1 /*
2  * Copyright (c) 2004 Topspin Communications.  All rights reserved.
3  * Copyright (c) 2005 Voltaire, Inc.  All rights reserved.
4  *
5  * This software is available to you under a choice of one of two
6  * licenses.  You may choose to be licensed under the terms of the GNU
7  * General Public License (GPL) Version 2, available from the file
8  * COPYING in the main directory of this source tree, or the
9  * OpenIB.org BSD license below:
10  *
11  *     Redistribution and use in source and binary forms, with or
12  *     without modification, are permitted provided that the following
13  *     conditions are met:
14  *
15  *      - Redistributions of source code must retain the above
16  *        copyright notice, this list of conditions and the following
17  *        disclaimer.
18  *
19  *      - Redistributions in binary form must reproduce the above
20  *        copyright notice, this list of conditions and the following
21  *        disclaimer in the documentation and/or other materials
22  *        provided with the distribution.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31  * SOFTWARE.
32  *
33  * $Id: sa_query.c 2811 2005-07-06 18:11:43Z halr $
34  */
35 
36 #include <linux/module.h>
37 #include <linux/init.h>
38 #include <linux/err.h>
39 #include <linux/random.h>
40 #include <linux/spinlock.h>
41 #include <linux/slab.h>
42 #include <linux/pci.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_sa.h>
50 
51 MODULE_AUTHOR("Roland Dreier");
52 MODULE_DESCRIPTION("InfiniBand subnet administration query support");
53 MODULE_LICENSE("Dual BSD/GPL");
54 
55 struct ib_sa_sm_ah {
56 	struct ib_ah        *ah;
57 	struct kref          ref;
58 };
59 
60 struct ib_sa_port {
61 	struct ib_mad_agent *agent;
62 	struct ib_sa_sm_ah  *sm_ah;
63 	struct work_struct   update_task;
64 	spinlock_t           ah_lock;
65 	u8                   port_num;
66 };
67 
68 struct ib_sa_device {
69 	int                     start_port, end_port;
70 	struct ib_event_handler event_handler;
71 	struct ib_sa_port port[0];
72 };
73 
74 struct ib_sa_query {
75 	void (*callback)(struct ib_sa_query *, int, struct ib_sa_mad *);
76 	void (*release)(struct ib_sa_query *);
77 	struct ib_sa_port      *port;
78 	struct ib_mad_send_buf *mad_buf;
79 	struct ib_sa_sm_ah     *sm_ah;
80 	int			id;
81 };
82 
83 struct ib_sa_service_query {
84 	void (*callback)(int, struct ib_sa_service_rec *, void *);
85 	void *context;
86 	struct ib_sa_query sa_query;
87 };
88 
89 struct ib_sa_path_query {
90 	void (*callback)(int, struct ib_sa_path_rec *, void *);
91 	void *context;
92 	struct ib_sa_query sa_query;
93 };
94 
95 struct ib_sa_mcmember_query {
96 	void (*callback)(int, struct ib_sa_mcmember_rec *, void *);
97 	void *context;
98 	struct ib_sa_query sa_query;
99 };
100 
101 static void ib_sa_add_one(struct ib_device *device);
102 static void ib_sa_remove_one(struct ib_device *device);
103 
104 static struct ib_client sa_client = {
105 	.name   = "sa",
106 	.add    = ib_sa_add_one,
107 	.remove = ib_sa_remove_one
108 };
109 
110 static spinlock_t idr_lock;
111 static DEFINE_IDR(query_idr);
112 
113 static spinlock_t tid_lock;
114 static u32 tid;
115 
116 #define PATH_REC_FIELD(field) \
117 	.struct_offset_bytes = offsetof(struct ib_sa_path_rec, field),		\
118 	.struct_size_bytes   = sizeof ((struct ib_sa_path_rec *) 0)->field,	\
119 	.field_name          = "sa_path_rec:" #field
120 
121 static const struct ib_field path_rec_table[] = {
122 	{ RESERVED,
123 	  .offset_words = 0,
124 	  .offset_bits  = 0,
125 	  .size_bits    = 32 },
126 	{ RESERVED,
127 	  .offset_words = 1,
128 	  .offset_bits  = 0,
129 	  .size_bits    = 32 },
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 	{ RESERVED,
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(void *port_ptr)
361 {
362 	struct ib_sa_port *port = port_ptr;
363 	struct ib_sa_sm_ah *new_ah, *old_ah;
364 	struct ib_port_attr port_attr;
365 	struct ib_ah_attr   ah_attr;
366 
367 	if (ib_query_port(port->agent->device, port->port_num, &port_attr)) {
368 		printk(KERN_WARNING "Couldn't query port\n");
369 		return;
370 	}
371 
372 	new_ah = kmalloc(sizeof *new_ah, GFP_KERNEL);
373 	if (!new_ah) {
374 		printk(KERN_WARNING "Couldn't allocate new SM AH\n");
375 		return;
376 	}
377 
378 	kref_init(&new_ah->ref);
379 
380 	memset(&ah_attr, 0, sizeof ah_attr);
381 	ah_attr.dlid     = port_attr.sm_lid;
382 	ah_attr.sl       = port_attr.sm_sl;
383 	ah_attr.port_num = port->port_num;
384 
385 	new_ah->ah = ib_create_ah(port->agent->qp->pd, &ah_attr);
386 	if (IS_ERR(new_ah->ah)) {
387 		printk(KERN_WARNING "Couldn't create new SM AH\n");
388 		kfree(new_ah);
389 		return;
390 	}
391 
392 	spin_lock_irq(&port->ah_lock);
393 	old_ah = port->sm_ah;
394 	port->sm_ah = new_ah;
395 	spin_unlock_irq(&port->ah_lock);
396 
397 	if (old_ah)
398 		kref_put(&old_ah->ref, free_sm_ah);
399 }
400 
401 static void ib_sa_event(struct ib_event_handler *handler, struct ib_event *event)
402 {
403 	if (event->event == IB_EVENT_PORT_ERR    ||
404 	    event->event == IB_EVENT_PORT_ACTIVE ||
405 	    event->event == IB_EVENT_LID_CHANGE  ||
406 	    event->event == IB_EVENT_PKEY_CHANGE ||
407 	    event->event == IB_EVENT_SM_CHANGE) {
408 		struct ib_sa_device *sa_dev;
409 		sa_dev = container_of(handler, typeof(*sa_dev), event_handler);
410 
411 		schedule_work(&sa_dev->port[event->element.port_num -
412 					    sa_dev->start_port].update_task);
413 	}
414 }
415 
416 /**
417  * ib_sa_cancel_query - try to cancel an SA query
418  * @id:ID of query to cancel
419  * @query:query pointer to cancel
420  *
421  * Try to cancel an SA query.  If the id and query don't match up or
422  * the query has already completed, nothing is done.  Otherwise the
423  * query is canceled and will complete with a status of -EINTR.
424  */
425 void ib_sa_cancel_query(int id, struct ib_sa_query *query)
426 {
427 	unsigned long flags;
428 	struct ib_mad_agent *agent;
429 	struct ib_mad_send_buf *mad_buf;
430 
431 	spin_lock_irqsave(&idr_lock, flags);
432 	if (idr_find(&query_idr, id) != query) {
433 		spin_unlock_irqrestore(&idr_lock, flags);
434 		return;
435 	}
436 	agent = query->port->agent;
437 	mad_buf = query->mad_buf;
438 	spin_unlock_irqrestore(&idr_lock, flags);
439 
440 	ib_cancel_mad(agent, mad_buf);
441 }
442 EXPORT_SYMBOL(ib_sa_cancel_query);
443 
444 static void init_mad(struct ib_sa_mad *mad, struct ib_mad_agent *agent)
445 {
446 	unsigned long flags;
447 
448 	memset(mad, 0, sizeof *mad);
449 
450 	mad->mad_hdr.base_version  = IB_MGMT_BASE_VERSION;
451 	mad->mad_hdr.mgmt_class    = IB_MGMT_CLASS_SUBN_ADM;
452 	mad->mad_hdr.class_version = IB_SA_CLASS_VERSION;
453 
454 	spin_lock_irqsave(&tid_lock, flags);
455 	mad->mad_hdr.tid           =
456 		cpu_to_be64(((u64) agent->hi_tid) << 32 | tid++);
457 	spin_unlock_irqrestore(&tid_lock, flags);
458 }
459 
460 static int send_mad(struct ib_sa_query *query, int timeout_ms)
461 {
462 	unsigned long flags;
463 	int ret, id;
464 
465 retry:
466 	if (!idr_pre_get(&query_idr, GFP_ATOMIC))
467 		return -ENOMEM;
468 	spin_lock_irqsave(&idr_lock, flags);
469 	ret = idr_get_new(&query_idr, query, &id);
470 	spin_unlock_irqrestore(&idr_lock, flags);
471 	if (ret == -EAGAIN)
472 		goto retry;
473 	if (ret)
474 		return ret;
475 
476 	query->mad_buf->timeout_ms  = timeout_ms;
477 	query->mad_buf->context[0] = query;
478 	query->id = id;
479 
480 	spin_lock_irqsave(&query->port->ah_lock, flags);
481 	kref_get(&query->port->sm_ah->ref);
482 	query->sm_ah = query->port->sm_ah;
483 	spin_unlock_irqrestore(&query->port->ah_lock, flags);
484 
485 	query->mad_buf->ah = query->sm_ah->ah;
486 
487 	ret = ib_post_send_mad(query->mad_buf, NULL);
488 	if (ret) {
489 		spin_lock_irqsave(&idr_lock, flags);
490 		idr_remove(&query_idr, id);
491 		spin_unlock_irqrestore(&idr_lock, flags);
492 
493 		kref_put(&query->sm_ah->ref, free_sm_ah);
494 	}
495 
496 	/*
497 	 * It's not safe to dereference query any more, because the
498 	 * send may already have completed and freed the query in
499 	 * another context.
500 	 */
501 	return ret ? ret : id;
502 }
503 
504 static void ib_sa_path_rec_callback(struct ib_sa_query *sa_query,
505 				    int status,
506 				    struct ib_sa_mad *mad)
507 {
508 	struct ib_sa_path_query *query =
509 		container_of(sa_query, struct ib_sa_path_query, sa_query);
510 
511 	if (mad) {
512 		struct ib_sa_path_rec rec;
513 
514 		ib_unpack(path_rec_table, ARRAY_SIZE(path_rec_table),
515 			  mad->data, &rec);
516 		query->callback(status, &rec, query->context);
517 	} else
518 		query->callback(status, NULL, query->context);
519 }
520 
521 static void ib_sa_path_rec_release(struct ib_sa_query *sa_query)
522 {
523 	kfree(container_of(sa_query, struct ib_sa_path_query, sa_query));
524 }
525 
526 /**
527  * ib_sa_path_rec_get - Start a Path get query
528  * @device:device to send query on
529  * @port_num: port number to send query on
530  * @rec:Path Record to send in query
531  * @comp_mask:component mask to send in query
532  * @timeout_ms:time to wait for response
533  * @gfp_mask:GFP mask to use for internal allocations
534  * @callback:function called when query completes, times out or is
535  * canceled
536  * @context:opaque user context passed to callback
537  * @sa_query:query context, used to cancel query
538  *
539  * Send a Path Record Get query to the SA to look up a path.  The
540  * callback function will be called when the query completes (or
541  * fails); status is 0 for a successful response, -EINTR if the query
542  * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
543  * occurred sending the query.  The resp parameter of the callback is
544  * only valid if status is 0.
545  *
546  * If the return value of ib_sa_path_rec_get() is negative, it is an
547  * error code.  Otherwise it is a query ID that can be used to cancel
548  * the query.
549  */
550 int ib_sa_path_rec_get(struct ib_device *device, u8 port_num,
551 		       struct ib_sa_path_rec *rec,
552 		       ib_sa_comp_mask comp_mask,
553 		       int timeout_ms, gfp_t gfp_mask,
554 		       void (*callback)(int status,
555 					struct ib_sa_path_rec *resp,
556 					void *context),
557 		       void *context,
558 		       struct ib_sa_query **sa_query)
559 {
560 	struct ib_sa_path_query *query;
561 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
562 	struct ib_sa_port   *port;
563 	struct ib_mad_agent *agent;
564 	struct ib_sa_mad *mad;
565 	int ret;
566 
567 	if (!sa_dev)
568 		return -ENODEV;
569 
570 	port  = &sa_dev->port[port_num - sa_dev->start_port];
571 	agent = port->agent;
572 
573 	query = kmalloc(sizeof *query, gfp_mask);
574 	if (!query)
575 		return -ENOMEM;
576 
577 	query->sa_query.mad_buf = ib_create_send_mad(agent, 1, 0,
578 						     0, IB_MGMT_SA_HDR,
579 						     IB_MGMT_SA_DATA, gfp_mask);
580 	if (!query->sa_query.mad_buf) {
581 		ret = -ENOMEM;
582 		goto err1;
583 	}
584 
585 	query->callback = callback;
586 	query->context  = context;
587 
588 	mad = query->sa_query.mad_buf->mad;
589 	init_mad(mad, agent);
590 
591 	query->sa_query.callback = callback ? ib_sa_path_rec_callback : NULL;
592 	query->sa_query.release  = ib_sa_path_rec_release;
593 	query->sa_query.port     = port;
594 	mad->mad_hdr.method	 = IB_MGMT_METHOD_GET;
595 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_PATH_REC);
596 	mad->sa_hdr.comp_mask	 = comp_mask;
597 
598 	ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table), rec, mad->data);
599 
600 	*sa_query = &query->sa_query;
601 
602 	ret = send_mad(&query->sa_query, timeout_ms);
603 	if (ret < 0)
604 		goto err2;
605 
606 	return ret;
607 
608 err2:
609 	*sa_query = NULL;
610 	ib_free_send_mad(query->sa_query.mad_buf);
611 
612 err1:
613 	kfree(query);
614 	return ret;
615 }
616 EXPORT_SYMBOL(ib_sa_path_rec_get);
617 
618 static void ib_sa_service_rec_callback(struct ib_sa_query *sa_query,
619 				    int status,
620 				    struct ib_sa_mad *mad)
621 {
622 	struct ib_sa_service_query *query =
623 		container_of(sa_query, struct ib_sa_service_query, sa_query);
624 
625 	if (mad) {
626 		struct ib_sa_service_rec rec;
627 
628 		ib_unpack(service_rec_table, ARRAY_SIZE(service_rec_table),
629 			  mad->data, &rec);
630 		query->callback(status, &rec, query->context);
631 	} else
632 		query->callback(status, NULL, query->context);
633 }
634 
635 static void ib_sa_service_rec_release(struct ib_sa_query *sa_query)
636 {
637 	kfree(container_of(sa_query, struct ib_sa_service_query, sa_query));
638 }
639 
640 /**
641  * ib_sa_service_rec_query - Start Service Record operation
642  * @device:device to send request on
643  * @port_num: port number to send request on
644  * @method:SA method - should be get, set, or delete
645  * @rec:Service Record to send in request
646  * @comp_mask:component mask to send in request
647  * @timeout_ms:time to wait for response
648  * @gfp_mask:GFP mask to use for internal allocations
649  * @callback:function called when request completes, times out or is
650  * canceled
651  * @context:opaque user context passed to callback
652  * @sa_query:request context, used to cancel request
653  *
654  * Send a Service Record set/get/delete to the SA to register,
655  * unregister or query a service record.
656  * The callback function will be called when the request completes (or
657  * fails); status is 0 for a successful response, -EINTR if the query
658  * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
659  * occurred sending the query.  The resp parameter of the callback is
660  * only valid if status is 0.
661  *
662  * If the return value of ib_sa_service_rec_query() is negative, it is an
663  * error code.  Otherwise it is a request ID that can be used to cancel
664  * the query.
665  */
666 int ib_sa_service_rec_query(struct ib_device *device, u8 port_num, u8 method,
667 			    struct ib_sa_service_rec *rec,
668 			    ib_sa_comp_mask comp_mask,
669 			    int timeout_ms, gfp_t gfp_mask,
670 			    void (*callback)(int status,
671 					     struct ib_sa_service_rec *resp,
672 					     void *context),
673 			    void *context,
674 			    struct ib_sa_query **sa_query)
675 {
676 	struct ib_sa_service_query *query;
677 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
678 	struct ib_sa_port   *port;
679 	struct ib_mad_agent *agent;
680 	struct ib_sa_mad *mad;
681 	int ret;
682 
683 	if (!sa_dev)
684 		return -ENODEV;
685 
686 	port  = &sa_dev->port[port_num - sa_dev->start_port];
687 	agent = port->agent;
688 
689 	if (method != IB_MGMT_METHOD_GET &&
690 	    method != IB_MGMT_METHOD_SET &&
691 	    method != IB_SA_METHOD_DELETE)
692 		return -EINVAL;
693 
694 	query = kmalloc(sizeof *query, gfp_mask);
695 	if (!query)
696 		return -ENOMEM;
697 
698 	query->sa_query.mad_buf = ib_create_send_mad(agent, 1, 0,
699 						     0, IB_MGMT_SA_HDR,
700 						     IB_MGMT_SA_DATA, gfp_mask);
701 	if (!query->sa_query.mad_buf) {
702 		ret = -ENOMEM;
703 		goto err1;
704 	}
705 
706 	query->callback = callback;
707 	query->context  = context;
708 
709 	mad = query->sa_query.mad_buf->mad;
710 	init_mad(mad, agent);
711 
712 	query->sa_query.callback = callback ? ib_sa_service_rec_callback : NULL;
713 	query->sa_query.release  = ib_sa_service_rec_release;
714 	query->sa_query.port     = port;
715 	mad->mad_hdr.method	 = method;
716 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_SERVICE_REC);
717 	mad->sa_hdr.comp_mask	 = comp_mask;
718 
719 	ib_pack(service_rec_table, ARRAY_SIZE(service_rec_table),
720 		rec, mad->data);
721 
722 	*sa_query = &query->sa_query;
723 
724 	ret = send_mad(&query->sa_query, timeout_ms);
725 	if (ret < 0)
726 		goto err2;
727 
728 	return ret;
729 
730 err2:
731 	*sa_query = NULL;
732 	ib_free_send_mad(query->sa_query.mad_buf);
733 
734 err1:
735 	kfree(query);
736 	return ret;
737 }
738 EXPORT_SYMBOL(ib_sa_service_rec_query);
739 
740 static void ib_sa_mcmember_rec_callback(struct ib_sa_query *sa_query,
741 					int status,
742 					struct ib_sa_mad *mad)
743 {
744 	struct ib_sa_mcmember_query *query =
745 		container_of(sa_query, struct ib_sa_mcmember_query, sa_query);
746 
747 	if (mad) {
748 		struct ib_sa_mcmember_rec rec;
749 
750 		ib_unpack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
751 			  mad->data, &rec);
752 		query->callback(status, &rec, query->context);
753 	} else
754 		query->callback(status, NULL, query->context);
755 }
756 
757 static void ib_sa_mcmember_rec_release(struct ib_sa_query *sa_query)
758 {
759 	kfree(container_of(sa_query, struct ib_sa_mcmember_query, sa_query));
760 }
761 
762 int ib_sa_mcmember_rec_query(struct ib_device *device, u8 port_num,
763 			     u8 method,
764 			     struct ib_sa_mcmember_rec *rec,
765 			     ib_sa_comp_mask comp_mask,
766 			     int timeout_ms, gfp_t gfp_mask,
767 			     void (*callback)(int status,
768 					      struct ib_sa_mcmember_rec *resp,
769 					      void *context),
770 			     void *context,
771 			     struct ib_sa_query **sa_query)
772 {
773 	struct ib_sa_mcmember_query *query;
774 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
775 	struct ib_sa_port   *port;
776 	struct ib_mad_agent *agent;
777 	struct ib_sa_mad *mad;
778 	int ret;
779 
780 	if (!sa_dev)
781 		return -ENODEV;
782 
783 	port  = &sa_dev->port[port_num - sa_dev->start_port];
784 	agent = port->agent;
785 
786 	query = kmalloc(sizeof *query, gfp_mask);
787 	if (!query)
788 		return -ENOMEM;
789 
790 	query->sa_query.mad_buf = ib_create_send_mad(agent, 1, 0,
791 						     0, IB_MGMT_SA_HDR,
792 						     IB_MGMT_SA_DATA, gfp_mask);
793 	if (!query->sa_query.mad_buf) {
794 		ret = -ENOMEM;
795 		goto err1;
796 	}
797 
798 	query->callback = callback;
799 	query->context  = context;
800 
801 	mad = query->sa_query.mad_buf->mad;
802 	init_mad(mad, agent);
803 
804 	query->sa_query.callback = callback ? ib_sa_mcmember_rec_callback : NULL;
805 	query->sa_query.release  = ib_sa_mcmember_rec_release;
806 	query->sa_query.port     = port;
807 	mad->mad_hdr.method	 = method;
808 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_MC_MEMBER_REC);
809 	mad->sa_hdr.comp_mask	 = comp_mask;
810 
811 	ib_pack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
812 		rec, mad->data);
813 
814 	*sa_query = &query->sa_query;
815 
816 	ret = send_mad(&query->sa_query, timeout_ms);
817 	if (ret < 0)
818 		goto err2;
819 
820 	return ret;
821 
822 err2:
823 	*sa_query = NULL;
824 	ib_free_send_mad(query->sa_query.mad_buf);
825 
826 err1:
827 	kfree(query);
828 	return ret;
829 }
830 EXPORT_SYMBOL(ib_sa_mcmember_rec_query);
831 
832 static void send_handler(struct ib_mad_agent *agent,
833 			 struct ib_mad_send_wc *mad_send_wc)
834 {
835 	struct ib_sa_query *query = mad_send_wc->send_buf->context[0];
836 	unsigned long flags;
837 
838 	if (query->callback)
839 		switch (mad_send_wc->status) {
840 		case IB_WC_SUCCESS:
841 			/* No callback -- already got recv */
842 			break;
843 		case IB_WC_RESP_TIMEOUT_ERR:
844 			query->callback(query, -ETIMEDOUT, NULL);
845 			break;
846 		case IB_WC_WR_FLUSH_ERR:
847 			query->callback(query, -EINTR, NULL);
848 			break;
849 		default:
850 			query->callback(query, -EIO, NULL);
851 			break;
852 		}
853 
854 	spin_lock_irqsave(&idr_lock, flags);
855 	idr_remove(&query_idr, query->id);
856 	spin_unlock_irqrestore(&idr_lock, flags);
857 
858         ib_free_send_mad(mad_send_wc->send_buf);
859 	kref_put(&query->sm_ah->ref, free_sm_ah);
860 	query->release(query);
861 }
862 
863 static void recv_handler(struct ib_mad_agent *mad_agent,
864 			 struct ib_mad_recv_wc *mad_recv_wc)
865 {
866 	struct ib_sa_query *query;
867 	struct ib_mad_send_buf *mad_buf;
868 
869 	mad_buf = (void *) (unsigned long) mad_recv_wc->wc->wr_id;
870 	query = mad_buf->context[0];
871 
872 	if (query->callback) {
873 		if (mad_recv_wc->wc->status == IB_WC_SUCCESS)
874 			query->callback(query,
875 					mad_recv_wc->recv_buf.mad->mad_hdr.status ?
876 					-EINVAL : 0,
877 					(struct ib_sa_mad *) mad_recv_wc->recv_buf.mad);
878 		else
879 			query->callback(query, -EIO, NULL);
880 	}
881 
882 	ib_free_recv_mad(mad_recv_wc);
883 }
884 
885 static void ib_sa_add_one(struct ib_device *device)
886 {
887 	struct ib_sa_device *sa_dev;
888 	int s, e, i;
889 
890 	if (device->node_type == IB_NODE_SWITCH)
891 		s = e = 0;
892 	else {
893 		s = 1;
894 		e = device->phys_port_cnt;
895 	}
896 
897 	sa_dev = kmalloc(sizeof *sa_dev +
898 			 (e - s + 1) * sizeof (struct ib_sa_port),
899 			 GFP_KERNEL);
900 	if (!sa_dev)
901 		return;
902 
903 	sa_dev->start_port = s;
904 	sa_dev->end_port   = e;
905 
906 	for (i = 0; i <= e - s; ++i) {
907 		sa_dev->port[i].sm_ah    = NULL;
908 		sa_dev->port[i].port_num = i + s;
909 		spin_lock_init(&sa_dev->port[i].ah_lock);
910 
911 		sa_dev->port[i].agent =
912 			ib_register_mad_agent(device, i + s, IB_QPT_GSI,
913 					      NULL, 0, send_handler,
914 					      recv_handler, sa_dev);
915 		if (IS_ERR(sa_dev->port[i].agent))
916 			goto err;
917 
918 		INIT_WORK(&sa_dev->port[i].update_task,
919 			  update_sm_ah, &sa_dev->port[i]);
920 	}
921 
922 	ib_set_client_data(device, &sa_client, sa_dev);
923 
924 	/*
925 	 * We register our event handler after everything is set up,
926 	 * and then update our cached info after the event handler is
927 	 * registered to avoid any problems if a port changes state
928 	 * during our initialization.
929 	 */
930 
931 	INIT_IB_EVENT_HANDLER(&sa_dev->event_handler, device, ib_sa_event);
932 	if (ib_register_event_handler(&sa_dev->event_handler))
933 		goto err;
934 
935 	for (i = 0; i <= e - s; ++i)
936 		update_sm_ah(&sa_dev->port[i]);
937 
938 	return;
939 
940 err:
941 	while (--i >= 0)
942 		ib_unregister_mad_agent(sa_dev->port[i].agent);
943 
944 	kfree(sa_dev);
945 
946 	return;
947 }
948 
949 static void ib_sa_remove_one(struct ib_device *device)
950 {
951 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
952 	int i;
953 
954 	if (!sa_dev)
955 		return;
956 
957 	ib_unregister_event_handler(&sa_dev->event_handler);
958 
959 	for (i = 0; i <= sa_dev->end_port - sa_dev->start_port; ++i) {
960 		ib_unregister_mad_agent(sa_dev->port[i].agent);
961 		kref_put(&sa_dev->port[i].sm_ah->ref, free_sm_ah);
962 	}
963 
964 	kfree(sa_dev);
965 }
966 
967 static int __init ib_sa_init(void)
968 {
969 	int ret;
970 
971 	spin_lock_init(&idr_lock);
972 	spin_lock_init(&tid_lock);
973 
974 	get_random_bytes(&tid, sizeof tid);
975 
976 	ret = ib_register_client(&sa_client);
977 	if (ret)
978 		printk(KERN_ERR "Couldn't register ib_sa client\n");
979 
980 	return ret;
981 }
982 
983 static void __exit ib_sa_cleanup(void)
984 {
985 	ib_unregister_client(&sa_client);
986 	idr_destroy(&query_idr);
987 }
988 
989 module_init(ib_sa_init);
990 module_exit(ib_sa_cleanup);
991