Merge ../linux-2.6
[linux-2.6] / drivers / infiniband / core / cm.c
1 /*
2  * Copyright (c) 2004, 2005 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 2821 2005-07-08 17:07:28Z 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/pci.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         struct workqueue_struct *wq;
77 } cm;
78
79 struct cm_port {
80         struct cm_device *cm_dev;
81         struct ib_mad_agent *mad_agent;
82         u8 port_num;
83 };
84
85 struct cm_device {
86         struct list_head list;
87         struct ib_device *device;
88         __be64 ca_guid;
89         struct cm_port port[0];
90 };
91
92 struct cm_av {
93         struct cm_port *port;
94         union ib_gid dgid;
95         struct ib_ah_attr ah_attr;
96         u16 pkey_index;
97         u8 packet_life_time;
98 };
99
100 struct cm_work {
101         struct work_struct work;
102         struct list_head list;
103         struct cm_port *port;
104         struct ib_mad_recv_wc *mad_recv_wc;     /* Received MADs */
105         __be32 local_id;                        /* Established / timewait */
106         __be32 remote_id;
107         struct ib_cm_event cm_event;
108         struct ib_sa_path_rec path[0];
109 };
110
111 struct cm_timewait_info {
112         struct cm_work work;                    /* Must be first. */
113         struct rb_node remote_qp_node;
114         struct rb_node remote_id_node;
115         __be64 remote_ca_guid;
116         __be32 remote_qpn;
117         u8 inserted_remote_qp;
118         u8 inserted_remote_id;
119 };
120
121 struct cm_id_private {
122         struct ib_cm_id id;
123
124         struct rb_node service_node;
125         struct rb_node sidr_id_node;
126         spinlock_t lock;        /* Do not acquire inside cm.lock */
127         struct completion comp;
128         atomic_t refcount;
129
130         struct ib_mad_send_buf *msg;
131         struct cm_timewait_info *timewait_info;
132         /* todo: use alternate port on send failure */
133         struct cm_av av;
134         struct cm_av alt_av;
135
136         void *private_data;
137         __be64 tid;
138         __be32 local_qpn;
139         __be32 remote_qpn;
140         enum ib_qp_type qp_type;
141         __be32 sq_psn;
142         __be32 rq_psn;
143         int timeout_ms;
144         enum ib_mtu path_mtu;
145         u8 private_data_len;
146         u8 max_cm_retries;
147         u8 peer_to_peer;
148         u8 responder_resources;
149         u8 initiator_depth;
150         u8 local_ack_timeout;
151         u8 retry_count;
152         u8 rnr_retry_count;
153         u8 service_timeout;
154
155         struct list_head work_list;
156         atomic_t work_count;
157 };
158
159 static void cm_work_handler(void *data);
160
161 static inline void cm_deref_id(struct cm_id_private *cm_id_priv)
162 {
163         if (atomic_dec_and_test(&cm_id_priv->refcount))
164                 complete(&cm_id_priv->comp);
165 }
166
167 static int cm_alloc_msg(struct cm_id_private *cm_id_priv,
168                         struct ib_mad_send_buf **msg)
169 {
170         struct ib_mad_agent *mad_agent;
171         struct ib_mad_send_buf *m;
172         struct ib_ah *ah;
173
174         mad_agent = cm_id_priv->av.port->mad_agent;
175         ah = ib_create_ah(mad_agent->qp->pd, &cm_id_priv->av.ah_attr);
176         if (IS_ERR(ah))
177                 return PTR_ERR(ah);
178
179         m = ib_create_send_mad(mad_agent, cm_id_priv->id.remote_cm_qpn, 
180                                cm_id_priv->av.pkey_index,
181                                0, IB_MGMT_MAD_HDR, IB_MGMT_MAD_DATA,
182                                GFP_ATOMIC);
183         if (IS_ERR(m)) {
184                 ib_destroy_ah(ah);
185                 return PTR_ERR(m);
186         }
187
188         /* Timeout set by caller if response is expected. */
189         m->ah = ah;
190         m->retries = cm_id_priv->max_cm_retries;
191
192         atomic_inc(&cm_id_priv->refcount);
193         m->context[0] = cm_id_priv;
194         *msg = m;
195         return 0;
196 }
197
198 static int cm_alloc_response_msg(struct cm_port *port,
199                                  struct ib_mad_recv_wc *mad_recv_wc,
200                                  struct ib_mad_send_buf **msg)
201 {
202         struct ib_mad_send_buf *m;
203         struct ib_ah *ah;
204
205         ah = ib_create_ah_from_wc(port->mad_agent->qp->pd, mad_recv_wc->wc,
206                                   mad_recv_wc->recv_buf.grh, port->port_num);
207         if (IS_ERR(ah))
208                 return PTR_ERR(ah);
209
210         m = ib_create_send_mad(port->mad_agent, 1, mad_recv_wc->wc->pkey_index,
211                                0, IB_MGMT_MAD_HDR, IB_MGMT_MAD_DATA,
212                                GFP_ATOMIC);
213         if (IS_ERR(m)) {
214                 ib_destroy_ah(ah);
215                 return PTR_ERR(m);
216         }
217         m->ah = ah;
218         *msg = m;
219         return 0;
220 }
221
222 static void cm_free_msg(struct ib_mad_send_buf *msg)
223 {
224         ib_destroy_ah(msg->ah);
225         if (msg->context[0])
226                 cm_deref_id(msg->context[0]);
227         ib_free_send_mad(msg);
228 }
229
230 static void * cm_copy_private_data(const void *private_data,
231                                    u8 private_data_len)
232 {
233         void *data;
234
235         if (!private_data || !private_data_len)
236                 return NULL;
237
238         data = kmalloc(private_data_len, GFP_KERNEL);
239         if (!data)
240                 return ERR_PTR(-ENOMEM);
241
242         memcpy(data, private_data, private_data_len);
243         return data;
244 }
245
246 static void cm_set_private_data(struct cm_id_private *cm_id_priv,
247                                  void *private_data, u8 private_data_len)
248 {
249         if (cm_id_priv->private_data && cm_id_priv->private_data_len)
250                 kfree(cm_id_priv->private_data);
251
252         cm_id_priv->private_data = private_data;
253         cm_id_priv->private_data_len = private_data_len;
254 }
255
256 static void cm_set_ah_attr(struct ib_ah_attr *ah_attr, u8 port_num,
257                            u16 dlid, u8 sl, u16 src_path_bits)
258 {
259         memset(ah_attr, 0, sizeof ah_attr);
260         ah_attr->dlid = dlid;
261         ah_attr->sl = sl;
262         ah_attr->src_path_bits = src_path_bits;
263         ah_attr->port_num = port_num;
264 }
265
266 static void cm_init_av_for_response(struct cm_port *port,
267                                     struct ib_wc *wc, struct cm_av *av)
268 {
269         av->port = port;
270         av->pkey_index = wc->pkey_index;
271         cm_set_ah_attr(&av->ah_attr, port->port_num, wc->slid,
272                        wc->sl, wc->dlid_path_bits);
273 }
274
275 static int cm_init_av_by_path(struct ib_sa_path_rec *path, struct cm_av *av)
276 {
277         struct cm_device *cm_dev;
278         struct cm_port *port = NULL;
279         unsigned long flags;
280         int ret;
281         u8 p;
282
283         read_lock_irqsave(&cm.device_lock, flags);
284         list_for_each_entry(cm_dev, &cm.device_list, list) {
285                 if (!ib_find_cached_gid(cm_dev->device, &path->sgid,
286                                         &p, NULL)) {
287                         port = &cm_dev->port[p-1];
288                         break;
289                 }
290         }
291         read_unlock_irqrestore(&cm.device_lock, flags);
292
293         if (!port)
294                 return -EINVAL;
295
296         ret = ib_find_cached_pkey(cm_dev->device, port->port_num,
297                                   be16_to_cpu(path->pkey), &av->pkey_index);
298         if (ret)
299                 return ret;
300
301         av->port = port;
302         cm_set_ah_attr(&av->ah_attr, av->port->port_num,
303                        be16_to_cpu(path->dlid), path->sl,
304                        be16_to_cpu(path->slid) & 0x7F);
305         av->packet_life_time = path->packet_life_time;
306         return 0;
307 }
308
309 static int cm_alloc_id(struct cm_id_private *cm_id_priv)
310 {
311         unsigned long flags;
312         int ret;
313         static int next_id;
314
315         do {
316                 spin_lock_irqsave(&cm.lock, flags);
317                 ret = idr_get_new_above(&cm.local_id_table, cm_id_priv, next_id++,
318                                         (__force int *) &cm_id_priv->id.local_id);
319                 spin_unlock_irqrestore(&cm.lock, flags);
320         } while( (ret == -EAGAIN) && idr_pre_get(&cm.local_id_table, GFP_KERNEL) );
321         return ret;
322 }
323
324 static void cm_free_id(__be32 local_id)
325 {
326         unsigned long flags;
327
328         spin_lock_irqsave(&cm.lock, flags);
329         idr_remove(&cm.local_id_table, (__force int) local_id);
330         spin_unlock_irqrestore(&cm.lock, flags);
331 }
332
333 static struct cm_id_private * cm_get_id(__be32 local_id, __be32 remote_id)
334 {
335         struct cm_id_private *cm_id_priv;
336
337         cm_id_priv = idr_find(&cm.local_id_table, (__force int) local_id);
338         if (cm_id_priv) {
339                 if (cm_id_priv->id.remote_id == remote_id)
340                         atomic_inc(&cm_id_priv->refcount);
341                 else
342                         cm_id_priv = NULL;
343         }
344
345         return cm_id_priv;
346 }
347
348 static struct cm_id_private * cm_acquire_id(__be32 local_id, __be32 remote_id)
349 {
350         struct cm_id_private *cm_id_priv;
351         unsigned long flags;
352
353         spin_lock_irqsave(&cm.lock, flags);
354         cm_id_priv = cm_get_id(local_id, remote_id);
355         spin_unlock_irqrestore(&cm.lock, flags);
356
357         return cm_id_priv;
358 }
359
360 static struct cm_id_private * cm_insert_listen(struct cm_id_private *cm_id_priv)
361 {
362         struct rb_node **link = &cm.listen_service_table.rb_node;
363         struct rb_node *parent = NULL;
364         struct cm_id_private *cur_cm_id_priv;
365         __be64 service_id = cm_id_priv->id.service_id;
366         __be64 service_mask = cm_id_priv->id.service_mask;
367
368         while (*link) {
369                 parent = *link;
370                 cur_cm_id_priv = rb_entry(parent, struct cm_id_private,
371                                           service_node);
372                 if ((cur_cm_id_priv->id.service_mask & service_id) ==
373                     (service_mask & cur_cm_id_priv->id.service_id) &&
374                     (cm_id_priv->id.device == cur_cm_id_priv->id.device))
375                         return cur_cm_id_priv;
376
377                 if (cm_id_priv->id.device < cur_cm_id_priv->id.device)
378                         link = &(*link)->rb_left;
379                 else if (cm_id_priv->id.device > cur_cm_id_priv->id.device)
380                         link = &(*link)->rb_right;
381                 else if (service_id < cur_cm_id_priv->id.service_id)
382                         link = &(*link)->rb_left;
383                 else
384                         link = &(*link)->rb_right;
385         }
386         rb_link_node(&cm_id_priv->service_node, parent, link);
387         rb_insert_color(&cm_id_priv->service_node, &cm.listen_service_table);
388         return NULL;
389 }
390
391 static struct cm_id_private * cm_find_listen(struct ib_device *device,
392                                              __be64 service_id)
393 {
394         struct rb_node *node = cm.listen_service_table.rb_node;
395         struct cm_id_private *cm_id_priv;
396
397         while (node) {
398                 cm_id_priv = rb_entry(node, struct cm_id_private, service_node);
399                 if ((cm_id_priv->id.service_mask & service_id) ==
400                      cm_id_priv->id.service_id &&
401                     (cm_id_priv->id.device == device))
402                         return cm_id_priv;
403
404                 if (device < cm_id_priv->id.device)
405                         node = node->rb_left;
406                 else if (device > cm_id_priv->id.device)
407                         node = node->rb_right;
408                 else if (service_id < cm_id_priv->id.service_id)
409                         node = node->rb_left;
410                 else
411                         node = node->rb_right;
412         }
413         return NULL;
414 }
415
416 static struct cm_timewait_info * cm_insert_remote_id(struct cm_timewait_info
417                                                      *timewait_info)
418 {
419         struct rb_node **link = &cm.remote_id_table.rb_node;
420         struct rb_node *parent = NULL;
421         struct cm_timewait_info *cur_timewait_info;
422         __be64 remote_ca_guid = timewait_info->remote_ca_guid;
423         __be32 remote_id = timewait_info->work.remote_id;
424
425         while (*link) {
426                 parent = *link;
427                 cur_timewait_info = rb_entry(parent, struct cm_timewait_info,
428                                              remote_id_node);
429                 if (remote_id < cur_timewait_info->work.remote_id)
430                         link = &(*link)->rb_left;
431                 else if (remote_id > cur_timewait_info->work.remote_id)
432                         link = &(*link)->rb_right;
433                 else if (remote_ca_guid < cur_timewait_info->remote_ca_guid)
434                         link = &(*link)->rb_left;
435                 else if (remote_ca_guid > cur_timewait_info->remote_ca_guid)
436                         link = &(*link)->rb_right;
437                 else
438                         return cur_timewait_info;
439         }
440         timewait_info->inserted_remote_id = 1;
441         rb_link_node(&timewait_info->remote_id_node, parent, link);
442         rb_insert_color(&timewait_info->remote_id_node, &cm.remote_id_table);
443         return NULL;
444 }
445
446 static struct cm_timewait_info * cm_find_remote_id(__be64 remote_ca_guid,
447                                                    __be32 remote_id)
448 {
449         struct rb_node *node = cm.remote_id_table.rb_node;
450         struct cm_timewait_info *timewait_info;
451
452         while (node) {
453                 timewait_info = rb_entry(node, struct cm_timewait_info,
454                                          remote_id_node);
455                 if (remote_id < timewait_info->work.remote_id)
456                         node = node->rb_left;
457                 else if (remote_id > timewait_info->work.remote_id)
458                         node = node->rb_right;
459                 else if (remote_ca_guid < timewait_info->remote_ca_guid)
460                         node = node->rb_left;
461                 else if (remote_ca_guid > timewait_info->remote_ca_guid)
462                         node = node->rb_right;
463                 else
464                         return timewait_info;
465         }
466         return NULL;
467 }
468
469 static struct cm_timewait_info * cm_insert_remote_qpn(struct cm_timewait_info
470                                                       *timewait_info)
471 {
472         struct rb_node **link = &cm.remote_qp_table.rb_node;
473         struct rb_node *parent = NULL;
474         struct cm_timewait_info *cur_timewait_info;
475         __be64 remote_ca_guid = timewait_info->remote_ca_guid;
476         __be32 remote_qpn = timewait_info->remote_qpn;
477
478         while (*link) {
479                 parent = *link;
480                 cur_timewait_info = rb_entry(parent, struct cm_timewait_info,
481                                              remote_qp_node);
482                 if (remote_qpn < cur_timewait_info->remote_qpn)
483                         link = &(*link)->rb_left;
484                 else if (remote_qpn > cur_timewait_info->remote_qpn)
485                         link = &(*link)->rb_right;
486                 else if (remote_ca_guid < cur_timewait_info->remote_ca_guid)
487                         link = &(*link)->rb_left;
488                 else if (remote_ca_guid > cur_timewait_info->remote_ca_guid)
489                         link = &(*link)->rb_right;
490                 else
491                         return cur_timewait_info;
492         }
493         timewait_info->inserted_remote_qp = 1;
494         rb_link_node(&timewait_info->remote_qp_node, parent, link);
495         rb_insert_color(&timewait_info->remote_qp_node, &cm.remote_qp_table);
496         return NULL;
497 }
498
499 static struct cm_id_private * cm_insert_remote_sidr(struct cm_id_private
500                                                     *cm_id_priv)
501 {
502         struct rb_node **link = &cm.remote_sidr_table.rb_node;
503         struct rb_node *parent = NULL;
504         struct cm_id_private *cur_cm_id_priv;
505         union ib_gid *port_gid = &cm_id_priv->av.dgid;
506         __be32 remote_id = cm_id_priv->id.remote_id;
507
508         while (*link) {
509                 parent = *link;
510                 cur_cm_id_priv = rb_entry(parent, struct cm_id_private,
511                                           sidr_id_node);
512                 if (remote_id < cur_cm_id_priv->id.remote_id)
513                         link = &(*link)->rb_left;
514                 else if (remote_id > cur_cm_id_priv->id.remote_id)
515                         link = &(*link)->rb_right;
516                 else {
517                         int cmp;
518                         cmp = memcmp(port_gid, &cur_cm_id_priv->av.dgid,
519                                      sizeof *port_gid);
520                         if (cmp < 0)
521                                 link = &(*link)->rb_left;
522                         else if (cmp > 0)
523                                 link = &(*link)->rb_right;
524                         else
525                                 return cur_cm_id_priv;
526                 }
527         }
528         rb_link_node(&cm_id_priv->sidr_id_node, parent, link);
529         rb_insert_color(&cm_id_priv->sidr_id_node, &cm.remote_sidr_table);
530         return NULL;
531 }
532
533 static void cm_reject_sidr_req(struct cm_id_private *cm_id_priv,
534                                enum ib_cm_sidr_status status)
535 {
536         struct ib_cm_sidr_rep_param param;
537
538         memset(&param, 0, sizeof param);
539         param.status = status;
540         ib_send_cm_sidr_rep(&cm_id_priv->id, &param);
541 }
542
543 struct ib_cm_id *ib_create_cm_id(struct ib_device *device,
544                                  ib_cm_handler cm_handler,
545                                  void *context)
546 {
547         struct cm_id_private *cm_id_priv;
548         int ret;
549
550         cm_id_priv = kzalloc(sizeof *cm_id_priv, GFP_KERNEL);
551         if (!cm_id_priv)
552                 return ERR_PTR(-ENOMEM);
553
554         cm_id_priv->id.state = IB_CM_IDLE;
555         cm_id_priv->id.device = device;
556         cm_id_priv->id.cm_handler = cm_handler;
557         cm_id_priv->id.context = context;
558         cm_id_priv->id.remote_cm_qpn = 1;
559         ret = cm_alloc_id(cm_id_priv);
560         if (ret)
561                 goto error;
562
563         spin_lock_init(&cm_id_priv->lock);
564         init_completion(&cm_id_priv->comp);
565         INIT_LIST_HEAD(&cm_id_priv->work_list);
566         atomic_set(&cm_id_priv->work_count, -1);
567         atomic_set(&cm_id_priv->refcount, 1);
568         return &cm_id_priv->id;
569
570 error:
571         kfree(cm_id_priv);
572         return ERR_PTR(-ENOMEM);
573 }
574 EXPORT_SYMBOL(ib_create_cm_id);
575
576 static struct cm_work * cm_dequeue_work(struct cm_id_private *cm_id_priv)
577 {
578         struct cm_work *work;
579
580         if (list_empty(&cm_id_priv->work_list))
581                 return NULL;
582
583         work = list_entry(cm_id_priv->work_list.next, struct cm_work, list);
584         list_del(&work->list);
585         return work;
586 }
587
588 static void cm_free_work(struct cm_work *work)
589 {
590         if (work->mad_recv_wc)
591                 ib_free_recv_mad(work->mad_recv_wc);
592         kfree(work);
593 }
594
595 static inline int cm_convert_to_ms(int iba_time)
596 {
597         /* approximate conversion to ms from 4.096us x 2^iba_time */
598         return 1 << max(iba_time - 8, 0);
599 }
600
601 static void cm_cleanup_timewait(struct cm_timewait_info *timewait_info)
602 {
603         unsigned long flags;
604
605         if (!timewait_info->inserted_remote_id &&
606             !timewait_info->inserted_remote_qp)
607             return;
608
609         spin_lock_irqsave(&cm.lock, flags);
610         if (timewait_info->inserted_remote_id) {
611                 rb_erase(&timewait_info->remote_id_node, &cm.remote_id_table);
612                 timewait_info->inserted_remote_id = 0;
613         }
614
615         if (timewait_info->inserted_remote_qp) {
616                 rb_erase(&timewait_info->remote_qp_node, &cm.remote_qp_table);
617                 timewait_info->inserted_remote_qp = 0;
618         }
619         spin_unlock_irqrestore(&cm.lock, flags);
620 }
621
622 static struct cm_timewait_info * cm_create_timewait_info(__be32 local_id)
623 {
624         struct cm_timewait_info *timewait_info;
625
626         timewait_info = kzalloc(sizeof *timewait_info, GFP_KERNEL);
627         if (!timewait_info)
628                 return ERR_PTR(-ENOMEM);
629
630         timewait_info->work.local_id = local_id;
631         INIT_WORK(&timewait_info->work.work, cm_work_handler,
632                   &timewait_info->work);
633         timewait_info->work.cm_event.event = IB_CM_TIMEWAIT_EXIT;
634         return timewait_info;
635 }
636
637 static void cm_enter_timewait(struct cm_id_private *cm_id_priv)
638 {
639         int wait_time;
640
641         /*
642          * The cm_id could be destroyed by the user before we exit timewait.
643          * To protect against this, we search for the cm_id after exiting
644          * timewait before notifying the user that we've exited timewait.
645          */
646         cm_id_priv->id.state = IB_CM_TIMEWAIT;
647         wait_time = cm_convert_to_ms(cm_id_priv->local_ack_timeout);
648         queue_delayed_work(cm.wq, &cm_id_priv->timewait_info->work.work,
649                            msecs_to_jiffies(wait_time));
650         cm_id_priv->timewait_info = NULL;
651 }
652
653 static void cm_reset_to_idle(struct cm_id_private *cm_id_priv)
654 {
655         cm_id_priv->id.state = IB_CM_IDLE;
656         if (cm_id_priv->timewait_info) {
657                 cm_cleanup_timewait(cm_id_priv->timewait_info);
658                 kfree(cm_id_priv->timewait_info);
659                 cm_id_priv->timewait_info = NULL;
660         }
661 }
662
663 void ib_destroy_cm_id(struct ib_cm_id *cm_id)
664 {
665         struct cm_id_private *cm_id_priv;
666         struct cm_work *work;
667         unsigned long flags;
668
669         cm_id_priv = container_of(cm_id, struct cm_id_private, id);
670 retest:
671         spin_lock_irqsave(&cm_id_priv->lock, flags);
672         switch (cm_id->state) {
673         case IB_CM_LISTEN:
674                 cm_id->state = IB_CM_IDLE;
675                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
676                 spin_lock_irqsave(&cm.lock, flags);
677                 rb_erase(&cm_id_priv->service_node, &cm.listen_service_table);
678                 spin_unlock_irqrestore(&cm.lock, flags);
679                 break;
680         case IB_CM_SIDR_REQ_SENT:
681                 cm_id->state = IB_CM_IDLE;
682                 ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg);
683                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
684                 break;
685         case IB_CM_SIDR_REQ_RCVD:
686                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
687                 cm_reject_sidr_req(cm_id_priv, IB_SIDR_REJECT);
688                 break;
689         case IB_CM_REQ_SENT:
690                 ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg);
691                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
692                 ib_send_cm_rej(cm_id, IB_CM_REJ_TIMEOUT,
693                                &cm_id_priv->av.port->cm_dev->ca_guid,
694                                sizeof cm_id_priv->av.port->cm_dev->ca_guid,
695                                NULL, 0);
696                 break;
697         case IB_CM_MRA_REQ_RCVD:
698         case IB_CM_REP_SENT:
699         case IB_CM_MRA_REP_RCVD:
700                 ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg);
701                 /* Fall through */
702         case IB_CM_REQ_RCVD:
703         case IB_CM_MRA_REQ_SENT:
704         case IB_CM_REP_RCVD:
705         case IB_CM_MRA_REP_SENT:
706                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
707                 ib_send_cm_rej(cm_id, IB_CM_REJ_CONSUMER_DEFINED,
708                                NULL, 0, NULL, 0);
709                 break;
710         case IB_CM_ESTABLISHED:
711                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
712                 ib_send_cm_dreq(cm_id, NULL, 0);
713                 goto retest;
714         case IB_CM_DREQ_SENT:
715                 ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg);
716                 cm_enter_timewait(cm_id_priv);
717                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
718                 break;
719         case IB_CM_DREQ_RCVD:
720                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
721                 ib_send_cm_drep(cm_id, NULL, 0);
722                 break;
723         default:
724                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
725                 break;
726         }
727
728         cm_free_id(cm_id->local_id);
729         cm_deref_id(cm_id_priv);
730         wait_for_completion(&cm_id_priv->comp);
731         while ((work = cm_dequeue_work(cm_id_priv)) != NULL)
732                 cm_free_work(work);
733         if (cm_id_priv->private_data && cm_id_priv->private_data_len)
734                 kfree(cm_id_priv->private_data);
735         kfree(cm_id_priv);
736 }
737 EXPORT_SYMBOL(ib_destroy_cm_id);
738
739 int ib_cm_listen(struct ib_cm_id *cm_id,
740                  __be64 service_id,
741                  __be64 service_mask)
742 {
743         struct cm_id_private *cm_id_priv, *cur_cm_id_priv;
744         unsigned long flags;
745         int ret = 0;
746
747         service_mask = service_mask ? service_mask :
748                        __constant_cpu_to_be64(~0ULL);
749         service_id &= service_mask;
750         if ((service_id & IB_SERVICE_ID_AGN_MASK) == IB_CM_ASSIGN_SERVICE_ID &&
751             (service_id != IB_CM_ASSIGN_SERVICE_ID))
752                 return -EINVAL;
753
754         cm_id_priv = container_of(cm_id, struct cm_id_private, id);
755         BUG_ON(cm_id->state != IB_CM_IDLE);
756
757         cm_id->state = IB_CM_LISTEN;
758
759         spin_lock_irqsave(&cm.lock, flags);
760         if (service_id == IB_CM_ASSIGN_SERVICE_ID) {
761                 cm_id->service_id = cpu_to_be64(cm.listen_service_id++);
762                 cm_id->service_mask = __constant_cpu_to_be64(~0ULL);
763         } else {
764                 cm_id->service_id = service_id;
765                 cm_id->service_mask = service_mask;
766         }
767         cur_cm_id_priv = cm_insert_listen(cm_id_priv);
768         spin_unlock_irqrestore(&cm.lock, flags);
769
770         if (cur_cm_id_priv) {
771                 cm_id->state = IB_CM_IDLE;
772                 ret = -EBUSY;
773         }
774         return ret;
775 }
776 EXPORT_SYMBOL(ib_cm_listen);
777
778 static __be64 cm_form_tid(struct cm_id_private *cm_id_priv,
779                           enum cm_msg_sequence msg_seq)
780 {
781         u64 hi_tid, low_tid;
782
783         hi_tid   = ((u64) cm_id_priv->av.port->mad_agent->hi_tid) << 32;
784         low_tid  = (u64) ((__force u32)cm_id_priv->id.local_id |
785                           (msg_seq << 30));
786         return cpu_to_be64(hi_tid | low_tid);
787 }
788
789 static void cm_format_mad_hdr(struct ib_mad_hdr *hdr,
790                               __be16 attr_id, __be64 tid)
791 {
792         hdr->base_version  = IB_MGMT_BASE_VERSION;
793         hdr->mgmt_class    = IB_MGMT_CLASS_CM;
794         hdr->class_version = IB_CM_CLASS_VERSION;
795         hdr->method        = IB_MGMT_METHOD_SEND;
796         hdr->attr_id       = attr_id;
797         hdr->tid           = tid;
798 }
799
800 static void cm_format_req(struct cm_req_msg *req_msg,
801                           struct cm_id_private *cm_id_priv,
802                           struct ib_cm_req_param *param)
803 {
804         cm_format_mad_hdr(&req_msg->hdr, CM_REQ_ATTR_ID,
805                           cm_form_tid(cm_id_priv, CM_MSG_SEQUENCE_REQ));
806
807         req_msg->local_comm_id = cm_id_priv->id.local_id;
808         req_msg->service_id = param->service_id;
809         req_msg->local_ca_guid = cm_id_priv->av.port->cm_dev->ca_guid;
810         cm_req_set_local_qpn(req_msg, cpu_to_be32(param->qp_num));
811         cm_req_set_resp_res(req_msg, param->responder_resources);
812         cm_req_set_init_depth(req_msg, param->initiator_depth);
813         cm_req_set_remote_resp_timeout(req_msg,
814                                        param->remote_cm_response_timeout);
815         cm_req_set_qp_type(req_msg, param->qp_type);
816         cm_req_set_flow_ctrl(req_msg, param->flow_control);
817         cm_req_set_starting_psn(req_msg, cpu_to_be32(param->starting_psn));
818         cm_req_set_local_resp_timeout(req_msg,
819                                       param->local_cm_response_timeout);
820         cm_req_set_retry_count(req_msg, param->retry_count);
821         req_msg->pkey = param->primary_path->pkey;
822         cm_req_set_path_mtu(req_msg, param->primary_path->mtu);
823         cm_req_set_rnr_retry_count(req_msg, param->rnr_retry_count);
824         cm_req_set_max_cm_retries(req_msg, param->max_cm_retries);
825         cm_req_set_srq(req_msg, param->srq);
826
827         req_msg->primary_local_lid = param->primary_path->slid;
828         req_msg->primary_remote_lid = param->primary_path->dlid;
829         req_msg->primary_local_gid = param->primary_path->sgid;
830         req_msg->primary_remote_gid = param->primary_path->dgid;
831         cm_req_set_primary_flow_label(req_msg, param->primary_path->flow_label);
832         cm_req_set_primary_packet_rate(req_msg, param->primary_path->rate);
833         req_msg->primary_traffic_class = param->primary_path->traffic_class;
834         req_msg->primary_hop_limit = param->primary_path->hop_limit;
835         cm_req_set_primary_sl(req_msg, param->primary_path->sl);
836         cm_req_set_primary_subnet_local(req_msg, 1); /* local only... */
837         cm_req_set_primary_local_ack_timeout(req_msg,
838                 min(31, param->primary_path->packet_life_time + 1));
839
840         if (param->alternate_path) {
841                 req_msg->alt_local_lid = param->alternate_path->slid;
842                 req_msg->alt_remote_lid = param->alternate_path->dlid;
843                 req_msg->alt_local_gid = param->alternate_path->sgid;
844                 req_msg->alt_remote_gid = param->alternate_path->dgid;
845                 cm_req_set_alt_flow_label(req_msg,
846                                           param->alternate_path->flow_label);
847                 cm_req_set_alt_packet_rate(req_msg, param->alternate_path->rate);
848                 req_msg->alt_traffic_class = param->alternate_path->traffic_class;
849                 req_msg->alt_hop_limit = param->alternate_path->hop_limit;
850                 cm_req_set_alt_sl(req_msg, param->alternate_path->sl);
851                 cm_req_set_alt_subnet_local(req_msg, 1); /* local only... */
852                 cm_req_set_alt_local_ack_timeout(req_msg,
853                         min(31, param->alternate_path->packet_life_time + 1));
854         }
855
856         if (param->private_data && param->private_data_len)
857                 memcpy(req_msg->private_data, param->private_data,
858                        param->private_data_len);
859 }
860
861 static int cm_validate_req_param(struct ib_cm_req_param *param)
862 {
863         /* peer-to-peer not supported */
864         if (param->peer_to_peer)
865                 return -EINVAL;
866
867         if (!param->primary_path)
868                 return -EINVAL;
869
870         if (param->qp_type != IB_QPT_RC && param->qp_type != IB_QPT_UC)
871                 return -EINVAL;
872
873         if (param->private_data &&
874             param->private_data_len > IB_CM_REQ_PRIVATE_DATA_SIZE)
875                 return -EINVAL;
876
877         if (param->alternate_path &&
878             (param->alternate_path->pkey != param->primary_path->pkey ||
879              param->alternate_path->mtu != param->primary_path->mtu))
880                 return -EINVAL;
881
882         return 0;
883 }
884
885 int ib_send_cm_req(struct ib_cm_id *cm_id,
886                    struct ib_cm_req_param *param)
887 {
888         struct cm_id_private *cm_id_priv;
889         struct cm_req_msg *req_msg;
890         unsigned long flags;
891         int ret;
892
893         ret = cm_validate_req_param(param);
894         if (ret)
895                 return ret;
896
897         /* Verify that we're not in timewait. */
898         cm_id_priv = container_of(cm_id, struct cm_id_private, id);
899         spin_lock_irqsave(&cm_id_priv->lock, flags);
900         if (cm_id->state != IB_CM_IDLE) {
901                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
902                 ret = -EINVAL;
903                 goto out;
904         }
905         spin_unlock_irqrestore(&cm_id_priv->lock, flags);
906
907         cm_id_priv->timewait_info = cm_create_timewait_info(cm_id_priv->
908                                                             id.local_id);
909         if (IS_ERR(cm_id_priv->timewait_info))
910                 goto out;
911
912         ret = cm_init_av_by_path(param->primary_path, &cm_id_priv->av);
913         if (ret)
914                 goto error1;
915         if (param->alternate_path) {
916                 ret = cm_init_av_by_path(param->alternate_path,
917                                          &cm_id_priv->alt_av);
918                 if (ret)
919                         goto error1;
920         }
921         cm_id->service_id = param->service_id;
922         cm_id->service_mask = __constant_cpu_to_be64(~0ULL);
923         cm_id_priv->timeout_ms = cm_convert_to_ms(
924                                     param->primary_path->packet_life_time) * 2 +
925                                  cm_convert_to_ms(
926                                     param->remote_cm_response_timeout);
927         cm_id_priv->max_cm_retries = param->max_cm_retries;
928         cm_id_priv->initiator_depth = param->initiator_depth;
929         cm_id_priv->responder_resources = param->responder_resources;
930         cm_id_priv->retry_count = param->retry_count;
931         cm_id_priv->path_mtu = param->primary_path->mtu;
932         cm_id_priv->qp_type = param->qp_type;
933
934         ret = cm_alloc_msg(cm_id_priv, &cm_id_priv->msg);
935         if (ret)
936                 goto error1;
937
938         req_msg = (struct cm_req_msg *) cm_id_priv->msg->mad;
939         cm_format_req(req_msg, cm_id_priv, param);
940         cm_id_priv->tid = req_msg->hdr.tid;
941         cm_id_priv->msg->timeout_ms = cm_id_priv->timeout_ms;
942         cm_id_priv->msg->context[1] = (void *) (unsigned long) IB_CM_REQ_SENT;
943
944         cm_id_priv->local_qpn = cm_req_get_local_qpn(req_msg);
945         cm_id_priv->rq_psn = cm_req_get_starting_psn(req_msg);
946         cm_id_priv->local_ack_timeout =
947                                 cm_req_get_primary_local_ack_timeout(req_msg);
948
949         spin_lock_irqsave(&cm_id_priv->lock, flags);
950         ret = ib_post_send_mad(cm_id_priv->msg, NULL);
951         if (ret) {
952                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
953                 goto error2;
954         }
955         BUG_ON(cm_id->state != IB_CM_IDLE);
956         cm_id->state = IB_CM_REQ_SENT;
957         spin_unlock_irqrestore(&cm_id_priv->lock, flags);
958         return 0;
959
960 error2: cm_free_msg(cm_id_priv->msg);
961 error1: kfree(cm_id_priv->timewait_info);
962 out:    return ret;
963 }
964 EXPORT_SYMBOL(ib_send_cm_req);
965
966 static int cm_issue_rej(struct cm_port *port,
967                         struct ib_mad_recv_wc *mad_recv_wc,
968                         enum ib_cm_rej_reason reason,
969                         enum cm_msg_response msg_rejected,
970                         void *ari, u8 ari_length)
971 {
972         struct ib_mad_send_buf *msg = NULL;
973         struct cm_rej_msg *rej_msg, *rcv_msg;
974         int ret;
975
976         ret = cm_alloc_response_msg(port, mad_recv_wc, &msg);
977         if (ret)
978                 return ret;
979
980         /* We just need common CM header information.  Cast to any message. */
981         rcv_msg = (struct cm_rej_msg *) mad_recv_wc->recv_buf.mad;
982         rej_msg = (struct cm_rej_msg *) msg->mad;
983
984         cm_format_mad_hdr(&rej_msg->hdr, CM_REJ_ATTR_ID, rcv_msg->hdr.tid);
985         rej_msg->remote_comm_id = rcv_msg->local_comm_id;
986         rej_msg->local_comm_id = rcv_msg->remote_comm_id;
987         cm_rej_set_msg_rejected(rej_msg, msg_rejected);
988         rej_msg->reason = cpu_to_be16(reason);
989
990         if (ari && ari_length) {
991                 cm_rej_set_reject_info_len(rej_msg, ari_length);
992                 memcpy(rej_msg->ari, ari, ari_length);
993         }
994
995         ret = ib_post_send_mad(msg, NULL);
996         if (ret)
997                 cm_free_msg(msg);
998
999         return ret;
1000 }
1001
1002 static inline int cm_is_active_peer(__be64 local_ca_guid, __be64 remote_ca_guid,
1003                                     __be32 local_qpn, __be32 remote_qpn)
1004 {
1005         return (be64_to_cpu(local_ca_guid) > be64_to_cpu(remote_ca_guid) ||
1006                 ((local_ca_guid == remote_ca_guid) &&
1007                  (be32_to_cpu(local_qpn) > be32_to_cpu(remote_qpn))));
1008 }
1009
1010 static void cm_format_paths_from_req(struct cm_req_msg *req_msg,
1011                                             struct ib_sa_path_rec *primary_path,
1012                                             struct ib_sa_path_rec *alt_path)
1013 {
1014         memset(primary_path, 0, sizeof *primary_path);
1015         primary_path->dgid = req_msg->primary_local_gid;
1016         primary_path->sgid = req_msg->primary_remote_gid;
1017         primary_path->dlid = req_msg->primary_local_lid;
1018         primary_path->slid = req_msg->primary_remote_lid;
1019         primary_path->flow_label = cm_req_get_primary_flow_label(req_msg);
1020         primary_path->hop_limit = req_msg->primary_hop_limit;
1021         primary_path->traffic_class = req_msg->primary_traffic_class;
1022         primary_path->reversible = 1;
1023         primary_path->pkey = req_msg->pkey;
1024         primary_path->sl = cm_req_get_primary_sl(req_msg);
1025         primary_path->mtu_selector = IB_SA_EQ;
1026         primary_path->mtu = cm_req_get_path_mtu(req_msg);
1027         primary_path->rate_selector = IB_SA_EQ;
1028         primary_path->rate = cm_req_get_primary_packet_rate(req_msg);
1029         primary_path->packet_life_time_selector = IB_SA_EQ;
1030         primary_path->packet_life_time =
1031                 cm_req_get_primary_local_ack_timeout(req_msg);
1032         primary_path->packet_life_time -= (primary_path->packet_life_time > 0);
1033
1034         if (req_msg->alt_local_lid) {
1035                 memset(alt_path, 0, sizeof *alt_path);
1036                 alt_path->dgid = req_msg->alt_local_gid;
1037                 alt_path->sgid = req_msg->alt_remote_gid;
1038                 alt_path->dlid = req_msg->alt_local_lid;
1039                 alt_path->slid = req_msg->alt_remote_lid;
1040                 alt_path->flow_label = cm_req_get_alt_flow_label(req_msg);
1041                 alt_path->hop_limit = req_msg->alt_hop_limit;
1042                 alt_path->traffic_class = req_msg->alt_traffic_class;
1043                 alt_path->reversible = 1;
1044                 alt_path->pkey = req_msg->pkey;
1045                 alt_path->sl = cm_req_get_alt_sl(req_msg);
1046                 alt_path->mtu_selector = IB_SA_EQ;
1047                 alt_path->mtu = cm_req_get_path_mtu(req_msg);
1048                 alt_path->rate_selector = IB_SA_EQ;
1049                 alt_path->rate = cm_req_get_alt_packet_rate(req_msg);
1050                 alt_path->packet_life_time_selector = IB_SA_EQ;
1051                 alt_path->packet_life_time =
1052                         cm_req_get_alt_local_ack_timeout(req_msg);
1053                 alt_path->packet_life_time -= (alt_path->packet_life_time > 0);
1054         }
1055 }
1056
1057 static void cm_format_req_event(struct cm_work *work,
1058                                 struct cm_id_private *cm_id_priv,
1059                                 struct ib_cm_id *listen_id)
1060 {
1061         struct cm_req_msg *req_msg;
1062         struct ib_cm_req_event_param *param;
1063
1064         req_msg = (struct cm_req_msg *)work->mad_recv_wc->recv_buf.mad;
1065         param = &work->cm_event.param.req_rcvd;
1066         param->listen_id = listen_id;
1067         param->port = cm_id_priv->av.port->port_num;
1068         param->primary_path = &work->path[0];
1069         if (req_msg->alt_local_lid)
1070                 param->alternate_path = &work->path[1];
1071         else
1072                 param->alternate_path = NULL;
1073         param->remote_ca_guid = req_msg->local_ca_guid;
1074         param->remote_qkey = be32_to_cpu(req_msg->local_qkey);
1075         param->remote_qpn = be32_to_cpu(cm_req_get_local_qpn(req_msg));
1076         param->qp_type = cm_req_get_qp_type(req_msg);
1077         param->starting_psn = be32_to_cpu(cm_req_get_starting_psn(req_msg));
1078         param->responder_resources = cm_req_get_init_depth(req_msg);
1079         param->initiator_depth = cm_req_get_resp_res(req_msg);
1080         param->local_cm_response_timeout =
1081                                         cm_req_get_remote_resp_timeout(req_msg);
1082         param->flow_control = cm_req_get_flow_ctrl(req_msg);
1083         param->remote_cm_response_timeout =
1084                                         cm_req_get_local_resp_timeout(req_msg);
1085         param->retry_count = cm_req_get_retry_count(req_msg);
1086         param->rnr_retry_count = cm_req_get_rnr_retry_count(req_msg);
1087         param->srq = cm_req_get_srq(req_msg);
1088         work->cm_event.private_data = &req_msg->private_data;
1089 }
1090
1091 static void cm_process_work(struct cm_id_private *cm_id_priv,
1092                             struct cm_work *work)
1093 {
1094         unsigned long flags;
1095         int ret;
1096
1097         /* We will typically only have the current event to report. */
1098         ret = cm_id_priv->id.cm_handler(&cm_id_priv->id, &work->cm_event);
1099         cm_free_work(work);
1100
1101         while (!ret && !atomic_add_negative(-1, &cm_id_priv->work_count)) {
1102                 spin_lock_irqsave(&cm_id_priv->lock, flags);
1103                 work = cm_dequeue_work(cm_id_priv);
1104                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1105                 BUG_ON(!work);
1106                 ret = cm_id_priv->id.cm_handler(&cm_id_priv->id,
1107                                                 &work->cm_event);
1108                 cm_free_work(work);
1109         }
1110         cm_deref_id(cm_id_priv);
1111         if (ret)
1112                 ib_destroy_cm_id(&cm_id_priv->id);
1113 }
1114
1115 static void cm_format_mra(struct cm_mra_msg *mra_msg,
1116                           struct cm_id_private *cm_id_priv,
1117                           enum cm_msg_response msg_mraed, u8 service_timeout,
1118                           const void *private_data, u8 private_data_len)
1119 {
1120         cm_format_mad_hdr(&mra_msg->hdr, CM_MRA_ATTR_ID, cm_id_priv->tid);
1121         cm_mra_set_msg_mraed(mra_msg, msg_mraed);
1122         mra_msg->local_comm_id = cm_id_priv->id.local_id;
1123         mra_msg->remote_comm_id = cm_id_priv->id.remote_id;
1124         cm_mra_set_service_timeout(mra_msg, service_timeout);
1125
1126         if (private_data && private_data_len)
1127                 memcpy(mra_msg->private_data, private_data, private_data_len);
1128 }
1129
1130 static void cm_format_rej(struct cm_rej_msg *rej_msg,
1131                           struct cm_id_private *cm_id_priv,
1132                           enum ib_cm_rej_reason reason,
1133                           void *ari,
1134                           u8 ari_length,
1135                           const void *private_data,
1136                           u8 private_data_len)
1137 {
1138         cm_format_mad_hdr(&rej_msg->hdr, CM_REJ_ATTR_ID, cm_id_priv->tid);
1139         rej_msg->remote_comm_id = cm_id_priv->id.remote_id;
1140
1141         switch(cm_id_priv->id.state) {
1142         case IB_CM_REQ_RCVD:
1143                 rej_msg->local_comm_id = 0;
1144                 cm_rej_set_msg_rejected(rej_msg, CM_MSG_RESPONSE_REQ);
1145                 break;
1146         case IB_CM_MRA_REQ_SENT:
1147                 rej_msg->local_comm_id = cm_id_priv->id.local_id;
1148                 cm_rej_set_msg_rejected(rej_msg, CM_MSG_RESPONSE_REQ);
1149                 break;
1150         case IB_CM_REP_RCVD:
1151         case IB_CM_MRA_REP_SENT:
1152                 rej_msg->local_comm_id = cm_id_priv->id.local_id;
1153                 cm_rej_set_msg_rejected(rej_msg, CM_MSG_RESPONSE_REP);
1154                 break;
1155         default:
1156                 rej_msg->local_comm_id = cm_id_priv->id.local_id;
1157                 cm_rej_set_msg_rejected(rej_msg, CM_MSG_RESPONSE_OTHER);
1158                 break;
1159         }
1160
1161         rej_msg->reason = cpu_to_be16(reason);
1162         if (ari && ari_length) {
1163                 cm_rej_set_reject_info_len(rej_msg, ari_length);
1164                 memcpy(rej_msg->ari, ari, ari_length);
1165         }
1166
1167         if (private_data && private_data_len)
1168                 memcpy(rej_msg->private_data, private_data, private_data_len);
1169 }
1170
1171 static void cm_dup_req_handler(struct cm_work *work,
1172                                struct cm_id_private *cm_id_priv)
1173 {
1174         struct ib_mad_send_buf *msg = NULL;
1175         unsigned long flags;
1176         int ret;
1177
1178         /* Quick state check to discard duplicate REQs. */
1179         if (cm_id_priv->id.state == IB_CM_REQ_RCVD)
1180                 return;
1181
1182         ret = cm_alloc_response_msg(work->port, work->mad_recv_wc, &msg);
1183         if (ret)
1184                 return;
1185
1186         spin_lock_irqsave(&cm_id_priv->lock, flags);
1187         switch (cm_id_priv->id.state) {
1188         case IB_CM_MRA_REQ_SENT:
1189                 cm_format_mra((struct cm_mra_msg *) msg->mad, cm_id_priv,
1190                               CM_MSG_RESPONSE_REQ, cm_id_priv->service_timeout,
1191                               cm_id_priv->private_data,
1192                               cm_id_priv->private_data_len);
1193                 break;
1194         case IB_CM_TIMEWAIT:
1195                 cm_format_rej((struct cm_rej_msg *) msg->mad, cm_id_priv,
1196                               IB_CM_REJ_STALE_CONN, NULL, 0, NULL, 0);
1197                 break;
1198         default:
1199                 goto unlock;
1200         }
1201         spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1202
1203         ret = ib_post_send_mad(msg, NULL);
1204         if (ret)
1205                 goto free;
1206         return;
1207
1208 unlock: spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1209 free:   cm_free_msg(msg);
1210 }
1211
1212 static struct cm_id_private * cm_match_req(struct cm_work *work,
1213                                            struct cm_id_private *cm_id_priv)
1214 {
1215         struct cm_id_private *listen_cm_id_priv, *cur_cm_id_priv;
1216         struct cm_timewait_info *timewait_info;
1217         struct cm_req_msg *req_msg;
1218         unsigned long flags;
1219
1220         req_msg = (struct cm_req_msg *)work->mad_recv_wc->recv_buf.mad;
1221
1222         /* Check for duplicate REQ and stale connections. */
1223         spin_lock_irqsave(&cm.lock, flags);
1224         timewait_info = cm_insert_remote_id(cm_id_priv->timewait_info);
1225         if (!timewait_info)
1226                 timewait_info = cm_insert_remote_qpn(cm_id_priv->timewait_info);
1227
1228         if (timewait_info) {
1229                 cur_cm_id_priv = cm_get_id(timewait_info->work.local_id,
1230                                            timewait_info->work.remote_id);
1231                 spin_unlock_irqrestore(&cm.lock, flags);
1232                 if (cur_cm_id_priv) {
1233                         cm_dup_req_handler(work, cur_cm_id_priv);
1234                         cm_deref_id(cur_cm_id_priv);
1235                 } else
1236                         cm_issue_rej(work->port, work->mad_recv_wc,
1237                                      IB_CM_REJ_STALE_CONN, CM_MSG_RESPONSE_REQ,
1238                                      NULL, 0);
1239                 goto error;
1240         }
1241
1242         /* Find matching listen request. */
1243         listen_cm_id_priv = cm_find_listen(cm_id_priv->id.device,
1244                                            req_msg->service_id);
1245         if (!listen_cm_id_priv) {
1246                 spin_unlock_irqrestore(&cm.lock, flags);
1247                 cm_issue_rej(work->port, work->mad_recv_wc,
1248                              IB_CM_REJ_INVALID_SERVICE_ID, CM_MSG_RESPONSE_REQ,
1249                              NULL, 0);
1250                 goto error;
1251         }
1252         atomic_inc(&listen_cm_id_priv->refcount);
1253         atomic_inc(&cm_id_priv->refcount);
1254         cm_id_priv->id.state = IB_CM_REQ_RCVD;
1255         atomic_inc(&cm_id_priv->work_count);
1256         spin_unlock_irqrestore(&cm.lock, flags);
1257         return listen_cm_id_priv;
1258
1259 error:  cm_cleanup_timewait(cm_id_priv->timewait_info);
1260         return NULL;
1261 }
1262
1263 static int cm_req_handler(struct cm_work *work)
1264 {
1265         struct ib_cm_id *cm_id;
1266         struct cm_id_private *cm_id_priv, *listen_cm_id_priv;
1267         struct cm_req_msg *req_msg;
1268         int ret;
1269
1270         req_msg = (struct cm_req_msg *)work->mad_recv_wc->recv_buf.mad;
1271
1272         cm_id = ib_create_cm_id(work->port->cm_dev->device, NULL, NULL);
1273         if (IS_ERR(cm_id))
1274                 return PTR_ERR(cm_id);
1275
1276         cm_id_priv = container_of(cm_id, struct cm_id_private, id);
1277         cm_id_priv->id.remote_id = req_msg->local_comm_id;
1278         cm_init_av_for_response(work->port, work->mad_recv_wc->wc,
1279                                 &cm_id_priv->av);
1280         cm_id_priv->timewait_info = cm_create_timewait_info(cm_id_priv->
1281                                                             id.local_id);
1282         if (IS_ERR(cm_id_priv->timewait_info)) {
1283                 ret = PTR_ERR(cm_id_priv->timewait_info);
1284                 goto error1;
1285         }
1286         cm_id_priv->timewait_info->work.remote_id = req_msg->local_comm_id;
1287         cm_id_priv->timewait_info->remote_ca_guid = req_msg->local_ca_guid;
1288         cm_id_priv->timewait_info->remote_qpn = cm_req_get_local_qpn(req_msg);
1289
1290         listen_cm_id_priv = cm_match_req(work, cm_id_priv);
1291         if (!listen_cm_id_priv) {
1292                 ret = -EINVAL;
1293                 goto error2;
1294         }
1295
1296         cm_id_priv->id.cm_handler = listen_cm_id_priv->id.cm_handler;
1297         cm_id_priv->id.context = listen_cm_id_priv->id.context;
1298         cm_id_priv->id.service_id = req_msg->service_id;
1299         cm_id_priv->id.service_mask = __constant_cpu_to_be64(~0ULL);
1300
1301         cm_format_paths_from_req(req_msg, &work->path[0], &work->path[1]);
1302         ret = cm_init_av_by_path(&work->path[0], &cm_id_priv->av);
1303         if (ret)
1304                 goto error3;
1305         if (req_msg->alt_local_lid) {
1306                 ret = cm_init_av_by_path(&work->path[1], &cm_id_priv->alt_av);
1307                 if (ret)
1308                         goto error3;
1309         }
1310         cm_id_priv->tid = req_msg->hdr.tid;
1311         cm_id_priv->timeout_ms = cm_convert_to_ms(
1312                                         cm_req_get_local_resp_timeout(req_msg));
1313         cm_id_priv->max_cm_retries = cm_req_get_max_cm_retries(req_msg);
1314         cm_id_priv->remote_qpn = cm_req_get_local_qpn(req_msg);
1315         cm_id_priv->initiator_depth = cm_req_get_resp_res(req_msg);
1316         cm_id_priv->responder_resources = cm_req_get_init_depth(req_msg);
1317         cm_id_priv->path_mtu = cm_req_get_path_mtu(req_msg);
1318         cm_id_priv->sq_psn = cm_req_get_starting_psn(req_msg);
1319         cm_id_priv->local_ack_timeout =
1320                                 cm_req_get_primary_local_ack_timeout(req_msg);
1321         cm_id_priv->retry_count = cm_req_get_retry_count(req_msg);
1322         cm_id_priv->rnr_retry_count = cm_req_get_rnr_retry_count(req_msg);
1323         cm_id_priv->qp_type = cm_req_get_qp_type(req_msg);
1324
1325         cm_format_req_event(work, cm_id_priv, &listen_cm_id_priv->id);
1326         cm_process_work(cm_id_priv, work);
1327         cm_deref_id(listen_cm_id_priv);
1328         return 0;
1329
1330 error3: atomic_dec(&cm_id_priv->refcount);
1331         cm_deref_id(listen_cm_id_priv);
1332         cm_cleanup_timewait(cm_id_priv->timewait_info);
1333 error2: kfree(cm_id_priv->timewait_info);
1334         cm_id_priv->timewait_info = NULL;
1335 error1: ib_destroy_cm_id(&cm_id_priv->id);
1336         return ret;
1337 }
1338
1339 static void cm_format_rep(struct cm_rep_msg *rep_msg,
1340                           struct cm_id_private *cm_id_priv,
1341                           struct ib_cm_rep_param *param)
1342 {
1343         cm_format_mad_hdr(&rep_msg->hdr, CM_REP_ATTR_ID, cm_id_priv->tid);
1344         rep_msg->local_comm_id = cm_id_priv->id.local_id;
1345         rep_msg->remote_comm_id = cm_id_priv->id.remote_id;
1346         cm_rep_set_local_qpn(rep_msg, cpu_to_be32(param->qp_num));
1347         cm_rep_set_starting_psn(rep_msg, cpu_to_be32(param->starting_psn));
1348         rep_msg->resp_resources = param->responder_resources;
1349         rep_msg->initiator_depth = param->initiator_depth;
1350         cm_rep_set_target_ack_delay(rep_msg, param->target_ack_delay);
1351         cm_rep_set_failover(rep_msg, param->failover_accepted);
1352         cm_rep_set_flow_ctrl(rep_msg, param->flow_control);
1353         cm_rep_set_rnr_retry_count(rep_msg, param->rnr_retry_count);
1354         cm_rep_set_srq(rep_msg, param->srq);
1355         rep_msg->local_ca_guid = cm_id_priv->av.port->cm_dev->ca_guid;
1356
1357         if (param->private_data && param->private_data_len)
1358                 memcpy(rep_msg->private_data, param->private_data,
1359                        param->private_data_len);
1360 }
1361
1362 int ib_send_cm_rep(struct ib_cm_id *cm_id,
1363                    struct ib_cm_rep_param *param)
1364 {
1365         struct cm_id_private *cm_id_priv;
1366         struct ib_mad_send_buf *msg;
1367         struct cm_rep_msg *rep_msg;
1368         unsigned long flags;
1369         int ret;
1370
1371         if (param->private_data &&
1372             param->private_data_len > IB_CM_REP_PRIVATE_DATA_SIZE)
1373                 return -EINVAL;
1374
1375         cm_id_priv = container_of(cm_id, struct cm_id_private, id);
1376         spin_lock_irqsave(&cm_id_priv->lock, flags);
1377         if (cm_id->state != IB_CM_REQ_RCVD &&
1378             cm_id->state != IB_CM_MRA_REQ_SENT) {
1379                 ret = -EINVAL;
1380                 goto out;
1381         }
1382
1383         ret = cm_alloc_msg(cm_id_priv, &msg);
1384         if (ret)
1385                 goto out;
1386
1387         rep_msg = (struct cm_rep_msg *) msg->mad;
1388         cm_format_rep(rep_msg, cm_id_priv, param);
1389         msg->timeout_ms = cm_id_priv->timeout_ms;
1390         msg->context[1] = (void *) (unsigned long) IB_CM_REP_SENT;
1391
1392         ret = ib_post_send_mad(msg, NULL);
1393         if (ret) {
1394                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1395                 cm_free_msg(msg);
1396                 return ret;
1397         }
1398
1399         cm_id->state = IB_CM_REP_SENT;
1400         cm_id_priv->msg = msg;
1401         cm_id_priv->initiator_depth = param->initiator_depth;
1402         cm_id_priv->responder_resources = param->responder_resources;
1403         cm_id_priv->rq_psn = cm_rep_get_starting_psn(rep_msg);
1404         cm_id_priv->local_qpn = cm_rep_get_local_qpn(rep_msg);
1405
1406 out:    spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1407         return ret;
1408 }
1409 EXPORT_SYMBOL(ib_send_cm_rep);
1410
1411 static void cm_format_rtu(struct cm_rtu_msg *rtu_msg,
1412                           struct cm_id_private *cm_id_priv,
1413                           const void *private_data,
1414                           u8 private_data_len)
1415 {
1416         cm_format_mad_hdr(&rtu_msg->hdr, CM_RTU_ATTR_ID, cm_id_priv->tid);
1417         rtu_msg->local_comm_id = cm_id_priv->id.local_id;
1418         rtu_msg->remote_comm_id = cm_id_priv->id.remote_id;
1419
1420         if (private_data && private_data_len)
1421                 memcpy(rtu_msg->private_data, private_data, private_data_len);
1422 }
1423
1424 int ib_send_cm_rtu(struct ib_cm_id *cm_id,
1425                    const void *private_data,
1426                    u8 private_data_len)
1427 {
1428         struct cm_id_private *cm_id_priv;
1429         struct ib_mad_send_buf *msg;
1430         unsigned long flags;
1431         void *data;
1432         int ret;
1433
1434         if (private_data && private_data_len > IB_CM_RTU_PRIVATE_DATA_SIZE)
1435                 return -EINVAL;
1436
1437         data = cm_copy_private_data(private_data, private_data_len);
1438         if (IS_ERR(data))
1439                 return PTR_ERR(data);
1440
1441         cm_id_priv = container_of(cm_id, struct cm_id_private, id);
1442         spin_lock_irqsave(&cm_id_priv->lock, flags);
1443         if (cm_id->state != IB_CM_REP_RCVD &&
1444             cm_id->state != IB_CM_MRA_REP_SENT) {
1445                 ret = -EINVAL;
1446                 goto error;
1447         }
1448
1449         ret = cm_alloc_msg(cm_id_priv, &msg);
1450         if (ret)
1451                 goto error;
1452
1453         cm_format_rtu((struct cm_rtu_msg *) msg->mad, cm_id_priv,
1454                       private_data, private_data_len);
1455
1456         ret = ib_post_send_mad(msg, NULL);
1457         if (ret) {
1458                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1459                 cm_free_msg(msg);
1460                 kfree(data);
1461                 return ret;
1462         }
1463
1464         cm_id->state = IB_CM_ESTABLISHED;
1465         cm_set_private_data(cm_id_priv, data, private_data_len);
1466         spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1467         return 0;
1468
1469 error:  spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1470         kfree(data);
1471         return ret;
1472 }
1473 EXPORT_SYMBOL(ib_send_cm_rtu);
1474
1475 static void cm_format_rep_event(struct cm_work *work)
1476 {
1477         struct cm_rep_msg *rep_msg;
1478         struct ib_cm_rep_event_param *param;
1479
1480         rep_msg = (struct cm_rep_msg *)work->mad_recv_wc->recv_buf.mad;
1481         param = &work->cm_event.param.rep_rcvd;
1482         param->remote_ca_guid = rep_msg->local_ca_guid;
1483         param->remote_qkey = be32_to_cpu(rep_msg->local_qkey);
1484         param->remote_qpn = be32_to_cpu(cm_rep_get_local_qpn(rep_msg));
1485         param->starting_psn = be32_to_cpu(cm_rep_get_starting_psn(rep_msg));
1486         param->responder_resources = rep_msg->initiator_depth;
1487         param->initiator_depth = rep_msg->resp_resources;
1488         param->target_ack_delay = cm_rep_get_target_ack_delay(rep_msg);
1489         param->failover_accepted = cm_rep_get_failover(rep_msg);
1490         param->flow_control = cm_rep_get_flow_ctrl(rep_msg);
1491         param->rnr_retry_count = cm_rep_get_rnr_retry_count(rep_msg);
1492         param->srq = cm_rep_get_srq(rep_msg);
1493         work->cm_event.private_data = &rep_msg->private_data;
1494 }
1495
1496 static void cm_dup_rep_handler(struct cm_work *work)
1497 {
1498         struct cm_id_private *cm_id_priv;
1499         struct cm_rep_msg *rep_msg;
1500         struct ib_mad_send_buf *msg = NULL;
1501         unsigned long flags;
1502         int ret;
1503
1504         rep_msg = (struct cm_rep_msg *) work->mad_recv_wc->recv_buf.mad;
1505         cm_id_priv = cm_acquire_id(rep_msg->remote_comm_id,
1506                                    rep_msg->local_comm_id);
1507         if (!cm_id_priv)
1508                 return;
1509
1510         ret = cm_alloc_response_msg(work->port, work->mad_recv_wc, &msg);
1511         if (ret)
1512                 goto deref;
1513
1514         spin_lock_irqsave(&cm_id_priv->lock, flags);
1515         if (cm_id_priv->id.state == IB_CM_ESTABLISHED)
1516                 cm_format_rtu((struct cm_rtu_msg *) msg->mad, cm_id_priv,
1517                               cm_id_priv->private_data,
1518                               cm_id_priv->private_data_len);
1519         else if (cm_id_priv->id.state == IB_CM_MRA_REP_SENT)
1520                 cm_format_mra((struct cm_mra_msg *) msg->mad, cm_id_priv,
1521                               CM_MSG_RESPONSE_REP, cm_id_priv->service_timeout,
1522                               cm_id_priv->private_data,
1523                               cm_id_priv->private_data_len);
1524         else
1525                 goto unlock;
1526         spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1527
1528         ret = ib_post_send_mad(msg, NULL);
1529         if (ret)
1530                 goto free;
1531         goto deref;
1532
1533 unlock: spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1534 free:   cm_free_msg(msg);
1535 deref:  cm_deref_id(cm_id_priv);
1536 }
1537
1538 static int cm_rep_handler(struct cm_work *work)
1539 {
1540         struct cm_id_private *cm_id_priv;
1541         struct cm_rep_msg *rep_msg;
1542         unsigned long flags;
1543         int ret;
1544
1545         rep_msg = (struct cm_rep_msg *)work->mad_recv_wc->recv_buf.mad;
1546         cm_id_priv = cm_acquire_id(rep_msg->remote_comm_id, 0);
1547         if (!cm_id_priv) {
1548                 cm_dup_rep_handler(work);
1549                 return -EINVAL;
1550         }
1551
1552         cm_format_rep_event(work);
1553
1554         spin_lock_irqsave(&cm_id_priv->lock, flags);
1555         switch (cm_id_priv->id.state) {
1556         case IB_CM_REQ_SENT:
1557         case IB_CM_MRA_REQ_RCVD:
1558                 break;
1559         default:
1560                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1561                 ret = -EINVAL;
1562                 goto error;
1563         }
1564
1565         cm_id_priv->timewait_info->work.remote_id = rep_msg->local_comm_id;
1566         cm_id_priv->timewait_info->remote_ca_guid = rep_msg->local_ca_guid;
1567         cm_id_priv->timewait_info->remote_qpn = cm_rep_get_local_qpn(rep_msg);
1568
1569         spin_lock(&cm.lock);
1570         /* Check for duplicate REP. */
1571         if (cm_insert_remote_id(cm_id_priv->timewait_info)) {
1572                 spin_unlock(&cm.lock);
1573                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1574                 ret = -EINVAL;
1575                 goto error;
1576         }
1577         /* Check for a stale connection. */
1578         if (cm_insert_remote_qpn(cm_id_priv->timewait_info)) {
1579                 rb_erase(&cm_id_priv->timewait_info->remote_id_node,
1580                          &cm.remote_id_table);
1581                 cm_id_priv->timewait_info->inserted_remote_id = 0;
1582                 spin_unlock(&cm.lock);
1583                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1584                 cm_issue_rej(work->port, work->mad_recv_wc,
1585                              IB_CM_REJ_STALE_CONN, CM_MSG_RESPONSE_REP,
1586                              NULL, 0);
1587                 ret = -EINVAL;
1588                 goto error;
1589         }
1590         spin_unlock(&cm.lock);
1591
1592         cm_id_priv->id.state = IB_CM_REP_RCVD;
1593         cm_id_priv->id.remote_id = rep_msg->local_comm_id;
1594         cm_id_priv->remote_qpn = cm_rep_get_local_qpn(rep_msg);
1595         cm_id_priv->initiator_depth = rep_msg->resp_resources;
1596         cm_id_priv->responder_resources = rep_msg->initiator_depth;
1597         cm_id_priv->sq_psn = cm_rep_get_starting_psn(rep_msg);
1598         cm_id_priv->rnr_retry_count = cm_rep_get_rnr_retry_count(rep_msg);
1599
1600         /* todo: handle peer_to_peer */
1601
1602         ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg);
1603         ret = atomic_inc_and_test(&cm_id_priv->work_count);
1604         if (!ret)
1605                 list_add_tail(&work->list, &cm_id_priv->work_list);
1606         spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1607
1608         if (ret)
1609                 cm_process_work(cm_id_priv, work);
1610         else
1611                 cm_deref_id(cm_id_priv);
1612         return 0;
1613
1614 error:
1615         cm_deref_id(cm_id_priv);
1616         return ret;
1617 }
1618
1619 static int cm_establish_handler(struct cm_work *work)
1620 {
1621         struct cm_id_private *cm_id_priv;
1622         unsigned long flags;
1623         int ret;
1624
1625         /* See comment in ib_cm_establish about lookup. */
1626         cm_id_priv = cm_acquire_id(work->local_id, work->remote_id);
1627         if (!cm_id_priv)
1628                 return -EINVAL;
1629
1630         spin_lock_irqsave(&cm_id_priv->lock, flags);
1631         if (cm_id_priv->id.state != IB_CM_ESTABLISHED) {
1632                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1633                 goto out;
1634         }
1635
1636         ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg);
1637         ret = atomic_inc_and_test(&cm_id_priv->work_count);
1638         if (!ret)
1639                 list_add_tail(&work->list, &cm_id_priv->work_list);
1640         spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1641
1642         if (ret)
1643                 cm_process_work(cm_id_priv, work);
1644         else
1645                 cm_deref_id(cm_id_priv);
1646         return 0;
1647 out:
1648         cm_deref_id(cm_id_priv);
1649         return -EINVAL;
1650 }
1651
1652 static int cm_rtu_handler(struct cm_work *work)
1653 {
1654         struct cm_id_private *cm_id_priv;
1655         struct cm_rtu_msg *rtu_msg;
1656         unsigned long flags;
1657         int ret;
1658
1659         rtu_msg = (struct cm_rtu_msg *)work->mad_recv_wc->recv_buf.mad;
1660         cm_id_priv = cm_acquire_id(rtu_msg->remote_comm_id,
1661                                    rtu_msg->local_comm_id);
1662         if (!cm_id_priv)
1663                 return -EINVAL;
1664
1665         work->cm_event.private_data = &rtu_msg->private_data;
1666
1667         spin_lock_irqsave(&cm_id_priv->lock, flags);
1668         if (cm_id_priv->id.state != IB_CM_REP_SENT &&
1669             cm_id_priv->id.state != IB_CM_MRA_REP_RCVD) {
1670                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1671                 goto out;
1672         }
1673         cm_id_priv->id.state = IB_CM_ESTABLISHED;
1674
1675         ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg);
1676         ret = atomic_inc_and_test(&cm_id_priv->work_count);
1677         if (!ret)
1678                 list_add_tail(&work->list, &cm_id_priv->work_list);
1679         spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1680
1681         if (ret)
1682                 cm_process_work(cm_id_priv, work);
1683         else
1684                 cm_deref_id(cm_id_priv);
1685         return 0;
1686 out:
1687         cm_deref_id(cm_id_priv);
1688         return -EINVAL;
1689 }
1690
1691 static void cm_format_dreq(struct cm_dreq_msg *dreq_msg,
1692                           struct cm_id_private *cm_id_priv,
1693                           const void *private_data,
1694                           u8 private_data_len)
1695 {
1696         cm_format_mad_hdr(&dreq_msg->hdr, CM_DREQ_ATTR_ID,
1697                           cm_form_tid(cm_id_priv, CM_MSG_SEQUENCE_DREQ));
1698         dreq_msg->local_comm_id = cm_id_priv->id.local_id;
1699         dreq_msg->remote_comm_id = cm_id_priv->id.remote_id;
1700         cm_dreq_set_remote_qpn(dreq_msg, cm_id_priv->remote_qpn);
1701
1702         if (private_data && private_data_len)
1703                 memcpy(dreq_msg->private_data, private_data, private_data_len);
1704 }
1705
1706 int ib_send_cm_dreq(struct ib_cm_id *cm_id,
1707                     const void *private_data,
1708                     u8 private_data_len)
1709 {
1710         struct cm_id_private *cm_id_priv;
1711         struct ib_mad_send_buf *msg;
1712         unsigned long flags;
1713         int ret;
1714
1715         if (private_data && private_data_len > IB_CM_DREQ_PRIVATE_DATA_SIZE)
1716                 return -EINVAL;
1717
1718         cm_id_priv = container_of(cm_id, struct cm_id_private, id);
1719         spin_lock_irqsave(&cm_id_priv->lock, flags);
1720         if (cm_id->state != IB_CM_ESTABLISHED) {
1721                 ret = -EINVAL;
1722                 goto out;
1723         }
1724
1725         ret = cm_alloc_msg(cm_id_priv, &msg);
1726         if (ret) {
1727                 cm_enter_timewait(cm_id_priv);
1728                 goto out;
1729         }
1730
1731         cm_format_dreq((struct cm_dreq_msg *) msg->mad, cm_id_priv,
1732                        private_data, private_data_len);
1733         msg->timeout_ms = cm_id_priv->timeout_ms;
1734         msg->context[1] = (void *) (unsigned long) IB_CM_DREQ_SENT;
1735
1736         ret = ib_post_send_mad(msg, NULL);
1737         if (ret) {
1738                 cm_enter_timewait(cm_id_priv);
1739                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1740                 cm_free_msg(msg);
1741                 return ret;
1742         }
1743
1744         cm_id->state = IB_CM_DREQ_SENT;
1745         cm_id_priv->msg = msg;
1746 out:    spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1747         return ret;
1748 }
1749 EXPORT_SYMBOL(ib_send_cm_dreq);
1750
1751 static void cm_format_drep(struct cm_drep_msg *drep_msg,
1752                           struct cm_id_private *cm_id_priv,
1753                           const void *private_data,
1754                           u8 private_data_len)
1755 {
1756         cm_format_mad_hdr(&drep_msg->hdr, CM_DREP_ATTR_ID, cm_id_priv->tid);
1757         drep_msg->local_comm_id = cm_id_priv->id.local_id;
1758         drep_msg->remote_comm_id = cm_id_priv->id.remote_id;
1759
1760         if (private_data && private_data_len)
1761                 memcpy(drep_msg->private_data, private_data, private_data_len);
1762 }
1763
1764 int ib_send_cm_drep(struct ib_cm_id *cm_id,
1765                     const void *private_data,
1766                     u8 private_data_len)
1767 {
1768         struct cm_id_private *cm_id_priv;
1769         struct ib_mad_send_buf *msg;
1770         unsigned long flags;
1771         void *data;
1772         int ret;
1773
1774         if (private_data && private_data_len > IB_CM_DREP_PRIVATE_DATA_SIZE)
1775                 return -EINVAL;
1776
1777         data = cm_copy_private_data(private_data, private_data_len);
1778         if (IS_ERR(data))
1779                 return PTR_ERR(data);
1780
1781         cm_id_priv = container_of(cm_id, struct cm_id_private, id);
1782         spin_lock_irqsave(&cm_id_priv->lock, flags);
1783         if (cm_id->state != IB_CM_DREQ_RCVD) {
1784                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1785                 kfree(data);
1786                 return -EINVAL;
1787         }
1788
1789         cm_set_private_data(cm_id_priv, data, private_data_len);
1790         cm_enter_timewait(cm_id_priv);
1791
1792         ret = cm_alloc_msg(cm_id_priv, &msg);
1793         if (ret)
1794                 goto out;
1795
1796         cm_format_drep((struct cm_drep_msg *) msg->mad, cm_id_priv,
1797                        private_data, private_data_len);
1798
1799         ret = ib_post_send_mad(msg, NULL);
1800         if (ret) {
1801                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1802                 cm_free_msg(msg);
1803                 return ret;
1804         }
1805
1806 out:    spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1807         return ret;
1808 }
1809 EXPORT_SYMBOL(ib_send_cm_drep);
1810
1811 static int cm_dreq_handler(struct cm_work *work)
1812 {
1813         struct cm_id_private *cm_id_priv;
1814         struct cm_dreq_msg *dreq_msg;
1815         struct ib_mad_send_buf *msg = NULL;
1816         unsigned long flags;
1817         int ret;
1818
1819         dreq_msg = (struct cm_dreq_msg *)work->mad_recv_wc->recv_buf.mad;
1820         cm_id_priv = cm_acquire_id(dreq_msg->remote_comm_id,
1821                                    dreq_msg->local_comm_id);
1822         if (!cm_id_priv)
1823                 return -EINVAL;
1824
1825         work->cm_event.private_data = &dreq_msg->private_data;
1826
1827         spin_lock_irqsave(&cm_id_priv->lock, flags);
1828         if (cm_id_priv->local_qpn != cm_dreq_get_remote_qpn(dreq_msg))
1829                 goto unlock;
1830
1831         switch (cm_id_priv->id.state) {
1832         case IB_CM_REP_SENT:
1833         case IB_CM_DREQ_SENT:
1834                 ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg);
1835                 break;
1836         case IB_CM_ESTABLISHED:
1837         case IB_CM_MRA_REP_RCVD:
1838                 break;
1839         case IB_CM_TIMEWAIT:
1840                 if (cm_alloc_response_msg(work->port, work->mad_recv_wc, &msg))
1841                         goto unlock;
1842
1843                 cm_format_drep((struct cm_drep_msg *) msg->mad, cm_id_priv,
1844                                cm_id_priv->private_data,
1845                                cm_id_priv->private_data_len);
1846                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1847
1848                 if (ib_post_send_mad(msg, NULL))
1849                         cm_free_msg(msg);
1850                 goto deref;
1851         default:
1852                 goto unlock;
1853         }
1854         cm_id_priv->id.state = IB_CM_DREQ_RCVD;
1855         cm_id_priv->tid = dreq_msg->hdr.tid;
1856         ret = atomic_inc_and_test(&cm_id_priv->work_count);
1857         if (!ret)
1858                 list_add_tail(&work->list, &cm_id_priv->work_list);
1859         spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1860
1861         if (ret)
1862                 cm_process_work(cm_id_priv, work);
1863         else
1864                 cm_deref_id(cm_id_priv);
1865         return 0;
1866
1867 unlock: spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1868 deref:  cm_deref_id(cm_id_priv);
1869         return -EINVAL;
1870 }
1871
1872 static int cm_drep_handler(struct cm_work *work)
1873 {
1874         struct cm_id_private *cm_id_priv;
1875         struct cm_drep_msg *drep_msg;
1876         unsigned long flags;
1877         int ret;
1878
1879         drep_msg = (struct cm_drep_msg *)work->mad_recv_wc->recv_buf.mad;
1880         cm_id_priv = cm_acquire_id(drep_msg->remote_comm_id,
1881                                    drep_msg->local_comm_id);
1882         if (!cm_id_priv)
1883                 return -EINVAL;
1884
1885         work->cm_event.private_data = &drep_msg->private_data;
1886
1887         spin_lock_irqsave(&cm_id_priv->lock, flags);
1888         if (cm_id_priv->id.state != IB_CM_DREQ_SENT &&
1889             cm_id_priv->id.state != IB_CM_DREQ_RCVD) {
1890                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1891                 goto out;
1892         }
1893         cm_enter_timewait(cm_id_priv);
1894
1895         ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg);
1896         ret = atomic_inc_and_test(&cm_id_priv->work_count);
1897         if (!ret)
1898                 list_add_tail(&work->list, &cm_id_priv->work_list);
1899         spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1900
1901         if (ret)
1902                 cm_process_work(cm_id_priv, work);
1903         else
1904                 cm_deref_id(cm_id_priv);
1905         return 0;
1906 out:
1907         cm_deref_id(cm_id_priv);
1908         return -EINVAL;
1909 }
1910
1911 int ib_send_cm_rej(struct ib_cm_id *cm_id,
1912                    enum ib_cm_rej_reason reason,
1913                    void *ari,
1914                    u8 ari_length,
1915                    const void *private_data,
1916                    u8 private_data_len)
1917 {
1918         struct cm_id_private *cm_id_priv;
1919         struct ib_mad_send_buf *msg;
1920         unsigned long flags;
1921         int ret;
1922
1923         if ((private_data && private_data_len > IB_CM_REJ_PRIVATE_DATA_SIZE) ||
1924             (ari && ari_length > IB_CM_REJ_ARI_LENGTH))
1925                 return -EINVAL;
1926
1927         cm_id_priv = container_of(cm_id, struct cm_id_private, id);
1928
1929         spin_lock_irqsave(&cm_id_priv->lock, flags);
1930         switch (cm_id->state) {
1931         case IB_CM_REQ_SENT:
1932         case IB_CM_MRA_REQ_RCVD:
1933         case IB_CM_REQ_RCVD:
1934         case IB_CM_MRA_REQ_SENT:
1935         case IB_CM_REP_RCVD:
1936         case IB_CM_MRA_REP_SENT:
1937                 ret = cm_alloc_msg(cm_id_priv, &msg);
1938                 if (!ret)
1939                         cm_format_rej((struct cm_rej_msg *) msg->mad,
1940                                       cm_id_priv, reason, ari, ari_length,
1941                                       private_data, private_data_len);
1942
1943                 cm_reset_to_idle(cm_id_priv);
1944                 break;
1945         case IB_CM_REP_SENT:
1946         case IB_CM_MRA_REP_RCVD:
1947                 ret = cm_alloc_msg(cm_id_priv, &msg);
1948                 if (!ret)
1949                         cm_format_rej((struct cm_rej_msg *) msg->mad,
1950                                       cm_id_priv, reason, ari, ari_length,
1951                                       private_data, private_data_len);
1952
1953                 cm_enter_timewait(cm_id_priv);
1954                 break;
1955         default:
1956                 ret = -EINVAL;
1957                 goto out;
1958         }
1959
1960         if (ret)
1961                 goto out;
1962
1963         ret = ib_post_send_mad(msg, NULL);
1964         if (ret)
1965                 cm_free_msg(msg);
1966
1967 out:    spin_unlock_irqrestore(&cm_id_priv->lock, flags);
1968         return ret;
1969 }
1970 EXPORT_SYMBOL(ib_send_cm_rej);
1971
1972 static void cm_format_rej_event(struct cm_work *work)
1973 {
1974         struct cm_rej_msg *rej_msg;
1975         struct ib_cm_rej_event_param *param;
1976
1977         rej_msg = (struct cm_rej_msg *)work->mad_recv_wc->recv_buf.mad;
1978         param = &work->cm_event.param.rej_rcvd;
1979         param->ari = rej_msg->ari;
1980         param->ari_length = cm_rej_get_reject_info_len(rej_msg);
1981         param->reason = __be16_to_cpu(rej_msg->reason);
1982         work->cm_event.private_data = &rej_msg->private_data;
1983 }
1984
1985 static struct cm_id_private * cm_acquire_rejected_id(struct cm_rej_msg *rej_msg)
1986 {
1987         struct cm_timewait_info *timewait_info;
1988         struct cm_id_private *cm_id_priv;
1989         unsigned long flags;
1990         __be32 remote_id;
1991
1992         remote_id = rej_msg->local_comm_id;
1993
1994         if (__be16_to_cpu(rej_msg->reason) == IB_CM_REJ_TIMEOUT) {
1995                 spin_lock_irqsave(&cm.lock, flags);
1996                 timewait_info = cm_find_remote_id( *((__be64 *) rej_msg->ari),
1997                                                   remote_id);
1998                 if (!timewait_info) {
1999                         spin_unlock_irqrestore(&cm.lock, flags);
2000                         return NULL;
2001                 }
2002                 cm_id_priv = idr_find(&cm.local_id_table,
2003                                       (__force int) timewait_info->work.local_id);
2004                 if (cm_id_priv) {
2005                         if (cm_id_priv->id.remote_id == remote_id)
2006                                 atomic_inc(&cm_id_priv->refcount);
2007                         else
2008                                 cm_id_priv = NULL;
2009                 }
2010                 spin_unlock_irqrestore(&cm.lock, flags);
2011         } else if (cm_rej_get_msg_rejected(rej_msg) == CM_MSG_RESPONSE_REQ)
2012                 cm_id_priv = cm_acquire_id(rej_msg->remote_comm_id, 0);
2013         else
2014                 cm_id_priv = cm_acquire_id(rej_msg->remote_comm_id, remote_id);
2015
2016         return cm_id_priv;
2017 }
2018
2019 static int cm_rej_handler(struct cm_work *work)
2020 {
2021         struct cm_id_private *cm_id_priv;
2022         struct cm_rej_msg *rej_msg;
2023         unsigned long flags;
2024         int ret;
2025
2026         rej_msg = (struct cm_rej_msg *)work->mad_recv_wc->recv_buf.mad;
2027         cm_id_priv = cm_acquire_rejected_id(rej_msg);
2028         if (!cm_id_priv)
2029                 return -EINVAL;
2030
2031         cm_format_rej_event(work);
2032
2033         spin_lock_irqsave(&cm_id_priv->lock, flags);
2034         switch (cm_id_priv->id.state) {
2035         case IB_CM_REQ_SENT:
2036         case IB_CM_MRA_REQ_RCVD:
2037         case IB_CM_REP_SENT:
2038         case IB_CM_MRA_REP_RCVD:
2039                 ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg);
2040                 /* fall through */
2041         case IB_CM_REQ_RCVD:
2042         case IB_CM_MRA_REQ_SENT:
2043                 if (__be16_to_cpu(rej_msg->reason) == IB_CM_REJ_STALE_CONN)
2044                         cm_enter_timewait(cm_id_priv);
2045                 else
2046                         cm_reset_to_idle(cm_id_priv);
2047                 break;
2048         case IB_CM_DREQ_SENT:
2049                 ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg);
2050                 /* fall through */
2051         case IB_CM_REP_RCVD:
2052         case IB_CM_MRA_REP_SENT:
2053         case IB_CM_ESTABLISHED:
2054                 cm_enter_timewait(cm_id_priv);
2055                 break;
2056         default:
2057                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2058                 ret = -EINVAL;
2059                 goto out;
2060         }
2061
2062         ret = atomic_inc_and_test(&cm_id_priv->work_count);
2063         if (!ret)
2064                 list_add_tail(&work->list, &cm_id_priv->work_list);
2065         spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2066
2067         if (ret)
2068                 cm_process_work(cm_id_priv, work);
2069         else
2070                 cm_deref_id(cm_id_priv);
2071         return 0;
2072 out:
2073         cm_deref_id(cm_id_priv);
2074         return -EINVAL;
2075 }
2076
2077 int ib_send_cm_mra(struct ib_cm_id *cm_id,
2078                    u8 service_timeout,
2079                    const void *private_data,
2080                    u8 private_data_len)
2081 {
2082         struct cm_id_private *cm_id_priv;
2083         struct ib_mad_send_buf *msg;
2084         void *data;
2085         unsigned long flags;
2086         int ret;
2087
2088         if (private_data && private_data_len > IB_CM_MRA_PRIVATE_DATA_SIZE)
2089                 return -EINVAL;
2090
2091         data = cm_copy_private_data(private_data, private_data_len);
2092         if (IS_ERR(data))
2093                 return PTR_ERR(data);
2094
2095         cm_id_priv = container_of(cm_id, struct cm_id_private, id);
2096
2097         spin_lock_irqsave(&cm_id_priv->lock, flags);
2098         switch(cm_id_priv->id.state) {
2099         case IB_CM_REQ_RCVD:
2100                 ret = cm_alloc_msg(cm_id_priv, &msg);
2101                 if (ret)
2102                         goto error1;
2103
2104                 cm_format_mra((struct cm_mra_msg *) msg->mad, cm_id_priv,
2105                               CM_MSG_RESPONSE_REQ, service_timeout,
2106                               private_data, private_data_len);
2107                 ret = ib_post_send_mad(msg, NULL);
2108                 if (ret)
2109                         goto error2;
2110                 cm_id->state = IB_CM_MRA_REQ_SENT;
2111                 break;
2112         case IB_CM_REP_RCVD:
2113                 ret = cm_alloc_msg(cm_id_priv, &msg);
2114                 if (ret)
2115                         goto error1;
2116
2117                 cm_format_mra((struct cm_mra_msg *) msg->mad, cm_id_priv,
2118                               CM_MSG_RESPONSE_REP, service_timeout,
2119                               private_data, private_data_len);
2120                 ret = ib_post_send_mad(msg, NULL);
2121                 if (ret)
2122                         goto error2;
2123                 cm_id->state = IB_CM_MRA_REP_SENT;
2124                 break;
2125         case IB_CM_ESTABLISHED:
2126                 ret = cm_alloc_msg(cm_id_priv, &msg);
2127                 if (ret)
2128                         goto error1;
2129
2130                 cm_format_mra((struct cm_mra_msg *) msg->mad, cm_id_priv,
2131                               CM_MSG_RESPONSE_OTHER, service_timeout,
2132                               private_data, private_data_len);
2133                 ret = ib_post_send_mad(msg, NULL);
2134                 if (ret)
2135                         goto error2;
2136                 cm_id->lap_state = IB_CM_MRA_LAP_SENT;
2137                 break;
2138         default:
2139                 ret = -EINVAL;
2140                 goto error1;
2141         }
2142         cm_id_priv->service_timeout = service_timeout;
2143         cm_set_private_data(cm_id_priv, data, private_data_len);
2144         spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2145         return 0;
2146
2147 error1: spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2148         kfree(data);
2149         return ret;
2150
2151 error2: spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2152         kfree(data);
2153         cm_free_msg(msg);
2154         return ret;
2155 }
2156 EXPORT_SYMBOL(ib_send_cm_mra);
2157
2158 static struct cm_id_private * cm_acquire_mraed_id(struct cm_mra_msg *mra_msg)
2159 {
2160         switch (cm_mra_get_msg_mraed(mra_msg)) {
2161         case CM_MSG_RESPONSE_REQ:
2162                 return cm_acquire_id(mra_msg->remote_comm_id, 0);
2163         case CM_MSG_RESPONSE_REP:
2164         case CM_MSG_RESPONSE_OTHER:
2165                 return cm_acquire_id(mra_msg->remote_comm_id,
2166                                      mra_msg->local_comm_id);
2167         default:
2168                 return NULL;
2169         }
2170 }
2171
2172 static int cm_mra_handler(struct cm_work *work)
2173 {
2174         struct cm_id_private *cm_id_priv;
2175         struct cm_mra_msg *mra_msg;
2176         unsigned long flags;
2177         int timeout, ret;
2178
2179         mra_msg = (struct cm_mra_msg *)work->mad_recv_wc->recv_buf.mad;
2180         cm_id_priv = cm_acquire_mraed_id(mra_msg);
2181         if (!cm_id_priv)
2182                 return -EINVAL;
2183
2184         work->cm_event.private_data = &mra_msg->private_data;
2185         work->cm_event.param.mra_rcvd.service_timeout =
2186                                         cm_mra_get_service_timeout(mra_msg);
2187         timeout = cm_convert_to_ms(cm_mra_get_service_timeout(mra_msg)) +
2188                   cm_convert_to_ms(cm_id_priv->av.packet_life_time);
2189
2190         spin_lock_irqsave(&cm_id_priv->lock, flags);
2191         switch (cm_id_priv->id.state) {
2192         case IB_CM_REQ_SENT:
2193                 if (cm_mra_get_msg_mraed(mra_msg) != CM_MSG_RESPONSE_REQ ||
2194                     ib_modify_mad(cm_id_priv->av.port->mad_agent,
2195                                   cm_id_priv->msg, timeout))
2196                         goto out;
2197                 cm_id_priv->id.state = IB_CM_MRA_REQ_RCVD;
2198                 break;
2199         case IB_CM_REP_SENT:
2200                 if (cm_mra_get_msg_mraed(mra_msg) != CM_MSG_RESPONSE_REP ||
2201                     ib_modify_mad(cm_id_priv->av.port->mad_agent,
2202                                   cm_id_priv->msg, timeout))
2203                         goto out;
2204                 cm_id_priv->id.state = IB_CM_MRA_REP_RCVD;
2205                 break;
2206         case IB_CM_ESTABLISHED:
2207                 if (cm_mra_get_msg_mraed(mra_msg) != CM_MSG_RESPONSE_OTHER ||
2208                     cm_id_priv->id.lap_state != IB_CM_LAP_SENT ||
2209                     ib_modify_mad(cm_id_priv->av.port->mad_agent,
2210                                   cm_id_priv->msg, timeout))
2211                         goto out;
2212                 cm_id_priv->id.lap_state = IB_CM_MRA_LAP_RCVD;
2213                 break;
2214         default:
2215                 goto out;
2216         }
2217
2218         cm_id_priv->msg->context[1] = (void *) (unsigned long)
2219                                       cm_id_priv->id.state;
2220         ret = atomic_inc_and_test(&cm_id_priv->work_count);
2221         if (!ret)
2222                 list_add_tail(&work->list, &cm_id_priv->work_list);
2223         spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2224
2225         if (ret)
2226                 cm_process_work(cm_id_priv, work);
2227         else
2228                 cm_deref_id(cm_id_priv);
2229         return 0;
2230 out:
2231         spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2232         cm_deref_id(cm_id_priv);
2233         return -EINVAL;
2234 }
2235
2236 static void cm_format_lap(struct cm_lap_msg *lap_msg,
2237                           struct cm_id_private *cm_id_priv,
2238                           struct ib_sa_path_rec *alternate_path,
2239                           const void *private_data,
2240                           u8 private_data_len)
2241 {
2242         cm_format_mad_hdr(&lap_msg->hdr, CM_LAP_ATTR_ID,
2243                           cm_form_tid(cm_id_priv, CM_MSG_SEQUENCE_LAP));
2244         lap_msg->local_comm_id = cm_id_priv->id.local_id;
2245         lap_msg->remote_comm_id = cm_id_priv->id.remote_id;
2246         cm_lap_set_remote_qpn(lap_msg, cm_id_priv->remote_qpn);
2247         /* todo: need remote CM response timeout */
2248         cm_lap_set_remote_resp_timeout(lap_msg, 0x1F);
2249         lap_msg->alt_local_lid = alternate_path->slid;
2250         lap_msg->alt_remote_lid = alternate_path->dlid;
2251         lap_msg->alt_local_gid = alternate_path->sgid;
2252         lap_msg->alt_remote_gid = alternate_path->dgid;
2253         cm_lap_set_flow_label(lap_msg, alternate_path->flow_label);
2254         cm_lap_set_traffic_class(lap_msg, alternate_path->traffic_class);
2255         lap_msg->alt_hop_limit = alternate_path->hop_limit;
2256         cm_lap_set_packet_rate(lap_msg, alternate_path->rate);
2257         cm_lap_set_sl(lap_msg, alternate_path->sl);
2258         cm_lap_set_subnet_local(lap_msg, 1); /* local only... */
2259         cm_lap_set_local_ack_timeout(lap_msg,
2260                 min(31, alternate_path->packet_life_time + 1));
2261
2262         if (private_data && private_data_len)
2263                 memcpy(lap_msg->private_data, private_data, private_data_len);
2264 }
2265
2266 int ib_send_cm_lap(struct ib_cm_id *cm_id,
2267                    struct ib_sa_path_rec *alternate_path,
2268                    const void *private_data,
2269                    u8 private_data_len)
2270 {
2271         struct cm_id_private *cm_id_priv;
2272         struct ib_mad_send_buf *msg;
2273         unsigned long flags;
2274         int ret;
2275
2276         if (private_data && private_data_len > IB_CM_LAP_PRIVATE_DATA_SIZE)
2277                 return -EINVAL;
2278
2279         cm_id_priv = container_of(cm_id, struct cm_id_private, id);
2280         spin_lock_irqsave(&cm_id_priv->lock, flags);
2281         if (cm_id->state != IB_CM_ESTABLISHED ||
2282             cm_id->lap_state != IB_CM_LAP_IDLE) {
2283                 ret = -EINVAL;
2284                 goto out;
2285         }
2286
2287         ret = cm_alloc_msg(cm_id_priv, &msg);
2288         if (ret)
2289                 goto out;
2290
2291         cm_format_lap((struct cm_lap_msg *) msg->mad, cm_id_priv,
2292                       alternate_path, private_data, private_data_len);
2293         msg->timeout_ms = cm_id_priv->timeout_ms;
2294         msg->context[1] = (void *) (unsigned long) IB_CM_ESTABLISHED;
2295
2296         ret = ib_post_send_mad(msg, NULL);
2297         if (ret) {
2298                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2299                 cm_free_msg(msg);
2300                 return ret;
2301         }
2302
2303         cm_id->lap_state = IB_CM_LAP_SENT;
2304         cm_id_priv->msg = msg;
2305
2306 out:    spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2307         return ret;
2308 }
2309 EXPORT_SYMBOL(ib_send_cm_lap);
2310
2311 static void cm_format_path_from_lap(struct ib_sa_path_rec *path,
2312                                     struct cm_lap_msg *lap_msg)
2313 {
2314         memset(path, 0, sizeof *path);
2315         path->dgid = lap_msg->alt_local_gid;
2316         path->sgid = lap_msg->alt_remote_gid;
2317         path->dlid = lap_msg->alt_local_lid;
2318         path->slid = lap_msg->alt_remote_lid;
2319         path->flow_label = cm_lap_get_flow_label(lap_msg);
2320         path->hop_limit = lap_msg->alt_hop_limit;
2321         path->traffic_class = cm_lap_get_traffic_class(lap_msg);
2322         path->reversible = 1;
2323         /* pkey is same as in REQ */
2324         path->sl = cm_lap_get_sl(lap_msg);
2325         path->mtu_selector = IB_SA_EQ;
2326         /* mtu is same as in REQ */
2327         path->rate_selector = IB_SA_EQ;
2328         path->rate = cm_lap_get_packet_rate(lap_msg);
2329         path->packet_life_time_selector = IB_SA_EQ;
2330         path->packet_life_time = cm_lap_get_local_ack_timeout(lap_msg);
2331         path->packet_life_time -= (path->packet_life_time > 0);
2332 }
2333
2334 static int cm_lap_handler(struct cm_work *work)
2335 {
2336         struct cm_id_private *cm_id_priv;
2337         struct cm_lap_msg *lap_msg;
2338         struct ib_cm_lap_event_param *param;
2339         struct ib_mad_send_buf *msg = NULL;
2340         unsigned long flags;
2341         int ret;
2342
2343         /* todo: verify LAP request and send reject APR if invalid. */
2344         lap_msg = (struct cm_lap_msg *)work->mad_recv_wc->recv_buf.mad;
2345         cm_id_priv = cm_acquire_id(lap_msg->remote_comm_id,
2346                                    lap_msg->local_comm_id);
2347         if (!cm_id_priv)
2348                 return -EINVAL;
2349
2350         param = &work->cm_event.param.lap_rcvd;
2351         param->alternate_path = &work->path[0];
2352         cm_format_path_from_lap(param->alternate_path, lap_msg);
2353         work->cm_event.private_data = &lap_msg->private_data;
2354
2355         spin_lock_irqsave(&cm_id_priv->lock, flags);
2356         if (cm_id_priv->id.state != IB_CM_ESTABLISHED)
2357                 goto unlock;
2358
2359         switch (cm_id_priv->id.lap_state) {
2360         case IB_CM_LAP_IDLE:
2361                 break;
2362         case IB_CM_MRA_LAP_SENT:
2363                 if (cm_alloc_response_msg(work->port, work->mad_recv_wc, &msg))
2364                         goto unlock;
2365
2366                 cm_format_mra((struct cm_mra_msg *) msg->mad, cm_id_priv,
2367                               CM_MSG_RESPONSE_OTHER,
2368                               cm_id_priv->service_timeout,
2369                               cm_id_priv->private_data,
2370                               cm_id_priv->private_data_len);
2371                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2372
2373                 if (ib_post_send_mad(msg, NULL))
2374                         cm_free_msg(msg);
2375                 goto deref;
2376         default:
2377                 goto unlock;
2378         }
2379
2380         cm_id_priv->id.lap_state = IB_CM_LAP_RCVD;
2381         cm_id_priv->tid = lap_msg->hdr.tid;
2382         ret = atomic_inc_and_test(&cm_id_priv->work_count);
2383         if (!ret)
2384                 list_add_tail(&work->list, &cm_id_priv->work_list);
2385         spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2386
2387         if (ret)
2388                 cm_process_work(cm_id_priv, work);
2389         else
2390                 cm_deref_id(cm_id_priv);
2391         return 0;
2392
2393 unlock: spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2394 deref:  cm_deref_id(cm_id_priv);
2395         return -EINVAL;
2396 }
2397
2398 static void cm_format_apr(struct cm_apr_msg *apr_msg,
2399                           struct cm_id_private *cm_id_priv,
2400                           enum ib_cm_apr_status status,
2401                           void *info,
2402                           u8 info_length,
2403                           const void *private_data,
2404                           u8 private_data_len)
2405 {
2406         cm_format_mad_hdr(&apr_msg->hdr, CM_APR_ATTR_ID, cm_id_priv->tid);
2407         apr_msg->local_comm_id = cm_id_priv->id.local_id;
2408         apr_msg->remote_comm_id = cm_id_priv->id.remote_id;
2409         apr_msg->ap_status = (u8) status;
2410
2411         if (info && info_length) {
2412                 apr_msg->info_length = info_length;
2413                 memcpy(apr_msg->info, info, info_length);
2414         }
2415
2416         if (private_data && private_data_len)
2417                 memcpy(apr_msg->private_data, private_data, private_data_len);
2418 }
2419
2420 int ib_send_cm_apr(struct ib_cm_id *cm_id,
2421                    enum ib_cm_apr_status status,
2422                    void *info,
2423                    u8 info_length,
2424                    const void *private_data,
2425                    u8 private_data_len)
2426 {
2427         struct cm_id_private *cm_id_priv;
2428         struct ib_mad_send_buf *msg;
2429         unsigned long flags;
2430         int ret;
2431
2432         if ((private_data && private_data_len > IB_CM_APR_PRIVATE_DATA_SIZE) ||
2433             (info && info_length > IB_CM_APR_INFO_LENGTH))
2434                 return -EINVAL;
2435
2436         cm_id_priv = container_of(cm_id, struct cm_id_private, id);
2437         spin_lock_irqsave(&cm_id_priv->lock, flags);
2438         if (cm_id->state != IB_CM_ESTABLISHED ||
2439             (cm_id->lap_state != IB_CM_LAP_RCVD &&
2440              cm_id->lap_state != IB_CM_MRA_LAP_SENT)) {
2441                 ret = -EINVAL;
2442                 goto out;
2443         }
2444
2445         ret = cm_alloc_msg(cm_id_priv, &msg);
2446         if (ret)
2447                 goto out;
2448
2449         cm_format_apr((struct cm_apr_msg *) msg->mad, cm_id_priv, status,
2450                       info, info_length, private_data, private_data_len);
2451         ret = ib_post_send_mad(msg, NULL);
2452         if (ret) {
2453                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2454                 cm_free_msg(msg);
2455                 return ret;
2456         }
2457
2458         cm_id->lap_state = IB_CM_LAP_IDLE;
2459 out:    spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2460         return ret;
2461 }
2462 EXPORT_SYMBOL(ib_send_cm_apr);
2463
2464 static int cm_apr_handler(struct cm_work *work)
2465 {
2466         struct cm_id_private *cm_id_priv;
2467         struct cm_apr_msg *apr_msg;
2468         unsigned long flags;
2469         int ret;
2470
2471         apr_msg = (struct cm_apr_msg *)work->mad_recv_wc->recv_buf.mad;
2472         cm_id_priv = cm_acquire_id(apr_msg->remote_comm_id,
2473                                    apr_msg->local_comm_id);
2474         if (!cm_id_priv)
2475                 return -EINVAL; /* Unmatched reply. */
2476
2477         work->cm_event.param.apr_rcvd.ap_status = apr_msg->ap_status;
2478         work->cm_event.param.apr_rcvd.apr_info = &apr_msg->info;
2479         work->cm_event.param.apr_rcvd.info_len = apr_msg->info_length;
2480         work->cm_event.private_data = &apr_msg->private_data;
2481
2482         spin_lock_irqsave(&cm_id_priv->lock, flags);
2483         if (cm_id_priv->id.state != IB_CM_ESTABLISHED ||
2484             (cm_id_priv->id.lap_state != IB_CM_LAP_SENT &&
2485              cm_id_priv->id.lap_state != IB_CM_MRA_LAP_RCVD)) {
2486                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2487                 goto out;
2488         }
2489         cm_id_priv->id.lap_state = IB_CM_LAP_IDLE;
2490         ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg);
2491         cm_id_priv->msg = NULL;
2492
2493         ret = atomic_inc_and_test(&cm_id_priv->work_count);
2494         if (!ret)
2495                 list_add_tail(&work->list, &cm_id_priv->work_list);
2496         spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2497
2498         if (ret)
2499                 cm_process_work(cm_id_priv, work);
2500         else
2501                 cm_deref_id(cm_id_priv);
2502         return 0;
2503 out:
2504         cm_deref_id(cm_id_priv);
2505         return -EINVAL;
2506 }
2507
2508 static int cm_timewait_handler(struct cm_work *work)
2509 {
2510         struct cm_timewait_info *timewait_info;
2511         struct cm_id_private *cm_id_priv;
2512         unsigned long flags;
2513         int ret;
2514
2515         timewait_info = (struct cm_timewait_info *)work;
2516         cm_cleanup_timewait(timewait_info);
2517
2518         cm_id_priv = cm_acquire_id(timewait_info->work.local_id,
2519                                    timewait_info->work.remote_id);
2520         if (!cm_id_priv)
2521                 return -EINVAL;
2522
2523         spin_lock_irqsave(&cm_id_priv->lock, flags);
2524         if (cm_id_priv->id.state != IB_CM_TIMEWAIT ||
2525             cm_id_priv->remote_qpn != timewait_info->remote_qpn) {
2526                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2527                 goto out;
2528         }
2529         cm_id_priv->id.state = IB_CM_IDLE;
2530         ret = atomic_inc_and_test(&cm_id_priv->work_count);
2531         if (!ret)
2532                 list_add_tail(&work->list, &cm_id_priv->work_list);
2533         spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2534
2535         if (ret)
2536                 cm_process_work(cm_id_priv, work);
2537         else
2538                 cm_deref_id(cm_id_priv);
2539         return 0;
2540 out:
2541         cm_deref_id(cm_id_priv);
2542         return -EINVAL;
2543 }
2544
2545 static void cm_format_sidr_req(struct cm_sidr_req_msg *sidr_req_msg,
2546                                struct cm_id_private *cm_id_priv,
2547                                struct ib_cm_sidr_req_param *param)
2548 {
2549         cm_format_mad_hdr(&sidr_req_msg->hdr, CM_SIDR_REQ_ATTR_ID,
2550                           cm_form_tid(cm_id_priv, CM_MSG_SEQUENCE_SIDR));
2551         sidr_req_msg->request_id = cm_id_priv->id.local_id;
2552         sidr_req_msg->pkey = cpu_to_be16(param->pkey);
2553         sidr_req_msg->service_id = param->service_id;
2554
2555         if (param->private_data && param->private_data_len)
2556                 memcpy(sidr_req_msg->private_data, param->private_data,
2557                        param->private_data_len);
2558 }
2559
2560 int ib_send_cm_sidr_req(struct ib_cm_id *cm_id,
2561                         struct ib_cm_sidr_req_param *param)
2562 {
2563         struct cm_id_private *cm_id_priv;
2564         struct ib_mad_send_buf *msg;
2565         unsigned long flags;
2566         int ret;
2567
2568         if (!param->path || (param->private_data &&
2569              param->private_data_len > IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE))
2570                 return -EINVAL;
2571
2572         cm_id_priv = container_of(cm_id, struct cm_id_private, id);
2573         ret = cm_init_av_by_path(param->path, &cm_id_priv->av);
2574         if (ret)
2575                 goto out;
2576
2577         cm_id->service_id = param->service_id;
2578         cm_id->service_mask = __constant_cpu_to_be64(~0ULL);
2579         cm_id_priv->timeout_ms = param->timeout_ms;
2580         cm_id_priv->max_cm_retries = param->max_cm_retries;
2581         ret = cm_alloc_msg(cm_id_priv, &msg);
2582         if (ret)
2583                 goto out;
2584
2585         cm_format_sidr_req((struct cm_sidr_req_msg *) msg->mad, cm_id_priv,
2586                            param);
2587         msg->timeout_ms = cm_id_priv->timeout_ms;
2588         msg->context[1] = (void *) (unsigned long) IB_CM_SIDR_REQ_SENT;
2589
2590         spin_lock_irqsave(&cm_id_priv->lock, flags);
2591         if (cm_id->state == IB_CM_IDLE)
2592                 ret = ib_post_send_mad(msg, NULL);
2593         else
2594                 ret = -EINVAL;
2595
2596         if (ret) {
2597                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2598                 cm_free_msg(msg);
2599                 goto out;
2600         }
2601         cm_id->state = IB_CM_SIDR_REQ_SENT;
2602         cm_id_priv->msg = msg;
2603         spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2604 out:
2605         return ret;
2606 }
2607 EXPORT_SYMBOL(ib_send_cm_sidr_req);
2608
2609 static void cm_format_sidr_req_event(struct cm_work *work,
2610                                      struct ib_cm_id *listen_id)
2611 {
2612         struct cm_sidr_req_msg *sidr_req_msg;
2613         struct ib_cm_sidr_req_event_param *param;
2614
2615         sidr_req_msg = (struct cm_sidr_req_msg *)
2616                                 work->mad_recv_wc->recv_buf.mad;
2617         param = &work->cm_event.param.sidr_req_rcvd;
2618         param->pkey = __be16_to_cpu(sidr_req_msg->pkey);
2619         param->listen_id = listen_id;
2620         param->port = work->port->port_num;
2621         work->cm_event.private_data = &sidr_req_msg->private_data;
2622 }
2623
2624 static int cm_sidr_req_handler(struct cm_work *work)
2625 {
2626         struct ib_cm_id *cm_id;
2627         struct cm_id_private *cm_id_priv, *cur_cm_id_priv;
2628         struct cm_sidr_req_msg *sidr_req_msg;
2629         struct ib_wc *wc;
2630         unsigned long flags;
2631
2632         cm_id = ib_create_cm_id(work->port->cm_dev->device, NULL, NULL);
2633         if (IS_ERR(cm_id))
2634                 return PTR_ERR(cm_id);
2635         cm_id_priv = container_of(cm_id, struct cm_id_private, id);
2636
2637         /* Record SGID/SLID and request ID for lookup. */
2638         sidr_req_msg = (struct cm_sidr_req_msg *)
2639                                 work->mad_recv_wc->recv_buf.mad;
2640         wc = work->mad_recv_wc->wc;
2641         cm_id_priv->av.dgid.global.subnet_prefix = cpu_to_be64(wc->slid);
2642         cm_id_priv->av.dgid.global.interface_id = 0;
2643         cm_init_av_for_response(work->port, work->mad_recv_wc->wc,
2644                                 &cm_id_priv->av);
2645         cm_id_priv->id.remote_id = sidr_req_msg->request_id;
2646         cm_id_priv->id.state = IB_CM_SIDR_REQ_RCVD;
2647         cm_id_priv->tid = sidr_req_msg->hdr.tid;
2648         atomic_inc(&cm_id_priv->work_count);
2649
2650         spin_lock_irqsave(&cm.lock, flags);
2651         cur_cm_id_priv = cm_insert_remote_sidr(cm_id_priv);
2652         if (cur_cm_id_priv) {
2653                 spin_unlock_irqrestore(&cm.lock, flags);
2654                 goto out; /* Duplicate message. */
2655         }
2656         cur_cm_id_priv = cm_find_listen(cm_id->device,
2657                                         sidr_req_msg->service_id);
2658         if (!cur_cm_id_priv) {
2659                 rb_erase(&cm_id_priv->sidr_id_node, &cm.remote_sidr_table);
2660                 spin_unlock_irqrestore(&cm.lock, flags);
2661                 /* todo: reply with no match */
2662                 goto out; /* No match. */
2663         }
2664         atomic_inc(&cur_cm_id_priv->refcount);
2665         spin_unlock_irqrestore(&cm.lock, flags);
2666
2667         cm_id_priv->id.cm_handler = cur_cm_id_priv->id.cm_handler;
2668         cm_id_priv->id.context = cur_cm_id_priv->id.context;
2669         cm_id_priv->id.service_id = sidr_req_msg->service_id;
2670         cm_id_priv->id.service_mask = __constant_cpu_to_be64(~0ULL);
2671
2672         cm_format_sidr_req_event(work, &cur_cm_id_priv->id);
2673         cm_process_work(cm_id_priv, work);
2674         cm_deref_id(cur_cm_id_priv);
2675         return 0;
2676 out:
2677         ib_destroy_cm_id(&cm_id_priv->id);
2678         return -EINVAL;
2679 }
2680
2681 static void cm_format_sidr_rep(struct cm_sidr_rep_msg *sidr_rep_msg,
2682                                struct cm_id_private *cm_id_priv,
2683                                struct ib_cm_sidr_rep_param *param)
2684 {
2685         cm_format_mad_hdr(&sidr_rep_msg->hdr, CM_SIDR_REP_ATTR_ID,
2686                           cm_id_priv->tid);
2687         sidr_rep_msg->request_id = cm_id_priv->id.remote_id;
2688         sidr_rep_msg->status = param->status;
2689         cm_sidr_rep_set_qpn(sidr_rep_msg, cpu_to_be32(param->qp_num));
2690         sidr_rep_msg->service_id = cm_id_priv->id.service_id;
2691         sidr_rep_msg->qkey = cpu_to_be32(param->qkey);
2692
2693         if (param->info && param->info_length)
2694                 memcpy(sidr_rep_msg->info, param->info, param->info_length);
2695
2696         if (param->private_data && param->private_data_len)
2697                 memcpy(sidr_rep_msg->private_data, param->private_data,
2698                        param->private_data_len);
2699 }
2700
2701 int ib_send_cm_sidr_rep(struct ib_cm_id *cm_id,
2702                         struct ib_cm_sidr_rep_param *param)
2703 {
2704         struct cm_id_private *cm_id_priv;
2705         struct ib_mad_send_buf *msg;
2706         unsigned long flags;
2707         int ret;
2708
2709         if ((param->info && param->info_length > IB_CM_SIDR_REP_INFO_LENGTH) ||
2710             (param->private_data &&
2711              param->private_data_len > IB_CM_SIDR_REP_PRIVATE_DATA_SIZE))
2712                 return -EINVAL;
2713
2714         cm_id_priv = container_of(cm_id, struct cm_id_private, id);
2715         spin_lock_irqsave(&cm_id_priv->lock, flags);
2716         if (cm_id->state != IB_CM_SIDR_REQ_RCVD) {
2717                 ret = -EINVAL;
2718                 goto error;
2719         }
2720
2721         ret = cm_alloc_msg(cm_id_priv, &msg);
2722         if (ret)
2723                 goto error;
2724
2725         cm_format_sidr_rep((struct cm_sidr_rep_msg *) msg->mad, cm_id_priv,
2726                            param);
2727         ret = ib_post_send_mad(msg, NULL);
2728         if (ret) {
2729                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2730                 cm_free_msg(msg);
2731                 return ret;
2732         }
2733         cm_id->state = IB_CM_IDLE;
2734         spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2735
2736         spin_lock_irqsave(&cm.lock, flags);
2737         rb_erase(&cm_id_priv->sidr_id_node, &cm.remote_sidr_table);
2738         spin_unlock_irqrestore(&cm.lock, flags);
2739         return 0;
2740
2741 error:  spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2742         return ret;
2743 }
2744 EXPORT_SYMBOL(ib_send_cm_sidr_rep);
2745
2746 static void cm_format_sidr_rep_event(struct cm_work *work)
2747 {
2748         struct cm_sidr_rep_msg *sidr_rep_msg;
2749         struct ib_cm_sidr_rep_event_param *param;
2750
2751         sidr_rep_msg = (struct cm_sidr_rep_msg *)
2752                                 work->mad_recv_wc->recv_buf.mad;
2753         param = &work->cm_event.param.sidr_rep_rcvd;
2754         param->status = sidr_rep_msg->status;
2755         param->qkey = be32_to_cpu(sidr_rep_msg->qkey);
2756         param->qpn = be32_to_cpu(cm_sidr_rep_get_qpn(sidr_rep_msg));
2757         param->info = &sidr_rep_msg->info;
2758         param->info_len = sidr_rep_msg->info_length;
2759         work->cm_event.private_data = &sidr_rep_msg->private_data;
2760 }
2761
2762 static int cm_sidr_rep_handler(struct cm_work *work)
2763 {
2764         struct cm_sidr_rep_msg *sidr_rep_msg;
2765         struct cm_id_private *cm_id_priv;
2766         unsigned long flags;
2767
2768         sidr_rep_msg = (struct cm_sidr_rep_msg *)
2769                                 work->mad_recv_wc->recv_buf.mad;
2770         cm_id_priv = cm_acquire_id(sidr_rep_msg->request_id, 0);
2771         if (!cm_id_priv)
2772                 return -EINVAL; /* Unmatched reply. */
2773
2774         spin_lock_irqsave(&cm_id_priv->lock, flags);
2775         if (cm_id_priv->id.state != IB_CM_SIDR_REQ_SENT) {
2776                 spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2777                 goto out;
2778         }
2779         cm_id_priv->id.state = IB_CM_IDLE;
2780         ib_cancel_mad(cm_id_priv->av.port->mad_agent, cm_id_priv->msg);
2781         spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2782
2783         cm_format_sidr_rep_event(work);
2784         cm_process_work(cm_id_priv, work);
2785         return 0;
2786 out:
2787         cm_deref_id(cm_id_priv);
2788         return -EINVAL;
2789 }
2790
2791 static void cm_process_send_error(struct ib_mad_send_buf *msg,
2792                                   enum ib_wc_status wc_status)
2793 {
2794         struct cm_id_private *cm_id_priv;
2795         struct ib_cm_event cm_event;
2796         enum ib_cm_state state;
2797         unsigned long flags;
2798         int ret;
2799
2800         memset(&cm_event, 0, sizeof cm_event);
2801         cm_id_priv = msg->context[0];
2802
2803         /* Discard old sends or ones without a response. */
2804         spin_lock_irqsave(&cm_id_priv->lock, flags);
2805         state = (enum ib_cm_state) (unsigned long) msg->context[1];
2806         if (msg != cm_id_priv->msg || state != cm_id_priv->id.state)
2807                 goto discard;
2808
2809         switch (state) {
2810         case IB_CM_REQ_SENT:
2811         case IB_CM_MRA_REQ_RCVD:
2812                 cm_reset_to_idle(cm_id_priv);
2813                 cm_event.event = IB_CM_REQ_ERROR;
2814                 break;
2815         case IB_CM_REP_SENT:
2816         case IB_CM_MRA_REP_RCVD:
2817                 cm_reset_to_idle(cm_id_priv);
2818                 cm_event.event = IB_CM_REP_ERROR;
2819                 break;
2820         case IB_CM_DREQ_SENT:
2821                 cm_enter_timewait(cm_id_priv);
2822                 cm_event.event = IB_CM_DREQ_ERROR;
2823                 break;
2824         case IB_CM_SIDR_REQ_SENT:
2825                 cm_id_priv->id.state = IB_CM_IDLE;
2826                 cm_event.event = IB_CM_SIDR_REQ_ERROR;
2827                 break;
2828         default:
2829                 goto discard;
2830         }
2831         spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2832         cm_event.param.send_status = wc_status;
2833
2834         /* No other events can occur on the cm_id at this point. */
2835         ret = cm_id_priv->id.cm_handler(&cm_id_priv->id, &cm_event);
2836         cm_free_msg(msg);
2837         if (ret)
2838                 ib_destroy_cm_id(&cm_id_priv->id);
2839         return;
2840 discard:
2841         spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2842         cm_free_msg(msg);
2843 }
2844
2845 static void cm_send_handler(struct ib_mad_agent *mad_agent,
2846                             struct ib_mad_send_wc *mad_send_wc)
2847 {
2848         struct ib_mad_send_buf *msg = mad_send_wc->send_buf;
2849
2850         switch (mad_send_wc->status) {
2851         case IB_WC_SUCCESS:
2852         case IB_WC_WR_FLUSH_ERR:
2853                 cm_free_msg(msg);
2854                 break;
2855         default:
2856                 if (msg->context[0] && msg->context[1])
2857                         cm_process_send_error(msg, mad_send_wc->status);
2858                 else
2859                         cm_free_msg(msg);
2860                 break;
2861         }
2862 }
2863
2864 static void cm_work_handler(void *data)
2865 {
2866         struct cm_work *work = data;
2867         int ret;
2868
2869         switch (work->cm_event.event) {
2870         case IB_CM_REQ_RECEIVED:
2871                 ret = cm_req_handler(work);
2872                 break;
2873         case IB_CM_MRA_RECEIVED:
2874                 ret = cm_mra_handler(work);
2875                 break;
2876         case IB_CM_REJ_RECEIVED:
2877                 ret = cm_rej_handler(work);
2878                 break;
2879         case IB_CM_REP_RECEIVED:
2880                 ret = cm_rep_handler(work);
2881                 break;
2882         case IB_CM_RTU_RECEIVED:
2883                 ret = cm_rtu_handler(work);
2884                 break;
2885         case IB_CM_USER_ESTABLISHED:
2886                 ret = cm_establish_handler(work);
2887                 break;
2888         case IB_CM_DREQ_RECEIVED:
2889                 ret = cm_dreq_handler(work);
2890                 break;
2891         case IB_CM_DREP_RECEIVED:
2892                 ret = cm_drep_handler(work);
2893                 break;
2894         case IB_CM_SIDR_REQ_RECEIVED:
2895                 ret = cm_sidr_req_handler(work);
2896                 break;
2897         case IB_CM_SIDR_REP_RECEIVED:
2898                 ret = cm_sidr_rep_handler(work);
2899                 break;
2900         case IB_CM_LAP_RECEIVED:
2901                 ret = cm_lap_handler(work);
2902                 break;
2903         case IB_CM_APR_RECEIVED:
2904                 ret = cm_apr_handler(work);
2905                 break;
2906         case IB_CM_TIMEWAIT_EXIT:
2907                 ret = cm_timewait_handler(work);
2908                 break;
2909         default:
2910                 ret = -EINVAL;
2911                 break;
2912         }
2913         if (ret)
2914                 cm_free_work(work);
2915 }
2916
2917 int ib_cm_establish(struct ib_cm_id *cm_id)
2918 {
2919         struct cm_id_private *cm_id_priv;
2920         struct cm_work *work;
2921         unsigned long flags;
2922         int ret = 0;
2923
2924         work = kmalloc(sizeof *work, GFP_ATOMIC);
2925         if (!work)
2926                 return -ENOMEM;
2927
2928         cm_id_priv = container_of(cm_id, struct cm_id_private, id);
2929         spin_lock_irqsave(&cm_id_priv->lock, flags);
2930         switch (cm_id->state)
2931         {
2932         case IB_CM_REP_SENT:
2933         case IB_CM_MRA_REP_RCVD:
2934                 cm_id->state = IB_CM_ESTABLISHED;
2935                 break;
2936         case IB_CM_ESTABLISHED:
2937                 ret = -EISCONN;
2938                 break;
2939         default:
2940                 ret = -EINVAL;
2941                 break;
2942         }
2943         spin_unlock_irqrestore(&cm_id_priv->lock, flags);
2944
2945         if (ret) {
2946                 kfree(work);
2947                 goto out;
2948         }
2949
2950         /*
2951          * The CM worker thread may try to destroy the cm_id before it
2952          * can execute this work item.  To prevent potential deadlock,
2953          * we need to find the cm_id once we're in the context of the
2954          * worker thread, rather than holding a reference on it.
2955          */
2956         INIT_WORK(&work->work, cm_work_handler, work);
2957         work->local_id = cm_id->local_id;
2958         work->remote_id = cm_id->remote_id;
2959         work->mad_recv_wc = NULL;
2960         work->cm_event.event = IB_CM_USER_ESTABLISHED;
2961         queue_work(cm.wq, &work->work);
2962 out:
2963         return ret;
2964 }
2965 EXPORT_SYMBOL(ib_cm_establish);
2966
2967 static void cm_recv_handler(struct ib_mad_agent *mad_agent,
2968                             struct ib_mad_recv_wc *mad_recv_wc)
2969 {
2970         struct cm_work *work;
2971         enum ib_cm_event_type event;
2972         int paths = 0;
2973
2974         switch (mad_recv_wc->recv_buf.mad->mad_hdr.attr_id) {
2975         case CM_REQ_ATTR_ID:
2976                 paths = 1 + (((struct cm_req_msg *) mad_recv_wc->recv_buf.mad)->
2977                                                     alt_local_lid != 0);
2978                 event = IB_CM_REQ_RECEIVED;
2979                 break;
2980         case CM_MRA_ATTR_ID:
2981                 event = IB_CM_MRA_RECEIVED;
2982                 break;
2983         case CM_REJ_ATTR_ID:
2984                 event = IB_CM_REJ_RECEIVED;
2985                 break;
2986         case CM_REP_ATTR_ID:
2987                 event = IB_CM_REP_RECEIVED;
2988                 break;
2989         case CM_RTU_ATTR_ID:
2990                 event = IB_CM_RTU_RECEIVED;
2991                 break;
2992         case CM_DREQ_ATTR_ID:
2993                 event = IB_CM_DREQ_RECEIVED;
2994                 break;
2995         case CM_DREP_ATTR_ID:
2996                 event = IB_CM_DREP_RECEIVED;
2997                 break;
2998         case CM_SIDR_REQ_ATTR_ID:
2999                 event = IB_CM_SIDR_REQ_RECEIVED;
3000                 break;
3001         case CM_SIDR_REP_ATTR_ID:
3002                 event = IB_CM_SIDR_REP_RECEIVED;
3003                 break;
3004         case CM_LAP_ATTR_ID:
3005                 paths = 1;
3006                 event = IB_CM_LAP_RECEIVED;
3007                 break;
3008         case CM_APR_ATTR_ID:
3009                 event = IB_CM_APR_RECEIVED;
3010                 break;
3011         default:
3012                 ib_free_recv_mad(mad_recv_wc);
3013                 return;
3014         }
3015
3016         work = kmalloc(sizeof *work + sizeof(struct ib_sa_path_rec) * paths,
3017                        GFP_KERNEL);
3018         if (!work) {
3019                 ib_free_recv_mad(mad_recv_wc);
3020                 return;
3021         }
3022
3023         INIT_WORK(&work->work, cm_work_handler, work);
3024         work->cm_event.event = event;
3025         work->mad_recv_wc = mad_recv_wc;
3026         work->port = (struct cm_port *)mad_agent->context;
3027         queue_work(cm.wq, &work->work);
3028 }
3029
3030 static int cm_init_qp_init_attr(struct cm_id_private *cm_id_priv,
3031                                 struct ib_qp_attr *qp_attr,
3032                                 int *qp_attr_mask)
3033 {
3034         unsigned long flags;
3035         int ret;
3036
3037         spin_lock_irqsave(&cm_id_priv->lock, flags);
3038         switch (cm_id_priv->id.state) {
3039         case IB_CM_REQ_SENT:
3040         case IB_CM_MRA_REQ_RCVD:
3041         case IB_CM_REQ_RCVD:
3042         case IB_CM_MRA_REQ_SENT:
3043         case IB_CM_REP_RCVD:
3044         case IB_CM_MRA_REP_SENT:
3045         case IB_CM_REP_SENT:
3046         case IB_CM_MRA_REP_RCVD:
3047         case IB_CM_ESTABLISHED:
3048                 *qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS |
3049                                 IB_QP_PKEY_INDEX | IB_QP_PORT;
3050                 qp_attr->qp_access_flags = IB_ACCESS_LOCAL_WRITE |
3051                                            IB_ACCESS_REMOTE_WRITE;
3052                 if (cm_id_priv->responder_resources)
3053                         qp_attr->qp_access_flags |= IB_ACCESS_REMOTE_READ;
3054                 qp_attr->pkey_index = cm_id_priv->av.pkey_index;
3055                 qp_attr->port_num = cm_id_priv->av.port->port_num;
3056                 ret = 0;
3057                 break;
3058         default:
3059                 ret = -EINVAL;
3060                 break;
3061         }
3062         spin_unlock_irqrestore(&cm_id_priv->lock, flags);
3063         return ret;
3064 }
3065
3066 static int cm_init_qp_rtr_attr(struct cm_id_private *cm_id_priv,
3067                                struct ib_qp_attr *qp_attr,
3068                                int *qp_attr_mask)
3069 {
3070         unsigned long flags;
3071         int ret;
3072
3073         spin_lock_irqsave(&cm_id_priv->lock, flags);
3074         switch (cm_id_priv->id.state) {
3075         case IB_CM_REQ_RCVD:
3076         case IB_CM_MRA_REQ_SENT:
3077         case IB_CM_REP_RCVD:
3078         case IB_CM_MRA_REP_SENT:
3079         case IB_CM_REP_SENT:
3080         case IB_CM_MRA_REP_RCVD:
3081         case IB_CM_ESTABLISHED:
3082                 *qp_attr_mask = IB_QP_STATE | IB_QP_AV | IB_QP_PATH_MTU |
3083                                 IB_QP_DEST_QPN | IB_QP_RQ_PSN;
3084                 qp_attr->ah_attr = cm_id_priv->av.ah_attr;
3085                 qp_attr->path_mtu = cm_id_priv->path_mtu;
3086                 qp_attr->dest_qp_num = be32_to_cpu(cm_id_priv->remote_qpn);
3087                 qp_attr->rq_psn = be32_to_cpu(cm_id_priv->rq_psn);
3088                 if (cm_id_priv->qp_type == IB_QPT_RC) {
3089                         *qp_attr_mask |= IB_QP_MAX_DEST_RD_ATOMIC |
3090                                          IB_QP_MIN_RNR_TIMER;
3091                         qp_attr->max_dest_rd_atomic =
3092                                         cm_id_priv->responder_resources;
3093                         qp_attr->min_rnr_timer = 0;
3094                 }
3095                 if (cm_id_priv->alt_av.ah_attr.dlid) {
3096                         *qp_attr_mask |= IB_QP_ALT_PATH;
3097                         qp_attr->alt_ah_attr = cm_id_priv->alt_av.ah_attr;
3098                 }
3099                 ret = 0;
3100                 break;
3101         default:
3102                 ret = -EINVAL;
3103                 break;
3104         }
3105         spin_unlock_irqrestore(&cm_id_priv->lock, flags);
3106         return ret;
3107 }
3108
3109 static int cm_init_qp_rts_attr(struct cm_id_private *cm_id_priv,
3110                                struct ib_qp_attr *qp_attr,
3111                                int *qp_attr_mask)
3112 {
3113         unsigned long flags;
3114         int ret;
3115
3116         spin_lock_irqsave(&cm_id_priv->lock, flags);
3117         switch (cm_id_priv->id.state) {
3118         case IB_CM_REP_RCVD:
3119         case IB_CM_MRA_REP_SENT:
3120         case IB_CM_REP_SENT:
3121         case IB_CM_MRA_REP_RCVD:
3122         case IB_CM_ESTABLISHED:
3123                 *qp_attr_mask = IB_QP_STATE | IB_QP_SQ_PSN;
3124                 qp_attr->sq_psn = be32_to_cpu(cm_id_priv->sq_psn);
3125                 if (cm_id_priv->qp_type == IB_QPT_RC) {
3126                         *qp_attr_mask |= IB_QP_TIMEOUT | IB_QP_RETRY_CNT |
3127                                          IB_QP_RNR_RETRY |
3128                                          IB_QP_MAX_QP_RD_ATOMIC;
3129                         qp_attr->timeout = cm_id_priv->local_ack_timeout;
3130                         qp_attr->retry_cnt = cm_id_priv->retry_count;
3131                         qp_attr->rnr_retry = cm_id_priv->rnr_retry_count;
3132                         qp_attr->max_rd_atomic = cm_id_priv->initiator_depth;
3133                 }
3134                 if (cm_id_priv->alt_av.ah_attr.dlid) {
3135                         *qp_attr_mask |= IB_QP_PATH_MIG_STATE;
3136                         qp_attr->path_mig_state = IB_MIG_REARM;
3137                 }
3138                 ret = 0;
3139                 break;
3140         default:
3141                 ret = -EINVAL;
3142                 break;
3143         }
3144         spin_unlock_irqrestore(&cm_id_priv->lock, flags);
3145         return ret;
3146 }
3147
3148 int ib_cm_init_qp_attr(struct ib_cm_id *cm_id,
3149                        struct ib_qp_attr *qp_attr,
3150                        int *qp_attr_mask)
3151 {
3152         struct cm_id_private *cm_id_priv;
3153         int ret;
3154
3155         cm_id_priv = container_of(cm_id, struct cm_id_private, id);
3156         switch (qp_attr->qp_state) {
3157         case IB_QPS_INIT:
3158                 ret = cm_init_qp_init_attr(cm_id_priv, qp_attr, qp_attr_mask);
3159                 break;
3160         case IB_QPS_RTR:
3161                 ret = cm_init_qp_rtr_attr(cm_id_priv, qp_attr, qp_attr_mask);
3162                 break;
3163         case IB_QPS_RTS:
3164                 ret = cm_init_qp_rts_attr(cm_id_priv, qp_attr, qp_attr_mask);
3165                 break;
3166         default:
3167                 ret = -EINVAL;
3168                 break;
3169         }
3170         return ret;
3171 }
3172 EXPORT_SYMBOL(ib_cm_init_qp_attr);
3173
3174 static void cm_add_one(struct ib_device *device)
3175 {
3176         struct cm_device *cm_dev;
3177         struct cm_port *port;
3178         struct ib_mad_reg_req reg_req = {
3179                 .mgmt_class = IB_MGMT_CLASS_CM,
3180                 .mgmt_class_version = IB_CM_CLASS_VERSION
3181         };
3182         struct ib_port_modify port_modify = {
3183                 .set_port_cap_mask = IB_PORT_CM_SUP
3184         };
3185         unsigned long flags;
3186         int ret;
3187         u8 i;
3188
3189         cm_dev = kmalloc(sizeof(*cm_dev) + sizeof(*port) *
3190                          device->phys_port_cnt, GFP_KERNEL);
3191         if (!cm_dev)
3192                 return;
3193
3194         cm_dev->device = device;
3195         cm_dev->ca_guid = device->node_guid;
3196
3197         set_bit(IB_MGMT_METHOD_SEND, reg_req.method_mask);
3198         for (i = 1; i <= device->phys_port_cnt; i++) {
3199                 port = &cm_dev->port[i-1];
3200                 port->cm_dev = cm_dev;
3201                 port->port_num = i;
3202                 port->mad_agent = ib_register_mad_agent(device, i,
3203                                                         IB_QPT_GSI,
3204                                                         &reg_req,
3205                                                         0,
3206                                                         cm_send_handler,
3207                                                         cm_recv_handler,
3208                                                         port);
3209                 if (IS_ERR(port->mad_agent))
3210                         goto error1;
3211
3212                 ret = ib_modify_port(device, i, 0, &port_modify);
3213                 if (ret)
3214                         goto error2;
3215         }
3216         ib_set_client_data(device, &cm_client, cm_dev);
3217
3218         write_lock_irqsave(&cm.device_lock, flags);
3219         list_add_tail(&cm_dev->list, &cm.device_list);
3220         write_unlock_irqrestore(&cm.device_lock, flags);
3221         return;
3222
3223 error2:
3224         ib_unregister_mad_agent(port->mad_agent);
3225 error1:
3226         port_modify.set_port_cap_mask = 0;
3227         port_modify.clr_port_cap_mask = IB_PORT_CM_SUP;
3228         while (--i) {
3229                 port = &cm_dev->port[i-1];
3230                 ib_modify_port(device, port->port_num, 0, &port_modify);
3231                 ib_unregister_mad_agent(port->mad_agent);
3232         }
3233         kfree(cm_dev);
3234 }
3235
3236 static void cm_remove_one(struct ib_device *device)
3237 {
3238         struct cm_device *cm_dev;
3239         struct cm_port *port;
3240         struct ib_port_modify port_modify = {
3241                 .clr_port_cap_mask = IB_PORT_CM_SUP
3242         };
3243         unsigned long flags;
3244         int i;
3245
3246         cm_dev = ib_get_client_data(device, &cm_client);
3247         if (!cm_dev)
3248                 return;
3249
3250         write_lock_irqsave(&cm.device_lock, flags);
3251         list_del(&cm_dev->list);
3252         write_unlock_irqrestore(&cm.device_lock, flags);
3253
3254         for (i = 1; i <= device->phys_port_cnt; i++) {
3255                 port = &cm_dev->port[i-1];
3256                 ib_modify_port(device, port->port_num, 0, &port_modify);
3257                 ib_unregister_mad_agent(port->mad_agent);
3258         }
3259         kfree(cm_dev);
3260 }
3261
3262 static int __init ib_cm_init(void)
3263 {
3264         int ret;
3265
3266         memset(&cm, 0, sizeof cm);
3267         INIT_LIST_HEAD(&cm.device_list);
3268         rwlock_init(&cm.device_lock);
3269         spin_lock_init(&cm.lock);
3270         cm.listen_service_table = RB_ROOT;
3271         cm.listen_service_id = __constant_be64_to_cpu(IB_CM_ASSIGN_SERVICE_ID);
3272         cm.remote_id_table = RB_ROOT;
3273         cm.remote_qp_table = RB_ROOT;
3274         cm.remote_sidr_table = RB_ROOT;
3275         idr_init(&cm.local_id_table);
3276         idr_pre_get(&cm.local_id_table, GFP_KERNEL);
3277
3278         cm.wq = create_workqueue("ib_cm");
3279         if (!cm.wq)
3280                 return -ENOMEM;
3281
3282         ret = ib_register_client(&cm_client);
3283         if (ret)
3284                 goto error;
3285
3286         return 0;
3287 error:
3288         destroy_workqueue(cm.wq);
3289         return ret;
3290 }
3291
3292 static void __exit ib_cm_cleanup(void)
3293 {
3294         flush_workqueue(cm.wq);
3295         destroy_workqueue(cm.wq);
3296         ib_unregister_client(&cm_client);
3297         idr_destroy(&cm.local_id_table);
3298 }
3299
3300 module_init(ib_cm_init);
3301 module_exit(ib_cm_cleanup);
3302