drivers/infiniband/hw/mlx4/cm.c

   1 /*
   2  * Copyright (c) 2012 Mellanox Technologies. All rights reserved.
   3  *
   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:
   9  *
  10  *     Redistribution and use in source and binary forms, with or
  11  *     without modification, are permitted provided that the following
  12  *     conditions are met:
  13  *
  14  *      - Redistributions of source code must retain the above
  15  *        copyright notice, this list of conditions and the following
  16  *        disclaimer.
  17  *
  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.
  22  *
  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
  30  * SOFTWARE.
  31  */
  32
  33 #include <rdma/ib_mad.h>
  34
  35 #include <linux/mlx4/cmd.h>
  36 #include <linux/rbtree.h>
  37 #include <linux/idr.h>
  38 #include <rdma/ib_cm.h>
  39
  40 #include "mlx4_ib.h"
  41
  42 #define CM_CLEANUP_CACHE_TIMEOUT  (30 * HZ)
  43
  44 struct id_map_entry {
  45         struct rb_node node;
  46
  47         u32 sl_cm_id;
  48         u32 pv_cm_id;
  49         int slave_id;
  50         int scheduled_delete;
  51         struct mlx4_ib_dev *dev;
  52
  53         struct list_head list;
  54         struct delayed_work timeout;
  55 };
  56
  57 struct cm_generic_msg {
  58         struct ib_mad_hdr hdr;
  59
  60         __be32 local_comm_id;
  61         __be32 remote_comm_id;
  62 };
  63
  64 struct cm_sidr_generic_msg {
  65         struct ib_mad_hdr hdr;
  66         __be32 request_id;
  67 };
  68
  69 struct cm_req_msg {
  70         unsigned char unused[0x60];
  71         union ib_gid primary_path_sgid;
  72 };
  73
  74
  75 static void set_local_comm_id(struct ib_mad *mad, u32 cm_id)
  76 {
  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");
  83                 return;
  84         } else {
  85                 struct cm_generic_msg *msg = (struct cm_generic_msg *)mad;
  86                 msg->local_comm_id = cpu_to_be32(cm_id);
  87         }
  88 }
  89
  90 static u32 get_local_comm_id(struct ib_mad *mad)
  91 {
  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");
  98                 return -1;
  99         } else {
 100                 struct cm_generic_msg *msg = (struct cm_generic_msg *)mad;
 101                 return be32_to_cpu(msg->local_comm_id);
 102         }
 103 }
 104
 105 static void set_remote_comm_id(struct ib_mad *mad, u32 cm_id)
 106 {
 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");
 113                 return;
 114         } else {
 115                 struct cm_generic_msg *msg = (struct cm_generic_msg *)mad;
 116                 msg->remote_comm_id = cpu_to_be32(cm_id);
 117         }
 118 }
 119
 120 static u32 get_remote_comm_id(struct ib_mad *mad)
 121 {
 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");
 128                 return -1;
 129         } else {
 130                 struct cm_generic_msg *msg = (struct cm_generic_msg *)mad;
 131                 return be32_to_cpu(msg->remote_comm_id);
 132         }
 133 }
 134
 135 static union ib_gid gid_from_req_msg(struct ib_device *ibdev, struct ib_mad *mad)
 136 {
 137         struct cm_req_msg *msg = (struct cm_req_msg *)mad;
 138
 139         return msg->primary_path_sgid;
 140 }
 141
 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)
 145 {
 146         struct rb_root *sl_id_map = &to_mdev(ibdev)->sriov.sl_id_map;
 147         struct rb_node *node = sl_id_map->rb_node;
 148
 149         while (node) {
 150                 struct id_map_entry *id_map_entry =
 151                         rb_entry(node, struct id_map_entry, node);
 152
 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;
 161                 else
 162                         return id_map_entry;
 163         }
 164         return NULL;
 165 }
 166
 167 static void id_map_ent_timeout(struct work_struct *work)
 168 {
 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 *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
 176         spin_lock(&sriov->id_map_lock);
 177         if (!xa_erase(&sriov->pv_id_table, ent->pv_cm_id))
 178                 goto out;
 179         found_ent = id_map_find_by_sl_id(&dev->ib_dev, ent->slave_id, ent->sl_cm_id);
 180         if (found_ent && found_ent == ent)
 181                 rb_erase(&found_ent->node, sl_id_map);
 182
 183 out:
 184         list_del(&ent->list);
 185         spin_unlock(&sriov->id_map_lock);
 186         kfree(ent);
 187 }
 188
 189 static void sl_id_map_add(struct ib_device *ibdev, struct id_map_entry *new)
 190 {
 191         struct rb_root *sl_id_map = &to_mdev(ibdev)->sriov.sl_id_map;
 192         struct rb_node **link = &sl_id_map->rb_node, *parent = NULL;
 193         struct id_map_entry *ent;
 194         int slave_id = new->slave_id;
 195         int sl_cm_id = new->sl_cm_id;
 196
 197         ent = id_map_find_by_sl_id(ibdev, slave_id, sl_cm_id);
 198         if (ent) {
 199                 pr_debug("overriding existing sl_id_map entry (cm_id = %x)\n",
 200                          sl_cm_id);
 201
 202                 rb_replace_node(&ent->node, &new->node, sl_id_map);
 203                 return;
 204         }
 205
 206         /* Go to the bottom of the tree */
 207         while (*link) {
 208                 parent = *link;
 209                 ent = rb_entry(parent, struct id_map_entry, node);
 210
 211                 if (ent->sl_cm_id > sl_cm_id || (ent->sl_cm_id == sl_cm_id && ent->slave_id > slave_id))
 212                         link = &(*link)->rb_left;
 213                 else
 214                         link = &(*link)->rb_right;
 215         }
 216
 217         rb_link_node(&new->node, parent, link);
 218         rb_insert_color(&new->node, sl_id_map);
 219 }
 220
 221 static struct id_map_entry *
 222 id_map_alloc(struct ib_device *ibdev, int slave_id, u32 sl_cm_id)
 223 {
 224         int ret;
 225         struct id_map_entry *ent;
 226         struct mlx4_ib_sriov *sriov = &to_mdev(ibdev)->sriov;
 227
 228         ent = kmalloc(sizeof (struct id_map_entry), GFP_KERNEL);
 229         if (!ent)
 230                 return ERR_PTR(-ENOMEM);
 231
 232         ent->sl_cm_id = sl_cm_id;
 233         ent->slave_id = slave_id;
 234         ent->scheduled_delete = 0;
 235         ent->dev = to_mdev(ibdev);
 236         INIT_DELAYED_WORK(&ent->timeout, id_map_ent_timeout);
 237
 238         ret = xa_alloc_cyclic(&sriov->pv_id_table, &ent->pv_cm_id, ent,
 239                         xa_limit_32b, &sriov->pv_id_next, GFP_KERNEL);
 240         if (ret >= 0) {
 241                 spin_lock(&sriov->id_map_lock);
 242                 sl_id_map_add(ibdev, ent);
 243                 list_add_tail(&ent->list, &sriov->cm_list);
 244                 spin_unlock(&sriov->id_map_lock);
 245                 return ent;
 246         }
 247
 248         /*error flow*/
 249         kfree(ent);
 250         mlx4_ib_warn(ibdev, "Allocation failed (err:0x%x)\n", ret);
 251         return ERR_PTR(-ENOMEM);
 252 }
 253
 254 static struct id_map_entry *
 255 id_map_get(struct ib_device *ibdev, int *pv_cm_id, int slave_id, int sl_cm_id)
 256 {
 257         struct id_map_entry *ent;
 258         struct mlx4_ib_sriov *sriov = &to_mdev(ibdev)->sriov;
 259
 260         spin_lock(&sriov->id_map_lock);
 261         if (*pv_cm_id == -1) {
 262                 ent = id_map_find_by_sl_id(ibdev, slave_id, sl_cm_id);
 263                 if (ent)
 264                         *pv_cm_id = (int) ent->pv_cm_id;
 265         } else
 266                 ent = xa_load(&sriov->pv_id_table, *pv_cm_id);
 267         spin_unlock(&sriov->id_map_lock);
 268
 269         return ent;
 270 }
 271
 272 static void schedule_delayed(struct ib_device *ibdev, struct id_map_entry *id)
 273 {
 274         struct mlx4_ib_sriov *sriov = &to_mdev(ibdev)->sriov;
 275         unsigned long flags;
 276
 277         spin_lock(&sriov->id_map_lock);
 278         spin_lock_irqsave(&sriov->going_down_lock, flags);
 279         /*make sure that there is no schedule inside the scheduled work.*/
 280         if (!sriov->is_going_down && !id->scheduled_delete) {
 281                 id->scheduled_delete = 1;
 282                 schedule_delayed_work(&id->timeout, CM_CLEANUP_CACHE_TIMEOUT);
 283         }
 284         spin_unlock_irqrestore(&sriov->going_down_lock, flags);
 285         spin_unlock(&sriov->id_map_lock);
 286 }
 287
 288 int mlx4_ib_multiplex_cm_handler(struct ib_device *ibdev, int port, int slave_id,
 289                 struct ib_mad *mad)
 290 {
 291         struct id_map_entry *id;
 292         u32 sl_cm_id;
 293         int pv_cm_id = -1;
 294
 295         if (mad->mad_hdr.attr_id == CM_REQ_ATTR_ID ||
 296                         mad->mad_hdr.attr_id == CM_REP_ATTR_ID ||
 297                         mad->mad_hdr.attr_id == CM_SIDR_REQ_ATTR_ID) {
 298                 sl_cm_id = get_local_comm_id(mad);
 299                 id = id_map_get(ibdev, &pv_cm_id, slave_id, sl_cm_id);
 300                 if (id)
 301                         goto cont;
 302                 id = id_map_alloc(ibdev, slave_id, sl_cm_id);
 303                 if (IS_ERR(id)) {
 304                         mlx4_ib_warn(ibdev, "%s: id{slave: %d, sl_cm_id: 0x%x} Failed to id_map_alloc\n",
 305                                 __func__, slave_id, sl_cm_id);
 306                         return PTR_ERR(id);
 307                 }
 308         } else if (mad->mad_hdr.attr_id == CM_REJ_ATTR_ID ||
 309                    mad->mad_hdr.attr_id == CM_SIDR_REP_ATTR_ID) {
 310                 return 0;
 311         } else {
 312                 sl_cm_id = get_local_comm_id(mad);
 313                 id = id_map_get(ibdev, &pv_cm_id, slave_id, sl_cm_id);
 314         }
 315
 316         if (!id) {
 317                 pr_debug("id{slave: %d, sl_cm_id: 0x%x} is NULL!\n",
 318                          slave_id, sl_cm_id);
 319                 return -EINVAL;
 320         }
 321
 322 cont:
 323         set_local_comm_id(mad, id->pv_cm_id);
 324
 325         if (mad->mad_hdr.attr_id == CM_DREQ_ATTR_ID)
 326                 schedule_delayed(ibdev, id);
 327         return 0;
 328 }
 329
 330 int mlx4_ib_demux_cm_handler(struct ib_device *ibdev, int port, int *slave,
 331                              struct ib_mad *mad)
 332 {
 333         u32 pv_cm_id;
 334         struct id_map_entry *id;
 335
 336         if (mad->mad_hdr.attr_id == CM_REQ_ATTR_ID ||
 337             mad->mad_hdr.attr_id == CM_SIDR_REQ_ATTR_ID) {
 338                 union ib_gid gid;
 339
 340                 if (!slave)
 341                         return 0;
 342
 343                 gid = gid_from_req_msg(ibdev, mad);
 344                 *slave = mlx4_ib_find_real_gid(ibdev, port, gid.global.interface_id);
 345                 if (*slave < 0) {
 346                         mlx4_ib_warn(ibdev, "failed matching slave_id by gid (0x%llx)\n",
 347                                      be64_to_cpu(gid.global.interface_id));
 348                         return -ENOENT;
 349                 }
 350                 return 0;
 351         }
 352
 353         pv_cm_id = get_remote_comm_id(mad);
 354         id = id_map_get(ibdev, (int *)&pv_cm_id, -1, -1);
 355
 356         if (!id) {
 357                 pr_debug("Couldn't find an entry for pv_cm_id 0x%x\n", pv_cm_id);
 358                 return -ENOENT;
 359         }
 360
 361         if (slave)
 362                 *slave = id->slave_id;
 363         set_remote_comm_id(mad, id->sl_cm_id);
 364
 365         if (mad->mad_hdr.attr_id == CM_DREQ_ATTR_ID ||
 366             mad->mad_hdr.attr_id == CM_REJ_ATTR_ID)
 367                 schedule_delayed(ibdev, id);
 368
 369         return 0;
 370 }
 371
 372 void mlx4_ib_cm_paravirt_init(struct mlx4_ib_dev *dev)
 373 {
 374         spin_lock_init(&dev->sriov.id_map_lock);
 375         INIT_LIST_HEAD(&dev->sriov.cm_list);
 376         dev->sriov.sl_id_map = RB_ROOT;
 377         xa_init_flags(&dev->sriov.pv_id_table, XA_FLAGS_ALLOC);
 378 }
 379
 380 /* slave = -1 ==> all slaves */
 381 /* TBD -- call paravirt clean for single slave.  Need for slave RESET event */
 382 void mlx4_ib_cm_paravirt_clean(struct mlx4_ib_dev *dev, int slave)
 383 {
 384         struct mlx4_ib_sriov *sriov = &dev->sriov;
 385         struct rb_root *sl_id_map = &sriov->sl_id_map;
 386         struct list_head lh;
 387         struct rb_node *nd;
 388         int need_flush = 0;
 389         struct id_map_entry *map, *tmp_map;
 390         /* cancel all delayed work queue entries */
 391         INIT_LIST_HEAD(&lh);
 392         spin_lock(&sriov->id_map_lock);
 393         list_for_each_entry_safe(map, tmp_map, &dev->sriov.cm_list, list) {
 394                 if (slave < 0 || slave == map->slave_id) {
 395                         if (map->scheduled_delete)
 396                                 need_flush |= !cancel_delayed_work(&map->timeout);
 397                 }
 398         }
 399
 400         spin_unlock(&sriov->id_map_lock);
 401
 402         if (need_flush)
 403                 flush_scheduled_work(); /* make sure all timers were flushed */
 404
 405         /* now, remove all leftover entries from databases*/
 406         spin_lock(&sriov->id_map_lock);
 407         if (slave < 0) {
 408                 while (rb_first(sl_id_map)) {
 409                         struct id_map_entry *ent =
 410                                 rb_entry(rb_first(sl_id_map),
 411                                          struct id_map_entry, node);
 412
 413                         rb_erase(&ent->node, sl_id_map);
 414                         xa_erase(&sriov->pv_id_table, ent->pv_cm_id);
 415                 }
 416                 list_splice_init(&dev->sriov.cm_list, &lh);
 417         } else {
 418                 /* first, move nodes belonging to slave to db remove list */
 419                 nd = rb_first(sl_id_map);
 420                 while (nd) {
 421                         struct id_map_entry *ent =
 422                                 rb_entry(nd, struct id_map_entry, node);
 423                         nd = rb_next(nd);
 424                         if (ent->slave_id == slave)
 425                                 list_move_tail(&ent->list, &lh);
 426                 }
 427                 /* remove those nodes from databases */
 428                 list_for_each_entry_safe(map, tmp_map, &lh, list) {
 429                         rb_erase(&map->node, sl_id_map);
 430                         xa_erase(&sriov->pv_id_table, map->pv_cm_id);
 431                 }
 432
 433                 /* add remaining nodes from cm_list */
 434                 list_for_each_entry_safe(map, tmp_map, &dev->sriov.cm_list, list) {
 435                         if (slave == map->slave_id)
 436                                 list_move_tail(&map->list, &lh);
 437                 }
 438         }
 439
 440         spin_unlock(&sriov->id_map_lock);
 441
 442         /* free any map entries left behind due to cancel_delayed_work above */
 443         list_for_each_entry_safe(map, tmp_map, &lh, list) {
 444                 list_del(&map->list);
 445                 kfree(map);
 446         }
 447 }