IB/SA: Add support to query opa classport info.
[linux-2.6-block.git] / drivers / infiniband / core / sa_query.c
1 /*
2  * Copyright (c) 2004 Topspin Communications.  All rights reserved.
3  * Copyright (c) 2005 Voltaire, Inc.  All rights reserved.
4  * Copyright (c) 2006 Intel Corporation.  All rights reserved.
5  *
6  * This software is available to you under a choice of one of two
7  * licenses.  You may choose to be licensed under the terms of the GNU
8  * General Public License (GPL) Version 2, available from the file
9  * COPYING in the main directory of this source tree, or the
10  * OpenIB.org BSD license below:
11  *
12  *     Redistribution and use in source and binary forms, with or
13  *     without modification, are permitted provided that the following
14  *     conditions are met:
15  *
16  *      - Redistributions of source code must retain the above
17  *        copyright notice, this list of conditions and the following
18  *        disclaimer.
19  *
20  *      - Redistributions in binary form must reproduce the above
21  *        copyright notice, this list of conditions and the following
22  *        disclaimer in the documentation and/or other materials
23  *        provided with the distribution.
24  *
25  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
26  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
27  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
28  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
29  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
30  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
31  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
32  * SOFTWARE.
33  */
34
35 #include <linux/module.h>
36 #include <linux/init.h>
37 #include <linux/err.h>
38 #include <linux/random.h>
39 #include <linux/spinlock.h>
40 #include <linux/slab.h>
41 #include <linux/dma-mapping.h>
42 #include <linux/kref.h>
43 #include <linux/idr.h>
44 #include <linux/workqueue.h>
45 #include <uapi/linux/if_ether.h>
46 #include <rdma/ib_pack.h>
47 #include <rdma/ib_cache.h>
48 #include <rdma/rdma_netlink.h>
49 #include <net/netlink.h>
50 #include <uapi/rdma/ib_user_sa.h>
51 #include <rdma/ib_marshall.h>
52 #include <rdma/ib_addr.h>
53 #include "sa.h"
54 #include "core_priv.h"
55
56 #define IB_SA_LOCAL_SVC_TIMEOUT_MIN             100
57 #define IB_SA_LOCAL_SVC_TIMEOUT_DEFAULT         2000
58 #define IB_SA_LOCAL_SVC_TIMEOUT_MAX             200000
59 #define IB_SA_CPI_MAX_RETRY_CNT                 3
60 #define IB_SA_CPI_RETRY_WAIT                    1000 /*msecs */
61 static int sa_local_svc_timeout_ms = IB_SA_LOCAL_SVC_TIMEOUT_DEFAULT;
62
63 struct ib_sa_sm_ah {
64         struct ib_ah        *ah;
65         struct kref          ref;
66         u16                  pkey_index;
67         u8                   src_path_mask;
68 };
69
70 enum rdma_class_port_info_type {
71         RDMA_CLASS_PORT_INFO_IB,
72         RDMA_CLASS_PORT_INFO_OPA
73 };
74
75 struct rdma_class_port_info {
76         enum rdma_class_port_info_type type;
77         union {
78                 struct ib_class_port_info ib;
79                 struct opa_class_port_info opa;
80         };
81 };
82
83 struct ib_sa_classport_cache {
84         bool valid;
85         int retry_cnt;
86         struct rdma_class_port_info data;
87 };
88
89 struct ib_sa_port {
90         struct ib_mad_agent *agent;
91         struct ib_sa_sm_ah  *sm_ah;
92         struct work_struct   update_task;
93         struct ib_sa_classport_cache classport_info;
94         struct delayed_work ib_cpi_work;
95         spinlock_t                   classport_lock; /* protects class port info set */
96         spinlock_t           ah_lock;
97         u8                   port_num;
98 };
99
100 struct ib_sa_device {
101         int                     start_port, end_port;
102         struct ib_event_handler event_handler;
103         struct ib_sa_port port[0];
104 };
105
106 struct ib_sa_query {
107         void (*callback)(struct ib_sa_query *, int, struct ib_sa_mad *);
108         void (*release)(struct ib_sa_query *);
109         struct ib_sa_client    *client;
110         struct ib_sa_port      *port;
111         struct ib_mad_send_buf *mad_buf;
112         struct ib_sa_sm_ah     *sm_ah;
113         int                     id;
114         u32                     flags;
115         struct list_head        list; /* Local svc request list */
116         u32                     seq; /* Local svc request sequence number */
117         unsigned long           timeout; /* Local svc timeout */
118         u8                      path_use; /* How will the pathrecord be used */
119 };
120
121 #define IB_SA_ENABLE_LOCAL_SERVICE      0x00000001
122 #define IB_SA_CANCEL                    0x00000002
123 #define IB_SA_QUERY_OPA                 0x00000004
124
125 struct ib_sa_service_query {
126         void (*callback)(int, struct ib_sa_service_rec *, void *);
127         void *context;
128         struct ib_sa_query sa_query;
129 };
130
131 struct ib_sa_path_query {
132         void (*callback)(int, struct ib_sa_path_rec *, void *);
133         void *context;
134         struct ib_sa_query sa_query;
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 static LIST_HEAD(ib_nl_request_list);
156 static DEFINE_SPINLOCK(ib_nl_request_lock);
157 static atomic_t ib_nl_sa_request_seq;
158 static struct workqueue_struct *ib_nl_wq;
159 static struct delayed_work ib_nl_timed_work;
160 static const struct nla_policy ib_nl_policy[LS_NLA_TYPE_MAX] = {
161         [LS_NLA_TYPE_PATH_RECORD]       = {.type = NLA_BINARY,
162                 .len = sizeof(struct ib_path_rec_data)},
163         [LS_NLA_TYPE_TIMEOUT]           = {.type = NLA_U32},
164         [LS_NLA_TYPE_SERVICE_ID]        = {.type = NLA_U64},
165         [LS_NLA_TYPE_DGID]              = {.type = NLA_BINARY,
166                 .len = sizeof(struct rdma_nla_ls_gid)},
167         [LS_NLA_TYPE_SGID]              = {.type = NLA_BINARY,
168                 .len = sizeof(struct rdma_nla_ls_gid)},
169         [LS_NLA_TYPE_TCLASS]            = {.type = NLA_U8},
170         [LS_NLA_TYPE_PKEY]              = {.type = NLA_U16},
171         [LS_NLA_TYPE_QOS_CLASS]         = {.type = NLA_U16},
172 };
173
174
175 static void ib_sa_add_one(struct ib_device *device);
176 static void ib_sa_remove_one(struct ib_device *device, void *client_data);
177
178 static struct ib_client sa_client = {
179         .name   = "sa",
180         .add    = ib_sa_add_one,
181         .remove = ib_sa_remove_one
182 };
183
184 static DEFINE_SPINLOCK(idr_lock);
185 static DEFINE_IDR(query_idr);
186
187 static DEFINE_SPINLOCK(tid_lock);
188 static u32 tid;
189
190 #define PATH_REC_FIELD(field) \
191         .struct_offset_bytes = offsetof(struct ib_sa_path_rec, field),          \
192         .struct_size_bytes   = sizeof ((struct ib_sa_path_rec *) 0)->field,     \
193         .field_name          = "sa_path_rec:" #field
194
195 static const struct ib_field path_rec_table[] = {
196         { PATH_REC_FIELD(service_id),
197           .offset_words = 0,
198           .offset_bits  = 0,
199           .size_bits    = 64 },
200         { PATH_REC_FIELD(dgid),
201           .offset_words = 2,
202           .offset_bits  = 0,
203           .size_bits    = 128 },
204         { PATH_REC_FIELD(sgid),
205           .offset_words = 6,
206           .offset_bits  = 0,
207           .size_bits    = 128 },
208         { PATH_REC_FIELD(dlid),
209           .offset_words = 10,
210           .offset_bits  = 0,
211           .size_bits    = 16 },
212         { PATH_REC_FIELD(slid),
213           .offset_words = 10,
214           .offset_bits  = 16,
215           .size_bits    = 16 },
216         { PATH_REC_FIELD(raw_traffic),
217           .offset_words = 11,
218           .offset_bits  = 0,
219           .size_bits    = 1 },
220         { RESERVED,
221           .offset_words = 11,
222           .offset_bits  = 1,
223           .size_bits    = 3 },
224         { PATH_REC_FIELD(flow_label),
225           .offset_words = 11,
226           .offset_bits  = 4,
227           .size_bits    = 20 },
228         { PATH_REC_FIELD(hop_limit),
229           .offset_words = 11,
230           .offset_bits  = 24,
231           .size_bits    = 8 },
232         { PATH_REC_FIELD(traffic_class),
233           .offset_words = 12,
234           .offset_bits  = 0,
235           .size_bits    = 8 },
236         { PATH_REC_FIELD(reversible),
237           .offset_words = 12,
238           .offset_bits  = 8,
239           .size_bits    = 1 },
240         { PATH_REC_FIELD(numb_path),
241           .offset_words = 12,
242           .offset_bits  = 9,
243           .size_bits    = 7 },
244         { PATH_REC_FIELD(pkey),
245           .offset_words = 12,
246           .offset_bits  = 16,
247           .size_bits    = 16 },
248         { PATH_REC_FIELD(qos_class),
249           .offset_words = 13,
250           .offset_bits  = 0,
251           .size_bits    = 12 },
252         { PATH_REC_FIELD(sl),
253           .offset_words = 13,
254           .offset_bits  = 12,
255           .size_bits    = 4 },
256         { PATH_REC_FIELD(mtu_selector),
257           .offset_words = 13,
258           .offset_bits  = 16,
259           .size_bits    = 2 },
260         { PATH_REC_FIELD(mtu),
261           .offset_words = 13,
262           .offset_bits  = 18,
263           .size_bits    = 6 },
264         { PATH_REC_FIELD(rate_selector),
265           .offset_words = 13,
266           .offset_bits  = 24,
267           .size_bits    = 2 },
268         { PATH_REC_FIELD(rate),
269           .offset_words = 13,
270           .offset_bits  = 26,
271           .size_bits    = 6 },
272         { PATH_REC_FIELD(packet_life_time_selector),
273           .offset_words = 14,
274           .offset_bits  = 0,
275           .size_bits    = 2 },
276         { PATH_REC_FIELD(packet_life_time),
277           .offset_words = 14,
278           .offset_bits  = 2,
279           .size_bits    = 6 },
280         { PATH_REC_FIELD(preference),
281           .offset_words = 14,
282           .offset_bits  = 8,
283           .size_bits    = 8 },
284         { RESERVED,
285           .offset_words = 14,
286           .offset_bits  = 16,
287           .size_bits    = 48 },
288 };
289
290 #define MCMEMBER_REC_FIELD(field) \
291         .struct_offset_bytes = offsetof(struct ib_sa_mcmember_rec, field),      \
292         .struct_size_bytes   = sizeof ((struct ib_sa_mcmember_rec *) 0)->field, \
293         .field_name          = "sa_mcmember_rec:" #field
294
295 static const struct ib_field mcmember_rec_table[] = {
296         { MCMEMBER_REC_FIELD(mgid),
297           .offset_words = 0,
298           .offset_bits  = 0,
299           .size_bits    = 128 },
300         { MCMEMBER_REC_FIELD(port_gid),
301           .offset_words = 4,
302           .offset_bits  = 0,
303           .size_bits    = 128 },
304         { MCMEMBER_REC_FIELD(qkey),
305           .offset_words = 8,
306           .offset_bits  = 0,
307           .size_bits    = 32 },
308         { MCMEMBER_REC_FIELD(mlid),
309           .offset_words = 9,
310           .offset_bits  = 0,
311           .size_bits    = 16 },
312         { MCMEMBER_REC_FIELD(mtu_selector),
313           .offset_words = 9,
314           .offset_bits  = 16,
315           .size_bits    = 2 },
316         { MCMEMBER_REC_FIELD(mtu),
317           .offset_words = 9,
318           .offset_bits  = 18,
319           .size_bits    = 6 },
320         { MCMEMBER_REC_FIELD(traffic_class),
321           .offset_words = 9,
322           .offset_bits  = 24,
323           .size_bits    = 8 },
324         { MCMEMBER_REC_FIELD(pkey),
325           .offset_words = 10,
326           .offset_bits  = 0,
327           .size_bits    = 16 },
328         { MCMEMBER_REC_FIELD(rate_selector),
329           .offset_words = 10,
330           .offset_bits  = 16,
331           .size_bits    = 2 },
332         { MCMEMBER_REC_FIELD(rate),
333           .offset_words = 10,
334           .offset_bits  = 18,
335           .size_bits    = 6 },
336         { MCMEMBER_REC_FIELD(packet_life_time_selector),
337           .offset_words = 10,
338           .offset_bits  = 24,
339           .size_bits    = 2 },
340         { MCMEMBER_REC_FIELD(packet_life_time),
341           .offset_words = 10,
342           .offset_bits  = 26,
343           .size_bits    = 6 },
344         { MCMEMBER_REC_FIELD(sl),
345           .offset_words = 11,
346           .offset_bits  = 0,
347           .size_bits    = 4 },
348         { MCMEMBER_REC_FIELD(flow_label),
349           .offset_words = 11,
350           .offset_bits  = 4,
351           .size_bits    = 20 },
352         { MCMEMBER_REC_FIELD(hop_limit),
353           .offset_words = 11,
354           .offset_bits  = 24,
355           .size_bits    = 8 },
356         { MCMEMBER_REC_FIELD(scope),
357           .offset_words = 12,
358           .offset_bits  = 0,
359           .size_bits    = 4 },
360         { MCMEMBER_REC_FIELD(join_state),
361           .offset_words = 12,
362           .offset_bits  = 4,
363           .size_bits    = 4 },
364         { MCMEMBER_REC_FIELD(proxy_join),
365           .offset_words = 12,
366           .offset_bits  = 8,
367           .size_bits    = 1 },
368         { RESERVED,
369           .offset_words = 12,
370           .offset_bits  = 9,
371           .size_bits    = 23 },
372 };
373
374 #define SERVICE_REC_FIELD(field) \
375         .struct_offset_bytes = offsetof(struct ib_sa_service_rec, field),       \
376         .struct_size_bytes   = sizeof ((struct ib_sa_service_rec *) 0)->field,  \
377         .field_name          = "sa_service_rec:" #field
378
379 static const struct ib_field service_rec_table[] = {
380         { SERVICE_REC_FIELD(id),
381           .offset_words = 0,
382           .offset_bits  = 0,
383           .size_bits    = 64 },
384         { SERVICE_REC_FIELD(gid),
385           .offset_words = 2,
386           .offset_bits  = 0,
387           .size_bits    = 128 },
388         { SERVICE_REC_FIELD(pkey),
389           .offset_words = 6,
390           .offset_bits  = 0,
391           .size_bits    = 16 },
392         { SERVICE_REC_FIELD(lease),
393           .offset_words = 7,
394           .offset_bits  = 0,
395           .size_bits    = 32 },
396         { SERVICE_REC_FIELD(key),
397           .offset_words = 8,
398           .offset_bits  = 0,
399           .size_bits    = 128 },
400         { SERVICE_REC_FIELD(name),
401           .offset_words = 12,
402           .offset_bits  = 0,
403           .size_bits    = 64*8 },
404         { SERVICE_REC_FIELD(data8),
405           .offset_words = 28,
406           .offset_bits  = 0,
407           .size_bits    = 16*8 },
408         { SERVICE_REC_FIELD(data16),
409           .offset_words = 32,
410           .offset_bits  = 0,
411           .size_bits    = 8*16 },
412         { SERVICE_REC_FIELD(data32),
413           .offset_words = 36,
414           .offset_bits  = 0,
415           .size_bits    = 4*32 },
416         { SERVICE_REC_FIELD(data64),
417           .offset_words = 40,
418           .offset_bits  = 0,
419           .size_bits    = 2*64 },
420 };
421
422 #define CLASSPORTINFO_REC_FIELD(field) \
423         .struct_offset_bytes = offsetof(struct ib_class_port_info, field),      \
424         .struct_size_bytes   = sizeof((struct ib_class_port_info *)0)->field,   \
425         .field_name          = "ib_class_port_info:" #field
426
427 static const struct ib_field ib_classport_info_rec_table[] = {
428         { CLASSPORTINFO_REC_FIELD(base_version),
429           .offset_words = 0,
430           .offset_bits  = 0,
431           .size_bits    = 8 },
432         { CLASSPORTINFO_REC_FIELD(class_version),
433           .offset_words = 0,
434           .offset_bits  = 8,
435           .size_bits    = 8 },
436         { CLASSPORTINFO_REC_FIELD(capability_mask),
437           .offset_words = 0,
438           .offset_bits  = 16,
439           .size_bits    = 16 },
440         { CLASSPORTINFO_REC_FIELD(cap_mask2_resp_time),
441           .offset_words = 1,
442           .offset_bits  = 0,
443           .size_bits    = 32 },
444         { CLASSPORTINFO_REC_FIELD(redirect_gid),
445           .offset_words = 2,
446           .offset_bits  = 0,
447           .size_bits    = 128 },
448         { CLASSPORTINFO_REC_FIELD(redirect_tcslfl),
449           .offset_words = 6,
450           .offset_bits  = 0,
451           .size_bits    = 32 },
452         { CLASSPORTINFO_REC_FIELD(redirect_lid),
453           .offset_words = 7,
454           .offset_bits  = 0,
455           .size_bits    = 16 },
456         { CLASSPORTINFO_REC_FIELD(redirect_pkey),
457           .offset_words = 7,
458           .offset_bits  = 16,
459           .size_bits    = 16 },
460
461         { CLASSPORTINFO_REC_FIELD(redirect_qp),
462           .offset_words = 8,
463           .offset_bits  = 0,
464           .size_bits    = 32 },
465         { CLASSPORTINFO_REC_FIELD(redirect_qkey),
466           .offset_words = 9,
467           .offset_bits  = 0,
468           .size_bits    = 32 },
469
470         { CLASSPORTINFO_REC_FIELD(trap_gid),
471           .offset_words = 10,
472           .offset_bits  = 0,
473           .size_bits    = 128 },
474         { CLASSPORTINFO_REC_FIELD(trap_tcslfl),
475           .offset_words = 14,
476           .offset_bits  = 0,
477           .size_bits    = 32 },
478
479         { CLASSPORTINFO_REC_FIELD(trap_lid),
480           .offset_words = 15,
481           .offset_bits  = 0,
482           .size_bits    = 16 },
483         { CLASSPORTINFO_REC_FIELD(trap_pkey),
484           .offset_words = 15,
485           .offset_bits  = 16,
486           .size_bits    = 16 },
487
488         { CLASSPORTINFO_REC_FIELD(trap_hlqp),
489           .offset_words = 16,
490           .offset_bits  = 0,
491           .size_bits    = 32 },
492         { CLASSPORTINFO_REC_FIELD(trap_qkey),
493           .offset_words = 17,
494           .offset_bits  = 0,
495           .size_bits    = 32 },
496 };
497
498 #define OPA_CLASSPORTINFO_REC_FIELD(field) \
499         .struct_offset_bytes =\
500                 offsetof(struct opa_class_port_info, field),    \
501         .struct_size_bytes   = \
502                 sizeof((struct opa_class_port_info *)0)->field, \
503         .field_name          = "opa_class_port_info:" #field
504
505 static const struct ib_field opa_classport_info_rec_table[] = {
506         { OPA_CLASSPORTINFO_REC_FIELD(base_version),
507           .offset_words = 0,
508           .offset_bits  = 0,
509           .size_bits    = 8 },
510         { OPA_CLASSPORTINFO_REC_FIELD(class_version),
511           .offset_words = 0,
512           .offset_bits  = 8,
513           .size_bits    = 8 },
514         { OPA_CLASSPORTINFO_REC_FIELD(cap_mask),
515           .offset_words = 0,
516           .offset_bits  = 16,
517           .size_bits    = 16 },
518         { OPA_CLASSPORTINFO_REC_FIELD(cap_mask2_resp_time),
519           .offset_words = 1,
520           .offset_bits  = 0,
521           .size_bits    = 32 },
522         { OPA_CLASSPORTINFO_REC_FIELD(redirect_gid),
523           .offset_words = 2,
524           .offset_bits  = 0,
525           .size_bits    = 128 },
526         { OPA_CLASSPORTINFO_REC_FIELD(redirect_tc_fl),
527           .offset_words = 6,
528           .offset_bits  = 0,
529           .size_bits    = 32 },
530         { OPA_CLASSPORTINFO_REC_FIELD(redirect_lid),
531           .offset_words = 7,
532           .offset_bits  = 0,
533           .size_bits    = 32 },
534         { OPA_CLASSPORTINFO_REC_FIELD(redirect_sl_qp),
535           .offset_words = 8,
536           .offset_bits  = 0,
537           .size_bits    = 32 },
538         { OPA_CLASSPORTINFO_REC_FIELD(redirect_qkey),
539           .offset_words = 9,
540           .offset_bits  = 0,
541           .size_bits    = 32 },
542         { OPA_CLASSPORTINFO_REC_FIELD(trap_gid),
543           .offset_words = 10,
544           .offset_bits  = 0,
545           .size_bits    = 128 },
546         { OPA_CLASSPORTINFO_REC_FIELD(trap_tc_fl),
547           .offset_words = 14,
548           .offset_bits  = 0,
549           .size_bits    = 32 },
550         { OPA_CLASSPORTINFO_REC_FIELD(trap_lid),
551           .offset_words = 15,
552           .offset_bits  = 0,
553           .size_bits    = 32 },
554         { OPA_CLASSPORTINFO_REC_FIELD(trap_hl_qp),
555           .offset_words = 16,
556           .offset_bits  = 0,
557           .size_bits    = 32 },
558         { OPA_CLASSPORTINFO_REC_FIELD(trap_qkey),
559           .offset_words = 17,
560           .offset_bits  = 0,
561           .size_bits    = 32 },
562         { OPA_CLASSPORTINFO_REC_FIELD(trap_pkey),
563           .offset_words = 18,
564           .offset_bits  = 0,
565           .size_bits    = 16 },
566         { OPA_CLASSPORTINFO_REC_FIELD(redirect_pkey),
567           .offset_words = 18,
568           .offset_bits  = 16,
569           .size_bits    = 16 },
570         { OPA_CLASSPORTINFO_REC_FIELD(trap_sl_rsvd),
571           .offset_words = 19,
572           .offset_bits  = 0,
573           .size_bits    = 8 },
574         { RESERVED,
575           .offset_words = 19,
576           .offset_bits  = 8,
577           .size_bits    = 24 },
578 };
579
580 #define GUIDINFO_REC_FIELD(field) \
581         .struct_offset_bytes = offsetof(struct ib_sa_guidinfo_rec, field),      \
582         .struct_size_bytes   = sizeof((struct ib_sa_guidinfo_rec *) 0)->field,  \
583         .field_name          = "sa_guidinfo_rec:" #field
584
585 static const struct ib_field guidinfo_rec_table[] = {
586         { GUIDINFO_REC_FIELD(lid),
587           .offset_words = 0,
588           .offset_bits  = 0,
589           .size_bits    = 16 },
590         { GUIDINFO_REC_FIELD(block_num),
591           .offset_words = 0,
592           .offset_bits  = 16,
593           .size_bits    = 8 },
594         { GUIDINFO_REC_FIELD(res1),
595           .offset_words = 0,
596           .offset_bits  = 24,
597           .size_bits    = 8 },
598         { GUIDINFO_REC_FIELD(res2),
599           .offset_words = 1,
600           .offset_bits  = 0,
601           .size_bits    = 32 },
602         { GUIDINFO_REC_FIELD(guid_info_list),
603           .offset_words = 2,
604           .offset_bits  = 0,
605           .size_bits    = 512 },
606 };
607
608 static inline void ib_sa_disable_local_svc(struct ib_sa_query *query)
609 {
610         query->flags &= ~IB_SA_ENABLE_LOCAL_SERVICE;
611 }
612
613 static inline int ib_sa_query_cancelled(struct ib_sa_query *query)
614 {
615         return (query->flags & IB_SA_CANCEL);
616 }
617
618 static void ib_nl_set_path_rec_attrs(struct sk_buff *skb,
619                                      struct ib_sa_query *query)
620 {
621         struct ib_sa_path_rec *sa_rec = query->mad_buf->context[1];
622         struct ib_sa_mad *mad = query->mad_buf->mad;
623         ib_sa_comp_mask comp_mask = mad->sa_hdr.comp_mask;
624         u16 val16;
625         u64 val64;
626         struct rdma_ls_resolve_header *header;
627
628         query->mad_buf->context[1] = NULL;
629
630         /* Construct the family header first */
631         header = (struct rdma_ls_resolve_header *)
632                 skb_put(skb, NLMSG_ALIGN(sizeof(*header)));
633         memcpy(header->device_name, query->port->agent->device->name,
634                LS_DEVICE_NAME_MAX);
635         header->port_num = query->port->port_num;
636
637         if ((comp_mask & IB_SA_PATH_REC_REVERSIBLE) &&
638             sa_rec->reversible != 0)
639                 query->path_use = LS_RESOLVE_PATH_USE_GMP;
640         else
641                 query->path_use = LS_RESOLVE_PATH_USE_UNIDIRECTIONAL;
642         header->path_use = query->path_use;
643
644         /* Now build the attributes */
645         if (comp_mask & IB_SA_PATH_REC_SERVICE_ID) {
646                 val64 = be64_to_cpu(sa_rec->service_id);
647                 nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_SERVICE_ID,
648                         sizeof(val64), &val64);
649         }
650         if (comp_mask & IB_SA_PATH_REC_DGID)
651                 nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_DGID,
652                         sizeof(sa_rec->dgid), &sa_rec->dgid);
653         if (comp_mask & IB_SA_PATH_REC_SGID)
654                 nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_SGID,
655                         sizeof(sa_rec->sgid), &sa_rec->sgid);
656         if (comp_mask & IB_SA_PATH_REC_TRAFFIC_CLASS)
657                 nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_TCLASS,
658                         sizeof(sa_rec->traffic_class), &sa_rec->traffic_class);
659
660         if (comp_mask & IB_SA_PATH_REC_PKEY) {
661                 val16 = be16_to_cpu(sa_rec->pkey);
662                 nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_PKEY,
663                         sizeof(val16), &val16);
664         }
665         if (comp_mask & IB_SA_PATH_REC_QOS_CLASS) {
666                 val16 = be16_to_cpu(sa_rec->qos_class);
667                 nla_put(skb, RDMA_NLA_F_MANDATORY | LS_NLA_TYPE_QOS_CLASS,
668                         sizeof(val16), &val16);
669         }
670 }
671
672 static int ib_nl_get_path_rec_attrs_len(ib_sa_comp_mask comp_mask)
673 {
674         int len = 0;
675
676         if (comp_mask & IB_SA_PATH_REC_SERVICE_ID)
677                 len += nla_total_size(sizeof(u64));
678         if (comp_mask & IB_SA_PATH_REC_DGID)
679                 len += nla_total_size(sizeof(struct rdma_nla_ls_gid));
680         if (comp_mask & IB_SA_PATH_REC_SGID)
681                 len += nla_total_size(sizeof(struct rdma_nla_ls_gid));
682         if (comp_mask & IB_SA_PATH_REC_TRAFFIC_CLASS)
683                 len += nla_total_size(sizeof(u8));
684         if (comp_mask & IB_SA_PATH_REC_PKEY)
685                 len += nla_total_size(sizeof(u16));
686         if (comp_mask & IB_SA_PATH_REC_QOS_CLASS)
687                 len += nla_total_size(sizeof(u16));
688
689         /*
690          * Make sure that at least some of the required comp_mask bits are
691          * set.
692          */
693         if (WARN_ON(len == 0))
694                 return len;
695
696         /* Add the family header */
697         len += NLMSG_ALIGN(sizeof(struct rdma_ls_resolve_header));
698
699         return len;
700 }
701
702 static int ib_nl_send_msg(struct ib_sa_query *query, gfp_t gfp_mask)
703 {
704         struct sk_buff *skb = NULL;
705         struct nlmsghdr *nlh;
706         void *data;
707         int ret = 0;
708         struct ib_sa_mad *mad;
709         int len;
710
711         mad = query->mad_buf->mad;
712         len = ib_nl_get_path_rec_attrs_len(mad->sa_hdr.comp_mask);
713         if (len <= 0)
714                 return -EMSGSIZE;
715
716         skb = nlmsg_new(len, gfp_mask);
717         if (!skb)
718                 return -ENOMEM;
719
720         /* Put nlmsg header only for now */
721         data = ibnl_put_msg(skb, &nlh, query->seq, 0, RDMA_NL_LS,
722                             RDMA_NL_LS_OP_RESOLVE, NLM_F_REQUEST);
723         if (!data) {
724                 nlmsg_free(skb);
725                 return -EMSGSIZE;
726         }
727
728         /* Add attributes */
729         ib_nl_set_path_rec_attrs(skb, query);
730
731         /* Repair the nlmsg header length */
732         nlmsg_end(skb, nlh);
733
734         ret = ibnl_multicast(skb, nlh, RDMA_NL_GROUP_LS, gfp_mask);
735         if (!ret)
736                 ret = len;
737         else
738                 ret = 0;
739
740         return ret;
741 }
742
743 static int ib_nl_make_request(struct ib_sa_query *query, gfp_t gfp_mask)
744 {
745         unsigned long flags;
746         unsigned long delay;
747         int ret;
748
749         INIT_LIST_HEAD(&query->list);
750         query->seq = (u32)atomic_inc_return(&ib_nl_sa_request_seq);
751
752         /* Put the request on the list first.*/
753         spin_lock_irqsave(&ib_nl_request_lock, flags);
754         delay = msecs_to_jiffies(sa_local_svc_timeout_ms);
755         query->timeout = delay + jiffies;
756         list_add_tail(&query->list, &ib_nl_request_list);
757         /* Start the timeout if this is the only request */
758         if (ib_nl_request_list.next == &query->list)
759                 queue_delayed_work(ib_nl_wq, &ib_nl_timed_work, delay);
760         spin_unlock_irqrestore(&ib_nl_request_lock, flags);
761
762         ret = ib_nl_send_msg(query, gfp_mask);
763         if (ret <= 0) {
764                 ret = -EIO;
765                 /* Remove the request */
766                 spin_lock_irqsave(&ib_nl_request_lock, flags);
767                 list_del(&query->list);
768                 spin_unlock_irqrestore(&ib_nl_request_lock, flags);
769         } else {
770                 ret = 0;
771         }
772
773         return ret;
774 }
775
776 static int ib_nl_cancel_request(struct ib_sa_query *query)
777 {
778         unsigned long flags;
779         struct ib_sa_query *wait_query;
780         int found = 0;
781
782         spin_lock_irqsave(&ib_nl_request_lock, flags);
783         list_for_each_entry(wait_query, &ib_nl_request_list, list) {
784                 /* Let the timeout to take care of the callback */
785                 if (query == wait_query) {
786                         query->flags |= IB_SA_CANCEL;
787                         query->timeout = jiffies;
788                         list_move(&query->list, &ib_nl_request_list);
789                         found = 1;
790                         mod_delayed_work(ib_nl_wq, &ib_nl_timed_work, 1);
791                         break;
792                 }
793         }
794         spin_unlock_irqrestore(&ib_nl_request_lock, flags);
795
796         return found;
797 }
798
799 static void send_handler(struct ib_mad_agent *agent,
800                          struct ib_mad_send_wc *mad_send_wc);
801
802 static void ib_nl_process_good_resolve_rsp(struct ib_sa_query *query,
803                                            const struct nlmsghdr *nlh)
804 {
805         struct ib_mad_send_wc mad_send_wc;
806         struct ib_sa_mad *mad = NULL;
807         const struct nlattr *head, *curr;
808         struct ib_path_rec_data  *rec;
809         int len, rem;
810         u32 mask = 0;
811         int status = -EIO;
812
813         if (query->callback) {
814                 head = (const struct nlattr *) nlmsg_data(nlh);
815                 len = nlmsg_len(nlh);
816                 switch (query->path_use) {
817                 case LS_RESOLVE_PATH_USE_UNIDIRECTIONAL:
818                         mask = IB_PATH_PRIMARY | IB_PATH_OUTBOUND;
819                         break;
820
821                 case LS_RESOLVE_PATH_USE_ALL:
822                 case LS_RESOLVE_PATH_USE_GMP:
823                 default:
824                         mask = IB_PATH_PRIMARY | IB_PATH_GMP |
825                                 IB_PATH_BIDIRECTIONAL;
826                         break;
827                 }
828                 nla_for_each_attr(curr, head, len, rem) {
829                         if (curr->nla_type == LS_NLA_TYPE_PATH_RECORD) {
830                                 rec = nla_data(curr);
831                                 /*
832                                  * Get the first one. In the future, we may
833                                  * need to get up to 6 pathrecords.
834                                  */
835                                 if ((rec->flags & mask) == mask) {
836                                         mad = query->mad_buf->mad;
837                                         mad->mad_hdr.method |=
838                                                 IB_MGMT_METHOD_RESP;
839                                         memcpy(mad->data, rec->path_rec,
840                                                sizeof(rec->path_rec));
841                                         status = 0;
842                                         break;
843                                 }
844                         }
845                 }
846                 query->callback(query, status, mad);
847         }
848
849         mad_send_wc.send_buf = query->mad_buf;
850         mad_send_wc.status = IB_WC_SUCCESS;
851         send_handler(query->mad_buf->mad_agent, &mad_send_wc);
852 }
853
854 static void ib_nl_request_timeout(struct work_struct *work)
855 {
856         unsigned long flags;
857         struct ib_sa_query *query;
858         unsigned long delay;
859         struct ib_mad_send_wc mad_send_wc;
860         int ret;
861
862         spin_lock_irqsave(&ib_nl_request_lock, flags);
863         while (!list_empty(&ib_nl_request_list)) {
864                 query = list_entry(ib_nl_request_list.next,
865                                    struct ib_sa_query, list);
866
867                 if (time_after(query->timeout, jiffies)) {
868                         delay = query->timeout - jiffies;
869                         if ((long)delay <= 0)
870                                 delay = 1;
871                         queue_delayed_work(ib_nl_wq, &ib_nl_timed_work, delay);
872                         break;
873                 }
874
875                 list_del(&query->list);
876                 ib_sa_disable_local_svc(query);
877                 /* Hold the lock to protect against query cancellation */
878                 if (ib_sa_query_cancelled(query))
879                         ret = -1;
880                 else
881                         ret = ib_post_send_mad(query->mad_buf, NULL);
882                 if (ret) {
883                         mad_send_wc.send_buf = query->mad_buf;
884                         mad_send_wc.status = IB_WC_WR_FLUSH_ERR;
885                         spin_unlock_irqrestore(&ib_nl_request_lock, flags);
886                         send_handler(query->port->agent, &mad_send_wc);
887                         spin_lock_irqsave(&ib_nl_request_lock, flags);
888                 }
889         }
890         spin_unlock_irqrestore(&ib_nl_request_lock, flags);
891 }
892
893 int ib_nl_handle_set_timeout(struct sk_buff *skb,
894                              struct netlink_callback *cb)
895 {
896         const struct nlmsghdr *nlh = (struct nlmsghdr *)cb->nlh;
897         int timeout, delta, abs_delta;
898         const struct nlattr *attr;
899         unsigned long flags;
900         struct ib_sa_query *query;
901         long delay = 0;
902         struct nlattr *tb[LS_NLA_TYPE_MAX];
903         int ret;
904
905         if (!(nlh->nlmsg_flags & NLM_F_REQUEST) ||
906             !(NETLINK_CB(skb).sk) ||
907             !netlink_capable(skb, CAP_NET_ADMIN))
908                 return -EPERM;
909
910         ret = nla_parse(tb, LS_NLA_TYPE_MAX - 1, nlmsg_data(nlh),
911                         nlmsg_len(nlh), ib_nl_policy, NULL);
912         attr = (const struct nlattr *)tb[LS_NLA_TYPE_TIMEOUT];
913         if (ret || !attr)
914                 goto settimeout_out;
915
916         timeout = *(int *) nla_data(attr);
917         if (timeout < IB_SA_LOCAL_SVC_TIMEOUT_MIN)
918                 timeout = IB_SA_LOCAL_SVC_TIMEOUT_MIN;
919         if (timeout > IB_SA_LOCAL_SVC_TIMEOUT_MAX)
920                 timeout = IB_SA_LOCAL_SVC_TIMEOUT_MAX;
921
922         delta = timeout - sa_local_svc_timeout_ms;
923         if (delta < 0)
924                 abs_delta = -delta;
925         else
926                 abs_delta = delta;
927
928         if (delta != 0) {
929                 spin_lock_irqsave(&ib_nl_request_lock, flags);
930                 sa_local_svc_timeout_ms = timeout;
931                 list_for_each_entry(query, &ib_nl_request_list, list) {
932                         if (delta < 0 && abs_delta > query->timeout)
933                                 query->timeout = 0;
934                         else
935                                 query->timeout += delta;
936
937                         /* Get the new delay from the first entry */
938                         if (!delay) {
939                                 delay = query->timeout - jiffies;
940                                 if (delay <= 0)
941                                         delay = 1;
942                         }
943                 }
944                 if (delay)
945                         mod_delayed_work(ib_nl_wq, &ib_nl_timed_work,
946                                          (unsigned long)delay);
947                 spin_unlock_irqrestore(&ib_nl_request_lock, flags);
948         }
949
950 settimeout_out:
951         return skb->len;
952 }
953
954 static inline int ib_nl_is_good_resolve_resp(const struct nlmsghdr *nlh)
955 {
956         struct nlattr *tb[LS_NLA_TYPE_MAX];
957         int ret;
958
959         if (nlh->nlmsg_flags & RDMA_NL_LS_F_ERR)
960                 return 0;
961
962         ret = nla_parse(tb, LS_NLA_TYPE_MAX - 1, nlmsg_data(nlh),
963                         nlmsg_len(nlh), ib_nl_policy, NULL);
964         if (ret)
965                 return 0;
966
967         return 1;
968 }
969
970 int ib_nl_handle_resolve_resp(struct sk_buff *skb,
971                               struct netlink_callback *cb)
972 {
973         const struct nlmsghdr *nlh = (struct nlmsghdr *)cb->nlh;
974         unsigned long flags;
975         struct ib_sa_query *query;
976         struct ib_mad_send_buf *send_buf;
977         struct ib_mad_send_wc mad_send_wc;
978         int found = 0;
979         int ret;
980
981         if ((nlh->nlmsg_flags & NLM_F_REQUEST) ||
982             !(NETLINK_CB(skb).sk) ||
983             !netlink_capable(skb, CAP_NET_ADMIN))
984                 return -EPERM;
985
986         spin_lock_irqsave(&ib_nl_request_lock, flags);
987         list_for_each_entry(query, &ib_nl_request_list, list) {
988                 /*
989                  * If the query is cancelled, let the timeout routine
990                  * take care of it.
991                  */
992                 if (nlh->nlmsg_seq == query->seq) {
993                         found = !ib_sa_query_cancelled(query);
994                         if (found)
995                                 list_del(&query->list);
996                         break;
997                 }
998         }
999
1000         if (!found) {
1001                 spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1002                 goto resp_out;
1003         }
1004
1005         send_buf = query->mad_buf;
1006
1007         if (!ib_nl_is_good_resolve_resp(nlh)) {
1008                 /* if the result is a failure, send out the packet via IB */
1009                 ib_sa_disable_local_svc(query);
1010                 ret = ib_post_send_mad(query->mad_buf, NULL);
1011                 spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1012                 if (ret) {
1013                         mad_send_wc.send_buf = send_buf;
1014                         mad_send_wc.status = IB_WC_GENERAL_ERR;
1015                         send_handler(query->port->agent, &mad_send_wc);
1016                 }
1017         } else {
1018                 spin_unlock_irqrestore(&ib_nl_request_lock, flags);
1019                 ib_nl_process_good_resolve_rsp(query, nlh);
1020         }
1021
1022 resp_out:
1023         return skb->len;
1024 }
1025
1026 static void free_sm_ah(struct kref *kref)
1027 {
1028         struct ib_sa_sm_ah *sm_ah = container_of(kref, struct ib_sa_sm_ah, ref);
1029
1030         ib_destroy_ah(sm_ah->ah);
1031         kfree(sm_ah);
1032 }
1033
1034 void ib_sa_register_client(struct ib_sa_client *client)
1035 {
1036         atomic_set(&client->users, 1);
1037         init_completion(&client->comp);
1038 }
1039 EXPORT_SYMBOL(ib_sa_register_client);
1040
1041 void ib_sa_unregister_client(struct ib_sa_client *client)
1042 {
1043         ib_sa_client_put(client);
1044         wait_for_completion(&client->comp);
1045 }
1046 EXPORT_SYMBOL(ib_sa_unregister_client);
1047
1048 /**
1049  * ib_sa_cancel_query - try to cancel an SA query
1050  * @id:ID of query to cancel
1051  * @query:query pointer to cancel
1052  *
1053  * Try to cancel an SA query.  If the id and query don't match up or
1054  * the query has already completed, nothing is done.  Otherwise the
1055  * query is canceled and will complete with a status of -EINTR.
1056  */
1057 void ib_sa_cancel_query(int id, struct ib_sa_query *query)
1058 {
1059         unsigned long flags;
1060         struct ib_mad_agent *agent;
1061         struct ib_mad_send_buf *mad_buf;
1062
1063         spin_lock_irqsave(&idr_lock, flags);
1064         if (idr_find(&query_idr, id) != query) {
1065                 spin_unlock_irqrestore(&idr_lock, flags);
1066                 return;
1067         }
1068         agent = query->port->agent;
1069         mad_buf = query->mad_buf;
1070         spin_unlock_irqrestore(&idr_lock, flags);
1071
1072         /*
1073          * If the query is still on the netlink request list, schedule
1074          * it to be cancelled by the timeout routine. Otherwise, it has been
1075          * sent to the MAD layer and has to be cancelled from there.
1076          */
1077         if (!ib_nl_cancel_request(query))
1078                 ib_cancel_mad(agent, mad_buf);
1079 }
1080 EXPORT_SYMBOL(ib_sa_cancel_query);
1081
1082 static u8 get_src_path_mask(struct ib_device *device, u8 port_num)
1083 {
1084         struct ib_sa_device *sa_dev;
1085         struct ib_sa_port   *port;
1086         unsigned long flags;
1087         u8 src_path_mask;
1088
1089         sa_dev = ib_get_client_data(device, &sa_client);
1090         if (!sa_dev)
1091                 return 0x7f;
1092
1093         port  = &sa_dev->port[port_num - sa_dev->start_port];
1094         spin_lock_irqsave(&port->ah_lock, flags);
1095         src_path_mask = port->sm_ah ? port->sm_ah->src_path_mask : 0x7f;
1096         spin_unlock_irqrestore(&port->ah_lock, flags);
1097
1098         return src_path_mask;
1099 }
1100
1101 int ib_init_ah_from_path(struct ib_device *device, u8 port_num,
1102                          struct ib_sa_path_rec *rec, struct ib_ah_attr *ah_attr)
1103 {
1104         int ret;
1105         u16 gid_index;
1106         int use_roce;
1107         struct net_device *ndev = NULL;
1108
1109         memset(ah_attr, 0, sizeof *ah_attr);
1110         ah_attr->dlid = be16_to_cpu(rec->dlid);
1111         ah_attr->sl = rec->sl;
1112         ah_attr->src_path_bits = be16_to_cpu(rec->slid) &
1113                                  get_src_path_mask(device, port_num);
1114         ah_attr->port_num = port_num;
1115         ah_attr->static_rate = rec->rate;
1116
1117         use_roce = rdma_cap_eth_ah(device, port_num);
1118
1119         if (use_roce) {
1120                 struct net_device *idev;
1121                 struct net_device *resolved_dev;
1122                 struct rdma_dev_addr dev_addr = {.bound_dev_if = rec->ifindex,
1123                                                  .net = rec->net ? rec->net :
1124                                                          &init_net};
1125                 union {
1126                         struct sockaddr     _sockaddr;
1127                         struct sockaddr_in  _sockaddr_in;
1128                         struct sockaddr_in6 _sockaddr_in6;
1129                 } sgid_addr, dgid_addr;
1130
1131                 if (!device->get_netdev)
1132                         return -EOPNOTSUPP;
1133
1134                 rdma_gid2ip(&sgid_addr._sockaddr, &rec->sgid);
1135                 rdma_gid2ip(&dgid_addr._sockaddr, &rec->dgid);
1136
1137                 /* validate the route */
1138                 ret = rdma_resolve_ip_route(&sgid_addr._sockaddr,
1139                                             &dgid_addr._sockaddr, &dev_addr);
1140                 if (ret)
1141                         return ret;
1142
1143                 if ((dev_addr.network == RDMA_NETWORK_IPV4 ||
1144                      dev_addr.network == RDMA_NETWORK_IPV6) &&
1145                     rec->gid_type != IB_GID_TYPE_ROCE_UDP_ENCAP)
1146                         return -EINVAL;
1147
1148                 idev = device->get_netdev(device, port_num);
1149                 if (!idev)
1150                         return -ENODEV;
1151
1152                 resolved_dev = dev_get_by_index(dev_addr.net,
1153                                                 dev_addr.bound_dev_if);
1154                 if (resolved_dev->flags & IFF_LOOPBACK) {
1155                         dev_put(resolved_dev);
1156                         resolved_dev = idev;
1157                         dev_hold(resolved_dev);
1158                 }
1159                 ndev = ib_get_ndev_from_path(rec);
1160                 rcu_read_lock();
1161                 if ((ndev && ndev != resolved_dev) ||
1162                     (resolved_dev != idev &&
1163                      !rdma_is_upper_dev_rcu(idev, resolved_dev)))
1164                         ret = -EHOSTUNREACH;
1165                 rcu_read_unlock();
1166                 dev_put(idev);
1167                 dev_put(resolved_dev);
1168                 if (ret) {
1169                         if (ndev)
1170                                 dev_put(ndev);
1171                         return ret;
1172                 }
1173         }
1174
1175         if (rec->hop_limit > 0 || use_roce) {
1176                 ah_attr->ah_flags = IB_AH_GRH;
1177                 ah_attr->grh.dgid = rec->dgid;
1178
1179                 ret = ib_find_cached_gid_by_port(device, &rec->sgid,
1180                                                  rec->gid_type, port_num, ndev,
1181                                                  &gid_index);
1182                 if (ret) {
1183                         if (ndev)
1184                                 dev_put(ndev);
1185                         return ret;
1186                 }
1187
1188                 ah_attr->grh.sgid_index    = gid_index;
1189                 ah_attr->grh.flow_label    = be32_to_cpu(rec->flow_label);
1190                 ah_attr->grh.hop_limit     = rec->hop_limit;
1191                 ah_attr->grh.traffic_class = rec->traffic_class;
1192                 if (ndev)
1193                         dev_put(ndev);
1194         }
1195
1196         if (use_roce)
1197                 memcpy(ah_attr->dmac, rec->dmac, ETH_ALEN);
1198
1199         return 0;
1200 }
1201 EXPORT_SYMBOL(ib_init_ah_from_path);
1202
1203 static int alloc_mad(struct ib_sa_query *query, gfp_t gfp_mask)
1204 {
1205         unsigned long flags;
1206
1207         spin_lock_irqsave(&query->port->ah_lock, flags);
1208         if (!query->port->sm_ah) {
1209                 spin_unlock_irqrestore(&query->port->ah_lock, flags);
1210                 return -EAGAIN;
1211         }
1212         kref_get(&query->port->sm_ah->ref);
1213         query->sm_ah = query->port->sm_ah;
1214         spin_unlock_irqrestore(&query->port->ah_lock, flags);
1215
1216         query->mad_buf = ib_create_send_mad(query->port->agent, 1,
1217                                             query->sm_ah->pkey_index,
1218                                             0, IB_MGMT_SA_HDR, IB_MGMT_SA_DATA,
1219                                             gfp_mask,
1220                                             ((query->flags & IB_SA_QUERY_OPA) ?
1221                                              OPA_MGMT_BASE_VERSION :
1222                                              IB_MGMT_BASE_VERSION));
1223         if (IS_ERR(query->mad_buf)) {
1224                 kref_put(&query->sm_ah->ref, free_sm_ah);
1225                 return -ENOMEM;
1226         }
1227
1228         query->mad_buf->ah = query->sm_ah->ah;
1229
1230         return 0;
1231 }
1232
1233 static void free_mad(struct ib_sa_query *query)
1234 {
1235         ib_free_send_mad(query->mad_buf);
1236         kref_put(&query->sm_ah->ref, free_sm_ah);
1237 }
1238
1239 static void init_mad(struct ib_sa_query *query, struct ib_mad_agent *agent)
1240 {
1241         struct ib_sa_mad *mad = query->mad_buf->mad;
1242         unsigned long flags;
1243
1244         memset(mad, 0, sizeof *mad);
1245
1246         if (query->flags & IB_SA_QUERY_OPA) {
1247                 mad->mad_hdr.base_version  = OPA_MGMT_BASE_VERSION;
1248                 mad->mad_hdr.class_version = OPA_SA_CLASS_VERSION;
1249         } else {
1250                 mad->mad_hdr.base_version  = IB_MGMT_BASE_VERSION;
1251                 mad->mad_hdr.class_version = IB_SA_CLASS_VERSION;
1252         }
1253         mad->mad_hdr.mgmt_class    = IB_MGMT_CLASS_SUBN_ADM;
1254         spin_lock_irqsave(&tid_lock, flags);
1255         mad->mad_hdr.tid           =
1256                 cpu_to_be64(((u64) agent->hi_tid) << 32 | tid++);
1257         spin_unlock_irqrestore(&tid_lock, flags);
1258 }
1259
1260 static int send_mad(struct ib_sa_query *query, int timeout_ms, gfp_t gfp_mask)
1261 {
1262         bool preload = gfpflags_allow_blocking(gfp_mask);
1263         unsigned long flags;
1264         int ret, id;
1265
1266         if (preload)
1267                 idr_preload(gfp_mask);
1268         spin_lock_irqsave(&idr_lock, flags);
1269
1270         id = idr_alloc(&query_idr, query, 0, 0, GFP_NOWAIT);
1271
1272         spin_unlock_irqrestore(&idr_lock, flags);
1273         if (preload)
1274                 idr_preload_end();
1275         if (id < 0)
1276                 return id;
1277
1278         query->mad_buf->timeout_ms  = timeout_ms;
1279         query->mad_buf->context[0] = query;
1280         query->id = id;
1281
1282         if (query->flags & IB_SA_ENABLE_LOCAL_SERVICE) {
1283                 if (!ibnl_chk_listeners(RDMA_NL_GROUP_LS)) {
1284                         if (!ib_nl_make_request(query, gfp_mask))
1285                                 return id;
1286                 }
1287                 ib_sa_disable_local_svc(query);
1288         }
1289
1290         ret = ib_post_send_mad(query->mad_buf, NULL);
1291         if (ret) {
1292                 spin_lock_irqsave(&idr_lock, flags);
1293                 idr_remove(&query_idr, id);
1294                 spin_unlock_irqrestore(&idr_lock, flags);
1295         }
1296
1297         /*
1298          * It's not safe to dereference query any more, because the
1299          * send may already have completed and freed the query in
1300          * another context.
1301          */
1302         return ret ? ret : id;
1303 }
1304
1305 void ib_sa_unpack_path(void *attribute, struct ib_sa_path_rec *rec)
1306 {
1307         ib_unpack(path_rec_table, ARRAY_SIZE(path_rec_table), attribute, rec);
1308 }
1309 EXPORT_SYMBOL(ib_sa_unpack_path);
1310
1311 void ib_sa_pack_path(struct ib_sa_path_rec *rec, void *attribute)
1312 {
1313         ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table), rec, attribute);
1314 }
1315 EXPORT_SYMBOL(ib_sa_pack_path);
1316
1317 static void ib_sa_path_rec_callback(struct ib_sa_query *sa_query,
1318                                     int status,
1319                                     struct ib_sa_mad *mad)
1320 {
1321         struct ib_sa_path_query *query =
1322                 container_of(sa_query, struct ib_sa_path_query, sa_query);
1323
1324         if (mad) {
1325                 struct ib_sa_path_rec rec;
1326
1327                 ib_unpack(path_rec_table, ARRAY_SIZE(path_rec_table),
1328                           mad->data, &rec);
1329                 rec.net = NULL;
1330                 rec.ifindex = 0;
1331                 rec.gid_type = IB_GID_TYPE_IB;
1332                 eth_zero_addr(rec.dmac);
1333                 query->callback(status, &rec, query->context);
1334         } else
1335                 query->callback(status, NULL, query->context);
1336 }
1337
1338 static void ib_sa_path_rec_release(struct ib_sa_query *sa_query)
1339 {
1340         kfree(container_of(sa_query, struct ib_sa_path_query, sa_query));
1341 }
1342
1343 /**
1344  * ib_sa_path_rec_get - Start a Path get query
1345  * @client:SA client
1346  * @device:device to send query on
1347  * @port_num: port number to send query on
1348  * @rec:Path Record to send in query
1349  * @comp_mask:component mask to send in query
1350  * @timeout_ms:time to wait for response
1351  * @gfp_mask:GFP mask to use for internal allocations
1352  * @callback:function called when query completes, times out or is
1353  * canceled
1354  * @context:opaque user context passed to callback
1355  * @sa_query:query context, used to cancel query
1356  *
1357  * Send a Path Record Get query to the SA to look up a path.  The
1358  * callback function will be called when the query completes (or
1359  * fails); status is 0 for a successful response, -EINTR if the query
1360  * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
1361  * occurred sending the query.  The resp parameter of the callback is
1362  * only valid if status is 0.
1363  *
1364  * If the return value of ib_sa_path_rec_get() is negative, it is an
1365  * error code.  Otherwise it is a query ID that can be used to cancel
1366  * the query.
1367  */
1368 int ib_sa_path_rec_get(struct ib_sa_client *client,
1369                        struct ib_device *device, u8 port_num,
1370                        struct ib_sa_path_rec *rec,
1371                        ib_sa_comp_mask comp_mask,
1372                        int timeout_ms, gfp_t gfp_mask,
1373                        void (*callback)(int status,
1374                                         struct ib_sa_path_rec *resp,
1375                                         void *context),
1376                        void *context,
1377                        struct ib_sa_query **sa_query)
1378 {
1379         struct ib_sa_path_query *query;
1380         struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1381         struct ib_sa_port   *port;
1382         struct ib_mad_agent *agent;
1383         struct ib_sa_mad *mad;
1384         int ret;
1385
1386         if (!sa_dev)
1387                 return -ENODEV;
1388
1389         port  = &sa_dev->port[port_num - sa_dev->start_port];
1390         agent = port->agent;
1391
1392         query = kzalloc(sizeof(*query), gfp_mask);
1393         if (!query)
1394                 return -ENOMEM;
1395
1396         query->sa_query.port     = port;
1397         ret = alloc_mad(&query->sa_query, gfp_mask);
1398         if (ret)
1399                 goto err1;
1400
1401         ib_sa_client_get(client);
1402         query->sa_query.client = client;
1403         query->callback        = callback;
1404         query->context         = context;
1405
1406         mad = query->sa_query.mad_buf->mad;
1407         init_mad(&query->sa_query, agent);
1408
1409         query->sa_query.callback = callback ? ib_sa_path_rec_callback : NULL;
1410         query->sa_query.release  = ib_sa_path_rec_release;
1411         mad->mad_hdr.method      = IB_MGMT_METHOD_GET;
1412         mad->mad_hdr.attr_id     = cpu_to_be16(IB_SA_ATTR_PATH_REC);
1413         mad->sa_hdr.comp_mask    = comp_mask;
1414
1415         ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table), rec, mad->data);
1416
1417         *sa_query = &query->sa_query;
1418
1419         query->sa_query.flags |= IB_SA_ENABLE_LOCAL_SERVICE;
1420         query->sa_query.mad_buf->context[1] = rec;
1421
1422         ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1423         if (ret < 0)
1424                 goto err2;
1425
1426         return ret;
1427
1428 err2:
1429         *sa_query = NULL;
1430         ib_sa_client_put(query->sa_query.client);
1431         free_mad(&query->sa_query);
1432
1433 err1:
1434         kfree(query);
1435         return ret;
1436 }
1437 EXPORT_SYMBOL(ib_sa_path_rec_get);
1438
1439 static void ib_sa_service_rec_callback(struct ib_sa_query *sa_query,
1440                                     int status,
1441                                     struct ib_sa_mad *mad)
1442 {
1443         struct ib_sa_service_query *query =
1444                 container_of(sa_query, struct ib_sa_service_query, sa_query);
1445
1446         if (mad) {
1447                 struct ib_sa_service_rec rec;
1448
1449                 ib_unpack(service_rec_table, ARRAY_SIZE(service_rec_table),
1450                           mad->data, &rec);
1451                 query->callback(status, &rec, query->context);
1452         } else
1453                 query->callback(status, NULL, query->context);
1454 }
1455
1456 static void ib_sa_service_rec_release(struct ib_sa_query *sa_query)
1457 {
1458         kfree(container_of(sa_query, struct ib_sa_service_query, sa_query));
1459 }
1460
1461 /**
1462  * ib_sa_service_rec_query - Start Service Record operation
1463  * @client:SA client
1464  * @device:device to send request on
1465  * @port_num: port number to send request on
1466  * @method:SA method - should be get, set, or delete
1467  * @rec:Service Record to send in request
1468  * @comp_mask:component mask to send in request
1469  * @timeout_ms:time to wait for response
1470  * @gfp_mask:GFP mask to use for internal allocations
1471  * @callback:function called when request completes, times out or is
1472  * canceled
1473  * @context:opaque user context passed to callback
1474  * @sa_query:request context, used to cancel request
1475  *
1476  * Send a Service Record set/get/delete to the SA to register,
1477  * unregister or query a service record.
1478  * The callback function will be called when the request completes (or
1479  * fails); status is 0 for a successful response, -EINTR if the query
1480  * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
1481  * occurred sending the query.  The resp parameter of the callback is
1482  * only valid if status is 0.
1483  *
1484  * If the return value of ib_sa_service_rec_query() is negative, it is an
1485  * error code.  Otherwise it is a request ID that can be used to cancel
1486  * the query.
1487  */
1488 int ib_sa_service_rec_query(struct ib_sa_client *client,
1489                             struct ib_device *device, u8 port_num, u8 method,
1490                             struct ib_sa_service_rec *rec,
1491                             ib_sa_comp_mask comp_mask,
1492                             int timeout_ms, gfp_t gfp_mask,
1493                             void (*callback)(int status,
1494                                              struct ib_sa_service_rec *resp,
1495                                              void *context),
1496                             void *context,
1497                             struct ib_sa_query **sa_query)
1498 {
1499         struct ib_sa_service_query *query;
1500         struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1501         struct ib_sa_port   *port;
1502         struct ib_mad_agent *agent;
1503         struct ib_sa_mad *mad;
1504         int ret;
1505
1506         if (!sa_dev)
1507                 return -ENODEV;
1508
1509         port  = &sa_dev->port[port_num - sa_dev->start_port];
1510         agent = port->agent;
1511
1512         if (method != IB_MGMT_METHOD_GET &&
1513             method != IB_MGMT_METHOD_SET &&
1514             method != IB_SA_METHOD_DELETE)
1515                 return -EINVAL;
1516
1517         query = kzalloc(sizeof(*query), gfp_mask);
1518         if (!query)
1519                 return -ENOMEM;
1520
1521         query->sa_query.port     = port;
1522         ret = alloc_mad(&query->sa_query, gfp_mask);
1523         if (ret)
1524                 goto err1;
1525
1526         ib_sa_client_get(client);
1527         query->sa_query.client = client;
1528         query->callback        = callback;
1529         query->context         = context;
1530
1531         mad = query->sa_query.mad_buf->mad;
1532         init_mad(&query->sa_query, agent);
1533
1534         query->sa_query.callback = callback ? ib_sa_service_rec_callback : NULL;
1535         query->sa_query.release  = ib_sa_service_rec_release;
1536         mad->mad_hdr.method      = method;
1537         mad->mad_hdr.attr_id     = cpu_to_be16(IB_SA_ATTR_SERVICE_REC);
1538         mad->sa_hdr.comp_mask    = comp_mask;
1539
1540         ib_pack(service_rec_table, ARRAY_SIZE(service_rec_table),
1541                 rec, mad->data);
1542
1543         *sa_query = &query->sa_query;
1544
1545         ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1546         if (ret < 0)
1547                 goto err2;
1548
1549         return ret;
1550
1551 err2:
1552         *sa_query = NULL;
1553         ib_sa_client_put(query->sa_query.client);
1554         free_mad(&query->sa_query);
1555
1556 err1:
1557         kfree(query);
1558         return ret;
1559 }
1560 EXPORT_SYMBOL(ib_sa_service_rec_query);
1561
1562 static void ib_sa_mcmember_rec_callback(struct ib_sa_query *sa_query,
1563                                         int status,
1564                                         struct ib_sa_mad *mad)
1565 {
1566         struct ib_sa_mcmember_query *query =
1567                 container_of(sa_query, struct ib_sa_mcmember_query, sa_query);
1568
1569         if (mad) {
1570                 struct ib_sa_mcmember_rec rec;
1571
1572                 ib_unpack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
1573                           mad->data, &rec);
1574                 query->callback(status, &rec, query->context);
1575         } else
1576                 query->callback(status, NULL, query->context);
1577 }
1578
1579 static void ib_sa_mcmember_rec_release(struct ib_sa_query *sa_query)
1580 {
1581         kfree(container_of(sa_query, struct ib_sa_mcmember_query, sa_query));
1582 }
1583
1584 int ib_sa_mcmember_rec_query(struct ib_sa_client *client,
1585                              struct ib_device *device, u8 port_num,
1586                              u8 method,
1587                              struct ib_sa_mcmember_rec *rec,
1588                              ib_sa_comp_mask comp_mask,
1589                              int timeout_ms, gfp_t gfp_mask,
1590                              void (*callback)(int status,
1591                                               struct ib_sa_mcmember_rec *resp,
1592                                               void *context),
1593                              void *context,
1594                              struct ib_sa_query **sa_query)
1595 {
1596         struct ib_sa_mcmember_query *query;
1597         struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1598         struct ib_sa_port   *port;
1599         struct ib_mad_agent *agent;
1600         struct ib_sa_mad *mad;
1601         int ret;
1602
1603         if (!sa_dev)
1604                 return -ENODEV;
1605
1606         port  = &sa_dev->port[port_num - sa_dev->start_port];
1607         agent = port->agent;
1608
1609         query = kzalloc(sizeof(*query), gfp_mask);
1610         if (!query)
1611                 return -ENOMEM;
1612
1613         query->sa_query.port     = port;
1614         ret = alloc_mad(&query->sa_query, gfp_mask);
1615         if (ret)
1616                 goto err1;
1617
1618         ib_sa_client_get(client);
1619         query->sa_query.client = client;
1620         query->callback        = callback;
1621         query->context         = context;
1622
1623         mad = query->sa_query.mad_buf->mad;
1624         init_mad(&query->sa_query, agent);
1625
1626         query->sa_query.callback = callback ? ib_sa_mcmember_rec_callback : NULL;
1627         query->sa_query.release  = ib_sa_mcmember_rec_release;
1628         mad->mad_hdr.method      = method;
1629         mad->mad_hdr.attr_id     = cpu_to_be16(IB_SA_ATTR_MC_MEMBER_REC);
1630         mad->sa_hdr.comp_mask    = comp_mask;
1631
1632         ib_pack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
1633                 rec, mad->data);
1634
1635         *sa_query = &query->sa_query;
1636
1637         ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1638         if (ret < 0)
1639                 goto err2;
1640
1641         return ret;
1642
1643 err2:
1644         *sa_query = NULL;
1645         ib_sa_client_put(query->sa_query.client);
1646         free_mad(&query->sa_query);
1647
1648 err1:
1649         kfree(query);
1650         return ret;
1651 }
1652
1653 /* Support GuidInfoRecord */
1654 static void ib_sa_guidinfo_rec_callback(struct ib_sa_query *sa_query,
1655                                         int status,
1656                                         struct ib_sa_mad *mad)
1657 {
1658         struct ib_sa_guidinfo_query *query =
1659                 container_of(sa_query, struct ib_sa_guidinfo_query, sa_query);
1660
1661         if (mad) {
1662                 struct ib_sa_guidinfo_rec rec;
1663
1664                 ib_unpack(guidinfo_rec_table, ARRAY_SIZE(guidinfo_rec_table),
1665                           mad->data, &rec);
1666                 query->callback(status, &rec, query->context);
1667         } else
1668                 query->callback(status, NULL, query->context);
1669 }
1670
1671 static void ib_sa_guidinfo_rec_release(struct ib_sa_query *sa_query)
1672 {
1673         kfree(container_of(sa_query, struct ib_sa_guidinfo_query, sa_query));
1674 }
1675
1676 int ib_sa_guid_info_rec_query(struct ib_sa_client *client,
1677                               struct ib_device *device, u8 port_num,
1678                               struct ib_sa_guidinfo_rec *rec,
1679                               ib_sa_comp_mask comp_mask, u8 method,
1680                               int timeout_ms, gfp_t gfp_mask,
1681                               void (*callback)(int status,
1682                                                struct ib_sa_guidinfo_rec *resp,
1683                                                void *context),
1684                               void *context,
1685                               struct ib_sa_query **sa_query)
1686 {
1687         struct ib_sa_guidinfo_query *query;
1688         struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1689         struct ib_sa_port *port;
1690         struct ib_mad_agent *agent;
1691         struct ib_sa_mad *mad;
1692         int ret;
1693
1694         if (!sa_dev)
1695                 return -ENODEV;
1696
1697         if (method != IB_MGMT_METHOD_GET &&
1698             method != IB_MGMT_METHOD_SET &&
1699             method != IB_SA_METHOD_DELETE) {
1700                 return -EINVAL;
1701         }
1702
1703         port  = &sa_dev->port[port_num - sa_dev->start_port];
1704         agent = port->agent;
1705
1706         query = kzalloc(sizeof(*query), gfp_mask);
1707         if (!query)
1708                 return -ENOMEM;
1709
1710         query->sa_query.port = port;
1711         ret = alloc_mad(&query->sa_query, gfp_mask);
1712         if (ret)
1713                 goto err1;
1714
1715         ib_sa_client_get(client);
1716         query->sa_query.client = client;
1717         query->callback        = callback;
1718         query->context         = context;
1719
1720         mad = query->sa_query.mad_buf->mad;
1721         init_mad(&query->sa_query, agent);
1722
1723         query->sa_query.callback = callback ? ib_sa_guidinfo_rec_callback : NULL;
1724         query->sa_query.release  = ib_sa_guidinfo_rec_release;
1725
1726         mad->mad_hdr.method      = method;
1727         mad->mad_hdr.attr_id     = cpu_to_be16(IB_SA_ATTR_GUID_INFO_REC);
1728         mad->sa_hdr.comp_mask    = comp_mask;
1729
1730         ib_pack(guidinfo_rec_table, ARRAY_SIZE(guidinfo_rec_table), rec,
1731                 mad->data);
1732
1733         *sa_query = &query->sa_query;
1734
1735         ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1736         if (ret < 0)
1737                 goto err2;
1738
1739         return ret;
1740
1741 err2:
1742         *sa_query = NULL;
1743         ib_sa_client_put(query->sa_query.client);
1744         free_mad(&query->sa_query);
1745
1746 err1:
1747         kfree(query);
1748         return ret;
1749 }
1750 EXPORT_SYMBOL(ib_sa_guid_info_rec_query);
1751
1752 bool ib_sa_sendonly_fullmem_support(struct ib_sa_client *client,
1753                                     struct ib_device *device,
1754                                     u8 port_num)
1755 {
1756         struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1757         struct ib_sa_port *port;
1758         bool ret = false;
1759         unsigned long flags;
1760
1761         if (!sa_dev)
1762                 return ret;
1763
1764         port  = &sa_dev->port[port_num - sa_dev->start_port];
1765
1766         spin_lock_irqsave(&port->classport_lock, flags);
1767         if ((port->classport_info.valid) &&
1768             (port->classport_info.data.type == RDMA_CLASS_PORT_INFO_IB))
1769                 ret = ib_get_cpi_capmask2(&port->classport_info.data.ib)
1770                         & IB_SA_CAP_MASK2_SENDONLY_FULL_MEM_SUPPORT;
1771         spin_unlock_irqrestore(&port->classport_lock, flags);
1772         return ret;
1773 }
1774 EXPORT_SYMBOL(ib_sa_sendonly_fullmem_support);
1775
1776 struct ib_classport_info_context {
1777         struct completion       done;
1778         struct ib_sa_query      *sa_query;
1779 };
1780
1781 static void ib_classportinfo_cb(void *context)
1782 {
1783         struct ib_classport_info_context *cb_ctx = context;
1784
1785         complete(&cb_ctx->done);
1786 }
1787
1788 static void ib_sa_classport_info_rec_callback(struct ib_sa_query *sa_query,
1789                                               int status,
1790                                               struct ib_sa_mad *mad)
1791 {
1792         unsigned long flags;
1793         struct ib_sa_classport_info_query *query =
1794                 container_of(sa_query, struct ib_sa_classport_info_query, sa_query);
1795         struct ib_sa_classport_cache *info = &sa_query->port->classport_info;
1796
1797         if (mad) {
1798                 if (sa_query->flags & IB_SA_QUERY_OPA) {
1799                         struct opa_class_port_info rec;
1800
1801                         ib_unpack(opa_classport_info_rec_table,
1802                                   ARRAY_SIZE(opa_classport_info_rec_table),
1803                                   mad->data, &rec);
1804
1805                         spin_lock_irqsave(&sa_query->port->classport_lock,
1806                                           flags);
1807                         if (!status && !info->valid) {
1808                                 memcpy(&info->data.opa, &rec,
1809                                        sizeof(info->data.opa));
1810
1811                                 info->valid = true;
1812                                 info->data.type = RDMA_CLASS_PORT_INFO_OPA;
1813                         }
1814                         spin_unlock_irqrestore(&sa_query->port->classport_lock,
1815                                                flags);
1816
1817                 } else {
1818                         struct ib_class_port_info rec;
1819
1820                         ib_unpack(ib_classport_info_rec_table,
1821                                   ARRAY_SIZE(ib_classport_info_rec_table),
1822                                   mad->data, &rec);
1823
1824                         spin_lock_irqsave(&sa_query->port->classport_lock,
1825                                           flags);
1826                         if (!status && !info->valid) {
1827                                 memcpy(&info->data.ib, &rec,
1828                                        sizeof(info->data.ib));
1829
1830                                 info->valid = true;
1831                                 info->data.type = RDMA_CLASS_PORT_INFO_IB;
1832                         }
1833                         spin_unlock_irqrestore(&sa_query->port->classport_lock,
1834                                                flags);
1835                 }
1836         }
1837         query->callback(query->context);
1838 }
1839
1840 static void ib_sa_classport_info_rec_release(struct ib_sa_query *sa_query)
1841 {
1842         kfree(container_of(sa_query, struct ib_sa_classport_info_query,
1843                            sa_query));
1844 }
1845
1846 static int ib_sa_classport_info_rec_query(struct ib_sa_port *port,
1847                                           int timeout_ms,
1848                                           void (*callback)(void *context),
1849                                           void *context,
1850                                           struct ib_sa_query **sa_query)
1851 {
1852         struct ib_mad_agent *agent;
1853         struct ib_sa_classport_info_query *query;
1854         struct ib_sa_mad *mad;
1855         gfp_t gfp_mask = GFP_KERNEL;
1856         int ret;
1857
1858         agent = port->agent;
1859
1860         query = kzalloc(sizeof(*query), gfp_mask);
1861         if (!query)
1862                 return -ENOMEM;
1863
1864         query->sa_query.port = port;
1865         query->sa_query.flags |= rdma_cap_opa_ah(port->agent->device,
1866                                                  port->port_num) ?
1867                                  IB_SA_QUERY_OPA : 0;
1868         ret = alloc_mad(&query->sa_query, gfp_mask);
1869         if (ret)
1870                 goto err_free;
1871
1872         query->callback = callback;
1873         query->context = context;
1874
1875         mad = query->sa_query.mad_buf->mad;
1876         init_mad(&query->sa_query, agent);
1877
1878         query->sa_query.callback = ib_sa_classport_info_rec_callback;
1879         query->sa_query.release  = ib_sa_classport_info_rec_release;
1880         mad->mad_hdr.method      = IB_MGMT_METHOD_GET;
1881         mad->mad_hdr.attr_id     = cpu_to_be16(IB_SA_ATTR_CLASS_PORTINFO);
1882         mad->sa_hdr.comp_mask    = 0;
1883         *sa_query = &query->sa_query;
1884
1885         ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1886         if (ret < 0)
1887                 goto err_free_mad;
1888
1889         return ret;
1890
1891 err_free_mad:
1892         *sa_query = NULL;
1893         free_mad(&query->sa_query);
1894
1895 err_free:
1896         kfree(query);
1897         return ret;
1898 }
1899
1900 static void update_ib_cpi(struct work_struct *work)
1901 {
1902         struct ib_sa_port *port =
1903                 container_of(work, struct ib_sa_port, ib_cpi_work.work);
1904         struct ib_classport_info_context *cb_context;
1905         unsigned long flags;
1906         int ret;
1907
1908         /* If the classport info is valid, nothing
1909          * to do here.
1910          */
1911         spin_lock_irqsave(&port->classport_lock, flags);
1912         if (port->classport_info.valid) {
1913                 spin_unlock_irqrestore(&port->classport_lock, flags);
1914                 return;
1915         }
1916         spin_unlock_irqrestore(&port->classport_lock, flags);
1917
1918         cb_context = kmalloc(sizeof(*cb_context), GFP_KERNEL);
1919         if (!cb_context)
1920                 goto err_nomem;
1921
1922         init_completion(&cb_context->done);
1923
1924         ret = ib_sa_classport_info_rec_query(port, 3000,
1925                                              ib_classportinfo_cb, cb_context,
1926                                              &cb_context->sa_query);
1927         if (ret < 0)
1928                 goto free_cb_err;
1929         wait_for_completion(&cb_context->done);
1930 free_cb_err:
1931         kfree(cb_context);
1932         spin_lock_irqsave(&port->classport_lock, flags);
1933
1934         /* If the classport info is still not valid, the query should have
1935          * failed for some reason. Retry issuing the query
1936          */
1937         if (!port->classport_info.valid) {
1938                 port->classport_info.retry_cnt++;
1939                 if (port->classport_info.retry_cnt <=
1940                     IB_SA_CPI_MAX_RETRY_CNT) {
1941                         unsigned long delay =
1942                                 msecs_to_jiffies(IB_SA_CPI_RETRY_WAIT);
1943
1944                         queue_delayed_work(ib_wq, &port->ib_cpi_work, delay);
1945                 }
1946         }
1947         spin_unlock_irqrestore(&port->classport_lock, flags);
1948
1949 err_nomem:
1950         return;
1951 }
1952
1953 static void send_handler(struct ib_mad_agent *agent,
1954                          struct ib_mad_send_wc *mad_send_wc)
1955 {
1956         struct ib_sa_query *query = mad_send_wc->send_buf->context[0];
1957         unsigned long flags;
1958
1959         if (query->callback)
1960                 switch (mad_send_wc->status) {
1961                 case IB_WC_SUCCESS:
1962                         /* No callback -- already got recv */
1963                         break;
1964                 case IB_WC_RESP_TIMEOUT_ERR:
1965                         query->callback(query, -ETIMEDOUT, NULL);
1966                         break;
1967                 case IB_WC_WR_FLUSH_ERR:
1968                         query->callback(query, -EINTR, NULL);
1969                         break;
1970                 default:
1971                         query->callback(query, -EIO, NULL);
1972                         break;
1973                 }
1974
1975         spin_lock_irqsave(&idr_lock, flags);
1976         idr_remove(&query_idr, query->id);
1977         spin_unlock_irqrestore(&idr_lock, flags);
1978
1979         free_mad(query);
1980         if (query->client)
1981                 ib_sa_client_put(query->client);
1982         query->release(query);
1983 }
1984
1985 static void recv_handler(struct ib_mad_agent *mad_agent,
1986                          struct ib_mad_send_buf *send_buf,
1987                          struct ib_mad_recv_wc *mad_recv_wc)
1988 {
1989         struct ib_sa_query *query;
1990
1991         if (!send_buf)
1992                 return;
1993
1994         query = send_buf->context[0];
1995         if (query->callback) {
1996                 if (mad_recv_wc->wc->status == IB_WC_SUCCESS)
1997                         query->callback(query,
1998                                         mad_recv_wc->recv_buf.mad->mad_hdr.status ?
1999                                         -EINVAL : 0,
2000                                         (struct ib_sa_mad *) mad_recv_wc->recv_buf.mad);
2001                 else
2002                         query->callback(query, -EIO, NULL);
2003         }
2004
2005         ib_free_recv_mad(mad_recv_wc);
2006 }
2007
2008 static void update_sm_ah(struct work_struct *work)
2009 {
2010         struct ib_sa_port *port =
2011                 container_of(work, struct ib_sa_port, update_task);
2012         struct ib_sa_sm_ah *new_ah;
2013         struct ib_port_attr port_attr;
2014         struct ib_ah_attr   ah_attr;
2015
2016         if (ib_query_port(port->agent->device, port->port_num, &port_attr)) {
2017                 pr_warn("Couldn't query port\n");
2018                 return;
2019         }
2020
2021         new_ah = kmalloc(sizeof(*new_ah), GFP_KERNEL);
2022         if (!new_ah)
2023                 return;
2024
2025         kref_init(&new_ah->ref);
2026         new_ah->src_path_mask = (1 << port_attr.lmc) - 1;
2027
2028         new_ah->pkey_index = 0;
2029         if (ib_find_pkey(port->agent->device, port->port_num,
2030                          IB_DEFAULT_PKEY_FULL, &new_ah->pkey_index))
2031                 pr_err("Couldn't find index for default PKey\n");
2032
2033         memset(&ah_attr, 0, sizeof(ah_attr));
2034         ah_attr.dlid     = port_attr.sm_lid;
2035         ah_attr.sl       = port_attr.sm_sl;
2036         ah_attr.port_num = port->port_num;
2037         if (port_attr.grh_required) {
2038                 ah_attr.ah_flags = IB_AH_GRH;
2039                 ah_attr.grh.dgid.global.subnet_prefix =
2040                         cpu_to_be64(port_attr.subnet_prefix);
2041                 ah_attr.grh.dgid.global.interface_id =
2042                         cpu_to_be64(IB_SA_WELL_KNOWN_GUID);
2043         }
2044
2045         new_ah->ah = ib_create_ah(port->agent->qp->pd, &ah_attr);
2046         if (IS_ERR(new_ah->ah)) {
2047                 pr_warn("Couldn't create new SM AH\n");
2048                 kfree(new_ah);
2049                 return;
2050         }
2051
2052         spin_lock_irq(&port->ah_lock);
2053         if (port->sm_ah)
2054                 kref_put(&port->sm_ah->ref, free_sm_ah);
2055         port->sm_ah = new_ah;
2056         spin_unlock_irq(&port->ah_lock);
2057 }
2058
2059 static void ib_sa_event(struct ib_event_handler *handler,
2060                         struct ib_event *event)
2061 {
2062         if (event->event == IB_EVENT_PORT_ERR    ||
2063             event->event == IB_EVENT_PORT_ACTIVE ||
2064             event->event == IB_EVENT_LID_CHANGE  ||
2065             event->event == IB_EVENT_PKEY_CHANGE ||
2066             event->event == IB_EVENT_SM_CHANGE   ||
2067             event->event == IB_EVENT_CLIENT_REREGISTER) {
2068                 unsigned long flags;
2069                 struct ib_sa_device *sa_dev =
2070                         container_of(handler, typeof(*sa_dev), event_handler);
2071                 u8 port_num = event->element.port_num - sa_dev->start_port;
2072                 struct ib_sa_port *port = &sa_dev->port[port_num];
2073
2074                 if (!rdma_cap_ib_sa(handler->device, port->port_num))
2075                         return;
2076
2077                 spin_lock_irqsave(&port->ah_lock, flags);
2078                 if (port->sm_ah)
2079                         kref_put(&port->sm_ah->ref, free_sm_ah);
2080                 port->sm_ah = NULL;
2081                 spin_unlock_irqrestore(&port->ah_lock, flags);
2082
2083                 if (event->event == IB_EVENT_SM_CHANGE ||
2084                     event->event == IB_EVENT_CLIENT_REREGISTER ||
2085                     event->event == IB_EVENT_LID_CHANGE ||
2086                     event->event == IB_EVENT_PORT_ACTIVE) {
2087                         unsigned long delay =
2088                                 msecs_to_jiffies(IB_SA_CPI_RETRY_WAIT);
2089
2090                         spin_lock_irqsave(&port->classport_lock, flags);
2091                         port->classport_info.valid = false;
2092                         port->classport_info.retry_cnt = 0;
2093                         spin_unlock_irqrestore(&port->classport_lock, flags);
2094                         queue_delayed_work(ib_wq,
2095                                            &port->ib_cpi_work, delay);
2096                 }
2097                 queue_work(ib_wq, &sa_dev->port[port_num].update_task);
2098         }
2099 }
2100
2101 static void ib_sa_add_one(struct ib_device *device)
2102 {
2103         struct ib_sa_device *sa_dev;
2104         int s, e, i;
2105         int count = 0;
2106
2107         s = rdma_start_port(device);
2108         e = rdma_end_port(device);
2109
2110         sa_dev = kzalloc(sizeof *sa_dev +
2111                          (e - s + 1) * sizeof (struct ib_sa_port),
2112                          GFP_KERNEL);
2113         if (!sa_dev)
2114                 return;
2115
2116         sa_dev->start_port = s;
2117         sa_dev->end_port   = e;
2118
2119         for (i = 0; i <= e - s; ++i) {
2120                 spin_lock_init(&sa_dev->port[i].ah_lock);
2121                 if (!rdma_cap_ib_sa(device, i + 1))
2122                         continue;
2123
2124                 sa_dev->port[i].sm_ah    = NULL;
2125                 sa_dev->port[i].port_num = i + s;
2126
2127                 spin_lock_init(&sa_dev->port[i].classport_lock);
2128                 sa_dev->port[i].classport_info.valid = false;
2129
2130                 sa_dev->port[i].agent =
2131                         ib_register_mad_agent(device, i + s, IB_QPT_GSI,
2132                                               NULL, 0, send_handler,
2133                                               recv_handler, sa_dev, 0);
2134                 if (IS_ERR(sa_dev->port[i].agent))
2135                         goto err;
2136
2137                 INIT_WORK(&sa_dev->port[i].update_task, update_sm_ah);
2138                 INIT_DELAYED_WORK(&sa_dev->port[i].ib_cpi_work,
2139                                   update_ib_cpi);
2140
2141                 count++;
2142         }
2143
2144         if (!count)
2145                 goto free;
2146
2147         ib_set_client_data(device, &sa_client, sa_dev);
2148
2149         /*
2150          * We register our event handler after everything is set up,
2151          * and then update our cached info after the event handler is
2152          * registered to avoid any problems if a port changes state
2153          * during our initialization.
2154          */
2155
2156         INIT_IB_EVENT_HANDLER(&sa_dev->event_handler, device, ib_sa_event);
2157         if (ib_register_event_handler(&sa_dev->event_handler))
2158                 goto err;
2159
2160         for (i = 0; i <= e - s; ++i) {
2161                 if (rdma_cap_ib_sa(device, i + 1))
2162                         update_sm_ah(&sa_dev->port[i].update_task);
2163         }
2164
2165         return;
2166
2167 err:
2168         while (--i >= 0) {
2169                 if (rdma_cap_ib_sa(device, i + 1))
2170                         ib_unregister_mad_agent(sa_dev->port[i].agent);
2171         }
2172 free:
2173         kfree(sa_dev);
2174         return;
2175 }
2176
2177 static void ib_sa_remove_one(struct ib_device *device, void *client_data)
2178 {
2179         struct ib_sa_device *sa_dev = client_data;
2180         int i;
2181
2182         if (!sa_dev)
2183                 return;
2184
2185         ib_unregister_event_handler(&sa_dev->event_handler);
2186         flush_workqueue(ib_wq);
2187
2188         for (i = 0; i <= sa_dev->end_port - sa_dev->start_port; ++i) {
2189                 if (rdma_cap_ib_sa(device, i + 1)) {
2190                         cancel_delayed_work_sync(&sa_dev->port[i].ib_cpi_work);
2191                         ib_unregister_mad_agent(sa_dev->port[i].agent);
2192                         if (sa_dev->port[i].sm_ah)
2193                                 kref_put(&sa_dev->port[i].sm_ah->ref, free_sm_ah);
2194                 }
2195
2196         }
2197
2198         kfree(sa_dev);
2199 }
2200
2201 int ib_sa_init(void)
2202 {
2203         int ret;
2204
2205         get_random_bytes(&tid, sizeof tid);
2206
2207         atomic_set(&ib_nl_sa_request_seq, 0);
2208
2209         ret = ib_register_client(&sa_client);
2210         if (ret) {
2211                 pr_err("Couldn't register ib_sa client\n");
2212                 goto err1;
2213         }
2214
2215         ret = mcast_init();
2216         if (ret) {
2217                 pr_err("Couldn't initialize multicast handling\n");
2218                 goto err2;
2219         }
2220
2221         ib_nl_wq = alloc_ordered_workqueue("ib_nl_sa_wq", WQ_MEM_RECLAIM);
2222         if (!ib_nl_wq) {
2223                 ret = -ENOMEM;
2224                 goto err3;
2225         }
2226
2227         INIT_DELAYED_WORK(&ib_nl_timed_work, ib_nl_request_timeout);
2228
2229         return 0;
2230
2231 err3:
2232         mcast_cleanup();
2233 err2:
2234         ib_unregister_client(&sa_client);
2235 err1:
2236         return ret;
2237 }
2238
2239 void ib_sa_cleanup(void)
2240 {
2241         cancel_delayed_work(&ib_nl_timed_work);
2242         flush_workqueue(ib_nl_wq);
2243         destroy_workqueue(ib_nl_wq);
2244         mcast_cleanup();
2245         ib_unregister_client(&sa_client);
2246         idr_destroy(&query_idr);
2247 }