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;
64 struct cm_sidr_generic_msg {
65 struct ib_mad_hdr hdr;
70 unsigned char unused[0x60];
71 union ib_gid primary_path_sgid;
75 static void set_local_comm_id(struct ib_mad *mad, u32 cm_id)
77 if (mad->mad_hdr.attr_id == CM_SIDR_REQ_ATTR_ID) {
78 struct cm_sidr_generic_msg *msg =
79 (struct cm_sidr_generic_msg *)mad;
80 msg->request_id = cpu_to_be32(cm_id);
81 } else if (mad->mad_hdr.attr_id == CM_SIDR_REP_ATTR_ID) {
82 pr_err("trying to set local_comm_id in SIDR_REP\n");
85 struct cm_generic_msg *msg = (struct cm_generic_msg *)mad;
86 msg->local_comm_id = cpu_to_be32(cm_id);
90 static u32 get_local_comm_id(struct ib_mad *mad)
92 if (mad->mad_hdr.attr_id == CM_SIDR_REQ_ATTR_ID) {
93 struct cm_sidr_generic_msg *msg =
94 (struct cm_sidr_generic_msg *)mad;
95 return be32_to_cpu(msg->request_id);
96 } else if (mad->mad_hdr.attr_id == CM_SIDR_REP_ATTR_ID) {
97 pr_err("trying to set local_comm_id in SIDR_REP\n");
100 struct cm_generic_msg *msg = (struct cm_generic_msg *)mad;
101 return be32_to_cpu(msg->local_comm_id);
105 static void set_remote_comm_id(struct ib_mad *mad, u32 cm_id)
107 if (mad->mad_hdr.attr_id == CM_SIDR_REP_ATTR_ID) {
108 struct cm_sidr_generic_msg *msg =
109 (struct cm_sidr_generic_msg *)mad;
110 msg->request_id = cpu_to_be32(cm_id);
111 } else if (mad->mad_hdr.attr_id == CM_SIDR_REQ_ATTR_ID) {
112 pr_err("trying to set remote_comm_id in SIDR_REQ\n");
115 struct cm_generic_msg *msg = (struct cm_generic_msg *)mad;
116 msg->remote_comm_id = cpu_to_be32(cm_id);
120 static u32 get_remote_comm_id(struct ib_mad *mad)
122 if (mad->mad_hdr.attr_id == CM_SIDR_REP_ATTR_ID) {
123 struct cm_sidr_generic_msg *msg =
124 (struct cm_sidr_generic_msg *)mad;
125 return be32_to_cpu(msg->request_id);
126 } else if (mad->mad_hdr.attr_id == CM_SIDR_REQ_ATTR_ID) {
127 pr_err("trying to set remote_comm_id in SIDR_REQ\n");
130 struct cm_generic_msg *msg = (struct cm_generic_msg *)mad;
131 return be32_to_cpu(msg->remote_comm_id);
135 static union ib_gid gid_from_req_msg(struct ib_device *ibdev, struct ib_mad *mad)
137 struct cm_req_msg *msg = (struct cm_req_msg *)mad;
139 return msg->primary_path_sgid;
142 /* Lock should be taken before called */
143 static struct id_map_entry *
144 id_map_find_by_sl_id(struct ib_device *ibdev, u32 slave_id, u32 sl_cm_id)
146 struct rb_root *sl_id_map = &to_mdev(ibdev)->sriov.sl_id_map;
147 struct rb_node *node = sl_id_map->rb_node;
150 struct id_map_entry *id_map_entry =
151 rb_entry(node, struct id_map_entry, node);
153 if (id_map_entry->sl_cm_id > sl_cm_id)
154 node = node->rb_left;
155 else if (id_map_entry->sl_cm_id < sl_cm_id)
156 node = node->rb_right;
157 else if (id_map_entry->slave_id > slave_id)
158 node = node->rb_left;
159 else if (id_map_entry->slave_id < slave_id)
160 node = node->rb_right;
167 static void id_map_ent_timeout(struct work_struct *work)
169 struct delayed_work *delay = to_delayed_work(work);
170 struct id_map_entry *ent = container_of(delay, struct id_map_entry, timeout);
171 struct id_map_entry *db_ent, *found_ent;
172 struct mlx4_ib_dev *dev = ent->dev;
173 struct mlx4_ib_sriov *sriov = &dev->sriov;
174 struct rb_root *sl_id_map = &sriov->sl_id_map;
175 int pv_id = (int) ent->pv_cm_id;
177 spin_lock(&sriov->id_map_lock);
178 db_ent = (struct id_map_entry *)idr_find(&sriov->pv_id_table, pv_id);
181 found_ent = id_map_find_by_sl_id(&dev->ib_dev, ent->slave_id, ent->sl_cm_id);
182 if (found_ent && found_ent == ent)
183 rb_erase(&found_ent->node, sl_id_map);
184 idr_remove(&sriov->pv_id_table, pv_id);
187 list_del(&ent->list);
188 spin_unlock(&sriov->id_map_lock);
192 static void id_map_find_del(struct ib_device *ibdev, int pv_cm_id)
194 struct mlx4_ib_sriov *sriov = &to_mdev(ibdev)->sriov;
195 struct rb_root *sl_id_map = &sriov->sl_id_map;
196 struct id_map_entry *ent, *found_ent;
198 spin_lock(&sriov->id_map_lock);
199 ent = (struct id_map_entry *)idr_find(&sriov->pv_id_table, pv_cm_id);
202 found_ent = id_map_find_by_sl_id(ibdev, ent->slave_id, ent->sl_cm_id);
203 if (found_ent && found_ent == ent)
204 rb_erase(&found_ent->node, sl_id_map);
205 idr_remove(&sriov->pv_id_table, pv_cm_id);
207 spin_unlock(&sriov->id_map_lock);
210 static void sl_id_map_add(struct ib_device *ibdev, struct id_map_entry *new)
212 struct rb_root *sl_id_map = &to_mdev(ibdev)->sriov.sl_id_map;
213 struct rb_node **link = &sl_id_map->rb_node, *parent = NULL;
214 struct id_map_entry *ent;
215 int slave_id = new->slave_id;
216 int sl_cm_id = new->sl_cm_id;
218 ent = id_map_find_by_sl_id(ibdev, slave_id, sl_cm_id);
220 pr_debug("overriding existing sl_id_map entry (cm_id = %x)\n",
223 rb_replace_node(&ent->node, &new->node, sl_id_map);
227 /* Go to the bottom of the tree */
230 ent = rb_entry(parent, struct id_map_entry, node);
232 if (ent->sl_cm_id > sl_cm_id || (ent->sl_cm_id == sl_cm_id && ent->slave_id > slave_id))
233 link = &(*link)->rb_left;
235 link = &(*link)->rb_right;
238 rb_link_node(&new->node, parent, link);
239 rb_insert_color(&new->node, sl_id_map);
242 static struct id_map_entry *
243 id_map_alloc(struct ib_device *ibdev, int slave_id, u32 sl_cm_id)
246 struct id_map_entry *ent;
247 struct mlx4_ib_sriov *sriov = &to_mdev(ibdev)->sriov;
249 ent = kmalloc(sizeof (struct id_map_entry), GFP_KERNEL);
251 return ERR_PTR(-ENOMEM);
253 ent->sl_cm_id = sl_cm_id;
254 ent->slave_id = slave_id;
255 ent->scheduled_delete = 0;
256 ent->dev = to_mdev(ibdev);
257 INIT_DELAYED_WORK(&ent->timeout, id_map_ent_timeout);
259 idr_preload(GFP_KERNEL);
260 spin_lock(&to_mdev(ibdev)->sriov.id_map_lock);
262 ret = idr_alloc_cyclic(&sriov->pv_id_table, ent, 0, 0, GFP_NOWAIT);
264 ent->pv_cm_id = (u32)ret;
265 sl_id_map_add(ibdev, ent);
266 list_add_tail(&ent->list, &sriov->cm_list);
269 spin_unlock(&sriov->id_map_lock);
277 mlx4_ib_warn(ibdev, "No more space in the idr (err:0x%x)\n", ret);
278 return ERR_PTR(-ENOMEM);
281 static struct id_map_entry *
282 id_map_get(struct ib_device *ibdev, int *pv_cm_id, int slave_id, int sl_cm_id)
284 struct id_map_entry *ent;
285 struct mlx4_ib_sriov *sriov = &to_mdev(ibdev)->sriov;
287 spin_lock(&sriov->id_map_lock);
288 if (*pv_cm_id == -1) {
289 ent = id_map_find_by_sl_id(ibdev, slave_id, sl_cm_id);
291 *pv_cm_id = (int) ent->pv_cm_id;
293 ent = (struct id_map_entry *)idr_find(&sriov->pv_id_table, *pv_cm_id);
294 spin_unlock(&sriov->id_map_lock);
299 static void schedule_delayed(struct ib_device *ibdev, struct id_map_entry *id)
301 struct mlx4_ib_sriov *sriov = &to_mdev(ibdev)->sriov;
304 spin_lock(&sriov->id_map_lock);
305 spin_lock_irqsave(&sriov->going_down_lock, flags);
306 /*make sure that there is no schedule inside the scheduled work.*/
307 if (!sriov->is_going_down) {
308 id->scheduled_delete = 1;
309 schedule_delayed_work(&id->timeout, CM_CLEANUP_CACHE_TIMEOUT);
311 spin_unlock_irqrestore(&sriov->going_down_lock, flags);
312 spin_unlock(&sriov->id_map_lock);
315 int mlx4_ib_multiplex_cm_handler(struct ib_device *ibdev, int port, int slave_id,
318 struct id_map_entry *id;
322 if (mad->mad_hdr.attr_id == CM_REQ_ATTR_ID ||
323 mad->mad_hdr.attr_id == CM_REP_ATTR_ID ||
324 mad->mad_hdr.attr_id == CM_SIDR_REQ_ATTR_ID) {
325 sl_cm_id = get_local_comm_id(mad);
326 id = id_map_alloc(ibdev, slave_id, sl_cm_id);
328 mlx4_ib_warn(ibdev, "%s: id{slave: %d, sl_cm_id: 0x%x} Failed to id_map_alloc\n",
329 __func__, slave_id, sl_cm_id);
332 } else if (mad->mad_hdr.attr_id == CM_REJ_ATTR_ID ||
333 mad->mad_hdr.attr_id == CM_SIDR_REP_ATTR_ID) {
336 sl_cm_id = get_local_comm_id(mad);
337 id = id_map_get(ibdev, &pv_cm_id, slave_id, sl_cm_id);
341 pr_debug("id{slave: %d, sl_cm_id: 0x%x} is NULL!\n",
346 set_local_comm_id(mad, id->pv_cm_id);
348 if (mad->mad_hdr.attr_id == CM_DREQ_ATTR_ID)
349 schedule_delayed(ibdev, id);
350 else if (mad->mad_hdr.attr_id == CM_DREP_ATTR_ID)
351 id_map_find_del(ibdev, pv_cm_id);
356 int mlx4_ib_demux_cm_handler(struct ib_device *ibdev, int port, int *slave,
360 struct id_map_entry *id;
362 if (mad->mad_hdr.attr_id == CM_REQ_ATTR_ID ||
363 mad->mad_hdr.attr_id == CM_SIDR_REQ_ATTR_ID) {
369 gid = gid_from_req_msg(ibdev, mad);
370 *slave = mlx4_ib_find_real_gid(ibdev, port, gid.global.interface_id);
372 mlx4_ib_warn(ibdev, "failed matching slave_id by gid (0x%llx)\n",
373 be64_to_cpu(gid.global.interface_id));
379 pv_cm_id = get_remote_comm_id(mad);
380 id = id_map_get(ibdev, (int *)&pv_cm_id, -1, -1);
383 pr_debug("Couldn't find an entry for pv_cm_id 0x%x\n", pv_cm_id);
388 *slave = id->slave_id;
389 set_remote_comm_id(mad, id->sl_cm_id);
391 if (mad->mad_hdr.attr_id == CM_DREQ_ATTR_ID)
392 schedule_delayed(ibdev, id);
393 else if (mad->mad_hdr.attr_id == CM_REJ_ATTR_ID ||
394 mad->mad_hdr.attr_id == CM_DREP_ATTR_ID) {
395 id_map_find_del(ibdev, (int) pv_cm_id);
401 void mlx4_ib_cm_paravirt_init(struct mlx4_ib_dev *dev)
403 spin_lock_init(&dev->sriov.id_map_lock);
404 INIT_LIST_HEAD(&dev->sriov.cm_list);
405 dev->sriov.sl_id_map = RB_ROOT;
406 idr_init(&dev->sriov.pv_id_table);
409 /* slave = -1 ==> all slaves */
410 /* TBD -- call paravirt clean for single slave. Need for slave RESET event */
411 void mlx4_ib_cm_paravirt_clean(struct mlx4_ib_dev *dev, int slave)
413 struct mlx4_ib_sriov *sriov = &dev->sriov;
414 struct rb_root *sl_id_map = &sriov->sl_id_map;
418 struct id_map_entry *map, *tmp_map;
419 /* cancel all delayed work queue entries */
421 spin_lock(&sriov->id_map_lock);
422 list_for_each_entry_safe(map, tmp_map, &dev->sriov.cm_list, list) {
423 if (slave < 0 || slave == map->slave_id) {
424 if (map->scheduled_delete)
425 need_flush |= !cancel_delayed_work(&map->timeout);
429 spin_unlock(&sriov->id_map_lock);
432 flush_scheduled_work(); /* make sure all timers were flushed */
434 /* now, remove all leftover entries from databases*/
435 spin_lock(&sriov->id_map_lock);
437 while (rb_first(sl_id_map)) {
438 struct id_map_entry *ent =
439 rb_entry(rb_first(sl_id_map),
440 struct id_map_entry, node);
442 rb_erase(&ent->node, sl_id_map);
443 idr_remove(&sriov->pv_id_table, (int) ent->pv_cm_id);
445 list_splice_init(&dev->sriov.cm_list, &lh);
447 /* first, move nodes belonging to slave to db remove list */
448 nd = rb_first(sl_id_map);
450 struct id_map_entry *ent =
451 rb_entry(nd, struct id_map_entry, node);
453 if (ent->slave_id == slave)
454 list_move_tail(&ent->list, &lh);
456 /* remove those nodes from databases */
457 list_for_each_entry_safe(map, tmp_map, &lh, list) {
458 rb_erase(&map->node, sl_id_map);
459 idr_remove(&sriov->pv_id_table, (int) map->pv_cm_id);
462 /* add remaining nodes from cm_list */
463 list_for_each_entry_safe(map, tmp_map, &dev->sriov.cm_list, list) {
464 if (slave == map->slave_id)
465 list_move_tail(&map->list, &lh);
469 spin_unlock(&sriov->id_map_lock);
471 /* free any map entries left behind due to cancel_delayed_work above */
472 list_for_each_entry_safe(map, tmp_map, &lh, list) {
473 list_del(&map->list);