2 * Copyright (c) 2012 Mellanox Technologies. All rights reserved.
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
33 #include <rdma/ib_mad.h>
35 #include <linux/mlx4/cmd.h>
36 #include <linux/rbtree.h>
37 #include <linux/idr.h>
38 #include <rdma/ib_cm.h>
42 #define CM_CLEANUP_CACHE_TIMEOUT (5 * HZ)
51 struct mlx4_ib_dev *dev;
53 struct list_head list;
54 struct delayed_work timeout;
57 struct cm_generic_msg {
58 struct ib_mad_hdr hdr;
61 __be32 remote_comm_id;
65 unsigned char unused[0x60];
66 union ib_gid primary_path_sgid;
70 static void set_local_comm_id(struct ib_mad *mad, u32 cm_id)
72 struct cm_generic_msg *msg = (struct cm_generic_msg *)mad;
73 msg->local_comm_id = cpu_to_be32(cm_id);
76 static u32 get_local_comm_id(struct ib_mad *mad)
78 struct cm_generic_msg *msg = (struct cm_generic_msg *)mad;
80 return be32_to_cpu(msg->local_comm_id);
83 static void set_remote_comm_id(struct ib_mad *mad, u32 cm_id)
85 struct cm_generic_msg *msg = (struct cm_generic_msg *)mad;
86 msg->remote_comm_id = cpu_to_be32(cm_id);
89 static u32 get_remote_comm_id(struct ib_mad *mad)
91 struct cm_generic_msg *msg = (struct cm_generic_msg *)mad;
93 return be32_to_cpu(msg->remote_comm_id);
96 static union ib_gid gid_from_req_msg(struct ib_device *ibdev, struct ib_mad *mad)
98 struct cm_req_msg *msg = (struct cm_req_msg *)mad;
100 return msg->primary_path_sgid;
103 /* Lock should be taken before called */
104 static struct id_map_entry *
105 id_map_find_by_sl_id(struct ib_device *ibdev, u32 slave_id, u32 sl_cm_id)
107 struct rb_root *sl_id_map = &to_mdev(ibdev)->sriov.sl_id_map;
108 struct rb_node *node = sl_id_map->rb_node;
111 struct id_map_entry *id_map_entry =
112 rb_entry(node, struct id_map_entry, node);
114 if (id_map_entry->sl_cm_id > sl_cm_id)
115 node = node->rb_left;
116 else if (id_map_entry->sl_cm_id < sl_cm_id)
117 node = node->rb_right;
118 else if (id_map_entry->slave_id > slave_id)
119 node = node->rb_left;
120 else if (id_map_entry->slave_id < slave_id)
121 node = node->rb_right;
128 static void id_map_ent_timeout(struct work_struct *work)
130 struct delayed_work *delay = to_delayed_work(work);
131 struct id_map_entry *ent = container_of(delay, struct id_map_entry, timeout);
132 struct id_map_entry *db_ent, *found_ent;
133 struct mlx4_ib_dev *dev = ent->dev;
134 struct mlx4_ib_sriov *sriov = &dev->sriov;
135 struct rb_root *sl_id_map = &sriov->sl_id_map;
136 int pv_id = (int) ent->pv_cm_id;
138 spin_lock(&sriov->id_map_lock);
139 db_ent = (struct id_map_entry *)idr_find(&sriov->pv_id_table, pv_id);
142 found_ent = id_map_find_by_sl_id(&dev->ib_dev, ent->slave_id, ent->sl_cm_id);
143 if (found_ent && found_ent == ent)
144 rb_erase(&found_ent->node, sl_id_map);
145 idr_remove(&sriov->pv_id_table, pv_id);
148 list_del(&ent->list);
149 spin_unlock(&sriov->id_map_lock);
153 static void id_map_find_del(struct ib_device *ibdev, int pv_cm_id)
155 struct mlx4_ib_sriov *sriov = &to_mdev(ibdev)->sriov;
156 struct rb_root *sl_id_map = &sriov->sl_id_map;
157 struct id_map_entry *ent, *found_ent;
159 spin_lock(&sriov->id_map_lock);
160 ent = (struct id_map_entry *)idr_find(&sriov->pv_id_table, pv_cm_id);
163 found_ent = id_map_find_by_sl_id(ibdev, ent->slave_id, ent->sl_cm_id);
164 if (found_ent && found_ent == ent)
165 rb_erase(&found_ent->node, sl_id_map);
166 idr_remove(&sriov->pv_id_table, pv_cm_id);
168 spin_unlock(&sriov->id_map_lock);
171 static void sl_id_map_add(struct ib_device *ibdev, struct id_map_entry *new)
173 struct rb_root *sl_id_map = &to_mdev(ibdev)->sriov.sl_id_map;
174 struct rb_node **link = &sl_id_map->rb_node, *parent = NULL;
175 struct id_map_entry *ent;
176 int slave_id = new->slave_id;
177 int sl_cm_id = new->sl_cm_id;
179 ent = id_map_find_by_sl_id(ibdev, slave_id, sl_cm_id);
181 pr_debug("overriding existing sl_id_map entry (cm_id = %x)\n",
184 rb_replace_node(&ent->node, &new->node, sl_id_map);
188 /* Go to the bottom of the tree */
191 ent = rb_entry(parent, struct id_map_entry, node);
193 if (ent->sl_cm_id > sl_cm_id || (ent->sl_cm_id == sl_cm_id && ent->slave_id > slave_id))
194 link = &(*link)->rb_left;
196 link = &(*link)->rb_right;
199 rb_link_node(&new->node, parent, link);
200 rb_insert_color(&new->node, sl_id_map);
203 static struct id_map_entry *
204 id_map_alloc(struct ib_device *ibdev, int slave_id, u32 sl_cm_id)
208 struct id_map_entry *ent;
209 struct mlx4_ib_sriov *sriov = &to_mdev(ibdev)->sriov;
211 ent = kmalloc(sizeof (struct id_map_entry), GFP_KERNEL);
213 mlx4_ib_warn(ibdev, "Couldn't allocate id cache entry - out of memory\n");
214 return ERR_PTR(-ENOMEM);
217 ent->sl_cm_id = sl_cm_id;
218 ent->slave_id = slave_id;
219 ent->scheduled_delete = 0;
220 ent->dev = to_mdev(ibdev);
221 INIT_DELAYED_WORK(&ent->timeout, id_map_ent_timeout);
223 idr_preload(GFP_KERNEL);
224 spin_lock(&to_mdev(ibdev)->sriov.id_map_lock);
226 ret = idr_alloc(&sriov->pv_id_table, ent, next_id, 0, GFP_NOWAIT);
228 next_id = max(ret + 1, 0);
229 ent->pv_cm_id = (u32)ret;
230 sl_id_map_add(ibdev, ent);
231 list_add_tail(&ent->list, &sriov->cm_list);
234 spin_unlock(&sriov->id_map_lock);
242 mlx4_ib_warn(ibdev, "No more space in the idr (err:0x%x)\n", ret);
243 return ERR_PTR(-ENOMEM);
246 static struct id_map_entry *
247 id_map_get(struct ib_device *ibdev, int *pv_cm_id, int sl_cm_id, int slave_id)
249 struct id_map_entry *ent;
250 struct mlx4_ib_sriov *sriov = &to_mdev(ibdev)->sriov;
252 spin_lock(&sriov->id_map_lock);
253 if (*pv_cm_id == -1) {
254 ent = id_map_find_by_sl_id(ibdev, sl_cm_id, slave_id);
256 *pv_cm_id = (int) ent->pv_cm_id;
258 ent = (struct id_map_entry *)idr_find(&sriov->pv_id_table, *pv_cm_id);
259 spin_unlock(&sriov->id_map_lock);
264 static void schedule_delayed(struct ib_device *ibdev, struct id_map_entry *id)
266 struct mlx4_ib_sriov *sriov = &to_mdev(ibdev)->sriov;
269 spin_lock(&sriov->id_map_lock);
270 spin_lock_irqsave(&sriov->going_down_lock, flags);
271 /*make sure that there is no schedule inside the scheduled work.*/
272 if (!sriov->is_going_down) {
273 id->scheduled_delete = 1;
274 schedule_delayed_work(&id->timeout, CM_CLEANUP_CACHE_TIMEOUT);
276 spin_unlock_irqrestore(&sriov->going_down_lock, flags);
277 spin_unlock(&sriov->id_map_lock);
280 int mlx4_ib_multiplex_cm_handler(struct ib_device *ibdev, int port, int slave_id,
283 struct id_map_entry *id;
287 sl_cm_id = get_local_comm_id(mad);
289 if (mad->mad_hdr.attr_id == CM_REQ_ATTR_ID ||
290 mad->mad_hdr.attr_id == CM_REP_ATTR_ID) {
291 id = id_map_alloc(ibdev, slave_id, sl_cm_id);
293 mlx4_ib_warn(ibdev, "%s: id{slave: %d, sl_cm_id: 0x%x} Failed to id_map_alloc\n",
294 __func__, slave_id, sl_cm_id);
297 } else if (mad->mad_hdr.attr_id == CM_REJ_ATTR_ID) {
300 id = id_map_get(ibdev, &pv_cm_id, slave_id, sl_cm_id);
304 pr_debug("id{slave: %d, sl_cm_id: 0x%x} is NULL!\n",
309 set_local_comm_id(mad, id->pv_cm_id);
311 if (mad->mad_hdr.attr_id == CM_DREQ_ATTR_ID)
312 schedule_delayed(ibdev, id);
313 else if (mad->mad_hdr.attr_id == CM_DREP_ATTR_ID)
314 id_map_find_del(ibdev, pv_cm_id);
319 int mlx4_ib_demux_cm_handler(struct ib_device *ibdev, int port, int *slave,
323 struct id_map_entry *id;
325 if (mad->mad_hdr.attr_id == CM_REQ_ATTR_ID) {
328 gid = gid_from_req_msg(ibdev, mad);
329 *slave = mlx4_ib_find_real_gid(ibdev, port, gid.global.interface_id);
331 mlx4_ib_warn(ibdev, "failed matching slave_id by gid (0x%llx)\n",
332 gid.global.interface_id);
338 pv_cm_id = get_remote_comm_id(mad);
339 id = id_map_get(ibdev, (int *)&pv_cm_id, -1, -1);
342 pr_debug("Couldn't find an entry for pv_cm_id 0x%x\n", pv_cm_id);
346 *slave = id->slave_id;
347 set_remote_comm_id(mad, id->sl_cm_id);
349 if (mad->mad_hdr.attr_id == CM_DREQ_ATTR_ID)
350 schedule_delayed(ibdev, id);
351 else if (mad->mad_hdr.attr_id == CM_REJ_ATTR_ID ||
352 mad->mad_hdr.attr_id == CM_DREP_ATTR_ID) {
353 id_map_find_del(ibdev, (int) pv_cm_id);
359 void mlx4_ib_cm_paravirt_init(struct mlx4_ib_dev *dev)
361 spin_lock_init(&dev->sriov.id_map_lock);
362 INIT_LIST_HEAD(&dev->sriov.cm_list);
363 dev->sriov.sl_id_map = RB_ROOT;
364 idr_init(&dev->sriov.pv_id_table);
365 idr_pre_get(&dev->sriov.pv_id_table, GFP_KERNEL);
368 /* slave = -1 ==> all slaves */
369 /* TBD -- call paravirt clean for single slave. Need for slave RESET event */
370 void mlx4_ib_cm_paravirt_clean(struct mlx4_ib_dev *dev, int slave)
372 struct mlx4_ib_sriov *sriov = &dev->sriov;
373 struct rb_root *sl_id_map = &sriov->sl_id_map;
377 struct id_map_entry *map, *tmp_map;
378 /* cancel all delayed work queue entries */
380 spin_lock(&sriov->id_map_lock);
381 list_for_each_entry_safe(map, tmp_map, &dev->sriov.cm_list, list) {
382 if (slave < 0 || slave == map->slave_id) {
383 if (map->scheduled_delete)
384 need_flush &= !!cancel_delayed_work(&map->timeout);
388 spin_unlock(&sriov->id_map_lock);
391 flush_scheduled_work(); /* make sure all timers were flushed */
393 /* now, remove all leftover entries from databases*/
394 spin_lock(&sriov->id_map_lock);
396 while (rb_first(sl_id_map)) {
397 struct id_map_entry *ent =
398 rb_entry(rb_first(sl_id_map),
399 struct id_map_entry, node);
401 rb_erase(&ent->node, sl_id_map);
402 idr_remove(&sriov->pv_id_table, (int) ent->pv_cm_id);
404 list_splice_init(&dev->sriov.cm_list, &lh);
406 /* first, move nodes belonging to slave to db remove list */
407 nd = rb_first(sl_id_map);
409 struct id_map_entry *ent =
410 rb_entry(nd, struct id_map_entry, node);
412 if (ent->slave_id == slave)
413 list_move_tail(&ent->list, &lh);
415 /* remove those nodes from databases */
416 list_for_each_entry_safe(map, tmp_map, &lh, list) {
417 rb_erase(&map->node, sl_id_map);
418 idr_remove(&sriov->pv_id_table, (int) map->pv_cm_id);
421 /* add remaining nodes from cm_list */
422 list_for_each_entry_safe(map, tmp_map, &dev->sriov.cm_list, list) {
423 if (slave == map->slave_id)
424 list_move_tail(&map->list, &lh);
428 spin_unlock(&sriov->id_map_lock);
430 /* free any map entries left behind due to cancel_delayed_work above */
431 list_for_each_entry_safe(map, tmp_map, &lh, list) {
432 list_del(&map->list);