xref: /linux/drivers/infiniband/core/sa_query.c (revision 8d765af51a099884bab37a51e211c7047f67f1f3)
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 	delta = timeout - sa_local_svc_timeout_ms;
1078 	if (delta < 0)
1079 		abs_delta = -delta;
1080 	else
1081 		abs_delta = delta;
1082 
1083 	if (delta != 0) {
1084 		spin_lock_irqsave(&ib_nl_request_lock, flags);
1085 		sa_local_svc_timeout_ms = timeout;
1086 		list_for_each_entry(query, &ib_nl_request_list, list) {
1087 			if (delta < 0 && abs_delta > query->timeout)
1088 				query->timeout = 0;
1089 			else
1090 				query->timeout += delta;
1091 
1092 			/* Get the new delay from the first entry */
1093 			if (!delay) {
1094 				delay = query->timeout - jiffies;
1095 				if (delay <= 0)
1096 					delay = 1;
1097 			}
1098 		}
1099 		if (delay)
1100 			mod_delayed_work(ib_nl_wq, &ib_nl_timed_work,
1101 					 (unsigned long)delay);
1102 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1103 	}
1104 
1105 settimeout_out:
1106 	return 0;
1107 }
1108 
1109 static inline int ib_nl_is_good_resolve_resp(const struct nlmsghdr *nlh)
1110 {
1111 	struct nlattr *tb[LS_NLA_TYPE_MAX];
1112 	int ret;
1113 
1114 	if (nlh->nlmsg_flags & RDMA_NL_LS_F_ERR)
1115 		return 0;
1116 
1117 	ret = nla_parse_deprecated(tb, LS_NLA_TYPE_MAX - 1, nlmsg_data(nlh),
1118 				   nlmsg_len(nlh), ib_nl_policy, NULL);
1119 	if (ret)
1120 		return 0;
1121 
1122 	return 1;
1123 }
1124 
1125 int ib_nl_handle_resolve_resp(struct sk_buff *skb,
1126 			      struct nlmsghdr *nlh,
1127 			      struct netlink_ext_ack *extack)
1128 {
1129 	unsigned long flags;
1130 	struct ib_sa_query *query = NULL, *iter;
1131 	struct ib_mad_send_buf *send_buf;
1132 	struct ib_mad_send_wc mad_send_wc;
1133 	int ret;
1134 
1135 	if ((nlh->nlmsg_flags & NLM_F_REQUEST) ||
1136 	    !(NETLINK_CB(skb).sk))
1137 		return -EPERM;
1138 
1139 	spin_lock_irqsave(&ib_nl_request_lock, flags);
1140 	list_for_each_entry(iter, &ib_nl_request_list, list) {
1141 		/*
1142 		 * If the query is cancelled, let the timeout routine
1143 		 * take care of it.
1144 		 */
1145 		if (nlh->nlmsg_seq == iter->seq) {
1146 			if (!ib_sa_query_cancelled(iter)) {
1147 				list_del(&iter->list);
1148 				query = iter;
1149 			}
1150 			break;
1151 		}
1152 	}
1153 
1154 	if (!query) {
1155 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1156 		goto resp_out;
1157 	}
1158 
1159 	send_buf = query->mad_buf;
1160 
1161 	if (!ib_nl_is_good_resolve_resp(nlh)) {
1162 		/* if the result is a failure, send out the packet via IB */
1163 		ib_sa_disable_local_svc(query);
1164 		ret = ib_post_send_mad(query->mad_buf, NULL);
1165 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1166 		if (ret) {
1167 			mad_send_wc.send_buf = send_buf;
1168 			mad_send_wc.status = IB_WC_GENERAL_ERR;
1169 			send_handler(query->port->agent, &mad_send_wc);
1170 		}
1171 	} else {
1172 		spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1173 		ib_nl_process_good_resolve_rsp(query, nlh);
1174 	}
1175 
1176 resp_out:
1177 	return 0;
1178 }
1179 
1180 static void free_sm_ah(struct kref *kref)
1181 {
1182 	struct ib_sa_sm_ah *sm_ah = container_of(kref, struct ib_sa_sm_ah, ref);
1183 
1184 	rdma_destroy_ah(sm_ah->ah, 0);
1185 	kfree(sm_ah);
1186 }
1187 
1188 void ib_sa_register_client(struct ib_sa_client *client)
1189 {
1190 	atomic_set(&client->users, 1);
1191 	init_completion(&client->comp);
1192 }
1193 EXPORT_SYMBOL(ib_sa_register_client);
1194 
1195 void ib_sa_unregister_client(struct ib_sa_client *client)
1196 {
1197 	ib_sa_client_put(client);
1198 	wait_for_completion(&client->comp);
1199 }
1200 EXPORT_SYMBOL(ib_sa_unregister_client);
1201 
1202 /**
1203  * ib_sa_cancel_query - try to cancel an SA query
1204  * @id:ID of query to cancel
1205  * @query:query pointer to cancel
1206  *
1207  * Try to cancel an SA query.  If the id and query don't match up or
1208  * the query has already completed, nothing is done.  Otherwise the
1209  * query is canceled and will complete with a status of -EINTR.
1210  */
1211 void ib_sa_cancel_query(int id, struct ib_sa_query *query)
1212 {
1213 	unsigned long flags;
1214 	struct ib_mad_send_buf *mad_buf;
1215 
1216 	xa_lock_irqsave(&queries, flags);
1217 	if (xa_load(&queries, id) != query) {
1218 		xa_unlock_irqrestore(&queries, flags);
1219 		return;
1220 	}
1221 	mad_buf = query->mad_buf;
1222 	xa_unlock_irqrestore(&queries, flags);
1223 
1224 	/*
1225 	 * If the query is still on the netlink request list, schedule
1226 	 * it to be cancelled by the timeout routine. Otherwise, it has been
1227 	 * sent to the MAD layer and has to be cancelled from there.
1228 	 */
1229 	if (!ib_nl_cancel_request(query))
1230 		ib_cancel_mad(mad_buf);
1231 }
1232 EXPORT_SYMBOL(ib_sa_cancel_query);
1233 
1234 static u8 get_src_path_mask(struct ib_device *device, u32 port_num)
1235 {
1236 	struct ib_sa_device *sa_dev;
1237 	struct ib_sa_port   *port;
1238 	unsigned long flags;
1239 	u8 src_path_mask;
1240 
1241 	sa_dev = ib_get_client_data(device, &sa_client);
1242 	if (!sa_dev)
1243 		return 0x7f;
1244 
1245 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1246 	spin_lock_irqsave(&port->ah_lock, flags);
1247 	src_path_mask = port->sm_ah ? port->sm_ah->src_path_mask : 0x7f;
1248 	spin_unlock_irqrestore(&port->ah_lock, flags);
1249 
1250 	return src_path_mask;
1251 }
1252 
1253 static int init_ah_attr_grh_fields(struct ib_device *device, u32 port_num,
1254 				   struct sa_path_rec *rec,
1255 				   struct rdma_ah_attr *ah_attr,
1256 				   const struct ib_gid_attr *gid_attr)
1257 {
1258 	enum ib_gid_type type = sa_conv_pathrec_to_gid_type(rec);
1259 
1260 	if (!gid_attr) {
1261 		gid_attr = rdma_find_gid_by_port(device, &rec->sgid, type,
1262 						 port_num, NULL);
1263 		if (IS_ERR(gid_attr))
1264 			return PTR_ERR(gid_attr);
1265 	} else
1266 		rdma_hold_gid_attr(gid_attr);
1267 
1268 	rdma_move_grh_sgid_attr(ah_attr, &rec->dgid,
1269 				be32_to_cpu(rec->flow_label),
1270 				rec->hop_limit,	rec->traffic_class,
1271 				gid_attr);
1272 	return 0;
1273 }
1274 
1275 /**
1276  * ib_init_ah_attr_from_path - Initialize address handle attributes based on
1277  *   an SA path record.
1278  * @device: Device associated ah attributes initialization.
1279  * @port_num: Port on the specified device.
1280  * @rec: path record entry to use for ah attributes initialization.
1281  * @ah_attr: address handle attributes to initialization from path record.
1282  * @gid_attr: SGID attribute to consider during initialization.
1283  *
1284  * When ib_init_ah_attr_from_path() returns success,
1285  * (a) for IB link layer it optionally contains a reference to SGID attribute
1286  * when GRH is present for IB link layer.
1287  * (b) for RoCE link layer it contains a reference to SGID attribute.
1288  * User must invoke rdma_destroy_ah_attr() to release reference to SGID
1289  * attributes which are initialized using ib_init_ah_attr_from_path().
1290  */
1291 int ib_init_ah_attr_from_path(struct ib_device *device, u32 port_num,
1292 			      struct sa_path_rec *rec,
1293 			      struct rdma_ah_attr *ah_attr,
1294 			      const struct ib_gid_attr *gid_attr)
1295 {
1296 	int ret = 0;
1297 
1298 	memset(ah_attr, 0, sizeof(*ah_attr));
1299 	ah_attr->type = rdma_ah_find_type(device, port_num);
1300 	rdma_ah_set_sl(ah_attr, rec->sl);
1301 	rdma_ah_set_port_num(ah_attr, port_num);
1302 	rdma_ah_set_static_rate(ah_attr, rec->rate);
1303 
1304 	if (sa_path_is_roce(rec)) {
1305 		ret = roce_resolve_route_from_path(rec, gid_attr);
1306 		if (ret)
1307 			return ret;
1308 
1309 		memcpy(ah_attr->roce.dmac, sa_path_get_dmac(rec), ETH_ALEN);
1310 	} else {
1311 		rdma_ah_set_dlid(ah_attr, be32_to_cpu(sa_path_get_dlid(rec)));
1312 		if (sa_path_is_opa(rec) &&
1313 		    rdma_ah_get_dlid(ah_attr) == be16_to_cpu(IB_LID_PERMISSIVE))
1314 			rdma_ah_set_make_grd(ah_attr, true);
1315 
1316 		rdma_ah_set_path_bits(ah_attr,
1317 				      be32_to_cpu(sa_path_get_slid(rec)) &
1318 				      get_src_path_mask(device, port_num));
1319 	}
1320 
1321 	if (rec->hop_limit > 0 || sa_path_is_roce(rec))
1322 		ret = init_ah_attr_grh_fields(device, port_num,
1323 					      rec, ah_attr, gid_attr);
1324 	return ret;
1325 }
1326 EXPORT_SYMBOL(ib_init_ah_attr_from_path);
1327 
1328 static int alloc_mad(struct ib_sa_query *query, gfp_t gfp_mask)
1329 {
1330 	struct rdma_ah_attr ah_attr;
1331 	unsigned long flags;
1332 
1333 	spin_lock_irqsave(&query->port->ah_lock, flags);
1334 	if (!query->port->sm_ah) {
1335 		spin_unlock_irqrestore(&query->port->ah_lock, flags);
1336 		return -EAGAIN;
1337 	}
1338 	kref_get(&query->port->sm_ah->ref);
1339 	query->sm_ah = query->port->sm_ah;
1340 	spin_unlock_irqrestore(&query->port->ah_lock, flags);
1341 
1342 	/*
1343 	 * Always check if sm_ah has valid dlid assigned,
1344 	 * before querying for class port info
1345 	 */
1346 	if ((rdma_query_ah(query->sm_ah->ah, &ah_attr) < 0) ||
1347 	    !rdma_is_valid_unicast_lid(&ah_attr)) {
1348 		kref_put(&query->sm_ah->ref, free_sm_ah);
1349 		return -EAGAIN;
1350 	}
1351 	query->mad_buf = ib_create_send_mad(query->port->agent, 1,
1352 					    query->sm_ah->pkey_index,
1353 					    0, IB_MGMT_SA_HDR, IB_MGMT_SA_DATA,
1354 					    gfp_mask,
1355 					    ((query->flags & IB_SA_QUERY_OPA) ?
1356 					     OPA_MGMT_BASE_VERSION :
1357 					     IB_MGMT_BASE_VERSION));
1358 	if (IS_ERR(query->mad_buf)) {
1359 		kref_put(&query->sm_ah->ref, free_sm_ah);
1360 		return -ENOMEM;
1361 	}
1362 
1363 	query->mad_buf->ah = query->sm_ah->ah;
1364 
1365 	return 0;
1366 }
1367 
1368 static void free_mad(struct ib_sa_query *query)
1369 {
1370 	ib_free_send_mad(query->mad_buf);
1371 	kref_put(&query->sm_ah->ref, free_sm_ah);
1372 }
1373 
1374 static void init_mad(struct ib_sa_query *query, struct ib_mad_agent *agent)
1375 {
1376 	struct ib_sa_mad *mad = query->mad_buf->mad;
1377 	unsigned long flags;
1378 
1379 	memset(mad, 0, sizeof *mad);
1380 
1381 	if (query->flags & IB_SA_QUERY_OPA) {
1382 		mad->mad_hdr.base_version  = OPA_MGMT_BASE_VERSION;
1383 		mad->mad_hdr.class_version = OPA_SA_CLASS_VERSION;
1384 	} else {
1385 		mad->mad_hdr.base_version  = IB_MGMT_BASE_VERSION;
1386 		mad->mad_hdr.class_version = IB_SA_CLASS_VERSION;
1387 	}
1388 	mad->mad_hdr.mgmt_class    = IB_MGMT_CLASS_SUBN_ADM;
1389 	spin_lock_irqsave(&tid_lock, flags);
1390 	mad->mad_hdr.tid           =
1391 		cpu_to_be64(((u64) agent->hi_tid) << 32 | tid++);
1392 	spin_unlock_irqrestore(&tid_lock, flags);
1393 }
1394 
1395 static int send_mad(struct ib_sa_query *query, unsigned long timeout_ms,
1396 		    gfp_t gfp_mask)
1397 {
1398 	unsigned long flags;
1399 	int ret, id;
1400 	const int nmbr_sa_query_retries = 10;
1401 
1402 	xa_lock_irqsave(&queries, flags);
1403 	ret = __xa_alloc(&queries, &id, query, xa_limit_32b, gfp_mask);
1404 	xa_unlock_irqrestore(&queries, flags);
1405 	if (ret < 0)
1406 		return ret;
1407 
1408 	query->mad_buf->timeout_ms  = timeout_ms / nmbr_sa_query_retries;
1409 	query->mad_buf->retries = nmbr_sa_query_retries;
1410 	if (!query->mad_buf->timeout_ms) {
1411 		/* Special case, very small timeout_ms */
1412 		query->mad_buf->timeout_ms = 1;
1413 		query->mad_buf->retries = timeout_ms;
1414 	}
1415 	query->mad_buf->context[0] = query;
1416 	query->id = id;
1417 
1418 	if ((query->flags & IB_SA_ENABLE_LOCAL_SERVICE) &&
1419 	    (!(query->flags & IB_SA_QUERY_OPA))) {
1420 		if (rdma_nl_chk_listeners(RDMA_NL_GROUP_LS)) {
1421 			if (!ib_nl_make_request(query, gfp_mask))
1422 				return id;
1423 		}
1424 		ib_sa_disable_local_svc(query);
1425 	}
1426 
1427 	ret = ib_post_send_mad(query->mad_buf, NULL);
1428 	if (ret) {
1429 		xa_lock_irqsave(&queries, flags);
1430 		__xa_erase(&queries, id);
1431 		xa_unlock_irqrestore(&queries, flags);
1432 	}
1433 
1434 	/*
1435 	 * It's not safe to dereference query any more, because the
1436 	 * send may already have completed and freed the query in
1437 	 * another context.
1438 	 */
1439 	return ret ? ret : id;
1440 }
1441 
1442 void ib_sa_unpack_path(void *attribute, struct sa_path_rec *rec)
1443 {
1444 	ib_unpack(path_rec_table, ARRAY_SIZE(path_rec_table), attribute, rec);
1445 }
1446 EXPORT_SYMBOL(ib_sa_unpack_path);
1447 
1448 void ib_sa_pack_path(struct sa_path_rec *rec, void *attribute)
1449 {
1450 	ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table), rec, attribute);
1451 }
1452 EXPORT_SYMBOL(ib_sa_pack_path);
1453 
1454 void ib_sa_pack_service(struct sa_service_rec *rec, void *attribute)
1455 {
1456 	ib_pack(service_rec_table, ARRAY_SIZE(service_rec_table), rec,
1457 		attribute);
1458 }
1459 EXPORT_SYMBOL(ib_sa_pack_service);
1460 
1461 void ib_sa_unpack_service(void *attribute, struct sa_service_rec *rec)
1462 {
1463 	ib_unpack(service_rec_table, ARRAY_SIZE(service_rec_table), attribute,
1464 		  rec);
1465 }
1466 EXPORT_SYMBOL(ib_sa_unpack_service);
1467 
1468 static bool ib_sa_opa_pathrecord_support(struct ib_sa_client *client,
1469 					 struct ib_sa_device *sa_dev,
1470 					 u32 port_num)
1471 {
1472 	struct ib_sa_port *port;
1473 	unsigned long flags;
1474 	bool ret = false;
1475 
1476 	port = &sa_dev->port[port_num - sa_dev->start_port];
1477 	spin_lock_irqsave(&port->classport_lock, flags);
1478 	if (!port->classport_info.valid)
1479 		goto ret;
1480 
1481 	if (port->classport_info.data.type == RDMA_CLASS_PORT_INFO_OPA)
1482 		ret = opa_get_cpi_capmask2(&port->classport_info.data.opa) &
1483 			OPA_CLASS_PORT_INFO_PR_SUPPORT;
1484 ret:
1485 	spin_unlock_irqrestore(&port->classport_lock, flags);
1486 	return ret;
1487 }
1488 
1489 enum opa_pr_supported {
1490 	PR_NOT_SUPPORTED,
1491 	PR_OPA_SUPPORTED,
1492 	PR_IB_SUPPORTED
1493 };
1494 
1495 /*
1496  * opa_pr_query_possible - Check if current PR query can be an OPA query.
1497  *
1498  * Returns PR_NOT_SUPPORTED if a path record query is not
1499  * possible, PR_OPA_SUPPORTED if an OPA path record query
1500  * is possible and PR_IB_SUPPORTED if an IB path record
1501  * query is possible.
1502  */
1503 static int opa_pr_query_possible(struct ib_sa_client *client,
1504 				 struct ib_sa_device *sa_dev,
1505 				 struct ib_device *device, u32 port_num)
1506 {
1507 	struct ib_port_attr port_attr;
1508 
1509 	if (ib_query_port(device, port_num, &port_attr))
1510 		return PR_NOT_SUPPORTED;
1511 
1512 	if (ib_sa_opa_pathrecord_support(client, sa_dev, port_num))
1513 		return PR_OPA_SUPPORTED;
1514 
1515 	if (port_attr.lid >= be16_to_cpu(IB_MULTICAST_LID_BASE))
1516 		return PR_NOT_SUPPORTED;
1517 	else
1518 		return PR_IB_SUPPORTED;
1519 }
1520 
1521 static void ib_sa_path_rec_callback(struct ib_sa_query *sa_query,
1522 				    int status, struct ib_sa_mad *mad)
1523 {
1524 	struct ib_sa_path_query *query =
1525 		container_of(sa_query, struct ib_sa_path_query, sa_query);
1526 	struct sa_path_rec rec = {};
1527 
1528 	if (!mad) {
1529 		query->callback(status, NULL, 0, query->context);
1530 		return;
1531 	}
1532 
1533 	if (sa_query->flags & IB_SA_QUERY_OPA) {
1534 		ib_unpack(opa_path_rec_table, ARRAY_SIZE(opa_path_rec_table),
1535 			  mad->data, &rec);
1536 		rec.rec_type = SA_PATH_REC_TYPE_OPA;
1537 		query->callback(status, &rec, 1, query->context);
1538 		return;
1539 	}
1540 
1541 	ib_unpack(path_rec_table, ARRAY_SIZE(path_rec_table),
1542 		  mad->data, &rec);
1543 	rec.rec_type = SA_PATH_REC_TYPE_IB;
1544 	sa_path_set_dmac_zero(&rec);
1545 
1546 	if (query->conv_pr) {
1547 		struct sa_path_rec opa;
1548 
1549 		memset(&opa, 0, sizeof(struct sa_path_rec));
1550 		sa_convert_path_ib_to_opa(&opa, &rec);
1551 		query->callback(status, &opa, 1, query->context);
1552 	} else {
1553 		query->callback(status, &rec, 1, query->context);
1554 	}
1555 }
1556 
1557 #define IB_SA_DATA_OFFS 56
1558 #define IB_SERVICE_REC_SZ 176
1559 
1560 static void ib_unpack_service_rmpp(struct sa_service_rec *rec,
1561 				   struct ib_mad_recv_wc *mad_wc,
1562 				   int num_services)
1563 {
1564 	unsigned int cp_sz, data_i, data_size, rec_i = 0, buf_i = 0;
1565 	struct ib_mad_recv_buf *mad_buf;
1566 	u8 buf[IB_SERVICE_REC_SZ];
1567 	u8 *data;
1568 
1569 	data_size = sizeof(((struct ib_sa_mad *) mad_buf->mad)->data);
1570 
1571 	list_for_each_entry(mad_buf, &mad_wc->rmpp_list, list) {
1572 		data = ((struct ib_sa_mad *) mad_buf->mad)->data;
1573 		data_i = 0;
1574 		while (data_i < data_size && rec_i < num_services) {
1575 			cp_sz = min(IB_SERVICE_REC_SZ - buf_i,
1576 				    data_size - data_i);
1577 			memcpy(buf + buf_i, data + data_i, cp_sz);
1578 			data_i += cp_sz;
1579 			buf_i += cp_sz;
1580 			if (buf_i == IB_SERVICE_REC_SZ) {
1581 				ib_sa_unpack_service(buf, rec + rec_i);
1582 				buf_i = 0;
1583 				rec_i++;
1584 			}
1585 		}
1586 	}
1587 }
1588 
1589 static void ib_sa_service_rec_callback(struct ib_sa_query *sa_query, int status,
1590 				       struct ib_mad_recv_wc *mad_wc)
1591 {
1592 	struct ib_sa_service_query *query =
1593 		container_of(sa_query, struct ib_sa_service_query, sa_query);
1594 	struct sa_service_rec *rec;
1595 	int num_services;
1596 
1597 	if (!mad_wc || !mad_wc->recv_buf.mad) {
1598 		query->callback(status, NULL, 0, query->context);
1599 		return;
1600 	}
1601 
1602 	num_services = (mad_wc->mad_len - IB_SA_DATA_OFFS) / IB_SERVICE_REC_SZ;
1603 	if (!num_services) {
1604 		query->callback(-ENODATA, NULL, 0, query->context);
1605 		return;
1606 	}
1607 
1608 	rec = kmalloc_array(num_services, sizeof(*rec), GFP_KERNEL);
1609 	if (!rec) {
1610 		query->callback(-ENOMEM, NULL, 0, query->context);
1611 		return;
1612 	}
1613 
1614 	ib_unpack_service_rmpp(rec, mad_wc, num_services);
1615 	query->callback(status, rec, num_services, query->context);
1616 	kfree(rec);
1617 }
1618 
1619 static void ib_sa_path_rec_release(struct ib_sa_query *sa_query)
1620 {
1621 	struct ib_sa_path_query *query =
1622 		container_of(sa_query, struct ib_sa_path_query, sa_query);
1623 
1624 	kfree(query->conv_pr);
1625 	kfree(query);
1626 }
1627 
1628 static void ib_sa_service_rec_release(struct ib_sa_query *sa_query)
1629 {
1630 	struct ib_sa_service_query *query =
1631 		container_of(sa_query, struct ib_sa_service_query, sa_query);
1632 
1633 	kfree(query);
1634 }
1635 
1636 /**
1637  * ib_sa_path_rec_get - Start a Path get query
1638  * @client:SA client
1639  * @device:device to send query on
1640  * @port_num: port number to send query on
1641  * @rec:Path Record to send in query
1642  * @comp_mask:component mask to send in query
1643  * @timeout_ms:time to wait for response
1644  * @gfp_mask:GFP mask to use for internal allocations
1645  * @callback:function called when query completes, times out or is
1646  * canceled
1647  * @context:opaque user context passed to callback
1648  * @sa_query:query context, used to cancel query
1649  *
1650  * Send a Path Record Get query to the SA to look up a path.  The
1651  * callback function will be called when the query completes (or
1652  * fails); status is 0 for a successful response, -EINTR if the query
1653  * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
1654  * occurred sending the query.  The resp parameter of the callback is
1655  * only valid if status is 0.
1656  *
1657  * If the return value of ib_sa_path_rec_get() is negative, it is an
1658  * error code.  Otherwise it is a query ID that can be used to cancel
1659  * the query.
1660  */
1661 int ib_sa_path_rec_get(struct ib_sa_client *client,
1662 		       struct ib_device *device, u32 port_num,
1663 		       struct sa_path_rec *rec,
1664 		       ib_sa_comp_mask comp_mask,
1665 		       unsigned long timeout_ms, gfp_t gfp_mask,
1666 		       void (*callback)(int status,
1667 					struct sa_path_rec *resp,
1668 					unsigned int num_paths, void *context),
1669 		       void *context,
1670 		       struct ib_sa_query **sa_query)
1671 {
1672 	struct ib_sa_path_query *query;
1673 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1674 	struct ib_sa_port   *port;
1675 	struct ib_mad_agent *agent;
1676 	struct ib_sa_mad *mad;
1677 	enum opa_pr_supported status;
1678 	int ret;
1679 
1680 	if (!sa_dev)
1681 		return -ENODEV;
1682 
1683 	if ((rec->rec_type != SA_PATH_REC_TYPE_IB) &&
1684 	    (rec->rec_type != SA_PATH_REC_TYPE_OPA))
1685 		return -EINVAL;
1686 
1687 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1688 	agent = port->agent;
1689 
1690 	query = kzalloc(sizeof(*query), gfp_mask);
1691 	if (!query)
1692 		return -ENOMEM;
1693 
1694 	query->sa_query.port     = port;
1695 	if (rec->rec_type == SA_PATH_REC_TYPE_OPA) {
1696 		status = opa_pr_query_possible(client, sa_dev, device, port_num);
1697 		if (status == PR_NOT_SUPPORTED) {
1698 			ret = -EINVAL;
1699 			goto err1;
1700 		} else if (status == PR_OPA_SUPPORTED) {
1701 			query->sa_query.flags |= IB_SA_QUERY_OPA;
1702 		} else {
1703 			query->conv_pr =
1704 				kmalloc(sizeof(*query->conv_pr), gfp_mask);
1705 			if (!query->conv_pr) {
1706 				ret = -ENOMEM;
1707 				goto err1;
1708 			}
1709 		}
1710 	}
1711 
1712 	ret = alloc_mad(&query->sa_query, gfp_mask);
1713 	if (ret)
1714 		goto err2;
1715 
1716 	ib_sa_client_get(client);
1717 	query->sa_query.client = client;
1718 	query->callback        = callback;
1719 	query->context         = context;
1720 
1721 	mad = query->sa_query.mad_buf->mad;
1722 	init_mad(&query->sa_query, agent);
1723 
1724 	query->sa_query.callback = callback ? ib_sa_path_rec_callback : NULL;
1725 	query->sa_query.release  = ib_sa_path_rec_release;
1726 	mad->mad_hdr.method	 = IB_MGMT_METHOD_GET;
1727 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_PATH_REC);
1728 	mad->sa_hdr.comp_mask	 = comp_mask;
1729 
1730 	if (query->sa_query.flags & IB_SA_QUERY_OPA) {
1731 		ib_pack(opa_path_rec_table, ARRAY_SIZE(opa_path_rec_table),
1732 			rec, mad->data);
1733 	} else if (query->conv_pr) {
1734 		sa_convert_path_opa_to_ib(query->conv_pr, rec);
1735 		ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table),
1736 			query->conv_pr, mad->data);
1737 	} else {
1738 		ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table),
1739 			rec, mad->data);
1740 	}
1741 
1742 	*sa_query = &query->sa_query;
1743 
1744 	query->sa_query.flags |= IB_SA_ENABLE_LOCAL_SERVICE;
1745 	query->sa_query.mad_buf->context[1] = (query->conv_pr) ?
1746 						query->conv_pr : rec;
1747 
1748 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1749 	if (ret < 0)
1750 		goto err3;
1751 
1752 	return ret;
1753 
1754 err3:
1755 	*sa_query = NULL;
1756 	ib_sa_client_put(query->sa_query.client);
1757 	free_mad(&query->sa_query);
1758 err2:
1759 	kfree(query->conv_pr);
1760 err1:
1761 	kfree(query);
1762 	return ret;
1763 }
1764 EXPORT_SYMBOL(ib_sa_path_rec_get);
1765 
1766 /**
1767  * ib_sa_service_rec_get - Start a Service get query
1768  * @client: SA client
1769  * @device: device to send query on
1770  * @port_num: port number to send query on
1771  * @rec: Service Record to send in query
1772  * @comp_mask: component mask to send in query
1773  * @timeout_ms: time to wait for response
1774  * @gfp_mask: GFP mask to use for internal allocations
1775  * @callback: function called when query completes, times out or is
1776  * canceled
1777  * @context: opaque user context passed to callback
1778  * @sa_query: query context, used to cancel query
1779  *
1780  * Send a Service Record Get query to the SA to look up a path.  The
1781  * callback function will be called when the query completes (or
1782  * fails); status is 0 for a successful response, -EINTR if the query
1783  * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
1784  * occurred sending the query.  The resp parameter of the callback is
1785  * only valid if status is 0.
1786  *
1787  * If the return value of ib_sa_service_rec_get() is negative, it is an
1788  * error code. Otherwise it is a query ID that can be used to cancel
1789  * the query.
1790  */
1791 int ib_sa_service_rec_get(struct ib_sa_client *client,
1792 			  struct ib_device *device, u32 port_num,
1793 			  struct sa_service_rec *rec,
1794 			  ib_sa_comp_mask comp_mask,
1795 			  unsigned long timeout_ms, gfp_t gfp_mask,
1796 			  void (*callback)(int status,
1797 					   struct sa_service_rec *resp,
1798 					   unsigned int num_services,
1799 					   void *context),
1800 			  void *context, struct ib_sa_query **sa_query)
1801 {
1802 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1803 	struct ib_sa_service_query *query;
1804 	struct ib_mad_agent *agent;
1805 	struct ib_sa_port   *port;
1806 	struct ib_sa_mad *mad;
1807 	int ret;
1808 
1809 	if (!sa_dev)
1810 		return -ENODEV;
1811 
1812 	port = &sa_dev->port[port_num - sa_dev->start_port];
1813 	agent = port->agent;
1814 
1815 	query = kzalloc(sizeof(*query), gfp_mask);
1816 	if (!query)
1817 		return -ENOMEM;
1818 
1819 	query->sa_query.port = port;
1820 
1821 	ret = alloc_mad(&query->sa_query, gfp_mask);
1822 	if (ret)
1823 		goto err1;
1824 
1825 	ib_sa_client_get(client);
1826 	query->sa_query.client = client;
1827 	query->callback        = callback;
1828 	query->context         = context;
1829 
1830 	mad = query->sa_query.mad_buf->mad;
1831 	init_mad(&query->sa_query, agent);
1832 
1833 	query->sa_query.rmpp_callback = callback ? ib_sa_service_rec_callback :
1834 		NULL;
1835 	query->sa_query.release = ib_sa_service_rec_release;
1836 	mad->mad_hdr.method	= IB_MGMT_METHOD_GET_TABLE;
1837 	mad->mad_hdr.attr_id	= cpu_to_be16(IB_SA_ATTR_SERVICE_REC);
1838 	mad->sa_hdr.comp_mask	= comp_mask;
1839 
1840 	ib_sa_pack_service(rec, mad->data);
1841 
1842 	*sa_query = &query->sa_query;
1843 	query->sa_query.mad_buf->context[1] = rec;
1844 
1845 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1846 	if (ret < 0)
1847 		goto err2;
1848 
1849 	return ret;
1850 
1851 err2:
1852 	*sa_query = NULL;
1853 	ib_sa_client_put(query->sa_query.client);
1854 	free_mad(&query->sa_query);
1855 err1:
1856 	kfree(query);
1857 	return ret;
1858 }
1859 EXPORT_SYMBOL(ib_sa_service_rec_get);
1860 
1861 static void ib_sa_mcmember_rec_callback(struct ib_sa_query *sa_query,
1862 					int status, struct ib_sa_mad *mad)
1863 {
1864 	struct ib_sa_mcmember_query *query =
1865 		container_of(sa_query, struct ib_sa_mcmember_query, sa_query);
1866 
1867 	if (mad) {
1868 		struct ib_sa_mcmember_rec rec;
1869 
1870 		ib_unpack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
1871 			  mad->data, &rec);
1872 		query->callback(status, &rec, query->context);
1873 	} else
1874 		query->callback(status, NULL, query->context);
1875 }
1876 
1877 static void ib_sa_mcmember_rec_release(struct ib_sa_query *sa_query)
1878 {
1879 	kfree(container_of(sa_query, struct ib_sa_mcmember_query, sa_query));
1880 }
1881 
1882 int ib_sa_mcmember_rec_query(struct ib_sa_client *client,
1883 			     struct ib_device *device, u32 port_num,
1884 			     u8 method,
1885 			     struct ib_sa_mcmember_rec *rec,
1886 			     ib_sa_comp_mask comp_mask,
1887 			     unsigned long timeout_ms, gfp_t gfp_mask,
1888 			     void (*callback)(int status,
1889 					      struct ib_sa_mcmember_rec *resp,
1890 					      void *context),
1891 			     void *context,
1892 			     struct ib_sa_query **sa_query)
1893 {
1894 	struct ib_sa_mcmember_query *query;
1895 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1896 	struct ib_sa_port   *port;
1897 	struct ib_mad_agent *agent;
1898 	struct ib_sa_mad *mad;
1899 	int ret;
1900 
1901 	if (!sa_dev)
1902 		return -ENODEV;
1903 
1904 	port  = &sa_dev->port[port_num - sa_dev->start_port];
1905 	agent = port->agent;
1906 
1907 	query = kzalloc(sizeof(*query), gfp_mask);
1908 	if (!query)
1909 		return -ENOMEM;
1910 
1911 	query->sa_query.port     = port;
1912 	ret = alloc_mad(&query->sa_query, gfp_mask);
1913 	if (ret)
1914 		goto err1;
1915 
1916 	ib_sa_client_get(client);
1917 	query->sa_query.client = client;
1918 	query->callback        = callback;
1919 	query->context         = context;
1920 
1921 	mad = query->sa_query.mad_buf->mad;
1922 	init_mad(&query->sa_query, agent);
1923 
1924 	query->sa_query.callback = callback ? ib_sa_mcmember_rec_callback : NULL;
1925 	query->sa_query.release  = ib_sa_mcmember_rec_release;
1926 	mad->mad_hdr.method	 = method;
1927 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_MC_MEMBER_REC);
1928 	mad->sa_hdr.comp_mask	 = comp_mask;
1929 
1930 	ib_pack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
1931 		rec, mad->data);
1932 
1933 	*sa_query = &query->sa_query;
1934 
1935 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1936 	if (ret < 0)
1937 		goto err2;
1938 
1939 	return ret;
1940 
1941 err2:
1942 	*sa_query = NULL;
1943 	ib_sa_client_put(query->sa_query.client);
1944 	free_mad(&query->sa_query);
1945 
1946 err1:
1947 	kfree(query);
1948 	return ret;
1949 }
1950 
1951 /* Support GuidInfoRecord */
1952 static void ib_sa_guidinfo_rec_callback(struct ib_sa_query *sa_query,
1953 					int status, struct ib_sa_mad *mad)
1954 {
1955 	struct ib_sa_guidinfo_query *query =
1956 		container_of(sa_query, struct ib_sa_guidinfo_query, sa_query);
1957 
1958 	if (mad) {
1959 		struct ib_sa_guidinfo_rec rec;
1960 
1961 		ib_unpack(guidinfo_rec_table, ARRAY_SIZE(guidinfo_rec_table),
1962 			  mad->data, &rec);
1963 		query->callback(status, &rec, query->context);
1964 	} else
1965 		query->callback(status, NULL, query->context);
1966 }
1967 
1968 static void ib_sa_guidinfo_rec_release(struct ib_sa_query *sa_query)
1969 {
1970 	kfree(container_of(sa_query, struct ib_sa_guidinfo_query, sa_query));
1971 }
1972 
1973 int ib_sa_guid_info_rec_query(struct ib_sa_client *client,
1974 			      struct ib_device *device, u32 port_num,
1975 			      struct ib_sa_guidinfo_rec *rec,
1976 			      ib_sa_comp_mask comp_mask, u8 method,
1977 			      unsigned long timeout_ms, gfp_t gfp_mask,
1978 			      void (*callback)(int status,
1979 					       struct ib_sa_guidinfo_rec *resp,
1980 					       void *context),
1981 			      void *context,
1982 			      struct ib_sa_query **sa_query)
1983 {
1984 	struct ib_sa_guidinfo_query *query;
1985 	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1986 	struct ib_sa_port *port;
1987 	struct ib_mad_agent *agent;
1988 	struct ib_sa_mad *mad;
1989 	int ret;
1990 
1991 	if (!sa_dev)
1992 		return -ENODEV;
1993 
1994 	if (method != IB_MGMT_METHOD_GET &&
1995 	    method != IB_MGMT_METHOD_SET &&
1996 	    method != IB_SA_METHOD_DELETE) {
1997 		return -EINVAL;
1998 	}
1999 
2000 	port  = &sa_dev->port[port_num - sa_dev->start_port];
2001 	agent = port->agent;
2002 
2003 	query = kzalloc(sizeof(*query), gfp_mask);
2004 	if (!query)
2005 		return -ENOMEM;
2006 
2007 	query->sa_query.port = port;
2008 	ret = alloc_mad(&query->sa_query, gfp_mask);
2009 	if (ret)
2010 		goto err1;
2011 
2012 	ib_sa_client_get(client);
2013 	query->sa_query.client = client;
2014 	query->callback        = callback;
2015 	query->context         = context;
2016 
2017 	mad = query->sa_query.mad_buf->mad;
2018 	init_mad(&query->sa_query, agent);
2019 
2020 	query->sa_query.callback = callback ? ib_sa_guidinfo_rec_callback : NULL;
2021 	query->sa_query.release  = ib_sa_guidinfo_rec_release;
2022 
2023 	mad->mad_hdr.method	 = method;
2024 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_GUID_INFO_REC);
2025 	mad->sa_hdr.comp_mask	 = comp_mask;
2026 
2027 	ib_pack(guidinfo_rec_table, ARRAY_SIZE(guidinfo_rec_table), rec,
2028 		mad->data);
2029 
2030 	*sa_query = &query->sa_query;
2031 
2032 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
2033 	if (ret < 0)
2034 		goto err2;
2035 
2036 	return ret;
2037 
2038 err2:
2039 	*sa_query = NULL;
2040 	ib_sa_client_put(query->sa_query.client);
2041 	free_mad(&query->sa_query);
2042 
2043 err1:
2044 	kfree(query);
2045 	return ret;
2046 }
2047 EXPORT_SYMBOL(ib_sa_guid_info_rec_query);
2048 
2049 struct ib_classport_info_context {
2050 	struct completion	done;
2051 	struct ib_sa_query	*sa_query;
2052 };
2053 
2054 static void ib_classportinfo_cb(void *context)
2055 {
2056 	struct ib_classport_info_context *cb_ctx = context;
2057 
2058 	complete(&cb_ctx->done);
2059 }
2060 
2061 static void ib_sa_classport_info_rec_callback(struct ib_sa_query *sa_query,
2062 					      int status, struct ib_sa_mad *mad)
2063 {
2064 	unsigned long flags;
2065 	struct ib_sa_classport_info_query *query =
2066 		container_of(sa_query, struct ib_sa_classport_info_query, sa_query);
2067 	struct ib_sa_classport_cache *info = &sa_query->port->classport_info;
2068 
2069 	if (mad) {
2070 		if (sa_query->flags & IB_SA_QUERY_OPA) {
2071 			struct opa_class_port_info rec;
2072 
2073 			ib_unpack(opa_classport_info_rec_table,
2074 				  ARRAY_SIZE(opa_classport_info_rec_table),
2075 				  mad->data, &rec);
2076 
2077 			spin_lock_irqsave(&sa_query->port->classport_lock,
2078 					  flags);
2079 			if (!status && !info->valid) {
2080 				memcpy(&info->data.opa, &rec,
2081 				       sizeof(info->data.opa));
2082 
2083 				info->valid = true;
2084 				info->data.type = RDMA_CLASS_PORT_INFO_OPA;
2085 			}
2086 			spin_unlock_irqrestore(&sa_query->port->classport_lock,
2087 					       flags);
2088 
2089 		} else {
2090 			struct ib_class_port_info rec;
2091 
2092 			ib_unpack(ib_classport_info_rec_table,
2093 				  ARRAY_SIZE(ib_classport_info_rec_table),
2094 				  mad->data, &rec);
2095 
2096 			spin_lock_irqsave(&sa_query->port->classport_lock,
2097 					  flags);
2098 			if (!status && !info->valid) {
2099 				memcpy(&info->data.ib, &rec,
2100 				       sizeof(info->data.ib));
2101 
2102 				info->valid = true;
2103 				info->data.type = RDMA_CLASS_PORT_INFO_IB;
2104 			}
2105 			spin_unlock_irqrestore(&sa_query->port->classport_lock,
2106 					       flags);
2107 		}
2108 	}
2109 	query->callback(query->context);
2110 }
2111 
2112 static void ib_sa_classport_info_rec_release(struct ib_sa_query *sa_query)
2113 {
2114 	kfree(container_of(sa_query, struct ib_sa_classport_info_query,
2115 			   sa_query));
2116 }
2117 
2118 static int ib_sa_classport_info_rec_query(struct ib_sa_port *port,
2119 					  unsigned long timeout_ms,
2120 					  void (*callback)(void *context),
2121 					  void *context,
2122 					  struct ib_sa_query **sa_query)
2123 {
2124 	struct ib_mad_agent *agent;
2125 	struct ib_sa_classport_info_query *query;
2126 	struct ib_sa_mad *mad;
2127 	gfp_t gfp_mask = GFP_KERNEL;
2128 	int ret;
2129 
2130 	agent = port->agent;
2131 
2132 	query = kzalloc(sizeof(*query), gfp_mask);
2133 	if (!query)
2134 		return -ENOMEM;
2135 
2136 	query->sa_query.port = port;
2137 	query->sa_query.flags |= rdma_cap_opa_ah(port->agent->device,
2138 						 port->port_num) ?
2139 				 IB_SA_QUERY_OPA : 0;
2140 	ret = alloc_mad(&query->sa_query, gfp_mask);
2141 	if (ret)
2142 		goto err_free;
2143 
2144 	query->callback = callback;
2145 	query->context = context;
2146 
2147 	mad = query->sa_query.mad_buf->mad;
2148 	init_mad(&query->sa_query, agent);
2149 
2150 	query->sa_query.callback = ib_sa_classport_info_rec_callback;
2151 	query->sa_query.release  = ib_sa_classport_info_rec_release;
2152 	mad->mad_hdr.method	 = IB_MGMT_METHOD_GET;
2153 	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_CLASS_PORTINFO);
2154 	mad->sa_hdr.comp_mask	 = 0;
2155 	*sa_query = &query->sa_query;
2156 
2157 	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
2158 	if (ret < 0)
2159 		goto err_free_mad;
2160 
2161 	return ret;
2162 
2163 err_free_mad:
2164 	*sa_query = NULL;
2165 	free_mad(&query->sa_query);
2166 
2167 err_free:
2168 	kfree(query);
2169 	return ret;
2170 }
2171 
2172 static void update_ib_cpi(struct work_struct *work)
2173 {
2174 	struct ib_sa_port *port =
2175 		container_of(work, struct ib_sa_port, ib_cpi_work.work);
2176 	struct ib_classport_info_context *cb_context;
2177 	unsigned long flags;
2178 	int ret;
2179 
2180 	/* If the classport info is valid, nothing
2181 	 * to do here.
2182 	 */
2183 	spin_lock_irqsave(&port->classport_lock, flags);
2184 	if (port->classport_info.valid) {
2185 		spin_unlock_irqrestore(&port->classport_lock, flags);
2186 		return;
2187 	}
2188 	spin_unlock_irqrestore(&port->classport_lock, flags);
2189 
2190 	cb_context = kmalloc(sizeof(*cb_context), GFP_KERNEL);
2191 	if (!cb_context)
2192 		goto err_nomem;
2193 
2194 	init_completion(&cb_context->done);
2195 
2196 	ret = ib_sa_classport_info_rec_query(port, 3000,
2197 					     ib_classportinfo_cb, cb_context,
2198 					     &cb_context->sa_query);
2199 	if (ret < 0)
2200 		goto free_cb_err;
2201 	wait_for_completion(&cb_context->done);
2202 free_cb_err:
2203 	kfree(cb_context);
2204 	spin_lock_irqsave(&port->classport_lock, flags);
2205 
2206 	/* If the classport info is still not valid, the query should have
2207 	 * failed for some reason. Retry issuing the query
2208 	 */
2209 	if (!port->classport_info.valid) {
2210 		port->classport_info.retry_cnt++;
2211 		if (port->classport_info.retry_cnt <=
2212 		    IB_SA_CPI_MAX_RETRY_CNT) {
2213 			unsigned long delay =
2214 				msecs_to_jiffies(IB_SA_CPI_RETRY_WAIT);
2215 
2216 			queue_delayed_work(ib_wq, &port->ib_cpi_work, delay);
2217 		}
2218 	}
2219 	spin_unlock_irqrestore(&port->classport_lock, flags);
2220 
2221 err_nomem:
2222 	return;
2223 }
2224 
2225 static void send_handler(struct ib_mad_agent *agent,
2226 			 struct ib_mad_send_wc *mad_send_wc)
2227 {
2228 	struct ib_sa_query *query = mad_send_wc->send_buf->context[0];
2229 	unsigned long flags;
2230 	int status = 0;
2231 
2232 	if (query->callback || query->rmpp_callback) {
2233 		switch (mad_send_wc->status) {
2234 		case IB_WC_SUCCESS:
2235 			/* No callback -- already got recv */
2236 			break;
2237 		case IB_WC_RESP_TIMEOUT_ERR:
2238 			status = -ETIMEDOUT;
2239 			break;
2240 		case IB_WC_WR_FLUSH_ERR:
2241 			status = -EINTR;
2242 			break;
2243 		default:
2244 			status = -EIO;
2245 			break;
2246 		}
2247 
2248 		if (status)
2249 			query->callback ? query->callback(query, status, NULL) :
2250 				query->rmpp_callback(query, status, NULL);
2251 	}
2252 
2253 	xa_lock_irqsave(&queries, flags);
2254 	__xa_erase(&queries, query->id);
2255 	xa_unlock_irqrestore(&queries, flags);
2256 
2257 	free_mad(query);
2258 	if (query->client)
2259 		ib_sa_client_put(query->client);
2260 	query->release(query);
2261 }
2262 
2263 static void recv_handler(struct ib_mad_agent *mad_agent,
2264 			 struct ib_mad_send_buf *send_buf,
2265 			 struct ib_mad_recv_wc *mad_recv_wc)
2266 {
2267 	struct ib_sa_query *query;
2268 	struct ib_mad *mad;
2269 
2270 
2271 	if (!send_buf)
2272 		return;
2273 
2274 	query = send_buf->context[0];
2275 	mad = mad_recv_wc->recv_buf.mad;
2276 
2277 	if (query->rmpp_callback) {
2278 		if (mad_recv_wc->wc->status == IB_WC_SUCCESS)
2279 			query->rmpp_callback(query, mad->mad_hdr.status ?
2280 					     -EINVAL : 0, mad_recv_wc);
2281 		else
2282 			query->rmpp_callback(query, -EIO, NULL);
2283 	} else if (query->callback) {
2284 		if (mad_recv_wc->wc->status == IB_WC_SUCCESS)
2285 			query->callback(query, mad->mad_hdr.status ?
2286 					-EINVAL : 0, (struct ib_sa_mad *)mad);
2287 		else
2288 			query->callback(query, -EIO, NULL);
2289 	}
2290 
2291 	ib_free_recv_mad(mad_recv_wc);
2292 }
2293 
2294 static void update_sm_ah(struct work_struct *work)
2295 {
2296 	struct ib_sa_port *port =
2297 		container_of(work, struct ib_sa_port, update_task);
2298 	struct ib_sa_sm_ah *new_ah;
2299 	struct ib_port_attr port_attr;
2300 	struct rdma_ah_attr   ah_attr;
2301 	bool grh_required;
2302 
2303 	if (ib_query_port(port->agent->device, port->port_num, &port_attr)) {
2304 		pr_warn("Couldn't query port\n");
2305 		return;
2306 	}
2307 
2308 	new_ah = kmalloc(sizeof(*new_ah), GFP_KERNEL);
2309 	if (!new_ah)
2310 		return;
2311 
2312 	kref_init(&new_ah->ref);
2313 	new_ah->src_path_mask = (1 << port_attr.lmc) - 1;
2314 
2315 	new_ah->pkey_index = 0;
2316 	if (ib_find_pkey(port->agent->device, port->port_num,
2317 			 IB_DEFAULT_PKEY_FULL, &new_ah->pkey_index))
2318 		pr_err("Couldn't find index for default PKey\n");
2319 
2320 	memset(&ah_attr, 0, sizeof(ah_attr));
2321 	ah_attr.type = rdma_ah_find_type(port->agent->device,
2322 					 port->port_num);
2323 	rdma_ah_set_dlid(&ah_attr, port_attr.sm_lid);
2324 	rdma_ah_set_sl(&ah_attr, port_attr.sm_sl);
2325 	rdma_ah_set_port_num(&ah_attr, port->port_num);
2326 
2327 	grh_required = rdma_is_grh_required(port->agent->device,
2328 					    port->port_num);
2329 
2330 	/*
2331 	 * The OPA sm_lid of 0xFFFF needs special handling so that it can be
2332 	 * differentiated from a permissive LID of 0xFFFF.  We set the
2333 	 * grh_required flag here so the SA can program the DGID in the
2334 	 * address handle appropriately
2335 	 */
2336 	if (ah_attr.type == RDMA_AH_ATTR_TYPE_OPA &&
2337 	    (grh_required ||
2338 	     port_attr.sm_lid == be16_to_cpu(IB_LID_PERMISSIVE)))
2339 		rdma_ah_set_make_grd(&ah_attr, true);
2340 
2341 	if (ah_attr.type == RDMA_AH_ATTR_TYPE_IB && grh_required) {
2342 		rdma_ah_set_ah_flags(&ah_attr, IB_AH_GRH);
2343 		rdma_ah_set_subnet_prefix(&ah_attr,
2344 					  cpu_to_be64(port_attr.subnet_prefix));
2345 		rdma_ah_set_interface_id(&ah_attr,
2346 					 cpu_to_be64(IB_SA_WELL_KNOWN_GUID));
2347 	}
2348 
2349 	new_ah->ah = rdma_create_ah(port->agent->qp->pd, &ah_attr,
2350 				    RDMA_CREATE_AH_SLEEPABLE);
2351 	if (IS_ERR(new_ah->ah)) {
2352 		pr_warn("Couldn't create new SM AH\n");
2353 		kfree(new_ah);
2354 		return;
2355 	}
2356 
2357 	spin_lock_irq(&port->ah_lock);
2358 	if (port->sm_ah)
2359 		kref_put(&port->sm_ah->ref, free_sm_ah);
2360 	port->sm_ah = new_ah;
2361 	spin_unlock_irq(&port->ah_lock);
2362 }
2363 
2364 static void ib_sa_event(struct ib_event_handler *handler,
2365 			struct ib_event *event)
2366 {
2367 	if (event->event == IB_EVENT_PORT_ERR    ||
2368 	    event->event == IB_EVENT_PORT_ACTIVE ||
2369 	    event->event == IB_EVENT_LID_CHANGE  ||
2370 	    event->event == IB_EVENT_PKEY_CHANGE ||
2371 	    event->event == IB_EVENT_SM_CHANGE   ||
2372 	    event->event == IB_EVENT_CLIENT_REREGISTER) {
2373 		unsigned long flags;
2374 		struct ib_sa_device *sa_dev =
2375 			container_of(handler, typeof(*sa_dev), event_handler);
2376 		u32 port_num = event->element.port_num - sa_dev->start_port;
2377 		struct ib_sa_port *port = &sa_dev->port[port_num];
2378 
2379 		if (!rdma_cap_ib_sa(handler->device, port->port_num))
2380 			return;
2381 
2382 		spin_lock_irqsave(&port->ah_lock, flags);
2383 		if (port->sm_ah)
2384 			kref_put(&port->sm_ah->ref, free_sm_ah);
2385 		port->sm_ah = NULL;
2386 		spin_unlock_irqrestore(&port->ah_lock, flags);
2387 
2388 		if (event->event == IB_EVENT_SM_CHANGE ||
2389 		    event->event == IB_EVENT_CLIENT_REREGISTER ||
2390 		    event->event == IB_EVENT_LID_CHANGE ||
2391 		    event->event == IB_EVENT_PORT_ACTIVE) {
2392 			unsigned long delay =
2393 				msecs_to_jiffies(IB_SA_CPI_RETRY_WAIT);
2394 
2395 			spin_lock_irqsave(&port->classport_lock, flags);
2396 			port->classport_info.valid = false;
2397 			port->classport_info.retry_cnt = 0;
2398 			spin_unlock_irqrestore(&port->classport_lock, flags);
2399 			queue_delayed_work(ib_wq,
2400 					   &port->ib_cpi_work, delay);
2401 		}
2402 		queue_work(ib_wq, &sa_dev->port[port_num].update_task);
2403 	}
2404 }
2405 
2406 static int ib_sa_add_one(struct ib_device *device)
2407 {
2408 	struct ib_sa_device *sa_dev;
2409 	int s, e, i;
2410 	int count = 0;
2411 	int ret;
2412 
2413 	s = rdma_start_port(device);
2414 	e = rdma_end_port(device);
2415 
2416 	sa_dev = kzalloc(struct_size(sa_dev, port,
2417 				     size_add(size_sub(e, s), 1)),
2418 			 GFP_KERNEL);
2419 	if (!sa_dev)
2420 		return -ENOMEM;
2421 
2422 	sa_dev->start_port = s;
2423 	sa_dev->end_port   = e;
2424 
2425 	for (i = 0; i <= e - s; ++i) {
2426 		spin_lock_init(&sa_dev->port[i].ah_lock);
2427 		if (!rdma_cap_ib_sa(device, i + 1))
2428 			continue;
2429 
2430 		sa_dev->port[i].sm_ah    = NULL;
2431 		sa_dev->port[i].port_num = i + s;
2432 
2433 		spin_lock_init(&sa_dev->port[i].classport_lock);
2434 		sa_dev->port[i].classport_info.valid = false;
2435 
2436 		sa_dev->port[i].agent =
2437 			ib_register_mad_agent(device, i + s, IB_QPT_GSI,
2438 					      NULL, IB_MGMT_RMPP_VERSION,
2439 					      send_handler, recv_handler,
2440 					      sa_dev, 0);
2441 		if (IS_ERR(sa_dev->port[i].agent)) {
2442 			ret = PTR_ERR(sa_dev->port[i].agent);
2443 			goto err;
2444 		}
2445 
2446 		INIT_WORK(&sa_dev->port[i].update_task, update_sm_ah);
2447 		INIT_DELAYED_WORK(&sa_dev->port[i].ib_cpi_work,
2448 				  update_ib_cpi);
2449 
2450 		count++;
2451 	}
2452 
2453 	if (!count) {
2454 		ret = -EOPNOTSUPP;
2455 		goto free;
2456 	}
2457 
2458 	ib_set_client_data(device, &sa_client, sa_dev);
2459 
2460 	/*
2461 	 * We register our event handler after everything is set up,
2462 	 * and then update our cached info after the event handler is
2463 	 * registered to avoid any problems if a port changes state
2464 	 * during our initialization.
2465 	 */
2466 
2467 	INIT_IB_EVENT_HANDLER(&sa_dev->event_handler, device, ib_sa_event);
2468 	ib_register_event_handler(&sa_dev->event_handler);
2469 
2470 	for (i = 0; i <= e - s; ++i) {
2471 		if (rdma_cap_ib_sa(device, i + 1))
2472 			update_sm_ah(&sa_dev->port[i].update_task);
2473 	}
2474 
2475 	return 0;
2476 
2477 err:
2478 	while (--i >= 0) {
2479 		if (rdma_cap_ib_sa(device, i + 1))
2480 			ib_unregister_mad_agent(sa_dev->port[i].agent);
2481 	}
2482 free:
2483 	kfree(sa_dev);
2484 	return ret;
2485 }
2486 
2487 static void ib_sa_remove_one(struct ib_device *device, void *client_data)
2488 {
2489 	struct ib_sa_device *sa_dev = client_data;
2490 	int i;
2491 
2492 	ib_unregister_event_handler(&sa_dev->event_handler);
2493 	flush_workqueue(ib_wq);
2494 
2495 	for (i = 0; i <= sa_dev->end_port - sa_dev->start_port; ++i) {
2496 		if (rdma_cap_ib_sa(device, i + 1)) {
2497 			cancel_delayed_work_sync(&sa_dev->port[i].ib_cpi_work);
2498 			ib_unregister_mad_agent(sa_dev->port[i].agent);
2499 			if (sa_dev->port[i].sm_ah)
2500 				kref_put(&sa_dev->port[i].sm_ah->ref, free_sm_ah);
2501 		}
2502 
2503 	}
2504 
2505 	kfree(sa_dev);
2506 }
2507 
2508 int ib_sa_init(void)
2509 {
2510 	int ret;
2511 
2512 	get_random_bytes(&tid, sizeof tid);
2513 
2514 	atomic_set(&ib_nl_sa_request_seq, 0);
2515 
2516 	ret = ib_register_client(&sa_client);
2517 	if (ret) {
2518 		pr_err("Couldn't register ib_sa client\n");
2519 		goto err1;
2520 	}
2521 
2522 	ret = mcast_init();
2523 	if (ret) {
2524 		pr_err("Couldn't initialize multicast handling\n");
2525 		goto err2;
2526 	}
2527 
2528 	ib_nl_wq = alloc_ordered_workqueue("ib_nl_sa_wq", WQ_MEM_RECLAIM);
2529 	if (!ib_nl_wq) {
2530 		ret = -ENOMEM;
2531 		goto err3;
2532 	}
2533 
2534 	INIT_DELAYED_WORK(&ib_nl_timed_work, ib_nl_request_timeout);
2535 
2536 	return 0;
2537 
2538 err3:
2539 	mcast_cleanup();
2540 err2:
2541 	ib_unregister_client(&sa_client);
2542 err1:
2543 	return ret;
2544 }
2545 
2546 void ib_sa_cleanup(void)
2547 {
2548 	cancel_delayed_work(&ib_nl_timed_work);
2549 	destroy_workqueue(ib_nl_wq);
2550 	mcast_cleanup();
2551 	ib_unregister_client(&sa_client);
2552 	WARN_ON(!xa_empty(&queries));
2553 }
2554