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
38 #include <linux/list.h>
39 #include <linux/skbuff.h>
40 #include <linux/netdevice.h>
41 #include <linux/workqueue.h>
42 #include <linux/kref.h>
43 #include <linux/if_infiniband.h>
44 #include <linux/mutex.h>
46 #include <net/neighbour.h>
48 #include <linux/atomic.h>
50 #include <rdma/ib_verbs.h>
51 #include <rdma/ib_pack.h>
52 #include <rdma/ib_sa.h>
53 #include <linux/sched.h>
57 enum ipoib_flush_level
{
66 IPOIB_UD_HEAD_SIZE
= IB_GRH_BYTES
+ IPOIB_ENCAP_LEN
,
67 IPOIB_UD_RX_SG
= 2, /* max buffer needed for 4K mtu */
69 IPOIB_CM_MTU
= 0x10000 - 0x10, /* padding to align header to 16 */
70 IPOIB_CM_BUF_SIZE
= IPOIB_CM_MTU
+ IPOIB_ENCAP_LEN
,
71 IPOIB_CM_HEAD_SIZE
= IPOIB_CM_BUF_SIZE
% PAGE_SIZE
,
72 IPOIB_CM_RX_SG
= ALIGN(IPOIB_CM_BUF_SIZE
, PAGE_SIZE
) / PAGE_SIZE
,
73 IPOIB_RX_RING_SIZE
= 256,
74 IPOIB_TX_RING_SIZE
= 128,
75 IPOIB_MAX_QUEUE_SIZE
= 8192,
76 IPOIB_MIN_QUEUE_SIZE
= 2,
77 IPOIB_CM_MAX_CONN_QP
= 4096,
81 IPOIB_MAX_PATH_REC_QUEUE
= 3,
82 IPOIB_MAX_MCAST_QUEUE
= 3,
84 IPOIB_FLAG_OPER_UP
= 0,
85 IPOIB_FLAG_INITIALIZED
= 1,
86 IPOIB_FLAG_ADMIN_UP
= 2,
87 IPOIB_PKEY_ASSIGNED
= 3,
89 IPOIB_FLAG_SUBINTERFACE
= 5,
91 IPOIB_STOP_REAPER
= 7,
92 IPOIB_FLAG_ADMIN_CM
= 9,
93 IPOIB_FLAG_UMCAST
= 10,
95 IPOIB_MAX_BACKOFF_SECONDS
= 16,
97 IPOIB_MCAST_FLAG_FOUND
= 0, /* used in set_multicast_list */
98 IPOIB_MCAST_FLAG_SENDONLY
= 1,
99 IPOIB_MCAST_FLAG_BUSY
= 2, /* joining or already joined */
100 IPOIB_MCAST_FLAG_ATTACHED
= 3,
103 IPOIB_CM_COPYBREAK
= 256,
106 #define IPOIB_OP_RECV (1ul << 31)
107 #ifdef CONFIG_INFINIBAND_IPOIB_CM
108 #define IPOIB_OP_CM (1ul << 30)
110 #define IPOIB_OP_CM (0)
115 struct ipoib_header
{
120 struct ipoib_pseudoheader
{
121 u8 hwaddr
[INFINIBAND_ALEN
];
124 /* Used for all multicast joins (broadcast, IPv4 mcast and IPv6 mcast) */
126 struct ib_sa_mcmember_rec mcmember
;
127 struct ib_sa_multicast
*mc
;
130 struct rb_node rb_node
;
131 struct list_head list
;
133 unsigned long created
;
134 unsigned long backoff
;
137 unsigned char logcount
;
139 struct list_head neigh_list
;
141 struct sk_buff_head pkt_queue
;
143 struct net_device
*dev
;
146 struct ipoib_rx_buf
{
148 u64 mapping
[IPOIB_UD_RX_SG
];
151 struct ipoib_tx_buf
{
153 u64 mapping
[MAX_SKB_FRAGS
+ 1];
156 struct ipoib_cm_tx_buf
{
163 struct ipoib_cm_data
{
164 __be32 qpn
; /* High byte MUST be ignored on receive */
169 * Quoting 10.3.1 Queue Pair and EE Context States:
171 * Note, for QPs that are associated with an SRQ, the Consumer should take the
172 * QP through the Error State before invoking a Destroy QP or a Modify QP to the
173 * Reset State. The Consumer may invoke the Destroy QP without first performing
174 * a Modify QP to the Error State and waiting for the Affiliated Asynchronous
175 * Last WQE Reached Event. However, if the Consumer does not wait for the
176 * Affiliated Asynchronous Last WQE Reached Event, then WQE and Data Segment
177 * leakage may occur. Therefore, it is good programming practice to tear down a
178 * QP that is associated with an SRQ by using the following process:
180 * - Put the QP in the Error State
181 * - Wait for the Affiliated Asynchronous Last WQE Reached Event;
183 * drain the CQ by invoking the Poll CQ verb and either wait for CQ
184 * to be empty or the number of Poll CQ operations has exceeded
187 * post another WR that completes on the same CQ and wait for this
188 * WR to return as a WC;
189 * - and then invoke a Destroy QP or Reset QP.
191 * We use the second option and wait for a completion on the
192 * same CQ before destroying QPs attached to our SRQ.
195 enum ipoib_cm_state
{
197 IPOIB_CM_RX_ERROR
, /* Ignored by stale task */
198 IPOIB_CM_RX_FLUSH
/* Last WQE Reached event observed */
204 struct ipoib_cm_rx_buf
*rx_ring
;
205 struct list_head list
;
206 struct net_device
*dev
;
207 unsigned long jiffies
;
208 enum ipoib_cm_state state
;
215 struct list_head list
;
216 struct net_device
*dev
;
217 struct ipoib_neigh
*neigh
;
218 struct ipoib_path
*path
;
219 struct ipoib_cm_tx_buf
*tx_ring
;
226 struct ipoib_cm_rx_buf
{
228 u64 mapping
[IPOIB_CM_RX_SG
];
231 struct ipoib_cm_dev_priv
{
233 struct ipoib_cm_rx_buf
*srq_ring
;
235 struct list_head passive_ids
; /* state: LIVE */
236 struct list_head rx_error_list
; /* state: ERROR */
237 struct list_head rx_flush_list
; /* state: FLUSH, drain not started */
238 struct list_head rx_drain_list
; /* state: FLUSH, drain started */
239 struct list_head rx_reap_list
; /* state: FLUSH, drain done */
240 struct work_struct start_task
;
241 struct work_struct reap_task
;
242 struct work_struct skb_task
;
243 struct work_struct rx_reap_task
;
244 struct delayed_work stale_task
;
245 struct sk_buff_head skb_queue
;
246 struct list_head start_list
;
247 struct list_head reap_list
;
248 struct ib_wc ibwc
[IPOIB_NUM_WC
];
249 struct ib_sge rx_sge
[IPOIB_CM_RX_SG
];
250 struct ib_recv_wr rx_wr
;
256 struct ipoib_ethtool_st
{
258 u16 max_coalesced_frames
;
262 * Device private locking: network stack tx_lock protects members used
263 * in TX fast path, lock protects everything else. lock nests inside
264 * of tx_lock (ie tx_lock must be acquired first if needed).
266 struct ipoib_dev_priv
{
269 struct net_device
*dev
;
271 struct napi_struct napi
;
275 struct mutex vlan_mutex
;
277 struct rb_root path_tree
;
278 struct list_head path_list
;
280 struct ipoib_mcast
*broadcast
;
281 struct list_head multicast_list
;
282 struct rb_root multicast_tree
;
284 struct delayed_work pkey_poll_task
;
285 struct delayed_work mcast_task
;
286 struct work_struct carrier_on_task
;
287 struct work_struct flush_light
;
288 struct work_struct flush_normal
;
289 struct work_struct flush_heavy
;
290 struct work_struct restart_task
;
291 struct delayed_work ah_reap_task
;
293 struct ib_device
*ca
;
299 struct ib_cq
*recv_cq
;
300 struct ib_cq
*send_cq
;
304 union ib_gid local_gid
;
307 unsigned int admin_mtu
;
308 unsigned int mcast_mtu
;
309 unsigned int max_ib_mtu
;
311 struct ipoib_rx_buf
*rx_ring
;
313 struct ipoib_tx_buf
*tx_ring
;
316 struct ib_sge tx_sge
[MAX_SKB_FRAGS
+ 1];
317 struct ib_send_wr tx_wr
;
318 unsigned tx_outstanding
;
319 struct ib_wc send_wc
[MAX_SEND_CQE
];
321 struct ib_recv_wr rx_wr
;
322 struct ib_sge rx_sge
[IPOIB_UD_RX_SG
];
324 struct ib_wc ibwc
[IPOIB_NUM_WC
];
326 struct list_head dead_ahs
;
328 struct ib_event_handler event_handler
;
330 struct net_device
*parent
;
331 struct list_head child_intfs
;
332 struct list_head list
;
334 #ifdef CONFIG_INFINIBAND_IPOIB_CM
335 struct ipoib_cm_dev_priv cm
;
338 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
339 struct list_head fs_list
;
340 struct dentry
*mcg_dentry
;
341 struct dentry
*path_dentry
;
344 struct ipoib_ethtool_st ethtool
;
345 struct timer_list poll_timer
;
349 struct net_device
*dev
;
351 struct list_head list
;
357 struct net_device
*dev
;
358 struct ib_sa_path_rec pathrec
;
360 struct sk_buff_head queue
;
362 struct list_head neigh_list
;
365 struct ib_sa_query
*query
;
366 struct completion done
;
368 struct rb_node rb_node
;
369 struct list_head list
;
375 #ifdef CONFIG_INFINIBAND_IPOIB_CM
376 struct ipoib_cm_tx
*cm
;
379 struct sk_buff_head queue
;
381 struct neighbour
*neighbour
;
382 struct net_device
*dev
;
384 struct list_head list
;
387 #define IPOIB_UD_MTU(ib_mtu) (ib_mtu - IPOIB_ENCAP_LEN)
388 #define IPOIB_UD_BUF_SIZE(ib_mtu) (ib_mtu + IB_GRH_BYTES)
390 static inline int ipoib_ud_need_sg(unsigned int ib_mtu
)
392 return IPOIB_UD_BUF_SIZE(ib_mtu
) > PAGE_SIZE
;
396 * We stash a pointer to our private neighbour information after our
397 * hardware address in neigh->ha. The ALIGN() expression here makes
398 * sure that this pointer is stored aligned so that an unaligned
399 * load is not needed to dereference it.
401 static inline struct ipoib_neigh
**to_ipoib_neigh(struct neighbour
*neigh
)
403 return (void*) neigh
+ ALIGN(offsetof(struct neighbour
, ha
) +
404 INFINIBAND_ALEN
, sizeof(void *));
407 struct ipoib_neigh
*ipoib_neigh_alloc(struct neighbour
*neigh
,
408 struct net_device
*dev
);
409 void ipoib_neigh_free(struct net_device
*dev
, struct ipoib_neigh
*neigh
);
411 extern struct workqueue_struct
*ipoib_workqueue
;
415 int ipoib_poll(struct napi_struct
*napi
, int budget
);
416 void ipoib_ib_completion(struct ib_cq
*cq
, void *dev_ptr
);
417 void ipoib_send_comp_handler(struct ib_cq
*cq
, void *dev_ptr
);
419 struct ipoib_ah
*ipoib_create_ah(struct net_device
*dev
,
420 struct ib_pd
*pd
, struct ib_ah_attr
*attr
);
421 void ipoib_free_ah(struct kref
*kref
);
422 static inline void ipoib_put_ah(struct ipoib_ah
*ah
)
424 kref_put(&ah
->ref
, ipoib_free_ah
);
427 int ipoib_open(struct net_device
*dev
);
428 int ipoib_add_pkey_attr(struct net_device
*dev
);
429 int ipoib_add_umcast_attr(struct net_device
*dev
);
431 void ipoib_send(struct net_device
*dev
, struct sk_buff
*skb
,
432 struct ipoib_ah
*address
, u32 qpn
);
433 void ipoib_reap_ah(struct work_struct
*work
);
435 void ipoib_mark_paths_invalid(struct net_device
*dev
);
436 void ipoib_flush_paths(struct net_device
*dev
);
437 struct ipoib_dev_priv
*ipoib_intf_alloc(const char *format
);
439 int ipoib_ib_dev_init(struct net_device
*dev
, struct ib_device
*ca
, int port
);
440 void ipoib_ib_dev_flush_light(struct work_struct
*work
);
441 void ipoib_ib_dev_flush_normal(struct work_struct
*work
);
442 void ipoib_ib_dev_flush_heavy(struct work_struct
*work
);
443 void ipoib_pkey_event(struct work_struct
*work
);
444 void ipoib_ib_dev_cleanup(struct net_device
*dev
);
446 int ipoib_ib_dev_open(struct net_device
*dev
);
447 int ipoib_ib_dev_up(struct net_device
*dev
);
448 int ipoib_ib_dev_down(struct net_device
*dev
, int flush
);
449 int ipoib_ib_dev_stop(struct net_device
*dev
, int flush
);
451 int ipoib_dev_init(struct net_device
*dev
, struct ib_device
*ca
, int port
);
452 void ipoib_dev_cleanup(struct net_device
*dev
);
454 void ipoib_mcast_join_task(struct work_struct
*work
);
455 void ipoib_mcast_carrier_on_task(struct work_struct
*work
);
456 void ipoib_mcast_send(struct net_device
*dev
, void *mgid
, struct sk_buff
*skb
);
458 void ipoib_mcast_restart_task(struct work_struct
*work
);
459 int ipoib_mcast_start_thread(struct net_device
*dev
);
460 int ipoib_mcast_stop_thread(struct net_device
*dev
, int flush
);
462 void ipoib_mcast_dev_down(struct net_device
*dev
);
463 void ipoib_mcast_dev_flush(struct net_device
*dev
);
465 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
466 struct ipoib_mcast_iter
*ipoib_mcast_iter_init(struct net_device
*dev
);
467 int ipoib_mcast_iter_next(struct ipoib_mcast_iter
*iter
);
468 void ipoib_mcast_iter_read(struct ipoib_mcast_iter
*iter
,
470 unsigned long *created
,
471 unsigned int *queuelen
,
472 unsigned int *complete
,
473 unsigned int *send_only
);
475 struct ipoib_path_iter
*ipoib_path_iter_init(struct net_device
*dev
);
476 int ipoib_path_iter_next(struct ipoib_path_iter
*iter
);
477 void ipoib_path_iter_read(struct ipoib_path_iter
*iter
,
478 struct ipoib_path
*path
);
481 int ipoib_mcast_attach(struct net_device
*dev
, u16 mlid
,
482 union ib_gid
*mgid
, int set_qkey
);
484 int ipoib_init_qp(struct net_device
*dev
);
485 int ipoib_transport_dev_init(struct net_device
*dev
, struct ib_device
*ca
);
486 void ipoib_transport_dev_cleanup(struct net_device
*dev
);
488 void ipoib_event(struct ib_event_handler
*handler
,
489 struct ib_event
*record
);
491 int ipoib_vlan_add(struct net_device
*pdev
, unsigned short pkey
);
492 int ipoib_vlan_delete(struct net_device
*pdev
, unsigned short pkey
);
494 void ipoib_pkey_poll(struct work_struct
*work
);
495 int ipoib_pkey_dev_delay_open(struct net_device
*dev
);
496 void ipoib_drain_cq(struct net_device
*dev
);
498 void ipoib_set_ethtool_ops(struct net_device
*dev
);
499 int ipoib_set_dev_features(struct ipoib_dev_priv
*priv
, struct ib_device
*hca
);
501 #ifdef CONFIG_INFINIBAND_IPOIB_CM
503 #define IPOIB_FLAGS_RC 0x80
504 #define IPOIB_FLAGS_UC 0x40
506 /* We don't support UC connections at the moment */
507 #define IPOIB_CM_SUPPORTED(ha) (ha[0] & (IPOIB_FLAGS_RC))
509 extern int ipoib_max_conn_qp
;
511 static inline int ipoib_cm_admin_enabled(struct net_device
*dev
)
513 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
514 return IPOIB_CM_SUPPORTED(dev
->dev_addr
) &&
515 test_bit(IPOIB_FLAG_ADMIN_CM
, &priv
->flags
);
518 static inline int ipoib_cm_enabled(struct net_device
*dev
, struct neighbour
*n
)
520 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
521 return IPOIB_CM_SUPPORTED(n
->ha
) &&
522 test_bit(IPOIB_FLAG_ADMIN_CM
, &priv
->flags
);
525 static inline int ipoib_cm_up(struct ipoib_neigh
*neigh
)
528 return test_bit(IPOIB_FLAG_OPER_UP
, &neigh
->cm
->flags
);
531 static inline struct ipoib_cm_tx
*ipoib_cm_get(struct ipoib_neigh
*neigh
)
536 static inline void ipoib_cm_set(struct ipoib_neigh
*neigh
, struct ipoib_cm_tx
*tx
)
541 static inline int ipoib_cm_has_srq(struct net_device
*dev
)
543 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
544 return !!priv
->cm
.srq
;
547 static inline unsigned int ipoib_cm_max_mtu(struct net_device
*dev
)
549 struct ipoib_dev_priv
*priv
= netdev_priv(dev
);
550 return priv
->cm
.max_cm_mtu
;
553 void ipoib_cm_send(struct net_device
*dev
, struct sk_buff
*skb
, struct ipoib_cm_tx
*tx
);
554 int ipoib_cm_dev_open(struct net_device
*dev
);
555 void ipoib_cm_dev_stop(struct net_device
*dev
);
556 int ipoib_cm_dev_init(struct net_device
*dev
);
557 int ipoib_cm_add_mode_attr(struct net_device
*dev
);
558 void ipoib_cm_dev_cleanup(struct net_device
*dev
);
559 struct ipoib_cm_tx
*ipoib_cm_create_tx(struct net_device
*dev
, struct ipoib_path
*path
,
560 struct ipoib_neigh
*neigh
);
561 void ipoib_cm_destroy_tx(struct ipoib_cm_tx
*tx
);
562 void ipoib_cm_skb_too_long(struct net_device
*dev
, struct sk_buff
*skb
,
564 void ipoib_cm_handle_rx_wc(struct net_device
*dev
, struct ib_wc
*wc
);
565 void ipoib_cm_handle_tx_wc(struct net_device
*dev
, struct ib_wc
*wc
);
570 #define ipoib_max_conn_qp 0
572 static inline int ipoib_cm_admin_enabled(struct net_device
*dev
)
576 static inline int ipoib_cm_enabled(struct net_device
*dev
, struct neighbour
*n
)
582 static inline int ipoib_cm_up(struct ipoib_neigh
*neigh
)
588 static inline struct ipoib_cm_tx
*ipoib_cm_get(struct ipoib_neigh
*neigh
)
593 static inline void ipoib_cm_set(struct ipoib_neigh
*neigh
, struct ipoib_cm_tx
*tx
)
597 static inline int ipoib_cm_has_srq(struct net_device
*dev
)
602 static inline unsigned int ipoib_cm_max_mtu(struct net_device
*dev
)
608 void ipoib_cm_send(struct net_device
*dev
, struct sk_buff
*skb
, struct ipoib_cm_tx
*tx
)
614 int ipoib_cm_dev_open(struct net_device
*dev
)
620 void ipoib_cm_dev_stop(struct net_device
*dev
)
626 int ipoib_cm_dev_init(struct net_device
*dev
)
632 void ipoib_cm_dev_cleanup(struct net_device
*dev
)
638 struct ipoib_cm_tx
*ipoib_cm_create_tx(struct net_device
*dev
, struct ipoib_path
*path
,
639 struct ipoib_neigh
*neigh
)
645 void ipoib_cm_destroy_tx(struct ipoib_cm_tx
*tx
)
651 int ipoib_cm_add_mode_attr(struct net_device
*dev
)
656 static inline void ipoib_cm_skb_too_long(struct net_device
*dev
, struct sk_buff
*skb
,
659 dev_kfree_skb_any(skb
);
662 static inline void ipoib_cm_handle_rx_wc(struct net_device
*dev
, struct ib_wc
*wc
)
666 static inline void ipoib_cm_handle_tx_wc(struct net_device
*dev
, struct ib_wc
*wc
)
671 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
672 void ipoib_create_debug_files(struct net_device
*dev
);
673 void ipoib_delete_debug_files(struct net_device
*dev
);
674 int ipoib_register_debugfs(void);
675 void ipoib_unregister_debugfs(void);
677 static inline void ipoib_create_debug_files(struct net_device
*dev
) { }
678 static inline void ipoib_delete_debug_files(struct net_device
*dev
) { }
679 static inline int ipoib_register_debugfs(void) { return 0; }
680 static inline void ipoib_unregister_debugfs(void) { }
683 #define ipoib_printk(level, priv, format, arg...) \
684 printk(level "%s: " format, ((struct ipoib_dev_priv *) priv)->dev->name , ## arg)
685 #define ipoib_warn(priv, format, arg...) \
686 ipoib_printk(KERN_WARNING, priv, format , ## arg)
688 extern int ipoib_sendq_size
;
689 extern int ipoib_recvq_size
;
691 extern struct ib_sa_client ipoib_sa_client
;
693 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
694 extern int ipoib_debug_level
;
696 #define ipoib_dbg(priv, format, arg...) \
698 if (ipoib_debug_level > 0) \
699 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
701 #define ipoib_dbg_mcast(priv, format, arg...) \
703 if (mcast_debug_level > 0) \
704 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
706 #else /* CONFIG_INFINIBAND_IPOIB_DEBUG */
707 #define ipoib_dbg(priv, format, arg...) \
708 do { (void) (priv); } while (0)
709 #define ipoib_dbg_mcast(priv, format, arg...) \
710 do { (void) (priv); } while (0)
711 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */
713 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG_DATA
714 #define ipoib_dbg_data(priv, format, arg...) \
716 if (data_debug_level > 0) \
717 ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
719 #else /* CONFIG_INFINIBAND_IPOIB_DEBUG_DATA */
720 #define ipoib_dbg_data(priv, format, arg...) \
721 do { (void) (priv); } while (0)
722 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG_DATA */
724 #define IPOIB_QPN(ha) (be32_to_cpup((__be32 *) ha) & 0xffffff)
726 #endif /* _IPOIB_H */