xref: /linux/drivers/infiniband/core/sa_query.c (revision 68a052239fc4b351e961f698b824f7654a346091)
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/init.h>
36 #include <linux/err.h>
37 #include <linux/random.h>
38 #include <linux/spinlock.h>
39 #include <linux/slab.h>
40 #include <linux/dma-mapping.h>
41 #include <linux/kref.h>
42 #include <linux/xarray.h>
43 #include <linux/workqueue.h>
44 #include <uapi/linux/if_ether.h>
45 #include <rdma/ib_pack.h>
46 #include <rdma/ib_cache.h>
47 #include <rdma/rdma_netlink.h>
48 #include <net/netlink.h>
49 #include <uapi/rdma/ib_user_sa.h>
50 #include <rdma/ib_marshall.h>
51 #include <rdma/ib_addr.h>
52 #include <rdma/opa_addr.h>
53 #include <rdma/rdma_cm.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 	u32		     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[];
105 };
106 
107 struct ib_sa_query {
108 	void (*callback)(struct ib_sa_query *sa_query, int status,
109 			 struct ib_sa_mad *mad);
110 	void (*rmpp_callback)(struct ib_sa_query *sa_query, int status,
111 			      struct ib_mad_recv_wc *mad);
112 	void (*release)(struct ib_sa_query *);
113 	struct ib_sa_client    *client;
114 	struct ib_sa_port      *port;
115 	struct ib_mad_send_buf *mad_buf;
116 	struct ib_sa_sm_ah     *sm_ah;
117 	int			id;
118 	u32			flags;
119 	struct list_head	list; /* Local svc request list */
120 	u32			seq; /* Local svc request sequence number */
121 	unsigned long		timeout; /* Local svc timeout */
122 	u8			path_use; /* How will the pathrecord be used */
123 };
124 
125 #define IB_SA_ENABLE_LOCAL_SERVICE	0x00000001
126 #define IB_SA_CANCEL			0x00000002
127 #define IB_SA_QUERY_OPA			0x00000004
128 
129 struct ib_sa_path_query {
130 	void (*callback)(int status, struct sa_path_rec *rec,
131 			 unsigned int num_paths, void *context);
132 	void *context;
133 	struct ib_sa_query sa_query;
134 	struct sa_path_rec *conv_pr;
135 };
136 
137 struct ib_sa_guidinfo_query {
138 	void (*callback)(int, struct ib_sa_guidinfo_rec *, void *);
139 	void *context;
140 	struct ib_sa_query sa_query;
141 };
142 
143 struct ib_sa_classport_info_query {
144 	void (*callback)(void *);
145 	void *context;
146 	struct ib_sa_query sa_query;
147 };
148 
149 struct ib_sa_mcmember_query {
150 	void (*callback)(int, struct ib_sa_mcmember_rec *, void *);
151 	void *context;
152 	struct ib_sa_query sa_query;
153 };
154 
155 struct ib_sa_service_query {
156 	void (*callback)(int status, struct sa_service_rec *rec,
157 			 unsigned int num_services, void *context);
158 	void *context;
159 	struct ib_sa_query sa_query;
160 };
161 
162 static LIST_HEAD(ib_nl_request_list);
163 static DEFINE_SPINLOCK(ib_nl_request_lock);
164 static atomic_t ib_nl_sa_request_seq;
165 static struct workqueue_struct *ib_nl_wq;
166 static struct delayed_work ib_nl_timed_work;
167 static const struct nla_policy ib_nl_policy[LS_NLA_TYPE_MAX] = {
168 	[LS_NLA_TYPE_PATH_RECORD]	= {.type = NLA_BINARY,
169 		.len = sizeof(struct ib_path_rec_data)},
170 	[LS_NLA_TYPE_TIMEOUT]		= {.type = NLA_U32},
171 	[LS_NLA_TYPE_SERVICE_ID]	= {.type = NLA_U64},
172 	[LS_NLA_TYPE_DGID]		= {.type = NLA_BINARY,
173 		.len = sizeof(struct rdma_nla_ls_gid)},
174 	[LS_NLA_TYPE_SGID]		= {.type = NLA_BINARY,
175 		.len = sizeof(struct rdma_nla_ls_gid)},
176 	[LS_NLA_TYPE_TCLASS]		= {.type = NLA_U8},
177 	[LS_NLA_TYPE_PKEY]		= {.type = NLA_U16},
178 	[LS_NLA_TYPE_QOS_CLASS]		= {.type = NLA_U16},
179 };
180 
181 
182 static int ib_sa_add_one(struct ib_device *device);
183 static void ib_sa_remove_one(struct ib_device *device, void *client_data);
184 
185 static struct ib_client sa_client = {
186 	.name   = "sa",
187 	.add    = ib_sa_add_one,
188 	.remove = ib_sa_remove_one
189 };
190 
191 static DEFINE_XARRAY_FLAGS(queries, XA_FLAGS_ALLOC | XA_FLAGS_LOCK_IRQ);
192 
193 static DEFINE_SPINLOCK(tid_lock);
194 static u32 tid;
195 
196 #define PATH_REC_FIELD(field) \
197 	.struct_offset_bytes = offsetof(struct sa_path_rec, field),	\
198 	.struct_size_bytes   = sizeof_field(struct sa_path_rec, field),	\
199 	.field_name          = "sa_path_rec:" #field
200 
201 static const struct ib_field path_rec_table[] = {
202 	{ PATH_REC_FIELD(service_id),
203 	  .offset_words = 0,
204 	  .offset_bits  = 0,
205 	  .size_bits    = 64 },
206 	{ PATH_REC_FIELD(dgid),
207 	  .offset_words = 2,
208 	  .offset_bits  = 0,
209 	  .size_bits    = 128 },
210 	{ PATH_REC_FIELD(sgid),
211 	  .offset_words = 6,
212 	  .offset_bits  = 0,
213 	  .size_bits    = 128 },
214 	{ PATH_REC_FIELD(ib.dlid),
215 	  .offset_words = 10,
216 	  .offset_bits  = 0,
217 	  .size_bits    = 16 },
218 	{ PATH_REC_FIELD(ib.slid),
219 	  .offset_words = 10,
220 	  .offset_bits  = 16,
221 	  .size_bits    = 16 },
222 	{ PATH_REC_FIELD(ib.raw_traffic),
223 	  .offset_words = 11,
224 	  .offset_bits  = 0,
225 	  .size_bits    = 1 },
226 	{ RESERVED,
227 	  .offset_words = 11,
228 	  .offset_bits  = 1,
229 	  .size_bits    = 3 },
230 	{ PATH_REC_FIELD(flow_label),
231 	  .offset_words = 11,
232 	  .offset_bits  = 4,
233 	  .size_bits    = 20 },
234 	{ PATH_REC_FIELD(hop_limit),
235 	  .offset_words = 11,
236 	  .offset_bits  = 24,
237 	  .size_bits    = 8 },
238 	{ PATH_REC_FIELD(traffic_class),
239 	  .offset_words = 12,
240 	  .offset_bits  = 0,
241 	  .size_bits    = 8 },
242 	{ PATH_REC_FIELD(reversible),
243 	  .offset_words = 12,
244 	  .offset_bits  = 8,
245 	  .size_bits    = 1 },
246 	{ PATH_REC_FIELD(numb_path),
247 	  .offset_words = 12,
248 	  .offset_bits  = 9,
249 	  .size_bits    = 7 },
250 	{ PATH_REC_FIELD(pkey),
251 	  .offset_words = 12,
252 	  .offset_bits  = 16,
253 	  .size_bits    = 16 },
254 	{ PATH_REC_FIELD(qos_class),
255 	  .offset_words = 13,
256 	  .offset_bits  = 0,
257 	  .size_bits    = 12 },
258 	{ PATH_REC_FIELD(sl),
259 	  .offset_words = 13,
260 	  .offset_bits  = 12,
261 	  .size_bits    = 4 },
262 	{ PATH_REC_FIELD(mtu_selector),
263 	  .offset_words = 13,
264 	  .offset_bits  = 16,
265 	  .size_bits    = 2 },
266 	{ PATH_REC_FIELD(mtu),
267 	  .offset_words = 13,
268 	  .offset_bits  = 18,
269 	  .size_bits    = 6 },
270 	{ PATH_REC_FIELD(rate_selector),
271 	  .offset_words = 13,
272 	  .offset_bits  = 24,
273 	  .size_bits    = 2 },
274 	{ PATH_REC_FIELD(rate),
275 	  .offset_words = 13,
276 	  .offset_bits  = 26,
277 	  .size_bits    = 6 },
278 	{ PATH_REC_FIELD(packet_life_time_selector),
279 	  .offset_words = 14,
280 	  .offset_bits  = 0,
281 	  .size_bits    = 2 },
282 	{ PATH_REC_FIELD(packet_life_time),
283 	  .offset_words = 14,
284 	  .offset_bits  = 2,
285 	  .size_bits    = 6 },
286 	{ PATH_REC_FIELD(preference),
287 	  .offset_words = 14,
288 	  .offset_bits  = 8,
289 	  .size_bits    = 8 },
290 	{ RESERVED,
291 	  .offset_words = 14,
292 	  .offset_bits  = 16,
293 	  .size_bits    = 48 },
294 };
295 
296 #define OPA_PATH_REC_FIELD(field) \
297 	.struct_offset_bytes = \
298 		offsetof(struct sa_path_rec, field), \
299 	.struct_size_bytes   = \
300 		sizeof_field(struct sa_path_rec, field),	\
301 	.field_name          = "sa_path_rec:" #field
302 
303 static const struct ib_field opa_path_rec_table[] = {
304 	{ OPA_PATH_REC_FIELD(service_id),
305 	  .offset_words = 0,
306 	  .offset_bits  = 0,
307 	  .size_bits    = 64 },
308 	{ OPA_PATH_REC_FIELD(dgid),
309 	  .offset_words = 2,
310 	  .offset_bits  = 0,
311 	  .size_bits    = 128 },
312 	{ OPA_PATH_REC_FIELD(sgid),
313 	  .offset_words = 6,
314 	  .offset_bits  = 0,
315 	  .size_bits    = 128 },
316 	{ OPA_PATH_REC_FIELD(opa.dlid),
317 	  .offset_words = 10,
318 	  .offset_bits  = 0,
319 	  .size_bits    = 32 },
320 	{ OPA_PATH_REC_FIELD(opa.slid),
321 	  .offset_words = 11,
322 	  .offset_bits  = 0,
323 	  .size_bits    = 32 },
324 	{ OPA_PATH_REC_FIELD(opa.raw_traffic),
325 	  .offset_words = 12,
326 	  .offset_bits  = 0,
327 	  .size_bits    = 1 },
328 	{ RESERVED,
329 	  .offset_words = 12,
330 	  .offset_bits  = 1,
331 	  .size_bits    = 3 },
332 	{ OPA_PATH_REC_FIELD(flow_label),
333 	  .offset_words = 12,
334 	  .offset_bits  = 4,
335 	  .size_bits    = 20 },
336 	{ OPA_PATH_REC_FIELD(hop_limit),
337 	  .offset_words = 12,
338 	  .offset_bits  = 24,
339 	  .size_bits    = 8 },
340 	{ OPA_PATH_REC_FIELD(traffic_class),
341 	  .offset_words = 13,
342 	  .offset_bits  = 0,
343 	  .size_bits    = 8 },
344 	{ OPA_PATH_REC_FIELD(reversible),
345 	  .offset_words = 13,
346 	  .offset_bits  = 8,
347 	  .size_bits    = 1 },
348 	{ OPA_PATH_REC_FIELD(numb_path),
349 	  .offset_words = 13,
350 	  .offset_bits  = 9,
351 	  .size_bits    = 7 },
352 	{ OPA_PATH_REC_FIELD(pkey),
353 	  .offset_words = 13,
354 	  .offset_bits  = 16,
355 	  .size_bits    = 16 },
356 	{ OPA_PATH_REC_FIELD(opa.l2_8B),
357 	  .offset_words = 14,
358 	  .offset_bits  = 0,
359 	  .size_bits    = 1 },
360 	{ OPA_PATH_REC_FIELD(opa.l2_10B),
361 	  .offset_words = 14,
362 	  .offset_bits  = 1,
363 	  .size_bits    = 1 },
364 	{ OPA_PATH_REC_FIELD(opa.l2_9B),
365 	  .offset_words = 14,
366 	  .offset_bits  = 2,
367 	  .size_bits    = 1 },
368 	{ OPA_PATH_REC_FIELD(opa.l2_16B),
369 	  .offset_words = 14,
370 	  .offset_bits  = 3,
371 	  .size_bits    = 1 },
372 	{ RESERVED,
373 	  .offset_words = 14,
374 	  .offset_bits  = 4,
375 	  .size_bits    = 2 },
376 	{ OPA_PATH_REC_FIELD(opa.qos_type),
377 	  .offset_words = 14,
378 	  .offset_bits  = 6,
379 	  .size_bits    = 2 },
380 	{ OPA_PATH_REC_FIELD(opa.qos_priority),
381 	  .offset_words = 14,
382 	  .offset_bits  = 8,
383 	  .size_bits    = 8 },
384 	{ RESERVED,
385 	  .offset_words = 14,
386 	  .offset_bits  = 16,
387 	  .size_bits    = 3 },
388 	{ OPA_PATH_REC_FIELD(sl),
389 	  .offset_words = 14,
390 	  .offset_bits  = 19,
391 	  .size_bits    = 5 },
392 	{ RESERVED,
393 	  .offset_words = 14,
394 	  .offset_bits  = 24,
395 	  .size_bits    = 8 },
396 	{ OPA_PATH_REC_FIELD(mtu_selector),
397 	  .offset_words = 15,
398 	  .offset_bits  = 0,
399 	  .size_bits    = 2 },
400 	{ OPA_PATH_REC_FIELD(mtu),
401 	  .offset_words = 15,
402 	  .offset_bits  = 2,
403 	  .size_bits    = 6 },
404 	{ OPA_PATH_REC_FIELD(rate_selector),
405 	  .offset_words = 15,
406 	  .offset_bits  = 8,
407 	  .size_bits    = 2 },
408 	{ OPA_PATH_REC_FIELD(rate),
409 	  .offset_words = 15,
410 	  .offset_bits  = 10,
411 	  .size_bits    = 6 },
412 	{ OPA_PATH_REC_FIELD(packet_life_time_selector),
413 	  .offset_words = 15,
414 	  .offset_bits  = 16,
415 	  .size_bits    = 2 },
416 	{ OPA_PATH_REC_FIELD(packet_life_time),
417 	  .offset_words = 15,
418 	  .offset_bits  = 18,
419 	  .size_bits    = 6 },
420 	{ OPA_PATH_REC_FIELD(preference),
421 	  .offset_words = 15,
422 	  .offset_bits  = 24,
423 	  .size_bits    = 8 },
424 };
425 
426 #define MCMEMBER_REC_FIELD(field) \
427 	.struct_offset_bytes = offsetof(struct ib_sa_mcmember_rec, field),	\
428 	.struct_size_bytes   = sizeof_field(struct ib_sa_mcmember_rec, field),	\
429 	.field_name          = "sa_mcmember_rec:" #field
430 
431 static const struct ib_field mcmember_rec_table[] = {
432 	{ MCMEMBER_REC_FIELD(mgid),
433 	  .offset_words = 0,
434 	  .offset_bits  = 0,
435 	  .size_bits    = 128 },
436 	{ MCMEMBER_REC_FIELD(port_gid),
437 	  .offset_words = 4,
438 	  .offset_bits  = 0,
439 	  .size_bits    = 128 },
440 	{ MCMEMBER_REC_FIELD(qkey),
441 	  .offset_words = 8,
442 	  .offset_bits  = 0,
443 	  .size_bits    = 32 },
444 	{ MCMEMBER_REC_FIELD(mlid),
445 	  .offset_words = 9,
446 	  .offset_bits  = 0,
447 	  .size_bits    = 16 },
448 	{ MCMEMBER_REC_FIELD(mtu_selector),
449 	  .offset_words = 9,
450 	  .offset_bits  = 16,
451 	  .size_bits    = 2 },
452 	{ MCMEMBER_REC_FIELD(mtu),
453 	  .offset_words = 9,
454 	  .offset_bits  = 18,
455 	  .size_bits    = 6 },
456 	{ MCMEMBER_REC_FIELD(traffic_class),
457 	  .offset_words = 9,
458 	  .offset_bits  = 24,
459 	  .size_bits    = 8 },
460 	{ MCMEMBER_REC_FIELD(pkey),
461 	  .offset_words = 10,
462 	  .offset_bits  = 0,
463 	  .size_bits    = 16 },
464 	{ MCMEMBER_REC_FIELD(rate_selector),
465 	  .offset_words = 10,
466 	  .offset_bits  = 16,
467 	  .size_bits    = 2 },
468 	{ MCMEMBER_REC_FIELD(rate),
469 	  .offset_words = 10,
470 	  .offset_bits  = 18,
471 	  .size_bits    = 6 },
472 	{ MCMEMBER_REC_FIELD(packet_life_time_selector),
473 	  .offset_words = 10,
474 	  .offset_bits  = 24,
475 	  .size_bits    = 2 },
476 	{ MCMEMBER_REC_FIELD(packet_life_time),
477 	  .offset_words = 10,
478 	  .offset_bits  = 26,
479 	  .size_bits    = 6 },
480 	{ MCMEMBER_REC_FIELD(sl),
481 	  .offset_words = 11,
482 	  .offset_bits  = 0,
483 	  .size_bits    = 4 },
484 	{ MCMEMBER_REC_FIELD(flow_label),
485 	  .offset_words = 11,
486 	  .offset_bits  = 4,
487 	  .size_bits    = 20 },
488 	{ MCMEMBER_REC_FIELD(hop_limit),
489 	  .offset_words = 11,
490 	  .offset_bits  = 24,
491 	  .size_bits    = 8 },
492 	{ MCMEMBER_REC_FIELD(scope),
493 	  .offset_words = 12,
494 	  .offset_bits  = 0,
495 	  .size_bits    = 4 },
496 	{ MCMEMBER_REC_FIELD(join_state),
497 	  .offset_words = 12,
498 	  .offset_bits  = 4,
499 	  .size_bits    = 4 },
500 	{ MCMEMBER_REC_FIELD(proxy_join),
501 	  .offset_words = 12,
502 	  .offset_bits  = 8,
503 	  .size_bits    = 1 },
504 	{ RESERVED,
505 	  .offset_words = 12,
506 	  .offset_bits  = 9,
507 	  .size_bits    = 23 },
508 };
509 
510 #define CLASSPORTINFO_REC_FIELD(field) \
511 	.struct_offset_bytes = offsetof(struct ib_class_port_info, field),	\
512 	.struct_size_bytes   = sizeof_field(struct ib_class_port_info, field),	\
513 	.field_name          = "ib_class_port_info:" #field
514 
515 static const struct ib_field ib_classport_info_rec_table[] = {
516 	{ CLASSPORTINFO_REC_FIELD(base_version),
517 	  .offset_words = 0,
518 	  .offset_bits  = 0,
519 	  .size_bits    = 8 },
520 	{ CLASSPORTINFO_REC_FIELD(class_version),
521 	  .offset_words = 0,
522 	  .offset_bits  = 8,
523 	  .size_bits    = 8 },
524 	{ CLASSPORTINFO_REC_FIELD(capability_mask),
525 	  .offset_words = 0,
526 	  .offset_bits  = 16,
527 	  .size_bits    = 16 },
528 	{ CLASSPORTINFO_REC_FIELD(cap_mask2_resp_time),
529 	  .offset_words = 1,
530 	  .offset_bits  = 0,
531 	  .size_bits    = 32 },
532 	{ CLASSPORTINFO_REC_FIELD(redirect_gid),
533 	  .offset_words = 2,
534 	  .offset_bits  = 0,
535 	  .size_bits    = 128 },
536 	{ CLASSPORTINFO_REC_FIELD(redirect_tcslfl),
537 	  .offset_words = 6,
538 	  .offset_bits  = 0,
539 	  .size_bits    = 32 },
540 	{ CLASSPORTINFO_REC_FIELD(redirect_lid),
541 	  .offset_words = 7,
542 	  .offset_bits  = 0,
543 	  .size_bits    = 16 },
544 	{ CLASSPORTINFO_REC_FIELD(redirect_pkey),
545 	  .offset_words = 7,
546 	  .offset_bits  = 16,
547 	  .size_bits    = 16 },
548 
549 	{ CLASSPORTINFO_REC_FIELD(redirect_qp),
550 	  .offset_words = 8,
551 	  .offset_bits  = 0,
552 	  .size_bits    = 32 },
553 	{ CLASSPORTINFO_REC_FIELD(redirect_qkey),
554 	  .offset_words = 9,
555 	  .offset_bits  = 0,
556 	  .size_bits    = 32 },
557 
558 	{ CLASSPORTINFO_REC_FIELD(trap_gid),
559 	  .offset_words = 10,
560 	  .offset_bits  = 0,
561 	  .size_bits    = 128 },
562 	{ CLASSPORTINFO_REC_FIELD(trap_tcslfl),
563 	  .offset_words = 14,
564 	  .offset_bits  = 0,
565 	  .size_bits    = 32 },
566 
567 	{ CLASSPORTINFO_REC_FIELD(trap_lid),
568 	  .offset_words = 15,
569 	  .offset_bits  = 0,
570 	  .size_bits    = 16 },
571 	{ CLASSPORTINFO_REC_FIELD(trap_pkey),
572 	  .offset_words = 15,
573 	  .offset_bits  = 16,
574 	  .size_bits    = 16 },
575 
576 	{ CLASSPORTINFO_REC_FIELD(trap_hlqp),
577 	  .offset_words = 16,
578 	  .offset_bits  = 0,
579 	  .size_bits    = 32 },
580 	{ CLASSPORTINFO_REC_FIELD(trap_qkey),
581 	  .offset_words = 17,
582 	  .offset_bits  = 0,
583 	  .size_bits    = 32 },
584 };
585 
586 #define OPA_CLASSPORTINFO_REC_FIELD(field) \
587 	.struct_offset_bytes =\
588 		offsetof(struct opa_class_port_info, field),	\
589 	.struct_size_bytes   = \
590 		sizeof_field(struct opa_class_port_info, field),	\
591 	.field_name          = "opa_class_port_info:" #field
592 
593 static const struct ib_field opa_classport_info_rec_table[] = {
594 	{ OPA_CLASSPORTINFO_REC_FIELD(base_version),
595 	  .offset_words = 0,
596 	  .offset_bits  = 0,
597 	  .size_bits    = 8 },
598 	{ OPA_CLASSPORTINFO_REC_FIELD(class_version),
599 	  .offset_words = 0,
600 	  .offset_bits  = 8,
601 	  .size_bits    = 8 },
602 	{ OPA_CLASSPORTINFO_REC_FIELD(cap_mask),
603 	  .offset_words = 0,
604 	  .offset_bits  = 16,
605 	  .size_bits    = 16 },
606 	{ OPA_CLASSPORTINFO_REC_FIELD(cap_mask2_resp_time),
607 	  .offset_words = 1,
608 	  .offset_bits  = 0,
609 	  .size_bits    = 32 },
610 	{ OPA_CLASSPORTINFO_REC_FIELD(redirect_gid),
611 	  .offset_words = 2,
612 	  .offset_bits  = 0,
613 	  .size_bits    = 128 },
614 	{ OPA_CLASSPORTINFO_REC_FIELD(redirect_tc_fl),
615 	  .offset_words = 6,
616 	  .offset_bits  = 0,
617 	  .size_bits    = 32 },
618 	{ OPA_CLASSPORTINFO_REC_FIELD(redirect_lid),
619 	  .offset_words = 7,
620 	  .offset_bits  = 0,
621 	  .size_bits    = 32 },
622 	{ OPA_CLASSPORTINFO_REC_FIELD(redirect_sl_qp),
623 	  .offset_words = 8,
624 	  .offset_bits  = 0,
625 	  .size_bits    = 32 },
626 	{ OPA_CLASSPORTINFO_REC_FIELD(redirect_qkey),
627 	  .offset_words = 9,
628 	  .offset_bits  = 0,
629 	  .size_bits    = 32 },
630 	{ OPA_CLASSPORTINFO_REC_FIELD(trap_gid),
631 	  .offset_words = 10,
632 	  .offset_bits  = 0,
633 	  .size_bits    = 128 },
634 	{ OPA_CLASSPORTINFO_REC_FIELD(trap_tc_fl),
635 	  .offset_words = 14,
636 	  .offset_bits  = 0,
637 	  .size_bits    = 32 },
638 	{ OPA_CLASSPORTINFO_REC_FIELD(trap_lid),
639 	  .offset_words = 15,
640 	  .offset_bits  = 0,
641 	  .size_bits    = 32 },
642 	{ OPA_CLASSPORTINFO_REC_FIELD(trap_hl_qp),
643 	  .offset_words = 16,
644 	  .offset_bits  = 0,
645 	  .size_bits    = 32 },
646 	{ OPA_CLASSPORTINFO_REC_FIELD(trap_qkey),
647 	  .offset_words = 17,
648 	  .offset_bits  = 0,
649 	  .size_bits    = 32 },
650 	{ OPA_CLASSPORTINFO_REC_FIELD(trap_pkey),
651 	  .offset_words = 18,
652 	  .offset_bits  = 0,
653 	  .size_bits    = 16 },
654 	{ OPA_CLASSPORTINFO_REC_FIELD(redirect_pkey),
655 	  .offset_words = 18,
656 	  .offset_bits  = 16,
657 	  .size_bits    = 16 },
658 	{ OPA_CLASSPORTINFO_REC_FIELD(trap_sl_rsvd),
659 	  .offset_words = 19,
660 	  .offset_bits  = 0,
661 	  .size_bits    = 8 },
662 	{ RESERVED,
663 	  .offset_words = 19,
664 	  .offset_bits  = 8,
665 	  .size_bits    = 24 },
666 };
667 
668 #define GUIDINFO_REC_FIELD(field) \
669 	.struct_offset_bytes = offsetof(struct ib_sa_guidinfo_rec, field),	\
670 	.struct_size_bytes   = sizeof_field(struct ib_sa_guidinfo_rec, field),	\
671 	.field_name          = "sa_guidinfo_rec:" #field
672 
673 static const struct ib_field guidinfo_rec_table[] = {
674 	{ GUIDINFO_REC_FIELD(lid),
675 	  .offset_words = 0,
676 	  .offset_bits  = 0,
677 	  .size_bits    = 16 },
678 	{ GUIDINFO_REC_FIELD(block_num),
679 	  .offset_words = 0,
680 	  .offset_bits  = 16,
681 	  .size_bits    = 8 },
682 	{ GUIDINFO_REC_FIELD(res1),
683 	  .offset_words = 0,
684 	  .offset_bits  = 24,
685 	  .size_bits    = 8 },
686 	{ GUIDINFO_REC_FIELD(res2),
687 	  .offset_words = 1,
688 	  .offset_bits  = 0,
689 	  .size_bits    = 32 },
690 	{ GUIDINFO_REC_FIELD(guid_info_list),
691 	  .offset_words = 2,
692 	  .offset_bits  = 0,
693 	  .size_bits    = 512 },
694 };
695 
696 #define SERVICE_REC_FIELD(field) \
697 	.struct_offset_bytes = offsetof(struct sa_service_rec, field),     \
698 	.struct_size_bytes   = sizeof_field(struct sa_service_rec, field), \
699 	.field_name          = "sa_service_rec:" #field
700 
701 static const struct ib_field service_rec_table[] = {
702 	{ SERVICE_REC_FIELD(id),
703 	  .offset_words = 0,
704 	  .offset_bits  = 0,
705 	  .size_bits    = 64 },
706 	{ SERVICE_REC_FIELD(gid),
707 	  .offset_words = 2,
708 	  .offset_bits  = 0,
709 	  .size_bits    = 128 },
710 	{ SERVICE_REC_FIELD(pkey),
711 	  .offset_words = 6,
712 	  .offset_bits  = 0,
713 	  .size_bits    = 16 },
714 	{ RESERVED,
715 	  .offset_words = 6,
716 	  .offset_bits  = 16,
717 	  .size_bits    = 16 },
718 	{ SERVICE_REC_FIELD(lease),
719 	  .offset_words = 7,
720 	  .offset_bits  = 0,
721 	  .size_bits    = 32 },
722 	{ SERVICE_REC_FIELD(key),
723 	  .offset_words = 8,
724 	  .offset_bits  = 0,
725 	  .size_bits    = 128 },
726 	{ SERVICE_REC_FIELD(name),
727 	  .offset_words = 12,
728 	  .offset_bits  = 0,
729 	  .size_bits    = 512 },
730 	{ SERVICE_REC_FIELD(data_8),
731 	  .offset_words = 28,
732 	  .offset_bits  = 0,
733 	  .size_bits    = 128 },
734 	{ SERVICE_REC_FIELD(data_16),
735 	  .offset_words = 32,
736 	  .offset_bits  = 0,
737 	  .size_bits    = 128 },
738 	{ SERVICE_REC_FIELD(data_32),
739 	  .offset_words = 36,
740 	  .offset_bits  = 0,
741 	  .size_bits    = 128 },
742 	{ SERVICE_REC_FIELD(data_64),
743 	  .offset_words = 40,
744 	  .offset_bits  = 0,
745 	  .size_bits    = 128 },
746 };
747 
748 #define RDMA_PRIMARY_PATH_MAX_REC_NUM 3
749 
750 static inline void ib_sa_disable_local_svc(struct ib_sa_query *query)
751 {
752 	query->flags &= ~IB_SA_ENABLE_LOCAL_SERVICE;
753 }
754 
755 static inline int ib_sa_query_cancelled(struct ib_sa_query *query)
756 {
757 	return (query->flags & IB_SA_CANCEL);
758 }
759 
760 static void ib_nl_set_path_rec_attrs(struct sk_buff *skb,
761 				     struct ib_sa_query *query)
762 {
763 	struct sa_path_rec *sa_rec = query->mad_buf->context[1];
764 	struct ib_sa_mad *mad = query->mad_buf->mad;
765 	ib_sa_comp_mask comp_mask = mad->sa_hdr.comp_mask;
766 	u16 val16;
767 	u64 val64;
768 	struct rdma_ls_resolve_header *header;
769 
770 	query->mad_buf->context[1] = NULL;
771 
772 	/* Construct the family header first */
773 	header = skb_put(skb, NLMSG_ALIGN(sizeof(*header)));
774 	strscpy_pad(header->device_name,
775 		    dev_name(&query->port->agent->device->dev),
776 		    LS_DEVICE_NAME_MAX);
777 	header->port_num = query->port->port_num;
778 
779 	if ((comp_mask & IB_SA_PATH_REC_REVERSIBLE) &&
780 	    sa_rec->reversible != 0)
781 		query->path_use = LS_RESOLVE_PATH_USE_ALL;
782 	else
783 		query->path_use = LS_RESOLVE_PATH_USE_UNIDIRECTIONAL;
784 	header->path_use = query->path_use;
785 
786 	/* Now build the attributes */
787 	if (comp_mask & IB_SA_PATH_REC_SERVICE_ID) {
788 		val64 = be64_to_cpu(sa_rec->service_id);
789 		nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_SERVICE_ID,
790 			sizeof(val64), &val64);
791 	}
792 	if (comp_mask & IB_SA_PATH_REC_DGID)
793 		nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_DGID,
794 			sizeof(sa_rec->dgid), &sa_rec->dgid);
795 	if (comp_mask & IB_SA_PATH_REC_SGID)
796 		nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_SGID,
797 			sizeof(sa_rec->sgid), &sa_rec->sgid);
798 	if (comp_mask & IB_SA_PATH_REC_TRAFFIC_CLASS)
799 		nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_TCLASS,
800 			sizeof(sa_rec->traffic_class), &sa_rec->traffic_class);
801 
802 	if (comp_mask & IB_SA_PATH_REC_PKEY) {
803 		val16 = be16_to_cpu(sa_rec->pkey);
804 		nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_PKEY,
805 			sizeof(val16), &val16);
806 	}
807 	if (comp_mask & IB_SA_PATH_REC_QOS_CLASS) {
808 		val16 = be16_to_cpu(sa_rec->qos_class);
809 		nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_QOS_CLASS,
810 			sizeof(val16), &val16);
811 	}
812 }
813 
814 static int ib_nl_get_path_rec_attrs_len(ib_sa_comp_mask comp_mask)
815 {
816 	int len = 0;
817 
818 	if (comp_mask & IB_SA_PATH_REC_SERVICE_ID)
819 		len += nla_total_size(sizeof(u64));
820 	if (comp_mask & IB_SA_PATH_REC_DGID)
821 		len += nla_total_size(sizeof(struct rdma_nla_ls_gid));
822 	if (comp_mask & IB_SA_PATH_REC_SGID)
823 		len += nla_total_size(sizeof(struct rdma_nla_ls_gid));
824 	if (comp_mask & IB_SA_PATH_REC_TRAFFIC_CLASS)
825 		len += nla_total_size(sizeof(u8));
826 	if (comp_mask & IB_SA_PATH_REC_PKEY)
827 		len += nla_total_size(sizeof(u16));
828 	if (comp_mask & IB_SA_PATH_REC_QOS_CLASS)
829 		len += nla_total_size(sizeof(u16));
830 
831 	/*
832 	 * Make sure that at least some of the required comp_mask bits are
833 	 * set.
834 	 */
835 	if (WARN_ON(len == 0))
836 		return len;
837 
838 	/* Add the family header */
839 	len += NLMSG_ALIGN(sizeof(struct rdma_ls_resolve_header));
840 
841 	return len;
842 }
843 
844 static int ib_nl_make_request(struct ib_sa_query *query, gfp_t gfp_mask)
845 {
846 	struct sk_buff *skb = NULL;
847 	struct nlmsghdr *nlh;
848 	void *data;
849 	struct ib_sa_mad *mad;
850 	int len;
851 	unsigned long flags;
852 	unsigned long delay;
853 	gfp_t gfp_flag;
854 	int ret;
855 
856 	INIT_LIST_HEAD(&query->list);
857 	query->seq = (u32)atomic_inc_return(&ib_nl_sa_request_seq);
858 
859 	mad = query->mad_buf->mad;
860 	len = ib_nl_get_path_rec_attrs_len(mad->sa_hdr.comp_mask);
861 	if (len <= 0)
862 		return -EMSGSIZE;
863 
864 	skb = nlmsg_new(len, gfp_mask);
865 	if (!skb)
866 		return -ENOMEM;
867 
868 	/* Put nlmsg header only for now */
869 	data = ibnl_put_msg(skb, &nlh, query->seq, 0, RDMA_NL_LS,
870 			    RDMA_NL_LS_OP_RESOLVE, NLM_F_REQUEST);
871 	if (!data) {
872 		nlmsg_free(skb);
873 		return -EMSGSIZE;
874 	}
875 
876 	/* Add attributes */
877 	ib_nl_set_path_rec_attrs(skb, query);
878 
879 	/* Repair the nlmsg header length */
880 	nlmsg_end(skb, nlh);
881 
882 	gfp_flag = ((gfp_mask & GFP_ATOMIC) == GFP_ATOMIC) ? GFP_ATOMIC :
883 		GFP_NOWAIT;
884 
885 	spin_lock_irqsave(&ib_nl_request_lock, flags);
886 	ret = rdma_nl_multicast(&init_net, skb, RDMA_NL_GROUP_LS, gfp_flag);
887 
888 	if (ret)
889 		goto out;
890 
891 	/* Put the request on the list.*/
892 	delay = msecs_to_jiffies(sa_local_svc_timeout_ms);
893 	query->timeout = delay + jiffies;
894 	list_add_tail(&query->list, &ib_nl_request_list);
895 	/* Start the timeout if this is the only request */
896 	if (ib_nl_request_list.next == &query->list)
897 		queue_delayed_work(ib_nl_wq, &ib_nl_timed_work, delay);
898 
899 out:
900 	spin_unlock_irqrestore(&ib_nl_request_lock, flags);
901 
902 	return ret;
903 }
904 
905 static int ib_nl_cancel_request(struct ib_sa_query *query)
906 {
907 	unsigned long flags;
908 	struct ib_sa_query *wait_query;
909 	int found = 0;
910 
911 	spin_lock_irqsave(&ib_nl_request_lock, flags);
912 	list_for_each_entry(wait_query, &ib_nl_request_list, list) {
913 		/* Let the timeout to take care of the callback */
914 		if (query == wait_query) {
915 			query->flags |= IB_SA_CANCEL;
916 			query->timeout = jiffies;
917 			list_move(&query->list, &ib_nl_request_list);
918 			found = 1;
919 			mod_delayed_work(ib_nl_wq, &ib_nl_timed_work, 1);
920 			break;
921 		}
922 	}
923 	spin_unlock_irqrestore(&ib_nl_request_lock, flags);
924 
925 	return found;
926 }
927 
928 static void send_handler(struct ib_mad_agent *agent,
929 			 struct ib_mad_send_wc *mad_send_wc);
930 
931 static void ib_nl_process_good_resolve_rsp(struct ib_sa_query *query,
932 					   const struct nlmsghdr *nlh)
933 {
934 	struct sa_path_rec recs[RDMA_PRIMARY_PATH_MAX_REC_NUM];
935 	struct ib_sa_path_query *path_query;
936 	struct ib_path_rec_data *rec_data;
937 	struct ib_mad_send_wc mad_send_wc;
938 	const struct nlattr *head, *curr;
939 	struct ib_sa_mad *mad = NULL;
940 	int len, rem, status = -EIO;
941 	unsigned int num_prs = 0;
942 	u32 mask = 0;
943 
944 	if (!query->callback)
945 		goto out;
946 
947 	path_query = container_of(query, struct ib_sa_path_query, sa_query);
948 	mad = query->mad_buf->mad;
949 
950 	head = (const struct nlattr *) nlmsg_data(nlh);
951 	len = nlmsg_len(nlh);
952 	switch (query->path_use) {
953 	case LS_RESOLVE_PATH_USE_UNIDIRECTIONAL:
954 		mask = IB_PATH_PRIMARY | IB_PATH_OUTBOUND;
955 		break;
956 
957 	case LS_RESOLVE_PATH_USE_ALL:
958 		mask = IB_PATH_PRIMARY;
959 		break;
960 
961 	case LS_RESOLVE_PATH_USE_GMP:
962 	default:
963 		mask = IB_PATH_PRIMARY | IB_PATH_GMP |
964 			IB_PATH_BIDIRECTIONAL;
965 		break;
966 	}
967 
968 	nla_for_each_attr(curr, head, len, rem) {
969 		if (curr->nla_type != LS_NLA_TYPE_PATH_RECORD)
970 			continue;
971 
972 		rec_data = nla_data(curr);
973 		if ((rec_data->flags & mask) != mask)
974 			continue;
975 
976 		if ((query->flags & IB_SA_QUERY_OPA) ||
977 		    path_query->conv_pr) {
978 			mad->mad_hdr.method |= IB_MGMT_METHOD_RESP;
979 			memcpy(mad->data, rec_data->path_rec,
980 			       sizeof(rec_data->path_rec));
981 			query->callback(query, 0, mad);
982 			goto out;
983 		}
984 
985 		status = 0;
986 		ib_unpack(path_rec_table, ARRAY_SIZE(path_rec_table),
987 			  rec_data->path_rec, &recs[num_prs]);
988 		recs[num_prs].flags = rec_data->flags;
989 		recs[num_prs].rec_type = SA_PATH_REC_TYPE_IB;
990 		sa_path_set_dmac_zero(&recs[num_prs]);
991 
992 		num_prs++;
993 		if (num_prs >= RDMA_PRIMARY_PATH_MAX_REC_NUM)
994 			break;
995 	}
996 
997 	if (!status) {
998 		mad->mad_hdr.method |= IB_MGMT_METHOD_RESP;
999 		path_query->callback(status, recs, num_prs,
1000 				     path_query->context);
1001 	} else
1002 		query->callback(query, status, mad);
1003 
1004 out:
1005 	mad_send_wc.send_buf = query->mad_buf;
1006 	mad_send_wc.status = IB_WC_SUCCESS;
1007 	send_handler(query->mad_buf->mad_agent, &mad_send_wc);
1008 }
1009 
1010 static void ib_nl_request_timeout(struct work_struct *work)
1011 {
1012 	unsigned long flags;
1013 	struct ib_sa_query *query;
1014 	unsigned long delay;
1015 	struct ib_mad_send_wc mad_send_wc;
1016 	int ret;
1017 
1018 	spin_lock_irqsave(&ib_nl_request_lock, flags);
1019 	while (!list_empty(&ib_nl_request_list)) {
1020 		query = list_entry(ib_nl_request_list.next,
1021 				   struct ib_sa_query, list);
1022 
1023 		if (time_after(query->timeout, jiffies)) {
1024 			delay = query->timeout - jiffies;
1025 			if ((long)delay <= 0)
1026 				delay = 1;
1027 			queue_delayed_work(ib_nl_wq, &ib_nl_timed_work, delay);
1028 			break;
1029 		}
1030 
1031 		list_del(&query->list);
1032 		ib_sa_disable_local_svc(query);
1033 		/* Hold the lock to protect against query cancellation */
1034 		if (ib_sa_query_cancelled(query))
1035 			ret = -1;
1036 		else
1037 			ret = ib_post_send_mad(query->mad_buf, NULL);
1038 		if (ret) {
1039 			mad_send_wc.send_buf = query->mad_buf;
1040 			mad_send_wc.status = IB_WC_WR_FLUSH_ERR;
1041 			spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1042 			send_handler(query->port->agent, &mad_send_wc);
1043 			spin_lock_irqsave(&ib_nl_request_lock, flags);
1044 		}
1045 	}
1046 	spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1047 }
1048 
1049 int ib_nl_handle_set_timeout(struct sk_buff *skb,
1050 			     struct nlmsghdr *nlh,
1051 			     struct netlink_ext_ack *extack)
1052 {
1053 	int timeout, delta, abs_delta;
1054 	const struct nlattr *attr;
1055 	unsigned long flags;
1056 	struct ib_sa_query *query;
1057 	long delay = 0;
1058 	struct nlattr *tb[LS_NLA_TYPE_MAX];
1059 	int ret;
1060 
1061 	if (!(nlh->nlmsg_flags & NLM_F_REQUEST) ||
1062 	    !(NETLINK_CB(skb).sk))
1063 		return -EPERM;
1064 
1065 	ret = nla_parse_deprecated(tb, LS_NLA_TYPE_MAX - 1, nlmsg_data(nlh),
1066 				   nlmsg_len(nlh), ib_nl_policy, NULL);
1067 	attr = (const struct nlattr *)tb[LS_NLA_TYPE_TIMEOUT];
1068 	if (ret || !attr)
1069 		goto settimeout_out;
1070 
1071 	timeout = *(int *) nla_data(attr);
1072 	if (timeout < IB_SA_LOCAL_SVC_TIMEOUT_MIN)
1073 		timeout = IB_SA_LOCAL_SVC_TIMEOUT_MIN;
1074 	if (timeout > IB_SA_LOCAL_SVC_TIMEOUT_MAX)
1075 		timeout = IB_SA_LOCAL_SVC_TIMEOUT_MAX;
1076 
1077 	spin_lock_irqsave(&ib_nl_request_lock, flags);
1078 
1079 	delta = timeout - sa_local_svc_timeout_ms;
1080 	if (delta < 0)
1081 		abs_delta = -delta;
1082 	else
1083 		abs_delta = delta;
1084 
1085 	if (delta != 0) {
1086 		sa_local_svc_timeout_ms = timeout;
1087 		list_for_each_entry(query, &ib_nl_request_list, list) {
1088 			if (delta < 0 && abs_delta > query->timeout)
1089 				query->timeout = 0;
1090 			else
1091 				query->timeout += delta;
1092 
1093 			/* Get the new delay from the first entry */
1094 			if (!delay) {
1095 				delay = query->timeout - jiffies;
1096 				if (delay <= 0)
1097 					delay = 1;
1098 			}
1099 		}
1100 		if (delay)
1101 			mod_delayed_work(ib_nl_wq, &ib_nl_timed_work,
1102 					 (unsigned long)delay);
1103 	}
1104 
1105 	spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1106 
1107 settimeout_out:
1108 	return 0;
1109 }
1110 
1111 static inline int ib_nl_is_good_resolve_resp(const struct nlmsghdr *nlh)
1112 {
1113 	struct nlattr *tb[LS_NLA_TYPE_MAX];
1114 	int ret;
1115 
1116 	if (nlh->nlmsg_flags & RDMA_NL_LS_F_ERR)
1117 		return 0;
1118 
1119 	ret = nla_parse_deprecated(tb, LS_NLA_TYPE_MAX - 1, nlmsg_data(nlh),
1120 				   nlmsg_len(nlh), ib_nl_policy, NULL);
1121 	if (ret)
1122 		return 0;
1123 
1124 	return 1;
1125 }
1126 
1127 int ib_nl_handle_resolve_resp(struct sk_buff *skb,
1128 			      struct nlmsghdr *nlh,
1129 			      struct netlink_ext_ack *extack)
1130 {
1131 	unsigned long flags;
1132 	struct ib_sa_query *query = NULL, *iter;
1133 	struct ib_mad_send_buf *send_buf;
1134 	struct ib_mad_send_wc mad_send_wc;
1135 	int ret;
1136 
1137 	if ((nlh->nlmsg_flags & NLM_F_REQUEST) ||
1138 	    !(NETLINK_CB(skb).sk))
1139 		return -EPERM;
1140 
1141 	spin_lock_irqsave(&ib_nl_request_lock, flags);
1142 	list_for_each_entry(iter, &ib_nl_request_list, list) {
1143 		/*
1144 		 * If the query is cancelled, let the timeout routine
1145 		 * take care of it.
1146 		 */
1147 		if (nlh->nlmsg_seq == iter->seq) {
1148 			if (!ib_sa_query_cancelled(iter)) {
1149 				list_del(&iter->list);
1150 				query = iter;
1151 			}
1152 			break;
1153 		}
1154 	}
1155 
1156 	if (!query) {
1157 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1158 		goto resp_out;
1159 	}
1160 
1161 	send_buf = query->mad_buf;
1162 
1163 	if (!ib_nl_is_good_resolve_resp(nlh)) {
1164 		/* if the result is a failure, send out the packet via IB */
1165 		ib_sa_disable_local_svc(query);
1166 		ret = ib_post_send_mad(query->mad_buf, NULL);
1167 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1168 		if (ret) {
1169 			mad_send_wc.send_buf = send_buf;
1170 			mad_send_wc.status = IB_WC_GENERAL_ERR;
1171 			send_handler(query->port->agent, &mad_send_wc);
1172 		}
1173 	} else {
1174 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1175 		ib_nl_process_good_resolve_rsp(query, nlh);
1176 	}
1177 
1178 resp_out:
1179 	return 0;
1180 }
1181 
1182 static void free_sm_ah(struct kref *kref)
1183 {
1184 	struct ib_sa_sm_ah *sm_ah = container_of(kref, struct ib_sa_sm_ah, ref);
1185 
1186 	rdma_destroy_ah(sm_ah->ah, 0);
1187 	kfree(sm_ah);
1188 }
1189 
1190 void ib_sa_register_client(struct ib_sa_client *client)
1191 {
1192 	atomic_set(&client->users, 1);
1193 	init_completion(&client->comp);
1194 }
1195 EXPORT_SYMBOL(ib_sa_register_client);
1196 
1197 void ib_sa_unregister_client(struct ib_sa_client *client)
1198 {
1199 	ib_sa_client_put(client);
1200 	wait_for_completion(&client->comp);
1201 }
1202 EXPORT_SYMBOL(ib_sa_unregister_client);
1203 
1204 /**
1205  * ib_sa_cancel_query - try to cancel an SA query
1206  * @id:ID of query to cancel
1207  * @query:query pointer to cancel
1208  *
1209  * Try to cancel an SA query.  If the id and query don't match up or
1210  * the query has already completed, nothing is done.  Otherwise the
1211  * query is canceled and will complete with a status of -EINTR.
1212  */
1213 void ib_sa_cancel_query(int id, struct ib_sa_query *query)
1214 {
1215 	unsigned long flags;
1216 	struct ib_mad_send_buf *mad_buf;
1217 
1218 	xa_lock_irqsave(&queries, flags);
1219 	if (xa_load(&queries, id) != query) {
1220 		xa_unlock_irqrestore(&queries, flags);
1221 		return;
1222 	}
1223 	mad_buf = query->mad_buf;
1224 	xa_unlock_irqrestore(&queries, flags);
1225 
1226 	/*
1227 	 * If the query is still on the netlink request list, schedule
1228 	 * it to be cancelled by the timeout routine. Otherwise, it has been
1229 	 * sent to the MAD layer and has to be cancelled from there.
1230 	 */
1231 	if (!ib_nl_cancel_request(query))
1232 		ib_cancel_mad(mad_buf);
1233 }
1234 EXPORT_SYMBOL(ib_sa_cancel_query);
1235 
1236 static u8 get_src_path_mask(struct ib_device *device, u32 port_num)
1237 {
1238 	struct ib_sa_device *sa_dev;
1239 	struct ib_sa_port   *port;
1240 	unsigned long flags;
1241 	u8 src_path_mask;
1242 
1243 	sa_dev = ib_get_client_data(device, &sa_client);
1244 	if (!sa_dev)
1245 		return 0x7f;
1246 
1247 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1248 	spin_lock_irqsave(&port->ah_lock, flags);
1249 	src_path_mask = port->sm_ah ? port->sm_ah->src_path_mask : 0x7f;
1250 	spin_unlock_irqrestore(&port->ah_lock, flags);
1251 
1252 	return src_path_mask;
1253 }
1254 
1255 static int init_ah_attr_grh_fields(struct ib_device *device, u32 port_num,
1256 				   struct sa_path_rec *rec,
1257 				   struct rdma_ah_attr *ah_attr,
1258 				   const struct ib_gid_attr *gid_attr)
1259 {
1260 	enum ib_gid_type type = sa_conv_pathrec_to_gid_type(rec);
1261 
1262 	if (!gid_attr) {
1263 		gid_attr = rdma_find_gid_by_port(device, &rec->sgid, type,
1264 						 port_num, NULL);
1265 		if (IS_ERR(gid_attr))
1266 			return PTR_ERR(gid_attr);
1267 	} else
1268 		rdma_hold_gid_attr(gid_attr);
1269 
1270 	rdma_move_grh_sgid_attr(ah_attr, &rec->dgid,
1271 				be32_to_cpu(rec->flow_label),
1272 				rec->hop_limit,	rec->traffic_class,
1273 				gid_attr);
1274 	return 0;
1275 }
1276 
1277 /**
1278  * ib_init_ah_attr_from_path - Initialize address handle attributes based on
1279  *   an SA path record.
1280  * @device: Device associated ah attributes initialization.
1281  * @port_num: Port on the specified device.
1282  * @rec: path record entry to use for ah attributes initialization.
1283  * @ah_attr: address handle attributes to initialization from path record.
1284  * @gid_attr: SGID attribute to consider during initialization.
1285  *
1286  * When ib_init_ah_attr_from_path() returns success,
1287  * (a) for IB link layer it optionally contains a reference to SGID attribute
1288  * when GRH is present for IB link layer.
1289  * (b) for RoCE link layer it contains a reference to SGID attribute.
1290  * User must invoke rdma_destroy_ah_attr() to release reference to SGID
1291  * attributes which are initialized using ib_init_ah_attr_from_path().
1292  */
1293 int ib_init_ah_attr_from_path(struct ib_device *device, u32 port_num,
1294 			      struct sa_path_rec *rec,
1295 			      struct rdma_ah_attr *ah_attr,
1296 			      const struct ib_gid_attr *gid_attr)
1297 {
1298 	int ret = 0;
1299 
1300 	memset(ah_attr, 0, sizeof(*ah_attr));
1301 	ah_attr->type = rdma_ah_find_type(device, port_num);
1302 	rdma_ah_set_sl(ah_attr, rec->sl);
1303 	rdma_ah_set_port_num(ah_attr, port_num);
1304 	rdma_ah_set_static_rate(ah_attr, rec->rate);
1305 
1306 	if (sa_path_is_roce(rec)) {
1307 		ret = roce_resolve_route_from_path(rec, gid_attr);
1308 		if (ret)
1309 			return ret;
1310 
1311 		memcpy(ah_attr->roce.dmac, sa_path_get_dmac(rec), ETH_ALEN);
1312 	} else {
1313 		rdma_ah_set_dlid(ah_attr, be32_to_cpu(sa_path_get_dlid(rec)));
1314 		if (sa_path_is_opa(rec) &&
1315 		    rdma_ah_get_dlid(ah_attr) == be16_to_cpu(IB_LID_PERMISSIVE))
1316 			rdma_ah_set_make_grd(ah_attr, true);
1317 
1318 		rdma_ah_set_path_bits(ah_attr,
1319 				      be32_to_cpu(sa_path_get_slid(rec)) &
1320 				      get_src_path_mask(device, port_num));
1321 	}
1322 
1323 	if (rec->hop_limit > 0 || sa_path_is_roce(rec))
1324 		ret = init_ah_attr_grh_fields(device, port_num,
1325 					      rec, ah_attr, gid_attr);
1326 	return ret;
1327 }
1328 EXPORT_SYMBOL(ib_init_ah_attr_from_path);
1329 
1330 static int alloc_mad(struct ib_sa_query *query, gfp_t gfp_mask)
1331 {
1332 	struct rdma_ah_attr ah_attr;
1333 	unsigned long flags;
1334 
1335 	spin_lock_irqsave(&query->port->ah_lock, flags);
1336 	if (!query->port->sm_ah) {
1337 		spin_unlock_irqrestore(&query->port->ah_lock, flags);
1338 		return -EAGAIN;
1339 	}
1340 	kref_get(&query->port->sm_ah->ref);
1341 	query->sm_ah = query->port->sm_ah;
1342 	spin_unlock_irqrestore(&query->port->ah_lock, flags);
1343 
1344 	/*
1345 	 * Always check if sm_ah has valid dlid assigned,
1346 	 * before querying for class port info
1347 	 */
1348 	if ((rdma_query_ah(query->sm_ah->ah, &ah_attr) < 0) ||
1349 	    !rdma_is_valid_unicast_lid(&ah_attr)) {
1350 		kref_put(&query->sm_ah->ref, free_sm_ah);
1351 		return -EAGAIN;
1352 	}
1353 	query->mad_buf = ib_create_send_mad(query->port->agent, 1,
1354 					    query->sm_ah->pkey_index,
1355 					    0, IB_MGMT_SA_HDR, IB_MGMT_SA_DATA,
1356 					    gfp_mask,
1357 					    ((query->flags & IB_SA_QUERY_OPA) ?
1358 					     OPA_MGMT_BASE_VERSION :
1359 					     IB_MGMT_BASE_VERSION));
1360 	if (IS_ERR(query->mad_buf)) {
1361 		kref_put(&query->sm_ah->ref, free_sm_ah);
1362 		return -ENOMEM;
1363 	}
1364 
1365 	query->mad_buf->ah = query->sm_ah->ah;
1366 
1367 	return 0;
1368 }
1369 
1370 static void free_mad(struct ib_sa_query *query)
1371 {
1372 	ib_free_send_mad(query->mad_buf);
1373 	kref_put(&query->sm_ah->ref, free_sm_ah);
1374 }
1375 
1376 static void init_mad(struct ib_sa_query *query, struct ib_mad_agent *agent)
1377 {
1378 	struct ib_sa_mad *mad = query->mad_buf->mad;
1379 	unsigned long flags;
1380 
1381 	memset(mad, 0, sizeof *mad);
1382 
1383 	if (query->flags & IB_SA_QUERY_OPA) {
1384 		mad->mad_hdr.base_version  = OPA_MGMT_BASE_VERSION;
1385 		mad->mad_hdr.class_version = OPA_SA_CLASS_VERSION;
1386 	} else {
1387 		mad->mad_hdr.base_version  = IB_MGMT_BASE_VERSION;
1388 		mad->mad_hdr.class_version = IB_SA_CLASS_VERSION;
1389 	}
1390 	mad->mad_hdr.mgmt_class    = IB_MGMT_CLASS_SUBN_ADM;
1391 	spin_lock_irqsave(&tid_lock, flags);
1392 	mad->mad_hdr.tid           =
1393 		cpu_to_be64(((u64) agent->hi_tid) << 32 | tid++);
1394 	spin_unlock_irqrestore(&tid_lock, flags);
1395 }
1396 
1397 static int send_mad(struct ib_sa_query *query, unsigned long timeout_ms,
1398 		    gfp_t gfp_mask)
1399 {
1400 	unsigned long flags;
1401 	int ret, id;
1402 	const int nmbr_sa_query_retries = 10;
1403 
1404 	xa_lock_irqsave(&queries, flags);
1405 	ret = __xa_alloc(&queries, &id, query, xa_limit_32b, gfp_mask);
1406 	xa_unlock_irqrestore(&queries, flags);
1407 	if (ret < 0)
1408 		return ret;
1409 
1410 	query->mad_buf->timeout_ms  = timeout_ms / nmbr_sa_query_retries;
1411 	query->mad_buf->retries = nmbr_sa_query_retries;
1412 	if (!query->mad_buf->timeout_ms) {
1413 		/* Special case, very small timeout_ms */
1414 		query->mad_buf->timeout_ms = 1;
1415 		query->mad_buf->retries = timeout_ms;
1416 	}
1417 	query->mad_buf->context[0] = query;
1418 	query->id = id;
1419 
1420 	if ((query->flags & IB_SA_ENABLE_LOCAL_SERVICE) &&
1421 	    (!(query->flags & IB_SA_QUERY_OPA))) {
1422 		if (rdma_nl_chk_listeners(RDMA_NL_GROUP_LS)) {
1423 			if (!ib_nl_make_request(query, gfp_mask))
1424 				return id;
1425 		}
1426 		ib_sa_disable_local_svc(query);
1427 	}
1428 
1429 	ret = ib_post_send_mad(query->mad_buf, NULL);
1430 	if (ret) {
1431 		xa_lock_irqsave(&queries, flags);
1432 		__xa_erase(&queries, id);
1433 		xa_unlock_irqrestore(&queries, flags);
1434 	}
1435 
1436 	/*
1437 	 * It's not safe to dereference query any more, because the
1438 	 * send may already have completed and freed the query in
1439 	 * another context.
1440 	 */
1441 	return ret ? ret : id;
1442 }
1443 
1444 void ib_sa_unpack_path(void *attribute, struct sa_path_rec *rec)
1445 {
1446 	ib_unpack(path_rec_table, ARRAY_SIZE(path_rec_table), attribute, rec);
1447 }
1448 EXPORT_SYMBOL(ib_sa_unpack_path);
1449 
1450 void ib_sa_pack_path(struct sa_path_rec *rec, void *attribute)
1451 {
1452 	ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table), rec, attribute);
1453 }
1454 EXPORT_SYMBOL(ib_sa_pack_path);
1455 
1456 void ib_sa_pack_service(struct sa_service_rec *rec, void *attribute)
1457 {
1458 	ib_pack(service_rec_table, ARRAY_SIZE(service_rec_table), rec,
1459 		attribute);
1460 }
1461 EXPORT_SYMBOL(ib_sa_pack_service);
1462 
1463 void ib_sa_unpack_service(void *attribute, struct sa_service_rec *rec)
1464 {
1465 	ib_unpack(service_rec_table, ARRAY_SIZE(service_rec_table), attribute,
1466 		  rec);
1467 }
1468 EXPORT_SYMBOL(ib_sa_unpack_service);
1469 
1470 static bool ib_sa_opa_pathrecord_support(struct ib_sa_client *client,
1471 					 struct ib_sa_device *sa_dev,
1472 					 u32 port_num)
1473 {
1474 	struct ib_sa_port *port;
1475 	unsigned long flags;
1476 	bool ret = false;
1477 
1478 	port = &sa_dev->port[port_num - sa_dev->start_port];
1479 	spin_lock_irqsave(&port->classport_lock, flags);
1480 	if (!port->classport_info.valid)
1481 		goto ret;
1482 
1483 	if (port->classport_info.data.type == RDMA_CLASS_PORT_INFO_OPA)
1484 		ret = opa_get_cpi_capmask2(&port->classport_info.data.opa) &
1485 			OPA_CLASS_PORT_INFO_PR_SUPPORT;
1486 ret:
1487 	spin_unlock_irqrestore(&port->classport_lock, flags);
1488 	return ret;
1489 }
1490 
1491 enum opa_pr_supported {
1492 	PR_NOT_SUPPORTED,
1493 	PR_OPA_SUPPORTED,
1494 	PR_IB_SUPPORTED
1495 };
1496 
1497 /*
1498  * opa_pr_query_possible - Check if current PR query can be an OPA query.
1499  *
1500  * Returns PR_NOT_SUPPORTED if a path record query is not
1501  * possible, PR_OPA_SUPPORTED if an OPA path record query
1502  * is possible and PR_IB_SUPPORTED if an IB path record
1503  * query is possible.
1504  */
1505 static int opa_pr_query_possible(struct ib_sa_client *client,
1506 				 struct ib_sa_device *sa_dev,
1507 				 struct ib_device *device, u32 port_num)
1508 {
1509 	struct ib_port_attr port_attr;
1510 
1511 	if (ib_query_port(device, port_num, &port_attr))
1512 		return PR_NOT_SUPPORTED;
1513 
1514 	if (ib_sa_opa_pathrecord_support(client, sa_dev, port_num))
1515 		return PR_OPA_SUPPORTED;
1516 
1517 	if (port_attr.lid >= be16_to_cpu(IB_MULTICAST_LID_BASE))
1518 		return PR_NOT_SUPPORTED;
1519 	else
1520 		return PR_IB_SUPPORTED;
1521 }
1522 
1523 static void ib_sa_path_rec_callback(struct ib_sa_query *sa_query,
1524 				    int status, struct ib_sa_mad *mad)
1525 {
1526 	struct ib_sa_path_query *query =
1527 		container_of(sa_query, struct ib_sa_path_query, sa_query);
1528 	struct sa_path_rec rec = {};
1529 
1530 	if (!mad) {
1531 		query->callback(status, NULL, 0, query->context);
1532 		return;
1533 	}
1534 
1535 	if (sa_query->flags & IB_SA_QUERY_OPA) {
1536 		ib_unpack(opa_path_rec_table, ARRAY_SIZE(opa_path_rec_table),
1537 			  mad->data, &rec);
1538 		rec.rec_type = SA_PATH_REC_TYPE_OPA;
1539 		query->callback(status, &rec, 1, query->context);
1540 		return;
1541 	}
1542 
1543 	ib_unpack(path_rec_table, ARRAY_SIZE(path_rec_table),
1544 		  mad->data, &rec);
1545 	rec.rec_type = SA_PATH_REC_TYPE_IB;
1546 	sa_path_set_dmac_zero(&rec);
1547 
1548 	if (query->conv_pr) {
1549 		struct sa_path_rec opa;
1550 
1551 		memset(&opa, 0, sizeof(struct sa_path_rec));
1552 		sa_convert_path_ib_to_opa(&opa, &rec);
1553 		query->callback(status, &opa, 1, query->context);
1554 	} else {
1555 		query->callback(status, &rec, 1, query->context);
1556 	}
1557 }
1558 
1559 #define IB_SA_DATA_OFFS 56
1560 #define IB_SERVICE_REC_SZ 176
1561 
1562 static void ib_unpack_service_rmpp(struct sa_service_rec *rec,
1563 				   struct ib_mad_recv_wc *mad_wc,
1564 				   int num_services)
1565 {
1566 	unsigned int cp_sz, data_i, data_size, rec_i = 0, buf_i = 0;
1567 	struct ib_mad_recv_buf *mad_buf;
1568 	u8 buf[IB_SERVICE_REC_SZ];
1569 	u8 *data;
1570 
1571 	data_size = sizeof(((struct ib_sa_mad *) mad_buf->mad)->data);
1572 
1573 	list_for_each_entry(mad_buf, &mad_wc->rmpp_list, list) {
1574 		data = ((struct ib_sa_mad *) mad_buf->mad)->data;
1575 		data_i = 0;
1576 		while (data_i < data_size && rec_i < num_services) {
1577 			cp_sz = min(IB_SERVICE_REC_SZ - buf_i,
1578 				    data_size - data_i);
1579 			memcpy(buf + buf_i, data + data_i, cp_sz);
1580 			data_i += cp_sz;
1581 			buf_i += cp_sz;
1582 			if (buf_i == IB_SERVICE_REC_SZ) {
1583 				ib_sa_unpack_service(buf, rec + rec_i);
1584 				buf_i = 0;
1585 				rec_i++;
1586 			}
1587 		}
1588 	}
1589 }
1590 
1591 static void ib_sa_service_rec_callback(struct ib_sa_query *sa_query, int status,
1592 				       struct ib_mad_recv_wc *mad_wc)
1593 {
1594 	struct ib_sa_service_query *query =
1595 		container_of(sa_query, struct ib_sa_service_query, sa_query);
1596 	struct sa_service_rec *rec;
1597 	int num_services;
1598 
1599 	if (!mad_wc || !mad_wc->recv_buf.mad) {
1600 		query->callback(status, NULL, 0, query->context);
1601 		return;
1602 	}
1603 
1604 	num_services = (mad_wc->mad_len - IB_SA_DATA_OFFS) / IB_SERVICE_REC_SZ;
1605 	if (!num_services) {
1606 		query->callback(-ENODATA, NULL, 0, query->context);
1607 		return;
1608 	}
1609 
1610 	rec = kmalloc_array(num_services, sizeof(*rec), GFP_KERNEL);
1611 	if (!rec) {
1612 		query->callback(-ENOMEM, NULL, 0, query->context);
1613 		return;
1614 	}
1615 
1616 	ib_unpack_service_rmpp(rec, mad_wc, num_services);
1617 	query->callback(status, rec, num_services, query->context);
1618 	kfree(rec);
1619 }
1620 
1621 static void ib_sa_path_rec_release(struct ib_sa_query *sa_query)
1622 {
1623 	struct ib_sa_path_query *query =
1624 		container_of(sa_query, struct ib_sa_path_query, sa_query);
1625 
1626 	kfree(query->conv_pr);
1627 	kfree(query);
1628 }
1629 
1630 static void ib_sa_service_rec_release(struct ib_sa_query *sa_query)
1631 {
1632 	struct ib_sa_service_query *query =
1633 		container_of(sa_query, struct ib_sa_service_query, sa_query);
1634 
1635 	kfree(query);
1636 }
1637 
1638 /**
1639  * ib_sa_path_rec_get - Start a Path get query
1640  * @client:SA client
1641  * @device:device to send query on
1642  * @port_num: port number to send query on
1643  * @rec:Path Record to send in query
1644  * @comp_mask:component mask to send in query
1645  * @timeout_ms:time to wait for response
1646  * @gfp_mask:GFP mask to use for internal allocations
1647  * @callback:function called when query completes, times out or is
1648  * canceled
1649  * @context:opaque user context passed to callback
1650  * @sa_query:query context, used to cancel query
1651  *
1652  * Send a Path Record Get query to the SA to look up a path.  The
1653  * callback function will be called when the query completes (or
1654  * fails); status is 0 for a successful response, -EINTR if the query
1655  * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
1656  * occurred sending the query.  The resp parameter of the callback is
1657  * only valid if status is 0.
1658  *
1659  * If the return value of ib_sa_path_rec_get() is negative, it is an
1660  * error code.  Otherwise it is a query ID that can be used to cancel
1661  * the query.
1662  */
1663 int ib_sa_path_rec_get(struct ib_sa_client *client,
1664 		       struct ib_device *device, u32 port_num,
1665 		       struct sa_path_rec *rec,
1666 		       ib_sa_comp_mask comp_mask,
1667 		       unsigned long timeout_ms, gfp_t gfp_mask,
1668 		       void (*callback)(int status,
1669 					struct sa_path_rec *resp,
1670 					unsigned int num_paths, void *context),
1671 		       void *context,
1672 		       struct ib_sa_query **sa_query)
1673 {
1674 	struct ib_sa_path_query *query;
1675 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1676 	struct ib_sa_port   *port;
1677 	struct ib_mad_agent *agent;
1678 	struct ib_sa_mad *mad;
1679 	enum opa_pr_supported status;
1680 	int ret;
1681 
1682 	if (!sa_dev)
1683 		return -ENODEV;
1684 
1685 	if ((rec->rec_type != SA_PATH_REC_TYPE_IB) &&
1686 	    (rec->rec_type != SA_PATH_REC_TYPE_OPA))
1687 		return -EINVAL;
1688 
1689 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1690 	agent = port->agent;
1691 
1692 	query = kzalloc(sizeof(*query), gfp_mask);
1693 	if (!query)
1694 		return -ENOMEM;
1695 
1696 	query->sa_query.port     = port;
1697 	if (rec->rec_type == SA_PATH_REC_TYPE_OPA) {
1698 		status = opa_pr_query_possible(client, sa_dev, device, port_num);
1699 		if (status == PR_NOT_SUPPORTED) {
1700 			ret = -EINVAL;
1701 			goto err1;
1702 		} else if (status == PR_OPA_SUPPORTED) {
1703 			query->sa_query.flags |= IB_SA_QUERY_OPA;
1704 		} else {
1705 			query->conv_pr =
1706 				kmalloc(sizeof(*query->conv_pr), gfp_mask);
1707 			if (!query->conv_pr) {
1708 				ret = -ENOMEM;
1709 				goto err1;
1710 			}
1711 		}
1712 	}
1713 
1714 	ret = alloc_mad(&query->sa_query, gfp_mask);
1715 	if (ret)
1716 		goto err2;
1717 
1718 	ib_sa_client_get(client);
1719 	query->sa_query.client = client;
1720 	query->callback        = callback;
1721 	query->context         = context;
1722 
1723 	mad = query->sa_query.mad_buf->mad;
1724 	init_mad(&query->sa_query, agent);
1725 
1726 	query->sa_query.callback = callback ? ib_sa_path_rec_callback : NULL;
1727 	query->sa_query.release  = ib_sa_path_rec_release;
1728 	mad->mad_hdr.method	 = IB_MGMT_METHOD_GET;
1729 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_PATH_REC);
1730 	mad->sa_hdr.comp_mask	 = comp_mask;
1731 
1732 	if (query->sa_query.flags & IB_SA_QUERY_OPA) {
1733 		ib_pack(opa_path_rec_table, ARRAY_SIZE(opa_path_rec_table),
1734 			rec, mad->data);
1735 	} else if (query->conv_pr) {
1736 		sa_convert_path_opa_to_ib(query->conv_pr, rec);
1737 		ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table),
1738 			query->conv_pr, mad->data);
1739 	} else {
1740 		ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table),
1741 			rec, mad->data);
1742 	}
1743 
1744 	*sa_query = &query->sa_query;
1745 
1746 	query->sa_query.flags |= IB_SA_ENABLE_LOCAL_SERVICE;
1747 	query->sa_query.mad_buf->context[1] = (query->conv_pr) ?
1748 						query->conv_pr : rec;
1749 
1750 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1751 	if (ret < 0)
1752 		goto err3;
1753 
1754 	return ret;
1755 
1756 err3:
1757 	*sa_query = NULL;
1758 	ib_sa_client_put(query->sa_query.client);
1759 	free_mad(&query->sa_query);
1760 err2:
1761 	kfree(query->conv_pr);
1762 err1:
1763 	kfree(query);
1764 	return ret;
1765 }
1766 EXPORT_SYMBOL(ib_sa_path_rec_get);
1767 
1768 /**
1769  * ib_sa_service_rec_get - Start a Service get query
1770  * @client: SA client
1771  * @device: device to send query on
1772  * @port_num: port number to send query on
1773  * @rec: Service Record to send in query
1774  * @comp_mask: component mask to send in query
1775  * @timeout_ms: time to wait for response
1776  * @gfp_mask: GFP mask to use for internal allocations
1777  * @callback: function called when query completes, times out or is
1778  * canceled
1779  * @context: opaque user context passed to callback
1780  * @sa_query: query context, used to cancel query
1781  *
1782  * Send a Service Record Get query to the SA to look up a path.  The
1783  * callback function will be called when the query completes (or
1784  * fails); status is 0 for a successful response, -EINTR if the query
1785  * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
1786  * occurred sending the query.  The resp parameter of the callback is
1787  * only valid if status is 0.
1788  *
1789  * If the return value of ib_sa_service_rec_get() is negative, it is an
1790  * error code. Otherwise it is a query ID that can be used to cancel
1791  * the query.
1792  */
1793 int ib_sa_service_rec_get(struct ib_sa_client *client,
1794 			  struct ib_device *device, u32 port_num,
1795 			  struct sa_service_rec *rec,
1796 			  ib_sa_comp_mask comp_mask,
1797 			  unsigned long timeout_ms, gfp_t gfp_mask,
1798 			  void (*callback)(int status,
1799 					   struct sa_service_rec *resp,
1800 					   unsigned int num_services,
1801 					   void *context),
1802 			  void *context, struct ib_sa_query **sa_query)
1803 {
1804 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1805 	struct ib_sa_service_query *query;
1806 	struct ib_mad_agent *agent;
1807 	struct ib_sa_port   *port;
1808 	struct ib_sa_mad *mad;
1809 	int ret;
1810 
1811 	if (!sa_dev)
1812 		return -ENODEV;
1813 
1814 	port = &sa_dev->port[port_num - sa_dev->start_port];
1815 	agent = port->agent;
1816 
1817 	query = kzalloc(sizeof(*query), gfp_mask);
1818 	if (!query)
1819 		return -ENOMEM;
1820 
1821 	query->sa_query.port = port;
1822 
1823 	ret = alloc_mad(&query->sa_query, gfp_mask);
1824 	if (ret)
1825 		goto err1;
1826 
1827 	ib_sa_client_get(client);
1828 	query->sa_query.client = client;
1829 	query->callback        = callback;
1830 	query->context         = context;
1831 
1832 	mad = query->sa_query.mad_buf->mad;
1833 	init_mad(&query->sa_query, agent);
1834 
1835 	query->sa_query.rmpp_callback = callback ? ib_sa_service_rec_callback :
1836 		NULL;
1837 	query->sa_query.release = ib_sa_service_rec_release;
1838 	mad->mad_hdr.method	= IB_MGMT_METHOD_GET_TABLE;
1839 	mad->mad_hdr.attr_id	= cpu_to_be16(IB_SA_ATTR_SERVICE_REC);
1840 	mad->sa_hdr.comp_mask	= comp_mask;
1841 
1842 	ib_sa_pack_service(rec, mad->data);
1843 
1844 	*sa_query = &query->sa_query;
1845 	query->sa_query.mad_buf->context[1] = rec;
1846 
1847 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1848 	if (ret < 0)
1849 		goto err2;
1850 
1851 	return ret;
1852 
1853 err2:
1854 	*sa_query = NULL;
1855 	ib_sa_client_put(query->sa_query.client);
1856 	free_mad(&query->sa_query);
1857 err1:
1858 	kfree(query);
1859 	return ret;
1860 }
1861 EXPORT_SYMBOL(ib_sa_service_rec_get);
1862 
1863 static void ib_sa_mcmember_rec_callback(struct ib_sa_query *sa_query,
1864 					int status, struct ib_sa_mad *mad)
1865 {
1866 	struct ib_sa_mcmember_query *query =
1867 		container_of(sa_query, struct ib_sa_mcmember_query, sa_query);
1868 
1869 	if (mad) {
1870 		struct ib_sa_mcmember_rec rec;
1871 
1872 		ib_unpack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
1873 			  mad->data, &rec);
1874 		query->callback(status, &rec, query->context);
1875 	} else
1876 		query->callback(status, NULL, query->context);
1877 }
1878 
1879 static void ib_sa_mcmember_rec_release(struct ib_sa_query *sa_query)
1880 {
1881 	kfree(container_of(sa_query, struct ib_sa_mcmember_query, sa_query));
1882 }
1883 
1884 int ib_sa_mcmember_rec_query(struct ib_sa_client *client,
1885 			     struct ib_device *device, u32 port_num,
1886 			     u8 method,
1887 			     struct ib_sa_mcmember_rec *rec,
1888 			     ib_sa_comp_mask comp_mask,
1889 			     unsigned long timeout_ms, gfp_t gfp_mask,
1890 			     void (*callback)(int status,
1891 					      struct ib_sa_mcmember_rec *resp,
1892 					      void *context),
1893 			     void *context,
1894 			     struct ib_sa_query **sa_query)
1895 {
1896 	struct ib_sa_mcmember_query *query;
1897 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1898 	struct ib_sa_port   *port;
1899 	struct ib_mad_agent *agent;
1900 	struct ib_sa_mad *mad;
1901 	int ret;
1902 
1903 	if (!sa_dev)
1904 		return -ENODEV;
1905 
1906 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1907 	agent = port->agent;
1908 
1909 	query = kzalloc(sizeof(*query), gfp_mask);
1910 	if (!query)
1911 		return -ENOMEM;
1912 
1913 	query->sa_query.port     = port;
1914 	ret = alloc_mad(&query->sa_query, gfp_mask);
1915 	if (ret)
1916 		goto err1;
1917 
1918 	ib_sa_client_get(client);
1919 	query->sa_query.client = client;
1920 	query->callback        = callback;
1921 	query->context         = context;
1922 
1923 	mad = query->sa_query.mad_buf->mad;
1924 	init_mad(&query->sa_query, agent);
1925 
1926 	query->sa_query.callback = callback ? ib_sa_mcmember_rec_callback : NULL;
1927 	query->sa_query.release  = ib_sa_mcmember_rec_release;
1928 	mad->mad_hdr.method	 = method;
1929 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_MC_MEMBER_REC);
1930 	mad->sa_hdr.comp_mask	 = comp_mask;
1931 
1932 	ib_pack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
1933 		rec, mad->data);
1934 
1935 	*sa_query = &query->sa_query;
1936 
1937 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1938 	if (ret < 0)
1939 		goto err2;
1940 
1941 	return ret;
1942 
1943 err2:
1944 	*sa_query = NULL;
1945 	ib_sa_client_put(query->sa_query.client);
1946 	free_mad(&query->sa_query);
1947 
1948 err1:
1949 	kfree(query);
1950 	return ret;
1951 }
1952 
1953 /* Support GuidInfoRecord */
1954 static void ib_sa_guidinfo_rec_callback(struct ib_sa_query *sa_query,
1955 					int status, struct ib_sa_mad *mad)
1956 {
1957 	struct ib_sa_guidinfo_query *query =
1958 		container_of(sa_query, struct ib_sa_guidinfo_query, sa_query);
1959 
1960 	if (mad) {
1961 		struct ib_sa_guidinfo_rec rec;
1962 
1963 		ib_unpack(guidinfo_rec_table, ARRAY_SIZE(guidinfo_rec_table),
1964 			  mad->data, &rec);
1965 		query->callback(status, &rec, query->context);
1966 	} else
1967 		query->callback(status, NULL, query->context);
1968 }
1969 
1970 static void ib_sa_guidinfo_rec_release(struct ib_sa_query *sa_query)
1971 {
1972 	kfree(container_of(sa_query, struct ib_sa_guidinfo_query, sa_query));
1973 }
1974 
1975 int ib_sa_guid_info_rec_query(struct ib_sa_client *client,
1976 			      struct ib_device *device, u32 port_num,
1977 			      struct ib_sa_guidinfo_rec *rec,
1978 			      ib_sa_comp_mask comp_mask, u8 method,
1979 			      unsigned long timeout_ms, gfp_t gfp_mask,
1980 			      void (*callback)(int status,
1981 					       struct ib_sa_guidinfo_rec *resp,
1982 					       void *context),
1983 			      void *context,
1984 			      struct ib_sa_query **sa_query)
1985 {
1986 	struct ib_sa_guidinfo_query *query;
1987 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1988 	struct ib_sa_port *port;
1989 	struct ib_mad_agent *agent;
1990 	struct ib_sa_mad *mad;
1991 	int ret;
1992 
1993 	if (!sa_dev)
1994 		return -ENODEV;
1995 
1996 	if (method != IB_MGMT_METHOD_GET &&
1997 	    method != IB_MGMT_METHOD_SET &&
1998 	    method != IB_SA_METHOD_DELETE) {
1999 		return -EINVAL;
2000 	}
2001 
2002 	port  = &sa_dev->port[port_num - sa_dev->start_port];
2003 	agent = port->agent;
2004 
2005 	query = kzalloc(sizeof(*query), gfp_mask);
2006 	if (!query)
2007 		return -ENOMEM;
2008 
2009 	query->sa_query.port = port;
2010 	ret = alloc_mad(&query->sa_query, gfp_mask);
2011 	if (ret)
2012 		goto err1;
2013 
2014 	ib_sa_client_get(client);
2015 	query->sa_query.client = client;
2016 	query->callback        = callback;
2017 	query->context         = context;
2018 
2019 	mad = query->sa_query.mad_buf->mad;
2020 	init_mad(&query->sa_query, agent);
2021 
2022 	query->sa_query.callback = callback ? ib_sa_guidinfo_rec_callback : NULL;
2023 	query->sa_query.release  = ib_sa_guidinfo_rec_release;
2024 
2025 	mad->mad_hdr.method	 = method;
2026 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_GUID_INFO_REC);
2027 	mad->sa_hdr.comp_mask	 = comp_mask;
2028 
2029 	ib_pack(guidinfo_rec_table, ARRAY_SIZE(guidinfo_rec_table), rec,
2030 		mad->data);
2031 
2032 	*sa_query = &query->sa_query;
2033 
2034 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
2035 	if (ret < 0)
2036 		goto err2;
2037 
2038 	return ret;
2039 
2040 err2:
2041 	*sa_query = NULL;
2042 	ib_sa_client_put(query->sa_query.client);
2043 	free_mad(&query->sa_query);
2044 
2045 err1:
2046 	kfree(query);
2047 	return ret;
2048 }
2049 EXPORT_SYMBOL(ib_sa_guid_info_rec_query);
2050 
2051 struct ib_classport_info_context {
2052 	struct completion	done;
2053 	struct ib_sa_query	*sa_query;
2054 };
2055 
2056 static void ib_classportinfo_cb(void *context)
2057 {
2058 	struct ib_classport_info_context *cb_ctx = context;
2059 
2060 	complete(&cb_ctx->done);
2061 }
2062 
2063 static void ib_sa_classport_info_rec_callback(struct ib_sa_query *sa_query,
2064 					      int status, struct ib_sa_mad *mad)
2065 {
2066 	unsigned long flags;
2067 	struct ib_sa_classport_info_query *query =
2068 		container_of(sa_query, struct ib_sa_classport_info_query, sa_query);
2069 	struct ib_sa_classport_cache *info = &sa_query->port->classport_info;
2070 
2071 	if (mad) {
2072 		if (sa_query->flags & IB_SA_QUERY_OPA) {
2073 			struct opa_class_port_info rec;
2074 
2075 			ib_unpack(opa_classport_info_rec_table,
2076 				  ARRAY_SIZE(opa_classport_info_rec_table),
2077 				  mad->data, &rec);
2078 
2079 			spin_lock_irqsave(&sa_query->port->classport_lock,
2080 					  flags);
2081 			if (!status && !info->valid) {
2082 				memcpy(&info->data.opa, &rec,
2083 				       sizeof(info->data.opa));
2084 
2085 				info->valid = true;
2086 				info->data.type = RDMA_CLASS_PORT_INFO_OPA;
2087 			}
2088 			spin_unlock_irqrestore(&sa_query->port->classport_lock,
2089 					       flags);
2090 
2091 		} else {
2092 			struct ib_class_port_info rec;
2093 
2094 			ib_unpack(ib_classport_info_rec_table,
2095 				  ARRAY_SIZE(ib_classport_info_rec_table),
2096 				  mad->data, &rec);
2097 
2098 			spin_lock_irqsave(&sa_query->port->classport_lock,
2099 					  flags);
2100 			if (!status && !info->valid) {
2101 				memcpy(&info->data.ib, &rec,
2102 				       sizeof(info->data.ib));
2103 
2104 				info->valid = true;
2105 				info->data.type = RDMA_CLASS_PORT_INFO_IB;
2106 			}
2107 			spin_unlock_irqrestore(&sa_query->port->classport_lock,
2108 					       flags);
2109 		}
2110 	}
2111 	query->callback(query->context);
2112 }
2113 
2114 static void ib_sa_classport_info_rec_release(struct ib_sa_query *sa_query)
2115 {
2116 	kfree(container_of(sa_query, struct ib_sa_classport_info_query,
2117 			   sa_query));
2118 }
2119 
2120 static int ib_sa_classport_info_rec_query(struct ib_sa_port *port,
2121 					  unsigned long timeout_ms,
2122 					  void (*callback)(void *context),
2123 					  void *context,
2124 					  struct ib_sa_query **sa_query)
2125 {
2126 	struct ib_mad_agent *agent;
2127 	struct ib_sa_classport_info_query *query;
2128 	struct ib_sa_mad *mad;
2129 	gfp_t gfp_mask = GFP_KERNEL;
2130 	int ret;
2131 
2132 	agent = port->agent;
2133 
2134 	query = kzalloc(sizeof(*query), gfp_mask);
2135 	if (!query)
2136 		return -ENOMEM;
2137 
2138 	query->sa_query.port = port;
2139 	query->sa_query.flags |= rdma_cap_opa_ah(port->agent->device,
2140 						 port->port_num) ?
2141 				 IB_SA_QUERY_OPA : 0;
2142 	ret = alloc_mad(&query->sa_query, gfp_mask);
2143 	if (ret)
2144 		goto err_free;
2145 
2146 	query->callback = callback;
2147 	query->context = context;
2148 
2149 	mad = query->sa_query.mad_buf->mad;
2150 	init_mad(&query->sa_query, agent);
2151 
2152 	query->sa_query.callback = ib_sa_classport_info_rec_callback;
2153 	query->sa_query.release  = ib_sa_classport_info_rec_release;
2154 	mad->mad_hdr.method	 = IB_MGMT_METHOD_GET;
2155 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_CLASS_PORTINFO);
2156 	mad->sa_hdr.comp_mask	 = 0;
2157 	*sa_query = &query->sa_query;
2158 
2159 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
2160 	if (ret < 0)
2161 		goto err_free_mad;
2162 
2163 	return ret;
2164 
2165 err_free_mad:
2166 	*sa_query = NULL;
2167 	free_mad(&query->sa_query);
2168 
2169 err_free:
2170 	kfree(query);
2171 	return ret;
2172 }
2173 
2174 static void update_ib_cpi(struct work_struct *work)
2175 {
2176 	struct ib_sa_port *port =
2177 		container_of(work, struct ib_sa_port, ib_cpi_work.work);
2178 	struct ib_classport_info_context *cb_context;
2179 	unsigned long flags;
2180 	int ret;
2181 
2182 	/* If the classport info is valid, nothing
2183 	 * to do here.
2184 	 */
2185 	spin_lock_irqsave(&port->classport_lock, flags);
2186 	if (port->classport_info.valid) {
2187 		spin_unlock_irqrestore(&port->classport_lock, flags);
2188 		return;
2189 	}
2190 	spin_unlock_irqrestore(&port->classport_lock, flags);
2191 
2192 	cb_context = kmalloc(sizeof(*cb_context), GFP_KERNEL);
2193 	if (!cb_context)
2194 		goto err_nomem;
2195 
2196 	init_completion(&cb_context->done);
2197 
2198 	ret = ib_sa_classport_info_rec_query(port, 3000,
2199 					     ib_classportinfo_cb, cb_context,
2200 					     &cb_context->sa_query);
2201 	if (ret < 0)
2202 		goto free_cb_err;
2203 	wait_for_completion(&cb_context->done);
2204 free_cb_err:
2205 	kfree(cb_context);
2206 	spin_lock_irqsave(&port->classport_lock, flags);
2207 
2208 	/* If the classport info is still not valid, the query should have
2209 	 * failed for some reason. Retry issuing the query
2210 	 */
2211 	if (!port->classport_info.valid) {
2212 		port->classport_info.retry_cnt++;
2213 		if (port->classport_info.retry_cnt <=
2214 		    IB_SA_CPI_MAX_RETRY_CNT) {
2215 			unsigned long delay =
2216 				msecs_to_jiffies(IB_SA_CPI_RETRY_WAIT);
2217 
2218 			queue_delayed_work(ib_wq, &port->ib_cpi_work, delay);
2219 		}
2220 	}
2221 	spin_unlock_irqrestore(&port->classport_lock, flags);
2222 
2223 err_nomem:
2224 	return;
2225 }
2226 
2227 static void send_handler(struct ib_mad_agent *agent,
2228 			 struct ib_mad_send_wc *mad_send_wc)
2229 {
2230 	struct ib_sa_query *query = mad_send_wc->send_buf->context[0];
2231 	unsigned long flags;
2232 	int status = 0;
2233 
2234 	if (query->callback || query->rmpp_callback) {
2235 		switch (mad_send_wc->status) {
2236 		case IB_WC_SUCCESS:
2237 			/* No callback -- already got recv */
2238 			break;
2239 		case IB_WC_RESP_TIMEOUT_ERR:
2240 			status = -ETIMEDOUT;
2241 			break;
2242 		case IB_WC_WR_FLUSH_ERR:
2243 			status = -EINTR;
2244 			break;
2245 		default:
2246 			status = -EIO;
2247 			break;
2248 		}
2249 
2250 		if (status)
2251 			query->callback ? query->callback(query, status, NULL) :
2252 				query->rmpp_callback(query, status, NULL);
2253 	}
2254 
2255 	xa_lock_irqsave(&queries, flags);
2256 	__xa_erase(&queries, query->id);
2257 	xa_unlock_irqrestore(&queries, flags);
2258 
2259 	free_mad(query);
2260 	if (query->client)
2261 		ib_sa_client_put(query->client);
2262 	query->release(query);
2263 }
2264 
2265 static void recv_handler(struct ib_mad_agent *mad_agent,
2266 			 struct ib_mad_send_buf *send_buf,
2267 			 struct ib_mad_recv_wc *mad_recv_wc)
2268 {
2269 	struct ib_sa_query *query;
2270 	struct ib_mad *mad;
2271 
2272 
2273 	if (!send_buf)
2274 		return;
2275 
2276 	query = send_buf->context[0];
2277 	mad = mad_recv_wc->recv_buf.mad;
2278 
2279 	if (query->rmpp_callback) {
2280 		if (mad_recv_wc->wc->status == IB_WC_SUCCESS)
2281 			query->rmpp_callback(query, mad->mad_hdr.status ?
2282 					     -EINVAL : 0, mad_recv_wc);
2283 		else
2284 			query->rmpp_callback(query, -EIO, NULL);
2285 	} else if (query->callback) {
2286 		if (mad_recv_wc->wc->status == IB_WC_SUCCESS)
2287 			query->callback(query, mad->mad_hdr.status ?
2288 					-EINVAL : 0, (struct ib_sa_mad *)mad);
2289 		else
2290 			query->callback(query, -EIO, NULL);
2291 	}
2292 
2293 	ib_free_recv_mad(mad_recv_wc);
2294 }
2295 
2296 static void update_sm_ah(struct work_struct *work)
2297 {
2298 	struct ib_sa_port *port =
2299 		container_of(work, struct ib_sa_port, update_task);
2300 	struct ib_sa_sm_ah *new_ah;
2301 	struct ib_port_attr port_attr;
2302 	struct rdma_ah_attr   ah_attr;
2303 	bool grh_required;
2304 
2305 	if (ib_query_port(port->agent->device, port->port_num, &port_attr)) {
2306 		pr_warn("Couldn't query port\n");
2307 		return;
2308 	}
2309 
2310 	new_ah = kmalloc(sizeof(*new_ah), GFP_KERNEL);
2311 	if (!new_ah)
2312 		return;
2313 
2314 	kref_init(&new_ah->ref);
2315 	new_ah->src_path_mask = (1 << port_attr.lmc) - 1;
2316 
2317 	new_ah->pkey_index = 0;
2318 	if (ib_find_pkey(port->agent->device, port->port_num,
2319 			 IB_DEFAULT_PKEY_FULL, &new_ah->pkey_index))
2320 		pr_err("Couldn't find index for default PKey\n");
2321 
2322 	memset(&ah_attr, 0, sizeof(ah_attr));
2323 	ah_attr.type = rdma_ah_find_type(port->agent->device,
2324 					 port->port_num);
2325 	rdma_ah_set_dlid(&ah_attr, port_attr.sm_lid);
2326 	rdma_ah_set_sl(&ah_attr, port_attr.sm_sl);
2327 	rdma_ah_set_port_num(&ah_attr, port->port_num);
2328 
2329 	grh_required = rdma_is_grh_required(port->agent->device,
2330 					    port->port_num);
2331 
2332 	/*
2333 	 * The OPA sm_lid of 0xFFFF needs special handling so that it can be
2334 	 * differentiated from a permissive LID of 0xFFFF.  We set the
2335 	 * grh_required flag here so the SA can program the DGID in the
2336 	 * address handle appropriately
2337 	 */
2338 	if (ah_attr.type == RDMA_AH_ATTR_TYPE_OPA &&
2339 	    (grh_required ||
2340 	     port_attr.sm_lid == be16_to_cpu(IB_LID_PERMISSIVE)))
2341 		rdma_ah_set_make_grd(&ah_attr, true);
2342 
2343 	if (ah_attr.type == RDMA_AH_ATTR_TYPE_IB && grh_required) {
2344 		rdma_ah_set_ah_flags(&ah_attr, IB_AH_GRH);
2345 		rdma_ah_set_subnet_prefix(&ah_attr,
2346 					  cpu_to_be64(port_attr.subnet_prefix));
2347 		rdma_ah_set_interface_id(&ah_attr,
2348 					 cpu_to_be64(IB_SA_WELL_KNOWN_GUID));
2349 	}
2350 
2351 	new_ah->ah = rdma_create_ah(port->agent->qp->pd, &ah_attr,
2352 				    RDMA_CREATE_AH_SLEEPABLE);
2353 	if (IS_ERR(new_ah->ah)) {
2354 		pr_warn("Couldn't create new SM AH\n");
2355 		kfree(new_ah);
2356 		return;
2357 	}
2358 
2359 	spin_lock_irq(&port->ah_lock);
2360 	if (port->sm_ah)
2361 		kref_put(&port->sm_ah->ref, free_sm_ah);
2362 	port->sm_ah = new_ah;
2363 	spin_unlock_irq(&port->ah_lock);
2364 }
2365 
2366 static void ib_sa_event(struct ib_event_handler *handler,
2367 			struct ib_event *event)
2368 {
2369 	if (event->event == IB_EVENT_PORT_ERR    ||
2370 	    event->event == IB_EVENT_PORT_ACTIVE ||
2371 	    event->event == IB_EVENT_LID_CHANGE  ||
2372 	    event->event == IB_EVENT_PKEY_CHANGE ||
2373 	    event->event == IB_EVENT_SM_CHANGE   ||
2374 	    event->event == IB_EVENT_CLIENT_REREGISTER) {
2375 		unsigned long flags;
2376 		struct ib_sa_device *sa_dev =
2377 			container_of(handler, typeof(*sa_dev), event_handler);
2378 		u32 port_num = event->element.port_num - sa_dev->start_port;
2379 		struct ib_sa_port *port = &sa_dev->port[port_num];
2380 
2381 		if (!rdma_cap_ib_sa(handler->device, port->port_num))
2382 			return;
2383 
2384 		spin_lock_irqsave(&port->ah_lock, flags);
2385 		if (port->sm_ah)
2386 			kref_put(&port->sm_ah->ref, free_sm_ah);
2387 		port->sm_ah = NULL;
2388 		spin_unlock_irqrestore(&port->ah_lock, flags);
2389 
2390 		if (event->event == IB_EVENT_SM_CHANGE ||
2391 		    event->event == IB_EVENT_CLIENT_REREGISTER ||
2392 		    event->event == IB_EVENT_LID_CHANGE ||
2393 		    event->event == IB_EVENT_PORT_ACTIVE) {
2394 			unsigned long delay =
2395 				msecs_to_jiffies(IB_SA_CPI_RETRY_WAIT);
2396 
2397 			spin_lock_irqsave(&port->classport_lock, flags);
2398 			port->classport_info.valid = false;
2399 			port->classport_info.retry_cnt = 0;
2400 			spin_unlock_irqrestore(&port->classport_lock, flags);
2401 			queue_delayed_work(ib_wq,
2402 					   &port->ib_cpi_work, delay);
2403 		}
2404 		queue_work(ib_wq, &sa_dev->port[port_num].update_task);
2405 	}
2406 }
2407 
2408 static int ib_sa_add_one(struct ib_device *device)
2409 {
2410 	struct ib_sa_device *sa_dev;
2411 	int s, e, i;
2412 	int count = 0;
2413 	int ret;
2414 
2415 	s = rdma_start_port(device);
2416 	e = rdma_end_port(device);
2417 
2418 	sa_dev = kzalloc(struct_size(sa_dev, port,
2419 				     size_add(size_sub(e, s), 1)),
2420 			 GFP_KERNEL);
2421 	if (!sa_dev)
2422 		return -ENOMEM;
2423 
2424 	sa_dev->start_port = s;
2425 	sa_dev->end_port   = e;
2426 
2427 	for (i = 0; i <= e - s; ++i) {
2428 		spin_lock_init(&sa_dev->port[i].ah_lock);
2429 		if (!rdma_cap_ib_sa(device, i + 1))
2430 			continue;
2431 
2432 		sa_dev->port[i].sm_ah    = NULL;
2433 		sa_dev->port[i].port_num = i + s;
2434 
2435 		spin_lock_init(&sa_dev->port[i].classport_lock);
2436 		sa_dev->port[i].classport_info.valid = false;
2437 
2438 		sa_dev->port[i].agent =
2439 			ib_register_mad_agent(device, i + s, IB_QPT_GSI,
2440 					      NULL, IB_MGMT_RMPP_VERSION,
2441 					      send_handler, recv_handler,
2442 					      sa_dev, 0);
2443 		if (IS_ERR(sa_dev->port[i].agent)) {
2444 			ret = PTR_ERR(sa_dev->port[i].agent);
2445 			goto err;
2446 		}
2447 
2448 		INIT_WORK(&sa_dev->port[i].update_task, update_sm_ah);
2449 		INIT_DELAYED_WORK(&sa_dev->port[i].ib_cpi_work,
2450 				  update_ib_cpi);
2451 
2452 		count++;
2453 	}
2454 
2455 	if (!count) {
2456 		ret = -EOPNOTSUPP;
2457 		goto free;
2458 	}
2459 
2460 	ib_set_client_data(device, &sa_client, sa_dev);
2461 
2462 	/*
2463 	 * We register our event handler after everything is set up,
2464 	 * and then update our cached info after the event handler is
2465 	 * registered to avoid any problems if a port changes state
2466 	 * during our initialization.
2467 	 */
2468 
2469 	INIT_IB_EVENT_HANDLER(&sa_dev->event_handler, device, ib_sa_event);
2470 	ib_register_event_handler(&sa_dev->event_handler);
2471 
2472 	for (i = 0; i <= e - s; ++i) {
2473 		if (rdma_cap_ib_sa(device, i + 1))
2474 			update_sm_ah(&sa_dev->port[i].update_task);
2475 	}
2476 
2477 	return 0;
2478 
2479 err:
2480 	while (--i >= 0) {
2481 		if (rdma_cap_ib_sa(device, i + 1))
2482 			ib_unregister_mad_agent(sa_dev->port[i].agent);
2483 	}
2484 free:
2485 	kfree(sa_dev);
2486 	return ret;
2487 }
2488 
2489 static void ib_sa_remove_one(struct ib_device *device, void *client_data)
2490 {
2491 	struct ib_sa_device *sa_dev = client_data;
2492 	int i;
2493 
2494 	ib_unregister_event_handler(&sa_dev->event_handler);
2495 	flush_workqueue(ib_wq);
2496 
2497 	for (i = 0; i <= sa_dev->end_port - sa_dev->start_port; ++i) {
2498 		if (rdma_cap_ib_sa(device, i + 1)) {
2499 			cancel_delayed_work_sync(&sa_dev->port[i].ib_cpi_work);
2500 			ib_unregister_mad_agent(sa_dev->port[i].agent);
2501 			if (sa_dev->port[i].sm_ah)
2502 				kref_put(&sa_dev->port[i].sm_ah->ref, free_sm_ah);
2503 		}
2504 
2505 	}
2506 
2507 	kfree(sa_dev);
2508 }
2509 
2510 int ib_sa_init(void)
2511 {
2512 	int ret;
2513 
2514 	get_random_bytes(&tid, sizeof tid);
2515 
2516 	atomic_set(&ib_nl_sa_request_seq, 0);
2517 
2518 	ret = ib_register_client(&sa_client);
2519 	if (ret) {
2520 		pr_err("Couldn't register ib_sa client\n");
2521 		goto err1;
2522 	}
2523 
2524 	ret = mcast_init();
2525 	if (ret) {
2526 		pr_err("Couldn't initialize multicast handling\n");
2527 		goto err2;
2528 	}
2529 
2530 	ib_nl_wq = alloc_ordered_workqueue("ib_nl_sa_wq", WQ_MEM_RECLAIM);
2531 	if (!ib_nl_wq) {
2532 		ret = -ENOMEM;
2533 		goto err3;
2534 	}
2535 
2536 	INIT_DELAYED_WORK(&ib_nl_timed_work, ib_nl_request_timeout);
2537 
2538 	return 0;
2539 
2540 err3:
2541 	mcast_cleanup();
2542 err2:
2543 	ib_unregister_client(&sa_client);
2544 err1:
2545 	return ret;
2546 }
2547 
2548 void ib_sa_cleanup(void)
2549 {
2550 	cancel_delayed_work(&ib_nl_timed_work);
2551 	destroy_workqueue(ib_nl_wq);
2552 	mcast_cleanup();
2553 	ib_unregister_client(&sa_client);
2554 	WARN_ON(!xa_empty(&queries));
2555 }
2556