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