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, query->port->agent->device->name,
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 	int ret = 0;
839 	struct ib_sa_mad *mad;
840 	int len;
841 
842 	mad = query->mad_buf->mad;
843 	len = ib_nl_get_path_rec_attrs_len(mad->sa_hdr.comp_mask);
844 	if (len <= 0)
845 		return -EMSGSIZE;
846 
847 	skb = nlmsg_new(len, gfp_mask);
848 	if (!skb)
849 		return -ENOMEM;
850 
851 	/* Put nlmsg header only for now */
852 	data = ibnl_put_msg(skb, &nlh, query->seq, 0, RDMA_NL_LS,
853 			    RDMA_NL_LS_OP_RESOLVE, NLM_F_REQUEST);
854 	if (!data) {
855 		nlmsg_free(skb);
856 		return -EMSGSIZE;
857 	}
858 
859 	/* Add attributes */
860 	ib_nl_set_path_rec_attrs(skb, query);
861 
862 	/* Repair the nlmsg header length */
863 	nlmsg_end(skb, nlh);
864 
865 	ret = rdma_nl_multicast(skb, RDMA_NL_GROUP_LS, gfp_mask);
866 	if (!ret)
867 		ret = len;
868 	else
869 		ret = 0;
870 
871 	return ret;
872 }
873 
874 static int ib_nl_make_request(struct ib_sa_query *query, gfp_t gfp_mask)
875 {
876 	unsigned long flags;
877 	unsigned long delay;
878 	int ret;
879 
880 	INIT_LIST_HEAD(&query->list);
881 	query->seq = (u32)atomic_inc_return(&ib_nl_sa_request_seq);
882 
883 	/* Put the request on the list first.*/
884 	spin_lock_irqsave(&ib_nl_request_lock, flags);
885 	delay = msecs_to_jiffies(sa_local_svc_timeout_ms);
886 	query->timeout = delay + jiffies;
887 	list_add_tail(&query->list, &ib_nl_request_list);
888 	/* Start the timeout if this is the only request */
889 	if (ib_nl_request_list.next == &query->list)
890 		queue_delayed_work(ib_nl_wq, &ib_nl_timed_work, delay);
891 	spin_unlock_irqrestore(&ib_nl_request_lock, flags);
892 
893 	ret = ib_nl_send_msg(query, gfp_mask);
894 	if (ret <= 0) {
895 		ret = -EIO;
896 		/* Remove the request */
897 		spin_lock_irqsave(&ib_nl_request_lock, flags);
898 		list_del(&query->list);
899 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
900 	} else {
901 		ret = 0;
902 	}
903 
904 	return ret;
905 }
906 
907 static int ib_nl_cancel_request(struct ib_sa_query *query)
908 {
909 	unsigned long flags;
910 	struct ib_sa_query *wait_query;
911 	int found = 0;
912 
913 	spin_lock_irqsave(&ib_nl_request_lock, flags);
914 	list_for_each_entry(wait_query, &ib_nl_request_list, list) {
915 		/* Let the timeout to take care of the callback */
916 		if (query == wait_query) {
917 			query->flags |= IB_SA_CANCEL;
918 			query->timeout = jiffies;
919 			list_move(&query->list, &ib_nl_request_list);
920 			found = 1;
921 			mod_delayed_work(ib_nl_wq, &ib_nl_timed_work, 1);
922 			break;
923 		}
924 	}
925 	spin_unlock_irqrestore(&ib_nl_request_lock, flags);
926 
927 	return found;
928 }
929 
930 static void send_handler(struct ib_mad_agent *agent,
931 			 struct ib_mad_send_wc *mad_send_wc);
932 
933 static void ib_nl_process_good_resolve_rsp(struct ib_sa_query *query,
934 					   const struct nlmsghdr *nlh)
935 {
936 	struct ib_mad_send_wc mad_send_wc;
937 	struct ib_sa_mad *mad = NULL;
938 	const struct nlattr *head, *curr;
939 	struct ib_path_rec_data  *rec;
940 	int len, rem;
941 	u32 mask = 0;
942 	int status = -EIO;
943 
944 	if (query->callback) {
945 		head = (const struct nlattr *) nlmsg_data(nlh);
946 		len = nlmsg_len(nlh);
947 		switch (query->path_use) {
948 		case LS_RESOLVE_PATH_USE_UNIDIRECTIONAL:
949 			mask = IB_PATH_PRIMARY | IB_PATH_OUTBOUND;
950 			break;
951 
952 		case LS_RESOLVE_PATH_USE_ALL:
953 		case LS_RESOLVE_PATH_USE_GMP:
954 		default:
955 			mask = IB_PATH_PRIMARY | IB_PATH_GMP |
956 				IB_PATH_BIDIRECTIONAL;
957 			break;
958 		}
959 		nla_for_each_attr(curr, head, len, rem) {
960 			if (curr->nla_type == LS_NLA_TYPE_PATH_RECORD) {
961 				rec = nla_data(curr);
962 				/*
963 				 * Get the first one. In the future, we may
964 				 * need to get up to 6 pathrecords.
965 				 */
966 				if ((rec->flags & mask) == mask) {
967 					mad = query->mad_buf->mad;
968 					mad->mad_hdr.method |=
969 						IB_MGMT_METHOD_RESP;
970 					memcpy(mad->data, rec->path_rec,
971 					       sizeof(rec->path_rec));
972 					status = 0;
973 					break;
974 				}
975 			}
976 		}
977 		query->callback(query, status, mad);
978 	}
979 
980 	mad_send_wc.send_buf = query->mad_buf;
981 	mad_send_wc.status = IB_WC_SUCCESS;
982 	send_handler(query->mad_buf->mad_agent, &mad_send_wc);
983 }
984 
985 static void ib_nl_request_timeout(struct work_struct *work)
986 {
987 	unsigned long flags;
988 	struct ib_sa_query *query;
989 	unsigned long delay;
990 	struct ib_mad_send_wc mad_send_wc;
991 	int ret;
992 
993 	spin_lock_irqsave(&ib_nl_request_lock, flags);
994 	while (!list_empty(&ib_nl_request_list)) {
995 		query = list_entry(ib_nl_request_list.next,
996 				   struct ib_sa_query, list);
997 
998 		if (time_after(query->timeout, jiffies)) {
999 			delay = query->timeout - jiffies;
1000 			if ((long)delay <= 0)
1001 				delay = 1;
1002 			queue_delayed_work(ib_nl_wq, &ib_nl_timed_work, delay);
1003 			break;
1004 		}
1005 
1006 		list_del(&query->list);
1007 		ib_sa_disable_local_svc(query);
1008 		/* Hold the lock to protect against query cancellation */
1009 		if (ib_sa_query_cancelled(query))
1010 			ret = -1;
1011 		else
1012 			ret = ib_post_send_mad(query->mad_buf, NULL);
1013 		if (ret) {
1014 			mad_send_wc.send_buf = query->mad_buf;
1015 			mad_send_wc.status = IB_WC_WR_FLUSH_ERR;
1016 			spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1017 			send_handler(query->port->agent, &mad_send_wc);
1018 			spin_lock_irqsave(&ib_nl_request_lock, flags);
1019 		}
1020 	}
1021 	spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1022 }
1023 
1024 int ib_nl_handle_set_timeout(struct sk_buff *skb,
1025 			     struct nlmsghdr *nlh,
1026 			     struct netlink_ext_ack *extack)
1027 {
1028 	int timeout, delta, abs_delta;
1029 	const struct nlattr *attr;
1030 	unsigned long flags;
1031 	struct ib_sa_query *query;
1032 	long delay = 0;
1033 	struct nlattr *tb[LS_NLA_TYPE_MAX];
1034 	int ret;
1035 
1036 	if (!(nlh->nlmsg_flags & NLM_F_REQUEST) ||
1037 	    !(NETLINK_CB(skb).sk))
1038 		return -EPERM;
1039 
1040 	ret = nla_parse(tb, LS_NLA_TYPE_MAX - 1, nlmsg_data(nlh),
1041 			nlmsg_len(nlh), ib_nl_policy, NULL);
1042 	attr = (const struct nlattr *)tb[LS_NLA_TYPE_TIMEOUT];
1043 	if (ret || !attr)
1044 		goto settimeout_out;
1045 
1046 	timeout = *(int *) nla_data(attr);
1047 	if (timeout < IB_SA_LOCAL_SVC_TIMEOUT_MIN)
1048 		timeout = IB_SA_LOCAL_SVC_TIMEOUT_MIN;
1049 	if (timeout > IB_SA_LOCAL_SVC_TIMEOUT_MAX)
1050 		timeout = IB_SA_LOCAL_SVC_TIMEOUT_MAX;
1051 
1052 	delta = timeout - sa_local_svc_timeout_ms;
1053 	if (delta < 0)
1054 		abs_delta = -delta;
1055 	else
1056 		abs_delta = delta;
1057 
1058 	if (delta != 0) {
1059 		spin_lock_irqsave(&ib_nl_request_lock, flags);
1060 		sa_local_svc_timeout_ms = timeout;
1061 		list_for_each_entry(query, &ib_nl_request_list, list) {
1062 			if (delta < 0 && abs_delta > query->timeout)
1063 				query->timeout = 0;
1064 			else
1065 				query->timeout += delta;
1066 
1067 			/* Get the new delay from the first entry */
1068 			if (!delay) {
1069 				delay = query->timeout - jiffies;
1070 				if (delay <= 0)
1071 					delay = 1;
1072 			}
1073 		}
1074 		if (delay)
1075 			mod_delayed_work(ib_nl_wq, &ib_nl_timed_work,
1076 					 (unsigned long)delay);
1077 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1078 	}
1079 
1080 settimeout_out:
1081 	return skb->len;
1082 }
1083 
1084 static inline int ib_nl_is_good_resolve_resp(const struct nlmsghdr *nlh)
1085 {
1086 	struct nlattr *tb[LS_NLA_TYPE_MAX];
1087 	int ret;
1088 
1089 	if (nlh->nlmsg_flags & RDMA_NL_LS_F_ERR)
1090 		return 0;
1091 
1092 	ret = nla_parse(tb, LS_NLA_TYPE_MAX - 1, nlmsg_data(nlh),
1093 			nlmsg_len(nlh), ib_nl_policy, NULL);
1094 	if (ret)
1095 		return 0;
1096 
1097 	return 1;
1098 }
1099 
1100 int ib_nl_handle_resolve_resp(struct sk_buff *skb,
1101 			      struct nlmsghdr *nlh,
1102 			      struct netlink_ext_ack *extack)
1103 {
1104 	unsigned long flags;
1105 	struct ib_sa_query *query;
1106 	struct ib_mad_send_buf *send_buf;
1107 	struct ib_mad_send_wc mad_send_wc;
1108 	int found = 0;
1109 	int ret;
1110 
1111 	if ((nlh->nlmsg_flags & NLM_F_REQUEST) ||
1112 	    !(NETLINK_CB(skb).sk))
1113 		return -EPERM;
1114 
1115 	spin_lock_irqsave(&ib_nl_request_lock, flags);
1116 	list_for_each_entry(query, &ib_nl_request_list, list) {
1117 		/*
1118 		 * If the query is cancelled, let the timeout routine
1119 		 * take care of it.
1120 		 */
1121 		if (nlh->nlmsg_seq == query->seq) {
1122 			found = !ib_sa_query_cancelled(query);
1123 			if (found)
1124 				list_del(&query->list);
1125 			break;
1126 		}
1127 	}
1128 
1129 	if (!found) {
1130 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1131 		goto resp_out;
1132 	}
1133 
1134 	send_buf = query->mad_buf;
1135 
1136 	if (!ib_nl_is_good_resolve_resp(nlh)) {
1137 		/* if the result is a failure, send out the packet via IB */
1138 		ib_sa_disable_local_svc(query);
1139 		ret = ib_post_send_mad(query->mad_buf, NULL);
1140 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1141 		if (ret) {
1142 			mad_send_wc.send_buf = send_buf;
1143 			mad_send_wc.status = IB_WC_GENERAL_ERR;
1144 			send_handler(query->port->agent, &mad_send_wc);
1145 		}
1146 	} else {
1147 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1148 		ib_nl_process_good_resolve_rsp(query, nlh);
1149 	}
1150 
1151 resp_out:
1152 	return skb->len;
1153 }
1154 
1155 static void free_sm_ah(struct kref *kref)
1156 {
1157 	struct ib_sa_sm_ah *sm_ah = container_of(kref, struct ib_sa_sm_ah, ref);
1158 
1159 	rdma_destroy_ah(sm_ah->ah);
1160 	kfree(sm_ah);
1161 }
1162 
1163 void ib_sa_register_client(struct ib_sa_client *client)
1164 {
1165 	atomic_set(&client->users, 1);
1166 	init_completion(&client->comp);
1167 }
1168 EXPORT_SYMBOL(ib_sa_register_client);
1169 
1170 void ib_sa_unregister_client(struct ib_sa_client *client)
1171 {
1172 	ib_sa_client_put(client);
1173 	wait_for_completion(&client->comp);
1174 }
1175 EXPORT_SYMBOL(ib_sa_unregister_client);
1176 
1177 /**
1178  * ib_sa_cancel_query - try to cancel an SA query
1179  * @id:ID of query to cancel
1180  * @query:query pointer to cancel
1181  *
1182  * Try to cancel an SA query.  If the id and query don't match up or
1183  * the query has already completed, nothing is done.  Otherwise the
1184  * query is canceled and will complete with a status of -EINTR.
1185  */
1186 void ib_sa_cancel_query(int id, struct ib_sa_query *query)
1187 {
1188 	unsigned long flags;
1189 	struct ib_mad_agent *agent;
1190 	struct ib_mad_send_buf *mad_buf;
1191 
1192 	spin_lock_irqsave(&idr_lock, flags);
1193 	if (idr_find(&query_idr, id) != query) {
1194 		spin_unlock_irqrestore(&idr_lock, flags);
1195 		return;
1196 	}
1197 	agent = query->port->agent;
1198 	mad_buf = query->mad_buf;
1199 	spin_unlock_irqrestore(&idr_lock, flags);
1200 
1201 	/*
1202 	 * If the query is still on the netlink request list, schedule
1203 	 * it to be cancelled by the timeout routine. Otherwise, it has been
1204 	 * sent to the MAD layer and has to be cancelled from there.
1205 	 */
1206 	if (!ib_nl_cancel_request(query))
1207 		ib_cancel_mad(agent, mad_buf);
1208 }
1209 EXPORT_SYMBOL(ib_sa_cancel_query);
1210 
1211 static u8 get_src_path_mask(struct ib_device *device, u8 port_num)
1212 {
1213 	struct ib_sa_device *sa_dev;
1214 	struct ib_sa_port   *port;
1215 	unsigned long flags;
1216 	u8 src_path_mask;
1217 
1218 	sa_dev = ib_get_client_data(device, &sa_client);
1219 	if (!sa_dev)
1220 		return 0x7f;
1221 
1222 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1223 	spin_lock_irqsave(&port->ah_lock, flags);
1224 	src_path_mask = port->sm_ah ? port->sm_ah->src_path_mask : 0x7f;
1225 	spin_unlock_irqrestore(&port->ah_lock, flags);
1226 
1227 	return src_path_mask;
1228 }
1229 
1230 static int
1231 roce_resolve_route_from_path(struct ib_device *device, u8 port_num,
1232 			     struct sa_path_rec *rec)
1233 {
1234 	struct net_device *resolved_dev;
1235 	struct net_device *ndev;
1236 	struct net_device *idev;
1237 	struct rdma_dev_addr dev_addr = {
1238 		.bound_dev_if = ((sa_path_get_ifindex(rec) >= 0) ?
1239 				 sa_path_get_ifindex(rec) : 0),
1240 		.net = sa_path_get_ndev(rec) ?
1241 			sa_path_get_ndev(rec) :
1242 			&init_net
1243 	};
1244 	union {
1245 		struct sockaddr     _sockaddr;
1246 		struct sockaddr_in  _sockaddr_in;
1247 		struct sockaddr_in6 _sockaddr_in6;
1248 	} sgid_addr, dgid_addr;
1249 	int ret;
1250 
1251 	if (rec->roce.route_resolved)
1252 		return 0;
1253 
1254 	if (!device->get_netdev)
1255 		return -EOPNOTSUPP;
1256 
1257 	rdma_gid2ip(&sgid_addr._sockaddr, &rec->sgid);
1258 	rdma_gid2ip(&dgid_addr._sockaddr, &rec->dgid);
1259 
1260 	/* validate the route */
1261 	ret = rdma_resolve_ip_route(&sgid_addr._sockaddr,
1262 				    &dgid_addr._sockaddr, &dev_addr);
1263 	if (ret)
1264 		return ret;
1265 
1266 	if ((dev_addr.network == RDMA_NETWORK_IPV4 ||
1267 	     dev_addr.network == RDMA_NETWORK_IPV6) &&
1268 	    rec->rec_type != SA_PATH_REC_TYPE_ROCE_V2)
1269 		return -EINVAL;
1270 
1271 	idev = device->get_netdev(device, port_num);
1272 	if (!idev)
1273 		return -ENODEV;
1274 
1275 	resolved_dev = dev_get_by_index(dev_addr.net,
1276 					dev_addr.bound_dev_if);
1277 	if (!resolved_dev) {
1278 		ret = -ENODEV;
1279 		goto done;
1280 	}
1281 	ndev = ib_get_ndev_from_path(rec);
1282 	rcu_read_lock();
1283 	if ((ndev && ndev != resolved_dev) ||
1284 	    (resolved_dev != idev &&
1285 	     !rdma_is_upper_dev_rcu(idev, resolved_dev)))
1286 		ret = -EHOSTUNREACH;
1287 	rcu_read_unlock();
1288 	dev_put(resolved_dev);
1289 	if (ndev)
1290 		dev_put(ndev);
1291 done:
1292 	dev_put(idev);
1293 	if (!ret)
1294 		rec->roce.route_resolved = true;
1295 	return ret;
1296 }
1297 
1298 static int init_ah_attr_grh_fields(struct ib_device *device, u8 port_num,
1299 				   struct sa_path_rec *rec,
1300 				   struct rdma_ah_attr *ah_attr)
1301 {
1302 	enum ib_gid_type type = sa_conv_pathrec_to_gid_type(rec);
1303 	struct net_device *ndev;
1304 	u16 gid_index;
1305 	int ret;
1306 
1307 	ndev = ib_get_ndev_from_path(rec);
1308 	ret = ib_find_cached_gid_by_port(device, &rec->sgid, type,
1309 					 port_num, ndev, &gid_index);
1310 	if (ndev)
1311 		dev_put(ndev);
1312 	if (ret)
1313 		return ret;
1314 
1315 	rdma_ah_set_grh(ah_attr, &rec->dgid,
1316 			be32_to_cpu(rec->flow_label),
1317 			gid_index, rec->hop_limit,
1318 			rec->traffic_class);
1319 	return 0;
1320 }
1321 
1322 int ib_init_ah_attr_from_path(struct ib_device *device, u8 port_num,
1323 			      struct sa_path_rec *rec,
1324 			      struct rdma_ah_attr *ah_attr)
1325 {
1326 	int ret = 0;
1327 
1328 	memset(ah_attr, 0, sizeof(*ah_attr));
1329 	ah_attr->type = rdma_ah_find_type(device, port_num);
1330 	rdma_ah_set_sl(ah_attr, rec->sl);
1331 	rdma_ah_set_port_num(ah_attr, port_num);
1332 	rdma_ah_set_static_rate(ah_attr, rec->rate);
1333 
1334 	if (sa_path_is_roce(rec)) {
1335 		ret = roce_resolve_route_from_path(device, port_num, rec);
1336 		if (ret)
1337 			return ret;
1338 
1339 		memcpy(ah_attr->roce.dmac, sa_path_get_dmac(rec), ETH_ALEN);
1340 	} else {
1341 		rdma_ah_set_dlid(ah_attr, be32_to_cpu(sa_path_get_dlid(rec)));
1342 		if (sa_path_is_opa(rec) &&
1343 		    rdma_ah_get_dlid(ah_attr) == be16_to_cpu(IB_LID_PERMISSIVE))
1344 			rdma_ah_set_make_grd(ah_attr, true);
1345 
1346 		rdma_ah_set_path_bits(ah_attr,
1347 				      be32_to_cpu(sa_path_get_slid(rec)) &
1348 				      get_src_path_mask(device, port_num));
1349 	}
1350 
1351 	if (rec->hop_limit > 0 || sa_path_is_roce(rec))
1352 		ret = init_ah_attr_grh_fields(device, port_num, rec, ah_attr);
1353 	return ret;
1354 }
1355 EXPORT_SYMBOL(ib_init_ah_attr_from_path);
1356 
1357 static int alloc_mad(struct ib_sa_query *query, gfp_t gfp_mask)
1358 {
1359 	struct rdma_ah_attr ah_attr;
1360 	unsigned long flags;
1361 
1362 	spin_lock_irqsave(&query->port->ah_lock, flags);
1363 	if (!query->port->sm_ah) {
1364 		spin_unlock_irqrestore(&query->port->ah_lock, flags);
1365 		return -EAGAIN;
1366 	}
1367 	kref_get(&query->port->sm_ah->ref);
1368 	query->sm_ah = query->port->sm_ah;
1369 	spin_unlock_irqrestore(&query->port->ah_lock, flags);
1370 
1371 	/*
1372 	 * Always check if sm_ah has valid dlid assigned,
1373 	 * before querying for class port info
1374 	 */
1375 	if ((rdma_query_ah(query->sm_ah->ah, &ah_attr) < 0) ||
1376 	    !rdma_is_valid_unicast_lid(&ah_attr)) {
1377 		kref_put(&query->sm_ah->ref, free_sm_ah);
1378 		return -EAGAIN;
1379 	}
1380 	query->mad_buf = ib_create_send_mad(query->port->agent, 1,
1381 					    query->sm_ah->pkey_index,
1382 					    0, IB_MGMT_SA_HDR, IB_MGMT_SA_DATA,
1383 					    gfp_mask,
1384 					    ((query->flags & IB_SA_QUERY_OPA) ?
1385 					     OPA_MGMT_BASE_VERSION :
1386 					     IB_MGMT_BASE_VERSION));
1387 	if (IS_ERR(query->mad_buf)) {
1388 		kref_put(&query->sm_ah->ref, free_sm_ah);
1389 		return -ENOMEM;
1390 	}
1391 
1392 	query->mad_buf->ah = query->sm_ah->ah;
1393 
1394 	return 0;
1395 }
1396 
1397 static void free_mad(struct ib_sa_query *query)
1398 {
1399 	ib_free_send_mad(query->mad_buf);
1400 	kref_put(&query->sm_ah->ref, free_sm_ah);
1401 }
1402 
1403 static void init_mad(struct ib_sa_query *query, struct ib_mad_agent *agent)
1404 {
1405 	struct ib_sa_mad *mad = query->mad_buf->mad;
1406 	unsigned long flags;
1407 
1408 	memset(mad, 0, sizeof *mad);
1409 
1410 	if (query->flags & IB_SA_QUERY_OPA) {
1411 		mad->mad_hdr.base_version  = OPA_MGMT_BASE_VERSION;
1412 		mad->mad_hdr.class_version = OPA_SA_CLASS_VERSION;
1413 	} else {
1414 		mad->mad_hdr.base_version  = IB_MGMT_BASE_VERSION;
1415 		mad->mad_hdr.class_version = IB_SA_CLASS_VERSION;
1416 	}
1417 	mad->mad_hdr.mgmt_class    = IB_MGMT_CLASS_SUBN_ADM;
1418 	spin_lock_irqsave(&tid_lock, flags);
1419 	mad->mad_hdr.tid           =
1420 		cpu_to_be64(((u64) agent->hi_tid) << 32 | tid++);
1421 	spin_unlock_irqrestore(&tid_lock, flags);
1422 }
1423 
1424 static int send_mad(struct ib_sa_query *query, int timeout_ms, gfp_t gfp_mask)
1425 {
1426 	bool preload = gfpflags_allow_blocking(gfp_mask);
1427 	unsigned long flags;
1428 	int ret, id;
1429 
1430 	if (preload)
1431 		idr_preload(gfp_mask);
1432 	spin_lock_irqsave(&idr_lock, flags);
1433 
1434 	id = idr_alloc(&query_idr, query, 0, 0, GFP_NOWAIT);
1435 
1436 	spin_unlock_irqrestore(&idr_lock, flags);
1437 	if (preload)
1438 		idr_preload_end();
1439 	if (id < 0)
1440 		return id;
1441 
1442 	query->mad_buf->timeout_ms  = timeout_ms;
1443 	query->mad_buf->context[0] = query;
1444 	query->id = id;
1445 
1446 	if ((query->flags & IB_SA_ENABLE_LOCAL_SERVICE) &&
1447 	    (!(query->flags & IB_SA_QUERY_OPA))) {
1448 		if (!rdma_nl_chk_listeners(RDMA_NL_GROUP_LS)) {
1449 			if (!ib_nl_make_request(query, gfp_mask))
1450 				return id;
1451 		}
1452 		ib_sa_disable_local_svc(query);
1453 	}
1454 
1455 	ret = ib_post_send_mad(query->mad_buf, NULL);
1456 	if (ret) {
1457 		spin_lock_irqsave(&idr_lock, flags);
1458 		idr_remove(&query_idr, id);
1459 		spin_unlock_irqrestore(&idr_lock, flags);
1460 	}
1461 
1462 	/*
1463 	 * It's not safe to dereference query any more, because the
1464 	 * send may already have completed and freed the query in
1465 	 * another context.
1466 	 */
1467 	return ret ? ret : id;
1468 }
1469 
1470 void ib_sa_unpack_path(void *attribute, struct sa_path_rec *rec)
1471 {
1472 	ib_unpack(path_rec_table, ARRAY_SIZE(path_rec_table), attribute, rec);
1473 }
1474 EXPORT_SYMBOL(ib_sa_unpack_path);
1475 
1476 void ib_sa_pack_path(struct sa_path_rec *rec, void *attribute)
1477 {
1478 	ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table), rec, attribute);
1479 }
1480 EXPORT_SYMBOL(ib_sa_pack_path);
1481 
1482 static bool ib_sa_opa_pathrecord_support(struct ib_sa_client *client,
1483 					 struct ib_device *device,
1484 					 u8 port_num)
1485 {
1486 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1487 	struct ib_sa_port *port;
1488 	unsigned long flags;
1489 	bool ret = false;
1490 
1491 	if (!sa_dev)
1492 		return ret;
1493 
1494 	port = &sa_dev->port[port_num - sa_dev->start_port];
1495 	spin_lock_irqsave(&port->classport_lock, flags);
1496 	if (!port->classport_info.valid)
1497 		goto ret;
1498 
1499 	if (port->classport_info.data.type == RDMA_CLASS_PORT_INFO_OPA)
1500 		ret = opa_get_cpi_capmask2(&port->classport_info.data.opa) &
1501 			OPA_CLASS_PORT_INFO_PR_SUPPORT;
1502 ret:
1503 	spin_unlock_irqrestore(&port->classport_lock, flags);
1504 	return ret;
1505 }
1506 
1507 enum opa_pr_supported {
1508 	PR_NOT_SUPPORTED,
1509 	PR_OPA_SUPPORTED,
1510 	PR_IB_SUPPORTED
1511 };
1512 
1513 /**
1514  * Check if current PR query can be an OPA query.
1515  * Retuns PR_NOT_SUPPORTED if a path record query is not
1516  * possible, PR_OPA_SUPPORTED if an OPA path record query
1517  * is possible and PR_IB_SUPPORTED if an IB path record
1518  * query is possible.
1519  */
1520 static int opa_pr_query_possible(struct ib_sa_client *client,
1521 				 struct ib_device *device,
1522 				 u8 port_num,
1523 				 struct sa_path_rec *rec)
1524 {
1525 	struct ib_port_attr port_attr;
1526 
1527 	if (ib_query_port(device, port_num, &port_attr))
1528 		return PR_NOT_SUPPORTED;
1529 
1530 	if (ib_sa_opa_pathrecord_support(client, device, port_num))
1531 		return PR_OPA_SUPPORTED;
1532 
1533 	if (port_attr.lid >= be16_to_cpu(IB_MULTICAST_LID_BASE))
1534 		return PR_NOT_SUPPORTED;
1535 	else
1536 		return PR_IB_SUPPORTED;
1537 }
1538 
1539 static void ib_sa_path_rec_callback(struct ib_sa_query *sa_query,
1540 				    int status,
1541 				    struct ib_sa_mad *mad)
1542 {
1543 	struct ib_sa_path_query *query =
1544 		container_of(sa_query, struct ib_sa_path_query, sa_query);
1545 
1546 	if (mad) {
1547 		struct sa_path_rec rec;
1548 
1549 		if (sa_query->flags & IB_SA_QUERY_OPA) {
1550 			ib_unpack(opa_path_rec_table,
1551 				  ARRAY_SIZE(opa_path_rec_table),
1552 				  mad->data, &rec);
1553 			rec.rec_type = SA_PATH_REC_TYPE_OPA;
1554 			query->callback(status, &rec, query->context);
1555 		} else {
1556 			ib_unpack(path_rec_table,
1557 				  ARRAY_SIZE(path_rec_table),
1558 				  mad->data, &rec);
1559 			rec.rec_type = SA_PATH_REC_TYPE_IB;
1560 			sa_path_set_ndev(&rec, NULL);
1561 			sa_path_set_ifindex(&rec, 0);
1562 			sa_path_set_dmac_zero(&rec);
1563 
1564 			if (query->conv_pr) {
1565 				struct sa_path_rec opa;
1566 
1567 				memset(&opa, 0, sizeof(struct sa_path_rec));
1568 				sa_convert_path_ib_to_opa(&opa, &rec);
1569 				query->callback(status, &opa, query->context);
1570 			} else {
1571 				query->callback(status, &rec, query->context);
1572 			}
1573 		}
1574 	} else
1575 		query->callback(status, NULL, query->context);
1576 }
1577 
1578 static void ib_sa_path_rec_release(struct ib_sa_query *sa_query)
1579 {
1580 	struct ib_sa_path_query *query =
1581 		container_of(sa_query, struct ib_sa_path_query, sa_query);
1582 
1583 	kfree(query->conv_pr);
1584 	kfree(query);
1585 }
1586 
1587 /**
1588  * ib_sa_path_rec_get - Start a Path get query
1589  * @client:SA client
1590  * @device:device to send query on
1591  * @port_num: port number to send query on
1592  * @rec:Path Record to send in query
1593  * @comp_mask:component mask to send in query
1594  * @timeout_ms:time to wait for response
1595  * @gfp_mask:GFP mask to use for internal allocations
1596  * @callback:function called when query completes, times out or is
1597  * canceled
1598  * @context:opaque user context passed to callback
1599  * @sa_query:query context, used to cancel query
1600  *
1601  * Send a Path Record Get query to the SA to look up a path.  The
1602  * callback function will be called when the query completes (or
1603  * fails); status is 0 for a successful response, -EINTR if the query
1604  * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
1605  * occurred sending the query.  The resp parameter of the callback is
1606  * only valid if status is 0.
1607  *
1608  * If the return value of ib_sa_path_rec_get() is negative, it is an
1609  * error code.  Otherwise it is a query ID that can be used to cancel
1610  * the query.
1611  */
1612 int ib_sa_path_rec_get(struct ib_sa_client *client,
1613 		       struct ib_device *device, u8 port_num,
1614 		       struct sa_path_rec *rec,
1615 		       ib_sa_comp_mask comp_mask,
1616 		       int timeout_ms, gfp_t gfp_mask,
1617 		       void (*callback)(int status,
1618 					struct sa_path_rec *resp,
1619 					void *context),
1620 		       void *context,
1621 		       struct ib_sa_query **sa_query)
1622 {
1623 	struct ib_sa_path_query *query;
1624 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1625 	struct ib_sa_port   *port;
1626 	struct ib_mad_agent *agent;
1627 	struct ib_sa_mad *mad;
1628 	enum opa_pr_supported status;
1629 	int ret;
1630 
1631 	if (!sa_dev)
1632 		return -ENODEV;
1633 
1634 	if ((rec->rec_type != SA_PATH_REC_TYPE_IB) &&
1635 	    (rec->rec_type != SA_PATH_REC_TYPE_OPA))
1636 		return -EINVAL;
1637 
1638 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1639 	agent = port->agent;
1640 
1641 	query = kzalloc(sizeof(*query), gfp_mask);
1642 	if (!query)
1643 		return -ENOMEM;
1644 
1645 	query->sa_query.port     = port;
1646 	if (rec->rec_type == SA_PATH_REC_TYPE_OPA) {
1647 		status = opa_pr_query_possible(client, device, port_num, rec);
1648 		if (status == PR_NOT_SUPPORTED) {
1649 			ret = -EINVAL;
1650 			goto err1;
1651 		} else if (status == PR_OPA_SUPPORTED) {
1652 			query->sa_query.flags |= IB_SA_QUERY_OPA;
1653 		} else {
1654 			query->conv_pr =
1655 				kmalloc(sizeof(*query->conv_pr), gfp_mask);
1656 			if (!query->conv_pr) {
1657 				ret = -ENOMEM;
1658 				goto err1;
1659 			}
1660 		}
1661 	}
1662 
1663 	ret = alloc_mad(&query->sa_query, gfp_mask);
1664 	if (ret)
1665 		goto err2;
1666 
1667 	ib_sa_client_get(client);
1668 	query->sa_query.client = client;
1669 	query->callback        = callback;
1670 	query->context         = context;
1671 
1672 	mad = query->sa_query.mad_buf->mad;
1673 	init_mad(&query->sa_query, agent);
1674 
1675 	query->sa_query.callback = callback ? ib_sa_path_rec_callback : NULL;
1676 	query->sa_query.release  = ib_sa_path_rec_release;
1677 	mad->mad_hdr.method	 = IB_MGMT_METHOD_GET;
1678 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_PATH_REC);
1679 	mad->sa_hdr.comp_mask	 = comp_mask;
1680 
1681 	if (query->sa_query.flags & IB_SA_QUERY_OPA) {
1682 		ib_pack(opa_path_rec_table, ARRAY_SIZE(opa_path_rec_table),
1683 			rec, mad->data);
1684 	} else if (query->conv_pr) {
1685 		sa_convert_path_opa_to_ib(query->conv_pr, rec);
1686 		ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table),
1687 			query->conv_pr, mad->data);
1688 	} else {
1689 		ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table),
1690 			rec, mad->data);
1691 	}
1692 
1693 	*sa_query = &query->sa_query;
1694 
1695 	query->sa_query.flags |= IB_SA_ENABLE_LOCAL_SERVICE;
1696 	query->sa_query.mad_buf->context[1] = (query->conv_pr) ?
1697 						query->conv_pr : rec;
1698 
1699 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1700 	if (ret < 0)
1701 		goto err3;
1702 
1703 	return ret;
1704 
1705 err3:
1706 	*sa_query = NULL;
1707 	ib_sa_client_put(query->sa_query.client);
1708 	free_mad(&query->sa_query);
1709 err2:
1710 	kfree(query->conv_pr);
1711 err1:
1712 	kfree(query);
1713 	return ret;
1714 }
1715 EXPORT_SYMBOL(ib_sa_path_rec_get);
1716 
1717 static void ib_sa_service_rec_callback(struct ib_sa_query *sa_query,
1718 				    int status,
1719 				    struct ib_sa_mad *mad)
1720 {
1721 	struct ib_sa_service_query *query =
1722 		container_of(sa_query, struct ib_sa_service_query, sa_query);
1723 
1724 	if (mad) {
1725 		struct ib_sa_service_rec rec;
1726 
1727 		ib_unpack(service_rec_table, ARRAY_SIZE(service_rec_table),
1728 			  mad->data, &rec);
1729 		query->callback(status, &rec, query->context);
1730 	} else
1731 		query->callback(status, NULL, query->context);
1732 }
1733 
1734 static void ib_sa_service_rec_release(struct ib_sa_query *sa_query)
1735 {
1736 	kfree(container_of(sa_query, struct ib_sa_service_query, sa_query));
1737 }
1738 
1739 /**
1740  * ib_sa_service_rec_query - Start Service Record operation
1741  * @client:SA client
1742  * @device:device to send request on
1743  * @port_num: port number to send request on
1744  * @method:SA method - should be get, set, or delete
1745  * @rec:Service Record to send in request
1746  * @comp_mask:component mask to send in request
1747  * @timeout_ms:time to wait for response
1748  * @gfp_mask:GFP mask to use for internal allocations
1749  * @callback:function called when request completes, times out or is
1750  * canceled
1751  * @context:opaque user context passed to callback
1752  * @sa_query:request context, used to cancel request
1753  *
1754  * Send a Service Record set/get/delete to the SA to register,
1755  * unregister or query a service record.
1756  * The callback function will be called when the request completes (or
1757  * fails); status is 0 for a successful response, -EINTR if the query
1758  * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
1759  * occurred sending the query.  The resp parameter of the callback is
1760  * only valid if status is 0.
1761  *
1762  * If the return value of ib_sa_service_rec_query() is negative, it is an
1763  * error code.  Otherwise it is a request ID that can be used to cancel
1764  * the query.
1765  */
1766 int ib_sa_service_rec_query(struct ib_sa_client *client,
1767 			    struct ib_device *device, u8 port_num, u8 method,
1768 			    struct ib_sa_service_rec *rec,
1769 			    ib_sa_comp_mask comp_mask,
1770 			    int timeout_ms, gfp_t gfp_mask,
1771 			    void (*callback)(int status,
1772 					     struct ib_sa_service_rec *resp,
1773 					     void *context),
1774 			    void *context,
1775 			    struct ib_sa_query **sa_query)
1776 {
1777 	struct ib_sa_service_query *query;
1778 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1779 	struct ib_sa_port   *port;
1780 	struct ib_mad_agent *agent;
1781 	struct ib_sa_mad *mad;
1782 	int ret;
1783 
1784 	if (!sa_dev)
1785 		return -ENODEV;
1786 
1787 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1788 	agent = port->agent;
1789 
1790 	if (method != IB_MGMT_METHOD_GET &&
1791 	    method != IB_MGMT_METHOD_SET &&
1792 	    method != IB_SA_METHOD_DELETE)
1793 		return -EINVAL;
1794 
1795 	query = kzalloc(sizeof(*query), gfp_mask);
1796 	if (!query)
1797 		return -ENOMEM;
1798 
1799 	query->sa_query.port     = port;
1800 	ret = alloc_mad(&query->sa_query, gfp_mask);
1801 	if (ret)
1802 		goto err1;
1803 
1804 	ib_sa_client_get(client);
1805 	query->sa_query.client = client;
1806 	query->callback        = callback;
1807 	query->context         = context;
1808 
1809 	mad = query->sa_query.mad_buf->mad;
1810 	init_mad(&query->sa_query, agent);
1811 
1812 	query->sa_query.callback = callback ? ib_sa_service_rec_callback : NULL;
1813 	query->sa_query.release  = ib_sa_service_rec_release;
1814 	mad->mad_hdr.method	 = method;
1815 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_SERVICE_REC);
1816 	mad->sa_hdr.comp_mask	 = comp_mask;
1817 
1818 	ib_pack(service_rec_table, ARRAY_SIZE(service_rec_table),
1819 		rec, mad->data);
1820 
1821 	*sa_query = &query->sa_query;
1822 
1823 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1824 	if (ret < 0)
1825 		goto err2;
1826 
1827 	return ret;
1828 
1829 err2:
1830 	*sa_query = NULL;
1831 	ib_sa_client_put(query->sa_query.client);
1832 	free_mad(&query->sa_query);
1833 
1834 err1:
1835 	kfree(query);
1836 	return ret;
1837 }
1838 EXPORT_SYMBOL(ib_sa_service_rec_query);
1839 
1840 static void ib_sa_mcmember_rec_callback(struct ib_sa_query *sa_query,
1841 					int status,
1842 					struct ib_sa_mad *mad)
1843 {
1844 	struct ib_sa_mcmember_query *query =
1845 		container_of(sa_query, struct ib_sa_mcmember_query, sa_query);
1846 
1847 	if (mad) {
1848 		struct ib_sa_mcmember_rec rec;
1849 
1850 		ib_unpack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
1851 			  mad->data, &rec);
1852 		query->callback(status, &rec, query->context);
1853 	} else
1854 		query->callback(status, NULL, query->context);
1855 }
1856 
1857 static void ib_sa_mcmember_rec_release(struct ib_sa_query *sa_query)
1858 {
1859 	kfree(container_of(sa_query, struct ib_sa_mcmember_query, sa_query));
1860 }
1861 
1862 int ib_sa_mcmember_rec_query(struct ib_sa_client *client,
1863 			     struct ib_device *device, u8 port_num,
1864 			     u8 method,
1865 			     struct ib_sa_mcmember_rec *rec,
1866 			     ib_sa_comp_mask comp_mask,
1867 			     int timeout_ms, gfp_t gfp_mask,
1868 			     void (*callback)(int status,
1869 					      struct ib_sa_mcmember_rec *resp,
1870 					      void *context),
1871 			     void *context,
1872 			     struct ib_sa_query **sa_query)
1873 {
1874 	struct ib_sa_mcmember_query *query;
1875 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1876 	struct ib_sa_port   *port;
1877 	struct ib_mad_agent *agent;
1878 	struct ib_sa_mad *mad;
1879 	int ret;
1880 
1881 	if (!sa_dev)
1882 		return -ENODEV;
1883 
1884 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1885 	agent = port->agent;
1886 
1887 	query = kzalloc(sizeof(*query), gfp_mask);
1888 	if (!query)
1889 		return -ENOMEM;
1890 
1891 	query->sa_query.port     = port;
1892 	ret = alloc_mad(&query->sa_query, gfp_mask);
1893 	if (ret)
1894 		goto err1;
1895 
1896 	ib_sa_client_get(client);
1897 	query->sa_query.client = client;
1898 	query->callback        = callback;
1899 	query->context         = context;
1900 
1901 	mad = query->sa_query.mad_buf->mad;
1902 	init_mad(&query->sa_query, agent);
1903 
1904 	query->sa_query.callback = callback ? ib_sa_mcmember_rec_callback : NULL;
1905 	query->sa_query.release  = ib_sa_mcmember_rec_release;
1906 	mad->mad_hdr.method	 = method;
1907 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_MC_MEMBER_REC);
1908 	mad->sa_hdr.comp_mask	 = comp_mask;
1909 
1910 	ib_pack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
1911 		rec, mad->data);
1912 
1913 	*sa_query = &query->sa_query;
1914 
1915 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1916 	if (ret < 0)
1917 		goto err2;
1918 
1919 	return ret;
1920 
1921 err2:
1922 	*sa_query = NULL;
1923 	ib_sa_client_put(query->sa_query.client);
1924 	free_mad(&query->sa_query);
1925 
1926 err1:
1927 	kfree(query);
1928 	return ret;
1929 }
1930 
1931 /* Support GuidInfoRecord */
1932 static void ib_sa_guidinfo_rec_callback(struct ib_sa_query *sa_query,
1933 					int status,
1934 					struct ib_sa_mad *mad)
1935 {
1936 	struct ib_sa_guidinfo_query *query =
1937 		container_of(sa_query, struct ib_sa_guidinfo_query, sa_query);
1938 
1939 	if (mad) {
1940 		struct ib_sa_guidinfo_rec rec;
1941 
1942 		ib_unpack(guidinfo_rec_table, ARRAY_SIZE(guidinfo_rec_table),
1943 			  mad->data, &rec);
1944 		query->callback(status, &rec, query->context);
1945 	} else
1946 		query->callback(status, NULL, query->context);
1947 }
1948 
1949 static void ib_sa_guidinfo_rec_release(struct ib_sa_query *sa_query)
1950 {
1951 	kfree(container_of(sa_query, struct ib_sa_guidinfo_query, sa_query));
1952 }
1953 
1954 int ib_sa_guid_info_rec_query(struct ib_sa_client *client,
1955 			      struct ib_device *device, u8 port_num,
1956 			      struct ib_sa_guidinfo_rec *rec,
1957 			      ib_sa_comp_mask comp_mask, u8 method,
1958 			      int timeout_ms, gfp_t gfp_mask,
1959 			      void (*callback)(int status,
1960 					       struct ib_sa_guidinfo_rec *resp,
1961 					       void *context),
1962 			      void *context,
1963 			      struct ib_sa_query **sa_query)
1964 {
1965 	struct ib_sa_guidinfo_query *query;
1966 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1967 	struct ib_sa_port *port;
1968 	struct ib_mad_agent *agent;
1969 	struct ib_sa_mad *mad;
1970 	int ret;
1971 
1972 	if (!sa_dev)
1973 		return -ENODEV;
1974 
1975 	if (method != IB_MGMT_METHOD_GET &&
1976 	    method != IB_MGMT_METHOD_SET &&
1977 	    method != IB_SA_METHOD_DELETE) {
1978 		return -EINVAL;
1979 	}
1980 
1981 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1982 	agent = port->agent;
1983 
1984 	query = kzalloc(sizeof(*query), gfp_mask);
1985 	if (!query)
1986 		return -ENOMEM;
1987 
1988 	query->sa_query.port = port;
1989 	ret = alloc_mad(&query->sa_query, gfp_mask);
1990 	if (ret)
1991 		goto err1;
1992 
1993 	ib_sa_client_get(client);
1994 	query->sa_query.client = client;
1995 	query->callback        = callback;
1996 	query->context         = context;
1997 
1998 	mad = query->sa_query.mad_buf->mad;
1999 	init_mad(&query->sa_query, agent);
2000 
2001 	query->sa_query.callback = callback ? ib_sa_guidinfo_rec_callback : NULL;
2002 	query->sa_query.release  = ib_sa_guidinfo_rec_release;
2003 
2004 	mad->mad_hdr.method	 = method;
2005 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_GUID_INFO_REC);
2006 	mad->sa_hdr.comp_mask	 = comp_mask;
2007 
2008 	ib_pack(guidinfo_rec_table, ARRAY_SIZE(guidinfo_rec_table), rec,
2009 		mad->data);
2010 
2011 	*sa_query = &query->sa_query;
2012 
2013 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
2014 	if (ret < 0)
2015 		goto err2;
2016 
2017 	return ret;
2018 
2019 err2:
2020 	*sa_query = NULL;
2021 	ib_sa_client_put(query->sa_query.client);
2022 	free_mad(&query->sa_query);
2023 
2024 err1:
2025 	kfree(query);
2026 	return ret;
2027 }
2028 EXPORT_SYMBOL(ib_sa_guid_info_rec_query);
2029 
2030 bool ib_sa_sendonly_fullmem_support(struct ib_sa_client *client,
2031 				    struct ib_device *device,
2032 				    u8 port_num)
2033 {
2034 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
2035 	struct ib_sa_port *port;
2036 	bool ret = false;
2037 	unsigned long flags;
2038 
2039 	if (!sa_dev)
2040 		return ret;
2041 
2042 	port  = &sa_dev->port[port_num - sa_dev->start_port];
2043 
2044 	spin_lock_irqsave(&port->classport_lock, flags);
2045 	if ((port->classport_info.valid) &&
2046 	    (port->classport_info.data.type == RDMA_CLASS_PORT_INFO_IB))
2047 		ret = ib_get_cpi_capmask2(&port->classport_info.data.ib)
2048 			& IB_SA_CAP_MASK2_SENDONLY_FULL_MEM_SUPPORT;
2049 	spin_unlock_irqrestore(&port->classport_lock, flags);
2050 	return ret;
2051 }
2052 EXPORT_SYMBOL(ib_sa_sendonly_fullmem_support);
2053 
2054 struct ib_classport_info_context {
2055 	struct completion	done;
2056 	struct ib_sa_query	*sa_query;
2057 };
2058 
2059 static void ib_classportinfo_cb(void *context)
2060 {
2061 	struct ib_classport_info_context *cb_ctx = context;
2062 
2063 	complete(&cb_ctx->done);
2064 }
2065 
2066 static void ib_sa_classport_info_rec_callback(struct ib_sa_query *sa_query,
2067 					      int status,
2068 					      struct ib_sa_mad *mad)
2069 {
2070 	unsigned long flags;
2071 	struct ib_sa_classport_info_query *query =
2072 		container_of(sa_query, struct ib_sa_classport_info_query, sa_query);
2073 	struct ib_sa_classport_cache *info = &sa_query->port->classport_info;
2074 
2075 	if (mad) {
2076 		if (sa_query->flags & IB_SA_QUERY_OPA) {
2077 			struct opa_class_port_info rec;
2078 
2079 			ib_unpack(opa_classport_info_rec_table,
2080 				  ARRAY_SIZE(opa_classport_info_rec_table),
2081 				  mad->data, &rec);
2082 
2083 			spin_lock_irqsave(&sa_query->port->classport_lock,
2084 					  flags);
2085 			if (!status && !info->valid) {
2086 				memcpy(&info->data.opa, &rec,
2087 				       sizeof(info->data.opa));
2088 
2089 				info->valid = true;
2090 				info->data.type = RDMA_CLASS_PORT_INFO_OPA;
2091 			}
2092 			spin_unlock_irqrestore(&sa_query->port->classport_lock,
2093 					       flags);
2094 
2095 		} else {
2096 			struct ib_class_port_info rec;
2097 
2098 			ib_unpack(ib_classport_info_rec_table,
2099 				  ARRAY_SIZE(ib_classport_info_rec_table),
2100 				  mad->data, &rec);
2101 
2102 			spin_lock_irqsave(&sa_query->port->classport_lock,
2103 					  flags);
2104 			if (!status && !info->valid) {
2105 				memcpy(&info->data.ib, &rec,
2106 				       sizeof(info->data.ib));
2107 
2108 				info->valid = true;
2109 				info->data.type = RDMA_CLASS_PORT_INFO_IB;
2110 			}
2111 			spin_unlock_irqrestore(&sa_query->port->classport_lock,
2112 					       flags);
2113 		}
2114 	}
2115 	query->callback(query->context);
2116 }
2117 
2118 static void ib_sa_classport_info_rec_release(struct ib_sa_query *sa_query)
2119 {
2120 	kfree(container_of(sa_query, struct ib_sa_classport_info_query,
2121 			   sa_query));
2122 }
2123 
2124 static int ib_sa_classport_info_rec_query(struct ib_sa_port *port,
2125 					  int timeout_ms,
2126 					  void (*callback)(void *context),
2127 					  void *context,
2128 					  struct ib_sa_query **sa_query)
2129 {
2130 	struct ib_mad_agent *agent;
2131 	struct ib_sa_classport_info_query *query;
2132 	struct ib_sa_mad *mad;
2133 	gfp_t gfp_mask = GFP_KERNEL;
2134 	int ret;
2135 
2136 	agent = port->agent;
2137 
2138 	query = kzalloc(sizeof(*query), gfp_mask);
2139 	if (!query)
2140 		return -ENOMEM;
2141 
2142 	query->sa_query.port = port;
2143 	query->sa_query.flags |= rdma_cap_opa_ah(port->agent->device,
2144 						 port->port_num) ?
2145 				 IB_SA_QUERY_OPA : 0;
2146 	ret = alloc_mad(&query->sa_query, gfp_mask);
2147 	if (ret)
2148 		goto err_free;
2149 
2150 	query->callback = callback;
2151 	query->context = context;
2152 
2153 	mad = query->sa_query.mad_buf->mad;
2154 	init_mad(&query->sa_query, agent);
2155 
2156 	query->sa_query.callback = ib_sa_classport_info_rec_callback;
2157 	query->sa_query.release  = ib_sa_classport_info_rec_release;
2158 	mad->mad_hdr.method	 = IB_MGMT_METHOD_GET;
2159 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_CLASS_PORTINFO);
2160 	mad->sa_hdr.comp_mask	 = 0;
2161 	*sa_query = &query->sa_query;
2162 
2163 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
2164 	if (ret < 0)
2165 		goto err_free_mad;
2166 
2167 	return ret;
2168 
2169 err_free_mad:
2170 	*sa_query = NULL;
2171 	free_mad(&query->sa_query);
2172 
2173 err_free:
2174 	kfree(query);
2175 	return ret;
2176 }
2177 
2178 static void update_ib_cpi(struct work_struct *work)
2179 {
2180 	struct ib_sa_port *port =
2181 		container_of(work, struct ib_sa_port, ib_cpi_work.work);
2182 	struct ib_classport_info_context *cb_context;
2183 	unsigned long flags;
2184 	int ret;
2185 
2186 	/* If the classport info is valid, nothing
2187 	 * to do here.
2188 	 */
2189 	spin_lock_irqsave(&port->classport_lock, flags);
2190 	if (port->classport_info.valid) {
2191 		spin_unlock_irqrestore(&port->classport_lock, flags);
2192 		return;
2193 	}
2194 	spin_unlock_irqrestore(&port->classport_lock, flags);
2195 
2196 	cb_context = kmalloc(sizeof(*cb_context), GFP_KERNEL);
2197 	if (!cb_context)
2198 		goto err_nomem;
2199 
2200 	init_completion(&cb_context->done);
2201 
2202 	ret = ib_sa_classport_info_rec_query(port, 3000,
2203 					     ib_classportinfo_cb, cb_context,
2204 					     &cb_context->sa_query);
2205 	if (ret < 0)
2206 		goto free_cb_err;
2207 	wait_for_completion(&cb_context->done);
2208 free_cb_err:
2209 	kfree(cb_context);
2210 	spin_lock_irqsave(&port->classport_lock, flags);
2211 
2212 	/* If the classport info is still not valid, the query should have
2213 	 * failed for some reason. Retry issuing the query
2214 	 */
2215 	if (!port->classport_info.valid) {
2216 		port->classport_info.retry_cnt++;
2217 		if (port->classport_info.retry_cnt <=
2218 		    IB_SA_CPI_MAX_RETRY_CNT) {
2219 			unsigned long delay =
2220 				msecs_to_jiffies(IB_SA_CPI_RETRY_WAIT);
2221 
2222 			queue_delayed_work(ib_wq, &port->ib_cpi_work, delay);
2223 		}
2224 	}
2225 	spin_unlock_irqrestore(&port->classport_lock, flags);
2226 
2227 err_nomem:
2228 	return;
2229 }
2230 
2231 static void send_handler(struct ib_mad_agent *agent,
2232 			 struct ib_mad_send_wc *mad_send_wc)
2233 {
2234 	struct ib_sa_query *query = mad_send_wc->send_buf->context[0];
2235 	unsigned long flags;
2236 
2237 	if (query->callback)
2238 		switch (mad_send_wc->status) {
2239 		case IB_WC_SUCCESS:
2240 			/* No callback -- already got recv */
2241 			break;
2242 		case IB_WC_RESP_TIMEOUT_ERR:
2243 			query->callback(query, -ETIMEDOUT, NULL);
2244 			break;
2245 		case IB_WC_WR_FLUSH_ERR:
2246 			query->callback(query, -EINTR, NULL);
2247 			break;
2248 		default:
2249 			query->callback(query, -EIO, NULL);
2250 			break;
2251 		}
2252 
2253 	spin_lock_irqsave(&idr_lock, flags);
2254 	idr_remove(&query_idr, query->id);
2255 	spin_unlock_irqrestore(&idr_lock, flags);
2256 
2257 	free_mad(query);
2258 	if (query->client)
2259 		ib_sa_client_put(query->client);
2260 	query->release(query);
2261 }
2262 
2263 static void recv_handler(struct ib_mad_agent *mad_agent,
2264 			 struct ib_mad_send_buf *send_buf,
2265 			 struct ib_mad_recv_wc *mad_recv_wc)
2266 {
2267 	struct ib_sa_query *query;
2268 
2269 	if (!send_buf)
2270 		return;
2271 
2272 	query = send_buf->context[0];
2273 	if (query->callback) {
2274 		if (mad_recv_wc->wc->status == IB_WC_SUCCESS)
2275 			query->callback(query,
2276 					mad_recv_wc->recv_buf.mad->mad_hdr.status ?
2277 					-EINVAL : 0,
2278 					(struct ib_sa_mad *) mad_recv_wc->recv_buf.mad);
2279 		else
2280 			query->callback(query, -EIO, NULL);
2281 	}
2282 
2283 	ib_free_recv_mad(mad_recv_wc);
2284 }
2285 
2286 static void update_sm_ah(struct work_struct *work)
2287 {
2288 	struct ib_sa_port *port =
2289 		container_of(work, struct ib_sa_port, update_task);
2290 	struct ib_sa_sm_ah *new_ah;
2291 	struct ib_port_attr port_attr;
2292 	struct rdma_ah_attr   ah_attr;
2293 
2294 	if (ib_query_port(port->agent->device, port->port_num, &port_attr)) {
2295 		pr_warn("Couldn't query port\n");
2296 		return;
2297 	}
2298 
2299 	new_ah = kmalloc(sizeof(*new_ah), GFP_KERNEL);
2300 	if (!new_ah)
2301 		return;
2302 
2303 	kref_init(&new_ah->ref);
2304 	new_ah->src_path_mask = (1 << port_attr.lmc) - 1;
2305 
2306 	new_ah->pkey_index = 0;
2307 	if (ib_find_pkey(port->agent->device, port->port_num,
2308 			 IB_DEFAULT_PKEY_FULL, &new_ah->pkey_index))
2309 		pr_err("Couldn't find index for default PKey\n");
2310 
2311 	memset(&ah_attr, 0, sizeof(ah_attr));
2312 	ah_attr.type = rdma_ah_find_type(port->agent->device,
2313 					 port->port_num);
2314 	rdma_ah_set_dlid(&ah_attr, port_attr.sm_lid);
2315 	rdma_ah_set_sl(&ah_attr, port_attr.sm_sl);
2316 	rdma_ah_set_port_num(&ah_attr, port->port_num);
2317 	if (port_attr.grh_required) {
2318 		if (ah_attr.type == RDMA_AH_ATTR_TYPE_OPA) {
2319 			rdma_ah_set_make_grd(&ah_attr, true);
2320 		} else {
2321 			rdma_ah_set_ah_flags(&ah_attr, IB_AH_GRH);
2322 			rdma_ah_set_subnet_prefix(&ah_attr,
2323 						  cpu_to_be64(port_attr.subnet_prefix));
2324 			rdma_ah_set_interface_id(&ah_attr,
2325 						 cpu_to_be64(IB_SA_WELL_KNOWN_GUID));
2326 		}
2327 	}
2328 
2329 	new_ah->ah = rdma_create_ah(port->agent->qp->pd, &ah_attr);
2330 	if (IS_ERR(new_ah->ah)) {
2331 		pr_warn("Couldn't create new SM AH\n");
2332 		kfree(new_ah);
2333 		return;
2334 	}
2335 
2336 	spin_lock_irq(&port->ah_lock);
2337 	if (port->sm_ah)
2338 		kref_put(&port->sm_ah->ref, free_sm_ah);
2339 	port->sm_ah = new_ah;
2340 	spin_unlock_irq(&port->ah_lock);
2341 }
2342 
2343 static void ib_sa_event(struct ib_event_handler *handler,
2344 			struct ib_event *event)
2345 {
2346 	if (event->event == IB_EVENT_PORT_ERR    ||
2347 	    event->event == IB_EVENT_PORT_ACTIVE ||
2348 	    event->event == IB_EVENT_LID_CHANGE  ||
2349 	    event->event == IB_EVENT_PKEY_CHANGE ||
2350 	    event->event == IB_EVENT_SM_CHANGE   ||
2351 	    event->event == IB_EVENT_CLIENT_REREGISTER) {
2352 		unsigned long flags;
2353 		struct ib_sa_device *sa_dev =
2354 			container_of(handler, typeof(*sa_dev), event_handler);
2355 		u8 port_num = event->element.port_num - sa_dev->start_port;
2356 		struct ib_sa_port *port = &sa_dev->port[port_num];
2357 
2358 		if (!rdma_cap_ib_sa(handler->device, port->port_num))
2359 			return;
2360 
2361 		spin_lock_irqsave(&port->ah_lock, flags);
2362 		if (port->sm_ah)
2363 			kref_put(&port->sm_ah->ref, free_sm_ah);
2364 		port->sm_ah = NULL;
2365 		spin_unlock_irqrestore(&port->ah_lock, flags);
2366 
2367 		if (event->event == IB_EVENT_SM_CHANGE ||
2368 		    event->event == IB_EVENT_CLIENT_REREGISTER ||
2369 		    event->event == IB_EVENT_LID_CHANGE ||
2370 		    event->event == IB_EVENT_PORT_ACTIVE) {
2371 			unsigned long delay =
2372 				msecs_to_jiffies(IB_SA_CPI_RETRY_WAIT);
2373 
2374 			spin_lock_irqsave(&port->classport_lock, flags);
2375 			port->classport_info.valid = false;
2376 			port->classport_info.retry_cnt = 0;
2377 			spin_unlock_irqrestore(&port->classport_lock, flags);
2378 			queue_delayed_work(ib_wq,
2379 					   &port->ib_cpi_work, delay);
2380 		}
2381 		queue_work(ib_wq, &sa_dev->port[port_num].update_task);
2382 	}
2383 }
2384 
2385 static void ib_sa_add_one(struct ib_device *device)
2386 {
2387 	struct ib_sa_device *sa_dev;
2388 	int s, e, i;
2389 	int count = 0;
2390 
2391 	s = rdma_start_port(device);
2392 	e = rdma_end_port(device);
2393 
2394 	sa_dev = kzalloc(sizeof *sa_dev +
2395 			 (e - s + 1) * sizeof (struct ib_sa_port),
2396 			 GFP_KERNEL);
2397 	if (!sa_dev)
2398 		return;
2399 
2400 	sa_dev->start_port = s;
2401 	sa_dev->end_port   = e;
2402 
2403 	for (i = 0; i <= e - s; ++i) {
2404 		spin_lock_init(&sa_dev->port[i].ah_lock);
2405 		if (!rdma_cap_ib_sa(device, i + 1))
2406 			continue;
2407 
2408 		sa_dev->port[i].sm_ah    = NULL;
2409 		sa_dev->port[i].port_num = i + s;
2410 
2411 		spin_lock_init(&sa_dev->port[i].classport_lock);
2412 		sa_dev->port[i].classport_info.valid = false;
2413 
2414 		sa_dev->port[i].agent =
2415 			ib_register_mad_agent(device, i + s, IB_QPT_GSI,
2416 					      NULL, 0, send_handler,
2417 					      recv_handler, sa_dev, 0);
2418 		if (IS_ERR(sa_dev->port[i].agent))
2419 			goto err;
2420 
2421 		INIT_WORK(&sa_dev->port[i].update_task, update_sm_ah);
2422 		INIT_DELAYED_WORK(&sa_dev->port[i].ib_cpi_work,
2423 				  update_ib_cpi);
2424 
2425 		count++;
2426 	}
2427 
2428 	if (!count)
2429 		goto free;
2430 
2431 	ib_set_client_data(device, &sa_client, sa_dev);
2432 
2433 	/*
2434 	 * We register our event handler after everything is set up,
2435 	 * and then update our cached info after the event handler is
2436 	 * registered to avoid any problems if a port changes state
2437 	 * during our initialization.
2438 	 */
2439 
2440 	INIT_IB_EVENT_HANDLER(&sa_dev->event_handler, device, ib_sa_event);
2441 	ib_register_event_handler(&sa_dev->event_handler);
2442 
2443 	for (i = 0; i <= e - s; ++i) {
2444 		if (rdma_cap_ib_sa(device, i + 1))
2445 			update_sm_ah(&sa_dev->port[i].update_task);
2446 	}
2447 
2448 	return;
2449 
2450 err:
2451 	while (--i >= 0) {
2452 		if (rdma_cap_ib_sa(device, i + 1))
2453 			ib_unregister_mad_agent(sa_dev->port[i].agent);
2454 	}
2455 free:
2456 	kfree(sa_dev);
2457 	return;
2458 }
2459 
2460 static void ib_sa_remove_one(struct ib_device *device, void *client_data)
2461 {
2462 	struct ib_sa_device *sa_dev = client_data;
2463 	int i;
2464 
2465 	if (!sa_dev)
2466 		return;
2467 
2468 	ib_unregister_event_handler(&sa_dev->event_handler);
2469 	flush_workqueue(ib_wq);
2470 
2471 	for (i = 0; i <= sa_dev->end_port - sa_dev->start_port; ++i) {
2472 		if (rdma_cap_ib_sa(device, i + 1)) {
2473 			cancel_delayed_work_sync(&sa_dev->port[i].ib_cpi_work);
2474 			ib_unregister_mad_agent(sa_dev->port[i].agent);
2475 			if (sa_dev->port[i].sm_ah)
2476 				kref_put(&sa_dev->port[i].sm_ah->ref, free_sm_ah);
2477 		}
2478 
2479 	}
2480 
2481 	kfree(sa_dev);
2482 }
2483 
2484 int ib_sa_init(void)
2485 {
2486 	int ret;
2487 
2488 	get_random_bytes(&tid, sizeof tid);
2489 
2490 	atomic_set(&ib_nl_sa_request_seq, 0);
2491 
2492 	ret = ib_register_client(&sa_client);
2493 	if (ret) {
2494 		pr_err("Couldn't register ib_sa client\n");
2495 		goto err1;
2496 	}
2497 
2498 	ret = mcast_init();
2499 	if (ret) {
2500 		pr_err("Couldn't initialize multicast handling\n");
2501 		goto err2;
2502 	}
2503 
2504 	ib_nl_wq = alloc_ordered_workqueue("ib_nl_sa_wq", WQ_MEM_RECLAIM);
2505 	if (!ib_nl_wq) {
2506 		ret = -ENOMEM;
2507 		goto err3;
2508 	}
2509 
2510 	INIT_DELAYED_WORK(&ib_nl_timed_work, ib_nl_request_timeout);
2511 
2512 	return 0;
2513 
2514 err3:
2515 	mcast_cleanup();
2516 err2:
2517 	ib_unregister_client(&sa_client);
2518 err1:
2519 	return ret;
2520 }
2521 
2522 void ib_sa_cleanup(void)
2523 {
2524 	cancel_delayed_work(&ib_nl_timed_work);
2525 	flush_workqueue(ib_nl_wq);
2526 	destroy_workqueue(ib_nl_wq);
2527 	mcast_cleanup();
2528 	ib_unregister_client(&sa_client);
2529 	idr_destroy(&query_idr);
2530 }
2531