16181d6558548837b22decaa9484d9c033527458
[linux-2.6.git] / drivers / infiniband / core / cm.c
1 /*
2  * Copyright (c) 2004-2006 Intel Corporation.  All rights reserved.
3  * Copyright (c) 2004 Topspin Corporation.  All rights reserved.
4  * Copyright (c) 2004, 2005 Voltaire Corporation.  All rights reserved.
5  * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
6  *
7  * This software is available to you under a choice of one of two
8  * licenses.  You may choose to be licensed under the terms of the GNU
9  * General Public License (GPL) Version 2, available from the file
10  * COPYING in the main directory of this source tree, or the
11  * OpenIB.org BSD license below:
12  *
13  *     Redistribution and use in source and binary forms, with or
14  *     without modification, are permitted provided that the following
15  *     conditions are met:
16  *
17  *      - Redistributions of source code must retain the above
18  *        copyright notice, this list of conditions and the following
19  *        disclaimer.
20  *
21  *      - Redistributions in binary form must reproduce the above
22  *        copyright notice, this list of conditions and the following
23  *        disclaimer in the documentation and/or other materials
24  *        provided with the distribution.
25  *
26  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
27  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
28  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
29  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
30  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
31  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
32  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
33  * SOFTWARE.
34  *
35  * $Id: cm.c 4311 2005-12-05 18:42:01Z sean.hefty $
36  */
37
38 #include <linux/completion.h>
39 #include <linux/dma-mapping.h>
40 #include <linux/err.h>
41 #include <linux/idr.h>
42 #include <linux/interrupt.h>
43 #include <linux/random.h>
44 #include <linux/rbtree.h>
45 #include <linux/spinlock.h>
46 #include <linux/workqueue.h>
47
48 #include <rdma/ib_cache.h>
49 #include <rdma/ib_cm.h>
50 #include "cm_msgs.h"
51
52 MODULE_AUTHOR("Sean Hefty");
53 MODULE_DESCRIPTION("InfiniBand CM");
54 MODULE_LICENSE("Dual BSD/GPL");
55
56 static void cm_add_one(struct ib_device *device);
57 static void cm_remove_one(struct ib_device *device);
58
59 static struct ib_client cm_client = {
60         .name   = "cm",
61         .add    = cm_add_one,
62         .remove = cm_remove_one
63 };
64
65 static struct ib_cm {
66         spinlock_t lock;
67         struct list_head device_list;
68         rwlock_t device_lock;
69         struct rb_root listen_service_table;
70         u64 listen_service_id;
71         /* struct rb_root peer_service_table; todo: fix peer to peer */
72         struct rb_root remote_qp_table;
73         struct rb_root remote_id_table;
74         struct rb_root remote_sidr_table;
75         struct idr local_id_table;
76         __be32 random_id_operand;
77         struct list_head timewait_list;
78         struct workqueue_struct *wq;
79 } cm;
80
81 struct cm_port {
82         struct cm_device *cm_dev;
83         struct ib_mad_agent *mad_agent;
84         u8 port_num;
85 };
86
87 struct cm_device {
88         struct list_head list;
89         struct ib_device *device;
90         struct cm_port port[0];
91 };
92
93 struct cm_av {
94         struct cm_port *port;
95         union ib_gid dgid;
96         struct ib_ah_attr ah_attr;
97         u16 pkey_index;
98         u8 packet_life_time;
99 };
100
101 struct cm_work {
102         struct delayed_work work;
103         struct list_head list;
104         struct cm_port *port;
105         struct ib_mad_recv_wc *mad_recv_wc;     /* Received MADs */
106         __be32 local_id;                        /* Established / timewait */
107         __be32 remote_id;
108         struct ib_cm_event cm_event;
109         struct ib_sa_path_rec path[0];
110 };
111
112 struct cm_timewait_info {
113         struct cm_work work;                    /* Must be first. */
114         struct list_head list;
115         struct rb_node remote_qp_node;
116         struct rb_node remote_id_node;
117         __be64 remote_ca_guid;
118         __be32 remote_qpn;
119         u8 inserted_remote_qp;
120         u8 inserted_remote_id;
121 };
122
123 struct cm_id_private {
124         struct ib_cm_id id;
125
126         struct rb_node service_node;
127         struct rb_node sidr_id_node;
128         spinlock_t lock;        /* Do not acquire inside cm.lock */
129         struct completion comp;
130         atomic_t refcount;
131
132         struct ib_mad_send_buf *msg;
133         struct cm_timewait_info *timewait_info;
134         /* todo: use alternate port on send failure */
135         struct cm_av av;
136         struct cm_av alt_av;
137         struct ib_cm_compare_data *compare_data;
138
139         void *private_data;
140         __be64 tid;
141         __be32 local_qpn;
142         __be32 remote_qpn;
143         enum ib_qp_type qp_type;
144         __be32 sq_psn;
145         __be32 rq_psn;
146         int timeout_ms;
147         enum ib_mtu path_mtu;
148         __be16 pkey;
149         u8 private_data_len;
150         u8 max_cm_retries;
151         u8 peer_to_peer;
152         u8 responder_resources;
153         u8 initiator_depth;
154         u8 retry_count;
155         u8 rnr_retry_count;
156         u8 service_timeout;
157
158         struct list_head work_list;
159         atomic_t work_count;
160 };
161
162 static void cm_work_handler(struct work_struct *work);
163
164 static inline void cm_deref_id(struct cm_id_private *cm_id_priv)
165 {
166         if (atomic_dec_and_test(&cm_id_priv->refcount))
167                 complete(&cm_id_priv->comp);
168 }
169
170 static int cm_alloc_msg(struct cm_id_private *cm_id_priv,
171                         struct ib_mad_send_buf **msg)
172 {
173         struct ib_mad_agent *mad_agent;
174         struct ib_mad_send_buf *m;
175         struct ib_ah *ah;
176
177         mad_agent = cm_id_priv->av.port->mad_agent;
178         ah = ib_create_ah(mad_agent->qp->pd, &cm_id_priv->av.ah_attr);
179         if (IS_ERR(ah))
180                 return PTR_ERR(ah);
181
182         m = ib_create_send_mad(mad_agent, cm_id_priv->id.remote_cm_qpn,
183                                cm_id_priv->av.pkey_index,
184                                0, IB_MGMT_MAD_HDR, IB_MGMT_MAD_DATA,
185                                GFP_ATOMIC);
186         if (IS_ERR(m)) {
187                 ib_destroy_ah(ah);
188                 return PTR_ERR(m);
189         }
190
191         /* Timeout set by caller if response is expected. */
192         m->ah = ah;
193         m->retries = cm_id_priv->max_cm_retries;
194
195         atomic_inc(&cm_id_priv->refcount);
196         m->context[0] = cm_id_priv;
197         *msg = m;
198         return 0;
199 }
200
201 static int cm_alloc_response_msg(struct cm_port *port,
202                                  struct ib_mad_recv_wc *mad_recv_wc,
203                                  struct ib_mad_send_buf **msg)
204 {
205         struct ib_mad_send_buf *m;
206         struct ib_ah *ah;
207
208         ah = ib_create_ah_from_wc(port->mad_agent->qp->pd, mad_recv_wc->wc,
209                                   mad_recv_wc->recv_buf.grh, port->port_num);
210         if (IS_ERR(ah))
211                 return PTR_ERR(ah);
212
213         m = ib_create_send_mad(port->mad_agent, 1, mad_recv_wc->wc->pkey_index,
214                                0, IB_MGMT_MAD_HDR, IB_MGMT_MAD_DATA,
215                                GFP_ATOMIC);
216         if (IS_ERR(m)) {
217                 ib_destroy_ah(ah);
218                 return PTR_ERR(m);
219         }
220         m->ah = ah;
221         *msg = m;
222         return 0;
223 }
224
225 static void cm_free_msg(struct ib_mad_send_buf *msg)
226 {
227         ib_destroy_ah(msg->ah);
228         if (msg->context[0])
229                 cm_deref_id(msg->context[0]);
230         ib_free_send_mad(msg);
231 }
232
233 static void * cm_copy_private_data(const void *private_data,
234                                    u8 private_data_len)
235 {
236         void *data;
237
238         if (!private_data || !private_data_len)
239                 return NULL;
240
241         data = kmemdup(private_data, private_data_len, GFP_KERNEL);
242         if (!data)
243                 return ERR_PTR(-ENOMEM);
244
245         return data;
246 }
247
248 static void cm_set_private_data(struct cm_id_private *cm_id_priv,
249                                  void *private_data, u8 private_data_len)
250 {
251         if (cm_id_priv->private_data && cm_id_priv->private_data_len)
252                 kfree(cm_id_priv->private_data);
253
254         cm_id_priv->private_data = private_data;
255         cm_id_priv->private_data_len = private_data_len;
256 }
257
258 static void cm_init_av_for_response(struct cm_port *port, struct ib_wc *wc,
259                                     struct ib_grh *grh, struct cm_av *av)
260 {
261         av->port = port;
262         av->pkey_index = wc->pkey_index;
263         ib_init_ah_from_wc(port->cm_dev->device, port->port_num, wc,
264                            grh, &av->ah_attr);
265 }
266
267 static int cm_init_av_by_path(struct ib_sa_path_rec *path, struct cm_av *av)
268 {
269         struct cm_device *cm_dev;
270         struct cm_port *port = NULL;
271         unsigned long flags;
272         int ret;
273         u8 p;
274
275         read_lock_irqsave(&cm.device_lock, flags);
276         list_for_each_entry(cm_dev, &cm.device_list, list) {
277                 if (!ib_find_cached_gid(cm_dev->device, &path->sgid,
278                                         &p, NULL)) {
279                         port = &cm_dev->port[p-1];
280                         break;
281                 }
282         }
283         read_unlock_irqrestore(&cm.device_lock, flags);
284
285         if (!port)
286                 return -EINVAL;
287
288         ret = ib_find_cached_pkey(cm_dev->device, port->port_num,
289                                   be16_to_cpu(path->pkey), &av->pkey_index);
290         if (ret)
291                 return ret;
292
293         av->port = port;
294         ib_init_ah_from_path(cm_dev->device, port->port_num, path,
295                              &av->ah_attr);
296         av->packet_life_time = path->packet_life_time;
297         return 0;
298 }
299
300 static int cm_alloc_id(struct cm_id_private *cm_id_priv)
301 {
302         unsigned long flags;
303         int ret, id;
304         static int next_id;
305
306         do {
307                 spin_lock_irqsave(&cm.lock, flags);
308                 ret = idr_get_new_above(&cm.local_id_table, cm_id_priv,
309                                         next_id, &id);
310                 if (!ret)
311                         next_id = ((unsigned) id + 1) & MAX_ID_MASK;
312                 spin_unlock_irqrestore(&cm.lock, flags);
313         } while( (ret == -EAGAIN) && idr_pre_get(&cm.local_id_table, GFP_KERNEL) );
314
315         cm_id_priv->id.local_id = (__force __be32) (id ^ cm.random_id_operand);
316         return ret;
317 }
318
319 static void cm_free_id(__be32 local_id)
320 {
321         spin_lock_irq(&cm.lock);
322         idr_remove(&cm.local_id_table,
323                    (__force int) (local_id ^ cm.random_id_operand));
324         spin_unlock_irq(&cm.lock);
325 }
326
327 static struct cm_id_private * cm_get_id(__be32 local_id, __be32 remote_id)
328 {
329         struct cm_id_private *cm_id_priv;
330
331         cm_id_priv = idr_find(&cm.local_id_table,
332                               (__force int) (local_id ^ cm.random_id_operand));
333         if (cm_id_priv) {
334                 if (cm_id_priv->id.remote_id == remote_id)
335                         atomic_inc(&cm_id_priv->refcount);
336                 else
337                         cm_id_priv = NULL;
338         }
339
340         return cm_id_priv;
341 }
342
343 static struct cm_id_private * cm_acquire_id(__be32 local_id, __be32 remote_id)
344 {
345         struct cm_id_private *cm_id_priv;
346
347         spin_lock_irq(&cm.lock);
348         cm_id_priv = cm_get_id(local_id, remote_id);
349         spin_unlock_irq(&cm.lock);
350
351         return cm_id_priv;
352 }
353
354 static void cm_mask_copy(u8 *dst, u8 *src, u8 *mask)
355 {
356         int i;
357
358         for (i = 0; i < IB_CM_COMPARE_SIZE / sizeof(unsigned long); i++)
359                 ((unsigned long *) dst)[i] = ((unsigned long *) src)[i] &
360                                              ((unsigned long *) mask)[i];
361 }
362
363 static int cm_compare_data(struct ib_cm_compare_data *src_data,
364                            struct ib_cm_compare_data *dst_data)
365 {
366         u8 src[IB_CM_COMPARE_SIZE];
367         u8 dst[IB_CM_COMPARE_SIZE];
368
369         if (!src_data || !dst_data)
370                 return 0;
371
372         cm_mask_copy(src, src_data->data, dst_data->mask);
373         cm_mask_copy(dst, dst_data->data, src_data->mask);
374         return memcmp(src, dst, IB_CM_COMPARE_SIZE);
375 }
376
377 static int cm_compare_private_data(u8 *private_data,
378                                    struct ib_cm_compare_data *dst_data)
379 {
380         u8 src[IB_CM_COMPARE_SIZE];
381
382         if (!dst_data)
383                 return 0;
384
385         cm_mask_copy(src, private_data, dst_data->mask);
386         return memcmp(src, dst_data->data, IB_CM_COMPARE_SIZE);
387 }
388
389 static struct cm_id_private * cm_insert_listen(struct cm_id_private *cm_id_priv)
390 {
391         struct rb_node **link = &cm.listen_service_table.rb_node;
392         struct rb_node *parent = NULL;
393         struct cm_id_private *cur_cm_id_priv;
394         __be64 service_id = cm_id_priv->id.service_id;
395         __be64 service_mask = cm_id_priv->id.service_mask;
396         int data_cmp;
397
398         while (*link) {
399                 parent = *link;
400                 cur_cm_id_priv = rb_entry(parent, struct cm_id_private,
401                                           service_node);
402                 data_cmp = cm_compare_data(cm_id_priv->compare_data,
403                                            cur_cm_id_priv->compare_data);
404                 if ((cur_cm_id_priv->id.service_mask & service_id) ==
405                     (service_mask & cur_cm_id_priv->id.service_id) &&
406                     (cm_id_priv->id.device == cur_cm_id_priv->id.device) &&
407                     !data_cmp)
408                         return cur_cm_id_priv;
409
410                 if (cm_id_priv->id.device < cur_cm_id_priv->id.device)
411                         link = &(*link)->rb_left;
412                 else if (cm_id_priv->id.device > cur_cm_id_priv->id.device)
413                         link = &(*link)->rb_right;
414                 else if (service_id < cur_cm_id_priv->id.service_id)
415                         link = &(*link)->rb_left;
416                 else if (service_id > cur_cm_id_priv->id.service_id)
417                         link = &(*link)->rb_right;
418                 else if (data_cmp < 0)
419                         link = &(*link)->rb_left;
420                 else
421                         link = &(*link)->rb_right;
422         }
423         rb_link_node(&cm_id_priv->service_node, parent, link);
424         rb_insert_color(&cm_id_priv->service_node, &cm.listen_service_table);
425         return NULL;
426 }
427
428 static struct cm_id_private * cm_find_listen(struct ib_device *device,
429                                              __be64 service_id,
430                                              u8 *private_data)
431 {
432         struct rb_node *node = cm.listen_service_table.rb_node;
433         struct cm_id_private *cm_id_priv;
434         int data_cmp;
435
436         while (node) {
437                 cm_id_priv = rb_entry(node, struct cm_id_private, service_node);
438                 data_cmp = cm_compare_private_data(private_data,
439                                                    cm_id_priv->compare_data);
440                 if ((cm_id_priv->id.service_mask & service_id) ==
441                      cm_id_priv->id.service_id &&
442                     (cm_id_priv->id.device == device) && !data_cmp)
443                         return cm_id_priv;
444
445                 if (device < cm_id_priv->id.device)
446                         node = node->rb_left;
447                 else if (device > cm_id_priv->id.device)
448                         node = node->rb_right;
449                 else if (service_id < cm_id_priv->id.service_id)
450                         node = node->rb_left;
451                 else if (service_id > cm_id_priv->id.service_id)
452                         node = node->rb_right;
453                 else if (data_cmp < 0)
454                         node = node->rb_left;
455                 else
456                         node = node->rb_right;
457         }
458         return NULL;
459 }
460
461 static struct cm_timewait_info * cm_insert_remote_id(struct cm_timewait_info
462                                                      *timewait_info)
463 {
464         struct rb_node **link = &cm.remote_id_table.rb_node;
465         struct rb_node *parent = NULL;
466         struct cm_timewait_info *cur_timewait_info;
467         __be64 remote_ca_guid = timewait_info->remote_ca_guid;
468         __be32 remote_id = timewait_info->work.remote_id;
469
470         while (*link) {
471                 parent = *link;
472                 cur_timewait_info = rb_entry(parent, struct cm_timewait_info,
473                                              remote_id_node);
474                 if (remote_id < cur_timewait_info->work.remote_id)
475                         link = &(*link)->rb_left;
476                 else if (remote_id > cur_timewait_info->work.remote_id)
477                         link = &(*link)->rb_right;
478                 else if (remote_ca_guid < cur_timewait_info->remote_ca_guid)
479                         link = &(*link)->rb_left;
480                 else if (remote_ca_guid > cur_timewait_info->remote_ca_guid)
481                         link = &(*link)->rb_right;
482                 else
483                         return cur_timewait_info;
484         }
485         timewait_info->inserted_remote_id = 1;
486         rb_link_node(&timewait_info->remote_id_node, parent, link);
487         rb_insert_color(&timewait_info->remote_id_node, &cm.remote_id_table);
488         return NULL;
489 }
490
491 static struct cm_timewait_info * cm_find_remote_id(__be64 remote_ca_guid,
492                                                    __be32 remote_id)
493 {
494         struct rb_node *node = cm.remote_id_table.rb_node;
495         struct cm_timewait_info *timewait_info;
496
497         while (node) {
498                 timewait_info = rb_entry(node, struct cm_timewait_info,
499                                          remote_id_node);
500                 if (remote_id < timewait_info->work.remote_id)
501                         node = node->rb_left;
502                 else if (remote_id > timewait_info->work.remote_id)
503                         node = node->rb_right;
504                 else if (remote_ca_guid < timewait_info->remote_ca_guid)
505                         node = node->rb_left;
506                 else if (remote_ca_guid > timewait_info->remote_ca_guid)
507                         node = node->rb_right;
508                 else
509                         return timewait_info;
510         }
511         return NULL;
512 }
513
514 static struct cm_timewait_info * cm_insert_remote_qpn(struct cm_timewait_info
515                                                       *timewait_info)
516 {
517         struct rb_node **link = &cm.remote_qp_table.rb_node;
518         struct rb_node *parent = NULL;
519         struct cm_timewait_info *cur_timewait_info;
520         __be64 remote_ca_guid = timewait_info->remote_ca_guid;
521         __be32 remote_qpn = timewait_info->remote_qpn;
522
523         while (*link) {
524                 parent = *link;
525                 cur_timewait_info = rb_entry(parent, struct cm_timewait_info,
526                                              remote_qp_node);
527                 if (remote_qpn < cur_timewait_info->remote_qpn)
528                         link = &(*link)->rb_left;
529                 else if (remote_qpn > cur_timewait_info->remote_qpn)
530                         link = &(*link)->rb_right;
531                 else if (remote_ca_guid < cur_timewait_info->remote_ca_guid)
532                         link = &(*link)->rb_left;
533                 else if (remote_ca_guid > cur_timewait_info->remote_ca_guid)
534                         link = &(*link)->rb_right;
535                 else
536                         return cur_timewait_info;
537         }
538         timewait_info->inserted_remote_qp = 1;
539         rb_link_node(&timewait_info->remote_qp_node, parent, link);
540         rb_insert_color(&timewait_info->remote_qp_node, &cm.remote_qp_table);
541         return NULL;
542 }
543
544 static struct cm_id_private * cm_insert_remote_sidr(struct cm_id_private
545                                                     *cm_id_priv)
546 {
547         struct rb_node **link = &cm.remote_sidr_table.rb_node;
548         struct rb_node *parent = NULL;
549         struct cm_id_private *cur_cm_id_priv;
550         union ib_gid *port_gid = &cm_id_priv->av.dgid;
551         __be32 remote_id = cm_id_priv->id.remote_id;
552
553         while (*link) {
554                 parent = *link;
555                 cur_cm_id_priv = rb_entry(parent, struct cm_id_private,
556                                           sidr_id_node);
557                 if (remote_id < cur_cm_id_priv->id.remote_id)
558                         link = &(*link)->rb_left;
559                 else if (remote_id > cur_cm_id_priv->id.remote_id)
560                         link = &(*link)->rb_right;
561                 else {
562                         int cmp;
563                         cmp = memcmp(port_gid, &cur_cm_id_priv->av.dgid,
564                                      sizeof *port_gid);
565                         if (cmp < 0)
566                                 link = &(*link)->rb_left;
567                         else if (cmp > 0)
568                                 link = &(*link)->rb_right;
569                         else
570                                 return cur_cm_id_priv;
571                 }
572         }
573         rb_link_node(&cm_id_priv->sidr_id_node, parent, link);
574         rb_insert_color(&cm_id_priv->sidr_id_node, &cm.remote_sidr_table);
575         return NULL;
576 }
577
578 static void cm_reject_sidr_req(struct cm_id_private *cm_id_priv,
579                                enum ib_cm_sidr_status status)
580 {
581         struct ib_cm_sidr_rep_param param;
582
583         memset(&param, 0, sizeof param);
584         param.status = status;
585         ib_send_cm_sidr_rep(&cm_id_priv->id, &param);
586 }
587
588 struct ib_cm_id *ib_create_cm_id(struct ib_device *device,
589                                  ib_cm_handler cm_handler,
590                                  void *context)
591 {
592         struct cm_id_private *cm_id_priv;
593         int ret;
594
595         cm_id_priv = kzalloc(sizeof *cm_id_priv, GFP_KERNEL);
596         if (!cm_id_priv)
597                 return ERR_PTR(-ENOMEM);
598
599         cm_id_priv->id.state = IB_CM_IDLE;
600         cm_id_priv->id.device = device;
601         cm_id_priv->id.cm_handler = cm_handler;
602         cm_id_priv->id.context = context;
603         cm_id_priv->id.remote_cm_qpn = 1;
604         ret = cm_alloc_id(cm_id_priv);
605         if (ret)
606                 goto error;
607
608         spin_lock_init(&cm_id_priv->lock);
609         init_completion(&cm_id_priv->comp);
610         INIT_LIST_HEAD(&cm_id_priv->work_list);
611         atomic_set(&cm_id_priv->work_count, -1);
612         atomic_set(&cm_id_priv->refcount, 1);
613         return &cm_id_priv->id;
614
615 error:
616         kfree(cm_id_priv);
617         return ERR_PTR(-ENOMEM);
618 }
619 EXPORT_SYMBOL(ib_create_cm_id);
620
621 static struct cm_work * cm_dequeue_work(struct cm_id_private *cm_id_priv)
622 {
623         struct cm_work *work;
624
625         if (list_empty(&cm_id_priv->work_list))
626                 return NULL;
627
628         work = list_entry(cm_id_priv->work_list.next, struct cm_work, list);
629         list_del(&work->list);
630         return work;
631 }
632
633 static void cm_free_work(struct cm_work *work)
634 {
635         if (work->mad_recv_wc)
636                 ib_free_recv_mad(work->mad_recv_wc);
637         kfree(work);
638 }
639
640 static inline int cm_convert_to_ms(int iba_time)
641 {
642         /* approximate conversion to ms from 4.096us x 2^iba_time */
643         return 1 << max(iba_time - 8, 0);
644 }
645
646 static void cm_cleanup_timewait(struct cm_timewait_info *timewait_info)
647 {
648         if (timewait_info->inserted_remote_id) {
649                 rb_erase(&timewait_info->remote_id_node, &cm.remote_id_table);
650                 timewait_info->inserted_remote_id = 0;
651         }
652
653         if (timewait_info->inserted_remote_qp) {
654                 rb_erase(&timewait_info->remote_qp_node, &cm.remote_qp_table);
655                 timewait_info->inserted_remote_qp = 0;
656         }
657 }
658
659 static struct cm_timewait_info * cm_create_timewait_info(__be32 local_id)
660 {
661         struct cm_timewait_info *timewait_info;
662
663         timewait_info = kzalloc(sizeof *timewait_info, GFP_KERNEL);
664         if (!timewait_info)
665                 return ERR_PTR(-ENOMEM);
666
667         timewait_info->work.local_id = local_id;
668         INIT_DELAYED_WORK(&timewait_info->work.work, cm_work_handler);
669         timewait_info->work.cm_event.event = IB_CM_TIMEWAIT_EXIT;
670         return timewait_info;
671 }
672
673 static void cm_enter_timewait(struct cm_id_private *cm_id_priv)
674 {
675         int wait_time;
676         unsigned long flags;
677
678         spin_lock_irqsave(&cm.lock, flags);
679         cm_cleanup_timewait(cm_id_priv->timewait_info);
680         list_add_tail(&cm_id_priv->timewait_info->list, &cm.timewait_list);
681         spin_unlock_irqrestore(&cm.lock, flags);
682
683         /*
684          * The cm_id could be destroyed by the user before we exit timewait.
685          * To protect against this, we search for the cm_id after exiting
686          * timewait before notifying the user that we've exited timewait.
687          */
688         cm_id_priv->id.state = IB_CM_TIMEWAIT;
689         wait_time = cm_convert_to_ms(cm_id_priv->av.packet_life_time + 1);
690         queue_delayed_work(cm.wq, &cm_id_priv->timewait_info->work.work,
691                            msecs_to_jiffies(wait_time));
692         cm_id_priv->timewait_info = NULL;
693 }
694
695 static void cm_reset_to_idle(struct cm_id_private *cm_id_priv)
696 {
697         unsigned long flags;
698
699         cm_id_priv->id.state = IB_CM_IDLE;
700         if (cm_id_priv->timewait_info) {
701                 spin_lock_irqsave(&cm.lock, flags);
702                 cm_cleanup_timewait(cm_id_priv->timewait_info);
703                 spin_unlock_irqrestore(&cm.lock, flags);
704                 kfree(cm_id_priv->timewait_info);
705                 cm_id_priv->timewait_info = NULL;
706         }
707 }
708
709 static void cm_destroy_id(struct ib_cm_id *cm_id, int err)
710 {
711         struct cm_id_private *cm_id_priv;
712         struct cm_work *work;
713
714         cm_id_priv = container_of(cm_id, struct cm_id_private, id);
715 retest:
716         spin_lock_irq(&cm_id_priv->lock);
717         switch (cm_id->state) {
718         case IB_CM_LISTEN:
719                 cm_id->state = IB_CM_IDLE;
720                 spin_unlock_irq(&cm_id_priv->lock);
721                 spin_lock_irq(&cm.lock);
722                 rb_erase(&cm_id_priv->service_node, &cm.listen_service_table);
723                 spin_unlock_irq(&cm.lock);
724                 break;
725         case IB_CM_SIDR_REQ_SENT:
726                 cm_id->state = IB_CM_IDLE;
727                 ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg);
728                 spin_unlock_irq(&cm_id_priv->lock);
729                 break;
730         case IB_CM_SIDR_REQ_RCVD:
731                 spin_unlock_irq(&cm_id_priv->lock);
732                 cm_reject_sidr_req(cm_id_priv, IB_SIDR_REJECT);
733                 break;
734         case IB_CM_REQ_SENT:
735                 ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg);
736                 spin_unlock_irq(&cm_id_priv->lock);
737                 ib_send_cm_rej(cm_id, IB_CM_REJ_TIMEOUT,
738                                &cm_id_priv->id.device->node_guid,
739                                sizeof cm_id_priv->id.device->node_guid,
740                                NULL, 0);
741                 break;
742         case IB_CM_REQ_RCVD:
743                 if (err == -ENOMEM) {
744                         /* Do not reject to allow future retries. */
745                         cm_reset_to_idle(cm_id_priv);
746                         spin_unlock_irq(&cm_id_priv->lock);
747                 } else {
748                         spin_unlock_irq(&cm_id_priv->lock);
749                         ib_send_cm_rej(cm_id, IB_CM_REJ_CONSUMER_DEFINED,
750                                        NULL, 0, NULL, 0);
751                 }
752                 break;
753         case IB_CM_MRA_REQ_RCVD:
754         case IB_CM_REP_SENT:
755         case IB_CM_MRA_REP_RCVD:
756                 ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg);
757                 /* Fall through */
758         case IB_CM_MRA_REQ_SENT:
759         case IB_CM_REP_RCVD:
760         case IB_CM_MRA_REP_SENT:
761                 spin_unlock_irq(&cm_id_priv->lock);
762                 ib_send_cm_rej(cm_id, IB_CM_REJ_CONSUMER_DEFINED,
763                                NULL, 0, NULL, 0);
764                 break;
765         case IB_CM_ESTABLISHED:
766                 spin_unlock_irq(&cm_id_priv->lock);
767                 ib_send_cm_dreq(cm_id, NULL, 0);
768                 goto retest;
769         case IB_CM_DREQ_SENT:
770                 ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg);
771                 cm_enter_timewait(cm_id_priv);
772                 spin_unlock_irq(&cm_id_priv->lock);
773                 break;
774         case IB_CM_DREQ_RCVD:
775                 spin_unlock_irq(&cm_id_priv->lock);
776                 ib_send_cm_drep(cm_id, NULL, 0);
777                 break;
778         default:
779                 spin_unlock_irq(&cm_id_priv->lock);
780                 break;
781         }
782
783         cm_free_id(cm_id->local_id);
784         cm_deref_id(cm_id_priv);
785         wait_for_completion(&cm_id_priv->comp);
786         while ((work = cm_dequeue_work(cm_id_priv)) != NULL)
787                 cm_free_work(work);
788         kfree(cm_id_priv->compare_data);
789         kfree(cm_id_priv->private_data);
790         kfree(cm_id_priv);
791 }
792
793 void ib_destroy_cm_id(struct ib_cm_id *cm_id)
794 {
795         cm_destroy_id(cm_id, 0);
796 }
797 EXPORT_SYMBOL(ib_destroy_cm_id);
798
799 int ib_cm_listen(struct ib_cm_id *cm_id, __be64 service_id, __be64 service_mask,
800                  struct ib_cm_compare_data *compare_data)
801 {
802         struct cm_id_private *cm_id_priv, *cur_cm_id_priv;
803         unsigned long flags;
804         int ret = 0;
805
806         service_mask = service_mask ? service_mask :
807                        __constant_cpu_to_be64(~0ULL);
808         service_id &= service_mask;
809         if ((service_id & IB_SERVICE_ID_AGN_MASK) == IB_CM_ASSIGN_SERVICE_ID &&
810             (service_id != IB_CM_ASSIGN_SERVICE_ID))
811                 return -EINVAL;
812
813         cm_id_priv = container_of(cm_id, struct cm_id_private, id);
814         if (cm_id->state != IB_CM_IDLE)
815                 return -EINVAL;
816
817         if (compare_data) {
818                 cm_id_priv->compare_data = kzalloc(sizeof *compare_data,
819                                                    GFP_KERNEL);
820                 if (!cm_id_priv->compare_data)
821                         return -ENOMEM;
822                 cm_mask_copy(cm_id_priv->compare_data->data,
823                              compare_data->data, compare_data->mask);
824                 memcpy(cm_id_priv->compare_data->mask, compare_data->mask,
825                        IB_CM_COMPARE_SIZE);
826         }
827
828         cm_id->state = IB_CM_LISTEN;
829
830         spin_lock_irqsave(&cm.lock, flags);
831         if (service_id == IB_CM_ASSIGN_SERVICE_ID) {
832                 cm_id->service_id = cpu_to_be64(cm.listen_service_id++);
833                 cm_id->service_mask = __constant_cpu_to_be64(~0ULL);
834         } else {
835                 cm_id->service_id = service_id;
836                 cm_id->service_mask = service_mask;
837         }
838         cur_cm_id_priv = cm_insert_listen(cm_id_priv);
839         spin_unlock_irqrestore(&cm.lock, flags);
840
841         if (cur_cm_id_priv) {
842                 cm_id->state = IB_CM_IDLE;
843                 kfree(cm_id_priv->compare_data);
844                 cm_id_priv->compare_data = NULL;
845                 ret = -EBUSY;
846         }
847         return ret;
848 }
849 EXPORT_SYMBOL(ib_cm_listen);
850
851 static __be64 cm_form_tid(struct cm_id_private *cm_id_priv,
852                           enum cm_msg_sequence msg_seq)
853 {
854         u64 hi_tid, low_tid;
855
856         hi_tid   = ((u64) cm_id_priv->av.port->mad_agent->hi_tid) << 32;
857         low_tid  = (u64) ((__force u32)cm_id_priv->id.local_id |
858                           (msg_seq << 30));
859         return cpu_to_be64(hi_tid | low_tid);
860 }
861
862 static void cm_format_mad_hdr(struct ib_mad_hdr *hdr,
863                               __be16 attr_id, __be64 tid)
864 {
865         hdr->base_version  = IB_MGMT_BASE_VERSION;
866         hdr->mgmt_class    = IB_MGMT_CLASS_CM;
867         hdr->class_version = IB_CM_CLASS_VERSION;
868         hdr->method        = IB_MGMT_METHOD_SEND;
869         hdr->attr_id       = attr_id;
870         hdr->tid           = tid;
871 }
872
873 static void cm_format_req(struct cm_req_msg *req_msg,
874                           struct cm_id_private *cm_id_priv,
875                           struct ib_cm_req_param *param)
876 {
877         cm_format_mad_hdr(&req_msg->hdr, CM_REQ_ATTR_ID,
878                           cm_form_tid(cm_id_priv, CM_MSG_SEQUENCE_REQ));
879
880         req_msg->local_comm_id = cm_id_priv->id.local_id;
881         req_msg->service_id = param->service_id;
882         req_msg->local_ca_guid = cm_id_priv->id.device->node_guid;
883         cm_req_set_local_qpn(req_msg, cpu_to_be32(param->qp_num));
884         cm_req_set_resp_res(req_msg, param->responder_resources);
885         cm_req_set_init_depth(req_msg, param->initiator_depth);
886         cm_req_set_remote_resp_timeout(req_msg,
887                                        param->remote_cm_response_timeout);
888         cm_req_set_qp_type(req_msg, param->qp_type);
889         cm_req_set_flow_ctrl(req_msg, param->flow_control);
890         cm_req_set_starting_psn(req_msg, cpu_to_be32(param->starting_psn));
891         cm_req_set_local_resp_timeout(req_msg,
892                                       param->local_cm_response_timeout);
893         cm_req_set_retry_count(req_msg, param->retry_count);
894         req_msg->pkey = param->primary_path->pkey;
895         cm_req_set_path_mtu(req_msg, param->primary_path->mtu);
896         cm_req_set_rnr_retry_count(req_msg, param->rnr_retry_count);
897         cm_req_set_max_cm_retries(req_msg, param->max_cm_retries);
898         cm_req_set_srq(req_msg, param->srq);
899
900         req_msg->primary_local_lid = param->primary_path->slid;
901         req_msg->primary_remote_lid = param->primary_path->dlid;
902         req_msg->primary_local_gid = param->primary_path->sgid;
903         req_msg->primary_remote_gid = param->primary_path->dgid;
904         cm_req_set_primary_flow_label(req_msg, param->primary_path->flow_label);
905         cm_req_set_primary_packet_rate(req_msg, param->primary_path->rate);
906         req_msg->primary_traffic_class = param->primary_path->traffic_class;
907         req_msg->primary_hop_limit = param->primary_path->hop_limit;
908         cm_req_set_primary_sl(req_msg, param->primary_path->sl);
909         cm_req_set_primary_subnet_local(req_msg, 1); /* local only... */
910         cm_req_set_primary_local_ack_timeout(req_msg,
911                 min(31, param->primary_path->packet_life_time + 1));
912
913         if (param->alternate_path) {
914                 req_msg->alt_local_lid = param->alternate_path->slid;
915                 req_msg->alt_remote_lid = param->alternate_path->dlid;
916                 req_msg->alt_local_gid = param->alternate_path->sgid;
917                 req_msg->alt_remote_gid = param->alternate_path->dgid;
918                 cm_req_set_alt_flow_label(req_msg,
919                                           param->alternate_path->flow_label);
920                 cm_req_set_alt_packet_rate(req_msg, param->alternate_path->rate);
921                 req_msg->alt_traffic_class = param->alternate_path->traffic_class;
922                 req_msg->alt_hop_limit = param->alternate_path->hop_limit;
923                 cm_req_set_alt_sl(req_msg, param->alternate_path->sl);
924                 cm_req_set_alt_subnet_local(req_msg, 1); /* local only... */
925                 cm_req_set_alt_local_ack_timeout(req_msg,
926                         min(31, param->alternate_path->packet_life_time + 1));
927         }
928
929         if (param->private_data && param->private_data_len)
930                 memcpy(req_msg->private_data, param->private_data,
931                        param->private_data_len);
932 }
933
934 static int cm_validate_req_param(struct ib_cm_req_param *param)
935 {
936         /* peer-to-peer not supported */
937         if (param->peer_to_peer)
938                 return -EINVAL;
939
940         if (!param->primary_path)
941                 return -EINVAL;
942
943         if (param->qp_type != IB_QPT_RC && param->qp_type != IB_QPT_UC)
944                 return -EINVAL;
945
946         if (param->private_data &&
947             param->private_data_len > IB_CM_REQ_PRIVATE_DATA_SIZE)
948                 return -EINVAL;
949
950         if (param->alternate_path &&
951             (param->alternate_path->pkey != param->primary_path->pkey ||
952              param->alternate_path->mtu != param->primary_path->mtu))
953                 return -EINVAL;
954
955         return 0;
956 }
957
958 int ib_send_cm_req(struct ib_cm_id *cm_id,
959                    struct ib_cm_req_param *param)
960 {
961         struct cm_id_private *cm_id_priv;
962         struct cm_req_msg *req_msg;
963         unsigned long flags;
964         int ret;
965
966         ret = cm_validate_req_param(param);
967         if (ret)
968                 return ret;
969
970         /* Verify that we're not in timewait. */
971         cm_id_priv = container_of(cm_id, struct cm_id_private, id);
972         spin_lock_irqsave(&cm_id_priv->lock, flags);
973         if (cm_id->state != IB_CM_IDLE) {
974                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
975                 ret = -EINVAL;
976                 goto out;
977         }
978         spin_unlock_irqrestore(&cm_id_priv->lock, flags);
979
980         cm_id_priv->timewait_info = cm_create_timewait_info(cm_id_priv->
981                                                             id.local_id);
982         if (IS_ERR(cm_id_priv->timewait_info)) {
983                 ret = PTR_ERR(cm_id_priv->timewait_info);
984                 goto out;
985         }
986
987         ret = cm_init_av_by_path(param->primary_path, &cm_id_priv->av);
988         if (ret)
989                 goto error1;
990         if (param->alternate_path) {
991                 ret = cm_init_av_by_path(param->alternate_path,
992                                          &cm_id_priv->alt_av);
993                 if (ret)
994                         goto error1;
995         }
996         cm_id->service_id = param->service_id;
997         cm_id->service_mask = __constant_cpu_to_be64(~0ULL);
998         cm_id_priv->timeout_ms = cm_convert_to_ms(
999                                     param->primary_path->packet_life_time) * 2 +
1000                                  cm_convert_to_ms(
1001                                     param->remote_cm_response_timeout);
1002         cm_id_priv->max_cm_retries = param->max_cm_retries;
1003         cm_id_priv->initiator_depth = param->initiator_depth;
1004         cm_id_priv->responder_resources = param->responder_resources;
1005         cm_id_priv->retry_count = param->retry_count;
1006         cm_id_priv->path_mtu = param->primary_path->mtu;
1007         cm_id_priv->pkey = param->primary_path->pkey;
1008         cm_id_priv->qp_type = param->qp_type;
1009
1010         ret = cm_alloc_msg(cm_id_priv, &cm_id_priv->msg);
1011         if (ret)
1012                 goto error1;
1013
1014         req_msg = (struct cm_req_msg *) cm_id_priv->msg->mad;
1015         cm_format_req(req_msg, cm_id_priv, param);
1016         cm_id_priv->tid = req_msg->hdr.tid;
1017         cm_id_priv->msg->timeout_ms = cm_id_priv->timeout_ms;
1018         cm_id_priv->msg->context[1] = (void *) (unsigned long) IB_CM_REQ_SENT;
1019
1020         cm_id_priv->local_qpn = cm_req_get_local_qpn(req_msg);
1021         cm_id_priv->rq_psn = cm_req_get_starting_psn(req_msg);
1022
1023         spin_lock_irqsave(&cm_id_priv->lock, flags);
1024         ret = ib_post_send_mad(cm_id_priv->msg, NULL);
1025         if (ret) {
1026                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1027                 goto error2;
1028         }
1029         BUG_ON(cm_id->state != IB_CM_IDLE);
1030         cm_id->state = IB_CM_REQ_SENT;
1031         spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1032         return 0;
1033
1034 error2: cm_free_msg(cm_id_priv->msg);
1035 error1: kfree(cm_id_priv->timewait_info);
1036 out:    return ret;
1037 }
1038 EXPORT_SYMBOL(ib_send_cm_req);
1039
1040 static int cm_issue_rej(struct cm_port *port,
1041                         struct ib_mad_recv_wc *mad_recv_wc,
1042                         enum ib_cm_rej_reason reason,
1043                         enum cm_msg_response msg_rejected,
1044                         void *ari, u8 ari_length)
1045 {
1046         struct ib_mad_send_buf *msg = NULL;
1047         struct cm_rej_msg *rej_msg, *rcv_msg;
1048         int ret;
1049
1050         ret = cm_alloc_response_msg(port, mad_recv_wc, &msg);
1051         if (ret)
1052                 return ret;
1053
1054         /* We just need common CM header information.  Cast to any message. */
1055         rcv_msg = (struct cm_rej_msg *) mad_recv_wc->recv_buf.mad;
1056         rej_msg = (struct cm_rej_msg *) msg->mad;
1057
1058         cm_format_mad_hdr(&rej_msg->hdr, CM_REJ_ATTR_ID, rcv_msg->hdr.tid);
1059         rej_msg->remote_comm_id = rcv_msg->local_comm_id;
1060         rej_msg->local_comm_id = rcv_msg->remote_comm_id;
1061         cm_rej_set_msg_rejected(rej_msg, msg_rejected);
1062         rej_msg->reason = cpu_to_be16(reason);
1063
1064         if (ari && ari_length) {
1065                 cm_rej_set_reject_info_len(rej_msg, ari_length);
1066                 memcpy(rej_msg->ari, ari, ari_length);
1067         }
1068
1069         ret = ib_post_send_mad(msg, NULL);
1070         if (ret)
1071                 cm_free_msg(msg);
1072
1073         return ret;
1074 }
1075
1076 static inline int cm_is_active_peer(__be64 local_ca_guid, __be64 remote_ca_guid,
1077                                     __be32 local_qpn, __be32 remote_qpn)
1078 {
1079         return (be64_to_cpu(local_ca_guid) > be64_to_cpu(remote_ca_guid) ||
1080                 ((local_ca_guid == remote_ca_guid) &&
1081                  (be32_to_cpu(local_qpn) > be32_to_cpu(remote_qpn))));
1082 }
1083
1084 static void cm_format_paths_from_req(struct cm_req_msg *req_msg,
1085                                             struct ib_sa_path_rec *primary_path,
1086                                             struct ib_sa_path_rec *alt_path)
1087 {
1088         memset(primary_path, 0, sizeof *primary_path);
1089         primary_path->dgid = req_msg->primary_local_gid;
1090         primary_path->sgid = req_msg->primary_remote_gid;
1091         primary_path->dlid = req_msg->primary_local_lid;
1092         primary_path->slid = req_msg->primary_remote_lid;
1093         primary_path->flow_label = cm_req_get_primary_flow_label(req_msg);
1094         primary_path->hop_limit = req_msg->primary_hop_limit;
1095         primary_path->traffic_class = req_msg->primary_traffic_class;
1096         primary_path->reversible = 1;
1097         primary_path->pkey = req_msg->pkey;
1098         primary_path->sl = cm_req_get_primary_sl(req_msg);
1099         primary_path->mtu_selector = IB_SA_EQ;
1100         primary_path->mtu = cm_req_get_path_mtu(req_msg);
1101         primary_path->rate_selector = IB_SA_EQ;
1102         primary_path->rate = cm_req_get_primary_packet_rate(req_msg);
1103         primary_path->packet_life_time_selector = IB_SA_EQ;
1104         primary_path->packet_life_time =
1105                 cm_req_get_primary_local_ack_timeout(req_msg);
1106         primary_path->packet_life_time -= (primary_path->packet_life_time > 0);
1107
1108         if (req_msg->alt_local_lid) {
1109                 memset(alt_path, 0, sizeof *alt_path);
1110                 alt_path->dgid = req_msg->alt_local_gid;
1111                 alt_path->sgid = req_msg->alt_remote_gid;
1112                 alt_path->dlid = req_msg->alt_local_lid;
1113                 alt_path->slid = req_msg->alt_remote_lid;
1114                 alt_path->flow_label = cm_req_get_alt_flow_label(req_msg);
1115                 alt_path->hop_limit = req_msg->alt_hop_limit;
1116                 alt_path->traffic_class = req_msg->alt_traffic_class;
1117                 alt_path->reversible = 1;
1118                 alt_path->pkey = req_msg->pkey;
1119                 alt_path->sl = cm_req_get_alt_sl(req_msg);
1120                 alt_path->mtu_selector = IB_SA_EQ;
1121                 alt_path->mtu = cm_req_get_path_mtu(req_msg);
1122                 alt_path->rate_selector = IB_SA_EQ;
1123                 alt_path->rate = cm_req_get_alt_packet_rate(req_msg);
1124                 alt_path->packet_life_time_selector = IB_SA_EQ;
1125                 alt_path->packet_life_time =
1126                         cm_req_get_alt_local_ack_timeout(req_msg);
1127                 alt_path->packet_life_time -= (alt_path->packet_life_time > 0);
1128         }
1129 }
1130
1131 static void cm_format_req_event(struct cm_work *work,
1132                                 struct cm_id_private *cm_id_priv,
1133                                 struct ib_cm_id *listen_id)
1134 {
1135         struct cm_req_msg *req_msg;
1136         struct ib_cm_req_event_param *param;
1137
1138         req_msg = (struct cm_req_msg *)work->mad_recv_wc->recv_buf.mad;
1139         param = &work->cm_event.param.req_rcvd;
1140         param->listen_id = listen_id;
1141         param->port = cm_id_priv->av.port->port_num;
1142         param->primary_path = &work->path[0];
1143         if (req_msg->alt_local_lid)
1144                 param->alternate_path = &work->path[1];
1145         else
1146                 param->alternate_path = NULL;
1147         param->remote_ca_guid = req_msg->local_ca_guid;
1148         param->remote_qkey = be32_to_cpu(req_msg->local_qkey);
1149         param->remote_qpn = be32_to_cpu(cm_req_get_local_qpn(req_msg));
1150         param->qp_type = cm_req_get_qp_type(req_msg);
1151         param->starting_psn = be32_to_cpu(cm_req_get_starting_psn(req_msg));
1152         param->responder_resources = cm_req_get_init_depth(req_msg);
1153         param->initiator_depth = cm_req_get_resp_res(req_msg);
1154         param->local_cm_response_timeout =
1155                                         cm_req_get_remote_resp_timeout(req_msg);
1156         param->flow_control = cm_req_get_flow_ctrl(req_msg);
1157         param->remote_cm_response_timeout =
1158                                         cm_req_get_local_resp_timeout(req_msg);
1159         param->retry_count = cm_req_get_retry_count(req_msg);
1160         param->rnr_retry_count = cm_req_get_rnr_retry_count(req_msg);
1161         param->srq = cm_req_get_srq(req_msg);
1162         work->cm_event.private_data = &req_msg->private_data;
1163 }
1164
1165 static void cm_process_work(struct cm_id_private *cm_id_priv,
1166                             struct cm_work *work)
1167 {
1168         int ret;
1169
1170         /* We will typically only have the current event to report. */
1171         ret = cm_id_priv->id.cm_handler(&cm_id_priv->id, &work->cm_event);
1172         cm_free_work(work);
1173
1174         while (!ret && !atomic_add_negative(-1, &cm_id_priv->work_count)) {
1175                 spin_lock_irq(&cm_id_priv->lock);
1176                 work = cm_dequeue_work(cm_id_priv);
1177                 spin_unlock_irq(&cm_id_priv->lock);
1178                 BUG_ON(!work);
1179                 ret = cm_id_priv->id.cm_handler(&cm_id_priv->id,
1180                                                 &work->cm_event);
1181                 cm_free_work(work);
1182         }
1183         cm_deref_id(cm_id_priv);
1184         if (ret)
1185                 cm_destroy_id(&cm_id_priv->id, ret);
1186 }
1187
1188 static void cm_format_mra(struct cm_mra_msg *mra_msg,
1189                           struct cm_id_private *cm_id_priv,
1190                           enum cm_msg_response msg_mraed, u8 service_timeout,
1191                           const void *private_data, u8 private_data_len)
1192 {
1193         cm_format_mad_hdr(&mra_msg->hdr, CM_MRA_ATTR_ID, cm_id_priv->tid);
1194         cm_mra_set_msg_mraed(mra_msg, msg_mraed);
1195         mra_msg->local_comm_id = cm_id_priv->id.local_id;
1196         mra_msg->remote_comm_id = cm_id_priv->id.remote_id;
1197         cm_mra_set_service_timeout(mra_msg, service_timeout);
1198
1199         if (private_data && private_data_len)
1200                 memcpy(mra_msg->private_data, private_data, private_data_len);
1201 }
1202
1203 static void cm_format_rej(struct cm_rej_msg *rej_msg,
1204                           struct cm_id_private *cm_id_priv,
1205                           enum ib_cm_rej_reason reason,
1206                           void *ari,
1207                           u8 ari_length,
1208                           const void *private_data,
1209                           u8 private_data_len)
1210 {
1211         cm_format_mad_hdr(&rej_msg->hdr, CM_REJ_ATTR_ID, cm_id_priv->tid);
1212         rej_msg->remote_comm_id = cm_id_priv->id.remote_id;
1213
1214         switch(cm_id_priv->id.state) {
1215         case IB_CM_REQ_RCVD:
1216                 rej_msg->local_comm_id = 0;
1217                 cm_rej_set_msg_rejected(rej_msg, CM_MSG_RESPONSE_REQ);
1218                 break;
1219         case IB_CM_MRA_REQ_SENT:
1220                 rej_msg->local_comm_id = cm_id_priv->id.local_id;
1221                 cm_rej_set_msg_rejected(rej_msg, CM_MSG_RESPONSE_REQ);
1222                 break;
1223         case IB_CM_REP_RCVD:
1224         case IB_CM_MRA_REP_SENT:
1225                 rej_msg->local_comm_id = cm_id_priv->id.local_id;
1226                 cm_rej_set_msg_rejected(rej_msg, CM_MSG_RESPONSE_REP);
1227                 break;
1228         default:
1229                 rej_msg->local_comm_id = cm_id_priv->id.local_id;
1230                 cm_rej_set_msg_rejected(rej_msg, CM_MSG_RESPONSE_OTHER);
1231                 break;
1232         }
1233
1234         rej_msg->reason = cpu_to_be16(reason);
1235         if (ari && ari_length) {
1236                 cm_rej_set_reject_info_len(rej_msg, ari_length);
1237                 memcpy(rej_msg->ari, ari, ari_length);
1238         }
1239
1240         if (private_data && private_data_len)
1241                 memcpy(rej_msg->private_data, private_data, private_data_len);
1242 }
1243
1244 static void cm_dup_req_handler(struct cm_work *work,
1245                                struct cm_id_private *cm_id_priv)
1246 {
1247         struct ib_mad_send_buf *msg = NULL;
1248         int ret;
1249
1250         /* Quick state check to discard duplicate REQs. */
1251         if (cm_id_priv->id.state == IB_CM_REQ_RCVD)
1252                 return;
1253
1254         ret = cm_alloc_response_msg(work->port, work->mad_recv_wc, &msg);
1255         if (ret)
1256                 return;
1257
1258         spin_lock_irq(&cm_id_priv->lock);
1259         switch (cm_id_priv->id.state) {
1260         case IB_CM_MRA_REQ_SENT:
1261                 cm_format_mra((struct cm_mra_msg *) msg->mad, cm_id_priv,
1262                               CM_MSG_RESPONSE_REQ, cm_id_priv->service_timeout,
1263                               cm_id_priv->private_data,
1264                               cm_id_priv->private_data_len);
1265                 break;
1266         case IB_CM_TIMEWAIT:
1267                 cm_format_rej((struct cm_rej_msg *) msg->mad, cm_id_priv,
1268                               IB_CM_REJ_STALE_CONN, NULL, 0, NULL, 0);
1269                 break;
1270         default:
1271                 goto unlock;
1272         }
1273         spin_unlock_irq(&cm_id_priv->lock);
1274
1275         ret = ib_post_send_mad(msg, NULL);
1276         if (ret)
1277                 goto free;
1278         return;
1279
1280 unlock: spin_unlock_irq(&cm_id_priv->lock);
1281 free:   cm_free_msg(msg);
1282 }
1283
1284 static struct cm_id_private * cm_match_req(struct cm_work *work,
1285                                            struct cm_id_private *cm_id_priv)
1286 {
1287         struct cm_id_private *listen_cm_id_priv, *cur_cm_id_priv;
1288         struct cm_timewait_info *timewait_info;
1289         struct cm_req_msg *req_msg;
1290
1291         req_msg = (struct cm_req_msg *)work->mad_recv_wc->recv_buf.mad;
1292
1293         /* Check for possible duplicate REQ. */
1294         spin_lock_irq(&cm.lock);
1295         timewait_info = cm_insert_remote_id(cm_id_priv->timewait_info);
1296         if (timewait_info) {
1297                 cur_cm_id_priv = cm_get_id(timewait_info->work.local_id,
1298                                            timewait_info->work.remote_id);
1299                 spin_unlock_irq(&cm.lock);
1300                 if (cur_cm_id_priv) {
1301                         cm_dup_req_handler(work, cur_cm_id_priv);
1302                         cm_deref_id(cur_cm_id_priv);
1303                 }
1304                 return NULL;
1305         }
1306
1307         /* Check for stale connections. */
1308         timewait_info = cm_insert_remote_qpn(cm_id_priv->timewait_info);
1309         if (timewait_info) {
1310                 cm_cleanup_timewait(cm_id_priv->timewait_info);
1311                 spin_unlock_irq(&cm.lock);
1312                 cm_issue_rej(work->port, work->mad_recv_wc,
1313                              IB_CM_REJ_STALE_CONN, CM_MSG_RESPONSE_REQ,
1314                              NULL, 0);
1315                 return NULL;
1316         }
1317
1318         /* Find matching listen request. */
1319         listen_cm_id_priv = cm_find_listen(cm_id_priv->id.device,
1320                                            req_msg->service_id,
1321                                            req_msg->private_data);
1322         if (!listen_cm_id_priv) {
1323                 cm_cleanup_timewait(cm_id_priv->timewait_info);
1324                 spin_unlock_irq(&cm.lock);
1325                 cm_issue_rej(work->port, work->mad_recv_wc,
1326                              IB_CM_REJ_INVALID_SERVICE_ID, CM_MSG_RESPONSE_REQ,
1327                              NULL, 0);
1328                 goto out;
1329         }
1330         atomic_inc(&listen_cm_id_priv->refcount);
1331         atomic_inc(&cm_id_priv->refcount);
1332         cm_id_priv->id.state = IB_CM_REQ_RCVD;
1333         atomic_inc(&cm_id_priv->work_count);
1334         spin_unlock_irq(&cm.lock);
1335 out:
1336         return listen_cm_id_priv;
1337 }
1338
1339 static int cm_req_handler(struct cm_work *work)
1340 {
1341         struct ib_cm_id *cm_id;
1342         struct cm_id_private *cm_id_priv, *listen_cm_id_priv;
1343         struct cm_req_msg *req_msg;
1344         int ret;
1345
1346         req_msg = (struct cm_req_msg *)work->mad_recv_wc->recv_buf.mad;
1347
1348         cm_id = ib_create_cm_id(work->port->cm_dev->device, NULL, NULL);
1349         if (IS_ERR(cm_id))
1350                 return PTR_ERR(cm_id);
1351
1352         cm_id_priv = container_of(cm_id, struct cm_id_private, id);
1353         cm_id_priv->id.remote_id = req_msg->local_comm_id;
1354         cm_init_av_for_response(work->port, work->mad_recv_wc->wc,
1355                                 work->mad_recv_wc->recv_buf.grh,
1356                                 &cm_id_priv->av);
1357         cm_id_priv->timewait_info = cm_create_timewait_info(cm_id_priv->
1358                                                             id.local_id);
1359         if (IS_ERR(cm_id_priv->timewait_info)) {
1360                 ret = PTR_ERR(cm_id_priv->timewait_info);
1361                 goto destroy;
1362         }
1363         cm_id_priv->timewait_info->work.remote_id = req_msg->local_comm_id;
1364         cm_id_priv->timewait_info->remote_ca_guid = req_msg->local_ca_guid;
1365         cm_id_priv->timewait_info->remote_qpn = cm_req_get_local_qpn(req_msg);
1366
1367         listen_cm_id_priv = cm_match_req(work, cm_id_priv);
1368         if (!listen_cm_id_priv) {
1369                 ret = -EINVAL;
1370                 kfree(cm_id_priv->timewait_info);
1371                 goto destroy;
1372         }
1373
1374         cm_id_priv->id.cm_handler = listen_cm_id_priv->id.cm_handler;
1375         cm_id_priv->id.context = listen_cm_id_priv->id.context;
1376         cm_id_priv->id.service_id = req_msg->service_id;
1377         cm_id_priv->id.service_mask = __constant_cpu_to_be64(~0ULL);
1378
1379         cm_format_paths_from_req(req_msg, &work->path[0], &work->path[1]);
1380         ret = cm_init_av_by_path(&work->path[0], &cm_id_priv->av);
1381         if (ret) {
1382                 ib_get_cached_gid(work->port->cm_dev->device,
1383                                   work->port->port_num, 0, &work->path[0].sgid);
1384                 ib_send_cm_rej(cm_id, IB_CM_REJ_INVALID_GID,
1385                                &work->path[0].sgid, sizeof work->path[0].sgid,
1386                                NULL, 0);
1387                 goto rejected;
1388         }
1389         if (req_msg->alt_local_lid) {
1390                 ret = cm_init_av_by_path(&work->path[1], &cm_id_priv->alt_av);
1391                 if (ret) {
1392                         ib_send_cm_rej(cm_id, IB_CM_REJ_INVALID_ALT_GID,
1393                                        &work->path[0].sgid,
1394                                        sizeof work->path[0].sgid, NULL, 0);
1395                         goto rejected;
1396                 }
1397         }
1398         cm_id_priv->tid = req_msg->hdr.tid;
1399         cm_id_priv->timeout_ms = cm_convert_to_ms(
1400                                         cm_req_get_local_resp_timeout(req_msg));
1401         cm_id_priv->max_cm_retries = cm_req_get_max_cm_retries(req_msg);
1402         cm_id_priv->remote_qpn = cm_req_get_local_qpn(req_msg);
1403         cm_id_priv->initiator_depth = cm_req_get_resp_res(req_msg);
1404         cm_id_priv->responder_resources = cm_req_get_init_depth(req_msg);
1405         cm_id_priv->path_mtu = cm_req_get_path_mtu(req_msg);
1406         cm_id_priv->pkey = req_msg->pkey;
1407         cm_id_priv->sq_psn = cm_req_get_starting_psn(req_msg);
1408         cm_id_priv->retry_count = cm_req_get_retry_count(req_msg);
1409         cm_id_priv->rnr_retry_count = cm_req_get_rnr_retry_count(req_msg);
1410         cm_id_priv->qp_type = cm_req_get_qp_type(req_msg);
1411
1412         cm_format_req_event(work, cm_id_priv, &listen_cm_id_priv->id);
1413         cm_process_work(cm_id_priv, work);
1414         cm_deref_id(listen_cm_id_priv);
1415         return 0;
1416
1417 rejected:
1418         atomic_dec(&cm_id_priv->refcount);
1419         cm_deref_id(listen_cm_id_priv);
1420 destroy:
1421         ib_destroy_cm_id(cm_id);
1422         return ret;
1423 }
1424
1425 static void cm_format_rep(struct cm_rep_msg *rep_msg,
1426                           struct cm_id_private *cm_id_priv,
1427                           struct ib_cm_rep_param *param)
1428 {
1429         cm_format_mad_hdr(&rep_msg->hdr, CM_REP_ATTR_ID, cm_id_priv->tid);
1430         rep_msg->local_comm_id = cm_id_priv->id.local_id;
1431         rep_msg->remote_comm_id = cm_id_priv->id.remote_id;
1432         cm_rep_set_local_qpn(rep_msg, cpu_to_be32(param->qp_num));
1433         cm_rep_set_starting_psn(rep_msg, cpu_to_be32(param->starting_psn));
1434         rep_msg->resp_resources = param->responder_resources;
1435         rep_msg->initiator_depth = param->initiator_depth;
1436         cm_rep_set_target_ack_delay(rep_msg, param->target_ack_delay);
1437         cm_rep_set_failover(rep_msg, param->failover_accepted);
1438         cm_rep_set_flow_ctrl(rep_msg, param->flow_control);
1439         cm_rep_set_rnr_retry_count(rep_msg, param->rnr_retry_count);
1440         cm_rep_set_srq(rep_msg, param->srq);
1441         rep_msg->local_ca_guid = cm_id_priv->id.device->node_guid;
1442
1443         if (param->private_data && param->private_data_len)
1444                 memcpy(rep_msg->private_data, param->private_data,
1445                        param->private_data_len);
1446 }
1447
1448 int ib_send_cm_rep(struct ib_cm_id *cm_id,
1449                    struct ib_cm_rep_param *param)
1450 {
1451         struct cm_id_private *cm_id_priv;
1452         struct ib_mad_send_buf *msg;
1453         struct cm_rep_msg *rep_msg;
1454         unsigned long flags;
1455         int ret;
1456
1457         if (param->private_data &&
1458             param->private_data_len > IB_CM_REP_PRIVATE_DATA_SIZE)
1459                 return -EINVAL;
1460
1461         cm_id_priv = container_of(cm_id, struct cm_id_private, id);
1462         spin_lock_irqsave(&cm_id_priv->lock, flags);
1463         if (cm_id->state != IB_CM_REQ_RCVD &&
1464             cm_id->state != IB_CM_MRA_REQ_SENT) {
1465                 ret = -EINVAL;
1466                 goto out;
1467         }
1468
1469         ret = cm_alloc_msg(cm_id_priv, &msg);
1470         if (ret)
1471                 goto out;
1472
1473         rep_msg = (struct cm_rep_msg *) msg->mad;
1474         cm_format_rep(rep_msg, cm_id_priv, param);
1475         msg->timeout_ms = cm_id_priv->timeout_ms;
1476         msg->context[1] = (void *) (unsigned long) IB_CM_REP_SENT;
1477
1478         ret = ib_post_send_mad(msg, NULL);
1479         if (ret) {
1480                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1481                 cm_free_msg(msg);
1482                 return ret;
1483         }
1484
1485         cm_id->state = IB_CM_REP_SENT;
1486         cm_id_priv->msg = msg;
1487         cm_id_priv->initiator_depth = param->initiator_depth;
1488         cm_id_priv->responder_resources = param->responder_resources;
1489         cm_id_priv->rq_psn = cm_rep_get_starting_psn(rep_msg);
1490         cm_id_priv->local_qpn = cm_rep_get_local_qpn(rep_msg);
1491
1492 out:    spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1493         return ret;
1494 }
1495 EXPORT_SYMBOL(ib_send_cm_rep);
1496
1497 static void cm_format_rtu(struct cm_rtu_msg *rtu_msg,
1498                           struct cm_id_private *cm_id_priv,
1499                           const void *private_data,
1500                           u8 private_data_len)
1501 {
1502         cm_format_mad_hdr(&rtu_msg->hdr, CM_RTU_ATTR_ID, cm_id_priv->tid);
1503         rtu_msg->local_comm_id = cm_id_priv->id.local_id;
1504         rtu_msg->remote_comm_id = cm_id_priv->id.remote_id;
1505
1506         if (private_data && private_data_len)
1507                 memcpy(rtu_msg->private_data, private_data, private_data_len);
1508 }
1509
1510 int ib_send_cm_rtu(struct ib_cm_id *cm_id,
1511                    const void *private_data,
1512                    u8 private_data_len)
1513 {
1514         struct cm_id_private *cm_id_priv;
1515         struct ib_mad_send_buf *msg;
1516         unsigned long flags;
1517         void *data;
1518         int ret;
1519
1520         if (private_data && private_data_len > IB_CM_RTU_PRIVATE_DATA_SIZE)
1521                 return -EINVAL;
1522
1523         data = cm_copy_private_data(private_data, private_data_len);
1524         if (IS_ERR(data))
1525                 return PTR_ERR(data);
1526
1527         cm_id_priv = container_of(cm_id, struct cm_id_private, id);
1528         spin_lock_irqsave(&cm_id_priv->lock, flags);
1529         if (cm_id->state != IB_CM_REP_RCVD &&
1530             cm_id->state != IB_CM_MRA_REP_SENT) {
1531                 ret = -EINVAL;
1532                 goto error;
1533         }
1534
1535         ret = cm_alloc_msg(cm_id_priv, &msg);
1536         if (ret)
1537                 goto error;
1538
1539         cm_format_rtu((struct cm_rtu_msg *) msg->mad, cm_id_priv,
1540                       private_data, private_data_len);
1541
1542         ret = ib_post_send_mad(msg, NULL);
1543         if (ret) {
1544                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1545                 cm_free_msg(msg);
1546                 kfree(data);
1547                 return ret;
1548         }
1549
1550         cm_id->state = IB_CM_ESTABLISHED;
1551         cm_set_private_data(cm_id_priv, data, private_data_len);
1552         spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1553         return 0;
1554
1555 error:  spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1556         kfree(data);
1557         return ret;
1558 }
1559 EXPORT_SYMBOL(ib_send_cm_rtu);
1560
1561 static void cm_format_rep_event(struct cm_work *work)
1562 {
1563         struct cm_rep_msg *rep_msg;
1564         struct ib_cm_rep_event_param *param;
1565
1566         rep_msg = (struct cm_rep_msg *)work->mad_recv_wc->recv_buf.mad;
1567         param = &work->cm_event.param.rep_rcvd;
1568         param->remote_ca_guid = rep_msg->local_ca_guid;
1569         param->remote_qkey = be32_to_cpu(rep_msg->local_qkey);
1570         param->remote_qpn = be32_to_cpu(cm_rep_get_local_qpn(rep_msg));
1571         param->starting_psn = be32_to_cpu(cm_rep_get_starting_psn(rep_msg));
1572         param->responder_resources = rep_msg->initiator_depth;
1573         param->initiator_depth = rep_msg->resp_resources;
1574         param->target_ack_delay = cm_rep_get_target_ack_delay(rep_msg);
1575         param->failover_accepted = cm_rep_get_failover(rep_msg);
1576         param->flow_control = cm_rep_get_flow_ctrl(rep_msg);
1577         param->rnr_retry_count = cm_rep_get_rnr_retry_count(rep_msg);
1578         param->srq = cm_rep_get_srq(rep_msg);
1579         work->cm_event.private_data = &rep_msg->private_data;
1580 }
1581
1582 static void cm_dup_rep_handler(struct cm_work *work)
1583 {
1584         struct cm_id_private *cm_id_priv;
1585         struct cm_rep_msg *rep_msg;
1586         struct ib_mad_send_buf *msg = NULL;
1587         int ret;
1588
1589         rep_msg = (struct cm_rep_msg *) work->mad_recv_wc->recv_buf.mad;
1590         cm_id_priv = cm_acquire_id(rep_msg->remote_comm_id,
1591                                    rep_msg->local_comm_id);
1592         if (!cm_id_priv)
1593                 return;
1594
1595         ret = cm_alloc_response_msg(work->port, work->mad_recv_wc, &msg);
1596         if (ret)
1597                 goto deref;
1598
1599         spin_lock_irq(&cm_id_priv->lock);
1600         if (cm_id_priv->id.state == IB_CM_ESTABLISHED)
1601                 cm_format_rtu((struct cm_rtu_msg *) msg->mad, cm_id_priv,
1602                               cm_id_priv->private_data,
1603                               cm_id_priv->private_data_len);
1604         else if (cm_id_priv->id.state == IB_CM_MRA_REP_SENT)
1605                 cm_format_mra((struct cm_mra_msg *) msg->mad, cm_id_priv,
1606                               CM_MSG_RESPONSE_REP, cm_id_priv->service_timeout,
1607                               cm_id_priv->private_data,
1608                               cm_id_priv->private_data_len);
1609         else
1610                 goto unlock;
1611         spin_unlock_irq(&cm_id_priv->lock);
1612
1613         ret = ib_post_send_mad(msg, NULL);
1614         if (ret)
1615                 goto free;
1616         goto deref;
1617
1618 unlock: spin_unlock_irq(&cm_id_priv->lock);
1619 free:   cm_free_msg(msg);
1620 deref:  cm_deref_id(cm_id_priv);
1621 }
1622
1623 static int cm_rep_handler(struct cm_work *work)
1624 {
1625         struct cm_id_private *cm_id_priv;
1626         struct cm_rep_msg *rep_msg;
1627         int ret;
1628
1629         rep_msg = (struct cm_rep_msg *)work->mad_recv_wc->recv_buf.mad;
1630         cm_id_priv = cm_acquire_id(rep_msg->remote_comm_id, 0);
1631         if (!cm_id_priv) {
1632                 cm_dup_rep_handler(work);
1633                 return -EINVAL;
1634         }
1635
1636         cm_format_rep_event(work);
1637
1638         spin_lock_irq(&cm_id_priv->lock);
1639         switch (cm_id_priv->id.state) {
1640         case IB_CM_REQ_SENT:
1641         case IB_CM_MRA_REQ_RCVD:
1642                 break;
1643         default:
1644                 spin_unlock_irq(&cm_id_priv->lock);
1645                 ret = -EINVAL;
1646                 goto error;
1647         }
1648
1649         cm_id_priv->timewait_info->work.remote_id = rep_msg->local_comm_id;
1650         cm_id_priv->timewait_info->remote_ca_guid = rep_msg->local_ca_guid;
1651         cm_id_priv->timewait_info->remote_qpn = cm_rep_get_local_qpn(rep_msg);
1652
1653         spin_lock(&cm.lock);
1654         /* Check for duplicate REP. */
1655         if (cm_insert_remote_id(cm_id_priv->timewait_info)) {
1656                 spin_unlock(&cm.lock);
1657                 spin_unlock_irq(&cm_id_priv->lock);
1658                 ret = -EINVAL;
1659                 goto error;
1660         }
1661         /* Check for a stale connection. */
1662         if (cm_insert_remote_qpn(cm_id_priv->timewait_info)) {
1663                 rb_erase(&cm_id_priv->timewait_info->remote_id_node,
1664                          &cm.remote_id_table);
1665                 cm_id_priv->timewait_info->inserted_remote_id = 0;
1666                 spin_unlock(&cm.lock);
1667                 spin_unlock_irq(&cm_id_priv->lock);
1668                 cm_issue_rej(work->port, work->mad_recv_wc,
1669                              IB_CM_REJ_STALE_CONN, CM_MSG_RESPONSE_REP,
1670                              NULL, 0);
1671                 ret = -EINVAL;
1672                 goto error;
1673         }
1674         spin_unlock(&cm.lock);
1675
1676         cm_id_priv->id.state = IB_CM_REP_RCVD;
1677         cm_id_priv->id.remote_id = rep_msg->local_comm_id;
1678         cm_id_priv->remote_qpn = cm_rep_get_local_qpn(rep_msg);
1679         cm_id_priv->initiator_depth = rep_msg->resp_resources;
1680         cm_id_priv->responder_resources = rep_msg->initiator_depth;
1681         cm_id_priv->sq_psn = cm_rep_get_starting_psn(rep_msg);
1682         cm_id_priv->rnr_retry_count = cm_rep_get_rnr_retry_count(rep_msg);
1683
1684         /* todo: handle peer_to_peer */
1685
1686         ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg);
1687         ret = atomic_inc_and_test(&cm_id_priv->work_count);
1688         if (!ret)
1689                 list_add_tail(&work->list, &cm_id_priv->work_list);
1690         spin_unlock_irq(&cm_id_priv->lock);
1691
1692         if (ret)
1693                 cm_process_work(cm_id_priv, work);
1694         else
1695                 cm_deref_id(cm_id_priv);
1696         return 0;
1697
1698 error:
1699         cm_deref_id(cm_id_priv);
1700         return ret;
1701 }
1702
1703 static int cm_establish_handler(struct cm_work *work)
1704 {
1705         struct cm_id_private *cm_id_priv;
1706         int ret;
1707
1708         /* See comment in cm_establish about lookup. */
1709         cm_id_priv = cm_acquire_id(work->local_id, work->remote_id);
1710         if (!cm_id_priv)
1711                 return -EINVAL;
1712
1713         spin_lock_irq(&cm_id_priv->lock);
1714         if (cm_id_priv->id.state != IB_CM_ESTABLISHED) {
1715                 spin_unlock_irq(&cm_id_priv->lock);
1716                 goto out;
1717         }
1718
1719         ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg);
1720         ret = atomic_inc_and_test(&cm_id_priv->work_count);
1721         if (!ret)
1722                 list_add_tail(&work->list, &cm_id_priv->work_list);
1723         spin_unlock_irq(&cm_id_priv->lock);
1724
1725         if (ret)
1726                 cm_process_work(cm_id_priv, work);
1727         else
1728                 cm_deref_id(cm_id_priv);
1729         return 0;
1730 out:
1731         cm_deref_id(cm_id_priv);
1732         return -EINVAL;
1733 }
1734
1735 static int cm_rtu_handler(struct cm_work *work)
1736 {
1737         struct cm_id_private *cm_id_priv;
1738         struct cm_rtu_msg *rtu_msg;
1739         int ret;
1740
1741         rtu_msg = (struct cm_rtu_msg *)work->mad_recv_wc->recv_buf.mad;
1742         cm_id_priv = cm_acquire_id(rtu_msg->remote_comm_id,
1743                                    rtu_msg->local_comm_id);
1744         if (!cm_id_priv)
1745                 return -EINVAL;
1746
1747         work->cm_event.private_data = &rtu_msg->private_data;
1748
1749         spin_lock_irq(&cm_id_priv->lock);
1750         if (cm_id_priv->id.state != IB_CM_REP_SENT &&
1751             cm_id_priv->id.state != IB_CM_MRA_REP_RCVD) {
1752                 spin_unlock_irq(&cm_id_priv->lock);
1753                 goto out;
1754         }
1755         cm_id_priv->id.state = IB_CM_ESTABLISHED;
1756
1757         ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg);
1758         ret = atomic_inc_and_test(&cm_id_priv->work_count);
1759         if (!ret)
1760                 list_add_tail(&work->list, &cm_id_priv->work_list);
1761         spin_unlock_irq(&cm_id_priv->lock);
1762
1763         if (ret)
1764                 cm_process_work(cm_id_priv, work);
1765         else
1766                 cm_deref_id(cm_id_priv);
1767         return 0;
1768 out:
1769         cm_deref_id(cm_id_priv);
1770         return -EINVAL;
1771 }
1772
1773 static void cm_format_dreq(struct cm_dreq_msg *dreq_msg,
1774                           struct cm_id_private *cm_id_priv,
1775                           const void *private_data,
1776                           u8 private_data_len)
1777 {
1778         cm_format_mad_hdr(&dreq_msg->hdr, CM_DREQ_ATTR_ID,
1779                           cm_form_tid(cm_id_priv, CM_MSG_SEQUENCE_DREQ));
1780         dreq_msg->local_comm_id = cm_id_priv->id.local_id;
1781         dreq_msg->remote_comm_id = cm_id_priv->id.remote_id;
1782         cm_dreq_set_remote_qpn(dreq_msg, cm_id_priv->remote_qpn);
1783
1784         if (private_data && private_data_len)
1785                 memcpy(dreq_msg->private_data, private_data, private_data_len);
1786 }
1787
1788 int ib_send_cm_dreq(struct ib_cm_id *cm_id,
1789                     const void *private_data,
1790                     u8 private_data_len)
1791 {
1792         struct cm_id_private *cm_id_priv;
1793         struct ib_mad_send_buf *msg;
1794         unsigned long flags;
1795         int ret;
1796
1797         if (private_data && private_data_len > IB_CM_DREQ_PRIVATE_DATA_SIZE)
1798                 return -EINVAL;
1799
1800         cm_id_priv = container_of(cm_id, struct cm_id_private, id);
1801         spin_lock_irqsave(&cm_id_priv->lock, flags);
1802         if (cm_id->state != IB_CM_ESTABLISHED) {
1803                 ret = -EINVAL;
1804                 goto out;
1805         }
1806
1807         ret = cm_alloc_msg(cm_id_priv, &msg);
1808         if (ret) {
1809                 cm_enter_timewait(cm_id_priv);
1810                 goto out;
1811         }
1812
1813         cm_format_dreq((struct cm_dreq_msg *) msg->mad, cm_id_priv,
1814                        private_data, private_data_len);
1815         msg->timeout_ms = cm_id_priv->timeout_ms;
1816         msg->context[1] = (void *) (unsigned long) IB_CM_DREQ_SENT;
1817
1818         ret = ib_post_send_mad(msg, NULL);
1819         if (ret) {
1820                 cm_enter_timewait(cm_id_priv);
1821                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1822                 cm_free_msg(msg);
1823                 return ret;
1824         }
1825
1826         cm_id->state = IB_CM_DREQ_SENT;
1827         cm_id_priv->msg = msg;
1828 out:    spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1829         return ret;
1830 }
1831 EXPORT_SYMBOL(ib_send_cm_dreq);
1832
1833 static void cm_format_drep(struct cm_drep_msg *drep_msg,
1834                           struct cm_id_private *cm_id_priv,
1835                           const void *private_data,
1836                           u8 private_data_len)
1837 {
1838         cm_format_mad_hdr(&drep_msg->hdr, CM_DREP_ATTR_ID, cm_id_priv->tid);
1839         drep_msg->local_comm_id = cm_id_priv->id.local_id;
1840         drep_msg->remote_comm_id = cm_id_priv->id.remote_id;
1841
1842         if (private_data && private_data_len)
1843                 memcpy(drep_msg->private_data, private_data, private_data_len);
1844 }
1845
1846 int ib_send_cm_drep(struct ib_cm_id *cm_id,
1847                     const void *private_data,
1848                     u8 private_data_len)
1849 {
1850         struct cm_id_private *cm_id_priv;
1851         struct ib_mad_send_buf *msg;
1852         unsigned long flags;
1853         void *data;
1854         int ret;
1855
1856         if (private_data && private_data_len > IB_CM_DREP_PRIVATE_DATA_SIZE)
1857                 return -EINVAL;
1858
1859         data = cm_copy_private_data(private_data, private_data_len);
1860         if (IS_ERR(data))
1861                 return PTR_ERR(data);
1862
1863         cm_id_priv = container_of(cm_id, struct cm_id_private, id);
1864         spin_lock_irqsave(&cm_id_priv->lock, flags);
1865         if (cm_id->state != IB_CM_DREQ_RCVD) {
1866                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1867                 kfree(data);
1868                 return -EINVAL;
1869         }
1870
1871         cm_set_private_data(cm_id_priv, data, private_data_len);
1872         cm_enter_timewait(cm_id_priv);
1873
1874         ret = cm_alloc_msg(cm_id_priv, &msg);
1875         if (ret)
1876                 goto out;
1877
1878         cm_format_drep((struct cm_drep_msg *) msg->mad, cm_id_priv,
1879                        private_data, private_data_len);
1880
1881         ret = ib_post_send_mad(msg, NULL);
1882         if (ret) {
1883                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1884                 cm_free_msg(msg);
1885                 return ret;
1886         }
1887
1888 out:    spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1889         return ret;
1890 }
1891 EXPORT_SYMBOL(ib_send_cm_drep);
1892
1893 static int cm_issue_drep(struct cm_port *port,
1894                          struct ib_mad_recv_wc *mad_recv_wc)
1895 {
1896         struct ib_mad_send_buf *msg = NULL;
1897         struct cm_dreq_msg *dreq_msg;
1898         struct cm_drep_msg *drep_msg;
1899         int ret;
1900
1901         ret = cm_alloc_response_msg(port, mad_recv_wc, &msg);
1902         if (ret)
1903                 return ret;
1904
1905         dreq_msg = (struct cm_dreq_msg *) mad_recv_wc->recv_buf.mad;
1906         drep_msg = (struct cm_drep_msg *) msg->mad;
1907
1908         cm_format_mad_hdr(&drep_msg->hdr, CM_DREP_ATTR_ID, dreq_msg->hdr.tid);
1909         drep_msg->remote_comm_id = dreq_msg->local_comm_id;
1910         drep_msg->local_comm_id = dreq_msg->remote_comm_id;
1911
1912         ret = ib_post_send_mad(msg, NULL);
1913         if (ret)
1914                 cm_free_msg(msg);
1915
1916         return ret;
1917 }
1918
1919 static int cm_dreq_handler(struct cm_work *work)
1920 {
1921         struct cm_id_private *cm_id_priv;
1922         struct cm_dreq_msg *dreq_msg;
1923         struct ib_mad_send_buf *msg = NULL;
1924         int ret;
1925
1926         dreq_msg = (struct cm_dreq_msg *)work->mad_recv_wc->recv_buf.mad;
1927         cm_id_priv = cm_acquire_id(dreq_msg->remote_comm_id,
1928                                    dreq_msg->local_comm_id);
1929         if (!cm_id_priv) {
1930                 cm_issue_drep(work->port, work->mad_recv_wc);
1931                 return -EINVAL;
1932         }
1933
1934         work->cm_event.private_data = &dreq_msg->private_data;
1935
1936         spin_lock_irq(&cm_id_priv->lock);
1937         if (cm_id_priv->local_qpn != cm_dreq_get_remote_qpn(dreq_msg))
1938                 goto unlock;
1939
1940         switch (cm_id_priv->id.state) {
1941         case IB_CM_REP_SENT:
1942         case IB_CM_DREQ_SENT:
1943                 ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg);
1944                 break;
1945         case IB_CM_ESTABLISHED:
1946         case IB_CM_MRA_REP_RCVD:
1947                 break;
1948         case IB_CM_TIMEWAIT:
1949                 if (cm_alloc_response_msg(work->port, work->mad_recv_wc, &msg))
1950                         goto unlock;
1951
1952                 cm_format_drep((struct cm_drep_msg *) msg->mad, cm_id_priv,
1953                                cm_id_priv->private_data,
1954                                cm_id_priv->private_data_len);
1955                 spin_unlock_irq(&cm_id_priv->lock);
1956
1957                 if (ib_post_send_mad(msg, NULL))
1958                         cm_free_msg(msg);
1959                 goto deref;
1960         default:
1961                 goto unlock;
1962         }
1963         cm_id_priv->id.state = IB_CM_DREQ_RCVD;
1964         cm_id_priv->tid = dreq_msg->hdr.tid;
1965         ret = atomic_inc_and_test(&cm_id_priv->work_count);
1966         if (!ret)
1967                 list_add_tail(&work->list, &cm_id_priv->work_list);
1968         spin_unlock_irq(&cm_id_priv->lock);
1969
1970         if (ret)
1971                 cm_process_work(cm_id_priv, work);
1972         else
1973                 cm_deref_id(cm_id_priv);
1974         return 0;
1975
1976 unlock: spin_unlock_irq(&cm_id_priv->lock);
1977 deref:  cm_deref_id(cm_id_priv);
1978         return -EINVAL;
1979 }
1980
1981 static int cm_drep_handler(struct cm_work *work)
1982 {
1983         struct cm_id_private *cm_id_priv;
1984         struct cm_drep_msg *drep_msg;
1985         int ret;
1986
1987         drep_msg = (struct cm_drep_msg *)work->mad_recv_wc->recv_buf.mad;
1988         cm_id_priv = cm_acquire_id(drep_msg->remote_comm_id,
1989                                    drep_msg->local_comm_id);
1990         if (!cm_id_priv)
1991                 return -EINVAL;
1992
1993         work->cm_event.private_data = &drep_msg->private_data;
1994
1995         spin_lock_irq(&cm_id_priv->lock);
1996         if (cm_id_priv->id.state != IB_CM_DREQ_SENT &&
1997             cm_id_priv->id.state != IB_CM_DREQ_RCVD) {
1998                 spin_unlock_irq(&cm_id_priv->lock);
1999                 goto out;
2000         }
2001         cm_enter_timewait(cm_id_priv);
2002
2003         ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg);
2004         ret = atomic_inc_and_test(&cm_id_priv->work_count);
2005         if (!ret)
2006                 list_add_tail(&work->list, &cm_id_priv->work_list);
2007         spin_unlock_irq(&cm_id_priv->lock);
2008
2009         if (ret)
2010                 cm_process_work(cm_id_priv, work);
2011         else
2012                 cm_deref_id(cm_id_priv);
2013         return 0;
2014 out:
2015         cm_deref_id(cm_id_priv);
2016         return -EINVAL;
2017 }
2018
2019 int ib_send_cm_rej(struct ib_cm_id *cm_id,
2020                    enum ib_cm_rej_reason reason,
2021                    void *ari,
2022                    u8 ari_length,
2023                    const void *private_data,
2024                    u8 private_data_len)
2025 {
2026         struct cm_id_private *cm_id_priv;
2027         struct ib_mad_send_buf *msg;
2028         unsigned long flags;
2029         int ret;
2030
2031         if ((private_data && private_data_len > IB_CM_REJ_PRIVATE_DATA_SIZE) ||
2032             (ari && ari_length > IB_CM_REJ_ARI_LENGTH))
2033                 return -EINVAL;
2034
2035         cm_id_priv = container_of(cm_id, struct cm_id_private, id);
2036
2037         spin_lock_irqsave(&cm_id_priv->lock, flags);
2038         switch (cm_id->state) {
2039         case IB_CM_REQ_SENT:
2040         case IB_CM_MRA_REQ_RCVD:
2041         case IB_CM_REQ_RCVD:
2042         case IB_CM_MRA_REQ_SENT:
2043         case IB_CM_REP_RCVD:
2044         case IB_CM_MRA_REP_SENT:
2045                 ret = cm_alloc_msg(cm_id_priv, &msg);
2046                 if (!ret)
2047                         cm_format_rej((struct cm_rej_msg *) msg->mad,
2048                                       cm_id_priv, reason, ari, ari_length,
2049                                       private_data, private_data_len);
2050
2051                 cm_reset_to_idle(cm_id_priv);
2052                 break;
2053         case IB_CM_REP_SENT:
2054         case IB_CM_MRA_REP_RCVD:
2055                 ret = cm_alloc_msg(cm_id_priv, &msg);
2056                 if (!ret)
2057                         cm_format_rej((struct cm_rej_msg *) msg->mad,
2058                                       cm_id_priv, reason, ari, ari_length,
2059                                       private_data, private_data_len);
2060
2061                 cm_enter_timewait(cm_id_priv);
2062                 break;
2063         default:
2064                 ret = -EINVAL;
2065                 goto out;
2066         }
2067
2068         if (ret)
2069                 goto out;
2070
2071         ret = ib_post_send_mad(msg, NULL);
2072         if (ret)
2073                 cm_free_msg(msg);
2074
2075 out:    spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2076         return ret;
2077 }
2078 EXPORT_SYMBOL(ib_send_cm_rej);
2079
2080 static void cm_format_rej_event(struct cm_work *work)
2081 {
2082         struct cm_rej_msg *rej_msg;
2083         struct ib_cm_rej_event_param *param;
2084
2085         rej_msg = (struct cm_rej_msg *)work->mad_recv_wc->recv_buf.mad;
2086         param = &work->cm_event.param.rej_rcvd;
2087         param->ari = rej_msg->ari;
2088         param->ari_length = cm_rej_get_reject_info_len(rej_msg);
2089         param->reason = __be16_to_cpu(rej_msg->reason);
2090         work->cm_event.private_data = &rej_msg->private_data;
2091 }
2092
2093 static struct cm_id_private * cm_acquire_rejected_id(struct cm_rej_msg *rej_msg)
2094 {
2095         struct cm_timewait_info *timewait_info;
2096         struct cm_id_private *cm_id_priv;
2097         __be32 remote_id;
2098
2099         remote_id = rej_msg->local_comm_id;
2100
2101         if (__be16_to_cpu(rej_msg->reason) == IB_CM_REJ_TIMEOUT) {
2102                 spin_lock_irq(&cm.lock);
2103                 timewait_info = cm_find_remote_id( *((__be64 *) rej_msg->ari),
2104                                                   remote_id);
2105                 if (!timewait_info) {
2106                         spin_unlock_irq(&cm.lock);
2107                         return NULL;
2108                 }
2109                 cm_id_priv = idr_find(&cm.local_id_table, (__force int)
2110                                       (timewait_info->work.local_id ^
2111                                        cm.random_id_operand));
2112                 if (cm_id_priv) {
2113                         if (cm_id_priv->id.remote_id == remote_id)
2114                                 atomic_inc(&cm_id_priv->refcount);
2115                         else
2116                                 cm_id_priv = NULL;
2117                 }
2118                 spin_unlock_irq(&cm.lock);
2119         } else if (cm_rej_get_msg_rejected(rej_msg) == CM_MSG_RESPONSE_REQ)
2120                 cm_id_priv = cm_acquire_id(rej_msg->remote_comm_id, 0);
2121         else
2122                 cm_id_priv = cm_acquire_id(rej_msg->remote_comm_id, remote_id);
2123
2124         return cm_id_priv;
2125 }
2126
2127 static int cm_rej_handler(struct cm_work *work)
2128 {
2129         struct cm_id_private *cm_id_priv;
2130         struct cm_rej_msg *rej_msg;
2131         int ret;
2132
2133         rej_msg = (struct cm_rej_msg *)work->mad_recv_wc->recv_buf.mad;
2134         cm_id_priv = cm_acquire_rejected_id(rej_msg);
2135         if (!cm_id_priv)
2136                 return -EINVAL;
2137
2138         cm_format_rej_event(work);
2139
2140         spin_lock_irq(&cm_id_priv->lock);
2141         switch (cm_id_priv->id.state) {
2142         case IB_CM_REQ_SENT:
2143         case IB_CM_MRA_REQ_RCVD:
2144         case IB_CM_REP_SENT:
2145         case IB_CM_MRA_REP_RCVD:
2146                 ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg);
2147                 /* fall through */
2148         case IB_CM_REQ_RCVD:
2149         case IB_CM_MRA_REQ_SENT:
2150                 if (__be16_to_cpu(rej_msg->reason) == IB_CM_REJ_STALE_CONN)
2151                         cm_enter_timewait(cm_id_priv);
2152                 else
2153                         cm_reset_to_idle(cm_id_priv);
2154                 break;
2155         case IB_CM_DREQ_SENT:
2156                 ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg);
2157                 /* fall through */
2158         case IB_CM_REP_RCVD:
2159         case IB_CM_MRA_REP_SENT:
2160         case IB_CM_ESTABLISHED:
2161                 cm_enter_timewait(cm_id_priv);
2162                 break;
2163         default:
2164                 spin_unlock_irq(&cm_id_priv->lock);
2165                 ret = -EINVAL;
2166                 goto out;
2167         }
2168
2169         ret = atomic_inc_and_test(&cm_id_priv->work_count);
2170         if (!ret)
2171                 list_add_tail(&work->list, &cm_id_priv->work_list);
2172         spin_unlock_irq(&cm_id_priv->lock);
2173
2174         if (ret)
2175                 cm_process_work(cm_id_priv, work);
2176         else
2177                 cm_deref_id(cm_id_priv);
2178         return 0;
2179 out:
2180         cm_deref_id(cm_id_priv);
2181         return -EINVAL;
2182 }
2183
2184 int ib_send_cm_mra(struct ib_cm_id *cm_id,
2185                    u8 service_timeout,
2186                    const void *private_data,
2187                    u8 private_data_len)
2188 {
2189         struct cm_id_private *cm_id_priv;
2190         struct ib_mad_send_buf *msg;
2191         void *data;
2192         unsigned long flags;
2193         int ret;
2194
2195         if (private_data && private_data_len > IB_CM_MRA_PRIVATE_DATA_SIZE)
2196                 return -EINVAL;
2197
2198         data = cm_copy_private_data(private_data, private_data_len);
2199         if (IS_ERR(data))
2200                 return PTR_ERR(data);
2201
2202         cm_id_priv = container_of(cm_id, struct cm_id_private, id);
2203
2204         spin_lock_irqsave(&cm_id_priv->lock, flags);
2205         switch(cm_id_priv->id.state) {
2206         case IB_CM_REQ_RCVD:
2207                 ret = cm_alloc_msg(cm_id_priv, &msg);
2208                 if (ret)
2209                         goto error1;
2210
2211                 cm_format_mra((struct cm_mra_msg *) msg->mad, cm_id_priv,
2212                               CM_MSG_RESPONSE_REQ, service_timeout,
2213                               private_data, private_data_len);
2214                 ret = ib_post_send_mad(msg, NULL);
2215                 if (ret)
2216                         goto error2;
2217                 cm_id->state = IB_CM_MRA_REQ_SENT;
2218                 break;
2219         case IB_CM_REP_RCVD:
2220                 ret = cm_alloc_msg(cm_id_priv, &msg);
2221                 if (ret)
2222                         goto error1;
2223
2224                 cm_format_mra((struct cm_mra_msg *) msg->mad, cm_id_priv,
2225                               CM_MSG_RESPONSE_REP, service_timeout,
2226                               private_data, private_data_len);
2227                 ret = ib_post_send_mad(msg, NULL);
2228                 if (ret)
2229                         goto error2;
2230                 cm_id->state = IB_CM_MRA_REP_SENT;
2231                 break;
2232         case IB_CM_ESTABLISHED:
2233                 ret = cm_alloc_msg(cm_id_priv, &msg);
2234                 if (ret)
2235                         goto error1;
2236
2237                 cm_format_mra((struct cm_mra_msg *) msg->mad, cm_id_priv,
2238                               CM_MSG_RESPONSE_OTHER, service_timeout,
2239                               private_data, private_data_len);
2240                 ret = ib_post_send_mad(msg, NULL);
2241                 if (ret)
2242                         goto error2;
2243                 cm_id->lap_state = IB_CM_MRA_LAP_SENT;
2244                 break;
2245         default:
2246                 ret = -EINVAL;
2247                 goto error1;
2248         }
2249         cm_id_priv->service_timeout = service_timeout;
2250         cm_set_private_data(cm_id_priv, data, private_data_len);
2251         spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2252         return 0;
2253
2254 error1: spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2255         kfree(data);
2256         return ret;
2257
2258 error2: spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2259         kfree(data);
2260         cm_free_msg(msg);
2261         return ret;
2262 }
2263 EXPORT_SYMBOL(ib_send_cm_mra);
2264
2265 static struct cm_id_private * cm_acquire_mraed_id(struct cm_mra_msg *mra_msg)
2266 {
2267         switch (cm_mra_get_msg_mraed(mra_msg)) {
2268         case CM_MSG_RESPONSE_REQ:
2269                 return cm_acquire_id(mra_msg->remote_comm_id, 0);
2270         case CM_MSG_RESPONSE_REP:
2271         case CM_MSG_RESPONSE_OTHER:
2272                 return cm_acquire_id(mra_msg->remote_comm_id,
2273                                      mra_msg->local_comm_id);
2274         default:
2275                 return NULL;
2276         }
2277 }
2278
2279 static int cm_mra_handler(struct cm_work *work)
2280 {
2281         struct cm_id_private *cm_id_priv;
2282         struct cm_mra_msg *mra_msg;
2283         int timeout, ret;
2284
2285         mra_msg = (struct cm_mra_msg *)work->mad_recv_wc->recv_buf.mad;
2286         cm_id_priv = cm_acquire_mraed_id(mra_msg);
2287         if (!cm_id_priv)
2288                 return -EINVAL;
2289
2290         work->cm_event.private_data = &mra_msg->private_data;
2291         work->cm_event.param.mra_rcvd.service_timeout =
2292                                         cm_mra_get_service_timeout(mra_msg);
2293         timeout = cm_convert_to_ms(cm_mra_get_service_timeout(mra_msg)) +
2294                   cm_convert_to_ms(cm_id_priv->av.packet_life_time);
2295
2296         spin_lock_irq(&cm_id_priv->lock);
2297         switch (cm_id_priv->id.state) {
2298         case IB_CM_REQ_SENT:
2299                 if (cm_mra_get_msg_mraed(mra_msg) != CM_MSG_RESPONSE_REQ ||
2300                     ib_modify_mad(cm_id_priv->av.port->mad_agent,
2301                                   cm_id_priv->msg, timeout))
2302                         goto out;
2303                 cm_id_priv->id.state = IB_CM_MRA_REQ_RCVD;
2304                 break;
2305         case IB_CM_REP_SENT:
2306                 if (cm_mra_get_msg_mraed(mra_msg) != CM_MSG_RESPONSE_REP ||
2307                     ib_modify_mad(cm_id_priv->av.port->mad_agent,
2308                                   cm_id_priv->msg, timeout))
2309                         goto out;
2310                 cm_id_priv->id.state = IB_CM_MRA_REP_RCVD;
2311                 break;
2312         case IB_CM_ESTABLISHED:
2313                 if (cm_mra_get_msg_mraed(mra_msg) != CM_MSG_RESPONSE_OTHER ||
2314                     cm_id_priv->id.lap_state != IB_CM_LAP_SENT ||
2315                     ib_modify_mad(cm_id_priv->av.port->mad_agent,
2316                                   cm_id_priv->msg, timeout))
2317                         goto out;
2318                 cm_id_priv->id.lap_state = IB_CM_MRA_LAP_RCVD;
2319                 break;
2320         default:
2321                 goto out;
2322         }
2323
2324         cm_id_priv->msg->context[1] = (void *) (unsigned long)
2325                                       cm_id_priv->id.state;
2326         ret = atomic_inc_and_test(&cm_id_priv->work_count);
2327         if (!ret)
2328                 list_add_tail(&work->list, &cm_id_priv->work_list);
2329         spin_unlock_irq(&cm_id_priv->lock);
2330
2331         if (ret)
2332                 cm_process_work(cm_id_priv, work);
2333         else
2334                 cm_deref_id(cm_id_priv);
2335         return 0;
2336 out:
2337         spin_unlock_irq(&cm_id_priv->lock);
2338         cm_deref_id(cm_id_priv);
2339         return -EINVAL;
2340 }
2341
2342 static void cm_format_lap(struct cm_lap_msg *lap_msg,
2343                           struct cm_id_private *cm_id_priv,
2344                           struct ib_sa_path_rec *alternate_path,
2345                           const void *private_data,
2346                           u8 private_data_len)
2347 {
2348         cm_format_mad_hdr(&lap_msg->hdr, CM_LAP_ATTR_ID,
2349                           cm_form_tid(cm_id_priv, CM_MSG_SEQUENCE_LAP));
2350         lap_msg->local_comm_id = cm_id_priv->id.local_id;
2351         lap_msg->remote_comm_id = cm_id_priv->id.remote_id;
2352         cm_lap_set_remote_qpn(lap_msg, cm_id_priv->remote_qpn);
2353         /* todo: need remote CM response timeout */
2354         cm_lap_set_remote_resp_timeout(lap_msg, 0x1F);
2355         lap_msg->alt_local_lid = alternate_path->slid;
2356         lap_msg->alt_remote_lid = alternate_path->dlid;
2357         lap_msg->alt_local_gid = alternate_path->sgid;
2358         lap_msg->alt_remote_gid = alternate_path->dgid;
2359         cm_lap_set_flow_label(lap_msg, alternate_path->flow_label);
2360         cm_lap_set_traffic_class(lap_msg, alternate_path->traffic_class);
2361         lap_msg->alt_hop_limit = alternate_path->hop_limit;
2362         cm_lap_set_packet_rate(lap_msg, alternate_path->rate);
2363         cm_lap_set_sl(lap_msg, alternate_path->sl);
2364         cm_lap_set_subnet_local(lap_msg, 1); /* local only... */
2365         cm_lap_set_local_ack_timeout(lap_msg,
2366                 min(31, alternate_path->packet_life_time + 1));
2367
2368         if (private_data && private_data_len)
2369                 memcpy(lap_msg->private_data, private_data, private_data_len);
2370 }
2371
2372 int ib_send_cm_lap(struct ib_cm_id *cm_id,
2373                    struct ib_sa_path_rec *alternate_path,
2374                    const void *private_data,
2375                    u8 private_data_len)
2376 {
2377         struct cm_id_private *cm_id_priv;
2378         struct ib_mad_send_buf *msg;
2379         unsigned long flags;
2380         int ret;
2381
2382         if (private_data && private_data_len > IB_CM_LAP_PRIVATE_DATA_SIZE)
2383                 return -EINVAL;
2384
2385         cm_id_priv = container_of(cm_id, struct cm_id_private, id);
2386         spin_lock_irqsave(&cm_id_priv->lock, flags);
2387         if (cm_id->state != IB_CM_ESTABLISHED ||
2388             (cm_id->lap_state != IB_CM_LAP_UNINIT &&
2389              cm_id->lap_state != IB_CM_LAP_IDLE)) {
2390                 ret = -EINVAL;
2391                 goto out;
2392         }
2393
2394         ret = cm_init_av_by_path(alternate_path, &cm_id_priv->alt_av);
2395         if (ret)
2396                 goto out;
2397
2398         ret = cm_alloc_msg(cm_id_priv, &msg);
2399         if (ret)
2400                 goto out;
2401
2402         cm_format_lap((struct cm_lap_msg *) msg->mad, cm_id_priv,
2403                       alternate_path, private_data, private_data_len);
2404         msg->timeout_ms = cm_id_priv->timeout_ms;
2405         msg->context[1] = (void *) (unsigned long) IB_CM_ESTABLISHED;
2406
2407         ret = ib_post_send_mad(msg, NULL);
2408         if (ret) {
2409                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2410                 cm_free_msg(msg);
2411                 return ret;
2412         }
2413
2414         cm_id->lap_state = IB_CM_LAP_SENT;
2415         cm_id_priv->msg = msg;
2416
2417 out:    spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2418         return ret;
2419 }
2420 EXPORT_SYMBOL(ib_send_cm_lap);
2421
2422 static void cm_format_path_from_lap(struct cm_id_private *cm_id_priv,
2423                                     struct ib_sa_path_rec *path,
2424                                     struct cm_lap_msg *lap_msg)
2425 {
2426         memset(path, 0, sizeof *path);
2427         path->dgid = lap_msg->alt_local_gid;
2428         path->sgid = lap_msg->alt_remote_gid;
2429         path->dlid = lap_msg->alt_local_lid;
2430         path->slid = lap_msg->alt_remote_lid;
2431         path->flow_label = cm_lap_get_flow_label(lap_msg);
2432         path->hop_limit = lap_msg->alt_hop_limit;
2433         path->traffic_class = cm_lap_get_traffic_class(lap_msg);
2434         path->reversible = 1;
2435         path->pkey = cm_id_priv->pkey;
2436         path->sl = cm_lap_get_sl(lap_msg);
2437         path->mtu_selector = IB_SA_EQ;
2438         path->mtu = cm_id_priv->path_mtu;
2439         path->rate_selector = IB_SA_EQ;
2440         path->rate = cm_lap_get_packet_rate(lap_msg);
2441         path->packet_life_time_selector = IB_SA_EQ;
2442         path->packet_life_time = cm_lap_get_local_ack_timeout(lap_msg);
2443         path->packet_life_time -= (path->packet_life_time > 0);
2444 }
2445
2446 static int cm_lap_handler(struct cm_work *work)
2447 {
2448         struct cm_id_private *cm_id_priv;
2449         struct cm_lap_msg *lap_msg;
2450         struct ib_cm_lap_event_param *param;
2451         struct ib_mad_send_buf *msg = NULL;
2452         int ret;
2453
2454         /* todo: verify LAP request and send reject APR if invalid. */
2455         lap_msg = (struct cm_lap_msg *)work->mad_recv_wc->recv_buf.mad;
2456         cm_id_priv = cm_acquire_id(lap_msg->remote_comm_id,
2457                                    lap_msg->local_comm_id);
2458         if (!cm_id_priv)
2459                 return -EINVAL;
2460
2461         param = &work->cm_event.param.lap_rcvd;
2462         param->alternate_path = &work->path[0];
2463         cm_format_path_from_lap(cm_id_priv, param->alternate_path, lap_msg);
2464         work->cm_event.private_data = &lap_msg->private_data;
2465
2466         spin_lock_irq(&cm_id_priv->lock);
2467         if (cm_id_priv->id.state != IB_CM_ESTABLISHED)
2468                 goto unlock;
2469
2470         switch (cm_id_priv->id.lap_state) {
2471         case IB_CM_LAP_UNINIT:
2472         case IB_CM_LAP_IDLE:
2473                 break;
2474         case IB_CM_MRA_LAP_SENT:
2475                 if (cm_alloc_response_msg(work->port, work->mad_recv_wc, &msg))
2476                         goto unlock;
2477
2478                 cm_format_mra((struct cm_mra_msg *) msg->mad, cm_id_priv,
2479                               CM_MSG_RESPONSE_OTHER,
2480                               cm_id_priv->service_timeout,
2481                               cm_id_priv->private_data,
2482                               cm_id_priv->private_data_len);
2483                 spin_unlock_irq(&cm_id_priv->lock);
2484
2485                 if (ib_post_send_mad(msg, NULL))
2486                         cm_free_msg(msg);
2487                 goto deref;
2488         default:
2489                 goto unlock;
2490         }
2491
2492         cm_id_priv->id.lap_state = IB_CM_LAP_RCVD;
2493         cm_id_priv->tid = lap_msg->hdr.tid;
2494         cm_init_av_for_response(work->port, work->mad_recv_wc->wc,
2495                                 work->mad_recv_wc->recv_buf.grh,
2496                                 &cm_id_priv->av);
2497         cm_init_av_by_path(param->alternate_path, &cm_id_priv->alt_av);
2498         ret = atomic_inc_and_test(&cm_id_priv->work_count);
2499         if (!ret)
2500                 list_add_tail(&work->list, &cm_id_priv->work_list);
2501         spin_unlock_irq(&cm_id_priv->lock);
2502
2503         if (ret)
2504                 cm_process_work(cm_id_priv, work);
2505         else
2506                 cm_deref_id(cm_id_priv);
2507         return 0;
2508
2509 unlock: spin_unlock_irq(&cm_id_priv->lock);
2510 deref:  cm_deref_id(cm_id_priv);
2511         return -EINVAL;
2512 }
2513
2514 static void cm_format_apr(struct cm_apr_msg *apr_msg,
2515                           struct cm_id_private *cm_id_priv,
2516                           enum ib_cm_apr_status status,
2517                           void *info,
2518                           u8 info_length,
2519                           const void *private_data,
2520                           u8 private_data_len)
2521 {
2522         cm_format_mad_hdr(&apr_msg->hdr, CM_APR_ATTR_ID, cm_id_priv->tid);
2523         apr_msg->local_comm_id = cm_id_priv->id.local_id;
2524         apr_msg->remote_comm_id = cm_id_priv->id.remote_id;
2525         apr_msg->ap_status = (u8) status;
2526
2527         if (info && info_length) {
2528                 apr_msg->info_length = info_length;
2529                 memcpy(apr_msg->info, info, info_length);
2530         }
2531
2532         if (private_data && private_data_len)
2533                 memcpy(apr_msg->private_data, private_data, private_data_len);
2534 }
2535
2536 int ib_send_cm_apr(struct ib_cm_id *cm_id,
2537                    enum ib_cm_apr_status status,
2538                    void *info,
2539                    u8 info_length,
2540                    const void *private_data,
2541                    u8 private_data_len)
2542 {
2543         struct cm_id_private *cm_id_priv;
2544         struct ib_mad_send_buf *msg;
2545         unsigned long flags;
2546         int ret;
2547
2548         if ((private_data && private_data_len > IB_CM_APR_PRIVATE_DATA_SIZE) ||
2549             (info && info_length > IB_CM_APR_INFO_LENGTH))
2550                 return -EINVAL;
2551
2552         cm_id_priv = container_of(cm_id, struct cm_id_private, id);
2553         spin_lock_irqsave(&cm_id_priv->lock, flags);
2554         if (cm_id->state != IB_CM_ESTABLISHED ||
2555             (cm_id->lap_state != IB_CM_LAP_RCVD &&
2556              cm_id->lap_state != IB_CM_MRA_LAP_SENT)) {
2557                 ret = -EINVAL;
2558                 goto out;
2559         }
2560
2561         ret = cm_alloc_msg(cm_id_priv, &msg);
2562         if (ret)
2563                 goto out;
2564
2565         cm_format_apr((struct cm_apr_msg *) msg->mad, cm_id_priv, status,
2566                       info, info_length, private_data, private_data_len);
2567         ret = ib_post_send_mad(msg, NULL);
2568         if (ret) {
2569                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2570                 cm_free_msg(msg);
2571                 return ret;
2572         }
2573
2574         cm_id->lap_state = IB_CM_LAP_IDLE;
2575 out:    spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2576         return ret;
2577 }
2578 EXPORT_SYMBOL(ib_send_cm_apr);
2579
2580 static int cm_apr_handler(struct cm_work *work)
2581 {
2582         struct cm_id_private *cm_id_priv;
2583         struct cm_apr_msg *apr_msg;
2584         int ret;
2585
2586         apr_msg = (struct cm_apr_msg *)work->mad_recv_wc->recv_buf.mad;
2587         cm_id_priv = cm_acquire_id(apr_msg->remote_comm_id,
2588                                    apr_msg->local_comm_id);
2589         if (!cm_id_priv)
2590                 return -EINVAL; /* Unmatched reply. */
2591
2592         work->cm_event.param.apr_rcvd.ap_status = apr_msg->ap_status;
2593         work->cm_event.param.apr_rcvd.apr_info = &apr_msg->info;
2594         work->cm_event.param.apr_rcvd.info_len = apr_msg->info_length;
2595         work->cm_event.private_data = &apr_msg->private_data;
2596
2597         spin_lock_irq(&cm_id_priv->lock);
2598         if (cm_id_priv->id.state != IB_CM_ESTABLISHED ||
2599             (cm_id_priv->id.lap_state != IB_CM_LAP_SENT &&
2600              cm_id_priv->id.lap_state != IB_CM_MRA_LAP_RCVD)) {
2601                 spin_unlock_irq(&cm_id_priv->lock);
2602                 goto out;
2603         }
2604         cm_id_priv->id.lap_state = IB_CM_LAP_IDLE;
2605         ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg);
2606         cm_id_priv->msg = NULL;
2607
2608         ret = atomic_inc_and_test(&cm_id_priv->work_count);
2609         if (!ret)
2610                 list_add_tail(&work->list, &cm_id_priv->work_list);
2611         spin_unlock_irq(&cm_id_priv->lock);
2612
2613         if (ret)
2614                 cm_process_work(cm_id_priv, work);
2615         else
2616                 cm_deref_id(cm_id_priv);
2617         return 0;
2618 out:
2619         cm_deref_id(cm_id_priv);
2620         return -EINVAL;
2621 }
2622
2623 static int cm_timewait_handler(struct cm_work *work)
2624 {
2625         struct cm_timewait_info *timewait_info;
2626         struct cm_id_private *cm_id_priv;
2627         int ret;
2628
2629         timewait_info = (struct cm_timewait_info *)work;
2630         spin_lock_irq(&cm.lock);
2631         list_del(&timewait_info->list);
2632         spin_unlock_irq(&cm.lock);
2633
2634         cm_id_priv = cm_acquire_id(timewait_info->work.local_id,
2635                                    timewait_info->work.remote_id);
2636         if (!cm_id_priv)
2637                 return -EINVAL;
2638
2639         spin_lock_irq(&cm_id_priv->lock);
2640         if (cm_id_priv->id.state != IB_CM_TIMEWAIT ||
2641             cm_id_priv->remote_qpn != timewait_info->remote_qpn) {
2642                 spin_unlock_irq(&cm_id_priv->lock);
2643                 goto out;
2644         }
2645         cm_id_priv->id.state = IB_CM_IDLE;
2646         ret = atomic_inc_and_test(&cm_id_priv->work_count);
2647         if (!ret)
2648                 list_add_tail(&work->list, &cm_id_priv->work_list);
2649         spin_unlock_irq(&cm_id_priv->lock);
2650
2651         if (ret)
2652                 cm_process_work(cm_id_priv, work);
2653         else
2654                 cm_deref_id(cm_id_priv);
2655         return 0;
2656 out:
2657         cm_deref_id(cm_id_priv);
2658         return -EINVAL;
2659 }
2660
2661 static void cm_format_sidr_req(struct cm_sidr_req_msg *sidr_req_msg,
2662                                struct cm_id_private *cm_id_priv,
2663                                struct ib_cm_sidr_req_param *param)
2664 {
2665         cm_format_mad_hdr(&sidr_req_msg->hdr, CM_SIDR_REQ_ATTR_ID,
2666                           cm_form_tid(cm_id_priv, CM_MSG_SEQUENCE_SIDR));
2667         sidr_req_msg->request_id = cm_id_priv->id.local_id;
2668         sidr_req_msg->pkey = cpu_to_be16(param->path->pkey);
2669         sidr_req_msg->service_id = param->service_id;
2670
2671         if (param->private_data && param->private_data_len)
2672                 memcpy(sidr_req_msg->private_data, param->private_data,
2673                        param->private_data_len);
2674 }
2675
2676 int ib_send_cm_sidr_req(struct ib_cm_id *cm_id,
2677                         struct ib_cm_sidr_req_param *param)
2678 {
2679         struct cm_id_private *cm_id_priv;
2680         struct ib_mad_send_buf *msg;
2681         unsigned long flags;
2682         int ret;
2683
2684         if (!param->path || (param->private_data &&
2685              param->private_data_len > IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE))
2686                 return -EINVAL;
2687
2688         cm_id_priv = container_of(cm_id, struct cm_id_private, id);
2689         ret = cm_init_av_by_path(param->path, &cm_id_priv->av);
2690         if (ret)
2691                 goto out;
2692
2693         cm_id->service_id = param->service_id;
2694         cm_id->service_mask = __constant_cpu_to_be64(~0ULL);
2695         cm_id_priv->timeout_ms = param->timeout_ms;
2696         cm_id_priv->max_cm_retries = param->max_cm_retries;
2697         ret = cm_alloc_msg(cm_id_priv, &msg);
2698         if (ret)
2699                 goto out;
2700
2701         cm_format_sidr_req((struct cm_sidr_req_msg *) msg->mad, cm_id_priv,
2702                            param);
2703         msg->timeout_ms = cm_id_priv->timeout_ms;
2704         msg->context[1] = (void *) (unsigned long) IB_CM_SIDR_REQ_SENT;
2705
2706         spin_lock_irqsave(&cm_id_priv->lock, flags);
2707         if (cm_id->state == IB_CM_IDLE)
2708                 ret = ib_post_send_mad(msg, NULL);
2709         else
2710                 ret = -EINVAL;
2711
2712         if (ret) {
2713                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2714                 cm_free_msg(msg);
2715                 goto out;
2716         }
2717         cm_id->state = IB_CM_SIDR_REQ_SENT;
2718         cm_id_priv->msg = msg;
2719         spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2720 out:
2721         return ret;
2722 }
2723 EXPORT_SYMBOL(ib_send_cm_sidr_req);
2724
2725 static void cm_format_sidr_req_event(struct cm_work *work,
2726                                      struct ib_cm_id *listen_id)
2727 {
2728         struct cm_sidr_req_msg *sidr_req_msg;
2729         struct ib_cm_sidr_req_event_param *param;
2730
2731         sidr_req_msg = (struct cm_sidr_req_msg *)
2732                                 work->mad_recv_wc->recv_buf.mad;
2733         param = &work->cm_event.param.sidr_req_rcvd;
2734         param->pkey = __be16_to_cpu(sidr_req_msg->pkey);
2735         param->listen_id = listen_id;
2736         param->port = work->port->port_num;
2737         work->cm_event.private_data = &sidr_req_msg->private_data;
2738 }
2739
2740 static int cm_sidr_req_handler(struct cm_work *work)
2741 {
2742         struct ib_cm_id *cm_id;
2743         struct cm_id_private *cm_id_priv, *cur_cm_id_priv;
2744         struct cm_sidr_req_msg *sidr_req_msg;
2745         struct ib_wc *wc;
2746
2747         cm_id = ib_create_cm_id(work->port->cm_dev->device, NULL, NULL);
2748         if (IS_ERR(cm_id))
2749                 return PTR_ERR(cm_id);
2750         cm_id_priv = container_of(cm_id, struct cm_id_private, id);
2751
2752         /* Record SGID/SLID and request ID for lookup. */
2753         sidr_req_msg = (struct cm_sidr_req_msg *)
2754                                 work->mad_recv_wc->recv_buf.mad;
2755         wc = work->mad_recv_wc->wc;
2756         cm_id_priv->av.dgid.global.subnet_prefix = cpu_to_be64(wc->slid);
2757         cm_id_priv->av.dgid.global.interface_id = 0;
2758         cm_init_av_for_response(work->port, work->mad_recv_wc->wc,
2759                                 work->mad_recv_wc->recv_buf.grh,
2760                                 &cm_id_priv->av);
2761         cm_id_priv->id.remote_id = sidr_req_msg->request_id;
2762         cm_id_priv->id.state = IB_CM_SIDR_REQ_RCVD;
2763         cm_id_priv->tid = sidr_req_msg->hdr.tid;
2764         atomic_inc(&cm_id_priv->work_count);
2765
2766         spin_lock_irq(&cm.lock);
2767         cur_cm_id_priv = cm_insert_remote_sidr(cm_id_priv);
2768         if (cur_cm_id_priv) {
2769                 spin_unlock_irq(&cm.lock);
2770                 goto out; /* Duplicate message. */
2771         }
2772         cur_cm_id_priv = cm_find_listen(cm_id->device,
2773                                         sidr_req_msg->service_id,
2774                                         sidr_req_msg->private_data);
2775         if (!cur_cm_id_priv) {
2776                 rb_erase(&cm_id_priv->sidr_id_node, &cm.remote_sidr_table);
2777                 spin_unlock_irq(&cm.lock);
2778                 /* todo: reply with no match */
2779                 goto out; /* No match. */
2780         }
2781         atomic_inc(&cur_cm_id_priv->refcount);
2782         spin_unlock_irq(&cm.lock);
2783
2784         cm_id_priv->id.cm_handler = cur_cm_id_priv->id.cm_handler;
2785         cm_id_priv->id.context = cur_cm_id_priv->id.context;
2786         cm_id_priv->id.service_id = sidr_req_msg->service_id;
2787         cm_id_priv->id.service_mask = __constant_cpu_to_be64(~0ULL);
2788
2789         cm_format_sidr_req_event(work, &cur_cm_id_priv->id);
2790         cm_process_work(cm_id_priv, work);
2791         cm_deref_id(cur_cm_id_priv);
2792         return 0;
2793 out:
2794         ib_destroy_cm_id(&cm_id_priv->id);
2795         return -EINVAL;
2796 }
2797
2798 static void cm_format_sidr_rep(struct cm_sidr_rep_msg *sidr_rep_msg,
2799                                struct cm_id_private *cm_id_priv,
2800                                struct ib_cm_sidr_rep_param *param)
2801 {
2802         cm_format_mad_hdr(&sidr_rep_msg->hdr, CM_SIDR_REP_ATTR_ID,
2803                           cm_id_priv->tid);
2804         sidr_rep_msg->request_id = cm_id_priv->id.remote_id;
2805         sidr_rep_msg->status = param->status;
2806         cm_sidr_rep_set_qpn(sidr_rep_msg, cpu_to_be32(param->qp_num));
2807         sidr_rep_msg->service_id = cm_id_priv->id.service_id;
2808         sidr_rep_msg->qkey = cpu_to_be32(param->qkey);
2809
2810         if (param->info && param->info_length)
2811                 memcpy(sidr_rep_msg->info, param->info, param->info_length);
2812
2813         if (param->private_data && param->private_data_len)
2814                 memcpy(sidr_rep_msg->private_data, param->private_data,
2815                        param->private_data_len);
2816 }
2817
2818 int ib_send_cm_sidr_rep(struct ib_cm_id *cm_id,
2819                         struct ib_cm_sidr_rep_param *param)
2820 {
2821         struct cm_id_private *cm_id_priv;
2822         struct ib_mad_send_buf *msg;
2823         unsigned long flags;
2824         int ret;
2825
2826         if ((param->info && param->info_length > IB_CM_SIDR_REP_INFO_LENGTH) ||
2827             (param->private_data &&
2828              param->private_data_len > IB_CM_SIDR_REP_PRIVATE_DATA_SIZE))
2829                 return -EINVAL;
2830
2831         cm_id_priv = container_of(cm_id, struct cm_id_private, id);
2832         spin_lock_irqsave(&cm_id_priv->lock, flags);
2833         if (cm_id->state != IB_CM_SIDR_REQ_RCVD) {
2834                 ret = -EINVAL;
2835                 goto error;
2836         }
2837
2838         ret = cm_alloc_msg(cm_id_priv, &msg);
2839         if (ret)
2840                 goto error;
2841
2842         cm_format_sidr_rep((struct cm_sidr_rep_msg *) msg->mad, cm_id_priv,
2843                            param);
2844         ret = ib_post_send_mad(msg, NULL);
2845         if (ret) {
2846                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2847                 cm_free_msg(msg);
2848                 return ret;
2849         }
2850         cm_id->state = IB_CM_IDLE;
2851         spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2852
2853         spin_lock_irqsave(&cm.lock, flags);
2854         rb_erase(&cm_id_priv->sidr_id_node, &cm.remote_sidr_table);
2855         spin_unlock_irqrestore(&cm.lock, flags);
2856         return 0;
2857
2858 error:  spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2859         return ret;
2860 }
2861 EXPORT_SYMBOL(ib_send_cm_sidr_rep);
2862
2863 static void cm_format_sidr_rep_event(struct cm_work *work)
2864 {
2865         struct cm_sidr_rep_msg *sidr_rep_msg;
2866         struct ib_cm_sidr_rep_event_param *param;
2867
2868         sidr_rep_msg = (struct cm_sidr_rep_msg *)
2869                                 work->mad_recv_wc->recv_buf.mad;
2870         param = &work->cm_event.param.sidr_rep_rcvd;
2871         param->status = sidr_rep_msg->status;
2872         param->qkey = be32_to_cpu(sidr_rep_msg->qkey);
2873         param->qpn = be32_to_cpu(cm_sidr_rep_get_qpn(sidr_rep_msg));
2874         param->info = &sidr_rep_msg->info;
2875         param->info_len = sidr_rep_msg->info_length;
2876         work->cm_event.private_data = &sidr_rep_msg->private_data;
2877 }
2878
2879 static int cm_sidr_rep_handler(struct cm_work *work)
2880 {
2881         struct cm_sidr_rep_msg *sidr_rep_msg;
2882         struct cm_id_private *cm_id_priv;
2883
2884         sidr_rep_msg = (struct cm_sidr_rep_msg *)
2885                                 work->mad_recv_wc->recv_buf.mad;
2886         cm_id_priv = cm_acquire_id(sidr_rep_msg->request_id, 0);
2887         if (!cm_id_priv)
2888                 return -EINVAL; /* Unmatched reply. */
2889
2890         spin_lock_irq(&cm_id_priv->lock);
2891         if (cm_id_priv->id.state != IB_CM_SIDR_REQ_SENT) {
2892                 spin_unlock_irq(&cm_id_priv->lock);
2893                 goto out;
2894         }
2895         cm_id_priv->id.state = IB_CM_IDLE;
2896         ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg);
2897         spin_unlock_irq(&cm_id_priv->lock);
2898
2899         cm_format_sidr_rep_event(work);
2900         cm_process_work(cm_id_priv, work);
2901         return 0;
2902 out:
2903         cm_deref_id(cm_id_priv);
2904         return -EINVAL;
2905 }
2906
2907 static void cm_process_send_error(struct ib_mad_send_buf *msg,
2908                                   enum ib_wc_status wc_status)
2909 {
2910         struct cm_id_private *cm_id_priv;
2911         struct ib_cm_event cm_event;
2912         enum ib_cm_state state;
2913         int ret;
2914
2915         memset(&cm_event, 0, sizeof cm_event);
2916         cm_id_priv = msg->context[0];
2917
2918         /* Discard old sends or ones without a response. */
2919         spin_lock_irq(&cm_id_priv->lock);
2920         state = (enum ib_cm_state) (unsigned long) msg->context[1];
2921         if (msg != cm_id_priv->msg || state != cm_id_priv->id.state)
2922                 goto discard;
2923
2924         switch (state) {
2925         case IB_CM_REQ_SENT:
2926         case IB_CM_MRA_REQ_RCVD:
2927                 cm_reset_to_idle(cm_id_priv);
2928                 cm_event.event = IB_CM_REQ_ERROR;
2929                 break;
2930         case IB_CM_REP_SENT:
2931         case IB_CM_MRA_REP_RCVD:
2932                 cm_reset_to_idle(cm_id_priv);
2933                 cm_event.event = IB_CM_REP_ERROR;
2934                 break;
2935         case IB_CM_DREQ_SENT:
2936                 cm_enter_timewait(cm_id_priv);
2937                 cm_event.event = IB_CM_DREQ_ERROR;
2938                 break;
2939         case IB_CM_SIDR_REQ_SENT:
2940                 cm_id_priv->id.state = IB_CM_IDLE;
2941                 cm_event.event = IB_CM_SIDR_REQ_ERROR;
2942                 break;
2943         default:
2944                 goto discard;
2945         }
2946         spin_unlock_irq(&cm_id_priv->lock);
2947         cm_event.param.send_status = wc_status;
2948
2949         /* No other events can occur on the cm_id at this point. */
2950         ret = cm_id_priv->id.cm_handler(&cm_id_priv->id, &cm_event);
2951         cm_free_msg(msg);
2952         if (ret)
2953                 ib_destroy_cm_id(&cm_id_priv->id);
2954         return;
2955 discard:
2956         spin_unlock_irq(&cm_id_priv->lock);
2957         cm_free_msg(msg);
2958 }
2959
2960 static void cm_send_handler(struct ib_mad_agent *mad_agent,
2961                             struct ib_mad_send_wc *mad_send_wc)
2962 {
2963         struct ib_mad_send_buf *msg = mad_send_wc->send_buf;
2964
2965         switch (mad_send_wc->status) {
2966         case IB_WC_SUCCESS:
2967         case IB_WC_WR_FLUSH_ERR:
2968                 cm_free_msg(msg);
2969                 break;
2970         default:
2971                 if (msg->context[0] && msg->context[1])
2972                         cm_process_send_error(msg, mad_send_wc->status);
2973                 else
2974                         cm_free_msg(msg);
2975                 break;
2976         }
2977 }
2978
2979 static void cm_work_handler(struct work_struct *_work)
2980 {
2981         struct cm_work *work = container_of(_work, struct cm_work, work.work);
2982         int ret;
2983
2984         switch (work->cm_event.event) {
2985         case IB_CM_REQ_RECEIVED:
2986                 ret = cm_req_handler(work);
2987                 break;
2988         case IB_CM_MRA_RECEIVED:
2989                 ret = cm_mra_handler(work);
2990                 break;
2991         case IB_CM_REJ_RECEIVED:
2992                 ret = cm_rej_handler(work);
2993                 break;
2994         case IB_CM_REP_RECEIVED:
2995                 ret = cm_rep_handler(work);
2996                 break;
2997         case IB_CM_RTU_RECEIVED:
2998                 ret = cm_rtu_handler(work);
2999                 break;
3000         case IB_CM_USER_ESTABLISHED:
3001                 ret = cm_establish_handler(work);
3002                 break;
3003         case IB_CM_DREQ_RECEIVED:
3004                 ret = cm_dreq_handler(work);
3005                 break;
3006         case IB_CM_DREP_RECEIVED:
3007                 ret = cm_drep_handler(work);
3008                 break;
3009         case IB_CM_SIDR_REQ_RECEIVED:
3010                 ret = cm_sidr_req_handler(work);
3011                 break;
3012         case IB_CM_SIDR_REP_RECEIVED:
3013                 ret = cm_sidr_rep_handler(work);
3014                 break;
3015         case IB_CM_LAP_RECEIVED:
3016                 ret = cm_lap_handler(work);
3017                 break;
3018         case IB_CM_APR_RECEIVED:
3019                 ret = cm_apr_handler(work);
3020                 break;
3021         case IB_CM_TIMEWAIT_EXIT:
3022                 ret = cm_timewait_handler(work);
3023                 break;
3024         default:
3025                 ret = -EINVAL;
3026                 break;
3027         }
3028         if (ret)
3029                 cm_free_work(work);
3030 }
3031
3032 static int cm_establish(struct ib_cm_id *cm_id)
3033 {
3034         struct cm_id_private *cm_id_priv;
3035         struct cm_work *work;
3036         unsigned long flags;
3037         int ret = 0;
3038
3039         work = kmalloc(sizeof *work, GFP_ATOMIC);
3040         if (!work)
3041                 return -ENOMEM;
3042
3043         cm_id_priv = container_of(cm_id, struct cm_id_private, id);
3044         spin_lock_irqsave(&cm_id_priv->lock, flags);
3045         switch (cm_id->state)
3046         {
3047         case IB_CM_REP_SENT:
3048         case IB_CM_MRA_REP_RCVD:
3049                 cm_id->state = IB_CM_ESTABLISHED;
3050                 break;
3051         case IB_CM_ESTABLISHED:
3052                 ret = -EISCONN;
3053                 break;
3054         default:
3055                 ret = -EINVAL;
3056                 break;
3057         }
3058         spin_unlock_irqrestore(&cm_id_priv->lock, flags);
3059
3060         if (ret) {
3061                 kfree(work);
3062                 goto out;
3063         }
3064
3065         /*
3066          * The CM worker thread may try to destroy the cm_id before it
3067          * can execute this work item.  To prevent potential deadlock,
3068          * we need to find the cm_id once we're in the context of the
3069          * worker thread, rather than holding a reference on it.
3070          */
3071         INIT_DELAYED_WORK(&work->work, cm_work_handler);
3072         work->local_id = cm_id->local_id;
3073         work->remote_id = cm_id->remote_id;
3074         work->mad_recv_wc = NULL;
3075         work->cm_event.event = IB_CM_USER_ESTABLISHED;
3076         queue_delayed_work(cm.wq, &work->work, 0);
3077 out:
3078         return ret;
3079 }
3080
3081 static int cm_migrate(struct ib_cm_id *cm_id)
3082 {
3083         struct cm_id_private *cm_id_priv;
3084         unsigned long flags;
3085         int ret = 0;
3086
3087         cm_id_priv = container_of(cm_id, struct cm_id_private, id);
3088         spin_lock_irqsave(&cm_id_priv->lock, flags);
3089         if (cm_id->state == IB_CM_ESTABLISHED &&
3090             (cm_id->lap_state == IB_CM_LAP_UNINIT ||
3091              cm_id->lap_state == IB_CM_LAP_IDLE)) {
3092                 cm_id->lap_state = IB_CM_LAP_IDLE;
3093                 cm_id_priv->av = cm_id_priv->alt_av;
3094         } else
3095                 ret = -EINVAL;
3096         spin_unlock_irqrestore(&cm_id_priv->lock, flags);
3097
3098         return ret;
3099 }
3100
3101 int ib_cm_notify(struct ib_cm_id *cm_id, enum ib_event_type event)
3102 {
3103         int ret;
3104
3105         switch (event) {
3106         case IB_EVENT_COMM_EST:
3107                 ret = cm_establish(cm_id);
3108                 break;
3109         case IB_EVENT_PATH_MIG:
3110                 ret = cm_migrate(cm_id);
3111                 break;
3112         default:
3113                 ret = -EINVAL;
3114         }
3115         return ret;
3116 }
3117 EXPORT_SYMBOL(ib_cm_notify);
3118
3119 static void cm_recv_handler(struct ib_mad_agent *mad_agent,
3120                             struct ib_mad_recv_wc *mad_recv_wc)
3121 {
3122         struct cm_work *work;
3123         enum ib_cm_event_type event;
3124         int paths = 0;
3125
3126         switch (mad_recv_wc->recv_buf.mad->mad_hdr.attr_id) {
3127         case CM_REQ_ATTR_ID:
3128                 paths = 1 + (((struct cm_req_msg *) mad_recv_wc->recv_buf.mad)->
3129                                                     alt_local_lid != 0);
3130                 event = IB_CM_REQ_RECEIVED;
3131                 break;
3132         case CM_MRA_ATTR_ID:
3133                 event = IB_CM_MRA_RECEIVED;
3134                 break;
3135         case CM_REJ_ATTR_ID:
3136                 event = IB_CM_REJ_RECEIVED;
3137                 break;
3138         case CM_REP_ATTR_ID:
3139                 event = IB_CM_REP_RECEIVED;
3140                 break;
3141         case CM_RTU_ATTR_ID:
3142                 event = IB_CM_RTU_RECEIVED;
3143                 break;
3144         case CM_DREQ_ATTR_ID:
3145                 event = IB_CM_DREQ_RECEIVED;
3146                 break;
3147         case CM_DREP_ATTR_ID:
3148                 event = IB_CM_DREP_RECEIVED;
3149                 break;
3150         case CM_SIDR_REQ_ATTR_ID:
3151                 event = IB_CM_SIDR_REQ_RECEIVED;
3152                 break;
3153         case CM_SIDR_REP_ATTR_ID:
3154                 event = IB_CM_SIDR_REP_RECEIVED;
3155                 break;
3156         case CM_LAP_ATTR_ID:
3157                 paths = 1;
3158                 event = IB_CM_LAP_RECEIVED;
3159                 break;
3160         case CM_APR_ATTR_ID:
3161                 event = IB_CM_APR_RECEIVED;
3162                 break;
3163         default:
3164                 ib_free_recv_mad(mad_recv_wc);
3165                 return;
3166         }
3167
3168         work = kmalloc(sizeof *work + sizeof(struct ib_sa_path_rec) * paths,
3169                        GFP_KERNEL);
3170         if (!work) {
3171                 ib_free_recv_mad(mad_recv_wc);
3172                 return;
3173         }
3174
3175         INIT_DELAYED_WORK(&work->work, cm_work_handler);
3176         work->cm_event.event = event;
3177         work->mad_recv_wc = mad_recv_wc;
3178         work->port = (struct cm_port *)mad_agent->context;
3179         queue_delayed_work(cm.wq, &work->work, 0);
3180 }
3181
3182 static int cm_init_qp_init_attr(struct cm_id_private *cm_id_priv,
3183                                 struct ib_qp_attr *qp_attr,
3184                                 int *qp_attr_mask)
3185 {
3186         unsigned long flags;
3187         int ret;
3188
3189         spin_lock_irqsave(&cm_id_priv->lock, flags);
3190         switch (cm_id_priv->id.state) {
3191         case IB_CM_REQ_SENT:
3192         case IB_CM_MRA_REQ_RCVD:
3193         case IB_CM_REQ_RCVD:
3194         case IB_CM_MRA_REQ_SENT:
3195         case IB_CM_REP_RCVD:
3196         case IB_CM_MRA_REP_SENT:
3197         case IB_CM_REP_SENT:
3198         case IB_CM_MRA_REP_RCVD:
3199         case IB_CM_ESTABLISHED:
3200                 *qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS |
3201                                 IB_QP_PKEY_INDEX | IB_QP_PORT;
3202                 qp_attr->qp_access_flags = IB_ACCESS_REMOTE_WRITE;
3203                 if (cm_id_priv->responder_resources)
3204                         qp_attr->qp_access_flags |= IB_ACCESS_REMOTE_READ |
3205                                                     IB_ACCESS_REMOTE_ATOMIC;
3206                 qp_attr->pkey_index = cm_id_priv->av.pkey_index;
3207                 qp_attr->port_num = cm_id_priv->av.port->port_num;
3208                 ret = 0;
3209                 break;
3210         default:
3211                 ret = -EINVAL;
3212                 break;
3213         }
3214         spin_unlock_irqrestore(&cm_id_priv->lock, flags);
3215         return ret;
3216 }
3217
3218 static int cm_init_qp_rtr_attr(struct cm_id_private *cm_id_priv,
3219                                struct ib_qp_attr *qp_attr,
3220                                int *qp_attr_mask)
3221 {
3222         unsigned long flags;
3223         int ret;
3224
3225         spin_lock_irqsave(&cm_id_priv->lock, flags);
3226         switch (cm_id_priv->id.state) {
3227         case IB_CM_REQ_RCVD:
3228         case IB_CM_MRA_REQ_SENT:
3229         case IB_CM_REP_RCVD:
3230         case IB_CM_MRA_REP_SENT:
3231         case IB_CM_REP_SENT:
3232         case IB_CM_MRA_REP_RCVD:
3233         case IB_CM_ESTABLISHED:
3234                 *qp_attr_mask = IB_QP_STATE | IB_QP_AV | IB_QP_PATH_MTU |
3235                                 IB_QP_DEST_QPN | IB_QP_RQ_PSN;
3236                 qp_attr->ah_attr = cm_id_priv->av.ah_attr;
3237                 qp_attr->path_mtu = cm_id_priv->path_mtu;
3238                 qp_attr->dest_qp_num = be32_to_cpu(cm_id_priv->remote_qpn);
3239                 qp_attr->rq_psn = be32_to_cpu(cm_id_priv->rq_psn);
3240                 if (cm_id_priv->qp_type == IB_QPT_RC) {
3241                         *qp_attr_mask |= IB_QP_MAX_DEST_RD_ATOMIC |
3242                                          IB_QP_MIN_RNR_TIMER;
3243                         qp_attr->max_dest_rd_atomic =
3244                                         cm_id_priv->responder_resources;
3245                         qp_attr->min_rnr_timer = 0;
3246                 }
3247                 if (cm_id_priv->alt_av.ah_attr.dlid) {
3248                         *qp_attr_mask |= IB_QP_ALT_PATH;
3249                         qp_attr->alt_port_num = cm_id_priv->alt_av.port->port_num;
3250                         qp_attr->alt_pkey_index = cm_id_priv->alt_av.pkey_index;
3251                         qp_attr->alt_timeout =
3252                                         cm_id_priv->alt_av.packet_life_time + 1;
3253                         qp_attr->alt_ah_attr = cm_id_priv->alt_av.ah_attr;
3254                 }
3255                 ret = 0;
3256                 break;
3257         default:
3258                 ret = -EINVAL;
3259                 break;
3260         }
3261         spin_unlock_irqrestore(&cm_id_priv->lock, flags);
3262         return ret;
3263 }
3264
3265 static int cm_init_qp_rts_attr(struct cm_id_private *cm_id_priv,
3266                                struct ib_qp_attr *qp_attr,
3267                                int *qp_attr_mask)
3268 {
3269         unsigned long flags;
3270         int ret;
3271
3272         spin_lock_irqsave(&cm_id_priv->lock, flags);
3273         switch (cm_id_priv->id.state) {
3274         /* Allow transition to RTS before sending REP */
3275         case IB_CM_REQ_RCVD:
3276         case IB_CM_MRA_REQ_SENT:
3277
3278         case IB_CM_REP_RCVD:
3279         case IB_CM_MRA_REP_SENT:
3280         case IB_CM_REP_SENT:
3281         case IB_CM_MRA_REP_RCVD:
3282         case IB_CM_ESTABLISHED:
3283                 if (cm_id_priv->id.lap_state == IB_CM_LAP_UNINIT) {
3284                         *qp_attr_mask = IB_QP_STATE | IB_QP_SQ_PSN;
3285                         qp_attr->sq_psn = be32_to_cpu(cm_id_priv->sq_psn);
3286                         if (cm_id_priv->qp_type == IB_QPT_RC) {
3287                                 *qp_attr_mask |= IB_QP_TIMEOUT | IB_QP_RETRY_CNT |
3288                                                  IB_QP_RNR_RETRY |
3289                                                  IB_QP_MAX_QP_RD_ATOMIC;
3290                                 qp_attr->timeout =
3291                                         cm_id_priv->av.packet_life_time + 1;
3292                                 qp_attr->retry_cnt = cm_id_priv->retry_count;
3293                                 qp_attr->rnr_retry = cm_id_priv->rnr_retry_count;
3294                                 qp_attr->max_rd_atomic =
3295                                         cm_id_priv->initiator_depth;
3296                         }
3297                         if (cm_id_priv->alt_av.ah_attr.dlid) {
3298                                 *qp_attr_mask |= IB_QP_PATH_MIG_STATE;
3299                                 qp_attr->path_mig_state = IB_MIG_REARM;
3300                         }
3301                 } else {
3302                         *qp_attr_mask = IB_QP_ALT_PATH | IB_QP_PATH_MIG_STATE;
3303                         qp_attr->alt_port_num = cm_id_priv->alt_av.port->port_num;
3304                         qp_attr->alt_pkey_index = cm_id_priv->alt_av.pkey_index;
3305                         qp_attr->alt_timeout =
3306                                 cm_id_priv->alt_av.packet_life_time + 1;
3307                         qp_attr->alt_ah_attr = cm_id_priv->alt_av.ah_attr;
3308                         qp_attr->path_mig_state = IB_MIG_REARM;
3309                 }
3310                 ret = 0;
3311                 break;
3312         default:
3313                 ret = -EINVAL;
3314                 break;
3315         }
3316         spin_unlock_irqrestore(&cm_id_priv->lock, flags);
3317         return ret;
3318 }
3319
3320 int ib_cm_init_qp_attr(struct ib_cm_id *cm_id,
3321                        struct ib_qp_attr *qp_attr,
3322                        int *qp_attr_mask)
3323 {
3324         struct cm_id_private *cm_id_priv;
3325         int ret;
3326
3327         cm_id_priv = container_of(cm_id, struct cm_id_private, id);
3328         switch (qp_attr->qp_state) {
3329         case IB_QPS_INIT:
3330                 ret = cm_init_qp_init_attr(cm_id_priv, qp_attr, qp_attr_mask);
3331                 break;
3332         case IB_QPS_RTR:
3333                 ret = cm_init_qp_rtr_attr(cm_id_priv, qp_attr, qp_attr_mask);
3334                 break;
3335         case IB_QPS_RTS:
3336                 ret = cm_init_qp_rts_attr(cm_id_priv, qp_attr, qp_attr_mask);
3337                 break;
3338         default:
3339                 ret = -EINVAL;
3340                 break;
3341         }
3342         return ret;
3343 }
3344 EXPORT_SYMBOL(ib_cm_init_qp_attr);
3345
3346 static void cm_add_one(struct ib_device *device)
3347 {
3348         struct cm_device *cm_dev;
3349         struct cm_port *port;
3350         struct ib_mad_reg_req reg_req = {
3351                 .mgmt_class = IB_MGMT_CLASS_CM,
3352                 .mgmt_class_version = IB_CM_CLASS_VERSION
3353         };
3354         struct ib_port_modify port_modify = {
3355                 .set_port_cap_mask = IB_PORT_CM_SUP
3356         };
3357         unsigned long flags;
3358         int ret;
3359         u8 i;
3360
3361         if (rdma_node_get_transport(device->node_type) != RDMA_TRANSPORT_IB)
3362                 return;
3363
3364         cm_dev = kmalloc(sizeof(*cm_dev) + sizeof(*port) *
3365                          device->phys_port_cnt, GFP_KERNEL);
3366         if (!cm_dev)
3367                 return;
3368
3369         cm_dev->device = device;
3370
3371         set_bit(IB_MGMT_METHOD_SEND, reg_req.method_mask);
3372         for (i = 1; i <= device->phys_port_cnt; i++) {
3373                 port = &cm_dev->port[i-1];
3374                 port->cm_dev = cm_dev;
3375                 port->port_num = i;
3376                 port->mad_agent = ib_register_mad_agent(device, i,
3377                                                         IB_QPT_GSI,
3378                                                         &reg_req,
3379                                                         0,
3380                                                         cm_send_handler,
3381                                                         cm_recv_handler,
3382                                                         port);
3383                 if (IS_ERR(port->mad_agent))
3384                         goto error1;
3385
3386                 ret = ib_modify_port(device, i, 0, &port_modify);
3387                 if (ret)
3388                         goto error2;
3389         }
3390         ib_set_client_data(device, &cm_client, cm_dev);
3391
3392         write_lock_irqsave(&cm.device_lock, flags);
3393         list_add_tail(&cm_dev->list, &cm.device_list);
3394         write_unlock_irqrestore(&cm.device_lock, flags);
3395         return;
3396
3397 error2:
3398         ib_unregister_mad_agent(port->mad_agent);
3399 error1:
3400         port_modify.set_port_cap_mask = 0;
3401         port_modify.clr_port_cap_mask = IB_PORT_CM_SUP;
3402         while (--i) {
3403                 port = &cm_dev->port[i-1];
3404                 ib_modify_port(device, port->port_num, 0, &port_modify);
3405                 ib_unregister_mad_agent(port->mad_agent);
3406         }
3407         kfree(cm_dev);
3408 }
3409
3410 static void cm_remove_one(struct ib_device *device)
3411 {
3412         struct cm_device *cm_dev;
3413         struct cm_port *port;
3414         struct ib_port_modify port_modify = {
3415                 .clr_port_cap_mask = IB_PORT_CM_SUP
3416         };
3417         unsigned long flags;
3418         int i;
3419
3420         cm_dev = ib_get_client_data(device, &cm_client);
3421         if (!cm_dev)
3422                 return;
3423
3424         write_lock_irqsave(&cm.device_lock, flags);
3425         list_del(&cm_dev->list);
3426         write_unlock_irqrestore(&cm.device_lock, flags);
3427
3428         for (i = 1; i <= device->phys_port_cnt; i++) {
3429                 port = &cm_dev->port[i-1];
3430                 ib_modify_port(device, port->port_num, 0, &port_modify);
3431                 ib_unregister_mad_agent(port->mad_agent);
3432         }
3433         kfree(cm_dev);
3434 }
3435
3436 static int __init ib_cm_init(void)
3437 {
3438         int ret;
3439
3440         memset(&cm, 0, sizeof cm);
3441         INIT_LIST_HEAD(&cm.device_list);
3442         rwlock_init(&cm.device_lock);
3443         spin_lock_init(&cm.lock);
3444         cm.listen_service_table = RB_ROOT;
3445         cm.listen_service_id = __constant_be64_to_cpu(IB_CM_ASSIGN_SERVICE_ID);
3446         cm.remote_id_table = RB_ROOT;
3447         cm.remote_qp_table = RB_ROOT;
3448         cm.remote_sidr_table = RB_ROOT;
3449         idr_init(&cm.local_id_table);
3450         get_random_bytes(&cm.random_id_operand, sizeof cm.random_id_operand);
3451         idr_pre_get(&cm.local_id_table, GFP_KERNEL);
3452         INIT_LIST_HEAD(&cm.timewait_list);
3453
3454         cm.wq = create_workqueue("ib_cm");
3455         if (!cm.wq)
3456                 return -ENOMEM;
3457
3458         ret = ib_register_client(&cm_client);
3459         if (ret)
3460                 goto error;
3461
3462         return 0;
3463 error:
3464         destroy_workqueue(cm.wq);
3465         return ret;
3466 }
3467
3468 static void __exit ib_cm_cleanup(void)
3469 {
3470         struct cm_timewait_info *timewait_info, *tmp;
3471
3472         spin_lock_irq(&cm.lock);
3473         list_for_each_entry(timewait_info, &cm.timewait_list, list)
3474                 cancel_delayed_work(&timewait_info->work.work);
3475         spin_unlock_irq(&cm.lock);
3476
3477         destroy_workqueue(cm.wq);
3478
3479         list_for_each_entry_safe(timewait_info, tmp, &cm.timewait_list, list) {
3480                 list_del(&timewait_info->list);
3481                 kfree(timewait_info);
3482         }
3483
3484         ib_unregister_client(&cm_client);
3485         idr_destroy(&cm.local_id_table);
3486 }
3487
3488 module_init(ib_cm_init);
3489 module_exit(ib_cm_cleanup);
3490