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 #include <uapi/linux/if_ether.h>
46 #include <rdma/ib_pack.h>
47 #include <rdma/ib_cache.h>
48 #include <rdma/rdma_netlink.h>
49 #include <net/netlink.h>
50 #include <uapi/rdma/ib_user_sa.h>
51 #include <rdma/ib_marshall.h>
52 #include <rdma/ib_addr.h>
53 #include "sa.h"
54 #include "core_priv.h"
55 
56 #define IB_SA_LOCAL_SVC_TIMEOUT_MIN		100
57 #define IB_SA_LOCAL_SVC_TIMEOUT_DEFAULT		2000
58 #define IB_SA_LOCAL_SVC_TIMEOUT_MAX		200000
59 #define IB_SA_CPI_MAX_RETRY_CNT			3
60 #define IB_SA_CPI_RETRY_WAIT			1000 /*msecs */
61 static int sa_local_svc_timeout_ms = IB_SA_LOCAL_SVC_TIMEOUT_DEFAULT;
62 
63 struct ib_sa_sm_ah {
64 	struct ib_ah        *ah;
65 	struct kref          ref;
66 	u16		     pkey_index;
67 	u8		     src_path_mask;
68 };
69 
70 enum rdma_class_port_info_type {
71 	RDMA_CLASS_PORT_INFO_IB,
72 	RDMA_CLASS_PORT_INFO_OPA
73 };
74 
75 struct rdma_class_port_info {
76 	enum rdma_class_port_info_type type;
77 	union {
78 		struct ib_class_port_info ib;
79 		struct opa_class_port_info opa;
80 	};
81 };
82 
83 struct ib_sa_classport_cache {
84 	bool valid;
85 	int retry_cnt;
86 	struct rdma_class_port_info data;
87 };
88 
89 struct ib_sa_port {
90 	struct ib_mad_agent *agent;
91 	struct ib_sa_sm_ah  *sm_ah;
92 	struct work_struct   update_task;
93 	struct ib_sa_classport_cache classport_info;
94 	struct delayed_work ib_cpi_work;
95 	spinlock_t                   classport_lock; /* protects class port info set */
96 	spinlock_t           ah_lock;
97 	u8                   port_num;
98 };
99 
100 struct ib_sa_device {
101 	int                     start_port, end_port;
102 	struct ib_event_handler event_handler;
103 	struct ib_sa_port port[0];
104 };
105 
106 struct ib_sa_query {
107 	void (*callback)(struct ib_sa_query *, int, struct ib_sa_mad *);
108 	void (*release)(struct ib_sa_query *);
109 	struct ib_sa_client    *client;
110 	struct ib_sa_port      *port;
111 	struct ib_mad_send_buf *mad_buf;
112 	struct ib_sa_sm_ah     *sm_ah;
113 	int			id;
114 	u32			flags;
115 	struct list_head	list; /* Local svc request list */
116 	u32			seq; /* Local svc request sequence number */
117 	unsigned long		timeout; /* Local svc timeout */
118 	u8			path_use; /* How will the pathrecord be used */
119 };
120 
121 #define IB_SA_ENABLE_LOCAL_SERVICE	0x00000001
122 #define IB_SA_CANCEL			0x00000002
123 #define IB_SA_QUERY_OPA			0x00000004
124 
125 struct ib_sa_service_query {
126 	void (*callback)(int, struct ib_sa_service_rec *, void *);
127 	void *context;
128 	struct ib_sa_query sa_query;
129 };
130 
131 struct ib_sa_path_query {
132 	void (*callback)(int, struct sa_path_rec *, void *);
133 	void *context;
134 	struct ib_sa_query sa_query;
135 	struct sa_path_rec *conv_pr;
136 };
137 
138 struct ib_sa_guidinfo_query {
139 	void (*callback)(int, struct ib_sa_guidinfo_rec *, void *);
140 	void *context;
141 	struct ib_sa_query sa_query;
142 };
143 
144 struct ib_sa_classport_info_query {
145 	void (*callback)(void *);
146 	void *context;
147 	struct ib_sa_query sa_query;
148 };
149 
150 struct ib_sa_mcmember_query {
151 	void (*callback)(int, struct ib_sa_mcmember_rec *, void *);
152 	void *context;
153 	struct ib_sa_query sa_query;
154 };
155 
156 static LIST_HEAD(ib_nl_request_list);
157 static DEFINE_SPINLOCK(ib_nl_request_lock);
158 static atomic_t ib_nl_sa_request_seq;
159 static struct workqueue_struct *ib_nl_wq;
160 static struct delayed_work ib_nl_timed_work;
161 static const struct nla_policy ib_nl_policy[LS_NLA_TYPE_MAX] = {
162 	[LS_NLA_TYPE_PATH_RECORD]	= {.type = NLA_BINARY,
163 		.len = sizeof(struct ib_path_rec_data)},
164 	[LS_NLA_TYPE_TIMEOUT]		= {.type = NLA_U32},
165 	[LS_NLA_TYPE_SERVICE_ID]	= {.type = NLA_U64},
166 	[LS_NLA_TYPE_DGID]		= {.type = NLA_BINARY,
167 		.len = sizeof(struct rdma_nla_ls_gid)},
168 	[LS_NLA_TYPE_SGID]		= {.type = NLA_BINARY,
169 		.len = sizeof(struct rdma_nla_ls_gid)},
170 	[LS_NLA_TYPE_TCLASS]		= {.type = NLA_U8},
171 	[LS_NLA_TYPE_PKEY]		= {.type = NLA_U16},
172 	[LS_NLA_TYPE_QOS_CLASS]		= {.type = NLA_U16},
173 };
174 
175 
176 static void ib_sa_add_one(struct ib_device *device);
177 static void ib_sa_remove_one(struct ib_device *device, void *client_data);
178 
179 static struct ib_client sa_client = {
180 	.name   = "sa",
181 	.add    = ib_sa_add_one,
182 	.remove = ib_sa_remove_one
183 };
184 
185 static DEFINE_SPINLOCK(idr_lock);
186 static DEFINE_IDR(query_idr);
187 
188 static DEFINE_SPINLOCK(tid_lock);
189 static u32 tid;
190 
191 #define PATH_REC_FIELD(field) \
192 	.struct_offset_bytes = offsetof(struct sa_path_rec, field),	\
193 	.struct_size_bytes   = sizeof((struct sa_path_rec *)0)->field,	\
194 	.field_name          = "sa_path_rec:" #field
195 
196 static const struct ib_field path_rec_table[] = {
197 	{ PATH_REC_FIELD(service_id),
198 	  .offset_words = 0,
199 	  .offset_bits  = 0,
200 	  .size_bits    = 64 },
201 	{ PATH_REC_FIELD(dgid),
202 	  .offset_words = 2,
203 	  .offset_bits  = 0,
204 	  .size_bits    = 128 },
205 	{ PATH_REC_FIELD(sgid),
206 	  .offset_words = 6,
207 	  .offset_bits  = 0,
208 	  .size_bits    = 128 },
209 	{ PATH_REC_FIELD(ib.dlid),
210 	  .offset_words = 10,
211 	  .offset_bits  = 0,
212 	  .size_bits    = 16 },
213 	{ PATH_REC_FIELD(ib.slid),
214 	  .offset_words = 10,
215 	  .offset_bits  = 16,
216 	  .size_bits    = 16 },
217 	{ PATH_REC_FIELD(ib.raw_traffic),
218 	  .offset_words = 11,
219 	  .offset_bits  = 0,
220 	  .size_bits    = 1 },
221 	{ RESERVED,
222 	  .offset_words = 11,
223 	  .offset_bits  = 1,
224 	  .size_bits    = 3 },
225 	{ PATH_REC_FIELD(flow_label),
226 	  .offset_words = 11,
227 	  .offset_bits  = 4,
228 	  .size_bits    = 20 },
229 	{ PATH_REC_FIELD(hop_limit),
230 	  .offset_words = 11,
231 	  .offset_bits  = 24,
232 	  .size_bits    = 8 },
233 	{ PATH_REC_FIELD(traffic_class),
234 	  .offset_words = 12,
235 	  .offset_bits  = 0,
236 	  .size_bits    = 8 },
237 	{ PATH_REC_FIELD(reversible),
238 	  .offset_words = 12,
239 	  .offset_bits  = 8,
240 	  .size_bits    = 1 },
241 	{ PATH_REC_FIELD(numb_path),
242 	  .offset_words = 12,
243 	  .offset_bits  = 9,
244 	  .size_bits    = 7 },
245 	{ PATH_REC_FIELD(pkey),
246 	  .offset_words = 12,
247 	  .offset_bits  = 16,
248 	  .size_bits    = 16 },
249 	{ PATH_REC_FIELD(qos_class),
250 	  .offset_words = 13,
251 	  .offset_bits  = 0,
252 	  .size_bits    = 12 },
253 	{ PATH_REC_FIELD(sl),
254 	  .offset_words = 13,
255 	  .offset_bits  = 12,
256 	  .size_bits    = 4 },
257 	{ PATH_REC_FIELD(mtu_selector),
258 	  .offset_words = 13,
259 	  .offset_bits  = 16,
260 	  .size_bits    = 2 },
261 	{ PATH_REC_FIELD(mtu),
262 	  .offset_words = 13,
263 	  .offset_bits  = 18,
264 	  .size_bits    = 6 },
265 	{ PATH_REC_FIELD(rate_selector),
266 	  .offset_words = 13,
267 	  .offset_bits  = 24,
268 	  .size_bits    = 2 },
269 	{ PATH_REC_FIELD(rate),
270 	  .offset_words = 13,
271 	  .offset_bits  = 26,
272 	  .size_bits    = 6 },
273 	{ PATH_REC_FIELD(packet_life_time_selector),
274 	  .offset_words = 14,
275 	  .offset_bits  = 0,
276 	  .size_bits    = 2 },
277 	{ PATH_REC_FIELD(packet_life_time),
278 	  .offset_words = 14,
279 	  .offset_bits  = 2,
280 	  .size_bits    = 6 },
281 	{ PATH_REC_FIELD(preference),
282 	  .offset_words = 14,
283 	  .offset_bits  = 8,
284 	  .size_bits    = 8 },
285 	{ RESERVED,
286 	  .offset_words = 14,
287 	  .offset_bits  = 16,
288 	  .size_bits    = 48 },
289 };
290 
291 #define OPA_PATH_REC_FIELD(field) \
292 	.struct_offset_bytes = \
293 		offsetof(struct sa_path_rec, field), \
294 	.struct_size_bytes   = \
295 		sizeof((struct sa_path_rec *)0)->field,	\
296 	.field_name          = "sa_path_rec:" #field
297 
298 static const struct ib_field opa_path_rec_table[] = {
299 	{ OPA_PATH_REC_FIELD(service_id),
300 	  .offset_words = 0,
301 	  .offset_bits  = 0,
302 	  .size_bits    = 64 },
303 	{ OPA_PATH_REC_FIELD(dgid),
304 	  .offset_words = 2,
305 	  .offset_bits  = 0,
306 	  .size_bits    = 128 },
307 	{ OPA_PATH_REC_FIELD(sgid),
308 	  .offset_words = 6,
309 	  .offset_bits  = 0,
310 	  .size_bits    = 128 },
311 	{ OPA_PATH_REC_FIELD(opa.dlid),
312 	  .offset_words = 10,
313 	  .offset_bits  = 0,
314 	  .size_bits    = 32 },
315 	{ OPA_PATH_REC_FIELD(opa.slid),
316 	  .offset_words = 11,
317 	  .offset_bits  = 0,
318 	  .size_bits    = 32 },
319 	{ OPA_PATH_REC_FIELD(opa.raw_traffic),
320 	  .offset_words = 12,
321 	  .offset_bits  = 0,
322 	  .size_bits    = 1 },
323 	{ RESERVED,
324 	  .offset_words = 12,
325 	  .offset_bits  = 1,
326 	  .size_bits    = 3 },
327 	{ OPA_PATH_REC_FIELD(flow_label),
328 	  .offset_words = 12,
329 	  .offset_bits  = 4,
330 	  .size_bits    = 20 },
331 	{ OPA_PATH_REC_FIELD(hop_limit),
332 	  .offset_words = 12,
333 	  .offset_bits  = 24,
334 	  .size_bits    = 8 },
335 	{ OPA_PATH_REC_FIELD(traffic_class),
336 	  .offset_words = 13,
337 	  .offset_bits  = 0,
338 	  .size_bits    = 8 },
339 	{ OPA_PATH_REC_FIELD(reversible),
340 	  .offset_words = 13,
341 	  .offset_bits  = 8,
342 	  .size_bits    = 1 },
343 	{ OPA_PATH_REC_FIELD(numb_path),
344 	  .offset_words = 13,
345 	  .offset_bits  = 9,
346 	  .size_bits    = 7 },
347 	{ OPA_PATH_REC_FIELD(pkey),
348 	  .offset_words = 13,
349 	  .offset_bits  = 16,
350 	  .size_bits    = 16 },
351 	{ OPA_PATH_REC_FIELD(opa.l2_8B),
352 	  .offset_words = 14,
353 	  .offset_bits  = 0,
354 	  .size_bits    = 1 },
355 	{ OPA_PATH_REC_FIELD(opa.l2_10B),
356 	  .offset_words = 14,
357 	  .offset_bits  = 1,
358 	  .size_bits    = 1 },
359 	{ OPA_PATH_REC_FIELD(opa.l2_9B),
360 	  .offset_words = 14,
361 	  .offset_bits  = 2,
362 	  .size_bits    = 1 },
363 	{ OPA_PATH_REC_FIELD(opa.l2_16B),
364 	  .offset_words = 14,
365 	  .offset_bits  = 3,
366 	  .size_bits    = 1 },
367 	{ RESERVED,
368 	  .offset_words = 14,
369 	  .offset_bits  = 4,
370 	  .size_bits    = 2 },
371 	{ OPA_PATH_REC_FIELD(opa.qos_type),
372 	  .offset_words = 14,
373 	  .offset_bits  = 6,
374 	  .size_bits    = 2 },
375 	{ OPA_PATH_REC_FIELD(opa.qos_priority),
376 	  .offset_words = 14,
377 	  .offset_bits  = 8,
378 	  .size_bits    = 8 },
379 	{ RESERVED,
380 	  .offset_words = 14,
381 	  .offset_bits  = 16,
382 	  .size_bits    = 3 },
383 	{ OPA_PATH_REC_FIELD(sl),
384 	  .offset_words = 14,
385 	  .offset_bits  = 19,
386 	  .size_bits    = 5 },
387 	{ RESERVED,
388 	  .offset_words = 14,
389 	  .offset_bits  = 24,
390 	  .size_bits    = 8 },
391 	{ OPA_PATH_REC_FIELD(mtu_selector),
392 	  .offset_words = 15,
393 	  .offset_bits  = 0,
394 	  .size_bits    = 2 },
395 	{ OPA_PATH_REC_FIELD(mtu),
396 	  .offset_words = 15,
397 	  .offset_bits  = 2,
398 	  .size_bits    = 6 },
399 	{ OPA_PATH_REC_FIELD(rate_selector),
400 	  .offset_words = 15,
401 	  .offset_bits  = 8,
402 	  .size_bits    = 2 },
403 	{ OPA_PATH_REC_FIELD(rate),
404 	  .offset_words = 15,
405 	  .offset_bits  = 10,
406 	  .size_bits    = 6 },
407 	{ OPA_PATH_REC_FIELD(packet_life_time_selector),
408 	  .offset_words = 15,
409 	  .offset_bits  = 16,
410 	  .size_bits    = 2 },
411 	{ OPA_PATH_REC_FIELD(packet_life_time),
412 	  .offset_words = 15,
413 	  .offset_bits  = 18,
414 	  .size_bits    = 6 },
415 	{ OPA_PATH_REC_FIELD(preference),
416 	  .offset_words = 15,
417 	  .offset_bits  = 24,
418 	  .size_bits    = 8 },
419 };
420 
421 #define MCMEMBER_REC_FIELD(field) \
422 	.struct_offset_bytes = offsetof(struct ib_sa_mcmember_rec, field),	\
423 	.struct_size_bytes   = sizeof ((struct ib_sa_mcmember_rec *) 0)->field,	\
424 	.field_name          = "sa_mcmember_rec:" #field
425 
426 static const struct ib_field mcmember_rec_table[] = {
427 	{ MCMEMBER_REC_FIELD(mgid),
428 	  .offset_words = 0,
429 	  .offset_bits  = 0,
430 	  .size_bits    = 128 },
431 	{ MCMEMBER_REC_FIELD(port_gid),
432 	  .offset_words = 4,
433 	  .offset_bits  = 0,
434 	  .size_bits    = 128 },
435 	{ MCMEMBER_REC_FIELD(qkey),
436 	  .offset_words = 8,
437 	  .offset_bits  = 0,
438 	  .size_bits    = 32 },
439 	{ MCMEMBER_REC_FIELD(mlid),
440 	  .offset_words = 9,
441 	  .offset_bits  = 0,
442 	  .size_bits    = 16 },
443 	{ MCMEMBER_REC_FIELD(mtu_selector),
444 	  .offset_words = 9,
445 	  .offset_bits  = 16,
446 	  .size_bits    = 2 },
447 	{ MCMEMBER_REC_FIELD(mtu),
448 	  .offset_words = 9,
449 	  .offset_bits  = 18,
450 	  .size_bits    = 6 },
451 	{ MCMEMBER_REC_FIELD(traffic_class),
452 	  .offset_words = 9,
453 	  .offset_bits  = 24,
454 	  .size_bits    = 8 },
455 	{ MCMEMBER_REC_FIELD(pkey),
456 	  .offset_words = 10,
457 	  .offset_bits  = 0,
458 	  .size_bits    = 16 },
459 	{ MCMEMBER_REC_FIELD(rate_selector),
460 	  .offset_words = 10,
461 	  .offset_bits  = 16,
462 	  .size_bits    = 2 },
463 	{ MCMEMBER_REC_FIELD(rate),
464 	  .offset_words = 10,
465 	  .offset_bits  = 18,
466 	  .size_bits    = 6 },
467 	{ MCMEMBER_REC_FIELD(packet_life_time_selector),
468 	  .offset_words = 10,
469 	  .offset_bits  = 24,
470 	  .size_bits    = 2 },
471 	{ MCMEMBER_REC_FIELD(packet_life_time),
472 	  .offset_words = 10,
473 	  .offset_bits  = 26,
474 	  .size_bits    = 6 },
475 	{ MCMEMBER_REC_FIELD(sl),
476 	  .offset_words = 11,
477 	  .offset_bits  = 0,
478 	  .size_bits    = 4 },
479 	{ MCMEMBER_REC_FIELD(flow_label),
480 	  .offset_words = 11,
481 	  .offset_bits  = 4,
482 	  .size_bits    = 20 },
483 	{ MCMEMBER_REC_FIELD(hop_limit),
484 	  .offset_words = 11,
485 	  .offset_bits  = 24,
486 	  .size_bits    = 8 },
487 	{ MCMEMBER_REC_FIELD(scope),
488 	  .offset_words = 12,
489 	  .offset_bits  = 0,
490 	  .size_bits    = 4 },
491 	{ MCMEMBER_REC_FIELD(join_state),
492 	  .offset_words = 12,
493 	  .offset_bits  = 4,
494 	  .size_bits    = 4 },
495 	{ MCMEMBER_REC_FIELD(proxy_join),
496 	  .offset_words = 12,
497 	  .offset_bits  = 8,
498 	  .size_bits    = 1 },
499 	{ RESERVED,
500 	  .offset_words = 12,
501 	  .offset_bits  = 9,
502 	  .size_bits    = 23 },
503 };
504 
505 #define SERVICE_REC_FIELD(field) \
506 	.struct_offset_bytes = offsetof(struct ib_sa_service_rec, field),	\
507 	.struct_size_bytes   = sizeof ((struct ib_sa_service_rec *) 0)->field,	\
508 	.field_name          = "sa_service_rec:" #field
509 
510 static const struct ib_field service_rec_table[] = {
511 	{ SERVICE_REC_FIELD(id),
512 	  .offset_words = 0,
513 	  .offset_bits  = 0,
514 	  .size_bits    = 64 },
515 	{ SERVICE_REC_FIELD(gid),
516 	  .offset_words = 2,
517 	  .offset_bits  = 0,
518 	  .size_bits    = 128 },
519 	{ SERVICE_REC_FIELD(pkey),
520 	  .offset_words = 6,
521 	  .offset_bits  = 0,
522 	  .size_bits    = 16 },
523 	{ SERVICE_REC_FIELD(lease),
524 	  .offset_words = 7,
525 	  .offset_bits  = 0,
526 	  .size_bits    = 32 },
527 	{ SERVICE_REC_FIELD(key),
528 	  .offset_words = 8,
529 	  .offset_bits  = 0,
530 	  .size_bits    = 128 },
531 	{ SERVICE_REC_FIELD(name),
532 	  .offset_words = 12,
533 	  .offset_bits  = 0,
534 	  .size_bits    = 64*8 },
535 	{ SERVICE_REC_FIELD(data8),
536 	  .offset_words = 28,
537 	  .offset_bits  = 0,
538 	  .size_bits    = 16*8 },
539 	{ SERVICE_REC_FIELD(data16),
540 	  .offset_words = 32,
541 	  .offset_bits  = 0,
542 	  .size_bits    = 8*16 },
543 	{ SERVICE_REC_FIELD(data32),
544 	  .offset_words = 36,
545 	  .offset_bits  = 0,
546 	  .size_bits    = 4*32 },
547 	{ SERVICE_REC_FIELD(data64),
548 	  .offset_words = 40,
549 	  .offset_bits  = 0,
550 	  .size_bits    = 2*64 },
551 };
552 
553 #define CLASSPORTINFO_REC_FIELD(field) \
554 	.struct_offset_bytes = offsetof(struct ib_class_port_info, field),	\
555 	.struct_size_bytes   = sizeof((struct ib_class_port_info *)0)->field,	\
556 	.field_name          = "ib_class_port_info:" #field
557 
558 static const struct ib_field ib_classport_info_rec_table[] = {
559 	{ CLASSPORTINFO_REC_FIELD(base_version),
560 	  .offset_words = 0,
561 	  .offset_bits  = 0,
562 	  .size_bits    = 8 },
563 	{ CLASSPORTINFO_REC_FIELD(class_version),
564 	  .offset_words = 0,
565 	  .offset_bits  = 8,
566 	  .size_bits    = 8 },
567 	{ CLASSPORTINFO_REC_FIELD(capability_mask),
568 	  .offset_words = 0,
569 	  .offset_bits  = 16,
570 	  .size_bits    = 16 },
571 	{ CLASSPORTINFO_REC_FIELD(cap_mask2_resp_time),
572 	  .offset_words = 1,
573 	  .offset_bits  = 0,
574 	  .size_bits    = 32 },
575 	{ CLASSPORTINFO_REC_FIELD(redirect_gid),
576 	  .offset_words = 2,
577 	  .offset_bits  = 0,
578 	  .size_bits    = 128 },
579 	{ CLASSPORTINFO_REC_FIELD(redirect_tcslfl),
580 	  .offset_words = 6,
581 	  .offset_bits  = 0,
582 	  .size_bits    = 32 },
583 	{ CLASSPORTINFO_REC_FIELD(redirect_lid),
584 	  .offset_words = 7,
585 	  .offset_bits  = 0,
586 	  .size_bits    = 16 },
587 	{ CLASSPORTINFO_REC_FIELD(redirect_pkey),
588 	  .offset_words = 7,
589 	  .offset_bits  = 16,
590 	  .size_bits    = 16 },
591 
592 	{ CLASSPORTINFO_REC_FIELD(redirect_qp),
593 	  .offset_words = 8,
594 	  .offset_bits  = 0,
595 	  .size_bits    = 32 },
596 	{ CLASSPORTINFO_REC_FIELD(redirect_qkey),
597 	  .offset_words = 9,
598 	  .offset_bits  = 0,
599 	  .size_bits    = 32 },
600 
601 	{ CLASSPORTINFO_REC_FIELD(trap_gid),
602 	  .offset_words = 10,
603 	  .offset_bits  = 0,
604 	  .size_bits    = 128 },
605 	{ CLASSPORTINFO_REC_FIELD(trap_tcslfl),
606 	  .offset_words = 14,
607 	  .offset_bits  = 0,
608 	  .size_bits    = 32 },
609 
610 	{ CLASSPORTINFO_REC_FIELD(trap_lid),
611 	  .offset_words = 15,
612 	  .offset_bits  = 0,
613 	  .size_bits    = 16 },
614 	{ CLASSPORTINFO_REC_FIELD(trap_pkey),
615 	  .offset_words = 15,
616 	  .offset_bits  = 16,
617 	  .size_bits    = 16 },
618 
619 	{ CLASSPORTINFO_REC_FIELD(trap_hlqp),
620 	  .offset_words = 16,
621 	  .offset_bits  = 0,
622 	  .size_bits    = 32 },
623 	{ CLASSPORTINFO_REC_FIELD(trap_qkey),
624 	  .offset_words = 17,
625 	  .offset_bits  = 0,
626 	  .size_bits    = 32 },
627 };
628 
629 #define OPA_CLASSPORTINFO_REC_FIELD(field) \
630 	.struct_offset_bytes =\
631 		offsetof(struct opa_class_port_info, field),	\
632 	.struct_size_bytes   = \
633 		sizeof((struct opa_class_port_info *)0)->field,	\
634 	.field_name          = "opa_class_port_info:" #field
635 
636 static const struct ib_field opa_classport_info_rec_table[] = {
637 	{ OPA_CLASSPORTINFO_REC_FIELD(base_version),
638 	  .offset_words = 0,
639 	  .offset_bits  = 0,
640 	  .size_bits    = 8 },
641 	{ OPA_CLASSPORTINFO_REC_FIELD(class_version),
642 	  .offset_words = 0,
643 	  .offset_bits  = 8,
644 	  .size_bits    = 8 },
645 	{ OPA_CLASSPORTINFO_REC_FIELD(cap_mask),
646 	  .offset_words = 0,
647 	  .offset_bits  = 16,
648 	  .size_bits    = 16 },
649 	{ OPA_CLASSPORTINFO_REC_FIELD(cap_mask2_resp_time),
650 	  .offset_words = 1,
651 	  .offset_bits  = 0,
652 	  .size_bits    = 32 },
653 	{ OPA_CLASSPORTINFO_REC_FIELD(redirect_gid),
654 	  .offset_words = 2,
655 	  .offset_bits  = 0,
656 	  .size_bits    = 128 },
657 	{ OPA_CLASSPORTINFO_REC_FIELD(redirect_tc_fl),
658 	  .offset_words = 6,
659 	  .offset_bits  = 0,
660 	  .size_bits    = 32 },
661 	{ OPA_CLASSPORTINFO_REC_FIELD(redirect_lid),
662 	  .offset_words = 7,
663 	  .offset_bits  = 0,
664 	  .size_bits    = 32 },
665 	{ OPA_CLASSPORTINFO_REC_FIELD(redirect_sl_qp),
666 	  .offset_words = 8,
667 	  .offset_bits  = 0,
668 	  .size_bits    = 32 },
669 	{ OPA_CLASSPORTINFO_REC_FIELD(redirect_qkey),
670 	  .offset_words = 9,
671 	  .offset_bits  = 0,
672 	  .size_bits    = 32 },
673 	{ OPA_CLASSPORTINFO_REC_FIELD(trap_gid),
674 	  .offset_words = 10,
675 	  .offset_bits  = 0,
676 	  .size_bits    = 128 },
677 	{ OPA_CLASSPORTINFO_REC_FIELD(trap_tc_fl),
678 	  .offset_words = 14,
679 	  .offset_bits  = 0,
680 	  .size_bits    = 32 },
681 	{ OPA_CLASSPORTINFO_REC_FIELD(trap_lid),
682 	  .offset_words = 15,
683 	  .offset_bits  = 0,
684 	  .size_bits    = 32 },
685 	{ OPA_CLASSPORTINFO_REC_FIELD(trap_hl_qp),
686 	  .offset_words = 16,
687 	  .offset_bits  = 0,
688 	  .size_bits    = 32 },
689 	{ OPA_CLASSPORTINFO_REC_FIELD(trap_qkey),
690 	  .offset_words = 17,
691 	  .offset_bits  = 0,
692 	  .size_bits    = 32 },
693 	{ OPA_CLASSPORTINFO_REC_FIELD(trap_pkey),
694 	  .offset_words = 18,
695 	  .offset_bits  = 0,
696 	  .size_bits    = 16 },
697 	{ OPA_CLASSPORTINFO_REC_FIELD(redirect_pkey),
698 	  .offset_words = 18,
699 	  .offset_bits  = 16,
700 	  .size_bits    = 16 },
701 	{ OPA_CLASSPORTINFO_REC_FIELD(trap_sl_rsvd),
702 	  .offset_words = 19,
703 	  .offset_bits  = 0,
704 	  .size_bits    = 8 },
705 	{ RESERVED,
706 	  .offset_words = 19,
707 	  .offset_bits  = 8,
708 	  .size_bits    = 24 },
709 };
710 
711 #define GUIDINFO_REC_FIELD(field) \
712 	.struct_offset_bytes = offsetof(struct ib_sa_guidinfo_rec, field),	\
713 	.struct_size_bytes   = sizeof((struct ib_sa_guidinfo_rec *) 0)->field,	\
714 	.field_name          = "sa_guidinfo_rec:" #field
715 
716 static const struct ib_field guidinfo_rec_table[] = {
717 	{ GUIDINFO_REC_FIELD(lid),
718 	  .offset_words = 0,
719 	  .offset_bits  = 0,
720 	  .size_bits    = 16 },
721 	{ GUIDINFO_REC_FIELD(block_num),
722 	  .offset_words = 0,
723 	  .offset_bits  = 16,
724 	  .size_bits    = 8 },
725 	{ GUIDINFO_REC_FIELD(res1),
726 	  .offset_words = 0,
727 	  .offset_bits  = 24,
728 	  .size_bits    = 8 },
729 	{ GUIDINFO_REC_FIELD(res2),
730 	  .offset_words = 1,
731 	  .offset_bits  = 0,
732 	  .size_bits    = 32 },
733 	{ GUIDINFO_REC_FIELD(guid_info_list),
734 	  .offset_words = 2,
735 	  .offset_bits  = 0,
736 	  .size_bits    = 512 },
737 };
738 
739 static inline void ib_sa_disable_local_svc(struct ib_sa_query *query)
740 {
741 	query->flags &= ~IB_SA_ENABLE_LOCAL_SERVICE;
742 }
743 
744 static inline int ib_sa_query_cancelled(struct ib_sa_query *query)
745 {
746 	return (query->flags & IB_SA_CANCEL);
747 }
748 
749 static void ib_nl_set_path_rec_attrs(struct sk_buff *skb,
750 				     struct ib_sa_query *query)
751 {
752 	struct sa_path_rec *sa_rec = query->mad_buf->context[1];
753 	struct ib_sa_mad *mad = query->mad_buf->mad;
754 	ib_sa_comp_mask comp_mask = mad->sa_hdr.comp_mask;
755 	u16 val16;
756 	u64 val64;
757 	struct rdma_ls_resolve_header *header;
758 
759 	query->mad_buf->context[1] = NULL;
760 
761 	/* Construct the family header first */
762 	header = skb_put(skb, NLMSG_ALIGN(sizeof(*header)));
763 	memcpy(header->device_name, query->port->agent->device->name,
764 	       LS_DEVICE_NAME_MAX);
765 	header->port_num = query->port->port_num;
766 
767 	if ((comp_mask & IB_SA_PATH_REC_REVERSIBLE) &&
768 	    sa_rec->reversible != 0)
769 		query->path_use = LS_RESOLVE_PATH_USE_GMP;
770 	else
771 		query->path_use = LS_RESOLVE_PATH_USE_UNIDIRECTIONAL;
772 	header->path_use = query->path_use;
773 
774 	/* Now build the attributes */
775 	if (comp_mask & IB_SA_PATH_REC_SERVICE_ID) {
776 		val64 = be64_to_cpu(sa_rec->service_id);
777 		nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_SERVICE_ID,
778 			sizeof(val64), &val64);
779 	}
780 	if (comp_mask & IB_SA_PATH_REC_DGID)
781 		nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_DGID,
782 			sizeof(sa_rec->dgid), &sa_rec->dgid);
783 	if (comp_mask & IB_SA_PATH_REC_SGID)
784 		nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_SGID,
785 			sizeof(sa_rec->sgid), &sa_rec->sgid);
786 	if (comp_mask & IB_SA_PATH_REC_TRAFFIC_CLASS)
787 		nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_TCLASS,
788 			sizeof(sa_rec->traffic_class), &sa_rec->traffic_class);
789 
790 	if (comp_mask & IB_SA_PATH_REC_PKEY) {
791 		val16 = be16_to_cpu(sa_rec->pkey);
792 		nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_PKEY,
793 			sizeof(val16), &val16);
794 	}
795 	if (comp_mask & IB_SA_PATH_REC_QOS_CLASS) {
796 		val16 = be16_to_cpu(sa_rec->qos_class);
797 		nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_QOS_CLASS,
798 			sizeof(val16), &val16);
799 	}
800 }
801 
802 static int ib_nl_get_path_rec_attrs_len(ib_sa_comp_mask comp_mask)
803 {
804 	int len = 0;
805 
806 	if (comp_mask & IB_SA_PATH_REC_SERVICE_ID)
807 		len += nla_total_size(sizeof(u64));
808 	if (comp_mask & IB_SA_PATH_REC_DGID)
809 		len += nla_total_size(sizeof(struct rdma_nla_ls_gid));
810 	if (comp_mask & IB_SA_PATH_REC_SGID)
811 		len += nla_total_size(sizeof(struct rdma_nla_ls_gid));
812 	if (comp_mask & IB_SA_PATH_REC_TRAFFIC_CLASS)
813 		len += nla_total_size(sizeof(u8));
814 	if (comp_mask & IB_SA_PATH_REC_PKEY)
815 		len += nla_total_size(sizeof(u16));
816 	if (comp_mask & IB_SA_PATH_REC_QOS_CLASS)
817 		len += nla_total_size(sizeof(u16));
818 
819 	/*
820 	 * Make sure that at least some of the required comp_mask bits are
821 	 * set.
822 	 */
823 	if (WARN_ON(len == 0))
824 		return len;
825 
826 	/* Add the family header */
827 	len += NLMSG_ALIGN(sizeof(struct rdma_ls_resolve_header));
828 
829 	return len;
830 }
831 
832 static int ib_nl_send_msg(struct ib_sa_query *query, gfp_t gfp_mask)
833 {
834 	struct sk_buff *skb = NULL;
835 	struct nlmsghdr *nlh;
836 	void *data;
837 	int ret = 0;
838 	struct ib_sa_mad *mad;
839 	int len;
840 
841 	mad = query->mad_buf->mad;
842 	len = ib_nl_get_path_rec_attrs_len(mad->sa_hdr.comp_mask);
843 	if (len <= 0)
844 		return -EMSGSIZE;
845 
846 	skb = nlmsg_new(len, gfp_mask);
847 	if (!skb)
848 		return -ENOMEM;
849 
850 	/* Put nlmsg header only for now */
851 	data = ibnl_put_msg(skb, &nlh, query->seq, 0, RDMA_NL_LS,
852 			    RDMA_NL_LS_OP_RESOLVE, NLM_F_REQUEST);
853 	if (!data) {
854 		nlmsg_free(skb);
855 		return -EMSGSIZE;
856 	}
857 
858 	/* Add attributes */
859 	ib_nl_set_path_rec_attrs(skb, query);
860 
861 	/* Repair the nlmsg header length */
862 	nlmsg_end(skb, nlh);
863 
864 	ret = ibnl_multicast(skb, nlh, RDMA_NL_GROUP_LS, gfp_mask);
865 	if (!ret)
866 		ret = len;
867 	else
868 		ret = 0;
869 
870 	return ret;
871 }
872 
873 static int ib_nl_make_request(struct ib_sa_query *query, gfp_t gfp_mask)
874 {
875 	unsigned long flags;
876 	unsigned long delay;
877 	int ret;
878 
879 	INIT_LIST_HEAD(&query->list);
880 	query->seq = (u32)atomic_inc_return(&ib_nl_sa_request_seq);
881 
882 	/* Put the request on the list first.*/
883 	spin_lock_irqsave(&ib_nl_request_lock, flags);
884 	delay = msecs_to_jiffies(sa_local_svc_timeout_ms);
885 	query->timeout = delay + jiffies;
886 	list_add_tail(&query->list, &ib_nl_request_list);
887 	/* Start the timeout if this is the only request */
888 	if (ib_nl_request_list.next == &query->list)
889 		queue_delayed_work(ib_nl_wq, &ib_nl_timed_work, delay);
890 	spin_unlock_irqrestore(&ib_nl_request_lock, flags);
891 
892 	ret = ib_nl_send_msg(query, gfp_mask);
893 	if (ret <= 0) {
894 		ret = -EIO;
895 		/* Remove the request */
896 		spin_lock_irqsave(&ib_nl_request_lock, flags);
897 		list_del(&query->list);
898 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
899 	} else {
900 		ret = 0;
901 	}
902 
903 	return ret;
904 }
905 
906 static int ib_nl_cancel_request(struct ib_sa_query *query)
907 {
908 	unsigned long flags;
909 	struct ib_sa_query *wait_query;
910 	int found = 0;
911 
912 	spin_lock_irqsave(&ib_nl_request_lock, flags);
913 	list_for_each_entry(wait_query, &ib_nl_request_list, list) {
914 		/* Let the timeout to take care of the callback */
915 		if (query == wait_query) {
916 			query->flags |= IB_SA_CANCEL;
917 			query->timeout = jiffies;
918 			list_move(&query->list, &ib_nl_request_list);
919 			found = 1;
920 			mod_delayed_work(ib_nl_wq, &ib_nl_timed_work, 1);
921 			break;
922 		}
923 	}
924 	spin_unlock_irqrestore(&ib_nl_request_lock, flags);
925 
926 	return found;
927 }
928 
929 static void send_handler(struct ib_mad_agent *agent,
930 			 struct ib_mad_send_wc *mad_send_wc);
931 
932 static void ib_nl_process_good_resolve_rsp(struct ib_sa_query *query,
933 					   const struct nlmsghdr *nlh)
934 {
935 	struct ib_mad_send_wc mad_send_wc;
936 	struct ib_sa_mad *mad = NULL;
937 	const struct nlattr *head, *curr;
938 	struct ib_path_rec_data  *rec;
939 	int len, rem;
940 	u32 mask = 0;
941 	int status = -EIO;
942 
943 	if (query->callback) {
944 		head = (const struct nlattr *) nlmsg_data(nlh);
945 		len = nlmsg_len(nlh);
946 		switch (query->path_use) {
947 		case LS_RESOLVE_PATH_USE_UNIDIRECTIONAL:
948 			mask = IB_PATH_PRIMARY | IB_PATH_OUTBOUND;
949 			break;
950 
951 		case LS_RESOLVE_PATH_USE_ALL:
952 		case LS_RESOLVE_PATH_USE_GMP:
953 		default:
954 			mask = IB_PATH_PRIMARY | IB_PATH_GMP |
955 				IB_PATH_BIDIRECTIONAL;
956 			break;
957 		}
958 		nla_for_each_attr(curr, head, len, rem) {
959 			if (curr->nla_type == LS_NLA_TYPE_PATH_RECORD) {
960 				rec = nla_data(curr);
961 				/*
962 				 * Get the first one. In the future, we may
963 				 * need to get up to 6 pathrecords.
964 				 */
965 				if ((rec->flags & mask) == mask) {
966 					mad = query->mad_buf->mad;
967 					mad->mad_hdr.method |=
968 						IB_MGMT_METHOD_RESP;
969 					memcpy(mad->data, rec->path_rec,
970 					       sizeof(rec->path_rec));
971 					status = 0;
972 					break;
973 				}
974 			}
975 		}
976 		query->callback(query, status, mad);
977 	}
978 
979 	mad_send_wc.send_buf = query->mad_buf;
980 	mad_send_wc.status = IB_WC_SUCCESS;
981 	send_handler(query->mad_buf->mad_agent, &mad_send_wc);
982 }
983 
984 static void ib_nl_request_timeout(struct work_struct *work)
985 {
986 	unsigned long flags;
987 	struct ib_sa_query *query;
988 	unsigned long delay;
989 	struct ib_mad_send_wc mad_send_wc;
990 	int ret;
991 
992 	spin_lock_irqsave(&ib_nl_request_lock, flags);
993 	while (!list_empty(&ib_nl_request_list)) {
994 		query = list_entry(ib_nl_request_list.next,
995 				   struct ib_sa_query, list);
996 
997 		if (time_after(query->timeout, jiffies)) {
998 			delay = query->timeout - jiffies;
999 			if ((long)delay <= 0)
1000 				delay = 1;
1001 			queue_delayed_work(ib_nl_wq, &ib_nl_timed_work, delay);
1002 			break;
1003 		}
1004 
1005 		list_del(&query->list);
1006 		ib_sa_disable_local_svc(query);
1007 		/* Hold the lock to protect against query cancellation */
1008 		if (ib_sa_query_cancelled(query))
1009 			ret = -1;
1010 		else
1011 			ret = ib_post_send_mad(query->mad_buf, NULL);
1012 		if (ret) {
1013 			mad_send_wc.send_buf = query->mad_buf;
1014 			mad_send_wc.status = IB_WC_WR_FLUSH_ERR;
1015 			spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1016 			send_handler(query->port->agent, &mad_send_wc);
1017 			spin_lock_irqsave(&ib_nl_request_lock, flags);
1018 		}
1019 	}
1020 	spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1021 }
1022 
1023 int ib_nl_handle_set_timeout(struct sk_buff *skb,
1024 			     struct netlink_callback *cb)
1025 {
1026 	const struct nlmsghdr *nlh = (struct nlmsghdr *)cb->nlh;
1027 	int timeout, delta, abs_delta;
1028 	const struct nlattr *attr;
1029 	unsigned long flags;
1030 	struct ib_sa_query *query;
1031 	long delay = 0;
1032 	struct nlattr *tb[LS_NLA_TYPE_MAX];
1033 	int ret;
1034 
1035 	if (!(nlh->nlmsg_flags & NLM_F_REQUEST) ||
1036 	    !(NETLINK_CB(skb).sk) ||
1037 	    !netlink_capable(skb, CAP_NET_ADMIN))
1038 		return -EPERM;
1039 
1040 	ret = nla_parse(tb, LS_NLA_TYPE_MAX - 1, nlmsg_data(nlh),
1041 			nlmsg_len(nlh), ib_nl_policy, NULL);
1042 	attr = (const struct nlattr *)tb[LS_NLA_TYPE_TIMEOUT];
1043 	if (ret || !attr)
1044 		goto settimeout_out;
1045 
1046 	timeout = *(int *) nla_data(attr);
1047 	if (timeout < IB_SA_LOCAL_SVC_TIMEOUT_MIN)
1048 		timeout = IB_SA_LOCAL_SVC_TIMEOUT_MIN;
1049 	if (timeout > IB_SA_LOCAL_SVC_TIMEOUT_MAX)
1050 		timeout = IB_SA_LOCAL_SVC_TIMEOUT_MAX;
1051 
1052 	delta = timeout - sa_local_svc_timeout_ms;
1053 	if (delta < 0)
1054 		abs_delta = -delta;
1055 	else
1056 		abs_delta = delta;
1057 
1058 	if (delta != 0) {
1059 		spin_lock_irqsave(&ib_nl_request_lock, flags);
1060 		sa_local_svc_timeout_ms = timeout;
1061 		list_for_each_entry(query, &ib_nl_request_list, list) {
1062 			if (delta < 0 && abs_delta > query->timeout)
1063 				query->timeout = 0;
1064 			else
1065 				query->timeout += delta;
1066 
1067 			/* Get the new delay from the first entry */
1068 			if (!delay) {
1069 				delay = query->timeout - jiffies;
1070 				if (delay <= 0)
1071 					delay = 1;
1072 			}
1073 		}
1074 		if (delay)
1075 			mod_delayed_work(ib_nl_wq, &ib_nl_timed_work,
1076 					 (unsigned long)delay);
1077 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1078 	}
1079 
1080 settimeout_out:
1081 	return skb->len;
1082 }
1083 
1084 static inline int ib_nl_is_good_resolve_resp(const struct nlmsghdr *nlh)
1085 {
1086 	struct nlattr *tb[LS_NLA_TYPE_MAX];
1087 	int ret;
1088 
1089 	if (nlh->nlmsg_flags & RDMA_NL_LS_F_ERR)
1090 		return 0;
1091 
1092 	ret = nla_parse(tb, LS_NLA_TYPE_MAX - 1, nlmsg_data(nlh),
1093 			nlmsg_len(nlh), ib_nl_policy, NULL);
1094 	if (ret)
1095 		return 0;
1096 
1097 	return 1;
1098 }
1099 
1100 int ib_nl_handle_resolve_resp(struct sk_buff *skb,
1101 			      struct netlink_callback *cb)
1102 {
1103 	const struct nlmsghdr *nlh = (struct nlmsghdr *)cb->nlh;
1104 	unsigned long flags;
1105 	struct ib_sa_query *query;
1106 	struct ib_mad_send_buf *send_buf;
1107 	struct ib_mad_send_wc mad_send_wc;
1108 	int found = 0;
1109 	int ret;
1110 
1111 	if ((nlh->nlmsg_flags & NLM_F_REQUEST) ||
1112 	    !(NETLINK_CB(skb).sk) ||
1113 	    !netlink_capable(skb, CAP_NET_ADMIN))
1114 		return -EPERM;
1115 
1116 	spin_lock_irqsave(&ib_nl_request_lock, flags);
1117 	list_for_each_entry(query, &ib_nl_request_list, list) {
1118 		/*
1119 		 * If the query is cancelled, let the timeout routine
1120 		 * take care of it.
1121 		 */
1122 		if (nlh->nlmsg_seq == query->seq) {
1123 			found = !ib_sa_query_cancelled(query);
1124 			if (found)
1125 				list_del(&query->list);
1126 			break;
1127 		}
1128 	}
1129 
1130 	if (!found) {
1131 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1132 		goto resp_out;
1133 	}
1134 
1135 	send_buf = query->mad_buf;
1136 
1137 	if (!ib_nl_is_good_resolve_resp(nlh)) {
1138 		/* if the result is a failure, send out the packet via IB */
1139 		ib_sa_disable_local_svc(query);
1140 		ret = ib_post_send_mad(query->mad_buf, NULL);
1141 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1142 		if (ret) {
1143 			mad_send_wc.send_buf = send_buf;
1144 			mad_send_wc.status = IB_WC_GENERAL_ERR;
1145 			send_handler(query->port->agent, &mad_send_wc);
1146 		}
1147 	} else {
1148 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1149 		ib_nl_process_good_resolve_rsp(query, nlh);
1150 	}
1151 
1152 resp_out:
1153 	return skb->len;
1154 }
1155 
1156 static void free_sm_ah(struct kref *kref)
1157 {
1158 	struct ib_sa_sm_ah *sm_ah = container_of(kref, struct ib_sa_sm_ah, ref);
1159 
1160 	rdma_destroy_ah(sm_ah->ah);
1161 	kfree(sm_ah);
1162 }
1163 
1164 void ib_sa_register_client(struct ib_sa_client *client)
1165 {
1166 	atomic_set(&client->users, 1);
1167 	init_completion(&client->comp);
1168 }
1169 EXPORT_SYMBOL(ib_sa_register_client);
1170 
1171 void ib_sa_unregister_client(struct ib_sa_client *client)
1172 {
1173 	ib_sa_client_put(client);
1174 	wait_for_completion(&client->comp);
1175 }
1176 EXPORT_SYMBOL(ib_sa_unregister_client);
1177 
1178 /**
1179  * ib_sa_cancel_query - try to cancel an SA query
1180  * @id:ID of query to cancel
1181  * @query:query pointer to cancel
1182  *
1183  * Try to cancel an SA query.  If the id and query don't match up or
1184  * the query has already completed, nothing is done.  Otherwise the
1185  * query is canceled and will complete with a status of -EINTR.
1186  */
1187 void ib_sa_cancel_query(int id, struct ib_sa_query *query)
1188 {
1189 	unsigned long flags;
1190 	struct ib_mad_agent *agent;
1191 	struct ib_mad_send_buf *mad_buf;
1192 
1193 	spin_lock_irqsave(&idr_lock, flags);
1194 	if (idr_find(&query_idr, id) != query) {
1195 		spin_unlock_irqrestore(&idr_lock, flags);
1196 		return;
1197 	}
1198 	agent = query->port->agent;
1199 	mad_buf = query->mad_buf;
1200 	spin_unlock_irqrestore(&idr_lock, flags);
1201 
1202 	/*
1203 	 * If the query is still on the netlink request list, schedule
1204 	 * it to be cancelled by the timeout routine. Otherwise, it has been
1205 	 * sent to the MAD layer and has to be cancelled from there.
1206 	 */
1207 	if (!ib_nl_cancel_request(query))
1208 		ib_cancel_mad(agent, mad_buf);
1209 }
1210 EXPORT_SYMBOL(ib_sa_cancel_query);
1211 
1212 static u8 get_src_path_mask(struct ib_device *device, u8 port_num)
1213 {
1214 	struct ib_sa_device *sa_dev;
1215 	struct ib_sa_port   *port;
1216 	unsigned long flags;
1217 	u8 src_path_mask;
1218 
1219 	sa_dev = ib_get_client_data(device, &sa_client);
1220 	if (!sa_dev)
1221 		return 0x7f;
1222 
1223 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1224 	spin_lock_irqsave(&port->ah_lock, flags);
1225 	src_path_mask = port->sm_ah ? port->sm_ah->src_path_mask : 0x7f;
1226 	spin_unlock_irqrestore(&port->ah_lock, flags);
1227 
1228 	return src_path_mask;
1229 }
1230 
1231 int ib_init_ah_from_path(struct ib_device *device, u8 port_num,
1232 			 struct sa_path_rec *rec,
1233 			 struct rdma_ah_attr *ah_attr)
1234 {
1235 	int ret;
1236 	u16 gid_index;
1237 	int use_roce;
1238 	struct net_device *ndev = NULL;
1239 
1240 	memset(ah_attr, 0, sizeof *ah_attr);
1241 	ah_attr->type = rdma_ah_find_type(device, port_num);
1242 
1243 	rdma_ah_set_dlid(ah_attr, be32_to_cpu(sa_path_get_dlid(rec)));
1244 	rdma_ah_set_sl(ah_attr, rec->sl);
1245 	rdma_ah_set_path_bits(ah_attr, be32_to_cpu(sa_path_get_slid(rec)) &
1246 			      get_src_path_mask(device, port_num));
1247 	rdma_ah_set_port_num(ah_attr, port_num);
1248 	rdma_ah_set_static_rate(ah_attr, rec->rate);
1249 	use_roce = rdma_cap_eth_ah(device, port_num);
1250 
1251 	if (use_roce) {
1252 		struct net_device *idev;
1253 		struct net_device *resolved_dev;
1254 		struct rdma_dev_addr dev_addr = {
1255 			.bound_dev_if = ((sa_path_get_ifindex(rec) >= 0) ?
1256 					 sa_path_get_ifindex(rec) : 0),
1257 			.net = sa_path_get_ndev(rec) ?
1258 				sa_path_get_ndev(rec) :
1259 				&init_net
1260 		};
1261 		union {
1262 			struct sockaddr     _sockaddr;
1263 			struct sockaddr_in  _sockaddr_in;
1264 			struct sockaddr_in6 _sockaddr_in6;
1265 		} sgid_addr, dgid_addr;
1266 
1267 		if (!device->get_netdev)
1268 			return -EOPNOTSUPP;
1269 
1270 		rdma_gid2ip(&sgid_addr._sockaddr, &rec->sgid);
1271 		rdma_gid2ip(&dgid_addr._sockaddr, &rec->dgid);
1272 
1273 		/* validate the route */
1274 		ret = rdma_resolve_ip_route(&sgid_addr._sockaddr,
1275 					    &dgid_addr._sockaddr, &dev_addr);
1276 		if (ret)
1277 			return ret;
1278 
1279 		if ((dev_addr.network == RDMA_NETWORK_IPV4 ||
1280 		     dev_addr.network == RDMA_NETWORK_IPV6) &&
1281 		    rec->rec_type != SA_PATH_REC_TYPE_ROCE_V2)
1282 			return -EINVAL;
1283 
1284 		idev = device->get_netdev(device, port_num);
1285 		if (!idev)
1286 			return -ENODEV;
1287 
1288 		resolved_dev = dev_get_by_index(dev_addr.net,
1289 						dev_addr.bound_dev_if);
1290 		if (resolved_dev->flags & IFF_LOOPBACK) {
1291 			dev_put(resolved_dev);
1292 			resolved_dev = idev;
1293 			dev_hold(resolved_dev);
1294 		}
1295 		ndev = ib_get_ndev_from_path(rec);
1296 		rcu_read_lock();
1297 		if ((ndev && ndev != resolved_dev) ||
1298 		    (resolved_dev != idev &&
1299 		     !rdma_is_upper_dev_rcu(idev, resolved_dev)))
1300 			ret = -EHOSTUNREACH;
1301 		rcu_read_unlock();
1302 		dev_put(idev);
1303 		dev_put(resolved_dev);
1304 		if (ret) {
1305 			if (ndev)
1306 				dev_put(ndev);
1307 			return ret;
1308 		}
1309 	}
1310 
1311 	if (rec->hop_limit > 0 || use_roce) {
1312 		enum ib_gid_type type = sa_conv_pathrec_to_gid_type(rec);
1313 
1314 		ret = ib_find_cached_gid_by_port(device, &rec->sgid, type,
1315 						 port_num, ndev, &gid_index);
1316 		if (ret) {
1317 			if (ndev)
1318 				dev_put(ndev);
1319 			return ret;
1320 		}
1321 
1322 		rdma_ah_set_grh(ah_attr, &rec->dgid,
1323 				be32_to_cpu(rec->flow_label),
1324 				gid_index, rec->hop_limit,
1325 				rec->traffic_class);
1326 		if (ndev)
1327 			dev_put(ndev);
1328 	}
1329 
1330 	if (use_roce) {
1331 		u8 *dmac = sa_path_get_dmac(rec);
1332 
1333 		if (!dmac)
1334 			return -EINVAL;
1335 		memcpy(ah_attr->roce.dmac, dmac, ETH_ALEN);
1336 	}
1337 
1338 	return 0;
1339 }
1340 EXPORT_SYMBOL(ib_init_ah_from_path);
1341 
1342 static int alloc_mad(struct ib_sa_query *query, gfp_t gfp_mask)
1343 {
1344 	unsigned long flags;
1345 
1346 	spin_lock_irqsave(&query->port->ah_lock, flags);
1347 	if (!query->port->sm_ah) {
1348 		spin_unlock_irqrestore(&query->port->ah_lock, flags);
1349 		return -EAGAIN;
1350 	}
1351 	kref_get(&query->port->sm_ah->ref);
1352 	query->sm_ah = query->port->sm_ah;
1353 	spin_unlock_irqrestore(&query->port->ah_lock, flags);
1354 
1355 	query->mad_buf = ib_create_send_mad(query->port->agent, 1,
1356 					    query->sm_ah->pkey_index,
1357 					    0, IB_MGMT_SA_HDR, IB_MGMT_SA_DATA,
1358 					    gfp_mask,
1359 					    ((query->flags & IB_SA_QUERY_OPA) ?
1360 					     OPA_MGMT_BASE_VERSION :
1361 					     IB_MGMT_BASE_VERSION));
1362 	if (IS_ERR(query->mad_buf)) {
1363 		kref_put(&query->sm_ah->ref, free_sm_ah);
1364 		return -ENOMEM;
1365 	}
1366 
1367 	query->mad_buf->ah = query->sm_ah->ah;
1368 
1369 	return 0;
1370 }
1371 
1372 static void free_mad(struct ib_sa_query *query)
1373 {
1374 	ib_free_send_mad(query->mad_buf);
1375 	kref_put(&query->sm_ah->ref, free_sm_ah);
1376 }
1377 
1378 static void init_mad(struct ib_sa_query *query, struct ib_mad_agent *agent)
1379 {
1380 	struct ib_sa_mad *mad = query->mad_buf->mad;
1381 	unsigned long flags;
1382 
1383 	memset(mad, 0, sizeof *mad);
1384 
1385 	if (query->flags & IB_SA_QUERY_OPA) {
1386 		mad->mad_hdr.base_version  = OPA_MGMT_BASE_VERSION;
1387 		mad->mad_hdr.class_version = OPA_SA_CLASS_VERSION;
1388 	} else {
1389 		mad->mad_hdr.base_version  = IB_MGMT_BASE_VERSION;
1390 		mad->mad_hdr.class_version = IB_SA_CLASS_VERSION;
1391 	}
1392 	mad->mad_hdr.mgmt_class    = IB_MGMT_CLASS_SUBN_ADM;
1393 	spin_lock_irqsave(&tid_lock, flags);
1394 	mad->mad_hdr.tid           =
1395 		cpu_to_be64(((u64) agent->hi_tid) << 32 | tid++);
1396 	spin_unlock_irqrestore(&tid_lock, flags);
1397 }
1398 
1399 static int send_mad(struct ib_sa_query *query, int timeout_ms, gfp_t gfp_mask)
1400 {
1401 	bool preload = gfpflags_allow_blocking(gfp_mask);
1402 	unsigned long flags;
1403 	int ret, id;
1404 
1405 	if (preload)
1406 		idr_preload(gfp_mask);
1407 	spin_lock_irqsave(&idr_lock, flags);
1408 
1409 	id = idr_alloc(&query_idr, query, 0, 0, GFP_NOWAIT);
1410 
1411 	spin_unlock_irqrestore(&idr_lock, flags);
1412 	if (preload)
1413 		idr_preload_end();
1414 	if (id < 0)
1415 		return id;
1416 
1417 	query->mad_buf->timeout_ms  = timeout_ms;
1418 	query->mad_buf->context[0] = query;
1419 	query->id = id;
1420 
1421 	if ((query->flags & IB_SA_ENABLE_LOCAL_SERVICE) &&
1422 	    (!(query->flags & IB_SA_QUERY_OPA))) {
1423 		if (!ibnl_chk_listeners(RDMA_NL_GROUP_LS)) {
1424 			if (!ib_nl_make_request(query, gfp_mask))
1425 				return id;
1426 		}
1427 		ib_sa_disable_local_svc(query);
1428 	}
1429 
1430 	ret = ib_post_send_mad(query->mad_buf, NULL);
1431 	if (ret) {
1432 		spin_lock_irqsave(&idr_lock, flags);
1433 		idr_remove(&query_idr, id);
1434 		spin_unlock_irqrestore(&idr_lock, flags);
1435 	}
1436 
1437 	/*
1438 	 * It's not safe to dereference query any more, because the
1439 	 * send may already have completed and freed the query in
1440 	 * another context.
1441 	 */
1442 	return ret ? ret : id;
1443 }
1444 
1445 void ib_sa_unpack_path(void *attribute, struct sa_path_rec *rec)
1446 {
1447 	ib_unpack(path_rec_table, ARRAY_SIZE(path_rec_table), attribute, rec);
1448 }
1449 EXPORT_SYMBOL(ib_sa_unpack_path);
1450 
1451 void ib_sa_pack_path(struct sa_path_rec *rec, void *attribute)
1452 {
1453 	ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table), rec, attribute);
1454 }
1455 EXPORT_SYMBOL(ib_sa_pack_path);
1456 
1457 static bool ib_sa_opa_pathrecord_support(struct ib_sa_client *client,
1458 					 struct ib_device *device,
1459 					 u8 port_num)
1460 {
1461 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1462 	struct ib_sa_port *port;
1463 	unsigned long flags;
1464 	bool ret = false;
1465 
1466 	if (!sa_dev)
1467 		return ret;
1468 
1469 	port = &sa_dev->port[port_num - sa_dev->start_port];
1470 	spin_lock_irqsave(&port->classport_lock, flags);
1471 	if (!port->classport_info.valid)
1472 		goto ret;
1473 
1474 	if (port->classport_info.data.type == RDMA_CLASS_PORT_INFO_OPA)
1475 		ret = opa_get_cpi_capmask2(&port->classport_info.data.opa) &
1476 			OPA_CLASS_PORT_INFO_PR_SUPPORT;
1477 ret:
1478 	spin_unlock_irqrestore(&port->classport_lock, flags);
1479 	return ret;
1480 }
1481 
1482 enum opa_pr_supported {
1483 	PR_NOT_SUPPORTED,
1484 	PR_OPA_SUPPORTED,
1485 	PR_IB_SUPPORTED
1486 };
1487 
1488 /**
1489  * Check if current PR query can be an OPA query.
1490  * Retuns PR_NOT_SUPPORTED if a path record query is not
1491  * possible, PR_OPA_SUPPORTED if an OPA path record query
1492  * is possible and PR_IB_SUPPORTED if an IB path record
1493  * query is possible.
1494  */
1495 static int opa_pr_query_possible(struct ib_sa_client *client,
1496 				 struct ib_device *device,
1497 				 u8 port_num,
1498 				 struct sa_path_rec *rec)
1499 {
1500 	struct ib_port_attr port_attr;
1501 
1502 	if (ib_query_port(device, port_num, &port_attr))
1503 		return PR_NOT_SUPPORTED;
1504 
1505 	if (ib_sa_opa_pathrecord_support(client, device, port_num))
1506 		return PR_OPA_SUPPORTED;
1507 
1508 	if (port_attr.lid >= be16_to_cpu(IB_MULTICAST_LID_BASE))
1509 		return PR_NOT_SUPPORTED;
1510 	else
1511 		return PR_IB_SUPPORTED;
1512 }
1513 
1514 static void ib_sa_path_rec_callback(struct ib_sa_query *sa_query,
1515 				    int status,
1516 				    struct ib_sa_mad *mad)
1517 {
1518 	struct ib_sa_path_query *query =
1519 		container_of(sa_query, struct ib_sa_path_query, sa_query);
1520 
1521 	if (mad) {
1522 		struct sa_path_rec rec;
1523 
1524 		if (sa_query->flags & IB_SA_QUERY_OPA) {
1525 			ib_unpack(opa_path_rec_table,
1526 				  ARRAY_SIZE(opa_path_rec_table),
1527 				  mad->data, &rec);
1528 			rec.rec_type = SA_PATH_REC_TYPE_OPA;
1529 			query->callback(status, &rec, query->context);
1530 		} else {
1531 			ib_unpack(path_rec_table,
1532 				  ARRAY_SIZE(path_rec_table),
1533 				  mad->data, &rec);
1534 			rec.rec_type = SA_PATH_REC_TYPE_IB;
1535 			sa_path_set_ndev(&rec, NULL);
1536 			sa_path_set_ifindex(&rec, 0);
1537 			sa_path_set_dmac_zero(&rec);
1538 
1539 			if (query->conv_pr) {
1540 				struct sa_path_rec opa;
1541 
1542 				memset(&opa, 0, sizeof(struct sa_path_rec));
1543 				sa_convert_path_ib_to_opa(&opa, &rec);
1544 				query->callback(status, &opa, query->context);
1545 			} else {
1546 				query->callback(status, &rec, query->context);
1547 			}
1548 		}
1549 	} else
1550 		query->callback(status, NULL, query->context);
1551 }
1552 
1553 static void ib_sa_path_rec_release(struct ib_sa_query *sa_query)
1554 {
1555 	struct ib_sa_path_query *query =
1556 		container_of(sa_query, struct ib_sa_path_query, sa_query);
1557 
1558 	kfree(query->conv_pr);
1559 	kfree(query);
1560 }
1561 
1562 /**
1563  * ib_sa_path_rec_get - Start a Path get query
1564  * @client:SA client
1565  * @device:device to send query on
1566  * @port_num: port number to send query on
1567  * @rec:Path Record to send in query
1568  * @comp_mask:component mask to send in query
1569  * @timeout_ms:time to wait for response
1570  * @gfp_mask:GFP mask to use for internal allocations
1571  * @callback:function called when query completes, times out or is
1572  * canceled
1573  * @context:opaque user context passed to callback
1574  * @sa_query:query context, used to cancel query
1575  *
1576  * Send a Path Record Get query to the SA to look up a path.  The
1577  * callback function will be called when the query completes (or
1578  * fails); status is 0 for a successful response, -EINTR if the query
1579  * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
1580  * occurred sending the query.  The resp parameter of the callback is
1581  * only valid if status is 0.
1582  *
1583  * If the return value of ib_sa_path_rec_get() is negative, it is an
1584  * error code.  Otherwise it is a query ID that can be used to cancel
1585  * the query.
1586  */
1587 int ib_sa_path_rec_get(struct ib_sa_client *client,
1588 		       struct ib_device *device, u8 port_num,
1589 		       struct sa_path_rec *rec,
1590 		       ib_sa_comp_mask comp_mask,
1591 		       int timeout_ms, gfp_t gfp_mask,
1592 		       void (*callback)(int status,
1593 					struct sa_path_rec *resp,
1594 					void *context),
1595 		       void *context,
1596 		       struct ib_sa_query **sa_query)
1597 {
1598 	struct ib_sa_path_query *query;
1599 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1600 	struct ib_sa_port   *port;
1601 	struct ib_mad_agent *agent;
1602 	struct ib_sa_mad *mad;
1603 	enum opa_pr_supported status;
1604 	int ret;
1605 
1606 	if (!sa_dev)
1607 		return -ENODEV;
1608 
1609 	if ((rec->rec_type != SA_PATH_REC_TYPE_IB) &&
1610 	    (rec->rec_type != SA_PATH_REC_TYPE_OPA))
1611 		return -EINVAL;
1612 
1613 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1614 	agent = port->agent;
1615 
1616 	query = kzalloc(sizeof(*query), gfp_mask);
1617 	if (!query)
1618 		return -ENOMEM;
1619 
1620 	query->sa_query.port     = port;
1621 	if (rec->rec_type == SA_PATH_REC_TYPE_OPA) {
1622 		status = opa_pr_query_possible(client, device, port_num, rec);
1623 		if (status == PR_NOT_SUPPORTED) {
1624 			ret = -EINVAL;
1625 			goto err1;
1626 		} else if (status == PR_OPA_SUPPORTED) {
1627 			query->sa_query.flags |= IB_SA_QUERY_OPA;
1628 		} else {
1629 			query->conv_pr =
1630 				kmalloc(sizeof(*query->conv_pr), gfp_mask);
1631 			if (!query->conv_pr) {
1632 				ret = -ENOMEM;
1633 				goto err1;
1634 			}
1635 		}
1636 	}
1637 
1638 	ret = alloc_mad(&query->sa_query, gfp_mask);
1639 	if (ret)
1640 		goto err2;
1641 
1642 	ib_sa_client_get(client);
1643 	query->sa_query.client = client;
1644 	query->callback        = callback;
1645 	query->context         = context;
1646 
1647 	mad = query->sa_query.mad_buf->mad;
1648 	init_mad(&query->sa_query, agent);
1649 
1650 	query->sa_query.callback = callback ? ib_sa_path_rec_callback : NULL;
1651 	query->sa_query.release  = ib_sa_path_rec_release;
1652 	mad->mad_hdr.method	 = IB_MGMT_METHOD_GET;
1653 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_PATH_REC);
1654 	mad->sa_hdr.comp_mask	 = comp_mask;
1655 
1656 	if (query->sa_query.flags & IB_SA_QUERY_OPA) {
1657 		ib_pack(opa_path_rec_table, ARRAY_SIZE(opa_path_rec_table),
1658 			rec, mad->data);
1659 	} else if (query->conv_pr) {
1660 		sa_convert_path_opa_to_ib(query->conv_pr, rec);
1661 		ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table),
1662 			query->conv_pr, mad->data);
1663 	} else {
1664 		ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table),
1665 			rec, mad->data);
1666 	}
1667 
1668 	*sa_query = &query->sa_query;
1669 
1670 	query->sa_query.flags |= IB_SA_ENABLE_LOCAL_SERVICE;
1671 	query->sa_query.mad_buf->context[1] = (query->conv_pr) ?
1672 						query->conv_pr : rec;
1673 
1674 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1675 	if (ret < 0)
1676 		goto err3;
1677 
1678 	return ret;
1679 
1680 err3:
1681 	*sa_query = NULL;
1682 	ib_sa_client_put(query->sa_query.client);
1683 	free_mad(&query->sa_query);
1684 err2:
1685 	kfree(query->conv_pr);
1686 err1:
1687 	kfree(query);
1688 	return ret;
1689 }
1690 EXPORT_SYMBOL(ib_sa_path_rec_get);
1691 
1692 static void ib_sa_service_rec_callback(struct ib_sa_query *sa_query,
1693 				    int status,
1694 				    struct ib_sa_mad *mad)
1695 {
1696 	struct ib_sa_service_query *query =
1697 		container_of(sa_query, struct ib_sa_service_query, sa_query);
1698 
1699 	if (mad) {
1700 		struct ib_sa_service_rec rec;
1701 
1702 		ib_unpack(service_rec_table, ARRAY_SIZE(service_rec_table),
1703 			  mad->data, &rec);
1704 		query->callback(status, &rec, query->context);
1705 	} else
1706 		query->callback(status, NULL, query->context);
1707 }
1708 
1709 static void ib_sa_service_rec_release(struct ib_sa_query *sa_query)
1710 {
1711 	kfree(container_of(sa_query, struct ib_sa_service_query, sa_query));
1712 }
1713 
1714 /**
1715  * ib_sa_service_rec_query - Start Service Record operation
1716  * @client:SA client
1717  * @device:device to send request on
1718  * @port_num: port number to send request on
1719  * @method:SA method - should be get, set, or delete
1720  * @rec:Service Record to send in request
1721  * @comp_mask:component mask to send in request
1722  * @timeout_ms:time to wait for response
1723  * @gfp_mask:GFP mask to use for internal allocations
1724  * @callback:function called when request completes, times out or is
1725  * canceled
1726  * @context:opaque user context passed to callback
1727  * @sa_query:request context, used to cancel request
1728  *
1729  * Send a Service Record set/get/delete to the SA to register,
1730  * unregister or query a service record.
1731  * The callback function will be called when the request completes (or
1732  * fails); status is 0 for a successful response, -EINTR if the query
1733  * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
1734  * occurred sending the query.  The resp parameter of the callback is
1735  * only valid if status is 0.
1736  *
1737  * If the return value of ib_sa_service_rec_query() is negative, it is an
1738  * error code.  Otherwise it is a request ID that can be used to cancel
1739  * the query.
1740  */
1741 int ib_sa_service_rec_query(struct ib_sa_client *client,
1742 			    struct ib_device *device, u8 port_num, u8 method,
1743 			    struct ib_sa_service_rec *rec,
1744 			    ib_sa_comp_mask comp_mask,
1745 			    int timeout_ms, gfp_t gfp_mask,
1746 			    void (*callback)(int status,
1747 					     struct ib_sa_service_rec *resp,
1748 					     void *context),
1749 			    void *context,
1750 			    struct ib_sa_query **sa_query)
1751 {
1752 	struct ib_sa_service_query *query;
1753 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1754 	struct ib_sa_port   *port;
1755 	struct ib_mad_agent *agent;
1756 	struct ib_sa_mad *mad;
1757 	int ret;
1758 
1759 	if (!sa_dev)
1760 		return -ENODEV;
1761 
1762 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1763 	agent = port->agent;
1764 
1765 	if (method != IB_MGMT_METHOD_GET &&
1766 	    method != IB_MGMT_METHOD_SET &&
1767 	    method != IB_SA_METHOD_DELETE)
1768 		return -EINVAL;
1769 
1770 	query = kzalloc(sizeof(*query), gfp_mask);
1771 	if (!query)
1772 		return -ENOMEM;
1773 
1774 	query->sa_query.port     = port;
1775 	ret = alloc_mad(&query->sa_query, gfp_mask);
1776 	if (ret)
1777 		goto err1;
1778 
1779 	ib_sa_client_get(client);
1780 	query->sa_query.client = client;
1781 	query->callback        = callback;
1782 	query->context         = context;
1783 
1784 	mad = query->sa_query.mad_buf->mad;
1785 	init_mad(&query->sa_query, agent);
1786 
1787 	query->sa_query.callback = callback ? ib_sa_service_rec_callback : NULL;
1788 	query->sa_query.release  = ib_sa_service_rec_release;
1789 	mad->mad_hdr.method	 = method;
1790 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_SERVICE_REC);
1791 	mad->sa_hdr.comp_mask	 = comp_mask;
1792 
1793 	ib_pack(service_rec_table, ARRAY_SIZE(service_rec_table),
1794 		rec, mad->data);
1795 
1796 	*sa_query = &query->sa_query;
1797 
1798 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1799 	if (ret < 0)
1800 		goto err2;
1801 
1802 	return ret;
1803 
1804 err2:
1805 	*sa_query = NULL;
1806 	ib_sa_client_put(query->sa_query.client);
1807 	free_mad(&query->sa_query);
1808 
1809 err1:
1810 	kfree(query);
1811 	return ret;
1812 }
1813 EXPORT_SYMBOL(ib_sa_service_rec_query);
1814 
1815 static void ib_sa_mcmember_rec_callback(struct ib_sa_query *sa_query,
1816 					int status,
1817 					struct ib_sa_mad *mad)
1818 {
1819 	struct ib_sa_mcmember_query *query =
1820 		container_of(sa_query, struct ib_sa_mcmember_query, sa_query);
1821 
1822 	if (mad) {
1823 		struct ib_sa_mcmember_rec rec;
1824 
1825 		ib_unpack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
1826 			  mad->data, &rec);
1827 		query->callback(status, &rec, query->context);
1828 	} else
1829 		query->callback(status, NULL, query->context);
1830 }
1831 
1832 static void ib_sa_mcmember_rec_release(struct ib_sa_query *sa_query)
1833 {
1834 	kfree(container_of(sa_query, struct ib_sa_mcmember_query, sa_query));
1835 }
1836 
1837 int ib_sa_mcmember_rec_query(struct ib_sa_client *client,
1838 			     struct ib_device *device, u8 port_num,
1839 			     u8 method,
1840 			     struct ib_sa_mcmember_rec *rec,
1841 			     ib_sa_comp_mask comp_mask,
1842 			     int timeout_ms, gfp_t gfp_mask,
1843 			     void (*callback)(int status,
1844 					      struct ib_sa_mcmember_rec *resp,
1845 					      void *context),
1846 			     void *context,
1847 			     struct ib_sa_query **sa_query)
1848 {
1849 	struct ib_sa_mcmember_query *query;
1850 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1851 	struct ib_sa_port   *port;
1852 	struct ib_mad_agent *agent;
1853 	struct ib_sa_mad *mad;
1854 	int ret;
1855 
1856 	if (!sa_dev)
1857 		return -ENODEV;
1858 
1859 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1860 	agent = port->agent;
1861 
1862 	query = kzalloc(sizeof(*query), gfp_mask);
1863 	if (!query)
1864 		return -ENOMEM;
1865 
1866 	query->sa_query.port     = port;
1867 	ret = alloc_mad(&query->sa_query, gfp_mask);
1868 	if (ret)
1869 		goto err1;
1870 
1871 	ib_sa_client_get(client);
1872 	query->sa_query.client = client;
1873 	query->callback        = callback;
1874 	query->context         = context;
1875 
1876 	mad = query->sa_query.mad_buf->mad;
1877 	init_mad(&query->sa_query, agent);
1878 
1879 	query->sa_query.callback = callback ? ib_sa_mcmember_rec_callback : NULL;
1880 	query->sa_query.release  = ib_sa_mcmember_rec_release;
1881 	mad->mad_hdr.method	 = method;
1882 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_MC_MEMBER_REC);
1883 	mad->sa_hdr.comp_mask	 = comp_mask;
1884 
1885 	ib_pack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
1886 		rec, mad->data);
1887 
1888 	*sa_query = &query->sa_query;
1889 
1890 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1891 	if (ret < 0)
1892 		goto err2;
1893 
1894 	return ret;
1895 
1896 err2:
1897 	*sa_query = NULL;
1898 	ib_sa_client_put(query->sa_query.client);
1899 	free_mad(&query->sa_query);
1900 
1901 err1:
1902 	kfree(query);
1903 	return ret;
1904 }
1905 
1906 /* Support GuidInfoRecord */
1907 static void ib_sa_guidinfo_rec_callback(struct ib_sa_query *sa_query,
1908 					int status,
1909 					struct ib_sa_mad *mad)
1910 {
1911 	struct ib_sa_guidinfo_query *query =
1912 		container_of(sa_query, struct ib_sa_guidinfo_query, sa_query);
1913 
1914 	if (mad) {
1915 		struct ib_sa_guidinfo_rec rec;
1916 
1917 		ib_unpack(guidinfo_rec_table, ARRAY_SIZE(guidinfo_rec_table),
1918 			  mad->data, &rec);
1919 		query->callback(status, &rec, query->context);
1920 	} else
1921 		query->callback(status, NULL, query->context);
1922 }
1923 
1924 static void ib_sa_guidinfo_rec_release(struct ib_sa_query *sa_query)
1925 {
1926 	kfree(container_of(sa_query, struct ib_sa_guidinfo_query, sa_query));
1927 }
1928 
1929 int ib_sa_guid_info_rec_query(struct ib_sa_client *client,
1930 			      struct ib_device *device, u8 port_num,
1931 			      struct ib_sa_guidinfo_rec *rec,
1932 			      ib_sa_comp_mask comp_mask, u8 method,
1933 			      int timeout_ms, gfp_t gfp_mask,
1934 			      void (*callback)(int status,
1935 					       struct ib_sa_guidinfo_rec *resp,
1936 					       void *context),
1937 			      void *context,
1938 			      struct ib_sa_query **sa_query)
1939 {
1940 	struct ib_sa_guidinfo_query *query;
1941 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1942 	struct ib_sa_port *port;
1943 	struct ib_mad_agent *agent;
1944 	struct ib_sa_mad *mad;
1945 	int ret;
1946 
1947 	if (!sa_dev)
1948 		return -ENODEV;
1949 
1950 	if (method != IB_MGMT_METHOD_GET &&
1951 	    method != IB_MGMT_METHOD_SET &&
1952 	    method != IB_SA_METHOD_DELETE) {
1953 		return -EINVAL;
1954 	}
1955 
1956 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1957 	agent = port->agent;
1958 
1959 	query = kzalloc(sizeof(*query), gfp_mask);
1960 	if (!query)
1961 		return -ENOMEM;
1962 
1963 	query->sa_query.port = port;
1964 	ret = alloc_mad(&query->sa_query, gfp_mask);
1965 	if (ret)
1966 		goto err1;
1967 
1968 	ib_sa_client_get(client);
1969 	query->sa_query.client = client;
1970 	query->callback        = callback;
1971 	query->context         = context;
1972 
1973 	mad = query->sa_query.mad_buf->mad;
1974 	init_mad(&query->sa_query, agent);
1975 
1976 	query->sa_query.callback = callback ? ib_sa_guidinfo_rec_callback : NULL;
1977 	query->sa_query.release  = ib_sa_guidinfo_rec_release;
1978 
1979 	mad->mad_hdr.method	 = method;
1980 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_GUID_INFO_REC);
1981 	mad->sa_hdr.comp_mask	 = comp_mask;
1982 
1983 	ib_pack(guidinfo_rec_table, ARRAY_SIZE(guidinfo_rec_table), rec,
1984 		mad->data);
1985 
1986 	*sa_query = &query->sa_query;
1987 
1988 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1989 	if (ret < 0)
1990 		goto err2;
1991 
1992 	return ret;
1993 
1994 err2:
1995 	*sa_query = NULL;
1996 	ib_sa_client_put(query->sa_query.client);
1997 	free_mad(&query->sa_query);
1998 
1999 err1:
2000 	kfree(query);
2001 	return ret;
2002 }
2003 EXPORT_SYMBOL(ib_sa_guid_info_rec_query);
2004 
2005 bool ib_sa_sendonly_fullmem_support(struct ib_sa_client *client,
2006 				    struct ib_device *device,
2007 				    u8 port_num)
2008 {
2009 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
2010 	struct ib_sa_port *port;
2011 	bool ret = false;
2012 	unsigned long flags;
2013 
2014 	if (!sa_dev)
2015 		return ret;
2016 
2017 	port  = &sa_dev->port[port_num - sa_dev->start_port];
2018 
2019 	spin_lock_irqsave(&port->classport_lock, flags);
2020 	if ((port->classport_info.valid) &&
2021 	    (port->classport_info.data.type == RDMA_CLASS_PORT_INFO_IB))
2022 		ret = ib_get_cpi_capmask2(&port->classport_info.data.ib)
2023 			& IB_SA_CAP_MASK2_SENDONLY_FULL_MEM_SUPPORT;
2024 	spin_unlock_irqrestore(&port->classport_lock, flags);
2025 	return ret;
2026 }
2027 EXPORT_SYMBOL(ib_sa_sendonly_fullmem_support);
2028 
2029 struct ib_classport_info_context {
2030 	struct completion	done;
2031 	struct ib_sa_query	*sa_query;
2032 };
2033 
2034 static void ib_classportinfo_cb(void *context)
2035 {
2036 	struct ib_classport_info_context *cb_ctx = context;
2037 
2038 	complete(&cb_ctx->done);
2039 }
2040 
2041 static void ib_sa_classport_info_rec_callback(struct ib_sa_query *sa_query,
2042 					      int status,
2043 					      struct ib_sa_mad *mad)
2044 {
2045 	unsigned long flags;
2046 	struct ib_sa_classport_info_query *query =
2047 		container_of(sa_query, struct ib_sa_classport_info_query, sa_query);
2048 	struct ib_sa_classport_cache *info = &sa_query->port->classport_info;
2049 
2050 	if (mad) {
2051 		if (sa_query->flags & IB_SA_QUERY_OPA) {
2052 			struct opa_class_port_info rec;
2053 
2054 			ib_unpack(opa_classport_info_rec_table,
2055 				  ARRAY_SIZE(opa_classport_info_rec_table),
2056 				  mad->data, &rec);
2057 
2058 			spin_lock_irqsave(&sa_query->port->classport_lock,
2059 					  flags);
2060 			if (!status && !info->valid) {
2061 				memcpy(&info->data.opa, &rec,
2062 				       sizeof(info->data.opa));
2063 
2064 				info->valid = true;
2065 				info->data.type = RDMA_CLASS_PORT_INFO_OPA;
2066 			}
2067 			spin_unlock_irqrestore(&sa_query->port->classport_lock,
2068 					       flags);
2069 
2070 		} else {
2071 			struct ib_class_port_info rec;
2072 
2073 			ib_unpack(ib_classport_info_rec_table,
2074 				  ARRAY_SIZE(ib_classport_info_rec_table),
2075 				  mad->data, &rec);
2076 
2077 			spin_lock_irqsave(&sa_query->port->classport_lock,
2078 					  flags);
2079 			if (!status && !info->valid) {
2080 				memcpy(&info->data.ib, &rec,
2081 				       sizeof(info->data.ib));
2082 
2083 				info->valid = true;
2084 				info->data.type = RDMA_CLASS_PORT_INFO_IB;
2085 			}
2086 			spin_unlock_irqrestore(&sa_query->port->classport_lock,
2087 					       flags);
2088 		}
2089 	}
2090 	query->callback(query->context);
2091 }
2092 
2093 static void ib_sa_classport_info_rec_release(struct ib_sa_query *sa_query)
2094 {
2095 	kfree(container_of(sa_query, struct ib_sa_classport_info_query,
2096 			   sa_query));
2097 }
2098 
2099 static int ib_sa_classport_info_rec_query(struct ib_sa_port *port,
2100 					  int timeout_ms,
2101 					  void (*callback)(void *context),
2102 					  void *context,
2103 					  struct ib_sa_query **sa_query)
2104 {
2105 	struct ib_mad_agent *agent;
2106 	struct ib_sa_classport_info_query *query;
2107 	struct ib_sa_mad *mad;
2108 	gfp_t gfp_mask = GFP_KERNEL;
2109 	int ret;
2110 
2111 	agent = port->agent;
2112 
2113 	query = kzalloc(sizeof(*query), gfp_mask);
2114 	if (!query)
2115 		return -ENOMEM;
2116 
2117 	query->sa_query.port = port;
2118 	query->sa_query.flags |= rdma_cap_opa_ah(port->agent->device,
2119 						 port->port_num) ?
2120 				 IB_SA_QUERY_OPA : 0;
2121 	ret = alloc_mad(&query->sa_query, gfp_mask);
2122 	if (ret)
2123 		goto err_free;
2124 
2125 	query->callback = callback;
2126 	query->context = context;
2127 
2128 	mad = query->sa_query.mad_buf->mad;
2129 	init_mad(&query->sa_query, agent);
2130 
2131 	query->sa_query.callback = ib_sa_classport_info_rec_callback;
2132 	query->sa_query.release  = ib_sa_classport_info_rec_release;
2133 	mad->mad_hdr.method	 = IB_MGMT_METHOD_GET;
2134 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_CLASS_PORTINFO);
2135 	mad->sa_hdr.comp_mask	 = 0;
2136 	*sa_query = &query->sa_query;
2137 
2138 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
2139 	if (ret < 0)
2140 		goto err_free_mad;
2141 
2142 	return ret;
2143 
2144 err_free_mad:
2145 	*sa_query = NULL;
2146 	free_mad(&query->sa_query);
2147 
2148 err_free:
2149 	kfree(query);
2150 	return ret;
2151 }
2152 
2153 static void update_ib_cpi(struct work_struct *work)
2154 {
2155 	struct ib_sa_port *port =
2156 		container_of(work, struct ib_sa_port, ib_cpi_work.work);
2157 	struct ib_classport_info_context *cb_context;
2158 	unsigned long flags;
2159 	int ret;
2160 
2161 	/* If the classport info is valid, nothing
2162 	 * to do here.
2163 	 */
2164 	spin_lock_irqsave(&port->classport_lock, flags);
2165 	if (port->classport_info.valid) {
2166 		spin_unlock_irqrestore(&port->classport_lock, flags);
2167 		return;
2168 	}
2169 	spin_unlock_irqrestore(&port->classport_lock, flags);
2170 
2171 	cb_context = kmalloc(sizeof(*cb_context), GFP_KERNEL);
2172 	if (!cb_context)
2173 		goto err_nomem;
2174 
2175 	init_completion(&cb_context->done);
2176 
2177 	ret = ib_sa_classport_info_rec_query(port, 3000,
2178 					     ib_classportinfo_cb, cb_context,
2179 					     &cb_context->sa_query);
2180 	if (ret < 0)
2181 		goto free_cb_err;
2182 	wait_for_completion(&cb_context->done);
2183 free_cb_err:
2184 	kfree(cb_context);
2185 	spin_lock_irqsave(&port->classport_lock, flags);
2186 
2187 	/* If the classport info is still not valid, the query should have
2188 	 * failed for some reason. Retry issuing the query
2189 	 */
2190 	if (!port->classport_info.valid) {
2191 		port->classport_info.retry_cnt++;
2192 		if (port->classport_info.retry_cnt <=
2193 		    IB_SA_CPI_MAX_RETRY_CNT) {
2194 			unsigned long delay =
2195 				msecs_to_jiffies(IB_SA_CPI_RETRY_WAIT);
2196 
2197 			queue_delayed_work(ib_wq, &port->ib_cpi_work, delay);
2198 		}
2199 	}
2200 	spin_unlock_irqrestore(&port->classport_lock, flags);
2201 
2202 err_nomem:
2203 	return;
2204 }
2205 
2206 static void send_handler(struct ib_mad_agent *agent,
2207 			 struct ib_mad_send_wc *mad_send_wc)
2208 {
2209 	struct ib_sa_query *query = mad_send_wc->send_buf->context[0];
2210 	unsigned long flags;
2211 
2212 	if (query->callback)
2213 		switch (mad_send_wc->status) {
2214 		case IB_WC_SUCCESS:
2215 			/* No callback -- already got recv */
2216 			break;
2217 		case IB_WC_RESP_TIMEOUT_ERR:
2218 			query->callback(query, -ETIMEDOUT, NULL);
2219 			break;
2220 		case IB_WC_WR_FLUSH_ERR:
2221 			query->callback(query, -EINTR, NULL);
2222 			break;
2223 		default:
2224 			query->callback(query, -EIO, NULL);
2225 			break;
2226 		}
2227 
2228 	spin_lock_irqsave(&idr_lock, flags);
2229 	idr_remove(&query_idr, query->id);
2230 	spin_unlock_irqrestore(&idr_lock, flags);
2231 
2232 	free_mad(query);
2233 	if (query->client)
2234 		ib_sa_client_put(query->client);
2235 	query->release(query);
2236 }
2237 
2238 static void recv_handler(struct ib_mad_agent *mad_agent,
2239 			 struct ib_mad_send_buf *send_buf,
2240 			 struct ib_mad_recv_wc *mad_recv_wc)
2241 {
2242 	struct ib_sa_query *query;
2243 
2244 	if (!send_buf)
2245 		return;
2246 
2247 	query = send_buf->context[0];
2248 	if (query->callback) {
2249 		if (mad_recv_wc->wc->status == IB_WC_SUCCESS)
2250 			query->callback(query,
2251 					mad_recv_wc->recv_buf.mad->mad_hdr.status ?
2252 					-EINVAL : 0,
2253 					(struct ib_sa_mad *) mad_recv_wc->recv_buf.mad);
2254 		else
2255 			query->callback(query, -EIO, NULL);
2256 	}
2257 
2258 	ib_free_recv_mad(mad_recv_wc);
2259 }
2260 
2261 static void update_sm_ah(struct work_struct *work)
2262 {
2263 	struct ib_sa_port *port =
2264 		container_of(work, struct ib_sa_port, update_task);
2265 	struct ib_sa_sm_ah *new_ah;
2266 	struct ib_port_attr port_attr;
2267 	struct rdma_ah_attr   ah_attr;
2268 
2269 	if (ib_query_port(port->agent->device, port->port_num, &port_attr)) {
2270 		pr_warn("Couldn't query port\n");
2271 		return;
2272 	}
2273 
2274 	new_ah = kmalloc(sizeof(*new_ah), GFP_KERNEL);
2275 	if (!new_ah)
2276 		return;
2277 
2278 	kref_init(&new_ah->ref);
2279 	new_ah->src_path_mask = (1 << port_attr.lmc) - 1;
2280 
2281 	new_ah->pkey_index = 0;
2282 	if (ib_find_pkey(port->agent->device, port->port_num,
2283 			 IB_DEFAULT_PKEY_FULL, &new_ah->pkey_index))
2284 		pr_err("Couldn't find index for default PKey\n");
2285 
2286 	memset(&ah_attr, 0, sizeof(ah_attr));
2287 	ah_attr.type = rdma_ah_find_type(port->agent->device,
2288 					 port->port_num);
2289 	rdma_ah_set_dlid(&ah_attr, port_attr.sm_lid);
2290 	rdma_ah_set_sl(&ah_attr, port_attr.sm_sl);
2291 	rdma_ah_set_port_num(&ah_attr, port->port_num);
2292 	if (port_attr.grh_required) {
2293 		rdma_ah_set_ah_flags(&ah_attr, IB_AH_GRH);
2294 
2295 		rdma_ah_set_subnet_prefix(&ah_attr,
2296 					  cpu_to_be64(port_attr.subnet_prefix));
2297 		rdma_ah_set_interface_id(&ah_attr,
2298 					 cpu_to_be64(IB_SA_WELL_KNOWN_GUID));
2299 	}
2300 
2301 	new_ah->ah = rdma_create_ah(port->agent->qp->pd, &ah_attr);
2302 	if (IS_ERR(new_ah->ah)) {
2303 		pr_warn("Couldn't create new SM AH\n");
2304 		kfree(new_ah);
2305 		return;
2306 	}
2307 
2308 	spin_lock_irq(&port->ah_lock);
2309 	if (port->sm_ah)
2310 		kref_put(&port->sm_ah->ref, free_sm_ah);
2311 	port->sm_ah = new_ah;
2312 	spin_unlock_irq(&port->ah_lock);
2313 }
2314 
2315 static void ib_sa_event(struct ib_event_handler *handler,
2316 			struct ib_event *event)
2317 {
2318 	if (event->event == IB_EVENT_PORT_ERR    ||
2319 	    event->event == IB_EVENT_PORT_ACTIVE ||
2320 	    event->event == IB_EVENT_LID_CHANGE  ||
2321 	    event->event == IB_EVENT_PKEY_CHANGE ||
2322 	    event->event == IB_EVENT_SM_CHANGE   ||
2323 	    event->event == IB_EVENT_CLIENT_REREGISTER) {
2324 		unsigned long flags;
2325 		struct ib_sa_device *sa_dev =
2326 			container_of(handler, typeof(*sa_dev), event_handler);
2327 		u8 port_num = event->element.port_num - sa_dev->start_port;
2328 		struct ib_sa_port *port = &sa_dev->port[port_num];
2329 
2330 		if (!rdma_cap_ib_sa(handler->device, port->port_num))
2331 			return;
2332 
2333 		spin_lock_irqsave(&port->ah_lock, flags);
2334 		if (port->sm_ah)
2335 			kref_put(&port->sm_ah->ref, free_sm_ah);
2336 		port->sm_ah = NULL;
2337 		spin_unlock_irqrestore(&port->ah_lock, flags);
2338 
2339 		if (event->event == IB_EVENT_SM_CHANGE ||
2340 		    event->event == IB_EVENT_CLIENT_REREGISTER ||
2341 		    event->event == IB_EVENT_LID_CHANGE ||
2342 		    event->event == IB_EVENT_PORT_ACTIVE) {
2343 			unsigned long delay =
2344 				msecs_to_jiffies(IB_SA_CPI_RETRY_WAIT);
2345 
2346 			spin_lock_irqsave(&port->classport_lock, flags);
2347 			port->classport_info.valid = false;
2348 			port->classport_info.retry_cnt = 0;
2349 			spin_unlock_irqrestore(&port->classport_lock, flags);
2350 			queue_delayed_work(ib_wq,
2351 					   &port->ib_cpi_work, delay);
2352 		}
2353 		queue_work(ib_wq, &sa_dev->port[port_num].update_task);
2354 	}
2355 }
2356 
2357 static void ib_sa_add_one(struct ib_device *device)
2358 {
2359 	struct ib_sa_device *sa_dev;
2360 	int s, e, i;
2361 	int count = 0;
2362 
2363 	s = rdma_start_port(device);
2364 	e = rdma_end_port(device);
2365 
2366 	sa_dev = kzalloc(sizeof *sa_dev +
2367 			 (e - s + 1) * sizeof (struct ib_sa_port),
2368 			 GFP_KERNEL);
2369 	if (!sa_dev)
2370 		return;
2371 
2372 	sa_dev->start_port = s;
2373 	sa_dev->end_port   = e;
2374 
2375 	for (i = 0; i <= e - s; ++i) {
2376 		spin_lock_init(&sa_dev->port[i].ah_lock);
2377 		if (!rdma_cap_ib_sa(device, i + 1))
2378 			continue;
2379 
2380 		sa_dev->port[i].sm_ah    = NULL;
2381 		sa_dev->port[i].port_num = i + s;
2382 
2383 		spin_lock_init(&sa_dev->port[i].classport_lock);
2384 		sa_dev->port[i].classport_info.valid = false;
2385 
2386 		sa_dev->port[i].agent =
2387 			ib_register_mad_agent(device, i + s, IB_QPT_GSI,
2388 					      NULL, 0, send_handler,
2389 					      recv_handler, sa_dev, 0);
2390 		if (IS_ERR(sa_dev->port[i].agent))
2391 			goto err;
2392 
2393 		INIT_WORK(&sa_dev->port[i].update_task, update_sm_ah);
2394 		INIT_DELAYED_WORK(&sa_dev->port[i].ib_cpi_work,
2395 				  update_ib_cpi);
2396 
2397 		count++;
2398 	}
2399 
2400 	if (!count)
2401 		goto free;
2402 
2403 	ib_set_client_data(device, &sa_client, sa_dev);
2404 
2405 	/*
2406 	 * We register our event handler after everything is set up,
2407 	 * and then update our cached info after the event handler is
2408 	 * registered to avoid any problems if a port changes state
2409 	 * during our initialization.
2410 	 */
2411 
2412 	INIT_IB_EVENT_HANDLER(&sa_dev->event_handler, device, ib_sa_event);
2413 	if (ib_register_event_handler(&sa_dev->event_handler))
2414 		goto err;
2415 
2416 	for (i = 0; i <= e - s; ++i) {
2417 		if (rdma_cap_ib_sa(device, i + 1))
2418 			update_sm_ah(&sa_dev->port[i].update_task);
2419 	}
2420 
2421 	return;
2422 
2423 err:
2424 	while (--i >= 0) {
2425 		if (rdma_cap_ib_sa(device, i + 1))
2426 			ib_unregister_mad_agent(sa_dev->port[i].agent);
2427 	}
2428 free:
2429 	kfree(sa_dev);
2430 	return;
2431 }
2432 
2433 static void ib_sa_remove_one(struct ib_device *device, void *client_data)
2434 {
2435 	struct ib_sa_device *sa_dev = client_data;
2436 	int i;
2437 
2438 	if (!sa_dev)
2439 		return;
2440 
2441 	ib_unregister_event_handler(&sa_dev->event_handler);
2442 	flush_workqueue(ib_wq);
2443 
2444 	for (i = 0; i <= sa_dev->end_port - sa_dev->start_port; ++i) {
2445 		if (rdma_cap_ib_sa(device, i + 1)) {
2446 			cancel_delayed_work_sync(&sa_dev->port[i].ib_cpi_work);
2447 			ib_unregister_mad_agent(sa_dev->port[i].agent);
2448 			if (sa_dev->port[i].sm_ah)
2449 				kref_put(&sa_dev->port[i].sm_ah->ref, free_sm_ah);
2450 		}
2451 
2452 	}
2453 
2454 	kfree(sa_dev);
2455 }
2456 
2457 int ib_sa_init(void)
2458 {
2459 	int ret;
2460 
2461 	get_random_bytes(&tid, sizeof tid);
2462 
2463 	atomic_set(&ib_nl_sa_request_seq, 0);
2464 
2465 	ret = ib_register_client(&sa_client);
2466 	if (ret) {
2467 		pr_err("Couldn't register ib_sa client\n");
2468 		goto err1;
2469 	}
2470 
2471 	ret = mcast_init();
2472 	if (ret) {
2473 		pr_err("Couldn't initialize multicast handling\n");
2474 		goto err2;
2475 	}
2476 
2477 	ib_nl_wq = alloc_ordered_workqueue("ib_nl_sa_wq", WQ_MEM_RECLAIM);
2478 	if (!ib_nl_wq) {
2479 		ret = -ENOMEM;
2480 		goto err3;
2481 	}
2482 
2483 	INIT_DELAYED_WORK(&ib_nl_timed_work, ib_nl_request_timeout);
2484 
2485 	return 0;
2486 
2487 err3:
2488 	mcast_cleanup();
2489 err2:
2490 	ib_unregister_client(&sa_client);
2491 err1:
2492 	return ret;
2493 }
2494 
2495 void ib_sa_cleanup(void)
2496 {
2497 	cancel_delayed_work(&ib_nl_timed_work);
2498 	flush_workqueue(ib_nl_wq);
2499 	destroy_workqueue(ib_nl_wq);
2500 	mcast_cleanup();
2501 	ib_unregister_client(&sa_client);
2502 	idr_destroy(&query_idr);
2503 }
2504