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 int ib_init_ah_attr_from_path(struct ib_device *device, u8 port_num,
1231 			      struct sa_path_rec *rec,
1232 			      struct rdma_ah_attr *ah_attr)
1233 {
1234 	int ret;
1235 	u16 gid_index;
1236 	int use_roce;
1237 	struct net_device *ndev = NULL;
1238 
1239 	memset(ah_attr, 0, sizeof *ah_attr);
1240 	ah_attr->type = rdma_ah_find_type(device, port_num);
1241 
1242 	rdma_ah_set_dlid(ah_attr, be32_to_cpu(sa_path_get_dlid(rec)));
1243 
1244 	if ((ah_attr->type == RDMA_AH_ATTR_TYPE_OPA) &&
1245 	    (rdma_ah_get_dlid(ah_attr) == be16_to_cpu(IB_LID_PERMISSIVE)))
1246 		rdma_ah_set_make_grd(ah_attr, true);
1247 
1248 	rdma_ah_set_sl(ah_attr, rec->sl);
1249 	rdma_ah_set_path_bits(ah_attr, be32_to_cpu(sa_path_get_slid(rec)) &
1250 			      get_src_path_mask(device, port_num));
1251 	rdma_ah_set_port_num(ah_attr, port_num);
1252 	rdma_ah_set_static_rate(ah_attr, rec->rate);
1253 	use_roce = rdma_cap_eth_ah(device, port_num);
1254 
1255 	if (use_roce) {
1256 		struct net_device *idev;
1257 		struct net_device *resolved_dev;
1258 		struct rdma_dev_addr dev_addr = {
1259 			.bound_dev_if = ((sa_path_get_ifindex(rec) >= 0) ?
1260 					 sa_path_get_ifindex(rec) : 0),
1261 			.net = sa_path_get_ndev(rec) ?
1262 				sa_path_get_ndev(rec) :
1263 				&init_net
1264 		};
1265 		union {
1266 			struct sockaddr     _sockaddr;
1267 			struct sockaddr_in  _sockaddr_in;
1268 			struct sockaddr_in6 _sockaddr_in6;
1269 		} sgid_addr, dgid_addr;
1270 
1271 		if (!device->get_netdev)
1272 			return -EOPNOTSUPP;
1273 
1274 		rdma_gid2ip(&sgid_addr._sockaddr, &rec->sgid);
1275 		rdma_gid2ip(&dgid_addr._sockaddr, &rec->dgid);
1276 
1277 		/* validate the route */
1278 		ret = rdma_resolve_ip_route(&sgid_addr._sockaddr,
1279 					    &dgid_addr._sockaddr, &dev_addr);
1280 		if (ret)
1281 			return ret;
1282 
1283 		if ((dev_addr.network == RDMA_NETWORK_IPV4 ||
1284 		     dev_addr.network == RDMA_NETWORK_IPV6) &&
1285 		    rec->rec_type != SA_PATH_REC_TYPE_ROCE_V2)
1286 			return -EINVAL;
1287 
1288 		idev = device->get_netdev(device, port_num);
1289 		if (!idev)
1290 			return -ENODEV;
1291 
1292 		resolved_dev = dev_get_by_index(dev_addr.net,
1293 						dev_addr.bound_dev_if);
1294 		if (resolved_dev->flags & IFF_LOOPBACK) {
1295 			dev_put(resolved_dev);
1296 			resolved_dev = idev;
1297 			dev_hold(resolved_dev);
1298 		}
1299 		ndev = ib_get_ndev_from_path(rec);
1300 		rcu_read_lock();
1301 		if ((ndev && ndev != resolved_dev) ||
1302 		    (resolved_dev != idev &&
1303 		     !rdma_is_upper_dev_rcu(idev, resolved_dev)))
1304 			ret = -EHOSTUNREACH;
1305 		rcu_read_unlock();
1306 		dev_put(idev);
1307 		dev_put(resolved_dev);
1308 		if (ret) {
1309 			if (ndev)
1310 				dev_put(ndev);
1311 			return ret;
1312 		}
1313 	}
1314 
1315 	if (rec->hop_limit > 0 || use_roce) {
1316 		enum ib_gid_type type = sa_conv_pathrec_to_gid_type(rec);
1317 
1318 		ret = ib_find_cached_gid_by_port(device, &rec->sgid, type,
1319 						 port_num, ndev, &gid_index);
1320 		if (ret) {
1321 			if (ndev)
1322 				dev_put(ndev);
1323 			return ret;
1324 		}
1325 
1326 		rdma_ah_set_grh(ah_attr, &rec->dgid,
1327 				be32_to_cpu(rec->flow_label),
1328 				gid_index, rec->hop_limit,
1329 				rec->traffic_class);
1330 		if (ndev)
1331 			dev_put(ndev);
1332 	}
1333 
1334 	if (use_roce) {
1335 		u8 *dmac = sa_path_get_dmac(rec);
1336 
1337 		if (!dmac)
1338 			return -EINVAL;
1339 		memcpy(ah_attr->roce.dmac, dmac, ETH_ALEN);
1340 	}
1341 
1342 	return 0;
1343 }
1344 EXPORT_SYMBOL(ib_init_ah_attr_from_path);
1345 
1346 static int alloc_mad(struct ib_sa_query *query, gfp_t gfp_mask)
1347 {
1348 	struct rdma_ah_attr ah_attr;
1349 	unsigned long flags;
1350 
1351 	spin_lock_irqsave(&query->port->ah_lock, flags);
1352 	if (!query->port->sm_ah) {
1353 		spin_unlock_irqrestore(&query->port->ah_lock, flags);
1354 		return -EAGAIN;
1355 	}
1356 	kref_get(&query->port->sm_ah->ref);
1357 	query->sm_ah = query->port->sm_ah;
1358 	spin_unlock_irqrestore(&query->port->ah_lock, flags);
1359 
1360 	/*
1361 	 * Always check if sm_ah has valid dlid assigned,
1362 	 * before querying for class port info
1363 	 */
1364 	if ((rdma_query_ah(query->sm_ah->ah, &ah_attr) < 0) ||
1365 	    !rdma_is_valid_unicast_lid(&ah_attr)) {
1366 		kref_put(&query->sm_ah->ref, free_sm_ah);
1367 		return -EAGAIN;
1368 	}
1369 	query->mad_buf = ib_create_send_mad(query->port->agent, 1,
1370 					    query->sm_ah->pkey_index,
1371 					    0, IB_MGMT_SA_HDR, IB_MGMT_SA_DATA,
1372 					    gfp_mask,
1373 					    ((query->flags & IB_SA_QUERY_OPA) ?
1374 					     OPA_MGMT_BASE_VERSION :
1375 					     IB_MGMT_BASE_VERSION));
1376 	if (IS_ERR(query->mad_buf)) {
1377 		kref_put(&query->sm_ah->ref, free_sm_ah);
1378 		return -ENOMEM;
1379 	}
1380 
1381 	query->mad_buf->ah = query->sm_ah->ah;
1382 
1383 	return 0;
1384 }
1385 
1386 static void free_mad(struct ib_sa_query *query)
1387 {
1388 	ib_free_send_mad(query->mad_buf);
1389 	kref_put(&query->sm_ah->ref, free_sm_ah);
1390 }
1391 
1392 static void init_mad(struct ib_sa_query *query, struct ib_mad_agent *agent)
1393 {
1394 	struct ib_sa_mad *mad = query->mad_buf->mad;
1395 	unsigned long flags;
1396 
1397 	memset(mad, 0, sizeof *mad);
1398 
1399 	if (query->flags & IB_SA_QUERY_OPA) {
1400 		mad->mad_hdr.base_version  = OPA_MGMT_BASE_VERSION;
1401 		mad->mad_hdr.class_version = OPA_SA_CLASS_VERSION;
1402 	} else {
1403 		mad->mad_hdr.base_version  = IB_MGMT_BASE_VERSION;
1404 		mad->mad_hdr.class_version = IB_SA_CLASS_VERSION;
1405 	}
1406 	mad->mad_hdr.mgmt_class    = IB_MGMT_CLASS_SUBN_ADM;
1407 	spin_lock_irqsave(&tid_lock, flags);
1408 	mad->mad_hdr.tid           =
1409 		cpu_to_be64(((u64) agent->hi_tid) << 32 | tid++);
1410 	spin_unlock_irqrestore(&tid_lock, flags);
1411 }
1412 
1413 static int send_mad(struct ib_sa_query *query, int timeout_ms, gfp_t gfp_mask)
1414 {
1415 	bool preload = gfpflags_allow_blocking(gfp_mask);
1416 	unsigned long flags;
1417 	int ret, id;
1418 
1419 	if (preload)
1420 		idr_preload(gfp_mask);
1421 	spin_lock_irqsave(&idr_lock, flags);
1422 
1423 	id = idr_alloc(&query_idr, query, 0, 0, GFP_NOWAIT);
1424 
1425 	spin_unlock_irqrestore(&idr_lock, flags);
1426 	if (preload)
1427 		idr_preload_end();
1428 	if (id < 0)
1429 		return id;
1430 
1431 	query->mad_buf->timeout_ms  = timeout_ms;
1432 	query->mad_buf->context[0] = query;
1433 	query->id = id;
1434 
1435 	if ((query->flags & IB_SA_ENABLE_LOCAL_SERVICE) &&
1436 	    (!(query->flags & IB_SA_QUERY_OPA))) {
1437 		if (!rdma_nl_chk_listeners(RDMA_NL_GROUP_LS)) {
1438 			if (!ib_nl_make_request(query, gfp_mask))
1439 				return id;
1440 		}
1441 		ib_sa_disable_local_svc(query);
1442 	}
1443 
1444 	ret = ib_post_send_mad(query->mad_buf, NULL);
1445 	if (ret) {
1446 		spin_lock_irqsave(&idr_lock, flags);
1447 		idr_remove(&query_idr, id);
1448 		spin_unlock_irqrestore(&idr_lock, flags);
1449 	}
1450 
1451 	/*
1452 	 * It's not safe to dereference query any more, because the
1453 	 * send may already have completed and freed the query in
1454 	 * another context.
1455 	 */
1456 	return ret ? ret : id;
1457 }
1458 
1459 void ib_sa_unpack_path(void *attribute, struct sa_path_rec *rec)
1460 {
1461 	ib_unpack(path_rec_table, ARRAY_SIZE(path_rec_table), attribute, rec);
1462 }
1463 EXPORT_SYMBOL(ib_sa_unpack_path);
1464 
1465 void ib_sa_pack_path(struct sa_path_rec *rec, void *attribute)
1466 {
1467 	ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table), rec, attribute);
1468 }
1469 EXPORT_SYMBOL(ib_sa_pack_path);
1470 
1471 static bool ib_sa_opa_pathrecord_support(struct ib_sa_client *client,
1472 					 struct ib_device *device,
1473 					 u8 port_num)
1474 {
1475 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1476 	struct ib_sa_port *port;
1477 	unsigned long flags;
1478 	bool ret = false;
1479 
1480 	if (!sa_dev)
1481 		return ret;
1482 
1483 	port = &sa_dev->port[port_num - sa_dev->start_port];
1484 	spin_lock_irqsave(&port->classport_lock, flags);
1485 	if (!port->classport_info.valid)
1486 		goto ret;
1487 
1488 	if (port->classport_info.data.type == RDMA_CLASS_PORT_INFO_OPA)
1489 		ret = opa_get_cpi_capmask2(&port->classport_info.data.opa) &
1490 			OPA_CLASS_PORT_INFO_PR_SUPPORT;
1491 ret:
1492 	spin_unlock_irqrestore(&port->classport_lock, flags);
1493 	return ret;
1494 }
1495 
1496 enum opa_pr_supported {
1497 	PR_NOT_SUPPORTED,
1498 	PR_OPA_SUPPORTED,
1499 	PR_IB_SUPPORTED
1500 };
1501 
1502 /**
1503  * Check if current PR query can be an OPA query.
1504  * Retuns PR_NOT_SUPPORTED if a path record query is not
1505  * possible, PR_OPA_SUPPORTED if an OPA path record query
1506  * is possible and PR_IB_SUPPORTED if an IB path record
1507  * query is possible.
1508  */
1509 static int opa_pr_query_possible(struct ib_sa_client *client,
1510 				 struct ib_device *device,
1511 				 u8 port_num,
1512 				 struct sa_path_rec *rec)
1513 {
1514 	struct ib_port_attr port_attr;
1515 
1516 	if (ib_query_port(device, port_num, &port_attr))
1517 		return PR_NOT_SUPPORTED;
1518 
1519 	if (ib_sa_opa_pathrecord_support(client, device, port_num))
1520 		return PR_OPA_SUPPORTED;
1521 
1522 	if (port_attr.lid >= be16_to_cpu(IB_MULTICAST_LID_BASE))
1523 		return PR_NOT_SUPPORTED;
1524 	else
1525 		return PR_IB_SUPPORTED;
1526 }
1527 
1528 static void ib_sa_path_rec_callback(struct ib_sa_query *sa_query,
1529 				    int status,
1530 				    struct ib_sa_mad *mad)
1531 {
1532 	struct ib_sa_path_query *query =
1533 		container_of(sa_query, struct ib_sa_path_query, sa_query);
1534 
1535 	if (mad) {
1536 		struct sa_path_rec rec;
1537 
1538 		if (sa_query->flags & IB_SA_QUERY_OPA) {
1539 			ib_unpack(opa_path_rec_table,
1540 				  ARRAY_SIZE(opa_path_rec_table),
1541 				  mad->data, &rec);
1542 			rec.rec_type = SA_PATH_REC_TYPE_OPA;
1543 			query->callback(status, &rec, query->context);
1544 		} else {
1545 			ib_unpack(path_rec_table,
1546 				  ARRAY_SIZE(path_rec_table),
1547 				  mad->data, &rec);
1548 			rec.rec_type = SA_PATH_REC_TYPE_IB;
1549 			sa_path_set_ndev(&rec, NULL);
1550 			sa_path_set_ifindex(&rec, 0);
1551 			sa_path_set_dmac_zero(&rec);
1552 
1553 			if (query->conv_pr) {
1554 				struct sa_path_rec opa;
1555 
1556 				memset(&opa, 0, sizeof(struct sa_path_rec));
1557 				sa_convert_path_ib_to_opa(&opa, &rec);
1558 				query->callback(status, &opa, query->context);
1559 			} else {
1560 				query->callback(status, &rec, query->context);
1561 			}
1562 		}
1563 	} else
1564 		query->callback(status, NULL, query->context);
1565 }
1566 
1567 static void ib_sa_path_rec_release(struct ib_sa_query *sa_query)
1568 {
1569 	struct ib_sa_path_query *query =
1570 		container_of(sa_query, struct ib_sa_path_query, sa_query);
1571 
1572 	kfree(query->conv_pr);
1573 	kfree(query);
1574 }
1575 
1576 /**
1577  * ib_sa_path_rec_get - Start a Path get query
1578  * @client:SA client
1579  * @device:device to send query on
1580  * @port_num: port number to send query on
1581  * @rec:Path Record to send in query
1582  * @comp_mask:component mask to send in query
1583  * @timeout_ms:time to wait for response
1584  * @gfp_mask:GFP mask to use for internal allocations
1585  * @callback:function called when query completes, times out or is
1586  * canceled
1587  * @context:opaque user context passed to callback
1588  * @sa_query:query context, used to cancel query
1589  *
1590  * Send a Path Record Get query to the SA to look up a path.  The
1591  * callback function will be called when the query completes (or
1592  * fails); status is 0 for a successful response, -EINTR if the query
1593  * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
1594  * occurred sending the query.  The resp parameter of the callback is
1595  * only valid if status is 0.
1596  *
1597  * If the return value of ib_sa_path_rec_get() is negative, it is an
1598  * error code.  Otherwise it is a query ID that can be used to cancel
1599  * the query.
1600  */
1601 int ib_sa_path_rec_get(struct ib_sa_client *client,
1602 		       struct ib_device *device, u8 port_num,
1603 		       struct sa_path_rec *rec,
1604 		       ib_sa_comp_mask comp_mask,
1605 		       int timeout_ms, gfp_t gfp_mask,
1606 		       void (*callback)(int status,
1607 					struct sa_path_rec *resp,
1608 					void *context),
1609 		       void *context,
1610 		       struct ib_sa_query **sa_query)
1611 {
1612 	struct ib_sa_path_query *query;
1613 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1614 	struct ib_sa_port   *port;
1615 	struct ib_mad_agent *agent;
1616 	struct ib_sa_mad *mad;
1617 	enum opa_pr_supported status;
1618 	int ret;
1619 
1620 	if (!sa_dev)
1621 		return -ENODEV;
1622 
1623 	if ((rec->rec_type != SA_PATH_REC_TYPE_IB) &&
1624 	    (rec->rec_type != SA_PATH_REC_TYPE_OPA))
1625 		return -EINVAL;
1626 
1627 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1628 	agent = port->agent;
1629 
1630 	query = kzalloc(sizeof(*query), gfp_mask);
1631 	if (!query)
1632 		return -ENOMEM;
1633 
1634 	query->sa_query.port     = port;
1635 	if (rec->rec_type == SA_PATH_REC_TYPE_OPA) {
1636 		status = opa_pr_query_possible(client, device, port_num, rec);
1637 		if (status == PR_NOT_SUPPORTED) {
1638 			ret = -EINVAL;
1639 			goto err1;
1640 		} else if (status == PR_OPA_SUPPORTED) {
1641 			query->sa_query.flags |= IB_SA_QUERY_OPA;
1642 		} else {
1643 			query->conv_pr =
1644 				kmalloc(sizeof(*query->conv_pr), gfp_mask);
1645 			if (!query->conv_pr) {
1646 				ret = -ENOMEM;
1647 				goto err1;
1648 			}
1649 		}
1650 	}
1651 
1652 	ret = alloc_mad(&query->sa_query, gfp_mask);
1653 	if (ret)
1654 		goto err2;
1655 
1656 	ib_sa_client_get(client);
1657 	query->sa_query.client = client;
1658 	query->callback        = callback;
1659 	query->context         = context;
1660 
1661 	mad = query->sa_query.mad_buf->mad;
1662 	init_mad(&query->sa_query, agent);
1663 
1664 	query->sa_query.callback = callback ? ib_sa_path_rec_callback : NULL;
1665 	query->sa_query.release  = ib_sa_path_rec_release;
1666 	mad->mad_hdr.method	 = IB_MGMT_METHOD_GET;
1667 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_PATH_REC);
1668 	mad->sa_hdr.comp_mask	 = comp_mask;
1669 
1670 	if (query->sa_query.flags & IB_SA_QUERY_OPA) {
1671 		ib_pack(opa_path_rec_table, ARRAY_SIZE(opa_path_rec_table),
1672 			rec, mad->data);
1673 	} else if (query->conv_pr) {
1674 		sa_convert_path_opa_to_ib(query->conv_pr, rec);
1675 		ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table),
1676 			query->conv_pr, mad->data);
1677 	} else {
1678 		ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table),
1679 			rec, mad->data);
1680 	}
1681 
1682 	*sa_query = &query->sa_query;
1683 
1684 	query->sa_query.flags |= IB_SA_ENABLE_LOCAL_SERVICE;
1685 	query->sa_query.mad_buf->context[1] = (query->conv_pr) ?
1686 						query->conv_pr : rec;
1687 
1688 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1689 	if (ret < 0)
1690 		goto err3;
1691 
1692 	return ret;
1693 
1694 err3:
1695 	*sa_query = NULL;
1696 	ib_sa_client_put(query->sa_query.client);
1697 	free_mad(&query->sa_query);
1698 err2:
1699 	kfree(query->conv_pr);
1700 err1:
1701 	kfree(query);
1702 	return ret;
1703 }
1704 EXPORT_SYMBOL(ib_sa_path_rec_get);
1705 
1706 static void ib_sa_service_rec_callback(struct ib_sa_query *sa_query,
1707 				    int status,
1708 				    struct ib_sa_mad *mad)
1709 {
1710 	struct ib_sa_service_query *query =
1711 		container_of(sa_query, struct ib_sa_service_query, sa_query);
1712 
1713 	if (mad) {
1714 		struct ib_sa_service_rec rec;
1715 
1716 		ib_unpack(service_rec_table, ARRAY_SIZE(service_rec_table),
1717 			  mad->data, &rec);
1718 		query->callback(status, &rec, query->context);
1719 	} else
1720 		query->callback(status, NULL, query->context);
1721 }
1722 
1723 static void ib_sa_service_rec_release(struct ib_sa_query *sa_query)
1724 {
1725 	kfree(container_of(sa_query, struct ib_sa_service_query, sa_query));
1726 }
1727 
1728 /**
1729  * ib_sa_service_rec_query - Start Service Record operation
1730  * @client:SA client
1731  * @device:device to send request on
1732  * @port_num: port number to send request on
1733  * @method:SA method - should be get, set, or delete
1734  * @rec:Service Record to send in request
1735  * @comp_mask:component mask to send in request
1736  * @timeout_ms:time to wait for response
1737  * @gfp_mask:GFP mask to use for internal allocations
1738  * @callback:function called when request completes, times out or is
1739  * canceled
1740  * @context:opaque user context passed to callback
1741  * @sa_query:request context, used to cancel request
1742  *
1743  * Send a Service Record set/get/delete to the SA to register,
1744  * unregister or query a service record.
1745  * The callback function will be called when the request completes (or
1746  * fails); status is 0 for a successful response, -EINTR if the query
1747  * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
1748  * occurred sending the query.  The resp parameter of the callback is
1749  * only valid if status is 0.
1750  *
1751  * If the return value of ib_sa_service_rec_query() is negative, it is an
1752  * error code.  Otherwise it is a request ID that can be used to cancel
1753  * the query.
1754  */
1755 int ib_sa_service_rec_query(struct ib_sa_client *client,
1756 			    struct ib_device *device, u8 port_num, u8 method,
1757 			    struct ib_sa_service_rec *rec,
1758 			    ib_sa_comp_mask comp_mask,
1759 			    int timeout_ms, gfp_t gfp_mask,
1760 			    void (*callback)(int status,
1761 					     struct ib_sa_service_rec *resp,
1762 					     void *context),
1763 			    void *context,
1764 			    struct ib_sa_query **sa_query)
1765 {
1766 	struct ib_sa_service_query *query;
1767 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1768 	struct ib_sa_port   *port;
1769 	struct ib_mad_agent *agent;
1770 	struct ib_sa_mad *mad;
1771 	int ret;
1772 
1773 	if (!sa_dev)
1774 		return -ENODEV;
1775 
1776 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1777 	agent = port->agent;
1778 
1779 	if (method != IB_MGMT_METHOD_GET &&
1780 	    method != IB_MGMT_METHOD_SET &&
1781 	    method != IB_SA_METHOD_DELETE)
1782 		return -EINVAL;
1783 
1784 	query = kzalloc(sizeof(*query), gfp_mask);
1785 	if (!query)
1786 		return -ENOMEM;
1787 
1788 	query->sa_query.port     = port;
1789 	ret = alloc_mad(&query->sa_query, gfp_mask);
1790 	if (ret)
1791 		goto err1;
1792 
1793 	ib_sa_client_get(client);
1794 	query->sa_query.client = client;
1795 	query->callback        = callback;
1796 	query->context         = context;
1797 
1798 	mad = query->sa_query.mad_buf->mad;
1799 	init_mad(&query->sa_query, agent);
1800 
1801 	query->sa_query.callback = callback ? ib_sa_service_rec_callback : NULL;
1802 	query->sa_query.release  = ib_sa_service_rec_release;
1803 	mad->mad_hdr.method	 = method;
1804 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_SERVICE_REC);
1805 	mad->sa_hdr.comp_mask	 = comp_mask;
1806 
1807 	ib_pack(service_rec_table, ARRAY_SIZE(service_rec_table),
1808 		rec, mad->data);
1809 
1810 	*sa_query = &query->sa_query;
1811 
1812 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1813 	if (ret < 0)
1814 		goto err2;
1815 
1816 	return ret;
1817 
1818 err2:
1819 	*sa_query = NULL;
1820 	ib_sa_client_put(query->sa_query.client);
1821 	free_mad(&query->sa_query);
1822 
1823 err1:
1824 	kfree(query);
1825 	return ret;
1826 }
1827 EXPORT_SYMBOL(ib_sa_service_rec_query);
1828 
1829 static void ib_sa_mcmember_rec_callback(struct ib_sa_query *sa_query,
1830 					int status,
1831 					struct ib_sa_mad *mad)
1832 {
1833 	struct ib_sa_mcmember_query *query =
1834 		container_of(sa_query, struct ib_sa_mcmember_query, sa_query);
1835 
1836 	if (mad) {
1837 		struct ib_sa_mcmember_rec rec;
1838 
1839 		ib_unpack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
1840 			  mad->data, &rec);
1841 		query->callback(status, &rec, query->context);
1842 	} else
1843 		query->callback(status, NULL, query->context);
1844 }
1845 
1846 static void ib_sa_mcmember_rec_release(struct ib_sa_query *sa_query)
1847 {
1848 	kfree(container_of(sa_query, struct ib_sa_mcmember_query, sa_query));
1849 }
1850 
1851 int ib_sa_mcmember_rec_query(struct ib_sa_client *client,
1852 			     struct ib_device *device, u8 port_num,
1853 			     u8 method,
1854 			     struct ib_sa_mcmember_rec *rec,
1855 			     ib_sa_comp_mask comp_mask,
1856 			     int timeout_ms, gfp_t gfp_mask,
1857 			     void (*callback)(int status,
1858 					      struct ib_sa_mcmember_rec *resp,
1859 					      void *context),
1860 			     void *context,
1861 			     struct ib_sa_query **sa_query)
1862 {
1863 	struct ib_sa_mcmember_query *query;
1864 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1865 	struct ib_sa_port   *port;
1866 	struct ib_mad_agent *agent;
1867 	struct ib_sa_mad *mad;
1868 	int ret;
1869 
1870 	if (!sa_dev)
1871 		return -ENODEV;
1872 
1873 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1874 	agent = port->agent;
1875 
1876 	query = kzalloc(sizeof(*query), gfp_mask);
1877 	if (!query)
1878 		return -ENOMEM;
1879 
1880 	query->sa_query.port     = port;
1881 	ret = alloc_mad(&query->sa_query, gfp_mask);
1882 	if (ret)
1883 		goto err1;
1884 
1885 	ib_sa_client_get(client);
1886 	query->sa_query.client = client;
1887 	query->callback        = callback;
1888 	query->context         = context;
1889 
1890 	mad = query->sa_query.mad_buf->mad;
1891 	init_mad(&query->sa_query, agent);
1892 
1893 	query->sa_query.callback = callback ? ib_sa_mcmember_rec_callback : NULL;
1894 	query->sa_query.release  = ib_sa_mcmember_rec_release;
1895 	mad->mad_hdr.method	 = method;
1896 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_MC_MEMBER_REC);
1897 	mad->sa_hdr.comp_mask	 = comp_mask;
1898 
1899 	ib_pack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
1900 		rec, mad->data);
1901 
1902 	*sa_query = &query->sa_query;
1903 
1904 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1905 	if (ret < 0)
1906 		goto err2;
1907 
1908 	return ret;
1909 
1910 err2:
1911 	*sa_query = NULL;
1912 	ib_sa_client_put(query->sa_query.client);
1913 	free_mad(&query->sa_query);
1914 
1915 err1:
1916 	kfree(query);
1917 	return ret;
1918 }
1919 
1920 /* Support GuidInfoRecord */
1921 static void ib_sa_guidinfo_rec_callback(struct ib_sa_query *sa_query,
1922 					int status,
1923 					struct ib_sa_mad *mad)
1924 {
1925 	struct ib_sa_guidinfo_query *query =
1926 		container_of(sa_query, struct ib_sa_guidinfo_query, sa_query);
1927 
1928 	if (mad) {
1929 		struct ib_sa_guidinfo_rec rec;
1930 
1931 		ib_unpack(guidinfo_rec_table, ARRAY_SIZE(guidinfo_rec_table),
1932 			  mad->data, &rec);
1933 		query->callback(status, &rec, query->context);
1934 	} else
1935 		query->callback(status, NULL, query->context);
1936 }
1937 
1938 static void ib_sa_guidinfo_rec_release(struct ib_sa_query *sa_query)
1939 {
1940 	kfree(container_of(sa_query, struct ib_sa_guidinfo_query, sa_query));
1941 }
1942 
1943 int ib_sa_guid_info_rec_query(struct ib_sa_client *client,
1944 			      struct ib_device *device, u8 port_num,
1945 			      struct ib_sa_guidinfo_rec *rec,
1946 			      ib_sa_comp_mask comp_mask, u8 method,
1947 			      int timeout_ms, gfp_t gfp_mask,
1948 			      void (*callback)(int status,
1949 					       struct ib_sa_guidinfo_rec *resp,
1950 					       void *context),
1951 			      void *context,
1952 			      struct ib_sa_query **sa_query)
1953 {
1954 	struct ib_sa_guidinfo_query *query;
1955 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1956 	struct ib_sa_port *port;
1957 	struct ib_mad_agent *agent;
1958 	struct ib_sa_mad *mad;
1959 	int ret;
1960 
1961 	if (!sa_dev)
1962 		return -ENODEV;
1963 
1964 	if (method != IB_MGMT_METHOD_GET &&
1965 	    method != IB_MGMT_METHOD_SET &&
1966 	    method != IB_SA_METHOD_DELETE) {
1967 		return -EINVAL;
1968 	}
1969 
1970 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1971 	agent = port->agent;
1972 
1973 	query = kzalloc(sizeof(*query), gfp_mask);
1974 	if (!query)
1975 		return -ENOMEM;
1976 
1977 	query->sa_query.port = port;
1978 	ret = alloc_mad(&query->sa_query, gfp_mask);
1979 	if (ret)
1980 		goto err1;
1981 
1982 	ib_sa_client_get(client);
1983 	query->sa_query.client = client;
1984 	query->callback        = callback;
1985 	query->context         = context;
1986 
1987 	mad = query->sa_query.mad_buf->mad;
1988 	init_mad(&query->sa_query, agent);
1989 
1990 	query->sa_query.callback = callback ? ib_sa_guidinfo_rec_callback : NULL;
1991 	query->sa_query.release  = ib_sa_guidinfo_rec_release;
1992 
1993 	mad->mad_hdr.method	 = method;
1994 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_GUID_INFO_REC);
1995 	mad->sa_hdr.comp_mask	 = comp_mask;
1996 
1997 	ib_pack(guidinfo_rec_table, ARRAY_SIZE(guidinfo_rec_table), rec,
1998 		mad->data);
1999 
2000 	*sa_query = &query->sa_query;
2001 
2002 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
2003 	if (ret < 0)
2004 		goto err2;
2005 
2006 	return ret;
2007 
2008 err2:
2009 	*sa_query = NULL;
2010 	ib_sa_client_put(query->sa_query.client);
2011 	free_mad(&query->sa_query);
2012 
2013 err1:
2014 	kfree(query);
2015 	return ret;
2016 }
2017 EXPORT_SYMBOL(ib_sa_guid_info_rec_query);
2018 
2019 bool ib_sa_sendonly_fullmem_support(struct ib_sa_client *client,
2020 				    struct ib_device *device,
2021 				    u8 port_num)
2022 {
2023 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
2024 	struct ib_sa_port *port;
2025 	bool ret = false;
2026 	unsigned long flags;
2027 
2028 	if (!sa_dev)
2029 		return ret;
2030 
2031 	port  = &sa_dev->port[port_num - sa_dev->start_port];
2032 
2033 	spin_lock_irqsave(&port->classport_lock, flags);
2034 	if ((port->classport_info.valid) &&
2035 	    (port->classport_info.data.type == RDMA_CLASS_PORT_INFO_IB))
2036 		ret = ib_get_cpi_capmask2(&port->classport_info.data.ib)
2037 			& IB_SA_CAP_MASK2_SENDONLY_FULL_MEM_SUPPORT;
2038 	spin_unlock_irqrestore(&port->classport_lock, flags);
2039 	return ret;
2040 }
2041 EXPORT_SYMBOL(ib_sa_sendonly_fullmem_support);
2042 
2043 struct ib_classport_info_context {
2044 	struct completion	done;
2045 	struct ib_sa_query	*sa_query;
2046 };
2047 
2048 static void ib_classportinfo_cb(void *context)
2049 {
2050 	struct ib_classport_info_context *cb_ctx = context;
2051 
2052 	complete(&cb_ctx->done);
2053 }
2054 
2055 static void ib_sa_classport_info_rec_callback(struct ib_sa_query *sa_query,
2056 					      int status,
2057 					      struct ib_sa_mad *mad)
2058 {
2059 	unsigned long flags;
2060 	struct ib_sa_classport_info_query *query =
2061 		container_of(sa_query, struct ib_sa_classport_info_query, sa_query);
2062 	struct ib_sa_classport_cache *info = &sa_query->port->classport_info;
2063 
2064 	if (mad) {
2065 		if (sa_query->flags & IB_SA_QUERY_OPA) {
2066 			struct opa_class_port_info rec;
2067 
2068 			ib_unpack(opa_classport_info_rec_table,
2069 				  ARRAY_SIZE(opa_classport_info_rec_table),
2070 				  mad->data, &rec);
2071 
2072 			spin_lock_irqsave(&sa_query->port->classport_lock,
2073 					  flags);
2074 			if (!status && !info->valid) {
2075 				memcpy(&info->data.opa, &rec,
2076 				       sizeof(info->data.opa));
2077 
2078 				info->valid = true;
2079 				info->data.type = RDMA_CLASS_PORT_INFO_OPA;
2080 			}
2081 			spin_unlock_irqrestore(&sa_query->port->classport_lock,
2082 					       flags);
2083 
2084 		} else {
2085 			struct ib_class_port_info rec;
2086 
2087 			ib_unpack(ib_classport_info_rec_table,
2088 				  ARRAY_SIZE(ib_classport_info_rec_table),
2089 				  mad->data, &rec);
2090 
2091 			spin_lock_irqsave(&sa_query->port->classport_lock,
2092 					  flags);
2093 			if (!status && !info->valid) {
2094 				memcpy(&info->data.ib, &rec,
2095 				       sizeof(info->data.ib));
2096 
2097 				info->valid = true;
2098 				info->data.type = RDMA_CLASS_PORT_INFO_IB;
2099 			}
2100 			spin_unlock_irqrestore(&sa_query->port->classport_lock,
2101 					       flags);
2102 		}
2103 	}
2104 	query->callback(query->context);
2105 }
2106 
2107 static void ib_sa_classport_info_rec_release(struct ib_sa_query *sa_query)
2108 {
2109 	kfree(container_of(sa_query, struct ib_sa_classport_info_query,
2110 			   sa_query));
2111 }
2112 
2113 static int ib_sa_classport_info_rec_query(struct ib_sa_port *port,
2114 					  int timeout_ms,
2115 					  void (*callback)(void *context),
2116 					  void *context,
2117 					  struct ib_sa_query **sa_query)
2118 {
2119 	struct ib_mad_agent *agent;
2120 	struct ib_sa_classport_info_query *query;
2121 	struct ib_sa_mad *mad;
2122 	gfp_t gfp_mask = GFP_KERNEL;
2123 	int ret;
2124 
2125 	agent = port->agent;
2126 
2127 	query = kzalloc(sizeof(*query), gfp_mask);
2128 	if (!query)
2129 		return -ENOMEM;
2130 
2131 	query->sa_query.port = port;
2132 	query->sa_query.flags |= rdma_cap_opa_ah(port->agent->device,
2133 						 port->port_num) ?
2134 				 IB_SA_QUERY_OPA : 0;
2135 	ret = alloc_mad(&query->sa_query, gfp_mask);
2136 	if (ret)
2137 		goto err_free;
2138 
2139 	query->callback = callback;
2140 	query->context = context;
2141 
2142 	mad = query->sa_query.mad_buf->mad;
2143 	init_mad(&query->sa_query, agent);
2144 
2145 	query->sa_query.callback = ib_sa_classport_info_rec_callback;
2146 	query->sa_query.release  = ib_sa_classport_info_rec_release;
2147 	mad->mad_hdr.method	 = IB_MGMT_METHOD_GET;
2148 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_CLASS_PORTINFO);
2149 	mad->sa_hdr.comp_mask	 = 0;
2150 	*sa_query = &query->sa_query;
2151 
2152 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
2153 	if (ret < 0)
2154 		goto err_free_mad;
2155 
2156 	return ret;
2157 
2158 err_free_mad:
2159 	*sa_query = NULL;
2160 	free_mad(&query->sa_query);
2161 
2162 err_free:
2163 	kfree(query);
2164 	return ret;
2165 }
2166 
2167 static void update_ib_cpi(struct work_struct *work)
2168 {
2169 	struct ib_sa_port *port =
2170 		container_of(work, struct ib_sa_port, ib_cpi_work.work);
2171 	struct ib_classport_info_context *cb_context;
2172 	unsigned long flags;
2173 	int ret;
2174 
2175 	/* If the classport info is valid, nothing
2176 	 * to do here.
2177 	 */
2178 	spin_lock_irqsave(&port->classport_lock, flags);
2179 	if (port->classport_info.valid) {
2180 		spin_unlock_irqrestore(&port->classport_lock, flags);
2181 		return;
2182 	}
2183 	spin_unlock_irqrestore(&port->classport_lock, flags);
2184 
2185 	cb_context = kmalloc(sizeof(*cb_context), GFP_KERNEL);
2186 	if (!cb_context)
2187 		goto err_nomem;
2188 
2189 	init_completion(&cb_context->done);
2190 
2191 	ret = ib_sa_classport_info_rec_query(port, 3000,
2192 					     ib_classportinfo_cb, cb_context,
2193 					     &cb_context->sa_query);
2194 	if (ret < 0)
2195 		goto free_cb_err;
2196 	wait_for_completion(&cb_context->done);
2197 free_cb_err:
2198 	kfree(cb_context);
2199 	spin_lock_irqsave(&port->classport_lock, flags);
2200 
2201 	/* If the classport info is still not valid, the query should have
2202 	 * failed for some reason. Retry issuing the query
2203 	 */
2204 	if (!port->classport_info.valid) {
2205 		port->classport_info.retry_cnt++;
2206 		if (port->classport_info.retry_cnt <=
2207 		    IB_SA_CPI_MAX_RETRY_CNT) {
2208 			unsigned long delay =
2209 				msecs_to_jiffies(IB_SA_CPI_RETRY_WAIT);
2210 
2211 			queue_delayed_work(ib_wq, &port->ib_cpi_work, delay);
2212 		}
2213 	}
2214 	spin_unlock_irqrestore(&port->classport_lock, flags);
2215 
2216 err_nomem:
2217 	return;
2218 }
2219 
2220 static void send_handler(struct ib_mad_agent *agent,
2221 			 struct ib_mad_send_wc *mad_send_wc)
2222 {
2223 	struct ib_sa_query *query = mad_send_wc->send_buf->context[0];
2224 	unsigned long flags;
2225 
2226 	if (query->callback)
2227 		switch (mad_send_wc->status) {
2228 		case IB_WC_SUCCESS:
2229 			/* No callback -- already got recv */
2230 			break;
2231 		case IB_WC_RESP_TIMEOUT_ERR:
2232 			query->callback(query, -ETIMEDOUT, NULL);
2233 			break;
2234 		case IB_WC_WR_FLUSH_ERR:
2235 			query->callback(query, -EINTR, NULL);
2236 			break;
2237 		default:
2238 			query->callback(query, -EIO, NULL);
2239 			break;
2240 		}
2241 
2242 	spin_lock_irqsave(&idr_lock, flags);
2243 	idr_remove(&query_idr, query->id);
2244 	spin_unlock_irqrestore(&idr_lock, flags);
2245 
2246 	free_mad(query);
2247 	if (query->client)
2248 		ib_sa_client_put(query->client);
2249 	query->release(query);
2250 }
2251 
2252 static void recv_handler(struct ib_mad_agent *mad_agent,
2253 			 struct ib_mad_send_buf *send_buf,
2254 			 struct ib_mad_recv_wc *mad_recv_wc)
2255 {
2256 	struct ib_sa_query *query;
2257 
2258 	if (!send_buf)
2259 		return;
2260 
2261 	query = send_buf->context[0];
2262 	if (query->callback) {
2263 		if (mad_recv_wc->wc->status == IB_WC_SUCCESS)
2264 			query->callback(query,
2265 					mad_recv_wc->recv_buf.mad->mad_hdr.status ?
2266 					-EINVAL : 0,
2267 					(struct ib_sa_mad *) mad_recv_wc->recv_buf.mad);
2268 		else
2269 			query->callback(query, -EIO, NULL);
2270 	}
2271 
2272 	ib_free_recv_mad(mad_recv_wc);
2273 }
2274 
2275 static void update_sm_ah(struct work_struct *work)
2276 {
2277 	struct ib_sa_port *port =
2278 		container_of(work, struct ib_sa_port, update_task);
2279 	struct ib_sa_sm_ah *new_ah;
2280 	struct ib_port_attr port_attr;
2281 	struct rdma_ah_attr   ah_attr;
2282 
2283 	if (ib_query_port(port->agent->device, port->port_num, &port_attr)) {
2284 		pr_warn("Couldn't query port\n");
2285 		return;
2286 	}
2287 
2288 	new_ah = kmalloc(sizeof(*new_ah), GFP_KERNEL);
2289 	if (!new_ah)
2290 		return;
2291 
2292 	kref_init(&new_ah->ref);
2293 	new_ah->src_path_mask = (1 << port_attr.lmc) - 1;
2294 
2295 	new_ah->pkey_index = 0;
2296 	if (ib_find_pkey(port->agent->device, port->port_num,
2297 			 IB_DEFAULT_PKEY_FULL, &new_ah->pkey_index))
2298 		pr_err("Couldn't find index for default PKey\n");
2299 
2300 	memset(&ah_attr, 0, sizeof(ah_attr));
2301 	ah_attr.type = rdma_ah_find_type(port->agent->device,
2302 					 port->port_num);
2303 	rdma_ah_set_dlid(&ah_attr, port_attr.sm_lid);
2304 	rdma_ah_set_sl(&ah_attr, port_attr.sm_sl);
2305 	rdma_ah_set_port_num(&ah_attr, port->port_num);
2306 	if (port_attr.grh_required) {
2307 		if (ah_attr.type == RDMA_AH_ATTR_TYPE_OPA) {
2308 			rdma_ah_set_make_grd(&ah_attr, true);
2309 		} else {
2310 			rdma_ah_set_ah_flags(&ah_attr, IB_AH_GRH);
2311 			rdma_ah_set_subnet_prefix(&ah_attr,
2312 						  cpu_to_be64(port_attr.subnet_prefix));
2313 			rdma_ah_set_interface_id(&ah_attr,
2314 						 cpu_to_be64(IB_SA_WELL_KNOWN_GUID));
2315 		}
2316 	}
2317 
2318 	new_ah->ah = rdma_create_ah(port->agent->qp->pd, &ah_attr);
2319 	if (IS_ERR(new_ah->ah)) {
2320 		pr_warn("Couldn't create new SM AH\n");
2321 		kfree(new_ah);
2322 		return;
2323 	}
2324 
2325 	spin_lock_irq(&port->ah_lock);
2326 	if (port->sm_ah)
2327 		kref_put(&port->sm_ah->ref, free_sm_ah);
2328 	port->sm_ah = new_ah;
2329 	spin_unlock_irq(&port->ah_lock);
2330 }
2331 
2332 static void ib_sa_event(struct ib_event_handler *handler,
2333 			struct ib_event *event)
2334 {
2335 	if (event->event == IB_EVENT_PORT_ERR    ||
2336 	    event->event == IB_EVENT_PORT_ACTIVE ||
2337 	    event->event == IB_EVENT_LID_CHANGE  ||
2338 	    event->event == IB_EVENT_PKEY_CHANGE ||
2339 	    event->event == IB_EVENT_SM_CHANGE   ||
2340 	    event->event == IB_EVENT_CLIENT_REREGISTER) {
2341 		unsigned long flags;
2342 		struct ib_sa_device *sa_dev =
2343 			container_of(handler, typeof(*sa_dev), event_handler);
2344 		u8 port_num = event->element.port_num - sa_dev->start_port;
2345 		struct ib_sa_port *port = &sa_dev->port[port_num];
2346 
2347 		if (!rdma_cap_ib_sa(handler->device, port->port_num))
2348 			return;
2349 
2350 		spin_lock_irqsave(&port->ah_lock, flags);
2351 		if (port->sm_ah)
2352 			kref_put(&port->sm_ah->ref, free_sm_ah);
2353 		port->sm_ah = NULL;
2354 		spin_unlock_irqrestore(&port->ah_lock, flags);
2355 
2356 		if (event->event == IB_EVENT_SM_CHANGE ||
2357 		    event->event == IB_EVENT_CLIENT_REREGISTER ||
2358 		    event->event == IB_EVENT_LID_CHANGE ||
2359 		    event->event == IB_EVENT_PORT_ACTIVE) {
2360 			unsigned long delay =
2361 				msecs_to_jiffies(IB_SA_CPI_RETRY_WAIT);
2362 
2363 			spin_lock_irqsave(&port->classport_lock, flags);
2364 			port->classport_info.valid = false;
2365 			port->classport_info.retry_cnt = 0;
2366 			spin_unlock_irqrestore(&port->classport_lock, flags);
2367 			queue_delayed_work(ib_wq,
2368 					   &port->ib_cpi_work, delay);
2369 		}
2370 		queue_work(ib_wq, &sa_dev->port[port_num].update_task);
2371 	}
2372 }
2373 
2374 static void ib_sa_add_one(struct ib_device *device)
2375 {
2376 	struct ib_sa_device *sa_dev;
2377 	int s, e, i;
2378 	int count = 0;
2379 
2380 	s = rdma_start_port(device);
2381 	e = rdma_end_port(device);
2382 
2383 	sa_dev = kzalloc(sizeof *sa_dev +
2384 			 (e - s + 1) * sizeof (struct ib_sa_port),
2385 			 GFP_KERNEL);
2386 	if (!sa_dev)
2387 		return;
2388 
2389 	sa_dev->start_port = s;
2390 	sa_dev->end_port   = e;
2391 
2392 	for (i = 0; i <= e - s; ++i) {
2393 		spin_lock_init(&sa_dev->port[i].ah_lock);
2394 		if (!rdma_cap_ib_sa(device, i + 1))
2395 			continue;
2396 
2397 		sa_dev->port[i].sm_ah    = NULL;
2398 		sa_dev->port[i].port_num = i + s;
2399 
2400 		spin_lock_init(&sa_dev->port[i].classport_lock);
2401 		sa_dev->port[i].classport_info.valid = false;
2402 
2403 		sa_dev->port[i].agent =
2404 			ib_register_mad_agent(device, i + s, IB_QPT_GSI,
2405 					      NULL, 0, send_handler,
2406 					      recv_handler, sa_dev, 0);
2407 		if (IS_ERR(sa_dev->port[i].agent))
2408 			goto err;
2409 
2410 		INIT_WORK(&sa_dev->port[i].update_task, update_sm_ah);
2411 		INIT_DELAYED_WORK(&sa_dev->port[i].ib_cpi_work,
2412 				  update_ib_cpi);
2413 
2414 		count++;
2415 	}
2416 
2417 	if (!count)
2418 		goto free;
2419 
2420 	ib_set_client_data(device, &sa_client, sa_dev);
2421 
2422 	/*
2423 	 * We register our event handler after everything is set up,
2424 	 * and then update our cached info after the event handler is
2425 	 * registered to avoid any problems if a port changes state
2426 	 * during our initialization.
2427 	 */
2428 
2429 	INIT_IB_EVENT_HANDLER(&sa_dev->event_handler, device, ib_sa_event);
2430 	ib_register_event_handler(&sa_dev->event_handler);
2431 
2432 	for (i = 0; i <= e - s; ++i) {
2433 		if (rdma_cap_ib_sa(device, i + 1))
2434 			update_sm_ah(&sa_dev->port[i].update_task);
2435 	}
2436 
2437 	return;
2438 
2439 err:
2440 	while (--i >= 0) {
2441 		if (rdma_cap_ib_sa(device, i + 1))
2442 			ib_unregister_mad_agent(sa_dev->port[i].agent);
2443 	}
2444 free:
2445 	kfree(sa_dev);
2446 	return;
2447 }
2448 
2449 static void ib_sa_remove_one(struct ib_device *device, void *client_data)
2450 {
2451 	struct ib_sa_device *sa_dev = client_data;
2452 	int i;
2453 
2454 	if (!sa_dev)
2455 		return;
2456 
2457 	ib_unregister_event_handler(&sa_dev->event_handler);
2458 	flush_workqueue(ib_wq);
2459 
2460 	for (i = 0; i <= sa_dev->end_port - sa_dev->start_port; ++i) {
2461 		if (rdma_cap_ib_sa(device, i + 1)) {
2462 			cancel_delayed_work_sync(&sa_dev->port[i].ib_cpi_work);
2463 			ib_unregister_mad_agent(sa_dev->port[i].agent);
2464 			if (sa_dev->port[i].sm_ah)
2465 				kref_put(&sa_dev->port[i].sm_ah->ref, free_sm_ah);
2466 		}
2467 
2468 	}
2469 
2470 	kfree(sa_dev);
2471 }
2472 
2473 int ib_sa_init(void)
2474 {
2475 	int ret;
2476 
2477 	get_random_bytes(&tid, sizeof tid);
2478 
2479 	atomic_set(&ib_nl_sa_request_seq, 0);
2480 
2481 	ret = ib_register_client(&sa_client);
2482 	if (ret) {
2483 		pr_err("Couldn't register ib_sa client\n");
2484 		goto err1;
2485 	}
2486 
2487 	ret = mcast_init();
2488 	if (ret) {
2489 		pr_err("Couldn't initialize multicast handling\n");
2490 		goto err2;
2491 	}
2492 
2493 	ib_nl_wq = alloc_ordered_workqueue("ib_nl_sa_wq", WQ_MEM_RECLAIM);
2494 	if (!ib_nl_wq) {
2495 		ret = -ENOMEM;
2496 		goto err3;
2497 	}
2498 
2499 	INIT_DELAYED_WORK(&ib_nl_timed_work, ib_nl_request_timeout);
2500 
2501 	return 0;
2502 
2503 err3:
2504 	mcast_cleanup();
2505 err2:
2506 	ib_unregister_client(&sa_client);
2507 err1:
2508 	return ret;
2509 }
2510 
2511 void ib_sa_cleanup(void)
2512 {
2513 	cancel_delayed_work(&ib_nl_timed_work);
2514 	flush_workqueue(ib_nl_wq);
2515 	destroy_workqueue(ib_nl_wq);
2516 	mcast_cleanup();
2517 	ib_unregister_client(&sa_client);
2518 	idr_destroy(&query_idr);
2519 }
2520