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/dma-mapping.h>
44 #include <linux/kref.h>
45 #include <linux/idr.h>
46 #include <linux/workqueue.h>
47 
48 #include <rdma/ib_pack.h>
49 #include <rdma/ib_cache.h>
50 #include "sa.h"
51 
52 MODULE_AUTHOR("Roland Dreier");
53 MODULE_DESCRIPTION("InfiniBand subnet administration query support");
54 MODULE_LICENSE("Dual BSD/GPL");
55 
56 struct ib_sa_sm_ah {
57 	struct ib_ah        *ah;
58 	struct kref          ref;
59 	u16		     pkey_index;
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 	{ PATH_REC_FIELD(service_id),
127 	  .offset_words = 0,
128 	  .offset_bits  = 0,
129 	  .size_bits    = 64 },
130 	{ PATH_REC_FIELD(dgid),
131 	  .offset_words = 2,
132 	  .offset_bits  = 0,
133 	  .size_bits    = 128 },
134 	{ PATH_REC_FIELD(sgid),
135 	  .offset_words = 6,
136 	  .offset_bits  = 0,
137 	  .size_bits    = 128 },
138 	{ PATH_REC_FIELD(dlid),
139 	  .offset_words = 10,
140 	  .offset_bits  = 0,
141 	  .size_bits    = 16 },
142 	{ PATH_REC_FIELD(slid),
143 	  .offset_words = 10,
144 	  .offset_bits  = 16,
145 	  .size_bits    = 16 },
146 	{ PATH_REC_FIELD(raw_traffic),
147 	  .offset_words = 11,
148 	  .offset_bits  = 0,
149 	  .size_bits    = 1 },
150 	{ RESERVED,
151 	  .offset_words = 11,
152 	  .offset_bits  = 1,
153 	  .size_bits    = 3 },
154 	{ PATH_REC_FIELD(flow_label),
155 	  .offset_words = 11,
156 	  .offset_bits  = 4,
157 	  .size_bits    = 20 },
158 	{ PATH_REC_FIELD(hop_limit),
159 	  .offset_words = 11,
160 	  .offset_bits  = 24,
161 	  .size_bits    = 8 },
162 	{ PATH_REC_FIELD(traffic_class),
163 	  .offset_words = 12,
164 	  .offset_bits  = 0,
165 	  .size_bits    = 8 },
166 	{ PATH_REC_FIELD(reversible),
167 	  .offset_words = 12,
168 	  .offset_bits  = 8,
169 	  .size_bits    = 1 },
170 	{ PATH_REC_FIELD(numb_path),
171 	  .offset_words = 12,
172 	  .offset_bits  = 9,
173 	  .size_bits    = 7 },
174 	{ PATH_REC_FIELD(pkey),
175 	  .offset_words = 12,
176 	  .offset_bits  = 16,
177 	  .size_bits    = 16 },
178 	{ PATH_REC_FIELD(qos_class),
179 	  .offset_words = 13,
180 	  .offset_bits  = 0,
181 	  .size_bits    = 12 },
182 	{ PATH_REC_FIELD(sl),
183 	  .offset_words = 13,
184 	  .offset_bits  = 12,
185 	  .size_bits    = 4 },
186 	{ PATH_REC_FIELD(mtu_selector),
187 	  .offset_words = 13,
188 	  .offset_bits  = 16,
189 	  .size_bits    = 2 },
190 	{ PATH_REC_FIELD(mtu),
191 	  .offset_words = 13,
192 	  .offset_bits  = 18,
193 	  .size_bits    = 6 },
194 	{ PATH_REC_FIELD(rate_selector),
195 	  .offset_words = 13,
196 	  .offset_bits  = 24,
197 	  .size_bits    = 2 },
198 	{ PATH_REC_FIELD(rate),
199 	  .offset_words = 13,
200 	  .offset_bits  = 26,
201 	  .size_bits    = 6 },
202 	{ PATH_REC_FIELD(packet_life_time_selector),
203 	  .offset_words = 14,
204 	  .offset_bits  = 0,
205 	  .size_bits    = 2 },
206 	{ PATH_REC_FIELD(packet_life_time),
207 	  .offset_words = 14,
208 	  .offset_bits  = 2,
209 	  .size_bits    = 6 },
210 	{ PATH_REC_FIELD(preference),
211 	  .offset_words = 14,
212 	  .offset_bits  = 8,
213 	  .size_bits    = 8 },
214 	{ RESERVED,
215 	  .offset_words = 14,
216 	  .offset_bits  = 16,
217 	  .size_bits    = 48 },
218 };
219 
220 #define MCMEMBER_REC_FIELD(field) \
221 	.struct_offset_bytes = offsetof(struct ib_sa_mcmember_rec, field),	\
222 	.struct_size_bytes   = sizeof ((struct ib_sa_mcmember_rec *) 0)->field,	\
223 	.field_name          = "sa_mcmember_rec:" #field
224 
225 static const struct ib_field mcmember_rec_table[] = {
226 	{ MCMEMBER_REC_FIELD(mgid),
227 	  .offset_words = 0,
228 	  .offset_bits  = 0,
229 	  .size_bits    = 128 },
230 	{ MCMEMBER_REC_FIELD(port_gid),
231 	  .offset_words = 4,
232 	  .offset_bits  = 0,
233 	  .size_bits    = 128 },
234 	{ MCMEMBER_REC_FIELD(qkey),
235 	  .offset_words = 8,
236 	  .offset_bits  = 0,
237 	  .size_bits    = 32 },
238 	{ MCMEMBER_REC_FIELD(mlid),
239 	  .offset_words = 9,
240 	  .offset_bits  = 0,
241 	  .size_bits    = 16 },
242 	{ MCMEMBER_REC_FIELD(mtu_selector),
243 	  .offset_words = 9,
244 	  .offset_bits  = 16,
245 	  .size_bits    = 2 },
246 	{ MCMEMBER_REC_FIELD(mtu),
247 	  .offset_words = 9,
248 	  .offset_bits  = 18,
249 	  .size_bits    = 6 },
250 	{ MCMEMBER_REC_FIELD(traffic_class),
251 	  .offset_words = 9,
252 	  .offset_bits  = 24,
253 	  .size_bits    = 8 },
254 	{ MCMEMBER_REC_FIELD(pkey),
255 	  .offset_words = 10,
256 	  .offset_bits  = 0,
257 	  .size_bits    = 16 },
258 	{ MCMEMBER_REC_FIELD(rate_selector),
259 	  .offset_words = 10,
260 	  .offset_bits  = 16,
261 	  .size_bits    = 2 },
262 	{ MCMEMBER_REC_FIELD(rate),
263 	  .offset_words = 10,
264 	  .offset_bits  = 18,
265 	  .size_bits    = 6 },
266 	{ MCMEMBER_REC_FIELD(packet_life_time_selector),
267 	  .offset_words = 10,
268 	  .offset_bits  = 24,
269 	  .size_bits    = 2 },
270 	{ MCMEMBER_REC_FIELD(packet_life_time),
271 	  .offset_words = 10,
272 	  .offset_bits  = 26,
273 	  .size_bits    = 6 },
274 	{ MCMEMBER_REC_FIELD(sl),
275 	  .offset_words = 11,
276 	  .offset_bits  = 0,
277 	  .size_bits    = 4 },
278 	{ MCMEMBER_REC_FIELD(flow_label),
279 	  .offset_words = 11,
280 	  .offset_bits  = 4,
281 	  .size_bits    = 20 },
282 	{ MCMEMBER_REC_FIELD(hop_limit),
283 	  .offset_words = 11,
284 	  .offset_bits  = 24,
285 	  .size_bits    = 8 },
286 	{ MCMEMBER_REC_FIELD(scope),
287 	  .offset_words = 12,
288 	  .offset_bits  = 0,
289 	  .size_bits    = 4 },
290 	{ MCMEMBER_REC_FIELD(join_state),
291 	  .offset_words = 12,
292 	  .offset_bits  = 4,
293 	  .size_bits    = 4 },
294 	{ MCMEMBER_REC_FIELD(proxy_join),
295 	  .offset_words = 12,
296 	  .offset_bits  = 8,
297 	  .size_bits    = 1 },
298 	{ RESERVED,
299 	  .offset_words = 12,
300 	  .offset_bits  = 9,
301 	  .size_bits    = 23 },
302 };
303 
304 #define SERVICE_REC_FIELD(field) \
305 	.struct_offset_bytes = offsetof(struct ib_sa_service_rec, field),	\
306 	.struct_size_bytes   = sizeof ((struct ib_sa_service_rec *) 0)->field,	\
307 	.field_name          = "sa_service_rec:" #field
308 
309 static const struct ib_field service_rec_table[] = {
310 	{ SERVICE_REC_FIELD(id),
311 	  .offset_words = 0,
312 	  .offset_bits  = 0,
313 	  .size_bits    = 64 },
314 	{ SERVICE_REC_FIELD(gid),
315 	  .offset_words = 2,
316 	  .offset_bits  = 0,
317 	  .size_bits    = 128 },
318 	{ SERVICE_REC_FIELD(pkey),
319 	  .offset_words = 6,
320 	  .offset_bits  = 0,
321 	  .size_bits    = 16 },
322 	{ SERVICE_REC_FIELD(lease),
323 	  .offset_words = 7,
324 	  .offset_bits  = 0,
325 	  .size_bits    = 32 },
326 	{ SERVICE_REC_FIELD(key),
327 	  .offset_words = 8,
328 	  .offset_bits  = 0,
329 	  .size_bits    = 128 },
330 	{ SERVICE_REC_FIELD(name),
331 	  .offset_words = 12,
332 	  .offset_bits  = 0,
333 	  .size_bits    = 64*8 },
334 	{ SERVICE_REC_FIELD(data8),
335 	  .offset_words = 28,
336 	  .offset_bits  = 0,
337 	  .size_bits    = 16*8 },
338 	{ SERVICE_REC_FIELD(data16),
339 	  .offset_words = 32,
340 	  .offset_bits  = 0,
341 	  .size_bits    = 8*16 },
342 	{ SERVICE_REC_FIELD(data32),
343 	  .offset_words = 36,
344 	  .offset_bits  = 0,
345 	  .size_bits    = 4*32 },
346 	{ SERVICE_REC_FIELD(data64),
347 	  .offset_words = 40,
348 	  .offset_bits  = 0,
349 	  .size_bits    = 2*64 },
350 };
351 
352 static void free_sm_ah(struct kref *kref)
353 {
354 	struct ib_sa_sm_ah *sm_ah = container_of(kref, struct ib_sa_sm_ah, ref);
355 
356 	ib_destroy_ah(sm_ah->ah);
357 	kfree(sm_ah);
358 }
359 
360 static void update_sm_ah(struct work_struct *work)
361 {
362 	struct ib_sa_port *port =
363 		container_of(work, struct ib_sa_port, update_task);
364 	struct ib_sa_sm_ah *new_ah, *old_ah;
365 	struct ib_port_attr port_attr;
366 	struct ib_ah_attr   ah_attr;
367 
368 	if (ib_query_port(port->agent->device, port->port_num, &port_attr)) {
369 		printk(KERN_WARNING "Couldn't query port\n");
370 		return;
371 	}
372 
373 	new_ah = kmalloc(sizeof *new_ah, GFP_KERNEL);
374 	if (!new_ah) {
375 		printk(KERN_WARNING "Couldn't allocate new SM AH\n");
376 		return;
377 	}
378 
379 	kref_init(&new_ah->ref);
380 	new_ah->src_path_mask = (1 << port_attr.lmc) - 1;
381 
382 	new_ah->pkey_index = 0;
383 	if (ib_find_pkey(port->agent->device, port->port_num,
384 			 IB_DEFAULT_PKEY_FULL, &new_ah->pkey_index))
385 		printk(KERN_ERR "Couldn't find index for default PKey\n");
386 
387 	memset(&ah_attr, 0, sizeof ah_attr);
388 	ah_attr.dlid     = port_attr.sm_lid;
389 	ah_attr.sl       = port_attr.sm_sl;
390 	ah_attr.port_num = port->port_num;
391 
392 	new_ah->ah = ib_create_ah(port->agent->qp->pd, &ah_attr);
393 	if (IS_ERR(new_ah->ah)) {
394 		printk(KERN_WARNING "Couldn't create new SM AH\n");
395 		kfree(new_ah);
396 		return;
397 	}
398 
399 	spin_lock_irq(&port->ah_lock);
400 	old_ah = port->sm_ah;
401 	port->sm_ah = new_ah;
402 	spin_unlock_irq(&port->ah_lock);
403 
404 	if (old_ah)
405 		kref_put(&old_ah->ref, free_sm_ah);
406 }
407 
408 static void ib_sa_event(struct ib_event_handler *handler, struct ib_event *event)
409 {
410 	if (event->event == IB_EVENT_PORT_ERR    ||
411 	    event->event == IB_EVENT_PORT_ACTIVE ||
412 	    event->event == IB_EVENT_LID_CHANGE  ||
413 	    event->event == IB_EVENT_PKEY_CHANGE ||
414 	    event->event == IB_EVENT_SM_CHANGE   ||
415 	    event->event == IB_EVENT_CLIENT_REREGISTER) {
416 		struct ib_sa_device *sa_dev;
417 		sa_dev = container_of(handler, typeof(*sa_dev), event_handler);
418 
419 		schedule_work(&sa_dev->port[event->element.port_num -
420 					    sa_dev->start_port].update_task);
421 	}
422 }
423 
424 void ib_sa_register_client(struct ib_sa_client *client)
425 {
426 	atomic_set(&client->users, 1);
427 	init_completion(&client->comp);
428 }
429 EXPORT_SYMBOL(ib_sa_register_client);
430 
431 void ib_sa_unregister_client(struct ib_sa_client *client)
432 {
433 	ib_sa_client_put(client);
434 	wait_for_completion(&client->comp);
435 }
436 EXPORT_SYMBOL(ib_sa_unregister_client);
437 
438 /**
439  * ib_sa_cancel_query - try to cancel an SA query
440  * @id:ID of query to cancel
441  * @query:query pointer to cancel
442  *
443  * Try to cancel an SA query.  If the id and query don't match up or
444  * the query has already completed, nothing is done.  Otherwise the
445  * query is canceled and will complete with a status of -EINTR.
446  */
447 void ib_sa_cancel_query(int id, struct ib_sa_query *query)
448 {
449 	unsigned long flags;
450 	struct ib_mad_agent *agent;
451 	struct ib_mad_send_buf *mad_buf;
452 
453 	spin_lock_irqsave(&idr_lock, flags);
454 	if (idr_find(&query_idr, id) != query) {
455 		spin_unlock_irqrestore(&idr_lock, flags);
456 		return;
457 	}
458 	agent = query->port->agent;
459 	mad_buf = query->mad_buf;
460 	spin_unlock_irqrestore(&idr_lock, flags);
461 
462 	ib_cancel_mad(agent, mad_buf);
463 }
464 EXPORT_SYMBOL(ib_sa_cancel_query);
465 
466 static u8 get_src_path_mask(struct ib_device *device, u8 port_num)
467 {
468 	struct ib_sa_device *sa_dev;
469 	struct ib_sa_port   *port;
470 	unsigned long flags;
471 	u8 src_path_mask;
472 
473 	sa_dev = ib_get_client_data(device, &sa_client);
474 	if (!sa_dev)
475 		return 0x7f;
476 
477 	port  = &sa_dev->port[port_num - sa_dev->start_port];
478 	spin_lock_irqsave(&port->ah_lock, flags);
479 	src_path_mask = port->sm_ah ? port->sm_ah->src_path_mask : 0x7f;
480 	spin_unlock_irqrestore(&port->ah_lock, flags);
481 
482 	return src_path_mask;
483 }
484 
485 int ib_init_ah_from_path(struct ib_device *device, u8 port_num,
486 			 struct ib_sa_path_rec *rec, struct ib_ah_attr *ah_attr)
487 {
488 	int ret;
489 	u16 gid_index;
490 
491 	memset(ah_attr, 0, sizeof *ah_attr);
492 	ah_attr->dlid = be16_to_cpu(rec->dlid);
493 	ah_attr->sl = rec->sl;
494 	ah_attr->src_path_bits = be16_to_cpu(rec->slid) &
495 				 get_src_path_mask(device, port_num);
496 	ah_attr->port_num = port_num;
497 	ah_attr->static_rate = rec->rate;
498 
499 	if (rec->hop_limit > 1) {
500 		ah_attr->ah_flags = IB_AH_GRH;
501 		ah_attr->grh.dgid = rec->dgid;
502 
503 		ret = ib_find_cached_gid(device, &rec->sgid, &port_num,
504 					 &gid_index);
505 		if (ret)
506 			return ret;
507 
508 		ah_attr->grh.sgid_index    = gid_index;
509 		ah_attr->grh.flow_label    = be32_to_cpu(rec->flow_label);
510 		ah_attr->grh.hop_limit     = rec->hop_limit;
511 		ah_attr->grh.traffic_class = rec->traffic_class;
512 	}
513 	return 0;
514 }
515 EXPORT_SYMBOL(ib_init_ah_from_path);
516 
517 static int alloc_mad(struct ib_sa_query *query, gfp_t gfp_mask)
518 {
519 	unsigned long flags;
520 
521 	spin_lock_irqsave(&query->port->ah_lock, flags);
522 	kref_get(&query->port->sm_ah->ref);
523 	query->sm_ah = query->port->sm_ah;
524 	spin_unlock_irqrestore(&query->port->ah_lock, flags);
525 
526 	query->mad_buf = ib_create_send_mad(query->port->agent, 1,
527 					    query->sm_ah->pkey_index,
528 					    0, IB_MGMT_SA_HDR, IB_MGMT_SA_DATA,
529 					    gfp_mask);
530 	if (IS_ERR(query->mad_buf)) {
531 		kref_put(&query->sm_ah->ref, free_sm_ah);
532 		return -ENOMEM;
533 	}
534 
535 	query->mad_buf->ah = query->sm_ah->ah;
536 
537 	return 0;
538 }
539 
540 static void free_mad(struct ib_sa_query *query)
541 {
542 	ib_free_send_mad(query->mad_buf);
543 	kref_put(&query->sm_ah->ref, free_sm_ah);
544 }
545 
546 static void init_mad(struct ib_sa_mad *mad, struct ib_mad_agent *agent)
547 {
548 	unsigned long flags;
549 
550 	memset(mad, 0, sizeof *mad);
551 
552 	mad->mad_hdr.base_version  = IB_MGMT_BASE_VERSION;
553 	mad->mad_hdr.mgmt_class    = IB_MGMT_CLASS_SUBN_ADM;
554 	mad->mad_hdr.class_version = IB_SA_CLASS_VERSION;
555 
556 	spin_lock_irqsave(&tid_lock, flags);
557 	mad->mad_hdr.tid           =
558 		cpu_to_be64(((u64) agent->hi_tid) << 32 | tid++);
559 	spin_unlock_irqrestore(&tid_lock, flags);
560 }
561 
562 static int send_mad(struct ib_sa_query *query, int timeout_ms, gfp_t gfp_mask)
563 {
564 	unsigned long flags;
565 	int ret, id;
566 
567 retry:
568 	if (!idr_pre_get(&query_idr, gfp_mask))
569 		return -ENOMEM;
570 	spin_lock_irqsave(&idr_lock, flags);
571 	ret = idr_get_new(&query_idr, query, &id);
572 	spin_unlock_irqrestore(&idr_lock, flags);
573 	if (ret == -EAGAIN)
574 		goto retry;
575 	if (ret)
576 		return ret;
577 
578 	query->mad_buf->timeout_ms  = timeout_ms;
579 	query->mad_buf->context[0] = query;
580 	query->id = id;
581 
582 	ret = ib_post_send_mad(query->mad_buf, NULL);
583 	if (ret) {
584 		spin_lock_irqsave(&idr_lock, flags);
585 		idr_remove(&query_idr, id);
586 		spin_unlock_irqrestore(&idr_lock, flags);
587 	}
588 
589 	/*
590 	 * It's not safe to dereference query any more, because the
591 	 * send may already have completed and freed the query in
592 	 * another context.
593 	 */
594 	return ret ? ret : id;
595 }
596 
597 static void ib_sa_path_rec_callback(struct ib_sa_query *sa_query,
598 				    int status,
599 				    struct ib_sa_mad *mad)
600 {
601 	struct ib_sa_path_query *query =
602 		container_of(sa_query, struct ib_sa_path_query, sa_query);
603 
604 	if (mad) {
605 		struct ib_sa_path_rec rec;
606 
607 		ib_unpack(path_rec_table, ARRAY_SIZE(path_rec_table),
608 			  mad->data, &rec);
609 		query->callback(status, &rec, query->context);
610 	} else
611 		query->callback(status, NULL, query->context);
612 }
613 
614 static void ib_sa_path_rec_release(struct ib_sa_query *sa_query)
615 {
616 	kfree(container_of(sa_query, struct ib_sa_path_query, sa_query));
617 }
618 
619 /**
620  * ib_sa_path_rec_get - Start a Path get query
621  * @client:SA client
622  * @device:device to send query on
623  * @port_num: port number to send query on
624  * @rec:Path Record to send in query
625  * @comp_mask:component mask to send in query
626  * @timeout_ms:time to wait for response
627  * @gfp_mask:GFP mask to use for internal allocations
628  * @callback:function called when query completes, times out or is
629  * canceled
630  * @context:opaque user context passed to callback
631  * @sa_query:query context, used to cancel query
632  *
633  * Send a Path Record Get query to the SA to look up a path.  The
634  * callback function will be called when the query completes (or
635  * fails); status is 0 for a successful response, -EINTR if the query
636  * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
637  * occurred sending the query.  The resp parameter of the callback is
638  * only valid if status is 0.
639  *
640  * If the return value of ib_sa_path_rec_get() is negative, it is an
641  * error code.  Otherwise it is a query ID that can be used to cancel
642  * the query.
643  */
644 int ib_sa_path_rec_get(struct ib_sa_client *client,
645 		       struct ib_device *device, u8 port_num,
646 		       struct ib_sa_path_rec *rec,
647 		       ib_sa_comp_mask comp_mask,
648 		       int timeout_ms, gfp_t gfp_mask,
649 		       void (*callback)(int status,
650 					struct ib_sa_path_rec *resp,
651 					void *context),
652 		       void *context,
653 		       struct ib_sa_query **sa_query)
654 {
655 	struct ib_sa_path_query *query;
656 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
657 	struct ib_sa_port   *port;
658 	struct ib_mad_agent *agent;
659 	struct ib_sa_mad *mad;
660 	int ret;
661 
662 	if (!sa_dev)
663 		return -ENODEV;
664 
665 	port  = &sa_dev->port[port_num - sa_dev->start_port];
666 	agent = port->agent;
667 
668 	query = kmalloc(sizeof *query, gfp_mask);
669 	if (!query)
670 		return -ENOMEM;
671 
672 	query->sa_query.port     = port;
673 	ret = alloc_mad(&query->sa_query, gfp_mask);
674 	if (ret)
675 		goto err1;
676 
677 	ib_sa_client_get(client);
678 	query->sa_query.client = client;
679 	query->callback        = callback;
680 	query->context         = context;
681 
682 	mad = query->sa_query.mad_buf->mad;
683 	init_mad(mad, agent);
684 
685 	query->sa_query.callback = callback ? ib_sa_path_rec_callback : NULL;
686 	query->sa_query.release  = ib_sa_path_rec_release;
687 	mad->mad_hdr.method	 = IB_MGMT_METHOD_GET;
688 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_PATH_REC);
689 	mad->sa_hdr.comp_mask	 = comp_mask;
690 
691 	ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table), rec, mad->data);
692 
693 	*sa_query = &query->sa_query;
694 
695 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
696 	if (ret < 0)
697 		goto err2;
698 
699 	return ret;
700 
701 err2:
702 	*sa_query = NULL;
703 	ib_sa_client_put(query->sa_query.client);
704 	free_mad(&query->sa_query);
705 
706 err1:
707 	kfree(query);
708 	return ret;
709 }
710 EXPORT_SYMBOL(ib_sa_path_rec_get);
711 
712 static void ib_sa_service_rec_callback(struct ib_sa_query *sa_query,
713 				    int status,
714 				    struct ib_sa_mad *mad)
715 {
716 	struct ib_sa_service_query *query =
717 		container_of(sa_query, struct ib_sa_service_query, sa_query);
718 
719 	if (mad) {
720 		struct ib_sa_service_rec rec;
721 
722 		ib_unpack(service_rec_table, ARRAY_SIZE(service_rec_table),
723 			  mad->data, &rec);
724 		query->callback(status, &rec, query->context);
725 	} else
726 		query->callback(status, NULL, query->context);
727 }
728 
729 static void ib_sa_service_rec_release(struct ib_sa_query *sa_query)
730 {
731 	kfree(container_of(sa_query, struct ib_sa_service_query, sa_query));
732 }
733 
734 /**
735  * ib_sa_service_rec_query - Start Service Record operation
736  * @client:SA client
737  * @device:device to send request on
738  * @port_num: port number to send request on
739  * @method:SA method - should be get, set, or delete
740  * @rec:Service Record to send in request
741  * @comp_mask:component mask to send in request
742  * @timeout_ms:time to wait for response
743  * @gfp_mask:GFP mask to use for internal allocations
744  * @callback:function called when request completes, times out or is
745  * canceled
746  * @context:opaque user context passed to callback
747  * @sa_query:request context, used to cancel request
748  *
749  * Send a Service Record set/get/delete to the SA to register,
750  * unregister or query a service record.
751  * The callback function will be called when the request completes (or
752  * fails); status is 0 for a successful response, -EINTR if the query
753  * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
754  * occurred sending the query.  The resp parameter of the callback is
755  * only valid if status is 0.
756  *
757  * If the return value of ib_sa_service_rec_query() is negative, it is an
758  * error code.  Otherwise it is a request ID that can be used to cancel
759  * the query.
760  */
761 int ib_sa_service_rec_query(struct ib_sa_client *client,
762 			    struct ib_device *device, u8 port_num, u8 method,
763 			    struct ib_sa_service_rec *rec,
764 			    ib_sa_comp_mask comp_mask,
765 			    int timeout_ms, gfp_t gfp_mask,
766 			    void (*callback)(int status,
767 					     struct ib_sa_service_rec *resp,
768 					     void *context),
769 			    void *context,
770 			    struct ib_sa_query **sa_query)
771 {
772 	struct ib_sa_service_query *query;
773 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
774 	struct ib_sa_port   *port;
775 	struct ib_mad_agent *agent;
776 	struct ib_sa_mad *mad;
777 	int ret;
778 
779 	if (!sa_dev)
780 		return -ENODEV;
781 
782 	port  = &sa_dev->port[port_num - sa_dev->start_port];
783 	agent = port->agent;
784 
785 	if (method != IB_MGMT_METHOD_GET &&
786 	    method != IB_MGMT_METHOD_SET &&
787 	    method != IB_SA_METHOD_DELETE)
788 		return -EINVAL;
789 
790 	query = kmalloc(sizeof *query, gfp_mask);
791 	if (!query)
792 		return -ENOMEM;
793 
794 	query->sa_query.port     = port;
795 	ret = alloc_mad(&query->sa_query, gfp_mask);
796 	if (ret)
797 		goto err1;
798 
799 	ib_sa_client_get(client);
800 	query->sa_query.client = client;
801 	query->callback        = callback;
802 	query->context         = context;
803 
804 	mad = query->sa_query.mad_buf->mad;
805 	init_mad(mad, agent);
806 
807 	query->sa_query.callback = callback ? ib_sa_service_rec_callback : NULL;
808 	query->sa_query.release  = ib_sa_service_rec_release;
809 	mad->mad_hdr.method	 = method;
810 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_SERVICE_REC);
811 	mad->sa_hdr.comp_mask	 = comp_mask;
812 
813 	ib_pack(service_rec_table, ARRAY_SIZE(service_rec_table),
814 		rec, mad->data);
815 
816 	*sa_query = &query->sa_query;
817 
818 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
819 	if (ret < 0)
820 		goto err2;
821 
822 	return ret;
823 
824 err2:
825 	*sa_query = NULL;
826 	ib_sa_client_put(query->sa_query.client);
827 	free_mad(&query->sa_query);
828 
829 err1:
830 	kfree(query);
831 	return ret;
832 }
833 EXPORT_SYMBOL(ib_sa_service_rec_query);
834 
835 static void ib_sa_mcmember_rec_callback(struct ib_sa_query *sa_query,
836 					int status,
837 					struct ib_sa_mad *mad)
838 {
839 	struct ib_sa_mcmember_query *query =
840 		container_of(sa_query, struct ib_sa_mcmember_query, sa_query);
841 
842 	if (mad) {
843 		struct ib_sa_mcmember_rec rec;
844 
845 		ib_unpack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
846 			  mad->data, &rec);
847 		query->callback(status, &rec, query->context);
848 	} else
849 		query->callback(status, NULL, query->context);
850 }
851 
852 static void ib_sa_mcmember_rec_release(struct ib_sa_query *sa_query)
853 {
854 	kfree(container_of(sa_query, struct ib_sa_mcmember_query, sa_query));
855 }
856 
857 int ib_sa_mcmember_rec_query(struct ib_sa_client *client,
858 			     struct ib_device *device, u8 port_num,
859 			     u8 method,
860 			     struct ib_sa_mcmember_rec *rec,
861 			     ib_sa_comp_mask comp_mask,
862 			     int timeout_ms, gfp_t gfp_mask,
863 			     void (*callback)(int status,
864 					      struct ib_sa_mcmember_rec *resp,
865 					      void *context),
866 			     void *context,
867 			     struct ib_sa_query **sa_query)
868 {
869 	struct ib_sa_mcmember_query *query;
870 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
871 	struct ib_sa_port   *port;
872 	struct ib_mad_agent *agent;
873 	struct ib_sa_mad *mad;
874 	int ret;
875 
876 	if (!sa_dev)
877 		return -ENODEV;
878 
879 	port  = &sa_dev->port[port_num - sa_dev->start_port];
880 	agent = port->agent;
881 
882 	query = kmalloc(sizeof *query, gfp_mask);
883 	if (!query)
884 		return -ENOMEM;
885 
886 	query->sa_query.port     = port;
887 	ret = alloc_mad(&query->sa_query, gfp_mask);
888 	if (ret)
889 		goto err1;
890 
891 	ib_sa_client_get(client);
892 	query->sa_query.client = client;
893 	query->callback        = callback;
894 	query->context         = context;
895 
896 	mad = query->sa_query.mad_buf->mad;
897 	init_mad(mad, agent);
898 
899 	query->sa_query.callback = callback ? ib_sa_mcmember_rec_callback : NULL;
900 	query->sa_query.release  = ib_sa_mcmember_rec_release;
901 	mad->mad_hdr.method	 = method;
902 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_MC_MEMBER_REC);
903 	mad->sa_hdr.comp_mask	 = comp_mask;
904 
905 	ib_pack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
906 		rec, mad->data);
907 
908 	*sa_query = &query->sa_query;
909 
910 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
911 	if (ret < 0)
912 		goto err2;
913 
914 	return ret;
915 
916 err2:
917 	*sa_query = NULL;
918 	ib_sa_client_put(query->sa_query.client);
919 	free_mad(&query->sa_query);
920 
921 err1:
922 	kfree(query);
923 	return ret;
924 }
925 
926 static void send_handler(struct ib_mad_agent *agent,
927 			 struct ib_mad_send_wc *mad_send_wc)
928 {
929 	struct ib_sa_query *query = mad_send_wc->send_buf->context[0];
930 	unsigned long flags;
931 
932 	if (query->callback)
933 		switch (mad_send_wc->status) {
934 		case IB_WC_SUCCESS:
935 			/* No callback -- already got recv */
936 			break;
937 		case IB_WC_RESP_TIMEOUT_ERR:
938 			query->callback(query, -ETIMEDOUT, NULL);
939 			break;
940 		case IB_WC_WR_FLUSH_ERR:
941 			query->callback(query, -EINTR, NULL);
942 			break;
943 		default:
944 			query->callback(query, -EIO, NULL);
945 			break;
946 		}
947 
948 	spin_lock_irqsave(&idr_lock, flags);
949 	idr_remove(&query_idr, query->id);
950 	spin_unlock_irqrestore(&idr_lock, flags);
951 
952 	free_mad(query);
953 	ib_sa_client_put(query->client);
954 	query->release(query);
955 }
956 
957 static void recv_handler(struct ib_mad_agent *mad_agent,
958 			 struct ib_mad_recv_wc *mad_recv_wc)
959 {
960 	struct ib_sa_query *query;
961 	struct ib_mad_send_buf *mad_buf;
962 
963 	mad_buf = (void *) (unsigned long) mad_recv_wc->wc->wr_id;
964 	query = mad_buf->context[0];
965 
966 	if (query->callback) {
967 		if (mad_recv_wc->wc->status == IB_WC_SUCCESS)
968 			query->callback(query,
969 					mad_recv_wc->recv_buf.mad->mad_hdr.status ?
970 					-EINVAL : 0,
971 					(struct ib_sa_mad *) mad_recv_wc->recv_buf.mad);
972 		else
973 			query->callback(query, -EIO, NULL);
974 	}
975 
976 	ib_free_recv_mad(mad_recv_wc);
977 }
978 
979 static void ib_sa_add_one(struct ib_device *device)
980 {
981 	struct ib_sa_device *sa_dev;
982 	int s, e, i;
983 
984 	if (rdma_node_get_transport(device->node_type) != RDMA_TRANSPORT_IB)
985 		return;
986 
987 	if (device->node_type == RDMA_NODE_IB_SWITCH)
988 		s = e = 0;
989 	else {
990 		s = 1;
991 		e = device->phys_port_cnt;
992 	}
993 
994 	sa_dev = kmalloc(sizeof *sa_dev +
995 			 (e - s + 1) * sizeof (struct ib_sa_port),
996 			 GFP_KERNEL);
997 	if (!sa_dev)
998 		return;
999 
1000 	sa_dev->start_port = s;
1001 	sa_dev->end_port   = e;
1002 
1003 	for (i = 0; i <= e - s; ++i) {
1004 		sa_dev->port[i].sm_ah    = NULL;
1005 		sa_dev->port[i].port_num = i + s;
1006 		spin_lock_init(&sa_dev->port[i].ah_lock);
1007 
1008 		sa_dev->port[i].agent =
1009 			ib_register_mad_agent(device, i + s, IB_QPT_GSI,
1010 					      NULL, 0, send_handler,
1011 					      recv_handler, sa_dev);
1012 		if (IS_ERR(sa_dev->port[i].agent))
1013 			goto err;
1014 
1015 		INIT_WORK(&sa_dev->port[i].update_task, update_sm_ah);
1016 	}
1017 
1018 	ib_set_client_data(device, &sa_client, sa_dev);
1019 
1020 	/*
1021 	 * We register our event handler after everything is set up,
1022 	 * and then update our cached info after the event handler is
1023 	 * registered to avoid any problems if a port changes state
1024 	 * during our initialization.
1025 	 */
1026 
1027 	INIT_IB_EVENT_HANDLER(&sa_dev->event_handler, device, ib_sa_event);
1028 	if (ib_register_event_handler(&sa_dev->event_handler))
1029 		goto err;
1030 
1031 	for (i = 0; i <= e - s; ++i)
1032 		update_sm_ah(&sa_dev->port[i].update_task);
1033 
1034 	return;
1035 
1036 err:
1037 	while (--i >= 0)
1038 		ib_unregister_mad_agent(sa_dev->port[i].agent);
1039 
1040 	kfree(sa_dev);
1041 
1042 	return;
1043 }
1044 
1045 static void ib_sa_remove_one(struct ib_device *device)
1046 {
1047 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1048 	int i;
1049 
1050 	if (!sa_dev)
1051 		return;
1052 
1053 	ib_unregister_event_handler(&sa_dev->event_handler);
1054 
1055 	flush_scheduled_work();
1056 
1057 	for (i = 0; i <= sa_dev->end_port - sa_dev->start_port; ++i) {
1058 		ib_unregister_mad_agent(sa_dev->port[i].agent);
1059 		kref_put(&sa_dev->port[i].sm_ah->ref, free_sm_ah);
1060 	}
1061 
1062 	kfree(sa_dev);
1063 }
1064 
1065 static int __init ib_sa_init(void)
1066 {
1067 	int ret;
1068 
1069 	spin_lock_init(&idr_lock);
1070 	spin_lock_init(&tid_lock);
1071 
1072 	get_random_bytes(&tid, sizeof tid);
1073 
1074 	ret = ib_register_client(&sa_client);
1075 	if (ret) {
1076 		printk(KERN_ERR "Couldn't register ib_sa client\n");
1077 		goto err1;
1078 	}
1079 
1080 	ret = mcast_init();
1081 	if (ret) {
1082 		printk(KERN_ERR "Couldn't initialize multicast handling\n");
1083 		goto err2;
1084 	}
1085 
1086 	return 0;
1087 err2:
1088 	ib_unregister_client(&sa_client);
1089 err1:
1090 	return ret;
1091 }
1092 
1093 static void __exit ib_sa_cleanup(void)
1094 {
1095 	mcast_cleanup();
1096 	ib_unregister_client(&sa_client);
1097 	idr_destroy(&query_idr);
1098 }
1099 
1100 module_init(ib_sa_init);
1101 module_exit(ib_sa_cleanup);
1102