1 /* 2 * Block device elevator/IO-scheduler. 3 * 4 * Copyright (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE 5 * 6 * 30042000 Jens Axboe <axboe@kernel.dk> : 7 * 8 * Split the elevator a bit so that it is possible to choose a different 9 * one or even write a new "plug in". There are three pieces: 10 * - elevator_fn, inserts a new request in the queue list 11 * - elevator_merge_fn, decides whether a new buffer can be merged with 12 * an existing request 13 * - elevator_dequeue_fn, called when a request is taken off the active list 14 * 15 * 20082000 Dave Jones <davej@suse.de> : 16 * Removed tests for max-bomb-segments, which was breaking elvtune 17 * when run without -bN 18 * 19 * Jens: 20 * - Rework again to work with bio instead of buffer_heads 21 * - loose bi_dev comparisons, partition handling is right now 22 * - completely modularize elevator setup and teardown 23 * 24 */ 25 #include <linux/kernel.h> 26 #include <linux/fs.h> 27 #include <linux/blkdev.h> 28 #include <linux/elevator.h> 29 #include <linux/bio.h> 30 #include <linux/module.h> 31 #include <linux/slab.h> 32 #include <linux/init.h> 33 #include <linux/compiler.h> 34 #include <linux/blktrace_api.h> 35 #include <linux/hash.h> 36 #include <linux/uaccess.h> 37 #include <linux/pm_runtime.h> 38 #include <linux/blk-cgroup.h> 39 40 #include <trace/events/block.h> 41 42 #include "blk.h" 43 #include "blk-mq-sched.h" 44 #include "blk-wbt.h" 45 46 static DEFINE_SPINLOCK(elv_list_lock); 47 static LIST_HEAD(elv_list); 48 49 /* 50 * Merge hash stuff. 51 */ 52 #define rq_hash_key(rq) (blk_rq_pos(rq) + blk_rq_sectors(rq)) 53 54 /* 55 * Query io scheduler to see if the current process issuing bio may be 56 * merged with rq. 57 */ 58 static int elv_iosched_allow_bio_merge(struct request *rq, struct bio *bio) 59 { 60 struct request_queue *q = rq->q; 61 struct elevator_queue *e = q->elevator; 62 63 if (e->uses_mq && e->type->ops.mq.allow_merge) 64 return e->type->ops.mq.allow_merge(q, rq, bio); 65 else if (!e->uses_mq && e->type->ops.sq.elevator_allow_bio_merge_fn) 66 return e->type->ops.sq.elevator_allow_bio_merge_fn(q, rq, bio); 67 68 return 1; 69 } 70 71 /* 72 * can we safely merge with this request? 73 */ 74 bool elv_bio_merge_ok(struct request *rq, struct bio *bio) 75 { 76 if (!blk_rq_merge_ok(rq, bio)) 77 return false; 78 79 if (!elv_iosched_allow_bio_merge(rq, bio)) 80 return false; 81 82 return true; 83 } 84 EXPORT_SYMBOL(elv_bio_merge_ok); 85 86 static bool elevator_match(const struct elevator_type *e, const char *name) 87 { 88 if (!strcmp(e->elevator_name, name)) 89 return true; 90 if (e->elevator_alias && !strcmp(e->elevator_alias, name)) 91 return true; 92 93 return false; 94 } 95 96 /* 97 * Return scheduler with name 'name' and with matching 'mq capability 98 */ 99 static struct elevator_type *elevator_find(const char *name, bool mq) 100 { 101 struct elevator_type *e; 102 103 list_for_each_entry(e, &elv_list, list) { 104 if (elevator_match(e, name) && (mq == e->uses_mq)) 105 return e; 106 } 107 108 return NULL; 109 } 110 111 static void elevator_put(struct elevator_type *e) 112 { 113 module_put(e->elevator_owner); 114 } 115 116 static struct elevator_type *elevator_get(struct request_queue *q, 117 const char *name, bool try_loading) 118 { 119 struct elevator_type *e; 120 121 spin_lock(&elv_list_lock); 122 123 e = elevator_find(name, q->mq_ops != NULL); 124 if (!e && try_loading) { 125 spin_unlock(&elv_list_lock); 126 request_module("%s-iosched", name); 127 spin_lock(&elv_list_lock); 128 e = elevator_find(name, q->mq_ops != NULL); 129 } 130 131 if (e && !try_module_get(e->elevator_owner)) 132 e = NULL; 133 134 spin_unlock(&elv_list_lock); 135 return e; 136 } 137 138 static char chosen_elevator[ELV_NAME_MAX]; 139 140 static int __init elevator_setup(char *str) 141 { 142 /* 143 * Be backwards-compatible with previous kernels, so users 144 * won't get the wrong elevator. 145 */ 146 strncpy(chosen_elevator, str, sizeof(chosen_elevator) - 1); 147 return 1; 148 } 149 150 __setup("elevator=", elevator_setup); 151 152 /* called during boot to load the elevator chosen by the elevator param */ 153 void __init load_default_elevator_module(void) 154 { 155 struct elevator_type *e; 156 157 if (!chosen_elevator[0]) 158 return; 159 160 /* 161 * Boot parameter is deprecated, we haven't supported that for MQ. 162 * Only look for non-mq schedulers from here. 163 */ 164 spin_lock(&elv_list_lock); 165 e = elevator_find(chosen_elevator, false); 166 spin_unlock(&elv_list_lock); 167 168 if (!e) 169 request_module("%s-iosched", chosen_elevator); 170 } 171 172 static struct kobj_type elv_ktype; 173 174 struct elevator_queue *elevator_alloc(struct request_queue *q, 175 struct elevator_type *e) 176 { 177 struct elevator_queue *eq; 178 179 eq = kzalloc_node(sizeof(*eq), GFP_KERNEL, q->node); 180 if (unlikely(!eq)) 181 return NULL; 182 183 eq->type = e; 184 kobject_init(&eq->kobj, &elv_ktype); 185 mutex_init(&eq->sysfs_lock); 186 hash_init(eq->hash); 187 eq->uses_mq = e->uses_mq; 188 189 return eq; 190 } 191 EXPORT_SYMBOL(elevator_alloc); 192 193 static void elevator_release(struct kobject *kobj) 194 { 195 struct elevator_queue *e; 196 197 e = container_of(kobj, struct elevator_queue, kobj); 198 elevator_put(e->type); 199 kfree(e); 200 } 201 202 /* 203 * Use the default elevator specified by config boot param for non-mq devices, 204 * or by config option. Don't try to load modules as we could be running off 205 * async and request_module() isn't allowed from async. 206 */ 207 int elevator_init(struct request_queue *q) 208 { 209 struct elevator_type *e = NULL; 210 int err = 0; 211 212 /* 213 * q->sysfs_lock must be held to provide mutual exclusion between 214 * elevator_switch() and here. 215 */ 216 mutex_lock(&q->sysfs_lock); 217 if (unlikely(q->elevator)) 218 goto out_unlock; 219 220 if (*chosen_elevator) { 221 e = elevator_get(q, chosen_elevator, false); 222 if (!e) 223 printk(KERN_ERR "I/O scheduler %s not found\n", 224 chosen_elevator); 225 } 226 227 if (!e) 228 e = elevator_get(q, CONFIG_DEFAULT_IOSCHED, false); 229 if (!e) { 230 printk(KERN_ERR 231 "Default I/O scheduler not found. Using noop.\n"); 232 e = elevator_get(q, "noop", false); 233 } 234 235 err = e->ops.sq.elevator_init_fn(q, e); 236 if (err) 237 elevator_put(e); 238 out_unlock: 239 mutex_unlock(&q->sysfs_lock); 240 return err; 241 } 242 243 void elevator_exit(struct request_queue *q, struct elevator_queue *e) 244 { 245 mutex_lock(&e->sysfs_lock); 246 if (e->uses_mq && e->type->ops.mq.exit_sched) 247 blk_mq_exit_sched(q, e); 248 else if (!e->uses_mq && e->type->ops.sq.elevator_exit_fn) 249 e->type->ops.sq.elevator_exit_fn(e); 250 mutex_unlock(&e->sysfs_lock); 251 252 kobject_put(&e->kobj); 253 } 254 255 static inline void __elv_rqhash_del(struct request *rq) 256 { 257 hash_del(&rq->hash); 258 rq->rq_flags &= ~RQF_HASHED; 259 } 260 261 void elv_rqhash_del(struct request_queue *q, struct request *rq) 262 { 263 if (ELV_ON_HASH(rq)) 264 __elv_rqhash_del(rq); 265 } 266 EXPORT_SYMBOL_GPL(elv_rqhash_del); 267 268 void elv_rqhash_add(struct request_queue *q, struct request *rq) 269 { 270 struct elevator_queue *e = q->elevator; 271 272 BUG_ON(ELV_ON_HASH(rq)); 273 hash_add(e->hash, &rq->hash, rq_hash_key(rq)); 274 rq->rq_flags |= RQF_HASHED; 275 } 276 EXPORT_SYMBOL_GPL(elv_rqhash_add); 277 278 void elv_rqhash_reposition(struct request_queue *q, struct request *rq) 279 { 280 __elv_rqhash_del(rq); 281 elv_rqhash_add(q, rq); 282 } 283 284 struct request *elv_rqhash_find(struct request_queue *q, sector_t offset) 285 { 286 struct elevator_queue *e = q->elevator; 287 struct hlist_node *next; 288 struct request *rq; 289 290 hash_for_each_possible_safe(e->hash, rq, next, hash, offset) { 291 BUG_ON(!ELV_ON_HASH(rq)); 292 293 if (unlikely(!rq_mergeable(rq))) { 294 __elv_rqhash_del(rq); 295 continue; 296 } 297 298 if (rq_hash_key(rq) == offset) 299 return rq; 300 } 301 302 return NULL; 303 } 304 305 /* 306 * RB-tree support functions for inserting/lookup/removal of requests 307 * in a sorted RB tree. 308 */ 309 void elv_rb_add(struct rb_root *root, struct request *rq) 310 { 311 struct rb_node **p = &root->rb_node; 312 struct rb_node *parent = NULL; 313 struct request *__rq; 314 315 while (*p) { 316 parent = *p; 317 __rq = rb_entry(parent, struct request, rb_node); 318 319 if (blk_rq_pos(rq) < blk_rq_pos(__rq)) 320 p = &(*p)->rb_left; 321 else if (blk_rq_pos(rq) >= blk_rq_pos(__rq)) 322 p = &(*p)->rb_right; 323 } 324 325 rb_link_node(&rq->rb_node, parent, p); 326 rb_insert_color(&rq->rb_node, root); 327 } 328 EXPORT_SYMBOL(elv_rb_add); 329 330 void elv_rb_del(struct rb_root *root, struct request *rq) 331 { 332 BUG_ON(RB_EMPTY_NODE(&rq->rb_node)); 333 rb_erase(&rq->rb_node, root); 334 RB_CLEAR_NODE(&rq->rb_node); 335 } 336 EXPORT_SYMBOL(elv_rb_del); 337 338 struct request *elv_rb_find(struct rb_root *root, sector_t sector) 339 { 340 struct rb_node *n = root->rb_node; 341 struct request *rq; 342 343 while (n) { 344 rq = rb_entry(n, struct request, rb_node); 345 346 if (sector < blk_rq_pos(rq)) 347 n = n->rb_left; 348 else if (sector > blk_rq_pos(rq)) 349 n = n->rb_right; 350 else 351 return rq; 352 } 353 354 return NULL; 355 } 356 EXPORT_SYMBOL(elv_rb_find); 357 358 /* 359 * Insert rq into dispatch queue of q. Queue lock must be held on 360 * entry. rq is sort instead into the dispatch queue. To be used by 361 * specific elevators. 362 */ 363 void elv_dispatch_sort(struct request_queue *q, struct request *rq) 364 { 365 sector_t boundary; 366 struct list_head *entry; 367 368 if (q->last_merge == rq) 369 q->last_merge = NULL; 370 371 elv_rqhash_del(q, rq); 372 373 q->nr_sorted--; 374 375 boundary = q->end_sector; 376 list_for_each_prev(entry, &q->queue_head) { 377 struct request *pos = list_entry_rq(entry); 378 379 if (req_op(rq) != req_op(pos)) 380 break; 381 if (rq_data_dir(rq) != rq_data_dir(pos)) 382 break; 383 if (pos->rq_flags & (RQF_STARTED | RQF_SOFTBARRIER)) 384 break; 385 if (blk_rq_pos(rq) >= boundary) { 386 if (blk_rq_pos(pos) < boundary) 387 continue; 388 } else { 389 if (blk_rq_pos(pos) >= boundary) 390 break; 391 } 392 if (blk_rq_pos(rq) >= blk_rq_pos(pos)) 393 break; 394 } 395 396 list_add(&rq->queuelist, entry); 397 } 398 EXPORT_SYMBOL(elv_dispatch_sort); 399 400 /* 401 * Insert rq into dispatch queue of q. Queue lock must be held on 402 * entry. rq is added to the back of the dispatch queue. To be used by 403 * specific elevators. 404 */ 405 void elv_dispatch_add_tail(struct request_queue *q, struct request *rq) 406 { 407 if (q->last_merge == rq) 408 q->last_merge = NULL; 409 410 elv_rqhash_del(q, rq); 411 412 q->nr_sorted--; 413 414 q->end_sector = rq_end_sector(rq); 415 q->boundary_rq = rq; 416 list_add_tail(&rq->queuelist, &q->queue_head); 417 } 418 EXPORT_SYMBOL(elv_dispatch_add_tail); 419 420 enum elv_merge elv_merge(struct request_queue *q, struct request **req, 421 struct bio *bio) 422 { 423 struct elevator_queue *e = q->elevator; 424 struct request *__rq; 425 426 /* 427 * Levels of merges: 428 * nomerges: No merges at all attempted 429 * noxmerges: Only simple one-hit cache try 430 * merges: All merge tries attempted 431 */ 432 if (blk_queue_nomerges(q) || !bio_mergeable(bio)) 433 return ELEVATOR_NO_MERGE; 434 435 /* 436 * First try one-hit cache. 437 */ 438 if (q->last_merge && elv_bio_merge_ok(q->last_merge, bio)) { 439 enum elv_merge ret = blk_try_merge(q->last_merge, bio); 440 441 if (ret != ELEVATOR_NO_MERGE) { 442 *req = q->last_merge; 443 return ret; 444 } 445 } 446 447 if (blk_queue_noxmerges(q)) 448 return ELEVATOR_NO_MERGE; 449 450 /* 451 * See if our hash lookup can find a potential backmerge. 452 */ 453 __rq = elv_rqhash_find(q, bio->bi_iter.bi_sector); 454 if (__rq && elv_bio_merge_ok(__rq, bio)) { 455 *req = __rq; 456 return ELEVATOR_BACK_MERGE; 457 } 458 459 if (e->uses_mq && e->type->ops.mq.request_merge) 460 return e->type->ops.mq.request_merge(q, req, bio); 461 else if (!e->uses_mq && e->type->ops.sq.elevator_merge_fn) 462 return e->type->ops.sq.elevator_merge_fn(q, req, bio); 463 464 return ELEVATOR_NO_MERGE; 465 } 466 467 /* 468 * Attempt to do an insertion back merge. Only check for the case where 469 * we can append 'rq' to an existing request, so we can throw 'rq' away 470 * afterwards. 471 * 472 * Returns true if we merged, false otherwise 473 */ 474 bool elv_attempt_insert_merge(struct request_queue *q, struct request *rq) 475 { 476 struct request *__rq; 477 bool ret; 478 479 if (blk_queue_nomerges(q)) 480 return false; 481 482 /* 483 * First try one-hit cache. 484 */ 485 if (q->last_merge && blk_attempt_req_merge(q, q->last_merge, rq)) 486 return true; 487 488 if (blk_queue_noxmerges(q)) 489 return false; 490 491 ret = false; 492 /* 493 * See if our hash lookup can find a potential backmerge. 494 */ 495 while (1) { 496 __rq = elv_rqhash_find(q, blk_rq_pos(rq)); 497 if (!__rq || !blk_attempt_req_merge(q, __rq, rq)) 498 break; 499 500 /* The merged request could be merged with others, try again */ 501 ret = true; 502 rq = __rq; 503 } 504 505 return ret; 506 } 507 508 void elv_merged_request(struct request_queue *q, struct request *rq, 509 enum elv_merge type) 510 { 511 struct elevator_queue *e = q->elevator; 512 513 if (e->uses_mq && e->type->ops.mq.request_merged) 514 e->type->ops.mq.request_merged(q, rq, type); 515 else if (!e->uses_mq && e->type->ops.sq.elevator_merged_fn) 516 e->type->ops.sq.elevator_merged_fn(q, rq, type); 517 518 if (type == ELEVATOR_BACK_MERGE) 519 elv_rqhash_reposition(q, rq); 520 521 q->last_merge = rq; 522 } 523 524 void elv_merge_requests(struct request_queue *q, struct request *rq, 525 struct request *next) 526 { 527 struct elevator_queue *e = q->elevator; 528 bool next_sorted = false; 529 530 if (e->uses_mq && e->type->ops.mq.requests_merged) 531 e->type->ops.mq.requests_merged(q, rq, next); 532 else if (e->type->ops.sq.elevator_merge_req_fn) { 533 next_sorted = (__force bool)(next->rq_flags & RQF_SORTED); 534 if (next_sorted) 535 e->type->ops.sq.elevator_merge_req_fn(q, rq, next); 536 } 537 538 elv_rqhash_reposition(q, rq); 539 540 if (next_sorted) { 541 elv_rqhash_del(q, next); 542 q->nr_sorted--; 543 } 544 545 q->last_merge = rq; 546 } 547 548 void elv_bio_merged(struct request_queue *q, struct request *rq, 549 struct bio *bio) 550 { 551 struct elevator_queue *e = q->elevator; 552 553 if (WARN_ON_ONCE(e->uses_mq)) 554 return; 555 556 if (e->type->ops.sq.elevator_bio_merged_fn) 557 e->type->ops.sq.elevator_bio_merged_fn(q, rq, bio); 558 } 559 560 #ifdef CONFIG_PM 561 static void blk_pm_requeue_request(struct request *rq) 562 { 563 if (rq->q->dev && !(rq->rq_flags & RQF_PM)) 564 rq->q->nr_pending--; 565 } 566 567 static void blk_pm_add_request(struct request_queue *q, struct request *rq) 568 { 569 if (q->dev && !(rq->rq_flags & RQF_PM) && q->nr_pending++ == 0 && 570 (q->rpm_status == RPM_SUSPENDED || q->rpm_status == RPM_SUSPENDING)) 571 pm_request_resume(q->dev); 572 } 573 #else 574 static inline void blk_pm_requeue_request(struct request *rq) {} 575 static inline void blk_pm_add_request(struct request_queue *q, 576 struct request *rq) 577 { 578 } 579 #endif 580 581 void elv_requeue_request(struct request_queue *q, struct request *rq) 582 { 583 /* 584 * it already went through dequeue, we need to decrement the 585 * in_flight count again 586 */ 587 if (blk_account_rq(rq)) { 588 q->in_flight[rq_is_sync(rq)]--; 589 if (rq->rq_flags & RQF_SORTED) 590 elv_deactivate_rq(q, rq); 591 } 592 593 rq->rq_flags &= ~RQF_STARTED; 594 595 blk_pm_requeue_request(rq); 596 597 __elv_add_request(q, rq, ELEVATOR_INSERT_REQUEUE); 598 } 599 600 void elv_drain_elevator(struct request_queue *q) 601 { 602 struct elevator_queue *e = q->elevator; 603 static int printed; 604 605 if (WARN_ON_ONCE(e->uses_mq)) 606 return; 607 608 lockdep_assert_held(q->queue_lock); 609 610 while (e->type->ops.sq.elevator_dispatch_fn(q, 1)) 611 ; 612 if (q->nr_sorted && printed++ < 10) { 613 printk(KERN_ERR "%s: forced dispatching is broken " 614 "(nr_sorted=%u), please report this\n", 615 q->elevator->type->elevator_name, q->nr_sorted); 616 } 617 } 618 619 void __elv_add_request(struct request_queue *q, struct request *rq, int where) 620 { 621 trace_block_rq_insert(q, rq); 622 623 blk_pm_add_request(q, rq); 624 625 rq->q = q; 626 627 if (rq->rq_flags & RQF_SOFTBARRIER) { 628 /* barriers are scheduling boundary, update end_sector */ 629 if (!blk_rq_is_passthrough(rq)) { 630 q->end_sector = rq_end_sector(rq); 631 q->boundary_rq = rq; 632 } 633 } else if (!(rq->rq_flags & RQF_ELVPRIV) && 634 (where == ELEVATOR_INSERT_SORT || 635 where == ELEVATOR_INSERT_SORT_MERGE)) 636 where = ELEVATOR_INSERT_BACK; 637 638 switch (where) { 639 case ELEVATOR_INSERT_REQUEUE: 640 case ELEVATOR_INSERT_FRONT: 641 rq->rq_flags |= RQF_SOFTBARRIER; 642 list_add(&rq->queuelist, &q->queue_head); 643 break; 644 645 case ELEVATOR_INSERT_BACK: 646 rq->rq_flags |= RQF_SOFTBARRIER; 647 elv_drain_elevator(q); 648 list_add_tail(&rq->queuelist, &q->queue_head); 649 /* 650 * We kick the queue here for the following reasons. 651 * - The elevator might have returned NULL previously 652 * to delay requests and returned them now. As the 653 * queue wasn't empty before this request, ll_rw_blk 654 * won't run the queue on return, resulting in hang. 655 * - Usually, back inserted requests won't be merged 656 * with anything. There's no point in delaying queue 657 * processing. 658 */ 659 __blk_run_queue(q); 660 break; 661 662 case ELEVATOR_INSERT_SORT_MERGE: 663 /* 664 * If we succeed in merging this request with one in the 665 * queue already, we are done - rq has now been freed, 666 * so no need to do anything further. 667 */ 668 if (elv_attempt_insert_merge(q, rq)) 669 break; 670 /* fall through */ 671 case ELEVATOR_INSERT_SORT: 672 BUG_ON(blk_rq_is_passthrough(rq)); 673 rq->rq_flags |= RQF_SORTED; 674 q->nr_sorted++; 675 if (rq_mergeable(rq)) { 676 elv_rqhash_add(q, rq); 677 if (!q->last_merge) 678 q->last_merge = rq; 679 } 680 681 /* 682 * Some ioscheds (cfq) run q->request_fn directly, so 683 * rq cannot be accessed after calling 684 * elevator_add_req_fn. 685 */ 686 q->elevator->type->ops.sq.elevator_add_req_fn(q, rq); 687 break; 688 689 case ELEVATOR_INSERT_FLUSH: 690 rq->rq_flags |= RQF_SOFTBARRIER; 691 blk_insert_flush(rq); 692 break; 693 default: 694 printk(KERN_ERR "%s: bad insertion point %d\n", 695 __func__, where); 696 BUG(); 697 } 698 } 699 EXPORT_SYMBOL(__elv_add_request); 700 701 void elv_add_request(struct request_queue *q, struct request *rq, int where) 702 { 703 unsigned long flags; 704 705 spin_lock_irqsave(q->queue_lock, flags); 706 __elv_add_request(q, rq, where); 707 spin_unlock_irqrestore(q->queue_lock, flags); 708 } 709 EXPORT_SYMBOL(elv_add_request); 710 711 struct request *elv_latter_request(struct request_queue *q, struct request *rq) 712 { 713 struct elevator_queue *e = q->elevator; 714 715 if (e->uses_mq && e->type->ops.mq.next_request) 716 return e->type->ops.mq.next_request(q, rq); 717 else if (!e->uses_mq && e->type->ops.sq.elevator_latter_req_fn) 718 return e->type->ops.sq.elevator_latter_req_fn(q, rq); 719 720 return NULL; 721 } 722 723 struct request *elv_former_request(struct request_queue *q, struct request *rq) 724 { 725 struct elevator_queue *e = q->elevator; 726 727 if (e->uses_mq && e->type->ops.mq.former_request) 728 return e->type->ops.mq.former_request(q, rq); 729 if (!e->uses_mq && e->type->ops.sq.elevator_former_req_fn) 730 return e->type->ops.sq.elevator_former_req_fn(q, rq); 731 return NULL; 732 } 733 734 int elv_set_request(struct request_queue *q, struct request *rq, 735 struct bio *bio, gfp_t gfp_mask) 736 { 737 struct elevator_queue *e = q->elevator; 738 739 if (WARN_ON_ONCE(e->uses_mq)) 740 return 0; 741 742 if (e->type->ops.sq.elevator_set_req_fn) 743 return e->type->ops.sq.elevator_set_req_fn(q, rq, bio, gfp_mask); 744 return 0; 745 } 746 747 void elv_put_request(struct request_queue *q, struct request *rq) 748 { 749 struct elevator_queue *e = q->elevator; 750 751 if (WARN_ON_ONCE(e->uses_mq)) 752 return; 753 754 if (e->type->ops.sq.elevator_put_req_fn) 755 e->type->ops.sq.elevator_put_req_fn(rq); 756 } 757 758 int elv_may_queue(struct request_queue *q, unsigned int op) 759 { 760 struct elevator_queue *e = q->elevator; 761 762 if (WARN_ON_ONCE(e->uses_mq)) 763 return 0; 764 765 if (e->type->ops.sq.elevator_may_queue_fn) 766 return e->type->ops.sq.elevator_may_queue_fn(q, op); 767 768 return ELV_MQUEUE_MAY; 769 } 770 771 void elv_completed_request(struct request_queue *q, struct request *rq) 772 { 773 struct elevator_queue *e = q->elevator; 774 775 if (WARN_ON_ONCE(e->uses_mq)) 776 return; 777 778 /* 779 * request is released from the driver, io must be done 780 */ 781 if (blk_account_rq(rq)) { 782 q->in_flight[rq_is_sync(rq)]--; 783 if ((rq->rq_flags & RQF_SORTED) && 784 e->type->ops.sq.elevator_completed_req_fn) 785 e->type->ops.sq.elevator_completed_req_fn(q, rq); 786 } 787 } 788 789 #define to_elv(atr) container_of((atr), struct elv_fs_entry, attr) 790 791 static ssize_t 792 elv_attr_show(struct kobject *kobj, struct attribute *attr, char *page) 793 { 794 struct elv_fs_entry *entry = to_elv(attr); 795 struct elevator_queue *e; 796 ssize_t error; 797 798 if (!entry->show) 799 return -EIO; 800 801 e = container_of(kobj, struct elevator_queue, kobj); 802 mutex_lock(&e->sysfs_lock); 803 error = e->type ? entry->show(e, page) : -ENOENT; 804 mutex_unlock(&e->sysfs_lock); 805 return error; 806 } 807 808 static ssize_t 809 elv_attr_store(struct kobject *kobj, struct attribute *attr, 810 const char *page, size_t length) 811 { 812 struct elv_fs_entry *entry = to_elv(attr); 813 struct elevator_queue *e; 814 ssize_t error; 815 816 if (!entry->store) 817 return -EIO; 818 819 e = container_of(kobj, struct elevator_queue, kobj); 820 mutex_lock(&e->sysfs_lock); 821 error = e->type ? entry->store(e, page, length) : -ENOENT; 822 mutex_unlock(&e->sysfs_lock); 823 return error; 824 } 825 826 static const struct sysfs_ops elv_sysfs_ops = { 827 .show = elv_attr_show, 828 .store = elv_attr_store, 829 }; 830 831 static struct kobj_type elv_ktype = { 832 .sysfs_ops = &elv_sysfs_ops, 833 .release = elevator_release, 834 }; 835 836 int elv_register_queue(struct request_queue *q) 837 { 838 struct elevator_queue *e = q->elevator; 839 int error; 840 841 lockdep_assert_held(&q->sysfs_lock); 842 843 error = kobject_add(&e->kobj, &q->kobj, "%s", "iosched"); 844 if (!error) { 845 struct elv_fs_entry *attr = e->type->elevator_attrs; 846 if (attr) { 847 while (attr->attr.name) { 848 if (sysfs_create_file(&e->kobj, &attr->attr)) 849 break; 850 attr++; 851 } 852 } 853 kobject_uevent(&e->kobj, KOBJ_ADD); 854 e->registered = 1; 855 if (!e->uses_mq && e->type->ops.sq.elevator_registered_fn) 856 e->type->ops.sq.elevator_registered_fn(q); 857 } 858 return error; 859 } 860 861 void elv_unregister_queue(struct request_queue *q) 862 { 863 lockdep_assert_held(&q->sysfs_lock); 864 865 if (q) { 866 struct elevator_queue *e = q->elevator; 867 868 kobject_uevent(&e->kobj, KOBJ_REMOVE); 869 kobject_del(&e->kobj); 870 e->registered = 0; 871 /* Re-enable throttling in case elevator disabled it */ 872 wbt_enable_default(q); 873 } 874 } 875 876 int elv_register(struct elevator_type *e) 877 { 878 char *def = ""; 879 880 /* create icq_cache if requested */ 881 if (e->icq_size) { 882 if (WARN_ON(e->icq_size < sizeof(struct io_cq)) || 883 WARN_ON(e->icq_align < __alignof__(struct io_cq))) 884 return -EINVAL; 885 886 snprintf(e->icq_cache_name, sizeof(e->icq_cache_name), 887 "%s_io_cq", e->elevator_name); 888 e->icq_cache = kmem_cache_create(e->icq_cache_name, e->icq_size, 889 e->icq_align, 0, NULL); 890 if (!e->icq_cache) 891 return -ENOMEM; 892 } 893 894 /* register, don't allow duplicate names */ 895 spin_lock(&elv_list_lock); 896 if (elevator_find(e->elevator_name, e->uses_mq)) { 897 spin_unlock(&elv_list_lock); 898 if (e->icq_cache) 899 kmem_cache_destroy(e->icq_cache); 900 return -EBUSY; 901 } 902 list_add_tail(&e->list, &elv_list); 903 spin_unlock(&elv_list_lock); 904 905 /* print pretty message */ 906 if (elevator_match(e, chosen_elevator) || 907 (!*chosen_elevator && 908 elevator_match(e, CONFIG_DEFAULT_IOSCHED))) 909 def = " (default)"; 910 911 printk(KERN_INFO "io scheduler %s registered%s\n", e->elevator_name, 912 def); 913 return 0; 914 } 915 EXPORT_SYMBOL_GPL(elv_register); 916 917 void elv_unregister(struct elevator_type *e) 918 { 919 /* unregister */ 920 spin_lock(&elv_list_lock); 921 list_del_init(&e->list); 922 spin_unlock(&elv_list_lock); 923 924 /* 925 * Destroy icq_cache if it exists. icq's are RCU managed. Make 926 * sure all RCU operations are complete before proceeding. 927 */ 928 if (e->icq_cache) { 929 rcu_barrier(); 930 kmem_cache_destroy(e->icq_cache); 931 e->icq_cache = NULL; 932 } 933 } 934 EXPORT_SYMBOL_GPL(elv_unregister); 935 936 static int elevator_switch_mq(struct request_queue *q, 937 struct elevator_type *new_e) 938 { 939 int ret; 940 941 lockdep_assert_held(&q->sysfs_lock); 942 943 blk_mq_freeze_queue(q); 944 blk_mq_quiesce_queue(q); 945 946 if (q->elevator) { 947 if (q->elevator->registered) 948 elv_unregister_queue(q); 949 ioc_clear_queue(q); 950 elevator_exit(q, q->elevator); 951 } 952 953 ret = blk_mq_init_sched(q, new_e); 954 if (ret) 955 goto out; 956 957 if (new_e) { 958 ret = elv_register_queue(q); 959 if (ret) { 960 elevator_exit(q, q->elevator); 961 goto out; 962 } 963 } 964 965 if (new_e) 966 blk_add_trace_msg(q, "elv switch: %s", new_e->elevator_name); 967 else 968 blk_add_trace_msg(q, "elv switch: none"); 969 970 out: 971 blk_mq_unquiesce_queue(q); 972 blk_mq_unfreeze_queue(q); 973 return ret; 974 } 975 976 /* 977 * For blk-mq devices, we default to using mq-deadline, if available, for single 978 * queue devices. If deadline isn't available OR we have multiple queues, 979 * default to "none". 980 */ 981 int elevator_init_mq(struct request_queue *q) 982 { 983 struct elevator_type *e; 984 int err = 0; 985 986 if (q->nr_hw_queues != 1) 987 return 0; 988 989 /* 990 * q->sysfs_lock must be held to provide mutual exclusion between 991 * elevator_switch() and here. 992 */ 993 mutex_lock(&q->sysfs_lock); 994 if (unlikely(q->elevator)) 995 goto out_unlock; 996 997 e = elevator_get(q, "mq-deadline", false); 998 if (!e) 999 goto out_unlock; 1000 1001 err = blk_mq_init_sched(q, e); 1002 if (err) 1003 elevator_put(e); 1004 out_unlock: 1005 mutex_unlock(&q->sysfs_lock); 1006 return err; 1007 } 1008 1009 1010 /* 1011 * switch to new_e io scheduler. be careful not to introduce deadlocks - 1012 * we don't free the old io scheduler, before we have allocated what we 1013 * need for the new one. this way we have a chance of going back to the old 1014 * one, if the new one fails init for some reason. 1015 */ 1016 static int elevator_switch(struct request_queue *q, struct elevator_type *new_e) 1017 { 1018 struct elevator_queue *old = q->elevator; 1019 bool old_registered = false; 1020 int err; 1021 1022 lockdep_assert_held(&q->sysfs_lock); 1023 1024 if (q->mq_ops) 1025 return elevator_switch_mq(q, new_e); 1026 1027 /* 1028 * Turn on BYPASS and drain all requests w/ elevator private data. 1029 * Block layer doesn't call into a quiesced elevator - all requests 1030 * are directly put on the dispatch list without elevator data 1031 * using INSERT_BACK. All requests have SOFTBARRIER set and no 1032 * merge happens either. 1033 */ 1034 if (old) { 1035 old_registered = old->registered; 1036 1037 blk_queue_bypass_start(q); 1038 1039 /* unregister and clear all auxiliary data of the old elevator */ 1040 if (old_registered) 1041 elv_unregister_queue(q); 1042 1043 ioc_clear_queue(q); 1044 } 1045 1046 /* allocate, init and register new elevator */ 1047 err = new_e->ops.sq.elevator_init_fn(q, new_e); 1048 if (err) 1049 goto fail_init; 1050 1051 err = elv_register_queue(q); 1052 if (err) 1053 goto fail_register; 1054 1055 /* done, kill the old one and finish */ 1056 if (old) { 1057 elevator_exit(q, old); 1058 blk_queue_bypass_end(q); 1059 } 1060 1061 blk_add_trace_msg(q, "elv switch: %s", new_e->elevator_name); 1062 1063 return 0; 1064 1065 fail_register: 1066 elevator_exit(q, q->elevator); 1067 fail_init: 1068 /* switch failed, restore and re-register old elevator */ 1069 if (old) { 1070 q->elevator = old; 1071 elv_register_queue(q); 1072 blk_queue_bypass_end(q); 1073 } 1074 1075 return err; 1076 } 1077 1078 /* 1079 * Switch this queue to the given IO scheduler. 1080 */ 1081 static int __elevator_change(struct request_queue *q, const char *name) 1082 { 1083 char elevator_name[ELV_NAME_MAX]; 1084 struct elevator_type *e; 1085 1086 /* Make sure queue is not in the middle of being removed */ 1087 if (!test_bit(QUEUE_FLAG_REGISTERED, &q->queue_flags)) 1088 return -ENOENT; 1089 1090 /* 1091 * Special case for mq, turn off scheduling 1092 */ 1093 if (q->mq_ops && !strncmp(name, "none", 4)) 1094 return elevator_switch(q, NULL); 1095 1096 strlcpy(elevator_name, name, sizeof(elevator_name)); 1097 e = elevator_get(q, strstrip(elevator_name), true); 1098 if (!e) 1099 return -EINVAL; 1100 1101 if (q->elevator && elevator_match(q->elevator->type, elevator_name)) { 1102 elevator_put(e); 1103 return 0; 1104 } 1105 1106 return elevator_switch(q, e); 1107 } 1108 1109 static inline bool elv_support_iosched(struct request_queue *q) 1110 { 1111 if (q->mq_ops && q->tag_set && (q->tag_set->flags & 1112 BLK_MQ_F_NO_SCHED)) 1113 return false; 1114 return true; 1115 } 1116 1117 ssize_t elv_iosched_store(struct request_queue *q, const char *name, 1118 size_t count) 1119 { 1120 int ret; 1121 1122 if (!(q->mq_ops || q->request_fn) || !elv_support_iosched(q)) 1123 return count; 1124 1125 ret = __elevator_change(q, name); 1126 if (!ret) 1127 return count; 1128 1129 return ret; 1130 } 1131 1132 ssize_t elv_iosched_show(struct request_queue *q, char *name) 1133 { 1134 struct elevator_queue *e = q->elevator; 1135 struct elevator_type *elv = NULL; 1136 struct elevator_type *__e; 1137 bool uses_mq = q->mq_ops != NULL; 1138 int len = 0; 1139 1140 if (!queue_is_rq_based(q)) 1141 return sprintf(name, "none\n"); 1142 1143 if (!q->elevator) 1144 len += sprintf(name+len, "[none] "); 1145 else 1146 elv = e->type; 1147 1148 spin_lock(&elv_list_lock); 1149 list_for_each_entry(__e, &elv_list, list) { 1150 if (elv && elevator_match(elv, __e->elevator_name) && 1151 (__e->uses_mq == uses_mq)) { 1152 len += sprintf(name+len, "[%s] ", elv->elevator_name); 1153 continue; 1154 } 1155 if (__e->uses_mq && q->mq_ops && elv_support_iosched(q)) 1156 len += sprintf(name+len, "%s ", __e->elevator_name); 1157 else if (!__e->uses_mq && !q->mq_ops) 1158 len += sprintf(name+len, "%s ", __e->elevator_name); 1159 } 1160 spin_unlock(&elv_list_lock); 1161 1162 if (q->mq_ops && q->elevator) 1163 len += sprintf(name+len, "none"); 1164 1165 len += sprintf(len+name, "\n"); 1166 return len; 1167 } 1168 1169 struct request *elv_rb_former_request(struct request_queue *q, 1170 struct request *rq) 1171 { 1172 struct rb_node *rbprev = rb_prev(&rq->rb_node); 1173 1174 if (rbprev) 1175 return rb_entry_rq(rbprev); 1176 1177 return NULL; 1178 } 1179 EXPORT_SYMBOL(elv_rb_former_request); 1180 1181 struct request *elv_rb_latter_request(struct request_queue *q, 1182 struct request *rq) 1183 { 1184 struct rb_node *rbnext = rb_next(&rq->rb_node); 1185 1186 if (rbnext) 1187 return rb_entry_rq(rbnext); 1188 1189 return NULL; 1190 } 1191 EXPORT_SYMBOL(elv_rb_latter_request); 1192