1 /*
2  * Copyright (c) 2004 Topspin Communications.  All rights reserved.
3  * Copyright (c) 2005 Voltaire, Inc.  All rights reserved.
4  * Copyright (c) 2006 Intel Corporation.  All rights reserved.
5  *
6  * This software is available to you under a choice of one of two
7  * licenses.  You may choose to be licensed under the terms of the GNU
8  * General Public License (GPL) Version 2, available from the file
9  * COPYING in the main directory of this source tree, or the
10  * OpenIB.org BSD license below:
11  *
12  *     Redistribution and use in source and binary forms, with or
13  *     without modification, are permitted provided that the following
14  *     conditions are met:
15  *
16  *      - Redistributions of source code must retain the above
17  *        copyright notice, this list of conditions and the following
18  *        disclaimer.
19  *
20  *      - Redistributions in binary form must reproduce the above
21  *        copyright notice, this list of conditions and the following
22  *        disclaimer in the documentation and/or other materials
23  *        provided with the distribution.
24  *
25  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
26  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
27  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
28  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
29  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
30  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
31  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
32  * SOFTWARE.
33  */
34 
35 #include <linux/module.h>
36 #include <linux/init.h>
37 #include <linux/err.h>
38 #include <linux/random.h>
39 #include <linux/spinlock.h>
40 #include <linux/slab.h>
41 #include <linux/dma-mapping.h>
42 #include <linux/kref.h>
43 #include <linux/idr.h>
44 #include <linux/workqueue.h>
45 
46 #include <rdma/ib_pack.h>
47 #include <rdma/ib_cache.h>
48 #include "sa.h"
49 
50 MODULE_AUTHOR("Roland Dreier");
51 MODULE_DESCRIPTION("InfiniBand subnet administration query support");
52 MODULE_LICENSE("Dual BSD/GPL");
53 
54 struct ib_sa_sm_ah {
55 	struct ib_ah        *ah;
56 	struct kref          ref;
57 	u16		     pkey_index;
58 	u8		     src_path_mask;
59 };
60 
61 struct ib_sa_port {
62 	struct ib_mad_agent *agent;
63 	struct ib_sa_sm_ah  *sm_ah;
64 	struct work_struct   update_task;
65 	spinlock_t           ah_lock;
66 	u8                   port_num;
67 };
68 
69 struct ib_sa_device {
70 	int                     start_port, end_port;
71 	struct ib_event_handler event_handler;
72 	struct ib_sa_port port[0];
73 };
74 
75 struct ib_sa_query {
76 	void (*callback)(struct ib_sa_query *, int, struct ib_sa_mad *);
77 	void (*release)(struct ib_sa_query *);
78 	struct ib_sa_client    *client;
79 	struct ib_sa_port      *port;
80 	struct ib_mad_send_buf *mad_buf;
81 	struct ib_sa_sm_ah     *sm_ah;
82 	int			id;
83 };
84 
85 struct ib_sa_service_query {
86 	void (*callback)(int, struct ib_sa_service_rec *, void *);
87 	void *context;
88 	struct ib_sa_query sa_query;
89 };
90 
91 struct ib_sa_path_query {
92 	void (*callback)(int, struct ib_sa_path_rec *, void *);
93 	void *context;
94 	struct ib_sa_query sa_query;
95 };
96 
97 struct ib_sa_mcmember_query {
98 	void (*callback)(int, struct ib_sa_mcmember_rec *, void *);
99 	void *context;
100 	struct ib_sa_query sa_query;
101 };
102 
103 static void ib_sa_add_one(struct ib_device *device);
104 static void ib_sa_remove_one(struct ib_device *device);
105 
106 static struct ib_client sa_client = {
107 	.name   = "sa",
108 	.add    = ib_sa_add_one,
109 	.remove = ib_sa_remove_one
110 };
111 
112 static DEFINE_SPINLOCK(idr_lock);
113 static DEFINE_IDR(query_idr);
114 
115 static DEFINE_SPINLOCK(tid_lock);
116 static u32 tid;
117 
118 #define PATH_REC_FIELD(field) \
119 	.struct_offset_bytes = offsetof(struct ib_sa_path_rec, field),		\
120 	.struct_size_bytes   = sizeof ((struct ib_sa_path_rec *) 0)->field,	\
121 	.field_name          = "sa_path_rec:" #field
122 
123 static const struct ib_field path_rec_table[] = {
124 	{ PATH_REC_FIELD(service_id),
125 	  .offset_words = 0,
126 	  .offset_bits  = 0,
127 	  .size_bits    = 64 },
128 	{ PATH_REC_FIELD(dgid),
129 	  .offset_words = 2,
130 	  .offset_bits  = 0,
131 	  .size_bits    = 128 },
132 	{ PATH_REC_FIELD(sgid),
133 	  .offset_words = 6,
134 	  .offset_bits  = 0,
135 	  .size_bits    = 128 },
136 	{ PATH_REC_FIELD(dlid),
137 	  .offset_words = 10,
138 	  .offset_bits  = 0,
139 	  .size_bits    = 16 },
140 	{ PATH_REC_FIELD(slid),
141 	  .offset_words = 10,
142 	  .offset_bits  = 16,
143 	  .size_bits    = 16 },
144 	{ PATH_REC_FIELD(raw_traffic),
145 	  .offset_words = 11,
146 	  .offset_bits  = 0,
147 	  .size_bits    = 1 },
148 	{ RESERVED,
149 	  .offset_words = 11,
150 	  .offset_bits  = 1,
151 	  .size_bits    = 3 },
152 	{ PATH_REC_FIELD(flow_label),
153 	  .offset_words = 11,
154 	  .offset_bits  = 4,
155 	  .size_bits    = 20 },
156 	{ PATH_REC_FIELD(hop_limit),
157 	  .offset_words = 11,
158 	  .offset_bits  = 24,
159 	  .size_bits    = 8 },
160 	{ PATH_REC_FIELD(traffic_class),
161 	  .offset_words = 12,
162 	  .offset_bits  = 0,
163 	  .size_bits    = 8 },
164 	{ PATH_REC_FIELD(reversible),
165 	  .offset_words = 12,
166 	  .offset_bits  = 8,
167 	  .size_bits    = 1 },
168 	{ PATH_REC_FIELD(numb_path),
169 	  .offset_words = 12,
170 	  .offset_bits  = 9,
171 	  .size_bits    = 7 },
172 	{ PATH_REC_FIELD(pkey),
173 	  .offset_words = 12,
174 	  .offset_bits  = 16,
175 	  .size_bits    = 16 },
176 	{ PATH_REC_FIELD(qos_class),
177 	  .offset_words = 13,
178 	  .offset_bits  = 0,
179 	  .size_bits    = 12 },
180 	{ PATH_REC_FIELD(sl),
181 	  .offset_words = 13,
182 	  .offset_bits  = 12,
183 	  .size_bits    = 4 },
184 	{ PATH_REC_FIELD(mtu_selector),
185 	  .offset_words = 13,
186 	  .offset_bits  = 16,
187 	  .size_bits    = 2 },
188 	{ PATH_REC_FIELD(mtu),
189 	  .offset_words = 13,
190 	  .offset_bits  = 18,
191 	  .size_bits    = 6 },
192 	{ PATH_REC_FIELD(rate_selector),
193 	  .offset_words = 13,
194 	  .offset_bits  = 24,
195 	  .size_bits    = 2 },
196 	{ PATH_REC_FIELD(rate),
197 	  .offset_words = 13,
198 	  .offset_bits  = 26,
199 	  .size_bits    = 6 },
200 	{ PATH_REC_FIELD(packet_life_time_selector),
201 	  .offset_words = 14,
202 	  .offset_bits  = 0,
203 	  .size_bits    = 2 },
204 	{ PATH_REC_FIELD(packet_life_time),
205 	  .offset_words = 14,
206 	  .offset_bits  = 2,
207 	  .size_bits    = 6 },
208 	{ PATH_REC_FIELD(preference),
209 	  .offset_words = 14,
210 	  .offset_bits  = 8,
211 	  .size_bits    = 8 },
212 	{ RESERVED,
213 	  .offset_words = 14,
214 	  .offset_bits  = 16,
215 	  .size_bits    = 48 },
216 };
217 
218 #define MCMEMBER_REC_FIELD(field) \
219 	.struct_offset_bytes = offsetof(struct ib_sa_mcmember_rec, field),	\
220 	.struct_size_bytes   = sizeof ((struct ib_sa_mcmember_rec *) 0)->field,	\
221 	.field_name          = "sa_mcmember_rec:" #field
222 
223 static const struct ib_field mcmember_rec_table[] = {
224 	{ MCMEMBER_REC_FIELD(mgid),
225 	  .offset_words = 0,
226 	  .offset_bits  = 0,
227 	  .size_bits    = 128 },
228 	{ MCMEMBER_REC_FIELD(port_gid),
229 	  .offset_words = 4,
230 	  .offset_bits  = 0,
231 	  .size_bits    = 128 },
232 	{ MCMEMBER_REC_FIELD(qkey),
233 	  .offset_words = 8,
234 	  .offset_bits  = 0,
235 	  .size_bits    = 32 },
236 	{ MCMEMBER_REC_FIELD(mlid),
237 	  .offset_words = 9,
238 	  .offset_bits  = 0,
239 	  .size_bits    = 16 },
240 	{ MCMEMBER_REC_FIELD(mtu_selector),
241 	  .offset_words = 9,
242 	  .offset_bits  = 16,
243 	  .size_bits    = 2 },
244 	{ MCMEMBER_REC_FIELD(mtu),
245 	  .offset_words = 9,
246 	  .offset_bits  = 18,
247 	  .size_bits    = 6 },
248 	{ MCMEMBER_REC_FIELD(traffic_class),
249 	  .offset_words = 9,
250 	  .offset_bits  = 24,
251 	  .size_bits    = 8 },
252 	{ MCMEMBER_REC_FIELD(pkey),
253 	  .offset_words = 10,
254 	  .offset_bits  = 0,
255 	  .size_bits    = 16 },
256 	{ MCMEMBER_REC_FIELD(rate_selector),
257 	  .offset_words = 10,
258 	  .offset_bits  = 16,
259 	  .size_bits    = 2 },
260 	{ MCMEMBER_REC_FIELD(rate),
261 	  .offset_words = 10,
262 	  .offset_bits  = 18,
263 	  .size_bits    = 6 },
264 	{ MCMEMBER_REC_FIELD(packet_life_time_selector),
265 	  .offset_words = 10,
266 	  .offset_bits  = 24,
267 	  .size_bits    = 2 },
268 	{ MCMEMBER_REC_FIELD(packet_life_time),
269 	  .offset_words = 10,
270 	  .offset_bits  = 26,
271 	  .size_bits    = 6 },
272 	{ MCMEMBER_REC_FIELD(sl),
273 	  .offset_words = 11,
274 	  .offset_bits  = 0,
275 	  .size_bits    = 4 },
276 	{ MCMEMBER_REC_FIELD(flow_label),
277 	  .offset_words = 11,
278 	  .offset_bits  = 4,
279 	  .size_bits    = 20 },
280 	{ MCMEMBER_REC_FIELD(hop_limit),
281 	  .offset_words = 11,
282 	  .offset_bits  = 24,
283 	  .size_bits    = 8 },
284 	{ MCMEMBER_REC_FIELD(scope),
285 	  .offset_words = 12,
286 	  .offset_bits  = 0,
287 	  .size_bits    = 4 },
288 	{ MCMEMBER_REC_FIELD(join_state),
289 	  .offset_words = 12,
290 	  .offset_bits  = 4,
291 	  .size_bits    = 4 },
292 	{ MCMEMBER_REC_FIELD(proxy_join),
293 	  .offset_words = 12,
294 	  .offset_bits  = 8,
295 	  .size_bits    = 1 },
296 	{ RESERVED,
297 	  .offset_words = 12,
298 	  .offset_bits  = 9,
299 	  .size_bits    = 23 },
300 };
301 
302 #define SERVICE_REC_FIELD(field) \
303 	.struct_offset_bytes = offsetof(struct ib_sa_service_rec, field),	\
304 	.struct_size_bytes   = sizeof ((struct ib_sa_service_rec *) 0)->field,	\
305 	.field_name          = "sa_service_rec:" #field
306 
307 static const struct ib_field service_rec_table[] = {
308 	{ SERVICE_REC_FIELD(id),
309 	  .offset_words = 0,
310 	  .offset_bits  = 0,
311 	  .size_bits    = 64 },
312 	{ SERVICE_REC_FIELD(gid),
313 	  .offset_words = 2,
314 	  .offset_bits  = 0,
315 	  .size_bits    = 128 },
316 	{ SERVICE_REC_FIELD(pkey),
317 	  .offset_words = 6,
318 	  .offset_bits  = 0,
319 	  .size_bits    = 16 },
320 	{ SERVICE_REC_FIELD(lease),
321 	  .offset_words = 7,
322 	  .offset_bits  = 0,
323 	  .size_bits    = 32 },
324 	{ SERVICE_REC_FIELD(key),
325 	  .offset_words = 8,
326 	  .offset_bits  = 0,
327 	  .size_bits    = 128 },
328 	{ SERVICE_REC_FIELD(name),
329 	  .offset_words = 12,
330 	  .offset_bits  = 0,
331 	  .size_bits    = 64*8 },
332 	{ SERVICE_REC_FIELD(data8),
333 	  .offset_words = 28,
334 	  .offset_bits  = 0,
335 	  .size_bits    = 16*8 },
336 	{ SERVICE_REC_FIELD(data16),
337 	  .offset_words = 32,
338 	  .offset_bits  = 0,
339 	  .size_bits    = 8*16 },
340 	{ SERVICE_REC_FIELD(data32),
341 	  .offset_words = 36,
342 	  .offset_bits  = 0,
343 	  .size_bits    = 4*32 },
344 	{ SERVICE_REC_FIELD(data64),
345 	  .offset_words = 40,
346 	  .offset_bits  = 0,
347 	  .size_bits    = 2*64 },
348 };
349 
350 static void free_sm_ah(struct kref *kref)
351 {
352 	struct ib_sa_sm_ah *sm_ah = container_of(kref, struct ib_sa_sm_ah, ref);
353 
354 	ib_destroy_ah(sm_ah->ah);
355 	kfree(sm_ah);
356 }
357 
358 static void update_sm_ah(struct work_struct *work)
359 {
360 	struct ib_sa_port *port =
361 		container_of(work, struct ib_sa_port, update_task);
362 	struct ib_sa_sm_ah *new_ah;
363 	struct ib_port_attr port_attr;
364 	struct ib_ah_attr   ah_attr;
365 
366 	if (ib_query_port(port->agent->device, port->port_num, &port_attr)) {
367 		printk(KERN_WARNING "Couldn't query port\n");
368 		return;
369 	}
370 
371 	new_ah = kmalloc(sizeof *new_ah, GFP_KERNEL);
372 	if (!new_ah) {
373 		printk(KERN_WARNING "Couldn't allocate new SM AH\n");
374 		return;
375 	}
376 
377 	kref_init(&new_ah->ref);
378 	new_ah->src_path_mask = (1 << port_attr.lmc) - 1;
379 
380 	new_ah->pkey_index = 0;
381 	if (ib_find_pkey(port->agent->device, port->port_num,
382 			 IB_DEFAULT_PKEY_FULL, &new_ah->pkey_index))
383 		printk(KERN_ERR "Couldn't find index for default PKey\n");
384 
385 	memset(&ah_attr, 0, sizeof ah_attr);
386 	ah_attr.dlid     = port_attr.sm_lid;
387 	ah_attr.sl       = port_attr.sm_sl;
388 	ah_attr.port_num = port->port_num;
389 
390 	new_ah->ah = ib_create_ah(port->agent->qp->pd, &ah_attr);
391 	if (IS_ERR(new_ah->ah)) {
392 		printk(KERN_WARNING "Couldn't create new SM AH\n");
393 		kfree(new_ah);
394 		return;
395 	}
396 
397 	spin_lock_irq(&port->ah_lock);
398 	if (port->sm_ah)
399 		kref_put(&port->sm_ah->ref, free_sm_ah);
400 	port->sm_ah = new_ah;
401 	spin_unlock_irq(&port->ah_lock);
402 
403 }
404 
405 static void ib_sa_event(struct ib_event_handler *handler, struct ib_event *event)
406 {
407 	if (event->event == IB_EVENT_PORT_ERR    ||
408 	    event->event == IB_EVENT_PORT_ACTIVE ||
409 	    event->event == IB_EVENT_LID_CHANGE  ||
410 	    event->event == IB_EVENT_PKEY_CHANGE ||
411 	    event->event == IB_EVENT_SM_CHANGE   ||
412 	    event->event == IB_EVENT_CLIENT_REREGISTER) {
413 		unsigned long flags;
414 		struct ib_sa_device *sa_dev =
415 			container_of(handler, typeof(*sa_dev), event_handler);
416 		struct ib_sa_port *port =
417 			&sa_dev->port[event->element.port_num - sa_dev->start_port];
418 
419 		if (rdma_port_get_link_layer(handler->device, port->port_num) != IB_LINK_LAYER_INFINIBAND)
420 			return;
421 
422 		spin_lock_irqsave(&port->ah_lock, flags);
423 		if (port->sm_ah)
424 			kref_put(&port->sm_ah->ref, free_sm_ah);
425 		port->sm_ah = NULL;
426 		spin_unlock_irqrestore(&port->ah_lock, flags);
427 
428 		queue_work(ib_wq, &sa_dev->port[event->element.port_num -
429 					    sa_dev->start_port].update_task);
430 	}
431 }
432 
433 void ib_sa_register_client(struct ib_sa_client *client)
434 {
435 	atomic_set(&client->users, 1);
436 	init_completion(&client->comp);
437 }
438 EXPORT_SYMBOL(ib_sa_register_client);
439 
440 void ib_sa_unregister_client(struct ib_sa_client *client)
441 {
442 	ib_sa_client_put(client);
443 	wait_for_completion(&client->comp);
444 }
445 EXPORT_SYMBOL(ib_sa_unregister_client);
446 
447 /**
448  * ib_sa_cancel_query - try to cancel an SA query
449  * @id:ID of query to cancel
450  * @query:query pointer to cancel
451  *
452  * Try to cancel an SA query.  If the id and query don't match up or
453  * the query has already completed, nothing is done.  Otherwise the
454  * query is canceled and will complete with a status of -EINTR.
455  */
456 void ib_sa_cancel_query(int id, struct ib_sa_query *query)
457 {
458 	unsigned long flags;
459 	struct ib_mad_agent *agent;
460 	struct ib_mad_send_buf *mad_buf;
461 
462 	spin_lock_irqsave(&idr_lock, flags);
463 	if (idr_find(&query_idr, id) != query) {
464 		spin_unlock_irqrestore(&idr_lock, flags);
465 		return;
466 	}
467 	agent = query->port->agent;
468 	mad_buf = query->mad_buf;
469 	spin_unlock_irqrestore(&idr_lock, flags);
470 
471 	ib_cancel_mad(agent, mad_buf);
472 }
473 EXPORT_SYMBOL(ib_sa_cancel_query);
474 
475 static u8 get_src_path_mask(struct ib_device *device, u8 port_num)
476 {
477 	struct ib_sa_device *sa_dev;
478 	struct ib_sa_port   *port;
479 	unsigned long flags;
480 	u8 src_path_mask;
481 
482 	sa_dev = ib_get_client_data(device, &sa_client);
483 	if (!sa_dev)
484 		return 0x7f;
485 
486 	port  = &sa_dev->port[port_num - sa_dev->start_port];
487 	spin_lock_irqsave(&port->ah_lock, flags);
488 	src_path_mask = port->sm_ah ? port->sm_ah->src_path_mask : 0x7f;
489 	spin_unlock_irqrestore(&port->ah_lock, flags);
490 
491 	return src_path_mask;
492 }
493 
494 int ib_init_ah_from_path(struct ib_device *device, u8 port_num,
495 			 struct ib_sa_path_rec *rec, struct ib_ah_attr *ah_attr)
496 {
497 	int ret;
498 	u16 gid_index;
499 	int force_grh;
500 
501 	memset(ah_attr, 0, sizeof *ah_attr);
502 	ah_attr->dlid = be16_to_cpu(rec->dlid);
503 	ah_attr->sl = rec->sl;
504 	ah_attr->src_path_bits = be16_to_cpu(rec->slid) &
505 				 get_src_path_mask(device, port_num);
506 	ah_attr->port_num = port_num;
507 	ah_attr->static_rate = rec->rate;
508 
509 	force_grh = rdma_port_get_link_layer(device, port_num) == IB_LINK_LAYER_ETHERNET;
510 
511 	if (rec->hop_limit > 1 || force_grh) {
512 		ah_attr->ah_flags = IB_AH_GRH;
513 		ah_attr->grh.dgid = rec->dgid;
514 
515 		ret = ib_find_cached_gid(device, &rec->sgid, &port_num,
516 					 &gid_index);
517 		if (ret)
518 			return ret;
519 
520 		ah_attr->grh.sgid_index    = gid_index;
521 		ah_attr->grh.flow_label    = be32_to_cpu(rec->flow_label);
522 		ah_attr->grh.hop_limit     = rec->hop_limit;
523 		ah_attr->grh.traffic_class = rec->traffic_class;
524 	}
525 	return 0;
526 }
527 EXPORT_SYMBOL(ib_init_ah_from_path);
528 
529 static int alloc_mad(struct ib_sa_query *query, gfp_t gfp_mask)
530 {
531 	unsigned long flags;
532 
533 	spin_lock_irqsave(&query->port->ah_lock, flags);
534 	if (!query->port->sm_ah) {
535 		spin_unlock_irqrestore(&query->port->ah_lock, flags);
536 		return -EAGAIN;
537 	}
538 	kref_get(&query->port->sm_ah->ref);
539 	query->sm_ah = query->port->sm_ah;
540 	spin_unlock_irqrestore(&query->port->ah_lock, flags);
541 
542 	query->mad_buf = ib_create_send_mad(query->port->agent, 1,
543 					    query->sm_ah->pkey_index,
544 					    0, IB_MGMT_SA_HDR, IB_MGMT_SA_DATA,
545 					    gfp_mask);
546 	if (IS_ERR(query->mad_buf)) {
547 		kref_put(&query->sm_ah->ref, free_sm_ah);
548 		return -ENOMEM;
549 	}
550 
551 	query->mad_buf->ah = query->sm_ah->ah;
552 
553 	return 0;
554 }
555 
556 static void free_mad(struct ib_sa_query *query)
557 {
558 	ib_free_send_mad(query->mad_buf);
559 	kref_put(&query->sm_ah->ref, free_sm_ah);
560 }
561 
562 static void init_mad(struct ib_sa_mad *mad, struct ib_mad_agent *agent)
563 {
564 	unsigned long flags;
565 
566 	memset(mad, 0, sizeof *mad);
567 
568 	mad->mad_hdr.base_version  = IB_MGMT_BASE_VERSION;
569 	mad->mad_hdr.mgmt_class    = IB_MGMT_CLASS_SUBN_ADM;
570 	mad->mad_hdr.class_version = IB_SA_CLASS_VERSION;
571 
572 	spin_lock_irqsave(&tid_lock, flags);
573 	mad->mad_hdr.tid           =
574 		cpu_to_be64(((u64) agent->hi_tid) << 32 | tid++);
575 	spin_unlock_irqrestore(&tid_lock, flags);
576 }
577 
578 static int send_mad(struct ib_sa_query *query, int timeout_ms, gfp_t gfp_mask)
579 {
580 	unsigned long flags;
581 	int ret, id;
582 
583 retry:
584 	if (!idr_pre_get(&query_idr, gfp_mask))
585 		return -ENOMEM;
586 	spin_lock_irqsave(&idr_lock, flags);
587 	ret = idr_get_new(&query_idr, query, &id);
588 	spin_unlock_irqrestore(&idr_lock, flags);
589 	if (ret == -EAGAIN)
590 		goto retry;
591 	if (ret)
592 		return ret;
593 
594 	query->mad_buf->timeout_ms  = timeout_ms;
595 	query->mad_buf->context[0] = query;
596 	query->id = id;
597 
598 	ret = ib_post_send_mad(query->mad_buf, NULL);
599 	if (ret) {
600 		spin_lock_irqsave(&idr_lock, flags);
601 		idr_remove(&query_idr, id);
602 		spin_unlock_irqrestore(&idr_lock, flags);
603 	}
604 
605 	/*
606 	 * It's not safe to dereference query any more, because the
607 	 * send may already have completed and freed the query in
608 	 * another context.
609 	 */
610 	return ret ? ret : id;
611 }
612 
613 void ib_sa_unpack_path(void *attribute, struct ib_sa_path_rec *rec)
614 {
615 	ib_unpack(path_rec_table, ARRAY_SIZE(path_rec_table), attribute, rec);
616 }
617 EXPORT_SYMBOL(ib_sa_unpack_path);
618 
619 static void ib_sa_path_rec_callback(struct ib_sa_query *sa_query,
620 				    int status,
621 				    struct ib_sa_mad *mad)
622 {
623 	struct ib_sa_path_query *query =
624 		container_of(sa_query, struct ib_sa_path_query, sa_query);
625 
626 	if (mad) {
627 		struct ib_sa_path_rec rec;
628 
629 		ib_unpack(path_rec_table, ARRAY_SIZE(path_rec_table),
630 			  mad->data, &rec);
631 		query->callback(status, &rec, query->context);
632 	} else
633 		query->callback(status, NULL, query->context);
634 }
635 
636 static void ib_sa_path_rec_release(struct ib_sa_query *sa_query)
637 {
638 	kfree(container_of(sa_query, struct ib_sa_path_query, sa_query));
639 }
640 
641 /**
642  * ib_sa_path_rec_get - Start a Path get query
643  * @client:SA client
644  * @device:device to send query on
645  * @port_num: port number to send query on
646  * @rec:Path Record to send in query
647  * @comp_mask:component mask to send in query
648  * @timeout_ms:time to wait for response
649  * @gfp_mask:GFP mask to use for internal allocations
650  * @callback:function called when query completes, times out or is
651  * canceled
652  * @context:opaque user context passed to callback
653  * @sa_query:query context, used to cancel query
654  *
655  * Send a Path Record Get query to the SA to look up a path.  The
656  * callback function will be called when the query 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_path_rec_get() is negative, it is an
663  * error code.  Otherwise it is a query ID that can be used to cancel
664  * the query.
665  */
666 int ib_sa_path_rec_get(struct ib_sa_client *client,
667 		       struct ib_device *device, u8 port_num,
668 		       struct ib_sa_path_rec *rec,
669 		       ib_sa_comp_mask comp_mask,
670 		       int timeout_ms, gfp_t gfp_mask,
671 		       void (*callback)(int status,
672 					struct ib_sa_path_rec *resp,
673 					void *context),
674 		       void *context,
675 		       struct ib_sa_query **sa_query)
676 {
677 	struct ib_sa_path_query *query;
678 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
679 	struct ib_sa_port   *port;
680 	struct ib_mad_agent *agent;
681 	struct ib_sa_mad *mad;
682 	int ret;
683 
684 	if (!sa_dev)
685 		return -ENODEV;
686 
687 	port  = &sa_dev->port[port_num - sa_dev->start_port];
688 	agent = port->agent;
689 
690 	query = kmalloc(sizeof *query, gfp_mask);
691 	if (!query)
692 		return -ENOMEM;
693 
694 	query->sa_query.port     = port;
695 	ret = alloc_mad(&query->sa_query, gfp_mask);
696 	if (ret)
697 		goto err1;
698 
699 	ib_sa_client_get(client);
700 	query->sa_query.client = client;
701 	query->callback        = callback;
702 	query->context         = context;
703 
704 	mad = query->sa_query.mad_buf->mad;
705 	init_mad(mad, agent);
706 
707 	query->sa_query.callback = callback ? ib_sa_path_rec_callback : NULL;
708 	query->sa_query.release  = ib_sa_path_rec_release;
709 	mad->mad_hdr.method	 = IB_MGMT_METHOD_GET;
710 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_PATH_REC);
711 	mad->sa_hdr.comp_mask	 = comp_mask;
712 
713 	ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table), rec, mad->data);
714 
715 	*sa_query = &query->sa_query;
716 
717 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
718 	if (ret < 0)
719 		goto err2;
720 
721 	return ret;
722 
723 err2:
724 	*sa_query = NULL;
725 	ib_sa_client_put(query->sa_query.client);
726 	free_mad(&query->sa_query);
727 
728 err1:
729 	kfree(query);
730 	return ret;
731 }
732 EXPORT_SYMBOL(ib_sa_path_rec_get);
733 
734 static void ib_sa_service_rec_callback(struct ib_sa_query *sa_query,
735 				    int status,
736 				    struct ib_sa_mad *mad)
737 {
738 	struct ib_sa_service_query *query =
739 		container_of(sa_query, struct ib_sa_service_query, sa_query);
740 
741 	if (mad) {
742 		struct ib_sa_service_rec rec;
743 
744 		ib_unpack(service_rec_table, ARRAY_SIZE(service_rec_table),
745 			  mad->data, &rec);
746 		query->callback(status, &rec, query->context);
747 	} else
748 		query->callback(status, NULL, query->context);
749 }
750 
751 static void ib_sa_service_rec_release(struct ib_sa_query *sa_query)
752 {
753 	kfree(container_of(sa_query, struct ib_sa_service_query, sa_query));
754 }
755 
756 /**
757  * ib_sa_service_rec_query - Start Service Record operation
758  * @client:SA client
759  * @device:device to send request on
760  * @port_num: port number to send request on
761  * @method:SA method - should be get, set, or delete
762  * @rec:Service Record to send in request
763  * @comp_mask:component mask to send in request
764  * @timeout_ms:time to wait for response
765  * @gfp_mask:GFP mask to use for internal allocations
766  * @callback:function called when request completes, times out or is
767  * canceled
768  * @context:opaque user context passed to callback
769  * @sa_query:request context, used to cancel request
770  *
771  * Send a Service Record set/get/delete to the SA to register,
772  * unregister or query a service record.
773  * The callback function will be called when the request completes (or
774  * fails); status is 0 for a successful response, -EINTR if the query
775  * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
776  * occurred sending the query.  The resp parameter of the callback is
777  * only valid if status is 0.
778  *
779  * If the return value of ib_sa_service_rec_query() is negative, it is an
780  * error code.  Otherwise it is a request ID that can be used to cancel
781  * the query.
782  */
783 int ib_sa_service_rec_query(struct ib_sa_client *client,
784 			    struct ib_device *device, u8 port_num, u8 method,
785 			    struct ib_sa_service_rec *rec,
786 			    ib_sa_comp_mask comp_mask,
787 			    int timeout_ms, gfp_t gfp_mask,
788 			    void (*callback)(int status,
789 					     struct ib_sa_service_rec *resp,
790 					     void *context),
791 			    void *context,
792 			    struct ib_sa_query **sa_query)
793 {
794 	struct ib_sa_service_query *query;
795 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
796 	struct ib_sa_port   *port;
797 	struct ib_mad_agent *agent;
798 	struct ib_sa_mad *mad;
799 	int ret;
800 
801 	if (!sa_dev)
802 		return -ENODEV;
803 
804 	port  = &sa_dev->port[port_num - sa_dev->start_port];
805 	agent = port->agent;
806 
807 	if (method != IB_MGMT_METHOD_GET &&
808 	    method != IB_MGMT_METHOD_SET &&
809 	    method != IB_SA_METHOD_DELETE)
810 		return -EINVAL;
811 
812 	query = kmalloc(sizeof *query, gfp_mask);
813 	if (!query)
814 		return -ENOMEM;
815 
816 	query->sa_query.port     = port;
817 	ret = alloc_mad(&query->sa_query, gfp_mask);
818 	if (ret)
819 		goto err1;
820 
821 	ib_sa_client_get(client);
822 	query->sa_query.client = client;
823 	query->callback        = callback;
824 	query->context         = context;
825 
826 	mad = query->sa_query.mad_buf->mad;
827 	init_mad(mad, agent);
828 
829 	query->sa_query.callback = callback ? ib_sa_service_rec_callback : NULL;
830 	query->sa_query.release  = ib_sa_service_rec_release;
831 	mad->mad_hdr.method	 = method;
832 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_SERVICE_REC);
833 	mad->sa_hdr.comp_mask	 = comp_mask;
834 
835 	ib_pack(service_rec_table, ARRAY_SIZE(service_rec_table),
836 		rec, mad->data);
837 
838 	*sa_query = &query->sa_query;
839 
840 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
841 	if (ret < 0)
842 		goto err2;
843 
844 	return ret;
845 
846 err2:
847 	*sa_query = NULL;
848 	ib_sa_client_put(query->sa_query.client);
849 	free_mad(&query->sa_query);
850 
851 err1:
852 	kfree(query);
853 	return ret;
854 }
855 EXPORT_SYMBOL(ib_sa_service_rec_query);
856 
857 static void ib_sa_mcmember_rec_callback(struct ib_sa_query *sa_query,
858 					int status,
859 					struct ib_sa_mad *mad)
860 {
861 	struct ib_sa_mcmember_query *query =
862 		container_of(sa_query, struct ib_sa_mcmember_query, sa_query);
863 
864 	if (mad) {
865 		struct ib_sa_mcmember_rec rec;
866 
867 		ib_unpack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
868 			  mad->data, &rec);
869 		query->callback(status, &rec, query->context);
870 	} else
871 		query->callback(status, NULL, query->context);
872 }
873 
874 static void ib_sa_mcmember_rec_release(struct ib_sa_query *sa_query)
875 {
876 	kfree(container_of(sa_query, struct ib_sa_mcmember_query, sa_query));
877 }
878 
879 int ib_sa_mcmember_rec_query(struct ib_sa_client *client,
880 			     struct ib_device *device, u8 port_num,
881 			     u8 method,
882 			     struct ib_sa_mcmember_rec *rec,
883 			     ib_sa_comp_mask comp_mask,
884 			     int timeout_ms, gfp_t gfp_mask,
885 			     void (*callback)(int status,
886 					      struct ib_sa_mcmember_rec *resp,
887 					      void *context),
888 			     void *context,
889 			     struct ib_sa_query **sa_query)
890 {
891 	struct ib_sa_mcmember_query *query;
892 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
893 	struct ib_sa_port   *port;
894 	struct ib_mad_agent *agent;
895 	struct ib_sa_mad *mad;
896 	int ret;
897 
898 	if (!sa_dev)
899 		return -ENODEV;
900 
901 	port  = &sa_dev->port[port_num - sa_dev->start_port];
902 	agent = port->agent;
903 
904 	query = kmalloc(sizeof *query, gfp_mask);
905 	if (!query)
906 		return -ENOMEM;
907 
908 	query->sa_query.port     = port;
909 	ret = alloc_mad(&query->sa_query, gfp_mask);
910 	if (ret)
911 		goto err1;
912 
913 	ib_sa_client_get(client);
914 	query->sa_query.client = client;
915 	query->callback        = callback;
916 	query->context         = context;
917 
918 	mad = query->sa_query.mad_buf->mad;
919 	init_mad(mad, agent);
920 
921 	query->sa_query.callback = callback ? ib_sa_mcmember_rec_callback : NULL;
922 	query->sa_query.release  = ib_sa_mcmember_rec_release;
923 	mad->mad_hdr.method	 = method;
924 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_MC_MEMBER_REC);
925 	mad->sa_hdr.comp_mask	 = comp_mask;
926 
927 	ib_pack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
928 		rec, mad->data);
929 
930 	*sa_query = &query->sa_query;
931 
932 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
933 	if (ret < 0)
934 		goto err2;
935 
936 	return ret;
937 
938 err2:
939 	*sa_query = NULL;
940 	ib_sa_client_put(query->sa_query.client);
941 	free_mad(&query->sa_query);
942 
943 err1:
944 	kfree(query);
945 	return ret;
946 }
947 
948 static void send_handler(struct ib_mad_agent *agent,
949 			 struct ib_mad_send_wc *mad_send_wc)
950 {
951 	struct ib_sa_query *query = mad_send_wc->send_buf->context[0];
952 	unsigned long flags;
953 
954 	if (query->callback)
955 		switch (mad_send_wc->status) {
956 		case IB_WC_SUCCESS:
957 			/* No callback -- already got recv */
958 			break;
959 		case IB_WC_RESP_TIMEOUT_ERR:
960 			query->callback(query, -ETIMEDOUT, NULL);
961 			break;
962 		case IB_WC_WR_FLUSH_ERR:
963 			query->callback(query, -EINTR, NULL);
964 			break;
965 		default:
966 			query->callback(query, -EIO, NULL);
967 			break;
968 		}
969 
970 	spin_lock_irqsave(&idr_lock, flags);
971 	idr_remove(&query_idr, query->id);
972 	spin_unlock_irqrestore(&idr_lock, flags);
973 
974 	free_mad(query);
975 	ib_sa_client_put(query->client);
976 	query->release(query);
977 }
978 
979 static void recv_handler(struct ib_mad_agent *mad_agent,
980 			 struct ib_mad_recv_wc *mad_recv_wc)
981 {
982 	struct ib_sa_query *query;
983 	struct ib_mad_send_buf *mad_buf;
984 
985 	mad_buf = (void *) (unsigned long) mad_recv_wc->wc->wr_id;
986 	query = mad_buf->context[0];
987 
988 	if (query->callback) {
989 		if (mad_recv_wc->wc->status == IB_WC_SUCCESS)
990 			query->callback(query,
991 					mad_recv_wc->recv_buf.mad->mad_hdr.status ?
992 					-EINVAL : 0,
993 					(struct ib_sa_mad *) mad_recv_wc->recv_buf.mad);
994 		else
995 			query->callback(query, -EIO, NULL);
996 	}
997 
998 	ib_free_recv_mad(mad_recv_wc);
999 }
1000 
1001 static void ib_sa_add_one(struct ib_device *device)
1002 {
1003 	struct ib_sa_device *sa_dev;
1004 	int s, e, i;
1005 
1006 	if (rdma_node_get_transport(device->node_type) != RDMA_TRANSPORT_IB)
1007 		return;
1008 
1009 	if (device->node_type == RDMA_NODE_IB_SWITCH)
1010 		s = e = 0;
1011 	else {
1012 		s = 1;
1013 		e = device->phys_port_cnt;
1014 	}
1015 
1016 	sa_dev = kzalloc(sizeof *sa_dev +
1017 			 (e - s + 1) * sizeof (struct ib_sa_port),
1018 			 GFP_KERNEL);
1019 	if (!sa_dev)
1020 		return;
1021 
1022 	sa_dev->start_port = s;
1023 	sa_dev->end_port   = e;
1024 
1025 	for (i = 0; i <= e - s; ++i) {
1026 		spin_lock_init(&sa_dev->port[i].ah_lock);
1027 		if (rdma_port_get_link_layer(device, i + 1) != IB_LINK_LAYER_INFINIBAND)
1028 			continue;
1029 
1030 		sa_dev->port[i].sm_ah    = NULL;
1031 		sa_dev->port[i].port_num = i + s;
1032 
1033 		sa_dev->port[i].agent =
1034 			ib_register_mad_agent(device, i + s, IB_QPT_GSI,
1035 					      NULL, 0, send_handler,
1036 					      recv_handler, sa_dev);
1037 		if (IS_ERR(sa_dev->port[i].agent))
1038 			goto err;
1039 
1040 		INIT_WORK(&sa_dev->port[i].update_task, update_sm_ah);
1041 	}
1042 
1043 	ib_set_client_data(device, &sa_client, sa_dev);
1044 
1045 	/*
1046 	 * We register our event handler after everything is set up,
1047 	 * and then update our cached info after the event handler is
1048 	 * registered to avoid any problems if a port changes state
1049 	 * during our initialization.
1050 	 */
1051 
1052 	INIT_IB_EVENT_HANDLER(&sa_dev->event_handler, device, ib_sa_event);
1053 	if (ib_register_event_handler(&sa_dev->event_handler))
1054 		goto err;
1055 
1056 	for (i = 0; i <= e - s; ++i)
1057 		if (rdma_port_get_link_layer(device, i + 1) == IB_LINK_LAYER_INFINIBAND)
1058 			update_sm_ah(&sa_dev->port[i].update_task);
1059 
1060 	return;
1061 
1062 err:
1063 	while (--i >= 0)
1064 		if (rdma_port_get_link_layer(device, i + 1) == IB_LINK_LAYER_INFINIBAND)
1065 			ib_unregister_mad_agent(sa_dev->port[i].agent);
1066 
1067 	kfree(sa_dev);
1068 
1069 	return;
1070 }
1071 
1072 static void ib_sa_remove_one(struct ib_device *device)
1073 {
1074 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1075 	int i;
1076 
1077 	if (!sa_dev)
1078 		return;
1079 
1080 	ib_unregister_event_handler(&sa_dev->event_handler);
1081 
1082 	flush_workqueue(ib_wq);
1083 
1084 	for (i = 0; i <= sa_dev->end_port - sa_dev->start_port; ++i) {
1085 		if (rdma_port_get_link_layer(device, i + 1) == IB_LINK_LAYER_INFINIBAND) {
1086 			ib_unregister_mad_agent(sa_dev->port[i].agent);
1087 			if (sa_dev->port[i].sm_ah)
1088 				kref_put(&sa_dev->port[i].sm_ah->ref, free_sm_ah);
1089 		}
1090 
1091 	}
1092 
1093 	kfree(sa_dev);
1094 }
1095 
1096 static int __init ib_sa_init(void)
1097 {
1098 	int ret;
1099 
1100 	get_random_bytes(&tid, sizeof tid);
1101 
1102 	ret = ib_register_client(&sa_client);
1103 	if (ret) {
1104 		printk(KERN_ERR "Couldn't register ib_sa client\n");
1105 		goto err1;
1106 	}
1107 
1108 	ret = mcast_init();
1109 	if (ret) {
1110 		printk(KERN_ERR "Couldn't initialize multicast handling\n");
1111 		goto err2;
1112 	}
1113 
1114 	return 0;
1115 err2:
1116 	ib_unregister_client(&sa_client);
1117 err1:
1118 	return ret;
1119 }
1120 
1121 static void __exit ib_sa_cleanup(void)
1122 {
1123 	mcast_cleanup();
1124 	ib_unregister_client(&sa_client);
1125 	idr_destroy(&query_idr);
1126 }
1127 
1128 module_init(ib_sa_init);
1129 module_exit(ib_sa_cleanup);
1130