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 44struct 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 57struct rej_tmout_entry { 58 int slave; 59 u32 rem_pv_cm_id; 60 struct delayed_work timeout; 61 struct xarray *xa_rej_tmout; 62}; 63 64struct cm_generic_msg { 65 struct ib_mad_hdr hdr; 66 67 __be32 local_comm_id; 68 __be32 remote_comm_id; 69 unsigned char unused[2]; 70 __be16 rej_reason; 71}; 72 73struct cm_sidr_generic_msg { 74 struct ib_mad_hdr hdr; 75 __be32 request_id; 76}; 77 78struct cm_req_msg { 79 unsigned char unused[0x60]; 80 union ib_gid primary_path_sgid; 81}; 82 83static struct workqueue_struct *cm_wq; 84 85static void set_local_comm_id(struct ib_mad *mad, u32 cm_id) 86{ 87 if (mad->mad_hdr.attr_id == CM_SIDR_REQ_ATTR_ID) { 88 struct cm_sidr_generic_msg *msg = 89 (struct cm_sidr_generic_msg *)mad; 90 msg->request_id = cpu_to_be32(cm_id); 91 } else if (mad->mad_hdr.attr_id == CM_SIDR_REP_ATTR_ID) { 92 pr_err("trying to set local_comm_id in SIDR_REP\n"); 93 return; 94 } else { 95 struct cm_generic_msg *msg = (struct cm_generic_msg *)mad; 96 msg->local_comm_id = cpu_to_be32(cm_id); 97 } 98} 99 100static u32 get_local_comm_id(struct ib_mad *mad) 101{ 102 if (mad->mad_hdr.attr_id == CM_SIDR_REQ_ATTR_ID) { 103 struct cm_sidr_generic_msg *msg = 104 (struct cm_sidr_generic_msg *)mad; 105 return be32_to_cpu(msg->request_id); 106 } else if (mad->mad_hdr.attr_id == CM_SIDR_REP_ATTR_ID) { 107 pr_err("trying to set local_comm_id in SIDR_REP\n"); 108 return -1; 109 } else { 110 struct cm_generic_msg *msg = (struct cm_generic_msg *)mad; 111 return be32_to_cpu(msg->local_comm_id); 112 } 113} 114 115static void set_remote_comm_id(struct ib_mad *mad, u32 cm_id) 116{ 117 if (mad->mad_hdr.attr_id == CM_SIDR_REP_ATTR_ID) { 118 struct cm_sidr_generic_msg *msg = 119 (struct cm_sidr_generic_msg *)mad; 120 msg->request_id = cpu_to_be32(cm_id); 121 } else if (mad->mad_hdr.attr_id == CM_SIDR_REQ_ATTR_ID) { 122 pr_err("trying to set remote_comm_id in SIDR_REQ\n"); 123 return; 124 } else { 125 struct cm_generic_msg *msg = (struct cm_generic_msg *)mad; 126 msg->remote_comm_id = cpu_to_be32(cm_id); 127 } 128} 129 130static u32 get_remote_comm_id(struct ib_mad *mad) 131{ 132 if (mad->mad_hdr.attr_id == CM_SIDR_REP_ATTR_ID) { 133 struct cm_sidr_generic_msg *msg = 134 (struct cm_sidr_generic_msg *)mad; 135 return be32_to_cpu(msg->request_id); 136 } else if (mad->mad_hdr.attr_id == CM_SIDR_REQ_ATTR_ID) { 137 pr_err("trying to set remote_comm_id in SIDR_REQ\n"); 138 return -1; 139 } else { 140 struct cm_generic_msg *msg = (struct cm_generic_msg *)mad; 141 return be32_to_cpu(msg->remote_comm_id); 142 } 143} 144 145static union ib_gid gid_from_req_msg(struct ib_device *ibdev, struct ib_mad *mad) 146{ 147 struct cm_req_msg *msg = (struct cm_req_msg *)mad; 148 149 return msg->primary_path_sgid; 150} 151 152/* Lock should be taken before called */ 153static struct id_map_entry * 154id_map_find_by_sl_id(struct ib_device *ibdev, u32 slave_id, u32 sl_cm_id) 155{ 156 struct rb_root *sl_id_map = &to_mdev(ibdev)->sriov.sl_id_map; 157 struct rb_node *node = sl_id_map->rb_node; 158 159 while (node) { 160 struct id_map_entry *id_map_entry = 161 rb_entry(node, struct id_map_entry, node); 162 163 if (id_map_entry->sl_cm_id > sl_cm_id) 164 node = node->rb_left; 165 else if (id_map_entry->sl_cm_id < sl_cm_id) 166 node = node->rb_right; 167 else if (id_map_entry->slave_id > slave_id) 168 node = node->rb_left; 169 else if (id_map_entry->slave_id < slave_id) 170 node = node->rb_right; 171 else 172 return id_map_entry; 173 } 174 return NULL; 175} 176 177static void id_map_ent_timeout(struct work_struct *work) 178{ 179 struct delayed_work *delay = to_delayed_work(work); 180 struct id_map_entry *ent = container_of(delay, struct id_map_entry, timeout); 181 struct id_map_entry *found_ent; 182 struct mlx4_ib_dev *dev = ent->dev; 183 struct mlx4_ib_sriov *sriov = &dev->sriov; 184 struct rb_root *sl_id_map = &sriov->sl_id_map; 185 186 spin_lock(&sriov->id_map_lock); 187 if (!xa_erase(&sriov->pv_id_table, ent->pv_cm_id)) 188 goto out; 189 found_ent = id_map_find_by_sl_id(&dev->ib_dev, ent->slave_id, ent->sl_cm_id); 190 if (found_ent && found_ent == ent) 191 rb_erase(&found_ent->node, sl_id_map); 192 193out: 194 list_del(&ent->list); 195 spin_unlock(&sriov->id_map_lock); 196 kfree(ent); 197} 198 199static void sl_id_map_add(struct ib_device *ibdev, struct id_map_entry *new) 200{ 201 struct rb_root *sl_id_map = &to_mdev(ibdev)->sriov.sl_id_map; 202 struct rb_node **link = &sl_id_map->rb_node, *parent = NULL; 203 struct id_map_entry *ent; 204 int slave_id = new->slave_id; 205 int sl_cm_id = new->sl_cm_id; 206 207 ent = id_map_find_by_sl_id(ibdev, slave_id, sl_cm_id); 208 if (ent) { 209 pr_debug("overriding existing sl_id_map entry (cm_id = %x)\n", 210 sl_cm_id); 211 212 rb_replace_node(&ent->node, &new->node, sl_id_map); 213 return; 214 } 215 216 /* Go to the bottom of the tree */ 217 while (*link) { 218 parent = *link; 219 ent = rb_entry(parent, struct id_map_entry, node); 220 221 if (ent->sl_cm_id > sl_cm_id || (ent->sl_cm_id == sl_cm_id && ent->slave_id > slave_id)) 222 link = &(*link)->rb_left; 223 else 224 link = &(*link)->rb_right; 225 } 226 227 rb_link_node(&new->node, parent, link); 228 rb_insert_color(&new->node, sl_id_map); 229} 230 231static struct id_map_entry * 232id_map_alloc(struct ib_device *ibdev, int slave_id, u32 sl_cm_id) 233{ 234 int ret; 235 struct id_map_entry *ent; 236 struct mlx4_ib_sriov *sriov = &to_mdev(ibdev)->sriov; 237 238 ent = kmalloc(sizeof (struct id_map_entry), GFP_KERNEL); 239 if (!ent) 240 return ERR_PTR(-ENOMEM); 241 242 ent->sl_cm_id = sl_cm_id; 243 ent->slave_id = slave_id; 244 ent->scheduled_delete = 0; 245 ent->dev = to_mdev(ibdev); 246 INIT_DELAYED_WORK(&ent->timeout, id_map_ent_timeout); 247 248 ret = xa_alloc_cyclic(&sriov->pv_id_table, &ent->pv_cm_id, ent, 249 xa_limit_32b, &sriov->pv_id_next, GFP_KERNEL); 250 if (ret >= 0) { 251 spin_lock(&sriov->id_map_lock); 252 sl_id_map_add(ibdev, ent); 253 list_add_tail(&ent->list, &sriov->cm_list); 254 spin_unlock(&sriov->id_map_lock); 255 return ent; 256 } 257 258 /*error flow*/ 259 kfree(ent); 260 mlx4_ib_warn(ibdev, "Allocation failed (err:0x%x)\n", ret); 261 return ERR_PTR(-ENOMEM); 262} 263 264static struct id_map_entry * 265id_map_get(struct ib_device *ibdev, int *pv_cm_id, int slave_id, int sl_cm_id) 266{ 267 struct id_map_entry *ent; 268 struct mlx4_ib_sriov *sriov = &to_mdev(ibdev)->sriov; 269 270 spin_lock(&sriov->id_map_lock); 271 if (*pv_cm_id == -1) { 272 ent = id_map_find_by_sl_id(ibdev, slave_id, sl_cm_id); 273 if (ent) 274 *pv_cm_id = (int) ent->pv_cm_id; 275 } else 276 ent = xa_load(&sriov->pv_id_table, *pv_cm_id); 277 spin_unlock(&sriov->id_map_lock); 278 279 return ent; 280} 281 282static void schedule_delayed(struct ib_device *ibdev, struct id_map_entry *id) 283{ 284 struct mlx4_ib_sriov *sriov = &to_mdev(ibdev)->sriov; 285 unsigned long flags; 286 287 spin_lock(&sriov->id_map_lock); 288 spin_lock_irqsave(&sriov->going_down_lock, flags); 289 /*make sure that there is no schedule inside the scheduled work.*/ 290 if (!sriov->is_going_down && !id->scheduled_delete) { 291 id->scheduled_delete = 1; 292 queue_delayed_work(cm_wq, &id->timeout, CM_CLEANUP_CACHE_TIMEOUT); 293 } else if (id->scheduled_delete) { 294 /* Adjust timeout if already scheduled */ 295 mod_delayed_work(cm_wq, &id->timeout, CM_CLEANUP_CACHE_TIMEOUT); 296 } 297 spin_unlock_irqrestore(&sriov->going_down_lock, flags); 298 spin_unlock(&sriov->id_map_lock); 299} 300 301#define REJ_REASON(m) be16_to_cpu(((struct cm_generic_msg *)(m))->rej_reason) 302int mlx4_ib_multiplex_cm_handler(struct ib_device *ibdev, int port, int slave_id, 303 struct ib_mad *mad) 304{ 305 struct id_map_entry *id; 306 u32 sl_cm_id; 307 int pv_cm_id = -1; 308 309 if (mad->mad_hdr.attr_id == CM_REQ_ATTR_ID || 310 mad->mad_hdr.attr_id == CM_REP_ATTR_ID || 311 mad->mad_hdr.attr_id == CM_MRA_ATTR_ID || 312 mad->mad_hdr.attr_id == CM_SIDR_REQ_ATTR_ID || 313 (mad->mad_hdr.attr_id == CM_REJ_ATTR_ID && REJ_REASON(mad) == IB_CM_REJ_TIMEOUT)) { 314 sl_cm_id = get_local_comm_id(mad); 315 id = id_map_get(ibdev, &pv_cm_id, slave_id, sl_cm_id); 316 if (id) 317 goto cont; 318 id = id_map_alloc(ibdev, slave_id, sl_cm_id); 319 if (IS_ERR(id)) { 320 mlx4_ib_warn(ibdev, "%s: id{slave: %d, sl_cm_id: 0x%x} Failed to id_map_alloc\n", 321 __func__, slave_id, sl_cm_id); 322 return PTR_ERR(id); 323 } 324 } else if (mad->mad_hdr.attr_id == CM_REJ_ATTR_ID || 325 mad->mad_hdr.attr_id == CM_SIDR_REP_ATTR_ID) { 326 return 0; 327 } else { 328 sl_cm_id = get_local_comm_id(mad); 329 id = id_map_get(ibdev, &pv_cm_id, slave_id, sl_cm_id); 330 } 331 332 if (!id) { 333 pr_debug("id{slave: %d, sl_cm_id: 0x%x} is NULL! attr_id: 0x%x\n", 334 slave_id, sl_cm_id, be16_to_cpu(mad->mad_hdr.attr_id)); 335 return -EINVAL; 336 } 337 338cont: 339 set_local_comm_id(mad, id->pv_cm_id); 340 341 if (mad->mad_hdr.attr_id == CM_DREQ_ATTR_ID) 342 schedule_delayed(ibdev, id); 343 return 0; 344} 345 346static void rej_tmout_timeout(struct work_struct *work) 347{ 348 struct delayed_work *delay = to_delayed_work(work); 349 struct rej_tmout_entry *item = container_of(delay, struct rej_tmout_entry, timeout); 350 struct rej_tmout_entry *deleted; 351 352 deleted = xa_cmpxchg(item->xa_rej_tmout, item->rem_pv_cm_id, item, NULL, 0); 353 354 if (deleted != item) 355 pr_debug("deleted(%p) != item(%p)\n", deleted, item); 356 357 kfree(item); 358} 359 360static int alloc_rej_tmout(struct mlx4_ib_sriov *sriov, u32 rem_pv_cm_id, int slave) 361{ 362 struct rej_tmout_entry *item; 363 struct rej_tmout_entry *old; 364 int ret = 0; 365 366 xa_lock(&sriov->xa_rej_tmout); 367 item = xa_load(&sriov->xa_rej_tmout, (unsigned long)rem_pv_cm_id); 368 369 if (item) { 370 if (xa_err(item)) 371 ret = xa_err(item); 372 else 373 /* If a retry, adjust delayed work */ 374 mod_delayed_work(cm_wq, &item->timeout, CM_CLEANUP_CACHE_TIMEOUT); 375 goto err_or_exists; 376 } 377 xa_unlock(&sriov->xa_rej_tmout); 378 379 item = kmalloc(sizeof(*item), GFP_KERNEL); 380 if (!item) 381 return -ENOMEM; 382 383 INIT_DELAYED_WORK(&item->timeout, rej_tmout_timeout); 384 item->slave = slave; 385 item->rem_pv_cm_id = rem_pv_cm_id; 386 item->xa_rej_tmout = &sriov->xa_rej_tmout; 387 388 old = xa_cmpxchg(&sriov->xa_rej_tmout, (unsigned long)rem_pv_cm_id, NULL, item, GFP_KERNEL); 389 if (old) { 390 pr_debug( 391 "Non-null old entry (%p) or error (%d) when inserting\n", 392 old, xa_err(old)); 393 kfree(item); 394 return xa_err(old); 395 } 396 397 queue_delayed_work(cm_wq, &item->timeout, CM_CLEANUP_CACHE_TIMEOUT); 398 399 return 0; 400 401err_or_exists: 402 xa_unlock(&sriov->xa_rej_tmout); 403 return ret; 404} 405 406static int lookup_rej_tmout_slave(struct mlx4_ib_sriov *sriov, u32 rem_pv_cm_id) 407{ 408 struct rej_tmout_entry *item; 409 int slave; 410 411 xa_lock(&sriov->xa_rej_tmout); 412 item = xa_load(&sriov->xa_rej_tmout, (unsigned long)rem_pv_cm_id); 413 414 if (!item || xa_err(item)) { 415 pr_debug("Could not find slave. rem_pv_cm_id 0x%x error: %d\n", 416 rem_pv_cm_id, xa_err(item)); 417 slave = !item ? -ENOENT : xa_err(item); 418 } else { 419 slave = item->slave; 420 } 421 xa_unlock(&sriov->xa_rej_tmout); 422 423 return slave; 424} 425 426int mlx4_ib_demux_cm_handler(struct ib_device *ibdev, int port, int *slave, 427 struct ib_mad *mad) 428{ 429 struct mlx4_ib_sriov *sriov = &to_mdev(ibdev)->sriov; 430 u32 rem_pv_cm_id = get_local_comm_id(mad); 431 u32 pv_cm_id; 432 struct id_map_entry *id; 433 int sts; 434 435 if (mad->mad_hdr.attr_id == CM_REQ_ATTR_ID || 436 mad->mad_hdr.attr_id == CM_SIDR_REQ_ATTR_ID) { 437 union ib_gid gid; 438 439 if (!slave) 440 return 0; 441 442 gid = gid_from_req_msg(ibdev, mad); 443 *slave = mlx4_ib_find_real_gid(ibdev, port, gid.global.interface_id); 444 if (*slave < 0) { 445 mlx4_ib_warn(ibdev, "failed matching slave_id by gid (0x%llx)\n", 446 be64_to_cpu(gid.global.interface_id)); 447 return -ENOENT; 448 } 449 450 sts = alloc_rej_tmout(sriov, rem_pv_cm_id, *slave); 451 if (sts) 452 /* Even if this fails, we pass on the REQ to the slave */ 453 pr_debug("Could not allocate rej_tmout entry. rem_pv_cm_id 0x%x slave %d status %d\n", 454 rem_pv_cm_id, *slave, sts); 455 456 return 0; 457 } 458 459 pv_cm_id = get_remote_comm_id(mad); 460 id = id_map_get(ibdev, (int *)&pv_cm_id, -1, -1); 461 462 if (!id) { 463 if (mad->mad_hdr.attr_id == CM_REJ_ATTR_ID && 464 REJ_REASON(mad) == IB_CM_REJ_TIMEOUT && slave) { 465 *slave = lookup_rej_tmout_slave(sriov, rem_pv_cm_id); 466 467 return (*slave < 0) ? *slave : 0; 468 } 469 pr_debug("Couldn't find an entry for pv_cm_id 0x%x, attr_id 0x%x\n", 470 pv_cm_id, be16_to_cpu(mad->mad_hdr.attr_id)); 471 return -ENOENT; 472 } 473 474 if (slave) 475 *slave = id->slave_id; 476 set_remote_comm_id(mad, id->sl_cm_id); 477 478 if (mad->mad_hdr.attr_id == CM_DREQ_ATTR_ID || 479 mad->mad_hdr.attr_id == CM_REJ_ATTR_ID) 480 schedule_delayed(ibdev, id); 481 482 return 0; 483} 484 485void mlx4_ib_cm_paravirt_init(struct mlx4_ib_dev *dev) 486{ 487 spin_lock_init(&dev->sriov.id_map_lock); 488 INIT_LIST_HEAD(&dev->sriov.cm_list); 489 dev->sriov.sl_id_map = RB_ROOT; 490 xa_init_flags(&dev->sriov.pv_id_table, XA_FLAGS_ALLOC); 491 xa_init(&dev->sriov.xa_rej_tmout); 492} 493 494static void rej_tmout_xa_cleanup(struct mlx4_ib_sriov *sriov, int slave) 495{ 496 struct rej_tmout_entry *item; 497 bool flush_needed = false; 498 unsigned long id; 499 int cnt = 0; 500 501 xa_lock(&sriov->xa_rej_tmout); 502 xa_for_each(&sriov->xa_rej_tmout, id, item) { 503 if (slave < 0 || slave == item->slave) { 504 mod_delayed_work(cm_wq, &item->timeout, 0); 505 flush_needed = true; 506 ++cnt; 507 } 508 } 509 xa_unlock(&sriov->xa_rej_tmout); 510 511 if (flush_needed) { 512 flush_workqueue(cm_wq); 513 pr_debug("Deleted %d entries in xarray for slave %d during cleanup\n", 514 cnt, slave); 515 } 516 517 if (slave < 0) 518 WARN_ON(!xa_empty(&sriov->xa_rej_tmout)); 519} 520 521/* slave = -1 ==> all slaves */ 522/* TBD -- call paravirt clean for single slave. Need for slave RESET event */ 523void mlx4_ib_cm_paravirt_clean(struct mlx4_ib_dev *dev, int slave) 524{ 525 struct mlx4_ib_sriov *sriov = &dev->sriov; 526 struct rb_root *sl_id_map = &sriov->sl_id_map; 527 struct list_head lh; 528 struct rb_node *nd; 529 int need_flush = 0; 530 struct id_map_entry *map, *tmp_map; 531 /* cancel all delayed work queue entries */ 532 INIT_LIST_HEAD(&lh); 533 spin_lock(&sriov->id_map_lock); 534 list_for_each_entry_safe(map, tmp_map, &dev->sriov.cm_list, list) { 535 if (slave < 0 || slave == map->slave_id) { 536 if (map->scheduled_delete) 537 need_flush |= !cancel_delayed_work(&map->timeout); 538 } 539 } 540 541 spin_unlock(&sriov->id_map_lock); 542 543 if (need_flush) 544 flush_workqueue(cm_wq); /* make sure all timers were flushed */ 545 546 /* now, remove all leftover entries from databases*/ 547 spin_lock(&sriov->id_map_lock); 548 if (slave < 0) { 549 while (rb_first(sl_id_map)) { 550 struct id_map_entry *ent = 551 rb_entry(rb_first(sl_id_map), 552 struct id_map_entry, node); 553 554 rb_erase(&ent->node, sl_id_map); 555 xa_erase(&sriov->pv_id_table, ent->pv_cm_id); 556 } 557 list_splice_init(&dev->sriov.cm_list, &lh); 558 } else { 559 /* first, move nodes belonging to slave to db remove list */ 560 nd = rb_first(sl_id_map); 561 while (nd) { 562 struct id_map_entry *ent = 563 rb_entry(nd, struct id_map_entry, node); 564 nd = rb_next(nd); 565 if (ent->slave_id == slave) 566 list_move_tail(&ent->list, &lh); 567 } 568 /* remove those nodes from databases */ 569 list_for_each_entry_safe(map, tmp_map, &lh, list) { 570 rb_erase(&map->node, sl_id_map); 571 xa_erase(&sriov->pv_id_table, map->pv_cm_id); 572 } 573 574 /* add remaining nodes from cm_list */ 575 list_for_each_entry_safe(map, tmp_map, &dev->sriov.cm_list, list) { 576 if (slave == map->slave_id) 577 list_move_tail(&map->list, &lh); 578 } 579 } 580 581 spin_unlock(&sriov->id_map_lock); 582 583 /* free any map entries left behind due to cancel_delayed_work above */ 584 list_for_each_entry_safe(map, tmp_map, &lh, list) { 585 list_del(&map->list); 586 kfree(map); 587 } 588 589 rej_tmout_xa_cleanup(sriov, slave); 590} 591 592int mlx4_ib_cm_init(void) 593{ 594 cm_wq = alloc_workqueue("mlx4_ib_cm", 0, 0); 595 if (!cm_wq) 596 return -ENOMEM; 597 598 return 0; 599} 600 601void mlx4_ib_cm_destroy(void) 602{ 603 destroy_workqueue(cm_wq); 604} 605