[go: nahoru, domu]

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 "sa.h"
49
50MODULE_AUTHOR("Roland Dreier");
51MODULE_DESCRIPTION("InfiniBand subnet administration query support");
52MODULE_LICENSE("Dual BSD/GPL");
53
54struct ib_sa_sm_ah {
55	struct ib_ah        *ah;
56	struct kref          ref;
57	u16		     pkey_index;
58	u8		     src_path_mask;
59};
60
61struct ib_sa_port {
62	struct ib_mad_agent *agent;
63	struct ib_sa_sm_ah  *sm_ah;
64	struct work_struct   update_task;
65	spinlock_t           ah_lock;
66	u8                   port_num;
67};
68
69struct ib_sa_device {
70	int                     start_port, end_port;
71	struct ib_event_handler event_handler;
72	struct ib_sa_port port[0];
73};
74
75struct ib_sa_query {
76	void (*callback)(struct ib_sa_query *, int, struct ib_sa_mad *);
77	void (*release)(struct ib_sa_query *);
78	struct ib_sa_client    *client;
79	struct ib_sa_port      *port;
80	struct ib_mad_send_buf *mad_buf;
81	struct ib_sa_sm_ah     *sm_ah;
82	int			id;
83};
84
85struct ib_sa_service_query {
86	void (*callback)(int, struct ib_sa_service_rec *, void *);
87	void *context;
88	struct ib_sa_query sa_query;
89};
90
91struct ib_sa_path_query {
92	void (*callback)(int, struct ib_sa_path_rec *, void *);
93	void *context;
94	struct ib_sa_query sa_query;
95};
96
97struct ib_sa_guidinfo_query {
98	void (*callback)(int, struct ib_sa_guidinfo_rec *, void *);
99	void *context;
100	struct ib_sa_query sa_query;
101};
102
103struct ib_sa_mcmember_query {
104	void (*callback)(int, struct ib_sa_mcmember_rec *, void *);
105	void *context;
106	struct ib_sa_query sa_query;
107};
108
109static void ib_sa_add_one(struct ib_device *device);
110static void ib_sa_remove_one(struct ib_device *device);
111
112static struct ib_client sa_client = {
113	.name   = "sa",
114	.add    = ib_sa_add_one,
115	.remove = ib_sa_remove_one
116};
117
118static DEFINE_SPINLOCK(idr_lock);
119static DEFINE_IDR(query_idr);
120
121static DEFINE_SPINLOCK(tid_lock);
122static u32 tid;
123
124#define PATH_REC_FIELD(field) \
125	.struct_offset_bytes = offsetof(struct ib_sa_path_rec, field),		\
126	.struct_size_bytes   = sizeof ((struct ib_sa_path_rec *) 0)->field,	\
127	.field_name          = "sa_path_rec:" #field
128
129static const struct ib_field path_rec_table[] = {
130	{ PATH_REC_FIELD(service_id),
131	  .offset_words = 0,
132	  .offset_bits  = 0,
133	  .size_bits    = 64 },
134	{ PATH_REC_FIELD(dgid),
135	  .offset_words = 2,
136	  .offset_bits  = 0,
137	  .size_bits    = 128 },
138	{ PATH_REC_FIELD(sgid),
139	  .offset_words = 6,
140	  .offset_bits  = 0,
141	  .size_bits    = 128 },
142	{ PATH_REC_FIELD(dlid),
143	  .offset_words = 10,
144	  .offset_bits  = 0,
145	  .size_bits    = 16 },
146	{ PATH_REC_FIELD(slid),
147	  .offset_words = 10,
148	  .offset_bits  = 16,
149	  .size_bits    = 16 },
150	{ PATH_REC_FIELD(raw_traffic),
151	  .offset_words = 11,
152	  .offset_bits  = 0,
153	  .size_bits    = 1 },
154	{ RESERVED,
155	  .offset_words = 11,
156	  .offset_bits  = 1,
157	  .size_bits    = 3 },
158	{ PATH_REC_FIELD(flow_label),
159	  .offset_words = 11,
160	  .offset_bits  = 4,
161	  .size_bits    = 20 },
162	{ PATH_REC_FIELD(hop_limit),
163	  .offset_words = 11,
164	  .offset_bits  = 24,
165	  .size_bits    = 8 },
166	{ PATH_REC_FIELD(traffic_class),
167	  .offset_words = 12,
168	  .offset_bits  = 0,
169	  .size_bits    = 8 },
170	{ PATH_REC_FIELD(reversible),
171	  .offset_words = 12,
172	  .offset_bits  = 8,
173	  .size_bits    = 1 },
174	{ PATH_REC_FIELD(numb_path),
175	  .offset_words = 12,
176	  .offset_bits  = 9,
177	  .size_bits    = 7 },
178	{ PATH_REC_FIELD(pkey),
179	  .offset_words = 12,
180	  .offset_bits  = 16,
181	  .size_bits    = 16 },
182	{ PATH_REC_FIELD(qos_class),
183	  .offset_words = 13,
184	  .offset_bits  = 0,
185	  .size_bits    = 12 },
186	{ PATH_REC_FIELD(sl),
187	  .offset_words = 13,
188	  .offset_bits  = 12,
189	  .size_bits    = 4 },
190	{ PATH_REC_FIELD(mtu_selector),
191	  .offset_words = 13,
192	  .offset_bits  = 16,
193	  .size_bits    = 2 },
194	{ PATH_REC_FIELD(mtu),
195	  .offset_words = 13,
196	  .offset_bits  = 18,
197	  .size_bits    = 6 },
198	{ PATH_REC_FIELD(rate_selector),
199	  .offset_words = 13,
200	  .offset_bits  = 24,
201	  .size_bits    = 2 },
202	{ PATH_REC_FIELD(rate),
203	  .offset_words = 13,
204	  .offset_bits  = 26,
205	  .size_bits    = 6 },
206	{ PATH_REC_FIELD(packet_life_time_selector),
207	  .offset_words = 14,
208	  .offset_bits  = 0,
209	  .size_bits    = 2 },
210	{ PATH_REC_FIELD(packet_life_time),
211	  .offset_words = 14,
212	  .offset_bits  = 2,
213	  .size_bits    = 6 },
214	{ PATH_REC_FIELD(preference),
215	  .offset_words = 14,
216	  .offset_bits  = 8,
217	  .size_bits    = 8 },
218	{ RESERVED,
219	  .offset_words = 14,
220	  .offset_bits  = 16,
221	  .size_bits    = 48 },
222};
223
224#define MCMEMBER_REC_FIELD(field) \
225	.struct_offset_bytes = offsetof(struct ib_sa_mcmember_rec, field),	\
226	.struct_size_bytes   = sizeof ((struct ib_sa_mcmember_rec *) 0)->field,	\
227	.field_name          = "sa_mcmember_rec:" #field
228
229static const struct ib_field mcmember_rec_table[] = {
230	{ MCMEMBER_REC_FIELD(mgid),
231	  .offset_words = 0,
232	  .offset_bits  = 0,
233	  .size_bits    = 128 },
234	{ MCMEMBER_REC_FIELD(port_gid),
235	  .offset_words = 4,
236	  .offset_bits  = 0,
237	  .size_bits    = 128 },
238	{ MCMEMBER_REC_FIELD(qkey),
239	  .offset_words = 8,
240	  .offset_bits  = 0,
241	  .size_bits    = 32 },
242	{ MCMEMBER_REC_FIELD(mlid),
243	  .offset_words = 9,
244	  .offset_bits  = 0,
245	  .size_bits    = 16 },
246	{ MCMEMBER_REC_FIELD(mtu_selector),
247	  .offset_words = 9,
248	  .offset_bits  = 16,
249	  .size_bits    = 2 },
250	{ MCMEMBER_REC_FIELD(mtu),
251	  .offset_words = 9,
252	  .offset_bits  = 18,
253	  .size_bits    = 6 },
254	{ MCMEMBER_REC_FIELD(traffic_class),
255	  .offset_words = 9,
256	  .offset_bits  = 24,
257	  .size_bits    = 8 },
258	{ MCMEMBER_REC_FIELD(pkey),
259	  .offset_words = 10,
260	  .offset_bits  = 0,
261	  .size_bits    = 16 },
262	{ MCMEMBER_REC_FIELD(rate_selector),
263	  .offset_words = 10,
264	  .offset_bits  = 16,
265	  .size_bits    = 2 },
266	{ MCMEMBER_REC_FIELD(rate),
267	  .offset_words = 10,
268	  .offset_bits  = 18,
269	  .size_bits    = 6 },
270	{ MCMEMBER_REC_FIELD(packet_life_time_selector),
271	  .offset_words = 10,
272	  .offset_bits  = 24,
273	  .size_bits    = 2 },
274	{ MCMEMBER_REC_FIELD(packet_life_time),
275	  .offset_words = 10,
276	  .offset_bits  = 26,
277	  .size_bits    = 6 },
278	{ MCMEMBER_REC_FIELD(sl),
279	  .offset_words = 11,
280	  .offset_bits  = 0,
281	  .size_bits    = 4 },
282	{ MCMEMBER_REC_FIELD(flow_label),
283	  .offset_words = 11,
284	  .offset_bits  = 4,
285	  .size_bits    = 20 },
286	{ MCMEMBER_REC_FIELD(hop_limit),
287	  .offset_words = 11,
288	  .offset_bits  = 24,
289	  .size_bits    = 8 },
290	{ MCMEMBER_REC_FIELD(scope),
291	  .offset_words = 12,
292	  .offset_bits  = 0,
293	  .size_bits    = 4 },
294	{ MCMEMBER_REC_FIELD(join_state),
295	  .offset_words = 12,
296	  .offset_bits  = 4,
297	  .size_bits    = 4 },
298	{ MCMEMBER_REC_FIELD(proxy_join),
299	  .offset_words = 12,
300	  .offset_bits  = 8,
301	  .size_bits    = 1 },
302	{ RESERVED,
303	  .offset_words = 12,
304	  .offset_bits  = 9,
305	  .size_bits    = 23 },
306};
307
308#define SERVICE_REC_FIELD(field) \
309	.struct_offset_bytes = offsetof(struct ib_sa_service_rec, field),	\
310	.struct_size_bytes   = sizeof ((struct ib_sa_service_rec *) 0)->field,	\
311	.field_name          = "sa_service_rec:" #field
312
313static const struct ib_field service_rec_table[] = {
314	{ SERVICE_REC_FIELD(id),
315	  .offset_words = 0,
316	  .offset_bits  = 0,
317	  .size_bits    = 64 },
318	{ SERVICE_REC_FIELD(gid),
319	  .offset_words = 2,
320	  .offset_bits  = 0,
321	  .size_bits    = 128 },
322	{ SERVICE_REC_FIELD(pkey),
323	  .offset_words = 6,
324	  .offset_bits  = 0,
325	  .size_bits    = 16 },
326	{ SERVICE_REC_FIELD(lease),
327	  .offset_words = 7,
328	  .offset_bits  = 0,
329	  .size_bits    = 32 },
330	{ SERVICE_REC_FIELD(key),
331	  .offset_words = 8,
332	  .offset_bits  = 0,
333	  .size_bits    = 128 },
334	{ SERVICE_REC_FIELD(name),
335	  .offset_words = 12,
336	  .offset_bits  = 0,
337	  .size_bits    = 64*8 },
338	{ SERVICE_REC_FIELD(data8),
339	  .offset_words = 28,
340	  .offset_bits  = 0,
341	  .size_bits    = 16*8 },
342	{ SERVICE_REC_FIELD(data16),
343	  .offset_words = 32,
344	  .offset_bits  = 0,
345	  .size_bits    = 8*16 },
346	{ SERVICE_REC_FIELD(data32),
347	  .offset_words = 36,
348	  .offset_bits  = 0,
349	  .size_bits    = 4*32 },
350	{ SERVICE_REC_FIELD(data64),
351	  .offset_words = 40,
352	  .offset_bits  = 0,
353	  .size_bits    = 2*64 },
354};
355
356#define GUIDINFO_REC_FIELD(field) \
357	.struct_offset_bytes = offsetof(struct ib_sa_guidinfo_rec, field),	\
358	.struct_size_bytes   = sizeof((struct ib_sa_guidinfo_rec *) 0)->field,	\
359	.field_name          = "sa_guidinfo_rec:" #field
360
361static const struct ib_field guidinfo_rec_table[] = {
362	{ GUIDINFO_REC_FIELD(lid),
363	  .offset_words = 0,
364	  .offset_bits  = 0,
365	  .size_bits    = 16 },
366	{ GUIDINFO_REC_FIELD(block_num),
367	  .offset_words = 0,
368	  .offset_bits  = 16,
369	  .size_bits    = 8 },
370	{ GUIDINFO_REC_FIELD(res1),
371	  .offset_words = 0,
372	  .offset_bits  = 24,
373	  .size_bits    = 8 },
374	{ GUIDINFO_REC_FIELD(res2),
375	  .offset_words = 1,
376	  .offset_bits  = 0,
377	  .size_bits    = 32 },
378	{ GUIDINFO_REC_FIELD(guid_info_list),
379	  .offset_words = 2,
380	  .offset_bits  = 0,
381	  .size_bits    = 512 },
382};
383
384static void free_sm_ah(struct kref *kref)
385{
386	struct ib_sa_sm_ah *sm_ah = container_of(kref, struct ib_sa_sm_ah, ref);
387
388	ib_destroy_ah(sm_ah->ah);
389	kfree(sm_ah);
390}
391
392static void update_sm_ah(struct work_struct *work)
393{
394	struct ib_sa_port *port =
395		container_of(work, struct ib_sa_port, update_task);
396	struct ib_sa_sm_ah *new_ah;
397	struct ib_port_attr port_attr;
398	struct ib_ah_attr   ah_attr;
399
400	if (ib_query_port(port->agent->device, port->port_num, &port_attr)) {
401		printk(KERN_WARNING "Couldn't query port\n");
402		return;
403	}
404
405	new_ah = kmalloc(sizeof *new_ah, GFP_KERNEL);
406	if (!new_ah) {
407		printk(KERN_WARNING "Couldn't allocate new SM AH\n");
408		return;
409	}
410
411	kref_init(&new_ah->ref);
412	new_ah->src_path_mask = (1 << port_attr.lmc) - 1;
413
414	new_ah->pkey_index = 0;
415	if (ib_find_pkey(port->agent->device, port->port_num,
416			 IB_DEFAULT_PKEY_FULL, &new_ah->pkey_index))
417		printk(KERN_ERR "Couldn't find index for default PKey\n");
418
419	memset(&ah_attr, 0, sizeof ah_attr);
420	ah_attr.dlid     = port_attr.sm_lid;
421	ah_attr.sl       = port_attr.sm_sl;
422	ah_attr.port_num = port->port_num;
423
424	new_ah->ah = ib_create_ah(port->agent->qp->pd, &ah_attr);
425	if (IS_ERR(new_ah->ah)) {
426		printk(KERN_WARNING "Couldn't create new SM AH\n");
427		kfree(new_ah);
428		return;
429	}
430
431	spin_lock_irq(&port->ah_lock);
432	if (port->sm_ah)
433		kref_put(&port->sm_ah->ref, free_sm_ah);
434	port->sm_ah = new_ah;
435	spin_unlock_irq(&port->ah_lock);
436
437}
438
439static void ib_sa_event(struct ib_event_handler *handler, struct ib_event *event)
440{
441	if (event->event == IB_EVENT_PORT_ERR    ||
442	    event->event == IB_EVENT_PORT_ACTIVE ||
443	    event->event == IB_EVENT_LID_CHANGE  ||
444	    event->event == IB_EVENT_PKEY_CHANGE ||
445	    event->event == IB_EVENT_SM_CHANGE   ||
446	    event->event == IB_EVENT_CLIENT_REREGISTER) {
447		unsigned long flags;
448		struct ib_sa_device *sa_dev =
449			container_of(handler, typeof(*sa_dev), event_handler);
450		struct ib_sa_port *port =
451			&sa_dev->port[event->element.port_num - sa_dev->start_port];
452
453		if (rdma_port_get_link_layer(handler->device, port->port_num) != IB_LINK_LAYER_INFINIBAND)
454			return;
455
456		spin_lock_irqsave(&port->ah_lock, flags);
457		if (port->sm_ah)
458			kref_put(&port->sm_ah->ref, free_sm_ah);
459		port->sm_ah = NULL;
460		spin_unlock_irqrestore(&port->ah_lock, flags);
461
462		queue_work(ib_wq, &sa_dev->port[event->element.port_num -
463					    sa_dev->start_port].update_task);
464	}
465}
466
467void ib_sa_register_client(struct ib_sa_client *client)
468{
469	atomic_set(&client->users, 1);
470	init_completion(&client->comp);
471}
472EXPORT_SYMBOL(ib_sa_register_client);
473
474void ib_sa_unregister_client(struct ib_sa_client *client)
475{
476	ib_sa_client_put(client);
477	wait_for_completion(&client->comp);
478}
479EXPORT_SYMBOL(ib_sa_unregister_client);
480
481/**
482 * ib_sa_cancel_query - try to cancel an SA query
483 * @id:ID of query to cancel
484 * @query:query pointer to cancel
485 *
486 * Try to cancel an SA query.  If the id and query don't match up or
487 * the query has already completed, nothing is done.  Otherwise the
488 * query is canceled and will complete with a status of -EINTR.
489 */
490void ib_sa_cancel_query(int id, struct ib_sa_query *query)
491{
492	unsigned long flags;
493	struct ib_mad_agent *agent;
494	struct ib_mad_send_buf *mad_buf;
495
496	spin_lock_irqsave(&idr_lock, flags);
497	if (idr_find(&query_idr, id) != query) {
498		spin_unlock_irqrestore(&idr_lock, flags);
499		return;
500	}
501	agent = query->port->agent;
502	mad_buf = query->mad_buf;
503	spin_unlock_irqrestore(&idr_lock, flags);
504
505	ib_cancel_mad(agent, mad_buf);
506}
507EXPORT_SYMBOL(ib_sa_cancel_query);
508
509static u8 get_src_path_mask(struct ib_device *device, u8 port_num)
510{
511	struct ib_sa_device *sa_dev;
512	struct ib_sa_port   *port;
513	unsigned long flags;
514	u8 src_path_mask;
515
516	sa_dev = ib_get_client_data(device, &sa_client);
517	if (!sa_dev)
518		return 0x7f;
519
520	port  = &sa_dev->port[port_num - sa_dev->start_port];
521	spin_lock_irqsave(&port->ah_lock, flags);
522	src_path_mask = port->sm_ah ? port->sm_ah->src_path_mask : 0x7f;
523	spin_unlock_irqrestore(&port->ah_lock, flags);
524
525	return src_path_mask;
526}
527
528int ib_init_ah_from_path(struct ib_device *device, u8 port_num,
529			 struct ib_sa_path_rec *rec, struct ib_ah_attr *ah_attr)
530{
531	int ret;
532	u16 gid_index;
533	int force_grh;
534
535	memset(ah_attr, 0, sizeof *ah_attr);
536	ah_attr->dlid = be16_to_cpu(rec->dlid);
537	ah_attr->sl = rec->sl;
538	ah_attr->src_path_bits = be16_to_cpu(rec->slid) &
539				 get_src_path_mask(device, port_num);
540	ah_attr->port_num = port_num;
541	ah_attr->static_rate = rec->rate;
542
543	force_grh = rdma_port_get_link_layer(device, port_num) == IB_LINK_LAYER_ETHERNET;
544
545	if (rec->hop_limit > 1 || force_grh) {
546		ah_attr->ah_flags = IB_AH_GRH;
547		ah_attr->grh.dgid = rec->dgid;
548
549		ret = ib_find_cached_gid(device, &rec->sgid, &port_num,
550					 &gid_index);
551		if (ret)
552			return ret;
553
554		ah_attr->grh.sgid_index    = gid_index;
555		ah_attr->grh.flow_label    = be32_to_cpu(rec->flow_label);
556		ah_attr->grh.hop_limit     = rec->hop_limit;
557		ah_attr->grh.traffic_class = rec->traffic_class;
558	}
559	if (force_grh) {
560		memcpy(ah_attr->dmac, rec->dmac, ETH_ALEN);
561		ah_attr->vlan_id = rec->vlan_id;
562	} else {
563		ah_attr->vlan_id = 0xffff;
564	}
565
566	return 0;
567}
568EXPORT_SYMBOL(ib_init_ah_from_path);
569
570static int alloc_mad(struct ib_sa_query *query, gfp_t gfp_mask)
571{
572	unsigned long flags;
573
574	spin_lock_irqsave(&query->port->ah_lock, flags);
575	if (!query->port->sm_ah) {
576		spin_unlock_irqrestore(&query->port->ah_lock, flags);
577		return -EAGAIN;
578	}
579	kref_get(&query->port->sm_ah->ref);
580	query->sm_ah = query->port->sm_ah;
581	spin_unlock_irqrestore(&query->port->ah_lock, flags);
582
583	query->mad_buf = ib_create_send_mad(query->port->agent, 1,
584					    query->sm_ah->pkey_index,
585					    0, IB_MGMT_SA_HDR, IB_MGMT_SA_DATA,
586					    gfp_mask);
587	if (IS_ERR(query->mad_buf)) {
588		kref_put(&query->sm_ah->ref, free_sm_ah);
589		return -ENOMEM;
590	}
591
592	query->mad_buf->ah = query->sm_ah->ah;
593
594	return 0;
595}
596
597static void free_mad(struct ib_sa_query *query)
598{
599	ib_free_send_mad(query->mad_buf);
600	kref_put(&query->sm_ah->ref, free_sm_ah);
601}
602
603static void init_mad(struct ib_sa_mad *mad, struct ib_mad_agent *agent)
604{
605	unsigned long flags;
606
607	memset(mad, 0, sizeof *mad);
608
609	mad->mad_hdr.base_version  = IB_MGMT_BASE_VERSION;
610	mad->mad_hdr.mgmt_class    = IB_MGMT_CLASS_SUBN_ADM;
611	mad->mad_hdr.class_version = IB_SA_CLASS_VERSION;
612
613	spin_lock_irqsave(&tid_lock, flags);
614	mad->mad_hdr.tid           =
615		cpu_to_be64(((u64) agent->hi_tid) << 32 | tid++);
616	spin_unlock_irqrestore(&tid_lock, flags);
617}
618
619static int send_mad(struct ib_sa_query *query, int timeout_ms, gfp_t gfp_mask)
620{
621	bool preload = !!(gfp_mask & __GFP_WAIT);
622	unsigned long flags;
623	int ret, id;
624
625	if (preload)
626		idr_preload(gfp_mask);
627	spin_lock_irqsave(&idr_lock, flags);
628
629	id = idr_alloc(&query_idr, query, 0, 0, GFP_NOWAIT);
630
631	spin_unlock_irqrestore(&idr_lock, flags);
632	if (preload)
633		idr_preload_end();
634	if (id < 0)
635		return id;
636
637	query->mad_buf->timeout_ms  = timeout_ms;
638	query->mad_buf->context[0] = query;
639	query->id = id;
640
641	ret = ib_post_send_mad(query->mad_buf, NULL);
642	if (ret) {
643		spin_lock_irqsave(&idr_lock, flags);
644		idr_remove(&query_idr, id);
645		spin_unlock_irqrestore(&idr_lock, flags);
646	}
647
648	/*
649	 * It's not safe to dereference query any more, because the
650	 * send may already have completed and freed the query in
651	 * another context.
652	 */
653	return ret ? ret : id;
654}
655
656void ib_sa_unpack_path(void *attribute, struct ib_sa_path_rec *rec)
657{
658	ib_unpack(path_rec_table, ARRAY_SIZE(path_rec_table), attribute, rec);
659}
660EXPORT_SYMBOL(ib_sa_unpack_path);
661
662void ib_sa_pack_path(struct ib_sa_path_rec *rec, void *attribute)
663{
664	ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table), rec, attribute);
665}
666EXPORT_SYMBOL(ib_sa_pack_path);
667
668static void ib_sa_path_rec_callback(struct ib_sa_query *sa_query,
669				    int status,
670				    struct ib_sa_mad *mad)
671{
672	struct ib_sa_path_query *query =
673		container_of(sa_query, struct ib_sa_path_query, sa_query);
674
675	if (mad) {
676		struct ib_sa_path_rec rec;
677
678		ib_unpack(path_rec_table, ARRAY_SIZE(path_rec_table),
679			  mad->data, &rec);
680		rec.vlan_id = 0xffff;
681		memset(rec.dmac, 0, ETH_ALEN);
682		memset(rec.smac, 0, ETH_ALEN);
683		query->callback(status, &rec, query->context);
684	} else
685		query->callback(status, NULL, query->context);
686}
687
688static void ib_sa_path_rec_release(struct ib_sa_query *sa_query)
689{
690	kfree(container_of(sa_query, struct ib_sa_path_query, sa_query));
691}
692
693/**
694 * ib_sa_path_rec_get - Start a Path get query
695 * @client:SA client
696 * @device:device to send query on
697 * @port_num: port number to send query on
698 * @rec:Path Record to send in query
699 * @comp_mask:component mask to send in query
700 * @timeout_ms:time to wait for response
701 * @gfp_mask:GFP mask to use for internal allocations
702 * @callback:function called when query completes, times out or is
703 * canceled
704 * @context:opaque user context passed to callback
705 * @sa_query:query context, used to cancel query
706 *
707 * Send a Path Record Get query to the SA to look up a path.  The
708 * callback function will be called when the query completes (or
709 * fails); status is 0 for a successful response, -EINTR if the query
710 * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
711 * occurred sending the query.  The resp parameter of the callback is
712 * only valid if status is 0.
713 *
714 * If the return value of ib_sa_path_rec_get() is negative, it is an
715 * error code.  Otherwise it is a query ID that can be used to cancel
716 * the query.
717 */
718int ib_sa_path_rec_get(struct ib_sa_client *client,
719		       struct ib_device *device, u8 port_num,
720		       struct ib_sa_path_rec *rec,
721		       ib_sa_comp_mask comp_mask,
722		       int timeout_ms, gfp_t gfp_mask,
723		       void (*callback)(int status,
724					struct ib_sa_path_rec *resp,
725					void *context),
726		       void *context,
727		       struct ib_sa_query **sa_query)
728{
729	struct ib_sa_path_query *query;
730	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
731	struct ib_sa_port   *port;
732	struct ib_mad_agent *agent;
733	struct ib_sa_mad *mad;
734	int ret;
735
736	if (!sa_dev)
737		return -ENODEV;
738
739	port  = &sa_dev->port[port_num - sa_dev->start_port];
740	agent = port->agent;
741
742	query = kmalloc(sizeof *query, gfp_mask);
743	if (!query)
744		return -ENOMEM;
745
746	query->sa_query.port     = port;
747	ret = alloc_mad(&query->sa_query, gfp_mask);
748	if (ret)
749		goto err1;
750
751	ib_sa_client_get(client);
752	query->sa_query.client = client;
753	query->callback        = callback;
754	query->context         = context;
755
756	mad = query->sa_query.mad_buf->mad;
757	init_mad(mad, agent);
758
759	query->sa_query.callback = callback ? ib_sa_path_rec_callback : NULL;
760	query->sa_query.release  = ib_sa_path_rec_release;
761	mad->mad_hdr.method	 = IB_MGMT_METHOD_GET;
762	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_PATH_REC);
763	mad->sa_hdr.comp_mask	 = comp_mask;
764
765	ib_pack(path_rec_table, ARRAY_SIZE(path_rec_table), rec, mad->data);
766
767	*sa_query = &query->sa_query;
768
769	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
770	if (ret < 0)
771		goto err2;
772
773	return ret;
774
775err2:
776	*sa_query = NULL;
777	ib_sa_client_put(query->sa_query.client);
778	free_mad(&query->sa_query);
779
780err1:
781	kfree(query);
782	return ret;
783}
784EXPORT_SYMBOL(ib_sa_path_rec_get);
785
786static void ib_sa_service_rec_callback(struct ib_sa_query *sa_query,
787				    int status,
788				    struct ib_sa_mad *mad)
789{
790	struct ib_sa_service_query *query =
791		container_of(sa_query, struct ib_sa_service_query, sa_query);
792
793	if (mad) {
794		struct ib_sa_service_rec rec;
795
796		ib_unpack(service_rec_table, ARRAY_SIZE(service_rec_table),
797			  mad->data, &rec);
798		query->callback(status, &rec, query->context);
799	} else
800		query->callback(status, NULL, query->context);
801}
802
803static void ib_sa_service_rec_release(struct ib_sa_query *sa_query)
804{
805	kfree(container_of(sa_query, struct ib_sa_service_query, sa_query));
806}
807
808/**
809 * ib_sa_service_rec_query - Start Service Record operation
810 * @client:SA client
811 * @device:device to send request on
812 * @port_num: port number to send request on
813 * @method:SA method - should be get, set, or delete
814 * @rec:Service Record to send in request
815 * @comp_mask:component mask to send in request
816 * @timeout_ms:time to wait for response
817 * @gfp_mask:GFP mask to use for internal allocations
818 * @callback:function called when request completes, times out or is
819 * canceled
820 * @context:opaque user context passed to callback
821 * @sa_query:request context, used to cancel request
822 *
823 * Send a Service Record set/get/delete to the SA to register,
824 * unregister or query a service record.
825 * The callback function will be called when the request completes (or
826 * fails); status is 0 for a successful response, -EINTR if the query
827 * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
828 * occurred sending the query.  The resp parameter of the callback is
829 * only valid if status is 0.
830 *
831 * If the return value of ib_sa_service_rec_query() is negative, it is an
832 * error code.  Otherwise it is a request ID that can be used to cancel
833 * the query.
834 */
835int ib_sa_service_rec_query(struct ib_sa_client *client,
836			    struct ib_device *device, u8 port_num, u8 method,
837			    struct ib_sa_service_rec *rec,
838			    ib_sa_comp_mask comp_mask,
839			    int timeout_ms, gfp_t gfp_mask,
840			    void (*callback)(int status,
841					     struct ib_sa_service_rec *resp,
842					     void *context),
843			    void *context,
844			    struct ib_sa_query **sa_query)
845{
846	struct ib_sa_service_query *query;
847	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
848	struct ib_sa_port   *port;
849	struct ib_mad_agent *agent;
850	struct ib_sa_mad *mad;
851	int ret;
852
853	if (!sa_dev)
854		return -ENODEV;
855
856	port  = &sa_dev->port[port_num - sa_dev->start_port];
857	agent = port->agent;
858
859	if (method != IB_MGMT_METHOD_GET &&
860	    method != IB_MGMT_METHOD_SET &&
861	    method != IB_SA_METHOD_DELETE)
862		return -EINVAL;
863
864	query = kmalloc(sizeof *query, gfp_mask);
865	if (!query)
866		return -ENOMEM;
867
868	query->sa_query.port     = port;
869	ret = alloc_mad(&query->sa_query, gfp_mask);
870	if (ret)
871		goto err1;
872
873	ib_sa_client_get(client);
874	query->sa_query.client = client;
875	query->callback        = callback;
876	query->context         = context;
877
878	mad = query->sa_query.mad_buf->mad;
879	init_mad(mad, agent);
880
881	query->sa_query.callback = callback ? ib_sa_service_rec_callback : NULL;
882	query->sa_query.release  = ib_sa_service_rec_release;
883	mad->mad_hdr.method	 = method;
884	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_SERVICE_REC);
885	mad->sa_hdr.comp_mask	 = comp_mask;
886
887	ib_pack(service_rec_table, ARRAY_SIZE(service_rec_table),
888		rec, mad->data);
889
890	*sa_query = &query->sa_query;
891
892	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
893	if (ret < 0)
894		goto err2;
895
896	return ret;
897
898err2:
899	*sa_query = NULL;
900	ib_sa_client_put(query->sa_query.client);
901	free_mad(&query->sa_query);
902
903err1:
904	kfree(query);
905	return ret;
906}
907EXPORT_SYMBOL(ib_sa_service_rec_query);
908
909static void ib_sa_mcmember_rec_callback(struct ib_sa_query *sa_query,
910					int status,
911					struct ib_sa_mad *mad)
912{
913	struct ib_sa_mcmember_query *query =
914		container_of(sa_query, struct ib_sa_mcmember_query, sa_query);
915
916	if (mad) {
917		struct ib_sa_mcmember_rec rec;
918
919		ib_unpack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
920			  mad->data, &rec);
921		query->callback(status, &rec, query->context);
922	} else
923		query->callback(status, NULL, query->context);
924}
925
926static void ib_sa_mcmember_rec_release(struct ib_sa_query *sa_query)
927{
928	kfree(container_of(sa_query, struct ib_sa_mcmember_query, sa_query));
929}
930
931int ib_sa_mcmember_rec_query(struct ib_sa_client *client,
932			     struct ib_device *device, u8 port_num,
933			     u8 method,
934			     struct ib_sa_mcmember_rec *rec,
935			     ib_sa_comp_mask comp_mask,
936			     int timeout_ms, gfp_t gfp_mask,
937			     void (*callback)(int status,
938					      struct ib_sa_mcmember_rec *resp,
939					      void *context),
940			     void *context,
941			     struct ib_sa_query **sa_query)
942{
943	struct ib_sa_mcmember_query *query;
944	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
945	struct ib_sa_port   *port;
946	struct ib_mad_agent *agent;
947	struct ib_sa_mad *mad;
948	int ret;
949
950	if (!sa_dev)
951		return -ENODEV;
952
953	port  = &sa_dev->port[port_num - sa_dev->start_port];
954	agent = port->agent;
955
956	query = kmalloc(sizeof *query, gfp_mask);
957	if (!query)
958		return -ENOMEM;
959
960	query->sa_query.port     = port;
961	ret = alloc_mad(&query->sa_query, gfp_mask);
962	if (ret)
963		goto err1;
964
965	ib_sa_client_get(client);
966	query->sa_query.client = client;
967	query->callback        = callback;
968	query->context         = context;
969
970	mad = query->sa_query.mad_buf->mad;
971	init_mad(mad, agent);
972
973	query->sa_query.callback = callback ? ib_sa_mcmember_rec_callback : NULL;
974	query->sa_query.release  = ib_sa_mcmember_rec_release;
975	mad->mad_hdr.method	 = method;
976	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_MC_MEMBER_REC);
977	mad->sa_hdr.comp_mask	 = comp_mask;
978
979	ib_pack(mcmember_rec_table, ARRAY_SIZE(mcmember_rec_table),
980		rec, mad->data);
981
982	*sa_query = &query->sa_query;
983
984	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
985	if (ret < 0)
986		goto err2;
987
988	return ret;
989
990err2:
991	*sa_query = NULL;
992	ib_sa_client_put(query->sa_query.client);
993	free_mad(&query->sa_query);
994
995err1:
996	kfree(query);
997	return ret;
998}
999
1000/* Support GuidInfoRecord */
1001static void ib_sa_guidinfo_rec_callback(struct ib_sa_query *sa_query,
1002					int status,
1003					struct ib_sa_mad *mad)
1004{
1005	struct ib_sa_guidinfo_query *query =
1006		container_of(sa_query, struct ib_sa_guidinfo_query, sa_query);
1007
1008	if (mad) {
1009		struct ib_sa_guidinfo_rec rec;
1010
1011		ib_unpack(guidinfo_rec_table, ARRAY_SIZE(guidinfo_rec_table),
1012			  mad->data, &rec);
1013		query->callback(status, &rec, query->context);
1014	} else
1015		query->callback(status, NULL, query->context);
1016}
1017
1018static void ib_sa_guidinfo_rec_release(struct ib_sa_query *sa_query)
1019{
1020	kfree(container_of(sa_query, struct ib_sa_guidinfo_query, sa_query));
1021}
1022
1023int ib_sa_guid_info_rec_query(struct ib_sa_client *client,
1024			      struct ib_device *device, u8 port_num,
1025			      struct ib_sa_guidinfo_rec *rec,
1026			      ib_sa_comp_mask comp_mask, u8 method,
1027			      int timeout_ms, gfp_t gfp_mask,
1028			      void (*callback)(int status,
1029					       struct ib_sa_guidinfo_rec *resp,
1030					       void *context),
1031			      void *context,
1032			      struct ib_sa_query **sa_query)
1033{
1034	struct ib_sa_guidinfo_query *query;
1035	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1036	struct ib_sa_port *port;
1037	struct ib_mad_agent *agent;
1038	struct ib_sa_mad *mad;
1039	int ret;
1040
1041	if (!sa_dev)
1042		return -ENODEV;
1043
1044	if (method != IB_MGMT_METHOD_GET &&
1045	    method != IB_MGMT_METHOD_SET &&
1046	    method != IB_SA_METHOD_DELETE) {
1047		return -EINVAL;
1048	}
1049
1050	port  = &sa_dev->port[port_num - sa_dev->start_port];
1051	agent = port->agent;
1052
1053	query = kmalloc(sizeof *query, gfp_mask);
1054	if (!query)
1055		return -ENOMEM;
1056
1057	query->sa_query.port = port;
1058	ret = alloc_mad(&query->sa_query, gfp_mask);
1059	if (ret)
1060		goto err1;
1061
1062	ib_sa_client_get(client);
1063	query->sa_query.client = client;
1064	query->callback        = callback;
1065	query->context         = context;
1066
1067	mad = query->sa_query.mad_buf->mad;
1068	init_mad(mad, agent);
1069
1070	query->sa_query.callback = callback ? ib_sa_guidinfo_rec_callback : NULL;
1071	query->sa_query.release  = ib_sa_guidinfo_rec_release;
1072
1073	mad->mad_hdr.method	 = method;
1074	mad->mad_hdr.attr_id	 = cpu_to_be16(IB_SA_ATTR_GUID_INFO_REC);
1075	mad->sa_hdr.comp_mask	 = comp_mask;
1076
1077	ib_pack(guidinfo_rec_table, ARRAY_SIZE(guidinfo_rec_table), rec,
1078		mad->data);
1079
1080	*sa_query = &query->sa_query;
1081
1082	ret = send_mad(&query->sa_query, timeout_ms, gfp_mask);
1083	if (ret < 0)
1084		goto err2;
1085
1086	return ret;
1087
1088err2:
1089	*sa_query = NULL;
1090	ib_sa_client_put(query->sa_query.client);
1091	free_mad(&query->sa_query);
1092
1093err1:
1094	kfree(query);
1095	return ret;
1096}
1097EXPORT_SYMBOL(ib_sa_guid_info_rec_query);
1098
1099static void send_handler(struct ib_mad_agent *agent,
1100			 struct ib_mad_send_wc *mad_send_wc)
1101{
1102	struct ib_sa_query *query = mad_send_wc->send_buf->context[0];
1103	unsigned long flags;
1104
1105	if (query->callback)
1106		switch (mad_send_wc->status) {
1107		case IB_WC_SUCCESS:
1108			/* No callback -- already got recv */
1109			break;
1110		case IB_WC_RESP_TIMEOUT_ERR:
1111			query->callback(query, -ETIMEDOUT, NULL);
1112			break;
1113		case IB_WC_WR_FLUSH_ERR:
1114			query->callback(query, -EINTR, NULL);
1115			break;
1116		default:
1117			query->callback(query, -EIO, NULL);
1118			break;
1119		}
1120
1121	spin_lock_irqsave(&idr_lock, flags);
1122	idr_remove(&query_idr, query->id);
1123	spin_unlock_irqrestore(&idr_lock, flags);
1124
1125	free_mad(query);
1126	ib_sa_client_put(query->client);
1127	query->release(query);
1128}
1129
1130static void recv_handler(struct ib_mad_agent *mad_agent,
1131			 struct ib_mad_recv_wc *mad_recv_wc)
1132{
1133	struct ib_sa_query *query;
1134	struct ib_mad_send_buf *mad_buf;
1135
1136	mad_buf = (void *) (unsigned long) mad_recv_wc->wc->wr_id;
1137	query = mad_buf->context[0];
1138
1139	if (query->callback) {
1140		if (mad_recv_wc->wc->status == IB_WC_SUCCESS)
1141			query->callback(query,
1142					mad_recv_wc->recv_buf.mad->mad_hdr.status ?
1143					-EINVAL : 0,
1144					(struct ib_sa_mad *) mad_recv_wc->recv_buf.mad);
1145		else
1146			query->callback(query, -EIO, NULL);
1147	}
1148
1149	ib_free_recv_mad(mad_recv_wc);
1150}
1151
1152static void ib_sa_add_one(struct ib_device *device)
1153{
1154	struct ib_sa_device *sa_dev;
1155	int s, e, i;
1156
1157	if (rdma_node_get_transport(device->node_type) != RDMA_TRANSPORT_IB)
1158		return;
1159
1160	if (device->node_type == RDMA_NODE_IB_SWITCH)
1161		s = e = 0;
1162	else {
1163		s = 1;
1164		e = device->phys_port_cnt;
1165	}
1166
1167	sa_dev = kzalloc(sizeof *sa_dev +
1168			 (e - s + 1) * sizeof (struct ib_sa_port),
1169			 GFP_KERNEL);
1170	if (!sa_dev)
1171		return;
1172
1173	sa_dev->start_port = s;
1174	sa_dev->end_port   = e;
1175
1176	for (i = 0; i <= e - s; ++i) {
1177		spin_lock_init(&sa_dev->port[i].ah_lock);
1178		if (rdma_port_get_link_layer(device, i + 1) != IB_LINK_LAYER_INFINIBAND)
1179			continue;
1180
1181		sa_dev->port[i].sm_ah    = NULL;
1182		sa_dev->port[i].port_num = i + s;
1183
1184		sa_dev->port[i].agent =
1185			ib_register_mad_agent(device, i + s, IB_QPT_GSI,
1186					      NULL, 0, send_handler,
1187					      recv_handler, sa_dev, 0);
1188		if (IS_ERR(sa_dev->port[i].agent))
1189			goto err;
1190
1191		INIT_WORK(&sa_dev->port[i].update_task, update_sm_ah);
1192	}
1193
1194	ib_set_client_data(device, &sa_client, sa_dev);
1195
1196	/*
1197	 * We register our event handler after everything is set up,
1198	 * and then update our cached info after the event handler is
1199	 * registered to avoid any problems if a port changes state
1200	 * during our initialization.
1201	 */
1202
1203	INIT_IB_EVENT_HANDLER(&sa_dev->event_handler, device, ib_sa_event);
1204	if (ib_register_event_handler(&sa_dev->event_handler))
1205		goto err;
1206
1207	for (i = 0; i <= e - s; ++i)
1208		if (rdma_port_get_link_layer(device, i + 1) == IB_LINK_LAYER_INFINIBAND)
1209			update_sm_ah(&sa_dev->port[i].update_task);
1210
1211	return;
1212
1213err:
1214	while (--i >= 0)
1215		if (rdma_port_get_link_layer(device, i + 1) == IB_LINK_LAYER_INFINIBAND)
1216			ib_unregister_mad_agent(sa_dev->port[i].agent);
1217
1218	kfree(sa_dev);
1219
1220	return;
1221}
1222
1223static void ib_sa_remove_one(struct ib_device *device)
1224{
1225	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
1226	int i;
1227
1228	if (!sa_dev)
1229		return;
1230
1231	ib_unregister_event_handler(&sa_dev->event_handler);
1232
1233	flush_workqueue(ib_wq);
1234
1235	for (i = 0; i <= sa_dev->end_port - sa_dev->start_port; ++i) {
1236		if (rdma_port_get_link_layer(device, i + 1) == IB_LINK_LAYER_INFINIBAND) {
1237			ib_unregister_mad_agent(sa_dev->port[i].agent);
1238			if (sa_dev->port[i].sm_ah)
1239				kref_put(&sa_dev->port[i].sm_ah->ref, free_sm_ah);
1240		}
1241
1242	}
1243
1244	kfree(sa_dev);
1245}
1246
1247static int __init ib_sa_init(void)
1248{
1249	int ret;
1250
1251	get_random_bytes(&tid, sizeof tid);
1252
1253	ret = ib_register_client(&sa_client);
1254	if (ret) {
1255		printk(KERN_ERR "Couldn't register ib_sa client\n");
1256		goto err1;
1257	}
1258
1259	ret = mcast_init();
1260	if (ret) {
1261		printk(KERN_ERR "Couldn't initialize multicast handling\n");
1262		goto err2;
1263	}
1264
1265	return 0;
1266err2:
1267	ib_unregister_client(&sa_client);
1268err1:
1269	return ret;
1270}
1271
1272static void __exit ib_sa_cleanup(void)
1273{
1274	mcast_cleanup();
1275	ib_unregister_client(&sa_client);
1276	idr_destroy(&query_idr);
1277}
1278
1279module_init(ib_sa_init);
1280module_exit(ib_sa_cleanup);
1281