoib_neigh
*neigh
)
554 static inline void ipoib_cm_set(struct ipoib_neigh
*neigh
, struct ipoib_cm_tx
*tx
)
558 static inline int ipoib_cm_has_srq(struct net_device
*dev
)
563 static inline unsigned int ipoib_cm_max_mtu(struct net_device
*dev
)
569 void ipoib_cm_send(struct net_device
*dev
, struct sk_buff
*skb
, struct ipoib_cm_tx
*tx
)
575 int ipoib_cm_dev_open(struct net_device
*dev
)
581 void ipoib_cm_dev_stop(struct net_device
*dev
)
587 int ipoib_cm_dev_init(struct net_device
*dev
)
593 void ipoib_cm_dev_cleanup(struct net_device
*dev
)
599 struct ipoib_cm_tx
*ipoib_cm_create_tx(struct net_device
*dev
, struct ipoib_path
*path
,
600 struct ipoib_neigh
*neigh
)
606 void ipoib_cm_destroy_tx(struct ipoib_cm_tx
*tx
)
612 int ipoib_cm_add_mode_attr(struct net_device
*dev
)
617 static inline void ipoib_cm_skb_too_long(struct net_device
*dev
, struct sk_buff
*skb
,
620 dev_kfree_skb_any(skb
);
623 static inline void ipoib_cm_handle_rx_wc(struct net_device
*dev
, struct ib_wc
*wc
)
627 static inline void ipoib_cm_handle_tx_wc(struct net_device
*dev
, struct ib_wc
*wc
)
632 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
633 void ipoib_create_debug_files(struct net_device
*dev
);
634 void ipoib_delete_debug_files(struct net_device
*dev
);
635 int ipoib_register_debugfs(void);
636 void ipoib_unregister_debugfs(void);
638 static inline void ipoib_create_debug_files(struct net_device
*dev
) { }
639 static inline void ipoib_delete_debug_files(struct net_device
*dev
) { }
640 static inline int ipoib_register_debugfs(void) { return 0; }
641 static inline void ipoib_unregister_debugfs(void) { }
645 #define ipoib_printk(level, priv, format, arg...) \
646 printk(level "%s: " format, ((struct ipoib_dev_priv *) priv)->dev->name , ## arg)
647 #define ipoib_warn(priv, format, arg...) \
648 ipoib_printk(KERN_WARNING, priv, format , ## arg)
650 extern int ipoib_sendq_size
;
651 extern int ipoib_recvq_size
;
653 extern struct ib_sa_client ipoib_sa_client
;
655 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
656 extern int ipoib_debug_level
;
658 #define ipoib_dbg(priv, format, arg...) \
660 if (ipoib_debug_level > 0) \
661 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
663 #define ipoib_dbg_mcast(priv, format, arg...) \
665 if (mcast_debug_level > 0) \
666 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
668 #else /* CONFIG_INFINIBAND_IPOIB_DEBUG */
669 #define ipoib_dbg(priv, format, arg...) \
670 do { (void) (priv); } while (0)
671 #define ipoib_dbg_mcast(priv, format, arg...) \
672 do { (void) (priv); } while (0)
673 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */
675 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG_DATA
676 #define ipoib_dbg_data(priv, format, arg...) \
678 if (data_debug_level > 0) \
679 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
681 #else /* CONFIG_INFINIBAND_IPOIB_DEBUG_DATA */
682 #define ipoib_dbg_data(priv, format, arg...) \
683 do { (void) (priv); } while (0)
684 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG_DATA */
687 #define IPOIB_GID_FMT "%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x:" \
688 "%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x:%2.2x%2.2x"
690 #define IPOIB_GID_RAW_ARG(gid) ((u8 *)(gid))[0], \
707 #define IPOIB_GID_ARG(gid) IPOIB_GID_RAW_ARG((gid).raw)
709 #define IPOIB_QPN(ha) (be32_to_cpup((__be32 *) ha) & 0xffffff)
711 #endif /* _IPOIB_H */