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/xarray.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_XARRAY_FLAGS(queries, XA_FLAGS_ALLOC | XA_FLAGS_LOCK_IRQ);
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, dev_name(&query->port->agent->device->dev),
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 	struct ib_sa_mad *mad;
838 	int len;
839 
840 	mad = query->mad_buf->mad;
841 	len = ib_nl_get_path_rec_attrs_len(mad->sa_hdr.comp_mask);
842 	if (len <= 0)
843 		return -EMSGSIZE;
844 
845 	skb = nlmsg_new(len, gfp_mask);
846 	if (!skb)
847 		return -ENOMEM;
848 
849 	/* Put nlmsg header only for now */
850 	data = ibnl_put_msg(skb, &nlh, query->seq, 0, RDMA_NL_LS,
851 			    RDMA_NL_LS_OP_RESOLVE, NLM_F_REQUEST);
852 	if (!data) {
853 		nlmsg_free(skb);
854 		return -EMSGSIZE;
855 	}
856 
857 	/* Add attributes */
858 	ib_nl_set_path_rec_attrs(skb, query);
859 
860 	/* Repair the nlmsg header length */
861 	nlmsg_end(skb, nlh);
862 
863 	return rdma_nl_multicast(&init_net, skb, RDMA_NL_GROUP_LS, gfp_mask);
864 }
865 
866 static int ib_nl_make_request(struct ib_sa_query *query, gfp_t gfp_mask)
867 {
868 	unsigned long flags;
869 	unsigned long delay;
870 	int ret;
871 
872 	INIT_LIST_HEAD(&query->list);
873 	query->seq = (u32)atomic_inc_return(&ib_nl_sa_request_seq);
874 
875 	/* Put the request on the list first.*/
876 	spin_lock_irqsave(&ib_nl_request_lock, flags);
877 	delay = msecs_to_jiffies(sa_local_svc_timeout_ms);
878 	query->timeout = delay + jiffies;
879 	list_add_tail(&query->list, &ib_nl_request_list);
880 	/* Start the timeout if this is the only request */
881 	if (ib_nl_request_list.next == &query->list)
882 		queue_delayed_work(ib_nl_wq, &ib_nl_timed_work, delay);
883 	spin_unlock_irqrestore(&ib_nl_request_lock, flags);
884 
885 	ret = ib_nl_send_msg(query, gfp_mask);
886 	if (ret) {
887 		ret = -EIO;
888 		/* Remove the request */
889 		spin_lock_irqsave(&ib_nl_request_lock, flags);
890 		list_del(&query->list);
891 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
892 	}
893 
894 	return ret;
895 }
896 
897 static int ib_nl_cancel_request(struct ib_sa_query *query)
898 {
899 	unsigned long flags;
900 	struct ib_sa_query *wait_query;
901 	int found = 0;
902 
903 	spin_lock_irqsave(&ib_nl_request_lock, flags);
904 	list_for_each_entry(wait_query, &ib_nl_request_list, list) {
905 		/* Let the timeout to take care of the callback */
906 		if (query == wait_query) {
907 			query->flags |= IB_SA_CANCEL;
908 			query->timeout = jiffies;
909 			list_move(&query->list, &ib_nl_request_list);
910 			found = 1;
911 			mod_delayed_work(ib_nl_wq, &ib_nl_timed_work, 1);
912 			break;
913 		}
914 	}
915 	spin_unlock_irqrestore(&ib_nl_request_lock, flags);
916 
917 	return found;
918 }
919 
920 static void send_handler(struct ib_mad_agent *agent,
921 			 struct ib_mad_send_wc *mad_send_wc);
922 
923 static void ib_nl_process_good_resolve_rsp(struct ib_sa_query *query,
924 					   const struct nlmsghdr *nlh)
925 {
926 	struct ib_mad_send_wc mad_send_wc;
927 	struct ib_sa_mad *mad = NULL;
928 	const struct nlattr *head, *curr;
929 	struct ib_path_rec_data  *rec;
930 	int len, rem;
931 	u32 mask = 0;
932 	int status = -EIO;
933 
934 	if (query->callback) {
935 		head = (const struct nlattr *) nlmsg_data(nlh);
936 		len = nlmsg_len(nlh);
937 		switch (query->path_use) {
938 		case LS_RESOLVE_PATH_USE_UNIDIRECTIONAL:
939 			mask = IB_PATH_PRIMARY | IB_PATH_OUTBOUND;
940 			break;
941 
942 		case LS_RESOLVE_PATH_USE_ALL:
943 		case LS_RESOLVE_PATH_USE_GMP:
944 		default:
945 			mask = IB_PATH_PRIMARY | IB_PATH_GMP |
946 				IB_PATH_BIDIRECTIONAL;
947 			break;
948 		}
949 		nla_for_each_attr(curr, head, len, rem) {
950 			if (curr->nla_type == LS_NLA_TYPE_PATH_RECORD) {
951 				rec = nla_data(curr);
952 				/*
953 				 * Get the first one. In the future, we may
954 				 * need to get up to 6 pathrecords.
955 				 */
956 				if ((rec->flags & mask) == mask) {
957 					mad = query->mad_buf->mad;
958 					mad->mad_hdr.method |=
959 						IB_MGMT_METHOD_RESP;
960 					memcpy(mad->data, rec->path_rec,
961 					       sizeof(rec->path_rec));
962 					status = 0;
963 					break;
964 				}
965 			}
966 		}
967 		query->callback(query, status, mad);
968 	}
969 
970 	mad_send_wc.send_buf = query->mad_buf;
971 	mad_send_wc.status = IB_WC_SUCCESS;
972 	send_handler(query->mad_buf->mad_agent, &mad_send_wc);
973 }
974 
975 static void ib_nl_request_timeout(struct work_struct *work)
976 {
977 	unsigned long flags;
978 	struct ib_sa_query *query;
979 	unsigned long delay;
980 	struct ib_mad_send_wc mad_send_wc;
981 	int ret;
982 
983 	spin_lock_irqsave(&ib_nl_request_lock, flags);
984 	while (!list_empty(&ib_nl_request_list)) {
985 		query = list_entry(ib_nl_request_list.next,
986 				   struct ib_sa_query, list);
987 
988 		if (time_after(query->timeout, jiffies)) {
989 			delay = query->timeout - jiffies;
990 			if ((long)delay <= 0)
991 				delay = 1;
992 			queue_delayed_work(ib_nl_wq, &ib_nl_timed_work, delay);
993 			break;
994 		}
995 
996 		list_del(&query->list);
997 		ib_sa_disable_local_svc(query);
998 		/* Hold the lock to protect against query cancellation */
999 		if (ib_sa_query_cancelled(query))
1000 			ret = -1;
1001 		else
1002 			ret = ib_post_send_mad(query->mad_buf, NULL);
1003 		if (ret) {
1004 			mad_send_wc.send_buf = query->mad_buf;
1005 			mad_send_wc.status = IB_WC_WR_FLUSH_ERR;
1006 			spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1007 			send_handler(query->port->agent, &mad_send_wc);
1008 			spin_lock_irqsave(&ib_nl_request_lock, flags);
1009 		}
1010 	}
1011 	spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1012 }
1013 
1014 int ib_nl_handle_set_timeout(struct sk_buff *skb,
1015 			     struct nlmsghdr *nlh,
1016 			     struct netlink_ext_ack *extack)
1017 {
1018 	int timeout, delta, abs_delta;
1019 	const struct nlattr *attr;
1020 	unsigned long flags;
1021 	struct ib_sa_query *query;
1022 	long delay = 0;
1023 	struct nlattr *tb[LS_NLA_TYPE_MAX];
1024 	int ret;
1025 
1026 	if (!(nlh->nlmsg_flags & NLM_F_REQUEST) ||
1027 	    !(NETLINK_CB(skb).sk))
1028 		return -EPERM;
1029 
1030 	ret = nla_parse_deprecated(tb, LS_NLA_TYPE_MAX - 1, nlmsg_data(nlh),
1031 				   nlmsg_len(nlh), ib_nl_policy, NULL);
1032 	attr = (const struct nlattr *)tb[LS_NLA_TYPE_TIMEOUT];
1033 	if (ret || !attr)
1034 		goto settimeout_out;
1035 
1036 	timeout = *(int *) nla_data(attr);
1037 	if (timeout < IB_SA_LOCAL_SVC_TIMEOUT_MIN)
1038 		timeout = IB_SA_LOCAL_SVC_TIMEOUT_MIN;
1039 	if (timeout > IB_SA_LOCAL_SVC_TIMEOUT_MAX)
1040 		timeout = IB_SA_LOCAL_SVC_TIMEOUT_MAX;
1041 
1042 	delta = timeout - sa_local_svc_timeout_ms;
1043 	if (delta < 0)
1044 		abs_delta = -delta;
1045 	else
1046 		abs_delta = delta;
1047 
1048 	if (delta != 0) {
1049 		spin_lock_irqsave(&ib_nl_request_lock, flags);
1050 		sa_local_svc_timeout_ms = timeout;
1051 		list_for_each_entry(query, &ib_nl_request_list, list) {
1052 			if (delta < 0 && abs_delta > query->timeout)
1053 				query->timeout = 0;
1054 			else
1055 				query->timeout += delta;
1056 
1057 			/* Get the new delay from the first entry */
1058 			if (!delay) {
1059 				delay = query->timeout - jiffies;
1060 				if (delay <= 0)
1061 					delay = 1;
1062 			}
1063 		}
1064 		if (delay)
1065 			mod_delayed_work(ib_nl_wq, &ib_nl_timed_work,
1066 					 (unsigned long)delay);
1067 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1068 	}
1069 
1070 settimeout_out:
1071 	return skb->len;
1072 }
1073 
1074 static inline int ib_nl_is_good_resolve_resp(const struct nlmsghdr *nlh)
1075 {
1076 	struct nlattr *tb[LS_NLA_TYPE_MAX];
1077 	int ret;
1078 
1079 	if (nlh->nlmsg_flags & RDMA_NL_LS_F_ERR)
1080 		return 0;
1081 
1082 	ret = nla_parse_deprecated(tb, LS_NLA_TYPE_MAX - 1, nlmsg_data(nlh),
1083 				   nlmsg_len(nlh), ib_nl_policy, NULL);
1084 	if (ret)
1085 		return 0;
1086 
1087 	return 1;
1088 }
1089 
1090 int ib_nl_handle_resolve_resp(struct sk_buff *skb,
1091 			      struct nlmsghdr *nlh,
1092 			      struct netlink_ext_ack *extack)
1093 {
1094 	unsigned long flags;
1095 	struct ib_sa_query *query;
1096 	struct ib_mad_send_buf *send_buf;
1097 	struct ib_mad_send_wc mad_send_wc;
1098 	int found = 0;
1099 	int ret;
1100 
1101 	if ((nlh->nlmsg_flags & NLM_F_REQUEST) ||
1102 	    !(NETLINK_CB(skb).sk))
1103 		return -EPERM;
1104 
1105 	spin_lock_irqsave(&ib_nl_request_lock, flags);
1106 	list_for_each_entry(query, &ib_nl_request_list, list) {
1107 		/*
1108 		 * If the query is cancelled, let the timeout routine
1109 		 * take care of it.
1110 		 */
1111 		if (nlh->nlmsg_seq == query->seq) {
1112 			found = !ib_sa_query_cancelled(query);
1113 			if (found)
1114 				list_del(&query->list);
1115 			break;
1116 		}
1117 	}
1118 
1119 	if (!found) {
1120 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1121 		goto resp_out;
1122 	}
1123 
1124 	send_buf = query->mad_buf;
1125 
1126 	if (!ib_nl_is_good_resolve_resp(nlh)) {
1127 		/* if the result is a failure, send out the packet via IB */
1128 		ib_sa_disable_local_svc(query);
1129 		ret = ib_post_send_mad(query->mad_buf, NULL);
1130 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1131 		if (ret) {
1132 			mad_send_wc.send_buf = send_buf;
1133 			mad_send_wc.status = IB_WC_GENERAL_ERR;
1134 			send_handler(query->port->agent, &mad_send_wc);
1135 		}
1136 	} else {
1137 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1138 		ib_nl_process_good_resolve_rsp(query, nlh);
1139 	}
1140 
1141 resp_out:
1142 	return skb->len;
1143 }
1144 
1145 static void free_sm_ah(struct kref *kref)
1146 {
1147 	struct ib_sa_sm_ah *sm_ah = container_of(kref, struct ib_sa_sm_ah, ref);
1148 
1149 	rdma_destroy_ah(sm_ah->ah, 0);
1150 	kfree(sm_ah);
1151 }
1152 
1153 void ib_sa_register_client(struct ib_sa_client *client)
1154 {
1155 	atomic_set(&client->users, 1);
1156 	init_completion(&client->comp);
1157 }
1158 EXPORT_SYMBOL(ib_sa_register_client);
1159 
1160 void ib_sa_unregister_client(struct ib_sa_client *client)
1161 {
1162 	ib_sa_client_put(client);
1163 	wait_for_completion(&client->comp);
1164 }
1165 EXPORT_SYMBOL(ib_sa_unregister_client);
1166 
1167 /**
1168  * ib_sa_cancel_query - try to cancel an SA query
1169  * @id:ID of query to cancel
1170  * @query:query pointer to cancel
1171  *
1172  * Try to cancel an SA query.  If the id and query don't match up or
1173  * the query has already completed, nothing is done.  Otherwise the
1174  * query is canceled and will complete with a status of -EINTR.
1175  */
1176 void ib_sa_cancel_query(int id, struct ib_sa_query *query)
1177 {
1178 	unsigned long flags;
1179 	struct ib_mad_agent *agent;
1180 	struct ib_mad_send_buf *mad_buf;
1181 
1182 	xa_lock_irqsave(&queries, flags);
1183 	if (xa_load(&queries, id) != query) {
1184 		xa_unlock_irqrestore(&queries, flags);
1185 		return;
1186 	}
1187 	agent = query->port->agent;
1188 	mad_buf = query->mad_buf;
1189 	xa_unlock_irqrestore(&queries, flags);
1190 
1191 	/*
1192 	 * If the query is still on the netlink request list, schedule
1193 	 * it to be cancelled by the timeout routine. Otherwise, it has been
1194 	 * sent to the MAD layer and has to be cancelled from there.
1195 	 */
1196 	if (!ib_nl_cancel_request(query))
1197 		ib_cancel_mad(agent, mad_buf);
1198 }
1199 EXPORT_SYMBOL(ib_sa_cancel_query);
1200 
1201 static u8 get_src_path_mask(struct ib_device *device, u8 port_num)
1202 {
1203 	struct ib_sa_device *sa_dev;
1204 	struct ib_sa_port   *port;
1205 	unsigned long flags;
1206 	u8 src_path_mask;
1207 
1208 	sa_dev = ib_get_client_data(device, &sa_client);
1209 	if (!sa_dev)
1210 		return 0x7f;
1211 
1212 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1213 	spin_lock_irqsave(&port->ah_lock, flags);
1214 	src_path_mask = port->sm_ah ? port->sm_ah->src_path_mask : 0x7f;
1215 	spin_unlock_irqrestore(&port->ah_lock, flags);
1216 
1217 	return src_path_mask;
1218 }
1219 
1220 static int init_ah_attr_grh_fields(struct ib_device *device, u8 port_num,
1221 				   struct sa_path_rec *rec,
1222 				   struct rdma_ah_attr *ah_attr,
1223 				   const struct ib_gid_attr *gid_attr)
1224 {
1225 	enum ib_gid_type type = sa_conv_pathrec_to_gid_type(rec);
1226 
1227 	if (!gid_attr) {
1228 		gid_attr = rdma_find_gid_by_port(device, &rec->sgid, type,
1229 						 port_num, NULL);
1230 		if (IS_ERR(gid_attr))
1231 			return PTR_ERR(gid_attr);
1232 	} else
1233 		rdma_hold_gid_attr(gid_attr);
1234 
1235 	rdma_move_grh_sgid_attr(ah_attr, &rec->dgid,
1236 				be32_to_cpu(rec->flow_label),
1237 				rec->hop_limit,	rec->traffic_class,
1238 				gid_attr);
1239 	return 0;
1240 }
1241 
1242 /**
1243  * ib_init_ah_attr_from_path - Initialize address handle attributes based on
1244  *   an SA path record.
1245  * @device: Device associated ah attributes initialization.
1246  * @port_num: Port on the specified device.
1247  * @rec: path record entry to use for ah attributes initialization.
1248  * @ah_attr: address handle attributes to initialization from path record.
1249  * @sgid_attr: SGID attribute to consider during initialization.
1250  *
1251  * When ib_init_ah_attr_from_path() returns success,
1252  * (a) for IB link layer it optionally contains a reference to SGID attribute
1253  * when GRH is present for IB link layer.
1254  * (b) for RoCE link layer it contains a reference to SGID attribute.
1255  * User must invoke rdma_destroy_ah_attr() to release reference to SGID
1256  * attributes which are initialized using ib_init_ah_attr_from_path().
1257  */
1258 int ib_init_ah_attr_from_path(struct ib_device *device, u8 port_num,
1259 			      struct sa_path_rec *rec,
1260 			      struct rdma_ah_attr *ah_attr,
1261 			      const struct ib_gid_attr *gid_attr)
1262 {
1263 	int ret = 0;
1264 
1265 	memset(ah_attr, 0, sizeof(*ah_attr));
1266 	ah_attr->type = rdma_ah_find_type(device, port_num);
1267 	rdma_ah_set_sl(ah_attr, rec->sl);
1268 	rdma_ah_set_port_num(ah_attr, port_num);
1269 	rdma_ah_set_static_rate(ah_attr, rec->rate);
1270 
1271 	if (sa_path_is_roce(rec)) {
1272 		ret = roce_resolve_route_from_path(rec, gid_attr);
1273 		if (ret)
1274 			return ret;
1275 
1276 		memcpy(ah_attr->roce.dmac, sa_path_get_dmac(rec), ETH_ALEN);
1277 	} else {
1278 		rdma_ah_set_dlid(ah_attr, be32_to_cpu(sa_path_get_dlid(rec)));
1279 		if (sa_path_is_opa(rec) &&
1280 		    rdma_ah_get_dlid(ah_attr) == be16_to_cpu(IB_LID_PERMISSIVE))
1281 			rdma_ah_set_make_grd(ah_attr, true);
1282 
1283 		rdma_ah_set_path_bits(ah_attr,
1284 				      be32_to_cpu(sa_path_get_slid(rec)) &
1285 				      get_src_path_mask(device, port_num));
1286 	}
1287 
1288 	if (rec->hop_limit > 0 || sa_path_is_roce(rec))
1289 		ret = init_ah_attr_grh_fields(device, port_num,
1290 					      rec, ah_attr, gid_attr);
1291 	return ret;
1292 }
1293 EXPORT_SYMBOL(ib_init_ah_attr_from_path);
1294 
1295 static int alloc_mad(struct ib_sa_query *query, gfp_t gfp_mask)
1296 {
1297 	struct rdma_ah_attr ah_attr;
1298 	unsigned long flags;
1299 
1300 	spin_lock_irqsave(&query->port->ah_lock, flags);
1301 	if (!query->port->sm_ah) {
1302 		spin_unlock_irqrestore(&query->port->ah_lock, flags);
1303 		return -EAGAIN;
1304 	}
1305 	kref_get(&query->port->sm_ah->ref);
1306 	query->sm_ah = query->port->sm_ah;
1307 	spin_unlock_irqrestore(&query->port->ah_lock, flags);
1308 
1309 	/*
1310 	 * Always check if sm_ah has valid dlid assigned,
1311 	 * before querying for class port info
1312 	 */
1313 	if ((rdma_query_ah(query->sm_ah->ah, &ah_attr) < 0) ||
1314 	    !rdma_is_valid_unicast_lid(&ah_attr)) {
1315 		kref_put(&query->sm_ah->ref, free_sm_ah);
1316 		return -EAGAIN;
1317 	}
1318 	query->mad_buf = ib_create_send_mad(query->port->agent, 1,
1319 					    query->sm_ah->pkey_index,
1320 					    0, IB_MGMT_SA_HDR, IB_MGMT_SA_DATA,
1321 					    gfp_mask,
1322 					    ((query->flags & IB_SA_QUERY_OPA) ?
1323 					     OPA_MGMT_BASE_VERSION :
1324 					     IB_MGMT_BASE_VERSION));
1325 	if (IS_ERR(query->mad_buf)) {
1326 		kref_put(&query->sm_ah->ref, free_sm_ah);
1327 		return -ENOMEM;
1328 	}
1329 
1330 	query->mad_buf->ah = query->sm_ah->ah;
1331 
1332 	return 0;
1333 }
1334 
1335 static void free_mad(struct ib_sa_query *query)
1336 {
1337 	ib_free_send_mad(query->mad_buf);
1338 	kref_put(&query->sm_ah->ref, free_sm_ah);
1339 }
1340 
1341 static void init_mad(struct ib_sa_query *query, struct ib_mad_agent *agent)
1342 {
1343 	struct ib_sa_mad *mad = query->mad_buf->mad;
1344 	unsigned long flags;
1345 
1346 	memset(mad, 0, sizeof *mad);
1347 
1348 	if (query->flags & IB_SA_QUERY_OPA) {
1349 		mad->mad_hdr.base_version  = OPA_MGMT_BASE_VERSION;
1350 		mad->mad_hdr.class_version = OPA_SA_CLASS_VERSION;
1351 	} else {
1352 		mad->mad_hdr.base_version  = IB_MGMT_BASE_VERSION;
1353 		mad->mad_hdr.class_version = IB_SA_CLASS_VERSION;
1354 	}
1355 	mad->mad_hdr.mgmt_class    = IB_MGMT_CLASS_SUBN_ADM;
1356 	spin_lock_irqsave(&tid_lock, flags);
1357 	mad->mad_hdr.tid           =
1358 		cpu_to_be64(((u64) agent->hi_tid) << 32 | tid++);
1359 	spin_unlock_irqrestore(&tid_lock, flags);
1360 }
1361 
1362 static int send_mad(struct ib_sa_query *query, unsigned long timeout_ms,
1363 		    gfp_t gfp_mask)
1364 {
1365 	unsigned long flags;
1366 	int ret, id;
1367 
1368 	xa_lock_irqsave(&queries, flags);
1369 	ret = __xa_alloc(&queries, &id, query, xa_limit_32b, gfp_mask);
1370 	xa_unlock_irqrestore(&queries, flags);
1371 	if (ret < 0)
1372 		return ret;
1373 
1374 	query->mad_buf->timeout_ms  = timeout_ms;
1375 	query->mad_buf->context[0] = query;
1376 	query->id = id;
1377 
1378 	if ((query->flags & IB_SA_ENABLE_LOCAL_SERVICE) &&
1379 	    (!(query->flags & IB_SA_QUERY_OPA))) {
1380 		if (rdma_nl_chk_listeners(RDMA_NL_GROUP_LS)) {
1381 			if (!ib_nl_make_request(query, gfp_mask))
1382 				return id;
1383 		}
1384 		ib_sa_disable_local_svc(query);
1385 	}
1386 
1387 	ret = ib_post_send_mad(query->mad_buf, NULL);
1388 	if (ret) {
1389 		xa_lock_irqsave(&queries, flags);
1390 		__xa_erase(&queries, id);
1391 		xa_unlock_irqrestore(&queries, flags);
1392 	}
1393 
1394 	/*
1395 	 * It's not safe to dereference query any more, because the
1396 	 * send may already have completed and freed the query in
1397 	 * another context.
1398 	 */
1399 	return ret ? ret : id;
1400 }
1401 
1402 void ib_sa_unpack_path(void *attribute, struct sa_path_rec *rec)
1403 {
1404 	ib_unpack(path_rec_table, ARRAY_SIZE(path_rec_table), attribute, rec);
1405 }
1406 EXPORT_SYMBOL(ib_sa_unpack_path);
1407 
1408 void ib_sa_pack_path(struct sa_path_rec *rec, void *attribute)
1409 {
1410 	ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table), rec, attribute);
1411 }
1412 EXPORT_SYMBOL(ib_sa_pack_path);
1413 
1414 static bool ib_sa_opa_pathrecord_support(struct ib_sa_client *client,
1415 					 struct ib_device *device,
1416 					 u8 port_num)
1417 {
1418 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1419 	struct ib_sa_port *port;
1420 	unsigned long flags;
1421 	bool ret = false;
1422 
1423 	if (!sa_dev)
1424 		return ret;
1425 
1426 	port = &sa_dev->port[port_num - sa_dev->start_port];
1427 	spin_lock_irqsave(&port->classport_lock, flags);
1428 	if (!port->classport_info.valid)
1429 		goto ret;
1430 
1431 	if (port->classport_info.data.type == RDMA_CLASS_PORT_INFO_OPA)
1432 		ret = opa_get_cpi_capmask2(&port->classport_info.data.opa) &
1433 			OPA_CLASS_PORT_INFO_PR_SUPPORT;
1434 ret:
1435 	spin_unlock_irqrestore(&port->classport_lock, flags);
1436 	return ret;
1437 }
1438 
1439 enum opa_pr_supported {
1440 	PR_NOT_SUPPORTED,
1441 	PR_OPA_SUPPORTED,
1442 	PR_IB_SUPPORTED
1443 };
1444 
1445 /**
1446  * Check if current PR query can be an OPA query.
1447  * Retuns PR_NOT_SUPPORTED if a path record query is not
1448  * possible, PR_OPA_SUPPORTED if an OPA path record query
1449  * is possible and PR_IB_SUPPORTED if an IB path record
1450  * query is possible.
1451  */
1452 static int opa_pr_query_possible(struct ib_sa_client *client,
1453 				 struct ib_device *device,
1454 				 u8 port_num,
1455 				 struct sa_path_rec *rec)
1456 {
1457 	struct ib_port_attr port_attr;
1458 
1459 	if (ib_query_port(device, port_num, &port_attr))
1460 		return PR_NOT_SUPPORTED;
1461 
1462 	if (ib_sa_opa_pathrecord_support(client, device, port_num))
1463 		return PR_OPA_SUPPORTED;
1464 
1465 	if (port_attr.lid >= be16_to_cpu(IB_MULTICAST_LID_BASE))
1466 		return PR_NOT_SUPPORTED;
1467 	else
1468 		return PR_IB_SUPPORTED;
1469 }
1470 
1471 static void ib_sa_path_rec_callback(struct ib_sa_query *sa_query,
1472 				    int status,
1473 				    struct ib_sa_mad *mad)
1474 {
1475 	struct ib_sa_path_query *query =
1476 		container_of(sa_query, struct ib_sa_path_query, sa_query);
1477 
1478 	if (mad) {
1479 		struct sa_path_rec rec;
1480 
1481 		if (sa_query->flags & IB_SA_QUERY_OPA) {
1482 			ib_unpack(opa_path_rec_table,
1483 				  ARRAY_SIZE(opa_path_rec_table),
1484 				  mad->data, &rec);
1485 			rec.rec_type = SA_PATH_REC_TYPE_OPA;
1486 			query->callback(status, &rec, query->context);
1487 		} else {
1488 			ib_unpack(path_rec_table,
1489 				  ARRAY_SIZE(path_rec_table),
1490 				  mad->data, &rec);
1491 			rec.rec_type = SA_PATH_REC_TYPE_IB;
1492 			sa_path_set_dmac_zero(&rec);
1493 
1494 			if (query->conv_pr) {
1495 				struct sa_path_rec opa;
1496 
1497 				memset(&opa, 0, sizeof(struct sa_path_rec));
1498 				sa_convert_path_ib_to_opa(&opa, &rec);
1499 				query->callback(status, &opa, query->context);
1500 			} else {
1501 				query->callback(status, &rec, query->context);
1502 			}
1503 		}
1504 	} else
1505 		query->callback(status, NULL, query->context);
1506 }
1507 
1508 static void ib_sa_path_rec_release(struct ib_sa_query *sa_query)
1509 {
1510 	struct ib_sa_path_query *query =
1511 		container_of(sa_query, struct ib_sa_path_query, sa_query);
1512 
1513 	kfree(query->conv_pr);
1514 	kfree(query);
1515 }
1516 
1517 /**
1518  * ib_sa_path_rec_get - Start a Path get query
1519  * @client:SA client
1520  * @device:device to send query on
1521  * @port_num: port number to send query on
1522  * @rec:Path Record to send in query
1523  * @comp_mask:component mask to send in query
1524  * @timeout_ms:time to wait for response
1525  * @gfp_mask:GFP mask to use for internal allocations
1526  * @callback:function called when query completes, times out or is
1527  * canceled
1528  * @context:opaque user context passed to callback
1529  * @sa_query:query context, used to cancel query
1530  *
1531  * Send a Path Record Get query to the SA to look up a path.  The
1532  * callback function will be called when the query completes (or
1533  * fails); status is 0 for a successful response, -EINTR if the query
1534  * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
1535  * occurred sending the query.  The resp parameter of the callback is
1536  * only valid if status is 0.
1537  *
1538  * If the return value of ib_sa_path_rec_get() is negative, it is an
1539  * error code.  Otherwise it is a query ID that can be used to cancel
1540  * the query.
1541  */
1542 int ib_sa_path_rec_get(struct ib_sa_client *client,
1543 		       struct ib_device *device, u8 port_num,
1544 		       struct sa_path_rec *rec,
1545 		       ib_sa_comp_mask comp_mask,
1546 		       unsigned long timeout_ms, gfp_t gfp_mask,
1547 		       void (*callback)(int status,
1548 					struct sa_path_rec *resp,
1549 					void *context),
1550 		       void *context,
1551 		       struct ib_sa_query **sa_query)
1552 {
1553 	struct ib_sa_path_query *query;
1554 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1555 	struct ib_sa_port   *port;
1556 	struct ib_mad_agent *agent;
1557 	struct ib_sa_mad *mad;
1558 	enum opa_pr_supported status;
1559 	int ret;
1560 
1561 	if (!sa_dev)
1562 		return -ENODEV;
1563 
1564 	if ((rec->rec_type != SA_PATH_REC_TYPE_IB) &&
1565 	    (rec->rec_type != SA_PATH_REC_TYPE_OPA))
1566 		return -EINVAL;
1567 
1568 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1569 	agent = port->agent;
1570 
1571 	query = kzalloc(sizeof(*query), gfp_mask);
1572 	if (!query)
1573 		return -ENOMEM;
1574 
1575 	query->sa_query.port     = port;
1576 	if (rec->rec_type == SA_PATH_REC_TYPE_OPA) {
1577 		status = opa_pr_query_possible(client, device, port_num, rec);
1578 		if (status == PR_NOT_SUPPORTED) {
1579 			ret = -EINVAL;
1580 			goto err1;
1581 		} else if (status == PR_OPA_SUPPORTED) {
1582 			query->sa_query.flags |= IB_SA_QUERY_OPA;
1583 		} else {
1584 			query->conv_pr =
1585 				kmalloc(sizeof(*query->conv_pr), gfp_mask);
1586 			if (!query->conv_pr) {
1587 				ret = -ENOMEM;
1588 				goto err1;
1589 			}
1590 		}
1591 	}
1592 
1593 	ret = alloc_mad(&query->sa_query, gfp_mask);
1594 	if (ret)
1595 		goto err2;
1596 
1597 	ib_sa_client_get(client);
1598 	query->sa_query.client = client;
1599 	query->callback        = callback;
1600 	query->context         = context;
1601 
1602 	mad = query->sa_query.mad_buf->mad;
1603 	init_mad(&query->sa_query, agent);
1604 
1605 	query->sa_query.callback = callback ? ib_sa_path_rec_callback : NULL;
1606 	query->sa_query.release  = ib_sa_path_rec_release;
1607 	mad->mad_hdr.method	 = IB_MGMT_METHOD_GET;
1608 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_PATH_REC);
1609 	mad->sa_hdr.comp_mask	 = comp_mask;
1610 
1611 	if (query->sa_query.flags & IB_SA_QUERY_OPA) {
1612 		ib_pack(opa_path_rec_table, ARRAY_SIZE(opa_path_rec_table),
1613 			rec, mad->data);
1614 	} else if (query->conv_pr) {
1615 		sa_convert_path_opa_to_ib(query->conv_pr, rec);
1616 		ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table),
1617 			query->conv_pr, mad->data);
1618 	} else {
1619 		ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table),
1620 			rec, mad->data);
1621 	}
1622 
1623 	*sa_query = &query->sa_query;
1624 
1625 	query->sa_query.flags |= IB_SA_ENABLE_LOCAL_SERVICE;
1626 	query->sa_query.mad_buf->context[1] = (query->conv_pr) ?
1627 						query->conv_pr : rec;
1628 
1629 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1630 	if (ret < 0)
1631 		goto err3;
1632 
1633 	return ret;
1634 
1635 err3:
1636 	*sa_query = NULL;
1637 	ib_sa_client_put(query->sa_query.client);
1638 	free_mad(&query->sa_query);
1639 err2:
1640 	kfree(query->conv_pr);
1641 err1:
1642 	kfree(query);
1643 	return ret;
1644 }
1645 EXPORT_SYMBOL(ib_sa_path_rec_get);
1646 
1647 static void ib_sa_service_rec_callback(struct ib_sa_query *sa_query,
1648 				    int status,
1649 				    struct ib_sa_mad *mad)
1650 {
1651 	struct ib_sa_service_query *query =
1652 		container_of(sa_query, struct ib_sa_service_query, sa_query);
1653 
1654 	if (mad) {
1655 		struct ib_sa_service_rec rec;
1656 
1657 		ib_unpack(service_rec_table, ARRAY_SIZE(service_rec_table),
1658 			  mad->data, &rec);
1659 		query->callback(status, &rec, query->context);
1660 	} else
1661 		query->callback(status, NULL, query->context);
1662 }
1663 
1664 static void ib_sa_service_rec_release(struct ib_sa_query *sa_query)
1665 {
1666 	kfree(container_of(sa_query, struct ib_sa_service_query, sa_query));
1667 }
1668 
1669 /**
1670  * ib_sa_service_rec_query - Start Service Record operation
1671  * @client:SA client
1672  * @device:device to send request on
1673  * @port_num: port number to send request on
1674  * @method:SA method - should be get, set, or delete
1675  * @rec:Service Record to send in request
1676  * @comp_mask:component mask to send in request
1677  * @timeout_ms:time to wait for response
1678  * @gfp_mask:GFP mask to use for internal allocations
1679  * @callback:function called when request completes, times out or is
1680  * canceled
1681  * @context:opaque user context passed to callback
1682  * @sa_query:request context, used to cancel request
1683  *
1684  * Send a Service Record set/get/delete to the SA to register,
1685  * unregister or query a service record.
1686  * The callback function will be called when the request completes (or
1687  * fails); status is 0 for a successful response, -EINTR if the query
1688  * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
1689  * occurred sending the query.  The resp parameter of the callback is
1690  * only valid if status is 0.
1691  *
1692  * If the return value of ib_sa_service_rec_query() is negative, it is an
1693  * error code.  Otherwise it is a request ID that can be used to cancel
1694  * the query.
1695  */
1696 int ib_sa_service_rec_query(struct ib_sa_client *client,
1697 			    struct ib_device *device, u8 port_num, u8 method,
1698 			    struct ib_sa_service_rec *rec,
1699 			    ib_sa_comp_mask comp_mask,
1700 			    unsigned long timeout_ms, gfp_t gfp_mask,
1701 			    void (*callback)(int status,
1702 					     struct ib_sa_service_rec *resp,
1703 					     void *context),
1704 			    void *context,
1705 			    struct ib_sa_query **sa_query)
1706 {
1707 	struct ib_sa_service_query *query;
1708 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1709 	struct ib_sa_port   *port;
1710 	struct ib_mad_agent *agent;
1711 	struct ib_sa_mad *mad;
1712 	int ret;
1713 
1714 	if (!sa_dev)
1715 		return -ENODEV;
1716 
1717 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1718 	agent = port->agent;
1719 
1720 	if (method != IB_MGMT_METHOD_GET &&
1721 	    method != IB_MGMT_METHOD_SET &&
1722 	    method != IB_SA_METHOD_DELETE)
1723 		return -EINVAL;
1724 
1725 	query = kzalloc(sizeof(*query), gfp_mask);
1726 	if (!query)
1727 		return -ENOMEM;
1728 
1729 	query->sa_query.port     = port;
1730 	ret = alloc_mad(&query->sa_query, gfp_mask);
1731 	if (ret)
1732 		goto err1;
1733 
1734 	ib_sa_client_get(client);
1735 	query->sa_query.client = client;
1736 	query->callback        = callback;
1737 	query->context         = context;
1738 
1739 	mad = query->sa_query.mad_buf->mad;
1740 	init_mad(&query->sa_query, agent);
1741 
1742 	query->sa_query.callback = callback ? ib_sa_service_rec_callback : NULL;
1743 	query->sa_query.release  = ib_sa_service_rec_release;
1744 	mad->mad_hdr.method	 = method;
1745 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_SERVICE_REC);
1746 	mad->sa_hdr.comp_mask	 = comp_mask;
1747 
1748 	ib_pack(service_rec_table, ARRAY_SIZE(service_rec_table),
1749 		rec, mad->data);
1750 
1751 	*sa_query = &query->sa_query;
1752 
1753 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1754 	if (ret < 0)
1755 		goto err2;
1756 
1757 	return ret;
1758 
1759 err2:
1760 	*sa_query = NULL;
1761 	ib_sa_client_put(query->sa_query.client);
1762 	free_mad(&query->sa_query);
1763 
1764 err1:
1765 	kfree(query);
1766 	return ret;
1767 }
1768 EXPORT_SYMBOL(ib_sa_service_rec_query);
1769 
1770 static void ib_sa_mcmember_rec_callback(struct ib_sa_query *sa_query,
1771 					int status,
1772 					struct ib_sa_mad *mad)
1773 {
1774 	struct ib_sa_mcmember_query *query =
1775 		container_of(sa_query, struct ib_sa_mcmember_query, sa_query);
1776 
1777 	if (mad) {
1778 		struct ib_sa_mcmember_rec rec;
1779 
1780 		ib_unpack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
1781 			  mad->data, &rec);
1782 		query->callback(status, &rec, query->context);
1783 	} else
1784 		query->callback(status, NULL, query->context);
1785 }
1786 
1787 static void ib_sa_mcmember_rec_release(struct ib_sa_query *sa_query)
1788 {
1789 	kfree(container_of(sa_query, struct ib_sa_mcmember_query, sa_query));
1790 }
1791 
1792 int ib_sa_mcmember_rec_query(struct ib_sa_client *client,
1793 			     struct ib_device *device, u8 port_num,
1794 			     u8 method,
1795 			     struct ib_sa_mcmember_rec *rec,
1796 			     ib_sa_comp_mask comp_mask,
1797 			     unsigned long timeout_ms, gfp_t gfp_mask,
1798 			     void (*callback)(int status,
1799 					      struct ib_sa_mcmember_rec *resp,
1800 					      void *context),
1801 			     void *context,
1802 			     struct ib_sa_query **sa_query)
1803 {
1804 	struct ib_sa_mcmember_query *query;
1805 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1806 	struct ib_sa_port   *port;
1807 	struct ib_mad_agent *agent;
1808 	struct ib_sa_mad *mad;
1809 	int ret;
1810 
1811 	if (!sa_dev)
1812 		return -ENODEV;
1813 
1814 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1815 	agent = port->agent;
1816 
1817 	query = kzalloc(sizeof(*query), gfp_mask);
1818 	if (!query)
1819 		return -ENOMEM;
1820 
1821 	query->sa_query.port     = port;
1822 	ret = alloc_mad(&query->sa_query, gfp_mask);
1823 	if (ret)
1824 		goto err1;
1825 
1826 	ib_sa_client_get(client);
1827 	query->sa_query.client = client;
1828 	query->callback        = callback;
1829 	query->context         = context;
1830 
1831 	mad = query->sa_query.mad_buf->mad;
1832 	init_mad(&query->sa_query, agent);
1833 
1834 	query->sa_query.callback = callback ? ib_sa_mcmember_rec_callback : NULL;
1835 	query->sa_query.release  = ib_sa_mcmember_rec_release;
1836 	mad->mad_hdr.method	 = method;
1837 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_MC_MEMBER_REC);
1838 	mad->sa_hdr.comp_mask	 = comp_mask;
1839 
1840 	ib_pack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
1841 		rec, mad->data);
1842 
1843 	*sa_query = &query->sa_query;
1844 
1845 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1846 	if (ret < 0)
1847 		goto err2;
1848 
1849 	return ret;
1850 
1851 err2:
1852 	*sa_query = NULL;
1853 	ib_sa_client_put(query->sa_query.client);
1854 	free_mad(&query->sa_query);
1855 
1856 err1:
1857 	kfree(query);
1858 	return ret;
1859 }
1860 
1861 /* Support GuidInfoRecord */
1862 static void ib_sa_guidinfo_rec_callback(struct ib_sa_query *sa_query,
1863 					int status,
1864 					struct ib_sa_mad *mad)
1865 {
1866 	struct ib_sa_guidinfo_query *query =
1867 		container_of(sa_query, struct ib_sa_guidinfo_query, sa_query);
1868 
1869 	if (mad) {
1870 		struct ib_sa_guidinfo_rec rec;
1871 
1872 		ib_unpack(guidinfo_rec_table, ARRAY_SIZE(guidinfo_rec_table),
1873 			  mad->data, &rec);
1874 		query->callback(status, &rec, query->context);
1875 	} else
1876 		query->callback(status, NULL, query->context);
1877 }
1878 
1879 static void ib_sa_guidinfo_rec_release(struct ib_sa_query *sa_query)
1880 {
1881 	kfree(container_of(sa_query, struct ib_sa_guidinfo_query, sa_query));
1882 }
1883 
1884 int ib_sa_guid_info_rec_query(struct ib_sa_client *client,
1885 			      struct ib_device *device, u8 port_num,
1886 			      struct ib_sa_guidinfo_rec *rec,
1887 			      ib_sa_comp_mask comp_mask, u8 method,
1888 			      unsigned long timeout_ms, gfp_t gfp_mask,
1889 			      void (*callback)(int status,
1890 					       struct ib_sa_guidinfo_rec *resp,
1891 					       void *context),
1892 			      void *context,
1893 			      struct ib_sa_query **sa_query)
1894 {
1895 	struct ib_sa_guidinfo_query *query;
1896 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1897 	struct ib_sa_port *port;
1898 	struct ib_mad_agent *agent;
1899 	struct ib_sa_mad *mad;
1900 	int ret;
1901 
1902 	if (!sa_dev)
1903 		return -ENODEV;
1904 
1905 	if (method != IB_MGMT_METHOD_GET &&
1906 	    method != IB_MGMT_METHOD_SET &&
1907 	    method != IB_SA_METHOD_DELETE) {
1908 		return -EINVAL;
1909 	}
1910 
1911 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1912 	agent = port->agent;
1913 
1914 	query = kzalloc(sizeof(*query), gfp_mask);
1915 	if (!query)
1916 		return -ENOMEM;
1917 
1918 	query->sa_query.port = port;
1919 	ret = alloc_mad(&query->sa_query, gfp_mask);
1920 	if (ret)
1921 		goto err1;
1922 
1923 	ib_sa_client_get(client);
1924 	query->sa_query.client = client;
1925 	query->callback        = callback;
1926 	query->context         = context;
1927 
1928 	mad = query->sa_query.mad_buf->mad;
1929 	init_mad(&query->sa_query, agent);
1930 
1931 	query->sa_query.callback = callback ? ib_sa_guidinfo_rec_callback : NULL;
1932 	query->sa_query.release  = ib_sa_guidinfo_rec_release;
1933 
1934 	mad->mad_hdr.method	 = method;
1935 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_GUID_INFO_REC);
1936 	mad->sa_hdr.comp_mask	 = comp_mask;
1937 
1938 	ib_pack(guidinfo_rec_table, ARRAY_SIZE(guidinfo_rec_table), rec,
1939 		mad->data);
1940 
1941 	*sa_query = &query->sa_query;
1942 
1943 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1944 	if (ret < 0)
1945 		goto err2;
1946 
1947 	return ret;
1948 
1949 err2:
1950 	*sa_query = NULL;
1951 	ib_sa_client_put(query->sa_query.client);
1952 	free_mad(&query->sa_query);
1953 
1954 err1:
1955 	kfree(query);
1956 	return ret;
1957 }
1958 EXPORT_SYMBOL(ib_sa_guid_info_rec_query);
1959 
1960 bool ib_sa_sendonly_fullmem_support(struct ib_sa_client *client,
1961 				    struct ib_device *device,
1962 				    u8 port_num)
1963 {
1964 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1965 	struct ib_sa_port *port;
1966 	bool ret = false;
1967 	unsigned long flags;
1968 
1969 	if (!sa_dev)
1970 		return ret;
1971 
1972 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1973 
1974 	spin_lock_irqsave(&port->classport_lock, flags);
1975 	if ((port->classport_info.valid) &&
1976 	    (port->classport_info.data.type == RDMA_CLASS_PORT_INFO_IB))
1977 		ret = ib_get_cpi_capmask2(&port->classport_info.data.ib)
1978 			& IB_SA_CAP_MASK2_SENDONLY_FULL_MEM_SUPPORT;
1979 	spin_unlock_irqrestore(&port->classport_lock, flags);
1980 	return ret;
1981 }
1982 EXPORT_SYMBOL(ib_sa_sendonly_fullmem_support);
1983 
1984 struct ib_classport_info_context {
1985 	struct completion	done;
1986 	struct ib_sa_query	*sa_query;
1987 };
1988 
1989 static void ib_classportinfo_cb(void *context)
1990 {
1991 	struct ib_classport_info_context *cb_ctx = context;
1992 
1993 	complete(&cb_ctx->done);
1994 }
1995 
1996 static void ib_sa_classport_info_rec_callback(struct ib_sa_query *sa_query,
1997 					      int status,
1998 					      struct ib_sa_mad *mad)
1999 {
2000 	unsigned long flags;
2001 	struct ib_sa_classport_info_query *query =
2002 		container_of(sa_query, struct ib_sa_classport_info_query, sa_query);
2003 	struct ib_sa_classport_cache *info = &sa_query->port->classport_info;
2004 
2005 	if (mad) {
2006 		if (sa_query->flags & IB_SA_QUERY_OPA) {
2007 			struct opa_class_port_info rec;
2008 
2009 			ib_unpack(opa_classport_info_rec_table,
2010 				  ARRAY_SIZE(opa_classport_info_rec_table),
2011 				  mad->data, &rec);
2012 
2013 			spin_lock_irqsave(&sa_query->port->classport_lock,
2014 					  flags);
2015 			if (!status && !info->valid) {
2016 				memcpy(&info->data.opa, &rec,
2017 				       sizeof(info->data.opa));
2018 
2019 				info->valid = true;
2020 				info->data.type = RDMA_CLASS_PORT_INFO_OPA;
2021 			}
2022 			spin_unlock_irqrestore(&sa_query->port->classport_lock,
2023 					       flags);
2024 
2025 		} else {
2026 			struct ib_class_port_info rec;
2027 
2028 			ib_unpack(ib_classport_info_rec_table,
2029 				  ARRAY_SIZE(ib_classport_info_rec_table),
2030 				  mad->data, &rec);
2031 
2032 			spin_lock_irqsave(&sa_query->port->classport_lock,
2033 					  flags);
2034 			if (!status && !info->valid) {
2035 				memcpy(&info->data.ib, &rec,
2036 				       sizeof(info->data.ib));
2037 
2038 				info->valid = true;
2039 				info->data.type = RDMA_CLASS_PORT_INFO_IB;
2040 			}
2041 			spin_unlock_irqrestore(&sa_query->port->classport_lock,
2042 					       flags);
2043 		}
2044 	}
2045 	query->callback(query->context);
2046 }
2047 
2048 static void ib_sa_classport_info_rec_release(struct ib_sa_query *sa_query)
2049 {
2050 	kfree(container_of(sa_query, struct ib_sa_classport_info_query,
2051 			   sa_query));
2052 }
2053 
2054 static int ib_sa_classport_info_rec_query(struct ib_sa_port *port,
2055 					  unsigned long timeout_ms,
2056 					  void (*callback)(void *context),
2057 					  void *context,
2058 					  struct ib_sa_query **sa_query)
2059 {
2060 	struct ib_mad_agent *agent;
2061 	struct ib_sa_classport_info_query *query;
2062 	struct ib_sa_mad *mad;
2063 	gfp_t gfp_mask = GFP_KERNEL;
2064 	int ret;
2065 
2066 	agent = port->agent;
2067 
2068 	query = kzalloc(sizeof(*query), gfp_mask);
2069 	if (!query)
2070 		return -ENOMEM;
2071 
2072 	query->sa_query.port = port;
2073 	query->sa_query.flags |= rdma_cap_opa_ah(port->agent->device,
2074 						 port->port_num) ?
2075 				 IB_SA_QUERY_OPA : 0;
2076 	ret = alloc_mad(&query->sa_query, gfp_mask);
2077 	if (ret)
2078 		goto err_free;
2079 
2080 	query->callback = callback;
2081 	query->context = context;
2082 
2083 	mad = query->sa_query.mad_buf->mad;
2084 	init_mad(&query->sa_query, agent);
2085 
2086 	query->sa_query.callback = ib_sa_classport_info_rec_callback;
2087 	query->sa_query.release  = ib_sa_classport_info_rec_release;
2088 	mad->mad_hdr.method	 = IB_MGMT_METHOD_GET;
2089 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_CLASS_PORTINFO);
2090 	mad->sa_hdr.comp_mask	 = 0;
2091 	*sa_query = &query->sa_query;
2092 
2093 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
2094 	if (ret < 0)
2095 		goto err_free_mad;
2096 
2097 	return ret;
2098 
2099 err_free_mad:
2100 	*sa_query = NULL;
2101 	free_mad(&query->sa_query);
2102 
2103 err_free:
2104 	kfree(query);
2105 	return ret;
2106 }
2107 
2108 static void update_ib_cpi(struct work_struct *work)
2109 {
2110 	struct ib_sa_port *port =
2111 		container_of(work, struct ib_sa_port, ib_cpi_work.work);
2112 	struct ib_classport_info_context *cb_context;
2113 	unsigned long flags;
2114 	int ret;
2115 
2116 	/* If the classport info is valid, nothing
2117 	 * to do here.
2118 	 */
2119 	spin_lock_irqsave(&port->classport_lock, flags);
2120 	if (port->classport_info.valid) {
2121 		spin_unlock_irqrestore(&port->classport_lock, flags);
2122 		return;
2123 	}
2124 	spin_unlock_irqrestore(&port->classport_lock, flags);
2125 
2126 	cb_context = kmalloc(sizeof(*cb_context), GFP_KERNEL);
2127 	if (!cb_context)
2128 		goto err_nomem;
2129 
2130 	init_completion(&cb_context->done);
2131 
2132 	ret = ib_sa_classport_info_rec_query(port, 3000,
2133 					     ib_classportinfo_cb, cb_context,
2134 					     &cb_context->sa_query);
2135 	if (ret < 0)
2136 		goto free_cb_err;
2137 	wait_for_completion(&cb_context->done);
2138 free_cb_err:
2139 	kfree(cb_context);
2140 	spin_lock_irqsave(&port->classport_lock, flags);
2141 
2142 	/* If the classport info is still not valid, the query should have
2143 	 * failed for some reason. Retry issuing the query
2144 	 */
2145 	if (!port->classport_info.valid) {
2146 		port->classport_info.retry_cnt++;
2147 		if (port->classport_info.retry_cnt <=
2148 		    IB_SA_CPI_MAX_RETRY_CNT) {
2149 			unsigned long delay =
2150 				msecs_to_jiffies(IB_SA_CPI_RETRY_WAIT);
2151 
2152 			queue_delayed_work(ib_wq, &port->ib_cpi_work, delay);
2153 		}
2154 	}
2155 	spin_unlock_irqrestore(&port->classport_lock, flags);
2156 
2157 err_nomem:
2158 	return;
2159 }
2160 
2161 static void send_handler(struct ib_mad_agent *agent,
2162 			 struct ib_mad_send_wc *mad_send_wc)
2163 {
2164 	struct ib_sa_query *query = mad_send_wc->send_buf->context[0];
2165 	unsigned long flags;
2166 
2167 	if (query->callback)
2168 		switch (mad_send_wc->status) {
2169 		case IB_WC_SUCCESS:
2170 			/* No callback -- already got recv */
2171 			break;
2172 		case IB_WC_RESP_TIMEOUT_ERR:
2173 			query->callback(query, -ETIMEDOUT, NULL);
2174 			break;
2175 		case IB_WC_WR_FLUSH_ERR:
2176 			query->callback(query, -EINTR, NULL);
2177 			break;
2178 		default:
2179 			query->callback(query, -EIO, NULL);
2180 			break;
2181 		}
2182 
2183 	xa_lock_irqsave(&queries, flags);
2184 	__xa_erase(&queries, query->id);
2185 	xa_unlock_irqrestore(&queries, flags);
2186 
2187 	free_mad(query);
2188 	if (query->client)
2189 		ib_sa_client_put(query->client);
2190 	query->release(query);
2191 }
2192 
2193 static void recv_handler(struct ib_mad_agent *mad_agent,
2194 			 struct ib_mad_send_buf *send_buf,
2195 			 struct ib_mad_recv_wc *mad_recv_wc)
2196 {
2197 	struct ib_sa_query *query;
2198 
2199 	if (!send_buf)
2200 		return;
2201 
2202 	query = send_buf->context[0];
2203 	if (query->callback) {
2204 		if (mad_recv_wc->wc->status == IB_WC_SUCCESS)
2205 			query->callback(query,
2206 					mad_recv_wc->recv_buf.mad->mad_hdr.status ?
2207 					-EINVAL : 0,
2208 					(struct ib_sa_mad *) mad_recv_wc->recv_buf.mad);
2209 		else
2210 			query->callback(query, -EIO, NULL);
2211 	}
2212 
2213 	ib_free_recv_mad(mad_recv_wc);
2214 }
2215 
2216 static void update_sm_ah(struct work_struct *work)
2217 {
2218 	struct ib_sa_port *port =
2219 		container_of(work, struct ib_sa_port, update_task);
2220 	struct ib_sa_sm_ah *new_ah;
2221 	struct ib_port_attr port_attr;
2222 	struct rdma_ah_attr   ah_attr;
2223 	bool grh_required;
2224 
2225 	if (ib_query_port(port->agent->device, port->port_num, &port_attr)) {
2226 		pr_warn("Couldn't query port\n");
2227 		return;
2228 	}
2229 
2230 	new_ah = kmalloc(sizeof(*new_ah), GFP_KERNEL);
2231 	if (!new_ah)
2232 		return;
2233 
2234 	kref_init(&new_ah->ref);
2235 	new_ah->src_path_mask = (1 << port_attr.lmc) - 1;
2236 
2237 	new_ah->pkey_index = 0;
2238 	if (ib_find_pkey(port->agent->device, port->port_num,
2239 			 IB_DEFAULT_PKEY_FULL, &new_ah->pkey_index))
2240 		pr_err("Couldn't find index for default PKey\n");
2241 
2242 	memset(&ah_attr, 0, sizeof(ah_attr));
2243 	ah_attr.type = rdma_ah_find_type(port->agent->device,
2244 					 port->port_num);
2245 	rdma_ah_set_dlid(&ah_attr, port_attr.sm_lid);
2246 	rdma_ah_set_sl(&ah_attr, port_attr.sm_sl);
2247 	rdma_ah_set_port_num(&ah_attr, port->port_num);
2248 
2249 	grh_required = rdma_is_grh_required(port->agent->device,
2250 					    port->port_num);
2251 
2252 	/*
2253 	 * The OPA sm_lid of 0xFFFF needs special handling so that it can be
2254 	 * differentiated from a permissive LID of 0xFFFF.  We set the
2255 	 * grh_required flag here so the SA can program the DGID in the
2256 	 * address handle appropriately
2257 	 */
2258 	if (ah_attr.type == RDMA_AH_ATTR_TYPE_OPA &&
2259 	    (grh_required ||
2260 	     port_attr.sm_lid == be16_to_cpu(IB_LID_PERMISSIVE)))
2261 		rdma_ah_set_make_grd(&ah_attr, true);
2262 
2263 	if (ah_attr.type == RDMA_AH_ATTR_TYPE_IB && grh_required) {
2264 		rdma_ah_set_ah_flags(&ah_attr, IB_AH_GRH);
2265 		rdma_ah_set_subnet_prefix(&ah_attr,
2266 					  cpu_to_be64(port_attr.subnet_prefix));
2267 		rdma_ah_set_interface_id(&ah_attr,
2268 					 cpu_to_be64(IB_SA_WELL_KNOWN_GUID));
2269 	}
2270 
2271 	new_ah->ah = rdma_create_ah(port->agent->qp->pd, &ah_attr,
2272 				    RDMA_CREATE_AH_SLEEPABLE);
2273 	if (IS_ERR(new_ah->ah)) {
2274 		pr_warn("Couldn't create new SM AH\n");
2275 		kfree(new_ah);
2276 		return;
2277 	}
2278 
2279 	spin_lock_irq(&port->ah_lock);
2280 	if (port->sm_ah)
2281 		kref_put(&port->sm_ah->ref, free_sm_ah);
2282 	port->sm_ah = new_ah;
2283 	spin_unlock_irq(&port->ah_lock);
2284 }
2285 
2286 static void ib_sa_event(struct ib_event_handler *handler,
2287 			struct ib_event *event)
2288 {
2289 	if (event->event == IB_EVENT_PORT_ERR    ||
2290 	    event->event == IB_EVENT_PORT_ACTIVE ||
2291 	    event->event == IB_EVENT_LID_CHANGE  ||
2292 	    event->event == IB_EVENT_PKEY_CHANGE ||
2293 	    event->event == IB_EVENT_SM_CHANGE   ||
2294 	    event->event == IB_EVENT_CLIENT_REREGISTER) {
2295 		unsigned long flags;
2296 		struct ib_sa_device *sa_dev =
2297 			container_of(handler, typeof(*sa_dev), event_handler);
2298 		u8 port_num = event->element.port_num - sa_dev->start_port;
2299 		struct ib_sa_port *port = &sa_dev->port[port_num];
2300 
2301 		if (!rdma_cap_ib_sa(handler->device, port->port_num))
2302 			return;
2303 
2304 		spin_lock_irqsave(&port->ah_lock, flags);
2305 		if (port->sm_ah)
2306 			kref_put(&port->sm_ah->ref, free_sm_ah);
2307 		port->sm_ah = NULL;
2308 		spin_unlock_irqrestore(&port->ah_lock, flags);
2309 
2310 		if (event->event == IB_EVENT_SM_CHANGE ||
2311 		    event->event == IB_EVENT_CLIENT_REREGISTER ||
2312 		    event->event == IB_EVENT_LID_CHANGE ||
2313 		    event->event == IB_EVENT_PORT_ACTIVE) {
2314 			unsigned long delay =
2315 				msecs_to_jiffies(IB_SA_CPI_RETRY_WAIT);
2316 
2317 			spin_lock_irqsave(&port->classport_lock, flags);
2318 			port->classport_info.valid = false;
2319 			port->classport_info.retry_cnt = 0;
2320 			spin_unlock_irqrestore(&port->classport_lock, flags);
2321 			queue_delayed_work(ib_wq,
2322 					   &port->ib_cpi_work, delay);
2323 		}
2324 		queue_work(ib_wq, &sa_dev->port[port_num].update_task);
2325 	}
2326 }
2327 
2328 static void ib_sa_add_one(struct ib_device *device)
2329 {
2330 	struct ib_sa_device *sa_dev;
2331 	int s, e, i;
2332 	int count = 0;
2333 
2334 	s = rdma_start_port(device);
2335 	e = rdma_end_port(device);
2336 
2337 	sa_dev = kzalloc(struct_size(sa_dev, port, e - s + 1), GFP_KERNEL);
2338 	if (!sa_dev)
2339 		return;
2340 
2341 	sa_dev->start_port = s;
2342 	sa_dev->end_port   = e;
2343 
2344 	for (i = 0; i <= e - s; ++i) {
2345 		spin_lock_init(&sa_dev->port[i].ah_lock);
2346 		if (!rdma_cap_ib_sa(device, i + 1))
2347 			continue;
2348 
2349 		sa_dev->port[i].sm_ah    = NULL;
2350 		sa_dev->port[i].port_num = i + s;
2351 
2352 		spin_lock_init(&sa_dev->port[i].classport_lock);
2353 		sa_dev->port[i].classport_info.valid = false;
2354 
2355 		sa_dev->port[i].agent =
2356 			ib_register_mad_agent(device, i + s, IB_QPT_GSI,
2357 					      NULL, 0, send_handler,
2358 					      recv_handler, sa_dev, 0);
2359 		if (IS_ERR(sa_dev->port[i].agent))
2360 			goto err;
2361 
2362 		INIT_WORK(&sa_dev->port[i].update_task, update_sm_ah);
2363 		INIT_DELAYED_WORK(&sa_dev->port[i].ib_cpi_work,
2364 				  update_ib_cpi);
2365 
2366 		count++;
2367 	}
2368 
2369 	if (!count)
2370 		goto free;
2371 
2372 	ib_set_client_data(device, &sa_client, sa_dev);
2373 
2374 	/*
2375 	 * We register our event handler after everything is set up,
2376 	 * and then update our cached info after the event handler is
2377 	 * registered to avoid any problems if a port changes state
2378 	 * during our initialization.
2379 	 */
2380 
2381 	INIT_IB_EVENT_HANDLER(&sa_dev->event_handler, device, ib_sa_event);
2382 	ib_register_event_handler(&sa_dev->event_handler);
2383 
2384 	for (i = 0; i <= e - s; ++i) {
2385 		if (rdma_cap_ib_sa(device, i + 1))
2386 			update_sm_ah(&sa_dev->port[i].update_task);
2387 	}
2388 
2389 	return;
2390 
2391 err:
2392 	while (--i >= 0) {
2393 		if (rdma_cap_ib_sa(device, i + 1))
2394 			ib_unregister_mad_agent(sa_dev->port[i].agent);
2395 	}
2396 free:
2397 	kfree(sa_dev);
2398 	return;
2399 }
2400 
2401 static void ib_sa_remove_one(struct ib_device *device, void *client_data)
2402 {
2403 	struct ib_sa_device *sa_dev = client_data;
2404 	int i;
2405 
2406 	if (!sa_dev)
2407 		return;
2408 
2409 	ib_unregister_event_handler(&sa_dev->event_handler);
2410 	flush_workqueue(ib_wq);
2411 
2412 	for (i = 0; i <= sa_dev->end_port - sa_dev->start_port; ++i) {
2413 		if (rdma_cap_ib_sa(device, i + 1)) {
2414 			cancel_delayed_work_sync(&sa_dev->port[i].ib_cpi_work);
2415 			ib_unregister_mad_agent(sa_dev->port[i].agent);
2416 			if (sa_dev->port[i].sm_ah)
2417 				kref_put(&sa_dev->port[i].sm_ah->ref, free_sm_ah);
2418 		}
2419 
2420 	}
2421 
2422 	kfree(sa_dev);
2423 }
2424 
2425 int ib_sa_init(void)
2426 {
2427 	int ret;
2428 
2429 	get_random_bytes(&tid, sizeof tid);
2430 
2431 	atomic_set(&ib_nl_sa_request_seq, 0);
2432 
2433 	ret = ib_register_client(&sa_client);
2434 	if (ret) {
2435 		pr_err("Couldn't register ib_sa client\n");
2436 		goto err1;
2437 	}
2438 
2439 	ret = mcast_init();
2440 	if (ret) {
2441 		pr_err("Couldn't initialize multicast handling\n");
2442 		goto err2;
2443 	}
2444 
2445 	ib_nl_wq = alloc_ordered_workqueue("ib_nl_sa_wq", WQ_MEM_RECLAIM);
2446 	if (!ib_nl_wq) {
2447 		ret = -ENOMEM;
2448 		goto err3;
2449 	}
2450 
2451 	INIT_DELAYED_WORK(&ib_nl_timed_work, ib_nl_request_timeout);
2452 
2453 	return 0;
2454 
2455 err3:
2456 	mcast_cleanup();
2457 err2:
2458 	ib_unregister_client(&sa_client);
2459 err1:
2460 	return ret;
2461 }
2462 
2463 void ib_sa_cleanup(void)
2464 {
2465 	cancel_delayed_work(&ib_nl_timed_work);
2466 	flush_workqueue(ib_nl_wq);
2467 	destroy_workqueue(ib_nl_wq);
2468 	mcast_cleanup();
2469 	ib_unregister_client(&sa_client);
2470 	WARN_ON(!xa_empty(&queries));
2471 }
2472