Linux Kernel  3.7.1
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
ipoib.h
Go to the documentation of this file.
1 /*
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.
5  *
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:
11  *
12  * Redistribution and use in source and binary forms, with or
13  * without modification, are permitted provided that the following
14  * conditions are met:
15  *
16  * - Redistributions of source code must retain the above
17  * copyright notice, this list of conditions and the following
18  * disclaimer.
19  *
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.
24  *
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
32  * SOFTWARE.
33  */
34 
35 #ifndef _IPOIB_H
36 #define _IPOIB_H
37 
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>
45 
46 #include <net/neighbour.h>
47 #include <net/sch_generic.h>
48 
49 #include <linux/atomic.h>
50 
51 #include <rdma/ib_verbs.h>
52 #include <rdma/ib_pack.h>
53 #include <rdma/ib_sa.h>
54 #include <linux/sched.h>
55 
56 /* constants */
57 
62 };
63 
64 enum {
66 
68  IPOIB_UD_RX_SG = 2, /* max buffer needed for 4K mtu */
69 
70  IPOIB_CM_MTU = 0x10000 - 0x10, /* padding to align header to 16 */
79 
81 
84 
97 
99 
100  IPOIB_MCAST_FLAG_FOUND = 0, /* used in set_multicast_list */
102  IPOIB_MCAST_FLAG_BUSY = 2, /* joining or already joined */
104 
107 
111 };
112 
113 #define IPOIB_OP_RECV (1ul << 31)
114 #ifdef CONFIG_INFINIBAND_IPOIB_CM
115 #define IPOIB_OP_CM (1ul << 30)
116 #else
117 #define IPOIB_OP_CM (0)
118 #endif
119 
120 /* structs */
121 
122 struct ipoib_header {
125 };
126 
127 struct ipoib_cb {
130 };
131 
132 /* Used for all multicast joins (broadcast, IPv4 mcast and IPv6 mcast) */
133 struct ipoib_mcast {
136  struct ipoib_ah *ah;
137 
138  struct rb_node rb_node;
139  struct list_head list;
140 
141  unsigned long created;
142  unsigned long backoff;
143 
144  unsigned long flags;
145  unsigned char logcount;
146 
148 
150 
151  struct net_device *dev;
152 };
153 
154 struct ipoib_rx_buf {
155  struct sk_buff *skb;
157 };
158 
159 struct ipoib_tx_buf {
160  struct sk_buff *skb;
162 };
163 
165  struct sk_buff *skb;
167 };
168 
169 struct ib_cm_id;
170 
172  __be32 qpn; /* High byte MUST be ignored on receive */
174 };
175 
176 /*
177  * Quoting 10.3.1 Queue Pair and EE Context States:
178  *
179  * Note, for QPs that are associated with an SRQ, the Consumer should take the
180  * QP through the Error State before invoking a Destroy QP or a Modify QP to the
181  * Reset State. The Consumer may invoke the Destroy QP without first performing
182  * a Modify QP to the Error State and waiting for the Affiliated Asynchronous
183  * Last WQE Reached Event. However, if the Consumer does not wait for the
184  * Affiliated Asynchronous Last WQE Reached Event, then WQE and Data Segment
185  * leakage may occur. Therefore, it is good programming practice to tear down a
186  * QP that is associated with an SRQ by using the following process:
187  *
188  * - Put the QP in the Error State
189  * - Wait for the Affiliated Asynchronous Last WQE Reached Event;
190  * - either:
191  * drain the CQ by invoking the Poll CQ verb and either wait for CQ
192  * to be empty or the number of Poll CQ operations has exceeded
193  * CQ capacity size;
194  * - or
195  * post another WR that completes on the same CQ and wait for this
196  * WR to return as a WC;
197  * - and then invoke a Destroy QP or Reset QP.
198  *
199  * We use the second option and wait for a completion on the
200  * same CQ before destroying QPs attached to our SRQ.
201  */
202 
205  IPOIB_CM_RX_ERROR, /* Ignored by stale task */
206  IPOIB_CM_RX_FLUSH /* Last WQE Reached event observed */
207 };
208 
209 struct ipoib_cm_rx {
210  struct ib_cm_id *id;
211  struct ib_qp *qp;
213  struct list_head list;
214  struct net_device *dev;
215  unsigned long jiffies;
218 };
219 
220 struct ipoib_cm_tx {
221  struct ib_cm_id *id;
222  struct ib_qp *qp;
223  struct list_head list;
224  struct net_device *dev;
226  struct ipoib_path *path;
228  unsigned tx_head;
229  unsigned tx_tail;
230  unsigned long flags;
232 };
233 
235  struct sk_buff *skb;
237 };
238 
240  struct ib_srq *srq;
242  struct ib_cm_id *id;
243  struct list_head passive_ids; /* state: LIVE */
244  struct list_head rx_error_list; /* state: ERROR */
245  struct list_head rx_flush_list; /* state: FLUSH, drain not started */
246  struct list_head rx_drain_list; /* state: FLUSH, drain started */
247  struct list_head rx_reap_list; /* state: FLUSH, drain done */
262 };
263 
267 };
268 
269 struct ipoib_neigh_table;
270 
274  struct rcu_head rcu;
277 };
278 
284 };
285 
286 /*
287  * Device private locking: network stack tx_lock protects members used
288  * in TX fast path, lock protects everything else. lock nests inside
289  * of tx_lock (ie tx_lock must be acquired first if needed).
290  */
293 
294  struct net_device *dev;
295 
297 
298  unsigned long flags;
299 
301 
304 
306 
310 
320  struct ib_device *ca;
324  struct ib_pd *pd;
325  struct ib_mr *mr;
326  struct ib_cq *recv_cq;
327  struct ib_cq *send_cq;
328  struct ib_qp *qp;
330 
333 
334  unsigned int admin_mtu;
335  unsigned int mcast_mtu;
336  unsigned int max_ib_mtu;
337 
339 
341  unsigned tx_head;
342  unsigned tx_tail;
345  unsigned tx_outstanding;
347 
350 
352 
354 
356 
359  struct list_head list;
361 
362 #ifdef CONFIG_INFINIBAND_IPOIB_CM
363  struct ipoib_cm_dev_priv cm;
364 #endif
365 
366 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
367  struct list_head fs_list;
368  struct dentry *mcg_dentry;
369  struct dentry *path_dentry;
370 #endif
371  int hca_caps;
374 };
375 
376 struct ipoib_ah {
377  struct net_device *dev;
378  struct ib_ah *ah;
379  struct list_head list;
380  struct kref ref;
381  unsigned last_send;
382 };
383 
384 struct ipoib_path {
385  struct net_device *dev;
387  struct ipoib_ah *ah;
389 
391 
392  int query_id;
394  struct completion done;
395 
396  struct rb_node rb_node;
397  struct list_head list;
398  int valid;
399 };
400 
401 struct ipoib_neigh {
402  struct ipoib_ah *ah;
403 #ifdef CONFIG_INFINIBAND_IPOIB_CM
404  struct ipoib_cm_tx *cm;
405 #endif
408 
409  struct net_device *dev;
410 
411  struct list_head list;
413  struct rcu_head rcu;
415  unsigned long alive;
416 };
417 
418 #define IPOIB_UD_MTU(ib_mtu) (ib_mtu - IPOIB_ENCAP_LEN)
419 #define IPOIB_UD_BUF_SIZE(ib_mtu) (ib_mtu + IB_GRH_BYTES)
420 
421 static inline int ipoib_ud_need_sg(unsigned int ib_mtu)
422 {
423  return IPOIB_UD_BUF_SIZE(ib_mtu) > PAGE_SIZE;
424 }
425 
426 void ipoib_neigh_dtor(struct ipoib_neigh *neigh);
427 static inline void ipoib_neigh_put(struct ipoib_neigh *neigh)
428 {
429  if (atomic_dec_and_test(&neigh->refcnt))
430  ipoib_neigh_dtor(neigh);
431 }
432 struct ipoib_neigh *ipoib_neigh_get(struct net_device *dev, u8 *daddr);
434  struct net_device *dev);
435 void ipoib_neigh_free(struct ipoib_neigh *neigh);
437 
438 extern struct workqueue_struct *ipoib_workqueue;
439 
440 /* functions */
441 
442 int ipoib_poll(struct napi_struct *napi, int budget);
443 void ipoib_ib_completion(struct ib_cq *cq, void *dev_ptr);
444 void ipoib_send_comp_handler(struct ib_cq *cq, void *dev_ptr);
445 
446 struct ipoib_ah *ipoib_create_ah(struct net_device *dev,
447  struct ib_pd *pd, struct ib_ah_attr *attr);
448 void ipoib_free_ah(struct kref *kref);
449 static inline void ipoib_put_ah(struct ipoib_ah *ah)
450 {
451  kref_put(&ah->ref, ipoib_free_ah);
452 }
453 int ipoib_open(struct net_device *dev);
454 int ipoib_add_pkey_attr(struct net_device *dev);
456 
457 void ipoib_send(struct net_device *dev, struct sk_buff *skb,
458  struct ipoib_ah *address, u32 qpn);
459 void ipoib_reap_ah(struct work_struct *work);
460 
462 void ipoib_flush_paths(struct net_device *dev);
463 struct ipoib_dev_priv *ipoib_intf_alloc(const char *format);
464 
465 int ipoib_ib_dev_init(struct net_device *dev, struct ib_device *ca, int port);
469 void ipoib_pkey_event(struct work_struct *work);
470 void ipoib_ib_dev_cleanup(struct net_device *dev);
471 
472 int ipoib_ib_dev_open(struct net_device *dev);
473 int ipoib_ib_dev_up(struct net_device *dev);
474 int ipoib_ib_dev_down(struct net_device *dev, int flush);
475 int ipoib_ib_dev_stop(struct net_device *dev, int flush);
476 
477 int ipoib_dev_init(struct net_device *dev, struct ib_device *ca, int port);
478 void ipoib_dev_cleanup(struct net_device *dev);
479 
482 void ipoib_mcast_send(struct net_device *dev, u8 *daddr, struct sk_buff *skb);
483 
486 int ipoib_mcast_stop_thread(struct net_device *dev, int flush);
487 
488 void ipoib_mcast_dev_down(struct net_device *dev);
489 void ipoib_mcast_dev_flush(struct net_device *dev);
490 
491 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
492 struct ipoib_mcast_iter *ipoib_mcast_iter_init(struct net_device *dev);
493 int ipoib_mcast_iter_next(struct ipoib_mcast_iter *iter);
494 void ipoib_mcast_iter_read(struct ipoib_mcast_iter *iter,
495  union ib_gid *gid,
496  unsigned long *created,
497  unsigned int *queuelen,
498  unsigned int *complete,
499  unsigned int *send_only);
500 
501 struct ipoib_path_iter *ipoib_path_iter_init(struct net_device *dev);
502 int ipoib_path_iter_next(struct ipoib_path_iter *iter);
503 void ipoib_path_iter_read(struct ipoib_path_iter *iter,
504  struct ipoib_path *path);
505 #endif
506 
507 int ipoib_mcast_attach(struct net_device *dev, u16 mlid,
508  union ib_gid *mgid, int set_qkey);
509 
510 int ipoib_init_qp(struct net_device *dev);
511 int ipoib_transport_dev_init(struct net_device *dev, struct ib_device *ca);
513 
514 void ipoib_event(struct ib_event_handler *handler,
515  struct ib_event *record);
516 
517 int ipoib_vlan_add(struct net_device *pdev, unsigned short pkey);
518 int ipoib_vlan_delete(struct net_device *pdev, unsigned short pkey);
519 
520 int __ipoib_vlan_add(struct ipoib_dev_priv *ppriv, struct ipoib_dev_priv *priv,
521  u16 pkey, int child_type);
522 
523 int __init ipoib_netlink_init(void);
524 void __exit ipoib_netlink_fini(void);
525 
526 void ipoib_set_umcast(struct net_device *ndev, int umcast_val);
527 int ipoib_set_mode(struct net_device *dev, const char *buf);
528 
529 void ipoib_setup(struct net_device *dev);
530 
531 void ipoib_pkey_poll(struct work_struct *work);
533 void ipoib_drain_cq(struct net_device *dev);
534 
535 void ipoib_set_ethtool_ops(struct net_device *dev);
536 int ipoib_set_dev_features(struct ipoib_dev_priv *priv, struct ib_device *hca);
537 
538 #define IPOIB_FLAGS_RC 0x80
539 #define IPOIB_FLAGS_UC 0x40
540 
541 /* We don't support UC connections at the moment */
542 #define IPOIB_CM_SUPPORTED(ha) (ha[0] & (IPOIB_FLAGS_RC))
543 
544 #ifdef CONFIG_INFINIBAND_IPOIB_CM
545 
546 extern int ipoib_max_conn_qp;
547 
548 static inline int ipoib_cm_admin_enabled(struct net_device *dev)
549 {
550  struct ipoib_dev_priv *priv = netdev_priv(dev);
551  return IPOIB_CM_SUPPORTED(dev->dev_addr) &&
553 }
554 
555 static inline int ipoib_cm_enabled(struct net_device *dev, u8 *hwaddr)
556 {
557  struct ipoib_dev_priv *priv = netdev_priv(dev);
558  return IPOIB_CM_SUPPORTED(hwaddr) &&
560 }
561 
562 static inline int ipoib_cm_up(struct ipoib_neigh *neigh)
563 
564 {
565  return test_bit(IPOIB_FLAG_OPER_UP, &neigh->cm->flags);
566 }
567 
568 static inline struct ipoib_cm_tx *ipoib_cm_get(struct ipoib_neigh *neigh)
569 {
570  return neigh->cm;
571 }
572 
573 static inline void ipoib_cm_set(struct ipoib_neigh *neigh, struct ipoib_cm_tx *tx)
574 {
575  neigh->cm = tx;
576 }
577 
578 static inline int ipoib_cm_has_srq(struct net_device *dev)
579 {
580  struct ipoib_dev_priv *priv = netdev_priv(dev);
581  return !!priv->cm.srq;
582 }
583 
584 static inline unsigned int ipoib_cm_max_mtu(struct net_device *dev)
585 {
586  struct ipoib_dev_priv *priv = netdev_priv(dev);
587  return priv->cm.max_cm_mtu;
588 }
589 
590 void ipoib_cm_send(struct net_device *dev, struct sk_buff *skb, struct ipoib_cm_tx *tx);
591 int ipoib_cm_dev_open(struct net_device *dev);
592 void ipoib_cm_dev_stop(struct net_device *dev);
593 int ipoib_cm_dev_init(struct net_device *dev);
595 void ipoib_cm_dev_cleanup(struct net_device *dev);
597  struct ipoib_neigh *neigh);
598 void ipoib_cm_destroy_tx(struct ipoib_cm_tx *tx);
599 void ipoib_cm_skb_too_long(struct net_device *dev, struct sk_buff *skb,
600  unsigned int mtu);
601 void ipoib_cm_handle_rx_wc(struct net_device *dev, struct ib_wc *wc);
602 void ipoib_cm_handle_tx_wc(struct net_device *dev, struct ib_wc *wc);
603 #else
604 
605 struct ipoib_cm_tx;
606 
607 #define ipoib_max_conn_qp 0
608 
609 static inline int ipoib_cm_admin_enabled(struct net_device *dev)
610 {
611  return 0;
612 }
613 static inline int ipoib_cm_enabled(struct net_device *dev, u8 *hwaddr)
614 
615 {
616  return 0;
617 }
618 
619 static inline int ipoib_cm_up(struct ipoib_neigh *neigh)
620 
621 {
622  return 0;
623 }
624 
625 static inline struct ipoib_cm_tx *ipoib_cm_get(struct ipoib_neigh *neigh)
626 {
627  return NULL;
628 }
629 
630 static inline void ipoib_cm_set(struct ipoib_neigh *neigh, struct ipoib_cm_tx *tx)
631 {
632 }
633 
634 static inline int ipoib_cm_has_srq(struct net_device *dev)
635 {
636  return 0;
637 }
638 
639 static inline unsigned int ipoib_cm_max_mtu(struct net_device *dev)
640 {
641  return 0;
642 }
643 
644 static inline
645 void ipoib_cm_send(struct net_device *dev, struct sk_buff *skb, struct ipoib_cm_tx *tx)
646 {
647  return;
648 }
649 
650 static inline
651 int ipoib_cm_dev_open(struct net_device *dev)
652 {
653  return 0;
654 }
655 
656 static inline
657 void ipoib_cm_dev_stop(struct net_device *dev)
658 {
659  return;
660 }
661 
662 static inline
663 int ipoib_cm_dev_init(struct net_device *dev)
664 {
665  return -ENOSYS;
666 }
667 
668 static inline
669 void ipoib_cm_dev_cleanup(struct net_device *dev)
670 {
671  return;
672 }
673 
674 static inline
676  struct ipoib_neigh *neigh)
677 {
678  return NULL;
679 }
680 
681 static inline
682 void ipoib_cm_destroy_tx(struct ipoib_cm_tx *tx)
683 {
684  return;
685 }
686 
687 static inline
689 {
690  return 0;
691 }
692 
693 static inline void ipoib_cm_skb_too_long(struct net_device *dev, struct sk_buff *skb,
694  unsigned int mtu)
695 {
696  dev_kfree_skb_any(skb);
697 }
698 
699 static inline void ipoib_cm_handle_rx_wc(struct net_device *dev, struct ib_wc *wc)
700 {
701 }
702 
703 static inline void ipoib_cm_handle_tx_wc(struct net_device *dev, struct ib_wc *wc)
704 {
705 }
706 #endif
707 
708 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
711 int ipoib_register_debugfs(void);
712 void ipoib_unregister_debugfs(void);
713 #else
714 static inline void ipoib_create_debug_files(struct net_device *dev) { }
715 static inline void ipoib_delete_debug_files(struct net_device *dev) { }
716 static inline int ipoib_register_debugfs(void) { return 0; }
717 static inline void ipoib_unregister_debugfs(void) { }
718 #endif
719 
720 #define ipoib_printk(level, priv, format, arg...) \
721  printk(level "%s: " format, ((struct ipoib_dev_priv *) priv)->dev->name , ## arg)
722 #define ipoib_warn(priv, format, arg...) \
723  ipoib_printk(KERN_WARNING, priv, format , ## arg)
724 
725 extern int ipoib_sendq_size;
726 extern int ipoib_recvq_size;
727 
728 extern struct ib_sa_client ipoib_sa_client;
729 
730 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
731 extern int ipoib_debug_level;
732 
733 #define ipoib_dbg(priv, format, arg...) \
734  do { \
735  if (ipoib_debug_level > 0) \
736  ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
737  } while (0)
738 #define ipoib_dbg_mcast(priv, format, arg...) \
739  do { \
740  if (mcast_debug_level > 0) \
741  ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
742  } while (0)
743 #else /* CONFIG_INFINIBAND_IPOIB_DEBUG */
744 #define ipoib_dbg(priv, format, arg...) \
745  do { (void) (priv); } while (0)
746 #define ipoib_dbg_mcast(priv, format, arg...) \
747  do { (void) (priv); } while (0)
748 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */
749 
750 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG_DATA
751 #define ipoib_dbg_data(priv, format, arg...) \
752  do { \
753  if (data_debug_level > 0) \
754  ipoib_printk(KERN_DEBUG, priv, format , ## arg); \
755  } while (0)
756 #else /* CONFIG_INFINIBAND_IPOIB_DEBUG_DATA */
757 #define ipoib_dbg_data(priv, format, arg...) \
758  do { (void) (priv); } while (0)
759 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG_DATA */
760 
761 #define IPOIB_QPN(ha) (be32_to_cpup((__be32 *) ha) & 0xffffff)
762 
763 #endif /* _IPOIB_H */