Merge /pub/scm/linux/kernel/git/torvalds/linux-2.6
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / drivers / infiniband / core / sa_query.c
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 * $Id: sa_query.c 2811 2005-07-06 18:11:43Z halr $
35 */
36
37 #include <linux/module.h>
38 #include <linux/init.h>
39 #include <linux/err.h>
40 #include <linux/random.h>
41 #include <linux/spinlock.h>
42 #include <linux/slab.h>
43 #include <linux/pci.h>
44 #include <linux/dma-mapping.h>
45 #include <linux/kref.h>
46 #include <linux/idr.h>
47 #include <linux/workqueue.h>
48
49 #include <rdma/ib_pack.h>
50 #include <rdma/ib_cache.h>
51 #include "sa.h"
52
53 MODULE_AUTHOR("Roland Dreier");
54 MODULE_DESCRIPTION("InfiniBand subnet administration query support");
55 MODULE_LICENSE("Dual BSD/GPL");
56
57 struct ib_sa_sm_ah {
58 struct ib_ah *ah;
59 struct kref ref;
60 u8 src_path_mask;
61 };
62
63 struct ib_sa_port {
64 struct ib_mad_agent *agent;
65 struct ib_sa_sm_ah *sm_ah;
66 struct work_struct update_task;
67 spinlock_t ah_lock;
68 u8 port_num;
69 };
70
71 struct ib_sa_device {
72 int start_port, end_port;
73 struct ib_event_handler event_handler;
74 struct ib_sa_port port[0];
75 };
76
77 struct ib_sa_query {
78 void (*callback)(struct ib_sa_query *, int, struct ib_sa_mad *);
79 void (*release)(struct ib_sa_query *);
80 struct ib_sa_client *client;
81 struct ib_sa_port *port;
82 struct ib_mad_send_buf *mad_buf;
83 struct ib_sa_sm_ah *sm_ah;
84 int id;
85 };
86
87 struct ib_sa_service_query {
88 void (*callback)(int, struct ib_sa_service_rec *, void *);
89 void *context;
90 struct ib_sa_query sa_query;
91 };
92
93 struct ib_sa_path_query {
94 void (*callback)(int, struct ib_sa_path_rec *, void *);
95 void *context;
96 struct ib_sa_query sa_query;
97 };
98
99 struct ib_sa_mcmember_query {
100 void (*callback)(int, struct ib_sa_mcmember_rec *, void *);
101 void *context;
102 struct ib_sa_query sa_query;
103 };
104
105 static void ib_sa_add_one(struct ib_device *device);
106 static void ib_sa_remove_one(struct ib_device *device);
107
108 static struct ib_client sa_client = {
109 .name = "sa",
110 .add = ib_sa_add_one,
111 .remove = ib_sa_remove_one
112 };
113
114 static spinlock_t idr_lock;
115 static DEFINE_IDR(query_idr);
116
117 static spinlock_t tid_lock;
118 static u32 tid;
119
120 #define PATH_REC_FIELD(field) \
121 .struct_offset_bytes = offsetof(struct ib_sa_path_rec, field), \
122 .struct_size_bytes = sizeof ((struct ib_sa_path_rec *) 0)->field, \
123 .field_name = "sa_path_rec:" #field
124
125 static const struct ib_field path_rec_table[] = {
126 { RESERVED,
127 .offset_words = 0,
128 .offset_bits = 0,
129 .size_bits = 32 },
130 { RESERVED,
131 .offset_words = 1,
132 .offset_bits = 0,
133 .size_bits = 32 },
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 },
162 { PATH_REC_FIELD(hop_limit),
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 { RESERVED,
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 },
238 { MCMEMBER_REC_FIELD(qkey),
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 },
258 { MCMEMBER_REC_FIELD(pkey),
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 },
266 { MCMEMBER_REC_FIELD(rate),
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 },
278 { MCMEMBER_REC_FIELD(sl),
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 },
286 { MCMEMBER_REC_FIELD(hop_limit),
287 .offset_words = 11,
288 .offset_bits = 24,
289 .size_bits = 8 },
290 { MCMEMBER_REC_FIELD(scope),
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 },
318 { SERVICE_REC_FIELD(gid),
319 .offset_words = 2,
320 .offset_bits = 0,
321 .size_bits = 128 },
322 { SERVICE_REC_FIELD(pkey),
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 },
330 { SERVICE_REC_FIELD(key),
331 .offset_words = 8,
332 .offset_bits = 0,
333 .size_bits = 128 },
334 { SERVICE_REC_FIELD(name),
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 static void free_sm_ah(struct kref *kref)
357 {
358 struct ib_sa_sm_ah *sm_ah = container_of(kref, struct ib_sa_sm_ah, ref);
359
360 ib_destroy_ah(sm_ah->ah);
361 kfree(sm_ah);
362 }
363
364 static void update_sm_ah(struct work_struct *work)
365 {
366 struct ib_sa_port *port =
367 container_of(work, struct ib_sa_port, update_task);
368 struct ib_sa_sm_ah *new_ah, *old_ah;
369 struct ib_port_attr port_attr;
370 struct ib_ah_attr ah_attr;
371
372 if (ib_query_port(port->agent->device, port->port_num, &port_attr)) {
373 printk(KERN_WARNING "Couldn't query port\n");
374 return;
375 }
376
377 new_ah = kmalloc(sizeof *new_ah, GFP_KERNEL);
378 if (!new_ah) {
379 printk(KERN_WARNING "Couldn't allocate new SM AH\n");
380 return;
381 }
382
383 kref_init(&new_ah->ref);
384 new_ah->src_path_mask = (1 << port_attr.lmc) - 1;
385
386 memset(&ah_attr, 0, sizeof ah_attr);
387 ah_attr.dlid = port_attr.sm_lid;
388 ah_attr.sl = port_attr.sm_sl;
389 ah_attr.port_num = port->port_num;
390
391 new_ah->ah = ib_create_ah(port->agent->qp->pd, &ah_attr);
392 if (IS_ERR(new_ah->ah)) {
393 printk(KERN_WARNING "Couldn't create new SM AH\n");
394 kfree(new_ah);
395 return;
396 }
397
398 spin_lock_irq(&port->ah_lock);
399 old_ah = port->sm_ah;
400 port->sm_ah = new_ah;
401 spin_unlock_irq(&port->ah_lock);
402
403 if (old_ah)
404 kref_put(&old_ah->ref, free_sm_ah);
405 }
406
407 static void ib_sa_event(struct ib_event_handler *handler, struct ib_event *event)
408 {
409 if (event->event == IB_EVENT_PORT_ERR ||
410 event->event == IB_EVENT_PORT_ACTIVE ||
411 event->event == IB_EVENT_LID_CHANGE ||
412 event->event == IB_EVENT_PKEY_CHANGE ||
413 event->event == IB_EVENT_SM_CHANGE ||
414 event->event == IB_EVENT_CLIENT_REREGISTER) {
415 struct ib_sa_device *sa_dev;
416 sa_dev = container_of(handler, typeof(*sa_dev), event_handler);
417
418 schedule_work(&sa_dev->port[event->element.port_num -
419 sa_dev->start_port].update_task);
420 }
421 }
422
423 void ib_sa_register_client(struct ib_sa_client *client)
424 {
425 atomic_set(&client->users, 1);
426 init_completion(&client->comp);
427 }
428 EXPORT_SYMBOL(ib_sa_register_client);
429
430 void ib_sa_unregister_client(struct ib_sa_client *client)
431 {
432 ib_sa_client_put(client);
433 wait_for_completion(&client->comp);
434 }
435 EXPORT_SYMBOL(ib_sa_unregister_client);
436
437 /**
438 * ib_sa_cancel_query - try to cancel an SA query
439 * @id:ID of query to cancel
440 * @query:query pointer to cancel
441 *
442 * Try to cancel an SA query. If the id and query don't match up or
443 * the query has already completed, nothing is done. Otherwise the
444 * query is canceled and will complete with a status of -EINTR.
445 */
446 void ib_sa_cancel_query(int id, struct ib_sa_query *query)
447 {
448 unsigned long flags;
449 struct ib_mad_agent *agent;
450 struct ib_mad_send_buf *mad_buf;
451
452 spin_lock_irqsave(&idr_lock, flags);
453 if (idr_find(&query_idr, id) != query) {
454 spin_unlock_irqrestore(&idr_lock, flags);
455 return;
456 }
457 agent = query->port->agent;
458 mad_buf = query->mad_buf;
459 spin_unlock_irqrestore(&idr_lock, flags);
460
461 ib_cancel_mad(agent, mad_buf);
462 }
463 EXPORT_SYMBOL(ib_sa_cancel_query);
464
465 static u8 get_src_path_mask(struct ib_device *device, u8 port_num)
466 {
467 struct ib_sa_device *sa_dev;
468 struct ib_sa_port *port;
469 unsigned long flags;
470 u8 src_path_mask;
471
472 sa_dev = ib_get_client_data(device, &sa_client);
473 if (!sa_dev)
474 return 0x7f;
475
476 port = &sa_dev->port[port_num - sa_dev->start_port];
477 spin_lock_irqsave(&port->ah_lock, flags);
478 src_path_mask = port->sm_ah ? port->sm_ah->src_path_mask : 0x7f;
479 spin_unlock_irqrestore(&port->ah_lock, flags);
480
481 return src_path_mask;
482 }
483
484 int ib_init_ah_from_path(struct ib_device *device, u8 port_num,
485 struct ib_sa_path_rec *rec, struct ib_ah_attr *ah_attr)
486 {
487 int ret;
488 u16 gid_index;
489
490 memset(ah_attr, 0, sizeof *ah_attr);
491 ah_attr->dlid = be16_to_cpu(rec->dlid);
492 ah_attr->sl = rec->sl;
493 ah_attr->src_path_bits = be16_to_cpu(rec->slid) &
494 get_src_path_mask(device, port_num);
495 ah_attr->port_num = port_num;
496 ah_attr->static_rate = rec->rate;
497
498 if (rec->hop_limit > 1) {
499 ah_attr->ah_flags = IB_AH_GRH;
500 ah_attr->grh.dgid = rec->dgid;
501
502 ret = ib_find_cached_gid(device, &rec->sgid, &port_num,
503 &gid_index);
504 if (ret)
505 return ret;
506
507 ah_attr->grh.sgid_index = gid_index;
508 ah_attr->grh.flow_label = be32_to_cpu(rec->flow_label);
509 ah_attr->grh.hop_limit = rec->hop_limit;
510 ah_attr->grh.traffic_class = rec->traffic_class;
511 }
512 return 0;
513 }
514 EXPORT_SYMBOL(ib_init_ah_from_path);
515
516 static void init_mad(struct ib_sa_mad *mad, struct ib_mad_agent *agent)
517 {
518 unsigned long flags;
519
520 memset(mad, 0, sizeof *mad);
521
522 mad->mad_hdr.base_version = IB_MGMT_BASE_VERSION;
523 mad->mad_hdr.mgmt_class = IB_MGMT_CLASS_SUBN_ADM;
524 mad->mad_hdr.class_version = IB_SA_CLASS_VERSION;
525
526 spin_lock_irqsave(&tid_lock, flags);
527 mad->mad_hdr.tid =
528 cpu_to_be64(((u64) agent->hi_tid) << 32 | tid++);
529 spin_unlock_irqrestore(&tid_lock, flags);
530 }
531
532 static int send_mad(struct ib_sa_query *query, int timeout_ms, gfp_t gfp_mask)
533 {
534 unsigned long flags;
535 int ret, id;
536
537 retry:
538 if (!idr_pre_get(&query_idr, gfp_mask))
539 return -ENOMEM;
540 spin_lock_irqsave(&idr_lock, flags);
541 ret = idr_get_new(&query_idr, query, &id);
542 spin_unlock_irqrestore(&idr_lock, flags);
543 if (ret == -EAGAIN)
544 goto retry;
545 if (ret)
546 return ret;
547
548 query->mad_buf->timeout_ms = timeout_ms;
549 query->mad_buf->context[0] = query;
550 query->id = id;
551
552 spin_lock_irqsave(&query->port->ah_lock, flags);
553 kref_get(&query->port->sm_ah->ref);
554 query->sm_ah = query->port->sm_ah;
555 spin_unlock_irqrestore(&query->port->ah_lock, flags);
556
557 query->mad_buf->ah = query->sm_ah->ah;
558
559 ret = ib_post_send_mad(query->mad_buf, NULL);
560 if (ret) {
561 spin_lock_irqsave(&idr_lock, flags);
562 idr_remove(&query_idr, id);
563 spin_unlock_irqrestore(&idr_lock, flags);
564
565 kref_put(&query->sm_ah->ref, free_sm_ah);
566 }
567
568 /*
569 * It's not safe to dereference query any more, because the
570 * send may already have completed and freed the query in
571 * another context.
572 */
573 return ret ? ret : id;
574 }
575
576 static void ib_sa_path_rec_callback(struct ib_sa_query *sa_query,
577 int status,
578 struct ib_sa_mad *mad)
579 {
580 struct ib_sa_path_query *query =
581 container_of(sa_query, struct ib_sa_path_query, sa_query);
582
583 if (mad) {
584 struct ib_sa_path_rec rec;
585
586 ib_unpack(path_rec_table, ARRAY_SIZE(path_rec_table),
587 mad->data, &rec);
588 query->callback(status, &rec, query->context);
589 } else
590 query->callback(status, NULL, query->context);
591 }
592
593 static void ib_sa_path_rec_release(struct ib_sa_query *sa_query)
594 {
595 kfree(container_of(sa_query, struct ib_sa_path_query, sa_query));
596 }
597
598 /**
599 * ib_sa_path_rec_get - Start a Path get query
600 * @client:SA client
601 * @device:device to send query on
602 * @port_num: port number to send query on
603 * @rec:Path Record to send in query
604 * @comp_mask:component mask to send in query
605 * @timeout_ms:time to wait for response
606 * @gfp_mask:GFP mask to use for internal allocations
607 * @callback:function called when query completes, times out or is
608 * canceled
609 * @context:opaque user context passed to callback
610 * @sa_query:query context, used to cancel query
611 *
612 * Send a Path Record Get query to the SA to look up a path. The
613 * callback function will be called when the query completes (or
614 * fails); status is 0 for a successful response, -EINTR if the query
615 * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
616 * occurred sending the query. The resp parameter of the callback is
617 * only valid if status is 0.
618 *
619 * If the return value of ib_sa_path_rec_get() is negative, it is an
620 * error code. Otherwise it is a query ID that can be used to cancel
621 * the query.
622 */
623 int ib_sa_path_rec_get(struct ib_sa_client *client,
624 struct ib_device *device, u8 port_num,
625 struct ib_sa_path_rec *rec,
626 ib_sa_comp_mask comp_mask,
627 int timeout_ms, gfp_t gfp_mask,
628 void (*callback)(int status,
629 struct ib_sa_path_rec *resp,
630 void *context),
631 void *context,
632 struct ib_sa_query **sa_query)
633 {
634 struct ib_sa_path_query *query;
635 struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
636 struct ib_sa_port *port;
637 struct ib_mad_agent *agent;
638 struct ib_sa_mad *mad;
639 int ret;
640
641 if (!sa_dev)
642 return -ENODEV;
643
644 port = &sa_dev->port[port_num - sa_dev->start_port];
645 agent = port->agent;
646
647 query = kmalloc(sizeof *query, gfp_mask);
648 if (!query)
649 return -ENOMEM;
650
651 query->sa_query.mad_buf = ib_create_send_mad(agent, 1, 0,
652 0, IB_MGMT_SA_HDR,
653 IB_MGMT_SA_DATA, gfp_mask);
654 if (!query->sa_query.mad_buf) {
655 ret = -ENOMEM;
656 goto err1;
657 }
658
659 ib_sa_client_get(client);
660 query->sa_query.client = client;
661 query->callback = callback;
662 query->context = context;
663
664 mad = query->sa_query.mad_buf->mad;
665 init_mad(mad, agent);
666
667 query->sa_query.callback = callback ? ib_sa_path_rec_callback : NULL;
668 query->sa_query.release = ib_sa_path_rec_release;
669 query->sa_query.port = port;
670 mad->mad_hdr.method = IB_MGMT_METHOD_GET;
671 mad->mad_hdr.attr_id = cpu_to_be16(IB_SA_ATTR_PATH_REC);
672 mad->sa_hdr.comp_mask = comp_mask;
673
674 ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table), rec, mad->data);
675
676 *sa_query = &query->sa_query;
677
678 ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
679 if (ret < 0)
680 goto err2;
681
682 return ret;
683
684 err2:
685 *sa_query = NULL;
686 ib_sa_client_put(query->sa_query.client);
687 ib_free_send_mad(query->sa_query.mad_buf);
688
689 err1:
690 kfree(query);
691 return ret;
692 }
693 EXPORT_SYMBOL(ib_sa_path_rec_get);
694
695 static void ib_sa_service_rec_callback(struct ib_sa_query *sa_query,
696 int status,
697 struct ib_sa_mad *mad)
698 {
699 struct ib_sa_service_query *query =
700 container_of(sa_query, struct ib_sa_service_query, sa_query);
701
702 if (mad) {
703 struct ib_sa_service_rec rec;
704
705 ib_unpack(service_rec_table, ARRAY_SIZE(service_rec_table),
706 mad->data, &rec);
707 query->callback(status, &rec, query->context);
708 } else
709 query->callback(status, NULL, query->context);
710 }
711
712 static void ib_sa_service_rec_release(struct ib_sa_query *sa_query)
713 {
714 kfree(container_of(sa_query, struct ib_sa_service_query, sa_query));
715 }
716
717 /**
718 * ib_sa_service_rec_query - Start Service Record operation
719 * @client:SA client
720 * @device:device to send request on
721 * @port_num: port number to send request on
722 * @method:SA method - should be get, set, or delete
723 * @rec:Service Record to send in request
724 * @comp_mask:component mask to send in request
725 * @timeout_ms:time to wait for response
726 * @gfp_mask:GFP mask to use for internal allocations
727 * @callback:function called when request completes, times out or is
728 * canceled
729 * @context:opaque user context passed to callback
730 * @sa_query:request context, used to cancel request
731 *
732 * Send a Service Record set/get/delete to the SA to register,
733 * unregister or query a service record.
734 * The callback function will be called when the request completes (or
735 * fails); status is 0 for a successful response, -EINTR if the query
736 * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
737 * occurred sending the query. The resp parameter of the callback is
738 * only valid if status is 0.
739 *
740 * If the return value of ib_sa_service_rec_query() is negative, it is an
741 * error code. Otherwise it is a request ID that can be used to cancel
742 * the query.
743 */
744 int ib_sa_service_rec_query(struct ib_sa_client *client,
745 struct ib_device *device, u8 port_num, u8 method,
746 struct ib_sa_service_rec *rec,
747 ib_sa_comp_mask comp_mask,
748 int timeout_ms, gfp_t gfp_mask,
749 void (*callback)(int status,
750 struct ib_sa_service_rec *resp,
751 void *context),
752 void *context,
753 struct ib_sa_query **sa_query)
754 {
755 struct ib_sa_service_query *query;
756 struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
757 struct ib_sa_port *port;
758 struct ib_mad_agent *agent;
759 struct ib_sa_mad *mad;
760 int ret;
761
762 if (!sa_dev)
763 return -ENODEV;
764
765 port = &sa_dev->port[port_num - sa_dev->start_port];
766 agent = port->agent;
767
768 if (method != IB_MGMT_METHOD_GET &&
769 method != IB_MGMT_METHOD_SET &&
770 method != IB_SA_METHOD_DELETE)
771 return -EINVAL;
772
773 query = kmalloc(sizeof *query, gfp_mask);
774 if (!query)
775 return -ENOMEM;
776
777 query->sa_query.mad_buf = ib_create_send_mad(agent, 1, 0,
778 0, IB_MGMT_SA_HDR,
779 IB_MGMT_SA_DATA, gfp_mask);
780 if (!query->sa_query.mad_buf) {
781 ret = -ENOMEM;
782 goto err1;
783 }
784
785 ib_sa_client_get(client);
786 query->sa_query.client = client;
787 query->callback = callback;
788 query->context = context;
789
790 mad = query->sa_query.mad_buf->mad;
791 init_mad(mad, agent);
792
793 query->sa_query.callback = callback ? ib_sa_service_rec_callback : NULL;
794 query->sa_query.release = ib_sa_service_rec_release;
795 query->sa_query.port = port;
796 mad->mad_hdr.method = method;
797 mad->mad_hdr.attr_id = cpu_to_be16(IB_SA_ATTR_SERVICE_REC);
798 mad->sa_hdr.comp_mask = comp_mask;
799
800 ib_pack(service_rec_table, ARRAY_SIZE(service_rec_table),
801 rec, mad->data);
802
803 *sa_query = &query->sa_query;
804
805 ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
806 if (ret < 0)
807 goto err2;
808
809 return ret;
810
811 err2:
812 *sa_query = NULL;
813 ib_sa_client_put(query->sa_query.client);
814 ib_free_send_mad(query->sa_query.mad_buf);
815
816 err1:
817 kfree(query);
818 return ret;
819 }
820 EXPORT_SYMBOL(ib_sa_service_rec_query);
821
822 static void ib_sa_mcmember_rec_callback(struct ib_sa_query *sa_query,
823 int status,
824 struct ib_sa_mad *mad)
825 {
826 struct ib_sa_mcmember_query *query =
827 container_of(sa_query, struct ib_sa_mcmember_query, sa_query);
828
829 if (mad) {
830 struct ib_sa_mcmember_rec rec;
831
832 ib_unpack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
833 mad->data, &rec);
834 query->callback(status, &rec, query->context);
835 } else
836 query->callback(status, NULL, query->context);
837 }
838
839 static void ib_sa_mcmember_rec_release(struct ib_sa_query *sa_query)
840 {
841 kfree(container_of(sa_query, struct ib_sa_mcmember_query, sa_query));
842 }
843
844 int ib_sa_mcmember_rec_query(struct ib_sa_client *client,
845 struct ib_device *device, u8 port_num,
846 u8 method,
847 struct ib_sa_mcmember_rec *rec,
848 ib_sa_comp_mask comp_mask,
849 int timeout_ms, gfp_t gfp_mask,
850 void (*callback)(int status,
851 struct ib_sa_mcmember_rec *resp,
852 void *context),
853 void *context,
854 struct ib_sa_query **sa_query)
855 {
856 struct ib_sa_mcmember_query *query;
857 struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
858 struct ib_sa_port *port;
859 struct ib_mad_agent *agent;
860 struct ib_sa_mad *mad;
861 int ret;
862
863 if (!sa_dev)
864 return -ENODEV;
865
866 port = &sa_dev->port[port_num - sa_dev->start_port];
867 agent = port->agent;
868
869 query = kmalloc(sizeof *query, gfp_mask);
870 if (!query)
871 return -ENOMEM;
872
873 query->sa_query.mad_buf = ib_create_send_mad(agent, 1, 0,
874 0, IB_MGMT_SA_HDR,
875 IB_MGMT_SA_DATA, gfp_mask);
876 if (!query->sa_query.mad_buf) {
877 ret = -ENOMEM;
878 goto err1;
879 }
880
881 ib_sa_client_get(client);
882 query->sa_query.client = client;
883 query->callback = callback;
884 query->context = context;
885
886 mad = query->sa_query.mad_buf->mad;
887 init_mad(mad, agent);
888
889 query->sa_query.callback = callback ? ib_sa_mcmember_rec_callback : NULL;
890 query->sa_query.release = ib_sa_mcmember_rec_release;
891 query->sa_query.port = port;
892 mad->mad_hdr.method = method;
893 mad->mad_hdr.attr_id = cpu_to_be16(IB_SA_ATTR_MC_MEMBER_REC);
894 mad->sa_hdr.comp_mask = comp_mask;
895
896 ib_pack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
897 rec, mad->data);
898
899 *sa_query = &query->sa_query;
900
901 ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
902 if (ret < 0)
903 goto err2;
904
905 return ret;
906
907 err2:
908 *sa_query = NULL;
909 ib_sa_client_put(query->sa_query.client);
910 ib_free_send_mad(query->sa_query.mad_buf);
911
912 err1:
913 kfree(query);
914 return ret;
915 }
916
917 static void send_handler(struct ib_mad_agent *agent,
918 struct ib_mad_send_wc *mad_send_wc)
919 {
920 struct ib_sa_query *query = mad_send_wc->send_buf->context[0];
921 unsigned long flags;
922
923 if (query->callback)
924 switch (mad_send_wc->status) {
925 case IB_WC_SUCCESS:
926 /* No callback -- already got recv */
927 break;
928 case IB_WC_RESP_TIMEOUT_ERR:
929 query->callback(query, -ETIMEDOUT, NULL);
930 break;
931 case IB_WC_WR_FLUSH_ERR:
932 query->callback(query, -EINTR, NULL);
933 break;
934 default:
935 query->callback(query, -EIO, NULL);
936 break;
937 }
938
939 spin_lock_irqsave(&idr_lock, flags);
940 idr_remove(&query_idr, query->id);
941 spin_unlock_irqrestore(&idr_lock, flags);
942
943 ib_free_send_mad(mad_send_wc->send_buf);
944 kref_put(&query->sm_ah->ref, free_sm_ah);
945 ib_sa_client_put(query->client);
946 query->release(query);
947 }
948
949 static void recv_handler(struct ib_mad_agent *mad_agent,
950 struct ib_mad_recv_wc *mad_recv_wc)
951 {
952 struct ib_sa_query *query;
953 struct ib_mad_send_buf *mad_buf;
954
955 mad_buf = (void *) (unsigned long) mad_recv_wc->wc->wr_id;
956 query = mad_buf->context[0];
957
958 if (query->callback) {
959 if (mad_recv_wc->wc->status == IB_WC_SUCCESS)
960 query->callback(query,
961 mad_recv_wc->recv_buf.mad->mad_hdr.status ?
962 -EINVAL : 0,
963 (struct ib_sa_mad *) mad_recv_wc->recv_buf.mad);
964 else
965 query->callback(query, -EIO, NULL);
966 }
967
968 ib_free_recv_mad(mad_recv_wc);
969 }
970
971 static void ib_sa_add_one(struct ib_device *device)
972 {
973 struct ib_sa_device *sa_dev;
974 int s, e, i;
975
976 if (rdma_node_get_transport(device->node_type) != RDMA_TRANSPORT_IB)
977 return;
978
979 if (device->node_type == RDMA_NODE_IB_SWITCH)
980 s = e = 0;
981 else {
982 s = 1;
983 e = device->phys_port_cnt;
984 }
985
986 sa_dev = kmalloc(sizeof *sa_dev +
987 (e - s + 1) * sizeof (struct ib_sa_port),
988 GFP_KERNEL);
989 if (!sa_dev)
990 return;
991
992 sa_dev->start_port = s;
993 sa_dev->end_port = e;
994
995 for (i = 0; i <= e - s; ++i) {
996 sa_dev->port[i].sm_ah = NULL;
997 sa_dev->port[i].port_num = i + s;
998 spin_lock_init(&sa_dev->port[i].ah_lock);
999
1000 sa_dev->port[i].agent =
1001 ib_register_mad_agent(device, i + s, IB_QPT_GSI,
1002 NULL, 0, send_handler,
1003 recv_handler, sa_dev);
1004 if (IS_ERR(sa_dev->port[i].agent))
1005 goto err;
1006
1007 INIT_WORK(&sa_dev->port[i].update_task, update_sm_ah);
1008 }
1009
1010 ib_set_client_data(device, &sa_client, sa_dev);
1011
1012 /*
1013 * We register our event handler after everything is set up,
1014 * and then update our cached info after the event handler is
1015 * registered to avoid any problems if a port changes state
1016 * during our initialization.
1017 */
1018
1019 INIT_IB_EVENT_HANDLER(&sa_dev->event_handler, device, ib_sa_event);
1020 if (ib_register_event_handler(&sa_dev->event_handler))
1021 goto err;
1022
1023 for (i = 0; i <= e - s; ++i)
1024 update_sm_ah(&sa_dev->port[i].update_task);
1025
1026 return;
1027
1028 err:
1029 while (--i >= 0)
1030 ib_unregister_mad_agent(sa_dev->port[i].agent);
1031
1032 kfree(sa_dev);
1033
1034 return;
1035 }
1036
1037 static void ib_sa_remove_one(struct ib_device *device)
1038 {
1039 struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1040 int i;
1041
1042 if (!sa_dev)
1043 return;
1044
1045 ib_unregister_event_handler(&sa_dev->event_handler);
1046
1047 flush_scheduled_work();
1048
1049 for (i = 0; i <= sa_dev->end_port - sa_dev->start_port; ++i) {
1050 ib_unregister_mad_agent(sa_dev->port[i].agent);
1051 kref_put(&sa_dev->port[i].sm_ah->ref, free_sm_ah);
1052 }
1053
1054 kfree(sa_dev);
1055 }
1056
1057 static int __init ib_sa_init(void)
1058 {
1059 int ret;
1060
1061 spin_lock_init(&idr_lock);
1062 spin_lock_init(&tid_lock);
1063
1064 get_random_bytes(&tid, sizeof tid);
1065
1066 ret = ib_register_client(&sa_client);
1067 if (ret) {
1068 printk(KERN_ERR "Couldn't register ib_sa client\n");
1069 goto err1;
1070 }
1071
1072 ret = mcast_init();
1073 if (ret) {
1074 printk(KERN_ERR "Couldn't initialize multicast handling\n");
1075 goto err2;
1076 }
1077
1078 return 0;
1079 err2:
1080 ib_unregister_client(&sa_client);
1081 err1:
1082 return ret;
1083 }
1084
1085 static void __exit ib_sa_cleanup(void)
1086 {
1087 mcast_cleanup();
1088 ib_unregister_client(&sa_client);
1089 idr_destroy(&query_idr);
1090 }
1091
1092 module_init(ib_sa_init);
1093 module_exit(ib_sa_cleanup);