Linux Kernel  3.7.1
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
sa_query.c
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2004 Topspin Communications. All rights reserved.
3  * Copyright (c) 2005 Voltaire, Inc. All rights reserved.
4  * Copyright (c) 2006 Intel Corporation. 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 #include <linux/module.h>
36 #include <linux/init.h>
37 #include <linux/err.h>
38 #include <linux/random.h>
39 #include <linux/spinlock.h>
40 #include <linux/slab.h>
41 #include <linux/dma-mapping.h>
42 #include <linux/kref.h>
43 #include <linux/idr.h>
44 #include <linux/workqueue.h>
45 
46 #include <rdma/ib_pack.h>
47 #include <rdma/ib_cache.h>
48 #include "sa.h"
49 
50 MODULE_AUTHOR("Roland Dreier");
51 MODULE_DESCRIPTION("InfiniBand subnet administration query support");
52 MODULE_LICENSE("Dual BSD/GPL");
53 
54 struct ib_sa_sm_ah {
55  struct ib_ah *ah;
56  struct kref ref;
59 };
60 
61 struct ib_sa_port {
63  struct ib_sa_sm_ah *sm_ah;
67 };
68 
69 struct ib_sa_device {
72  struct ib_sa_port port[0];
73 };
74 
75 struct ib_sa_query {
76  void (*callback)(struct ib_sa_query *, int, struct ib_sa_mad *);
77  void (*release)(struct ib_sa_query *);
79  struct ib_sa_port *port;
81  struct ib_sa_sm_ah *sm_ah;
82  int id;
83 };
84 
86  void (*callback)(int, struct ib_sa_service_rec *, void *);
87  void *context;
89 };
90 
92  void (*callback)(int, struct ib_sa_path_rec *, void *);
93  void *context;
95 };
96 
98  void (*callback)(int, struct ib_sa_guidinfo_rec *, void *);
99  void *context;
101 };
102 
104  void (*callback)(int, struct ib_sa_mcmember_rec *, void *);
105  void *context;
107 };
108 
109 static void ib_sa_add_one(struct ib_device *device);
110 static void ib_sa_remove_one(struct ib_device *device);
111 
112 static struct ib_client sa_client = {
113  .name = "sa",
114  .add = ib_sa_add_one,
115  .remove = ib_sa_remove_one
116 };
117 
118 static DEFINE_SPINLOCK(idr_lock);
119 static DEFINE_IDR(query_idr);
120 
121 static DEFINE_SPINLOCK(tid_lock);
122 static u32 tid;
123 
124 #define PATH_REC_FIELD(field) \
125  .struct_offset_bytes = offsetof(struct ib_sa_path_rec, field), \
126  .struct_size_bytes = sizeof ((struct ib_sa_path_rec *) 0)->field, \
127  .field_name = "sa_path_rec:" #field
128 
129 static const struct ib_field path_rec_table[] = {
131  .offset_words = 0,
132  .offset_bits = 0,
133  .size_bits = 64 },
134  { PATH_REC_FIELD(dgid),
135  .offset_words = 2,
136  .offset_bits = 0,
137  .size_bits = 128 },
138  { PATH_REC_FIELD(sgid),
139  .offset_words = 6,
140  .offset_bits = 0,
141  .size_bits = 128 },
142  { PATH_REC_FIELD(dlid),
143  .offset_words = 10,
144  .offset_bits = 0,
145  .size_bits = 16 },
146  { PATH_REC_FIELD(slid),
147  .offset_words = 10,
148  .offset_bits = 16,
149  .size_bits = 16 },
150  { PATH_REC_FIELD(raw_traffic),
151  .offset_words = 11,
152  .offset_bits = 0,
153  .size_bits = 1 },
154  { RESERVED,
155  .offset_words = 11,
156  .offset_bits = 1,
157  .size_bits = 3 },
158  { PATH_REC_FIELD(flow_label),
159  .offset_words = 11,
160  .offset_bits = 4,
161  .size_bits = 20 },
163  .offset_words = 11,
164  .offset_bits = 24,
165  .size_bits = 8 },
166  { PATH_REC_FIELD(traffic_class),
167  .offset_words = 12,
168  .offset_bits = 0,
169  .size_bits = 8 },
170  { PATH_REC_FIELD(reversible),
171  .offset_words = 12,
172  .offset_bits = 8,
173  .size_bits = 1 },
174  { PATH_REC_FIELD(numb_path),
175  .offset_words = 12,
176  .offset_bits = 9,
177  .size_bits = 7 },
178  { PATH_REC_FIELD(pkey),
179  .offset_words = 12,
180  .offset_bits = 16,
181  .size_bits = 16 },
182  { PATH_REC_FIELD(qos_class),
183  .offset_words = 13,
184  .offset_bits = 0,
185  .size_bits = 12 },
186  { PATH_REC_FIELD(sl),
187  .offset_words = 13,
188  .offset_bits = 12,
189  .size_bits = 4 },
190  { PATH_REC_FIELD(mtu_selector),
191  .offset_words = 13,
192  .offset_bits = 16,
193  .size_bits = 2 },
194  { PATH_REC_FIELD(mtu),
195  .offset_words = 13,
196  .offset_bits = 18,
197  .size_bits = 6 },
198  { PATH_REC_FIELD(rate_selector),
199  .offset_words = 13,
200  .offset_bits = 24,
201  .size_bits = 2 },
202  { PATH_REC_FIELD(rate),
203  .offset_words = 13,
204  .offset_bits = 26,
205  .size_bits = 6 },
206  { PATH_REC_FIELD(packet_life_time_selector),
207  .offset_words = 14,
208  .offset_bits = 0,
209  .size_bits = 2 },
210  { PATH_REC_FIELD(packet_life_time),
211  .offset_words = 14,
212  .offset_bits = 2,
213  .size_bits = 6 },
214  { PATH_REC_FIELD(preference),
215  .offset_words = 14,
216  .offset_bits = 8,
217  .size_bits = 8 },
218  { RESERVED,
219  .offset_words = 14,
220  .offset_bits = 16,
221  .size_bits = 48 },
222 };
223 
224 #define MCMEMBER_REC_FIELD(field) \
225  .struct_offset_bytes = offsetof(struct ib_sa_mcmember_rec, field), \
226  .struct_size_bytes = sizeof ((struct ib_sa_mcmember_rec *) 0)->field, \
227  .field_name = "sa_mcmember_rec:" #field
228 
229 static const struct ib_field mcmember_rec_table[] = {
230  { MCMEMBER_REC_FIELD(mgid),
231  .offset_words = 0,
232  .offset_bits = 0,
233  .size_bits = 128 },
234  { MCMEMBER_REC_FIELD(port_gid),
235  .offset_words = 4,
236  .offset_bits = 0,
237  .size_bits = 128 },
239  .offset_words = 8,
240  .offset_bits = 0,
241  .size_bits = 32 },
242  { MCMEMBER_REC_FIELD(mlid),
243  .offset_words = 9,
244  .offset_bits = 0,
245  .size_bits = 16 },
246  { MCMEMBER_REC_FIELD(mtu_selector),
247  .offset_words = 9,
248  .offset_bits = 16,
249  .size_bits = 2 },
250  { MCMEMBER_REC_FIELD(mtu),
251  .offset_words = 9,
252  .offset_bits = 18,
253  .size_bits = 6 },
254  { MCMEMBER_REC_FIELD(traffic_class),
255  .offset_words = 9,
256  .offset_bits = 24,
257  .size_bits = 8 },
259  .offset_words = 10,
260  .offset_bits = 0,
261  .size_bits = 16 },
262  { MCMEMBER_REC_FIELD(rate_selector),
263  .offset_words = 10,
264  .offset_bits = 16,
265  .size_bits = 2 },
267  .offset_words = 10,
268  .offset_bits = 18,
269  .size_bits = 6 },
270  { MCMEMBER_REC_FIELD(packet_life_time_selector),
271  .offset_words = 10,
272  .offset_bits = 24,
273  .size_bits = 2 },
274  { MCMEMBER_REC_FIELD(packet_life_time),
275  .offset_words = 10,
276  .offset_bits = 26,
277  .size_bits = 6 },
279  .offset_words = 11,
280  .offset_bits = 0,
281  .size_bits = 4 },
282  { MCMEMBER_REC_FIELD(flow_label),
283  .offset_words = 11,
284  .offset_bits = 4,
285  .size_bits = 20 },
287  .offset_words = 11,
288  .offset_bits = 24,
289  .size_bits = 8 },
291  .offset_words = 12,
292  .offset_bits = 0,
293  .size_bits = 4 },
294  { MCMEMBER_REC_FIELD(join_state),
295  .offset_words = 12,
296  .offset_bits = 4,
297  .size_bits = 4 },
298  { MCMEMBER_REC_FIELD(proxy_join),
299  .offset_words = 12,
300  .offset_bits = 8,
301  .size_bits = 1 },
302  { RESERVED,
303  .offset_words = 12,
304  .offset_bits = 9,
305  .size_bits = 23 },
306 };
307 
308 #define SERVICE_REC_FIELD(field) \
309  .struct_offset_bytes = offsetof(struct ib_sa_service_rec, field), \
310  .struct_size_bytes = sizeof ((struct ib_sa_service_rec *) 0)->field, \
311  .field_name = "sa_service_rec:" #field
312 
313 static const struct ib_field service_rec_table[] = {
314  { SERVICE_REC_FIELD(id),
315  .offset_words = 0,
316  .offset_bits = 0,
317  .size_bits = 64 },
319  .offset_words = 2,
320  .offset_bits = 0,
321  .size_bits = 128 },
323  .offset_words = 6,
324  .offset_bits = 0,
325  .size_bits = 16 },
326  { SERVICE_REC_FIELD(lease),
327  .offset_words = 7,
328  .offset_bits = 0,
329  .size_bits = 32 },
331  .offset_words = 8,
332  .offset_bits = 0,
333  .size_bits = 128 },
335  .offset_words = 12,
336  .offset_bits = 0,
337  .size_bits = 64*8 },
338  { SERVICE_REC_FIELD(data8),
339  .offset_words = 28,
340  .offset_bits = 0,
341  .size_bits = 16*8 },
342  { SERVICE_REC_FIELD(data16),
343  .offset_words = 32,
344  .offset_bits = 0,
345  .size_bits = 8*16 },
346  { SERVICE_REC_FIELD(data32),
347  .offset_words = 36,
348  .offset_bits = 0,
349  .size_bits = 4*32 },
350  { SERVICE_REC_FIELD(data64),
351  .offset_words = 40,
352  .offset_bits = 0,
353  .size_bits = 2*64 },
354 };
355 
356 #define GUIDINFO_REC_FIELD(field) \
357  .struct_offset_bytes = offsetof(struct ib_sa_guidinfo_rec, field), \
358  .struct_size_bytes = sizeof((struct ib_sa_guidinfo_rec *) 0)->field, \
359  .field_name = "sa_guidinfo_rec:" #field
360 
361 static const struct ib_field guidinfo_rec_table[] = {
363  .offset_words = 0,
364  .offset_bits = 0,
365  .size_bits = 16 },
366  { GUIDINFO_REC_FIELD(block_num),
367  .offset_words = 0,
368  .offset_bits = 16,
369  .size_bits = 8 },
371  .offset_words = 0,
372  .offset_bits = 24,
373  .size_bits = 8 },
375  .offset_words = 1,
376  .offset_bits = 0,
377  .size_bits = 32 },
378  { GUIDINFO_REC_FIELD(guid_info_list),
379  .offset_words = 2,
380  .offset_bits = 0,
381  .size_bits = 512 },
382 };
383 
384 static void free_sm_ah(struct kref *kref)
385 {
386  struct ib_sa_sm_ah *sm_ah = container_of(kref, struct ib_sa_sm_ah, ref);
387 
388  ib_destroy_ah(sm_ah->ah);
389  kfree(sm_ah);
390 }
391 
392 static void update_sm_ah(struct work_struct *work)
393 {
394  struct ib_sa_port *port =
395  container_of(work, struct ib_sa_port, update_task);
396  struct ib_sa_sm_ah *new_ah;
397  struct ib_port_attr port_attr;
398  struct ib_ah_attr ah_attr;
399 
400  if (ib_query_port(port->agent->device, port->port_num, &port_attr)) {
401  printk(KERN_WARNING "Couldn't query port\n");
402  return;
403  }
404 
405  new_ah = kmalloc(sizeof *new_ah, GFP_KERNEL);
406  if (!new_ah) {
407  printk(KERN_WARNING "Couldn't allocate new SM AH\n");
408  return;
409  }
410 
411  kref_init(&new_ah->ref);
412  new_ah->src_path_mask = (1 << port_attr.lmc) - 1;
413 
414  new_ah->pkey_index = 0;
415  if (ib_find_pkey(port->agent->device, port->port_num,
416  IB_DEFAULT_PKEY_FULL, &new_ah->pkey_index))
417  printk(KERN_ERR "Couldn't find index for default PKey\n");
418 
419  memset(&ah_attr, 0, sizeof ah_attr);
420  ah_attr.dlid = port_attr.sm_lid;
421  ah_attr.sl = port_attr.sm_sl;
422  ah_attr.port_num = port->port_num;
423 
424  new_ah->ah = ib_create_ah(port->agent->qp->pd, &ah_attr);
425  if (IS_ERR(new_ah->ah)) {
426  printk(KERN_WARNING "Couldn't create new SM AH\n");
427  kfree(new_ah);
428  return;
429  }
430 
431  spin_lock_irq(&port->ah_lock);
432  if (port->sm_ah)
433  kref_put(&port->sm_ah->ref, free_sm_ah);
434  port->sm_ah = new_ah;
435  spin_unlock_irq(&port->ah_lock);
436 
437 }
438 
439 static void ib_sa_event(struct ib_event_handler *handler, struct ib_event *event)
440 {
441  if (event->event == IB_EVENT_PORT_ERR ||
442  event->event == IB_EVENT_PORT_ACTIVE ||
443  event->event == IB_EVENT_LID_CHANGE ||
444  event->event == IB_EVENT_PKEY_CHANGE ||
445  event->event == IB_EVENT_SM_CHANGE ||
446  event->event == IB_EVENT_CLIENT_REREGISTER) {
447  unsigned long flags;
448  struct ib_sa_device *sa_dev =
449  container_of(handler, typeof(*sa_dev), event_handler);
450  struct ib_sa_port *port =
451  &sa_dev->port[event->element.port_num - sa_dev->start_port];
452 
454  return;
455 
456  spin_lock_irqsave(&port->ah_lock, flags);
457  if (port->sm_ah)
458  kref_put(&port->sm_ah->ref, free_sm_ah);
459  port->sm_ah = NULL;
460  spin_unlock_irqrestore(&port->ah_lock, flags);
461 
462  queue_work(ib_wq, &sa_dev->port[event->element.port_num -
463  sa_dev->start_port].update_task);
464  }
465 }
466 
468 {
469  atomic_set(&client->users, 1);
470  init_completion(&client->comp);
471 }
473 
475 {
476  ib_sa_client_put(client);
477  wait_for_completion(&client->comp);
478 }
480 
490 void ib_sa_cancel_query(int id, struct ib_sa_query *query)
491 {
492  unsigned long flags;
493  struct ib_mad_agent *agent;
494  struct ib_mad_send_buf *mad_buf;
495 
496  spin_lock_irqsave(&idr_lock, flags);
497  if (idr_find(&query_idr, id) != query) {
498  spin_unlock_irqrestore(&idr_lock, flags);
499  return;
500  }
501  agent = query->port->agent;
502  mad_buf = query->mad_buf;
503  spin_unlock_irqrestore(&idr_lock, flags);
504 
505  ib_cancel_mad(agent, mad_buf);
506 }
508 
509 static u8 get_src_path_mask(struct ib_device *device, u8 port_num)
510 {
511  struct ib_sa_device *sa_dev;
512  struct ib_sa_port *port;
513  unsigned long flags;
514  u8 src_path_mask;
515 
516  sa_dev = ib_get_client_data(device, &sa_client);
517  if (!sa_dev)
518  return 0x7f;
519 
520  port = &sa_dev->port[port_num - sa_dev->start_port];
521  spin_lock_irqsave(&port->ah_lock, flags);
522  src_path_mask = port->sm_ah ? port->sm_ah->src_path_mask : 0x7f;
523  spin_unlock_irqrestore(&port->ah_lock, flags);
524 
525  return src_path_mask;
526 }
527 
528 int ib_init_ah_from_path(struct ib_device *device, u8 port_num,
529  struct ib_sa_path_rec *rec, struct ib_ah_attr *ah_attr)
530 {
531  int ret;
532  u16 gid_index;
533  int force_grh;
534 
535  memset(ah_attr, 0, sizeof *ah_attr);
536  ah_attr->dlid = be16_to_cpu(rec->dlid);
537  ah_attr->sl = rec->sl;
538  ah_attr->src_path_bits = be16_to_cpu(rec->slid) &
539  get_src_path_mask(device, port_num);
540  ah_attr->port_num = port_num;
541  ah_attr->static_rate = rec->rate;
542 
543  force_grh = rdma_port_get_link_layer(device, port_num) == IB_LINK_LAYER_ETHERNET;
544 
545  if (rec->hop_limit > 1 || force_grh) {
546  ah_attr->ah_flags = IB_AH_GRH;
547  ah_attr->grh.dgid = rec->dgid;
548 
549  ret = ib_find_cached_gid(device, &rec->sgid, &port_num,
550  &gid_index);
551  if (ret)
552  return ret;
553 
554  ah_attr->grh.sgid_index = gid_index;
555  ah_attr->grh.flow_label = be32_to_cpu(rec->flow_label);
556  ah_attr->grh.hop_limit = rec->hop_limit;
557  ah_attr->grh.traffic_class = rec->traffic_class;
558  }
559  return 0;
560 }
562 
563 static int alloc_mad(struct ib_sa_query *query, gfp_t gfp_mask)
564 {
565  unsigned long flags;
566 
567  spin_lock_irqsave(&query->port->ah_lock, flags);
568  if (!query->port->sm_ah) {
569  spin_unlock_irqrestore(&query->port->ah_lock, flags);
570  return -EAGAIN;
571  }
572  kref_get(&query->port->sm_ah->ref);
573  query->sm_ah = query->port->sm_ah;
574  spin_unlock_irqrestore(&query->port->ah_lock, flags);
575 
576  query->mad_buf = ib_create_send_mad(query->port->agent, 1,
577  query->sm_ah->pkey_index,
579  gfp_mask);
580  if (IS_ERR(query->mad_buf)) {
581  kref_put(&query->sm_ah->ref, free_sm_ah);
582  return -ENOMEM;
583  }
584 
585  query->mad_buf->ah = query->sm_ah->ah;
586 
587  return 0;
588 }
589 
590 static void free_mad(struct ib_sa_query *query)
591 {
592  ib_free_send_mad(query->mad_buf);
593  kref_put(&query->sm_ah->ref, free_sm_ah);
594 }
595 
596 static void init_mad(struct ib_sa_mad *mad, struct ib_mad_agent *agent)
597 {
598  unsigned long flags;
599 
600  memset(mad, 0, sizeof *mad);
601 
602  mad->mad_hdr.base_version = IB_MGMT_BASE_VERSION;
603  mad->mad_hdr.mgmt_class = IB_MGMT_CLASS_SUBN_ADM;
604  mad->mad_hdr.class_version = IB_SA_CLASS_VERSION;
605 
606  spin_lock_irqsave(&tid_lock, flags);
607  mad->mad_hdr.tid =
608  cpu_to_be64(((u64) agent->hi_tid) << 32 | tid++);
609  spin_unlock_irqrestore(&tid_lock, flags);
610 }
611 
612 static int send_mad(struct ib_sa_query *query, int timeout_ms, gfp_t gfp_mask)
613 {
614  unsigned long flags;
615  int ret, id;
616 
617 retry:
618  if (!idr_pre_get(&query_idr, gfp_mask))
619  return -ENOMEM;
620  spin_lock_irqsave(&idr_lock, flags);
621  ret = idr_get_new(&query_idr, query, &id);
622  spin_unlock_irqrestore(&idr_lock, flags);
623  if (ret == -EAGAIN)
624  goto retry;
625  if (ret)
626  return ret;
627 
628  query->mad_buf->timeout_ms = timeout_ms;
629  query->mad_buf->context[0] = query;
630  query->id = id;
631 
632  ret = ib_post_send_mad(query->mad_buf, NULL);
633  if (ret) {
634  spin_lock_irqsave(&idr_lock, flags);
635  idr_remove(&query_idr, id);
636  spin_unlock_irqrestore(&idr_lock, flags);
637  }
638 
639  /*
640  * It's not safe to dereference query any more, because the
641  * send may already have completed and freed the query in
642  * another context.
643  */
644  return ret ? ret : id;
645 }
646 
648 {
649  ib_unpack(path_rec_table, ARRAY_SIZE(path_rec_table), attribute, rec);
650 }
652 
653 static void ib_sa_path_rec_callback(struct ib_sa_query *sa_query,
654  int status,
655  struct ib_sa_mad *mad)
656 {
657  struct ib_sa_path_query *query =
658  container_of(sa_query, struct ib_sa_path_query, sa_query);
659 
660  if (mad) {
661  struct ib_sa_path_rec rec;
662 
663  ib_unpack(path_rec_table, ARRAY_SIZE(path_rec_table),
664  mad->data, &rec);
665  query->callback(status, &rec, query->context);
666  } else
667  query->callback(status, NULL, query->context);
668 }
669 
670 static void ib_sa_path_rec_release(struct ib_sa_query *sa_query)
671 {
672  kfree(container_of(sa_query, struct ib_sa_path_query, sa_query));
673 }
674 
701  struct ib_device *device, u8 port_num,
702  struct ib_sa_path_rec *rec,
704  int timeout_ms, gfp_t gfp_mask,
705  void (*callback)(int status,
706  struct ib_sa_path_rec *resp,
707  void *context),
708  void *context,
709  struct ib_sa_query **sa_query)
710 {
711  struct ib_sa_path_query *query;
712  struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
713  struct ib_sa_port *port;
714  struct ib_mad_agent *agent;
715  struct ib_sa_mad *mad;
716  int ret;
717 
718  if (!sa_dev)
719  return -ENODEV;
720 
721  port = &sa_dev->port[port_num - sa_dev->start_port];
722  agent = port->agent;
723 
724  query = kmalloc(sizeof *query, gfp_mask);
725  if (!query)
726  return -ENOMEM;
727 
728  query->sa_query.port = port;
729  ret = alloc_mad(&query->sa_query, gfp_mask);
730  if (ret)
731  goto err1;
732 
733  ib_sa_client_get(client);
734  query->sa_query.client = client;
735  query->callback = callback;
736  query->context = context;
737 
738  mad = query->sa_query.mad_buf->mad;
739  init_mad(mad, agent);
740 
741  query->sa_query.callback = callback ? ib_sa_path_rec_callback : NULL;
742  query->sa_query.release = ib_sa_path_rec_release;
743  mad->mad_hdr.method = IB_MGMT_METHOD_GET;
744  mad->mad_hdr.attr_id = cpu_to_be16(IB_SA_ATTR_PATH_REC);
745  mad->sa_hdr.comp_mask = comp_mask;
746 
747  ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table), rec, mad->data);
748 
749  *sa_query = &query->sa_query;
750 
751  ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
752  if (ret < 0)
753  goto err2;
754 
755  return ret;
756 
757 err2:
758  *sa_query = NULL;
759  ib_sa_client_put(query->sa_query.client);
760  free_mad(&query->sa_query);
761 
762 err1:
763  kfree(query);
764  return ret;
765 }
767 
768 static void ib_sa_service_rec_callback(struct ib_sa_query *sa_query,
769  int status,
770  struct ib_sa_mad *mad)
771 {
772  struct ib_sa_service_query *query =
773  container_of(sa_query, struct ib_sa_service_query, sa_query);
774 
775  if (mad) {
776  struct ib_sa_service_rec rec;
777 
778  ib_unpack(service_rec_table, ARRAY_SIZE(service_rec_table),
779  mad->data, &rec);
780  query->callback(status, &rec, query->context);
781  } else
782  query->callback(status, NULL, query->context);
783 }
784 
785 static void ib_sa_service_rec_release(struct ib_sa_query *sa_query)
786 {
787  kfree(container_of(sa_query, struct ib_sa_service_query, sa_query));
788 }
789 
818  struct ib_device *device, u8 port_num, u8 method,
819  struct ib_sa_service_rec *rec,
820  ib_sa_comp_mask comp_mask,
821  int timeout_ms, gfp_t gfp_mask,
822  void (*callback)(int status,
823  struct ib_sa_service_rec *resp,
824  void *context),
825  void *context,
826  struct ib_sa_query **sa_query)
827 {
828  struct ib_sa_service_query *query;
829  struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
830  struct ib_sa_port *port;
831  struct ib_mad_agent *agent;
832  struct ib_sa_mad *mad;
833  int ret;
834 
835  if (!sa_dev)
836  return -ENODEV;
837 
838  port = &sa_dev->port[port_num - sa_dev->start_port];
839  agent = port->agent;
840 
841  if (method != IB_MGMT_METHOD_GET &&
842  method != IB_MGMT_METHOD_SET &&
843  method != IB_SA_METHOD_DELETE)
844  return -EINVAL;
845 
846  query = kmalloc(sizeof *query, gfp_mask);
847  if (!query)
848  return -ENOMEM;
849 
850  query->sa_query.port = port;
851  ret = alloc_mad(&query->sa_query, gfp_mask);
852  if (ret)
853  goto err1;
854 
855  ib_sa_client_get(client);
856  query->sa_query.client = client;
857  query->callback = callback;
858  query->context = context;
859 
860  mad = query->sa_query.mad_buf->mad;
861  init_mad(mad, agent);
862 
863  query->sa_query.callback = callback ? ib_sa_service_rec_callback : NULL;
864  query->sa_query.release = ib_sa_service_rec_release;
865  mad->mad_hdr.method = method;
867  mad->sa_hdr.comp_mask = comp_mask;
868 
869  ib_pack(service_rec_table, ARRAY_SIZE(service_rec_table),
870  rec, mad->data);
871 
872  *sa_query = &query->sa_query;
873 
874  ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
875  if (ret < 0)
876  goto err2;
877 
878  return ret;
879 
880 err2:
881  *sa_query = NULL;
882  ib_sa_client_put(query->sa_query.client);
883  free_mad(&query->sa_query);
884 
885 err1:
886  kfree(query);
887  return ret;
888 }
890 
891 static void ib_sa_mcmember_rec_callback(struct ib_sa_query *sa_query,
892  int status,
893  struct ib_sa_mad *mad)
894 {
895  struct ib_sa_mcmember_query *query =
896  container_of(sa_query, struct ib_sa_mcmember_query, sa_query);
897 
898  if (mad) {
899  struct ib_sa_mcmember_rec rec;
900 
901  ib_unpack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
902  mad->data, &rec);
903  query->callback(status, &rec, query->context);
904  } else
905  query->callback(status, NULL, query->context);
906 }
907 
908 static void ib_sa_mcmember_rec_release(struct ib_sa_query *sa_query)
909 {
910  kfree(container_of(sa_query, struct ib_sa_mcmember_query, sa_query));
911 }
912 
914  struct ib_device *device, u8 port_num,
915  u8 method,
916  struct ib_sa_mcmember_rec *rec,
917  ib_sa_comp_mask comp_mask,
918  int timeout_ms, gfp_t gfp_mask,
919  void (*callback)(int status,
920  struct ib_sa_mcmember_rec *resp,
921  void *context),
922  void *context,
923  struct ib_sa_query **sa_query)
924 {
925  struct ib_sa_mcmember_query *query;
926  struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
927  struct ib_sa_port *port;
928  struct ib_mad_agent *agent;
929  struct ib_sa_mad *mad;
930  int ret;
931 
932  if (!sa_dev)
933  return -ENODEV;
934 
935  port = &sa_dev->port[port_num - sa_dev->start_port];
936  agent = port->agent;
937 
938  query = kmalloc(sizeof *query, gfp_mask);
939  if (!query)
940  return -ENOMEM;
941 
942  query->sa_query.port = port;
943  ret = alloc_mad(&query->sa_query, gfp_mask);
944  if (ret)
945  goto err1;
946 
947  ib_sa_client_get(client);
948  query->sa_query.client = client;
949  query->callback = callback;
950  query->context = context;
951 
952  mad = query->sa_query.mad_buf->mad;
953  init_mad(mad, agent);
954 
955  query->sa_query.callback = callback ? ib_sa_mcmember_rec_callback : NULL;
956  query->sa_query.release = ib_sa_mcmember_rec_release;
957  mad->mad_hdr.method = method;
959  mad->sa_hdr.comp_mask = comp_mask;
960 
961  ib_pack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
962  rec, mad->data);
963 
964  *sa_query = &query->sa_query;
965 
966  ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
967  if (ret < 0)
968  goto err2;
969 
970  return ret;
971 
972 err2:
973  *sa_query = NULL;
974  ib_sa_client_put(query->sa_query.client);
975  free_mad(&query->sa_query);
976 
977 err1:
978  kfree(query);
979  return ret;
980 }
981 
982 /* Support GuidInfoRecord */
983 static void ib_sa_guidinfo_rec_callback(struct ib_sa_query *sa_query,
984  int status,
985  struct ib_sa_mad *mad)
986 {
987  struct ib_sa_guidinfo_query *query =
988  container_of(sa_query, struct ib_sa_guidinfo_query, sa_query);
989 
990  if (mad) {
991  struct ib_sa_guidinfo_rec rec;
992 
993  ib_unpack(guidinfo_rec_table, ARRAY_SIZE(guidinfo_rec_table),
994  mad->data, &rec);
995  query->callback(status, &rec, query->context);
996  } else
997  query->callback(status, NULL, query->context);
998 }
999 
1000 static void ib_sa_guidinfo_rec_release(struct ib_sa_query *sa_query)
1001 {
1002  kfree(container_of(sa_query, struct ib_sa_guidinfo_query, sa_query));
1003 }
1004 
1006  struct ib_device *device, u8 port_num,
1007  struct ib_sa_guidinfo_rec *rec,
1008  ib_sa_comp_mask comp_mask, u8 method,
1009  int timeout_ms, gfp_t gfp_mask,
1010  void (*callback)(int status,
1011  struct ib_sa_guidinfo_rec *resp,
1012  void *context),
1013  void *context,
1014  struct ib_sa_query **sa_query)
1015 {
1016  struct ib_sa_guidinfo_query *query;
1017  struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1018  struct ib_sa_port *port;
1019  struct ib_mad_agent *agent;
1020  struct ib_sa_mad *mad;
1021  int ret;
1022 
1023  if (!sa_dev)
1024  return -ENODEV;
1025 
1026  if (method != IB_MGMT_METHOD_GET &&
1027  method != IB_MGMT_METHOD_SET &&
1028  method != IB_SA_METHOD_DELETE) {
1029  return -EINVAL;
1030  }
1031 
1032  port = &sa_dev->port[port_num - sa_dev->start_port];
1033  agent = port->agent;
1034 
1035  query = kmalloc(sizeof *query, gfp_mask);
1036  if (!query)
1037  return -ENOMEM;
1038 
1039  query->sa_query.port = port;
1040  ret = alloc_mad(&query->sa_query, gfp_mask);
1041  if (ret)
1042  goto err1;
1043 
1044  ib_sa_client_get(client);
1045  query->sa_query.client = client;
1046  query->callback = callback;
1047  query->context = context;
1048 
1049  mad = query->sa_query.mad_buf->mad;
1050  init_mad(mad, agent);
1051 
1052  query->sa_query.callback = callback ? ib_sa_guidinfo_rec_callback : NULL;
1053  query->sa_query.release = ib_sa_guidinfo_rec_release;
1054 
1055  mad->mad_hdr.method = method;
1057  mad->sa_hdr.comp_mask = comp_mask;
1058 
1059  ib_pack(guidinfo_rec_table, ARRAY_SIZE(guidinfo_rec_table), rec,
1060  mad->data);
1061 
1062  *sa_query = &query->sa_query;
1063 
1064  ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1065  if (ret < 0)
1066  goto err2;
1067 
1068  return ret;
1069 
1070 err2:
1071  *sa_query = NULL;
1072  ib_sa_client_put(query->sa_query.client);
1073  free_mad(&query->sa_query);
1074 
1075 err1:
1076  kfree(query);
1077  return ret;
1078 }
1080 
1081 static void send_handler(struct ib_mad_agent *agent,
1082  struct ib_mad_send_wc *mad_send_wc)
1083 {
1084  struct ib_sa_query *query = mad_send_wc->send_buf->context[0];
1085  unsigned long flags;
1086 
1087  if (query->callback)
1088  switch (mad_send_wc->status) {
1089  case IB_WC_SUCCESS:
1090  /* No callback -- already got recv */
1091  break;
1093  query->callback(query, -ETIMEDOUT, NULL);
1094  break;
1095  case IB_WC_WR_FLUSH_ERR:
1096  query->callback(query, -EINTR, NULL);
1097  break;
1098  default:
1099  query->callback(query, -EIO, NULL);
1100  break;
1101  }
1102 
1103  spin_lock_irqsave(&idr_lock, flags);
1104  idr_remove(&query_idr, query->id);
1105  spin_unlock_irqrestore(&idr_lock, flags);
1106 
1107  free_mad(query);
1108  ib_sa_client_put(query->client);
1109  query->release(query);
1110 }
1111 
1112 static void recv_handler(struct ib_mad_agent *mad_agent,
1113  struct ib_mad_recv_wc *mad_recv_wc)
1114 {
1115  struct ib_sa_query *query;
1116  struct ib_mad_send_buf *mad_buf;
1117 
1118  mad_buf = (void *) (unsigned long) mad_recv_wc->wc->wr_id;
1119  query = mad_buf->context[0];
1120 
1121  if (query->callback) {
1122  if (mad_recv_wc->wc->status == IB_WC_SUCCESS)
1123  query->callback(query,
1124  mad_recv_wc->recv_buf.mad->mad_hdr.status ?
1125  -EINVAL : 0,
1126  (struct ib_sa_mad *) mad_recv_wc->recv_buf.mad);
1127  else
1128  query->callback(query, -EIO, NULL);
1129  }
1130 
1131  ib_free_recv_mad(mad_recv_wc);
1132 }
1133 
1134 static void ib_sa_add_one(struct ib_device *device)
1135 {
1136  struct ib_sa_device *sa_dev;
1137  int s, e, i;
1138 
1140  return;
1141 
1142  if (device->node_type == RDMA_NODE_IB_SWITCH)
1143  s = e = 0;
1144  else {
1145  s = 1;
1146  e = device->phys_port_cnt;
1147  }
1148 
1149  sa_dev = kzalloc(sizeof *sa_dev +
1150  (e - s + 1) * sizeof (struct ib_sa_port),
1151  GFP_KERNEL);
1152  if (!sa_dev)
1153  return;
1154 
1155  sa_dev->start_port = s;
1156  sa_dev->end_port = e;
1157 
1158  for (i = 0; i <= e - s; ++i) {
1159  spin_lock_init(&sa_dev->port[i].ah_lock);
1160  if (rdma_port_get_link_layer(device, i + 1) != IB_LINK_LAYER_INFINIBAND)
1161  continue;
1162 
1163  sa_dev->port[i].sm_ah = NULL;
1164  sa_dev->port[i].port_num = i + s;
1165 
1166  sa_dev->port[i].agent =
1167  ib_register_mad_agent(device, i + s, IB_QPT_GSI,
1168  NULL, 0, send_handler,
1169  recv_handler, sa_dev);
1170  if (IS_ERR(sa_dev->port[i].agent))
1171  goto err;
1172 
1173  INIT_WORK(&sa_dev->port[i].update_task, update_sm_ah);
1174  }
1175 
1176  ib_set_client_data(device, &sa_client, sa_dev);
1177 
1178  /*
1179  * We register our event handler after everything is set up,
1180  * and then update our cached info after the event handler is
1181  * registered to avoid any problems if a port changes state
1182  * during our initialization.
1183  */
1184 
1185  INIT_IB_EVENT_HANDLER(&sa_dev->event_handler, device, ib_sa_event);
1187  goto err;
1188 
1189  for (i = 0; i <= e - s; ++i)
1190  if (rdma_port_get_link_layer(device, i + 1) == IB_LINK_LAYER_INFINIBAND)
1191  update_sm_ah(&sa_dev->port[i].update_task);
1192 
1193  return;
1194 
1195 err:
1196  while (--i >= 0)
1197  if (rdma_port_get_link_layer(device, i + 1) == IB_LINK_LAYER_INFINIBAND)
1198  ib_unregister_mad_agent(sa_dev->port[i].agent);
1199 
1200  kfree(sa_dev);
1201 
1202  return;
1203 }
1204 
1205 static void ib_sa_remove_one(struct ib_device *device)
1206 {
1207  struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1208  int i;
1209 
1210  if (!sa_dev)
1211  return;
1212 
1214 
1216 
1217  for (i = 0; i <= sa_dev->end_port - sa_dev->start_port; ++i) {
1218  if (rdma_port_get_link_layer(device, i + 1) == IB_LINK_LAYER_INFINIBAND) {
1219  ib_unregister_mad_agent(sa_dev->port[i].agent);
1220  if (sa_dev->port[i].sm_ah)
1221  kref_put(&sa_dev->port[i].sm_ah->ref, free_sm_ah);
1222  }
1223 
1224  }
1225 
1226  kfree(sa_dev);
1227 }
1228 
1229 static int __init ib_sa_init(void)
1230 {
1231  int ret;
1232 
1233  get_random_bytes(&tid, sizeof tid);
1234 
1235  ret = ib_register_client(&sa_client);
1236  if (ret) {
1237  printk(KERN_ERR "Couldn't register ib_sa client\n");
1238  goto err1;
1239  }
1240 
1241  ret = mcast_init();
1242  if (ret) {
1243  printk(KERN_ERR "Couldn't initialize multicast handling\n");
1244  goto err2;
1245  }
1246 
1247  return 0;
1248 err2:
1249  ib_unregister_client(&sa_client);
1250 err1:
1251  return ret;
1252 }
1253 
1254 static void __exit ib_sa_cleanup(void)
1255 {
1256  mcast_cleanup();
1257  ib_unregister_client(&sa_client);
1258  idr_destroy(&query_idr);
1259 }
1260 
1261 module_init(ib_sa_init);
1262 module_exit(ib_sa_cleanup);