1/* 2 * Header file for the BFQ I/O scheduler: data structures and 3 * prototypes of interface functions among BFQ components. 4 * 5 * This program is free software; you can redistribute it and/or 6 * modify it under the terms of the GNU General Public License as 7 * published by the Free Software Foundation; either version 2 of the 8 * License, or (at your option) any later version. 9 * 10 * This program is distributed in the hope that it will be useful, 11 * but WITHOUT ANY WARRANTY; without even the implied warranty of 12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 13 * General Public License for more details. 14 */ 15#ifndef _BFQ_H 16#define _BFQ_H 17 18#include <linux/blktrace_api.h> 19#include <linux/hrtimer.h> 20#include <linux/blk-cgroup.h> 21 22#include "blk-cgroup-rwstat.h" 23 24#define BFQ_IOPRIO_CLASSES 3 25#define BFQ_CL_IDLE_TIMEOUT (HZ/5) 26 27#define BFQ_MIN_WEIGHT 1 28#define BFQ_MAX_WEIGHT 1000 29#define BFQ_WEIGHT_CONVERSION_COEFF 10 30 31#define BFQ_DEFAULT_QUEUE_IOPRIO 4 32 33#define BFQ_WEIGHT_LEGACY_DFL 100 34#define BFQ_DEFAULT_GRP_IOPRIO 0 35#define BFQ_DEFAULT_GRP_CLASS IOPRIO_CLASS_BE 36 37#define MAX_PID_STR_LENGTH 12 38 39/* 40 * Soft real-time applications are extremely more latency sensitive 41 * than interactive ones. Over-raise the weight of the former to 42 * privilege them against the latter. 43 */ 44#define BFQ_SOFTRT_WEIGHT_FACTOR 100 45 46struct bfq_entity; 47 48/** 49 * struct bfq_service_tree - per ioprio_class service tree. 50 * 51 * Each service tree represents a B-WF2Q+ scheduler on its own. Each 52 * ioprio_class has its own independent scheduler, and so its own 53 * bfq_service_tree. All the fields are protected by the queue lock 54 * of the containing bfqd. 55 */ 56struct bfq_service_tree { 57 /* tree for active entities (i.e., those backlogged) */ 58 struct rb_root active; 59 /* tree for idle entities (i.e., not backlogged, with V < F_i)*/ 60 struct rb_root idle; 61 62 /* idle entity with minimum F_i */ 63 struct bfq_entity *first_idle; 64 /* idle entity with maximum F_i */ 65 struct bfq_entity *last_idle; 66 67 /* scheduler virtual time */ 68 u64 vtime; 69 /* scheduler weight sum; active and idle entities contribute to it */ 70 unsigned long wsum; 71}; 72 73/** 74 * struct bfq_sched_data - multi-class scheduler. 75 * 76 * bfq_sched_data is the basic scheduler queue. It supports three 77 * ioprio_classes, and can be used either as a toplevel queue or as an 78 * intermediate queue in a hierarchical setup. 79 * 80 * The supported ioprio_classes are the same as in CFQ, in descending 81 * priority order, IOPRIO_CLASS_RT, IOPRIO_CLASS_BE, IOPRIO_CLASS_IDLE. 82 * Requests from higher priority queues are served before all the 83 * requests from lower priority queues; among requests of the same 84 * queue requests are served according to B-WF2Q+. 85 * 86 * The schedule is implemented by the service trees, plus the field 87 * @next_in_service, which points to the entity on the active trees 88 * that will be served next, if 1) no changes in the schedule occurs 89 * before the current in-service entity is expired, 2) the in-service 90 * queue becomes idle when it expires, and 3) if the entity pointed by 91 * in_service_entity is not a queue, then the in-service child entity 92 * of the entity pointed by in_service_entity becomes idle on 93 * expiration. This peculiar definition allows for the following 94 * optimization, not yet exploited: while a given entity is still in 95 * service, we already know which is the best candidate for next 96 * service among the other active entitities in the same parent 97 * entity. We can then quickly compare the timestamps of the 98 * in-service entity with those of such best candidate. 99 * 100 * All fields are protected by the lock of the containing bfqd. 101 */ 102struct bfq_sched_data { 103 /* entity in service */ 104 struct bfq_entity *in_service_entity; 105 /* head-of-line entity (see comments above) */ 106 struct bfq_entity *next_in_service; 107 /* array of service trees, one per ioprio_class */ 108 struct bfq_service_tree service_tree[BFQ_IOPRIO_CLASSES]; 109 /* last time CLASS_IDLE was served */ 110 unsigned long bfq_class_idle_last_service; 111 112}; 113 114/** 115 * struct bfq_weight_counter - counter of the number of all active queues 116 * with a given weight. 117 */ 118struct bfq_weight_counter { 119 unsigned int weight; /* weight of the queues this counter refers to */ 120 unsigned int num_active; /* nr of active queues with this weight */ 121 /* 122 * Weights tree member (see bfq_data's @queue_weights_tree) 123 */ 124 struct rb_node weights_node; 125}; 126 127/** 128 * struct bfq_entity - schedulable entity. 129 * 130 * A bfq_entity is used to represent either a bfq_queue (leaf node in the 131 * cgroup hierarchy) or a bfq_group into the upper level scheduler. Each 132 * entity belongs to the sched_data of the parent group in the cgroup 133 * hierarchy. Non-leaf entities have also their own sched_data, stored 134 * in @my_sched_data. 135 * 136 * Each entity stores independently its priority values; this would 137 * allow different weights on different devices, but this 138 * functionality is not exported to userspace by now. Priorities and 139 * weights are updated lazily, first storing the new values into the 140 * new_* fields, then setting the @prio_changed flag. As soon as 141 * there is a transition in the entity state that allows the priority 142 * update to take place the effective and the requested priority 143 * values are synchronized. 144 * 145 * Unless cgroups are used, the weight value is calculated from the 146 * ioprio to export the same interface as CFQ. When dealing with 147 * ``well-behaved'' queues (i.e., queues that do not spend too much 148 * time to consume their budget and have true sequential behavior, and 149 * when there are no external factors breaking anticipation) the 150 * relative weights at each level of the cgroups hierarchy should be 151 * guaranteed. All the fields are protected by the queue lock of the 152 * containing bfqd. 153 */ 154struct bfq_entity { 155 /* service_tree member */ 156 struct rb_node rb_node; 157 158 /* 159 * Flag, true if the entity is on a tree (either the active or 160 * the idle one of its service_tree) or is in service. 161 */ 162 bool on_st_or_in_serv; 163 164 /* B-WF2Q+ start and finish timestamps [sectors/weight] */ 165 u64 start, finish; 166 167 /* tree the entity is enqueued into; %NULL if not on a tree */ 168 struct rb_root *tree; 169 170 /* 171 * minimum start time of the (active) subtree rooted at this 172 * entity; used for O(log N) lookups into active trees 173 */ 174 u64 min_start; 175 176 /* amount of service received during the last service slot */ 177 int service; 178 179 /* budget, used also to calculate F_i: F_i = S_i + @budget / @weight */ 180 int budget; 181 182 /* device weight, if non-zero, it overrides the default weight of 183 * bfq_group_data */ 184 int dev_weight; 185 /* weight of the queue */ 186 int weight; 187 /* next weight if a change is in progress */ 188 int new_weight; 189 190 /* original weight, used to implement weight boosting */ 191 int orig_weight; 192 193 /* parent entity, for hierarchical scheduling */ 194 struct bfq_entity *parent; 195 196 /* 197 * For non-leaf nodes in the hierarchy, the associated 198 * scheduler queue, %NULL on leaf nodes. 199 */ 200 struct bfq_sched_data *my_sched_data; 201 /* the scheduler queue this entity belongs to */ 202 struct bfq_sched_data *sched_data; 203 204 /* flag, set to request a weight, ioprio or ioprio_class change */ 205 int prio_changed; 206 207 /* flag, set if the entity is counted in groups_with_pending_reqs */ 208 bool in_groups_with_pending_reqs; 209}; 210 211struct bfq_group; 212 213/** 214 * struct bfq_ttime - per process thinktime stats. 215 */ 216struct bfq_ttime { 217 /* completion time of the last request */ 218 u64 last_end_request; 219 220 /* total process thinktime */ 221 u64 ttime_total; 222 /* number of thinktime samples */ 223 unsigned long ttime_samples; 224 /* average process thinktime */ 225 u64 ttime_mean; 226}; 227 228/** 229 * struct bfq_queue - leaf schedulable entity. 230 * 231 * A bfq_queue is a leaf request queue; it can be associated with an 232 * io_context or more, if it is async or shared between cooperating 233 * processes. @cgroup holds a reference to the cgroup, to be sure that it 234 * does not disappear while a bfqq still references it (mostly to avoid 235 * races between request issuing and task migration followed by cgroup 236 * destruction). 237 * All the fields are protected by the queue lock of the containing bfqd. 238 */ 239struct bfq_queue { 240 /* reference counter */ 241 int ref; 242 /* parent bfq_data */ 243 struct bfq_data *bfqd; 244 245 /* current ioprio and ioprio class */ 246 unsigned short ioprio, ioprio_class; 247 /* next ioprio and ioprio class if a change is in progress */ 248 unsigned short new_ioprio, new_ioprio_class; 249 250 /* last total-service-time sample, see bfq_update_inject_limit() */ 251 u64 last_serv_time_ns; 252 /* limit for request injection */ 253 unsigned int inject_limit; 254 /* last time the inject limit has been decreased, in jiffies */ 255 unsigned long decrease_time_jif; 256 257 /* 258 * Shared bfq_queue if queue is cooperating with one or more 259 * other queues. 260 */ 261 struct bfq_queue *new_bfqq; 262 /* request-position tree member (see bfq_group's @rq_pos_tree) */ 263 struct rb_node pos_node; 264 /* request-position tree root (see bfq_group's @rq_pos_tree) */ 265 struct rb_root *pos_root; 266 267 /* sorted list of pending requests */ 268 struct rb_root sort_list; 269 /* if fifo isn't expired, next request to serve */ 270 struct request *next_rq; 271 /* number of sync and async requests queued */ 272 int queued[2]; 273 /* number of requests currently allocated */ 274 int allocated; 275 /* number of pending metadata requests */ 276 int meta_pending; 277 /* fifo list of requests in sort_list */ 278 struct list_head fifo; 279 280 /* entity representing this queue in the scheduler */ 281 struct bfq_entity entity; 282 283 /* pointer to the weight counter associated with this entity */ 284 struct bfq_weight_counter *weight_counter; 285 286 /* maximum budget allowed from the feedback mechanism */ 287 int max_budget; 288 /* budget expiration (in jiffies) */ 289 unsigned long budget_timeout; 290 291 /* number of requests on the dispatch list or inside driver */ 292 int dispatched; 293 294 /* status flags */ 295 unsigned long flags; 296 297 /* node for active/idle bfqq list inside parent bfqd */ 298 struct list_head bfqq_list; 299 300 /* associated @bfq_ttime struct */ 301 struct bfq_ttime ttime; 302 303 /* when bfqq started to do I/O within the last observation window */ 304 u64 io_start_time; 305 /* how long bfqq has remained empty during the last observ. window */ 306 u64 tot_idle_time; 307 308 /* bit vector: a 1 for each seeky requests in history */ 309 u32 seek_history; 310 311 /* node for the device's burst list */ 312 struct hlist_node burst_list_node; 313 314 /* position of the last request enqueued */ 315 sector_t last_request_pos; 316 317 /* Number of consecutive pairs of request completion and 318 * arrival, such that the queue becomes idle after the 319 * completion, but the next request arrives within an idle 320 * time slice; used only if the queue's IO_bound flag has been 321 * cleared. 322 */ 323 unsigned int requests_within_timer; 324 325 /* pid of the process owning the queue, used for logging purposes */ 326 pid_t pid; 327 328 /* 329 * Pointer to the bfq_io_cq owning the bfq_queue, set to %NULL 330 * if the queue is shared. 331 */ 332 struct bfq_io_cq *bic; 333 334 /* current maximum weight-raising time for this queue */ 335 unsigned long wr_cur_max_time; 336 /* 337 * Minimum time instant such that, only if a new request is 338 * enqueued after this time instant in an idle @bfq_queue with 339 * no outstanding requests, then the task associated with the 340 * queue it is deemed as soft real-time (see the comments on 341 * the function bfq_bfqq_softrt_next_start()) 342 */ 343 unsigned long soft_rt_next_start; 344 /* 345 * Start time of the current weight-raising period if 346 * the @bfq-queue is being weight-raised, otherwise 347 * finish time of the last weight-raising period. 348 */ 349 unsigned long last_wr_start_finish; 350 /* factor by which the weight of this queue is multiplied */ 351 unsigned int wr_coeff; 352 /* 353 * Time of the last transition of the @bfq_queue from idle to 354 * backlogged. 355 */ 356 unsigned long last_idle_bklogged; 357 /* 358 * Cumulative service received from the @bfq_queue since the 359 * last transition from idle to backlogged. 360 */ 361 unsigned long service_from_backlogged; 362 /* 363 * Cumulative service received from the @bfq_queue since its 364 * last transition to weight-raised state. 365 */ 366 unsigned long service_from_wr; 367 368 /* 369 * Value of wr start time when switching to soft rt 370 */ 371 unsigned long wr_start_at_switch_to_srt; 372 373 unsigned long split_time; /* time of last split */ 374 375 unsigned long first_IO_time; /* time of first I/O for this queue */ 376 377 /* max service rate measured so far */ 378 u32 max_service_rate; 379 380 /* 381 * Pointer to the waker queue for this queue, i.e., to the 382 * queue Q such that this queue happens to get new I/O right 383 * after some I/O request of Q is completed. For details, see 384 * the comments on the choice of the queue for injection in 385 * bfq_select_queue(). 386 */ 387 struct bfq_queue *waker_bfqq; 388 /* pointer to the curr. tentative waker queue, see bfq_check_waker() */ 389 struct bfq_queue *tentative_waker_bfqq; 390 /* number of times the same tentative waker has been detected */ 391 unsigned int num_waker_detections; 392 393 /* node for woken_list, see below */ 394 struct hlist_node woken_list_node; 395 /* 396 * Head of the list of the woken queues for this queue, i.e., 397 * of the list of the queues for which this queue is a waker 398 * queue. This list is used to reset the waker_bfqq pointer in 399 * the woken queues when this queue exits. 400 */ 401 struct hlist_head woken_list; 402}; 403 404/** 405 * struct bfq_io_cq - per (request_queue, io_context) structure. 406 */ 407struct bfq_io_cq { 408 /* associated io_cq structure */ 409 struct io_cq icq; /* must be the first member */ 410 /* array of two process queues, the sync and the async */ 411 struct bfq_queue *bfqq[2]; 412 /* per (request_queue, blkcg) ioprio */ 413 int ioprio; 414#ifdef CONFIG_BFQ_GROUP_IOSCHED 415 uint64_t blkcg_serial_nr; /* the current blkcg serial */ 416#endif 417 /* 418 * Snapshot of the has_short_time flag before merging; taken 419 * to remember its value while the queue is merged, so as to 420 * be able to restore it in case of split. 421 */ 422 bool saved_has_short_ttime; 423 /* 424 * Same purpose as the previous two fields for the I/O bound 425 * classification of a queue. 426 */ 427 bool saved_IO_bound; 428 429 u64 saved_io_start_time; 430 u64 saved_tot_idle_time; 431 432 /* 433 * Same purpose as the previous fields for the value of the 434 * field keeping the queue's belonging to a large burst 435 */ 436 bool saved_in_large_burst; 437 /* 438 * True if the queue belonged to a burst list before its merge 439 * with another cooperating queue. 440 */ 441 bool was_in_burst_list; 442 443 /* 444 * Save the weight when a merge occurs, to be able 445 * to restore it in case of split. If the weight is not 446 * correctly resumed when the queue is recycled, 447 * then the weight of the recycled queue could differ 448 * from the weight of the original queue. 449 */ 450 unsigned int saved_weight; 451 452 /* 453 * Similar to previous fields: save wr information. 454 */ 455 unsigned long saved_wr_coeff; 456 unsigned long saved_last_wr_start_finish; 457 unsigned long saved_service_from_wr; 458 unsigned long saved_wr_start_at_switch_to_srt; 459 unsigned int saved_wr_cur_max_time; 460 struct bfq_ttime saved_ttime; 461 462 /* Save also injection state */ 463 u64 saved_last_serv_time_ns; 464 unsigned int saved_inject_limit; 465 unsigned long saved_decrease_time_jif; 466}; 467 468/** 469 * struct bfq_data - per-device data structure. 470 * 471 * All the fields are protected by @lock. 472 */ 473struct bfq_data { 474 /* device request queue */ 475 struct request_queue *queue; 476 /* dispatch queue */ 477 struct list_head dispatch; 478 479 /* root bfq_group for the device */ 480 struct bfq_group *root_group; 481 482 /* 483 * rbtree of weight counters of @bfq_queues, sorted by 484 * weight. Used to keep track of whether all @bfq_queues have 485 * the same weight. The tree contains one counter for each 486 * distinct weight associated to some active and not 487 * weight-raised @bfq_queue (see the comments to the functions 488 * bfq_weights_tree_[add|remove] for further details). 489 */ 490 struct rb_root_cached queue_weights_tree; 491 492 /* 493 * Number of groups with at least one descendant process that 494 * has at least one request waiting for completion. Note that 495 * this accounts for also requests already dispatched, but not 496 * yet completed. Therefore this number of groups may differ 497 * (be larger) than the number of active groups, as a group is 498 * considered active only if its corresponding entity has 499 * descendant queues with at least one request queued. This 500 * number is used to decide whether a scenario is symmetric. 501 * For a detailed explanation see comments on the computation 502 * of the variable asymmetric_scenario in the function 503 * bfq_better_to_idle(). 504 * 505 * However, it is hard to compute this number exactly, for 506 * groups with multiple descendant processes. Consider a group 507 * that is inactive, i.e., that has no descendant process with 508 * pending I/O inside BFQ queues. Then suppose that 509 * num_groups_with_pending_reqs is still accounting for this 510 * group, because the group has descendant processes with some 511 * I/O request still in flight. num_groups_with_pending_reqs 512 * should be decremented when the in-flight request of the 513 * last descendant process is finally completed (assuming that 514 * nothing else has changed for the group in the meantime, in 515 * terms of composition of the group and active/inactive state of child 516 * groups and processes). To accomplish this, an additional 517 * pending-request counter must be added to entities, and must 518 * be updated correctly. To avoid this additional field and operations, 519 * we resort to the following tradeoff between simplicity and 520 * accuracy: for an inactive group that is still counted in 521 * num_groups_with_pending_reqs, we decrement 522 * num_groups_with_pending_reqs when the first descendant 523 * process of the group remains with no request waiting for 524 * completion. 525 * 526 * Even this simpler decrement strategy requires a little 527 * carefulness: to avoid multiple decrements, we flag a group, 528 * more precisely an entity representing a group, as still 529 * counted in num_groups_with_pending_reqs when it becomes 530 * inactive. Then, when the first descendant queue of the 531 * entity remains with no request waiting for completion, 532 * num_groups_with_pending_reqs is decremented, and this flag 533 * is reset. After this flag is reset for the entity, 534 * num_groups_with_pending_reqs won't be decremented any 535 * longer in case a new descendant queue of the entity remains 536 * with no request waiting for completion. 537 */ 538 unsigned int num_groups_with_pending_reqs; 539 540 /* 541 * Per-class (RT, BE, IDLE) number of bfq_queues containing 542 * requests (including the queue in service, even if it is 543 * idling). 544 */ 545 unsigned int busy_queues[3]; 546 /* number of weight-raised busy @bfq_queues */ 547 int wr_busy_queues; 548 /* number of queued requests */ 549 int queued; 550 /* number of requests dispatched and waiting for completion */ 551 int rq_in_driver; 552 553 /* true if the device is non rotational and performs queueing */ 554 bool nonrot_with_queueing; 555 556 /* 557 * Maximum number of requests in driver in the last 558 * @hw_tag_samples completed requests. 559 */ 560 int max_rq_in_driver; 561 /* number of samples used to calculate hw_tag */ 562 int hw_tag_samples; 563 /* flag set to one if the driver is showing a queueing behavior */ 564 int hw_tag; 565 566 /* number of budgets assigned */ 567 int budgets_assigned; 568 569 /* 570 * Timer set when idling (waiting) for the next request from 571 * the queue in service. 572 */ 573 struct hrtimer idle_slice_timer; 574 575 /* bfq_queue in service */ 576 struct bfq_queue *in_service_queue; 577 578 /* on-disk position of the last served request */ 579 sector_t last_position; 580 581 /* position of the last served request for the in-service queue */ 582 sector_t in_serv_last_pos; 583 584 /* time of last request completion (ns) */ 585 u64 last_completion; 586 587 /* bfqq owning the last completed rq */ 588 struct bfq_queue *last_completed_rq_bfqq; 589 590 /* time of last transition from empty to non-empty (ns) */ 591 u64 last_empty_occupied_ns; 592 593 /* 594 * Flag set to activate the sampling of the total service time 595 * of a just-arrived first I/O request (see 596 * bfq_update_inject_limit()). This will cause the setting of 597 * waited_rq when the request is finally dispatched. 598 */ 599 bool wait_dispatch; 600 /* 601 * If set, then bfq_update_inject_limit() is invoked when 602 * waited_rq is eventually completed. 603 */ 604 struct request *waited_rq; 605 /* 606 * True if some request has been injected during the last service hole. 607 */ 608 bool rqs_injected; 609 610 /* time of first rq dispatch in current observation interval (ns) */ 611 u64 first_dispatch; 612 /* time of last rq dispatch in current observation interval (ns) */ 613 u64 last_dispatch; 614 615 /* beginning of the last budget */ 616 ktime_t last_budget_start; 617 /* beginning of the last idle slice */ 618 ktime_t last_idling_start; 619 unsigned long last_idling_start_jiffies; 620 621 /* number of samples in current observation interval */ 622 int peak_rate_samples; 623 /* num of samples of seq dispatches in current observation interval */ 624 u32 sequential_samples; 625 /* total num of sectors transferred in current observation interval */ 626 u64 tot_sectors_dispatched; 627 /* max rq size seen during current observation interval (sectors) */ 628 u32 last_rq_max_size; 629 /* time elapsed from first dispatch in current observ. interval (us) */ 630 u64 delta_from_first; 631 /* 632 * Current estimate of the device peak rate, measured in 633 * [(sectors/usec) / 2^BFQ_RATE_SHIFT]. The left-shift by 634 * BFQ_RATE_SHIFT is performed to increase precision in 635 * fixed-point calculations. 636 */ 637 u32 peak_rate; 638 639 /* maximum budget allotted to a bfq_queue before rescheduling */ 640 int bfq_max_budget; 641 642 /* list of all the bfq_queues active on the device */ 643 struct list_head active_list; 644 /* list of all the bfq_queues idle on the device */ 645 struct list_head idle_list; 646 647 /* 648 * Timeout for async/sync requests; when it fires, requests 649 * are served in fifo order. 650 */ 651 u64 bfq_fifo_expire[2]; 652 /* weight of backward seeks wrt forward ones */ 653 unsigned int bfq_back_penalty; 654 /* maximum allowed backward seek */ 655 unsigned int bfq_back_max; 656 /* maximum idling time */ 657 u32 bfq_slice_idle; 658 659 /* user-configured max budget value (0 for auto-tuning) */ 660 int bfq_user_max_budget; 661 /* 662 * Timeout for bfq_queues to consume their budget; used to 663 * prevent seeky queues from imposing long latencies to 664 * sequential or quasi-sequential ones (this also implies that 665 * seeky queues cannot receive guarantees in the service 666 * domain; after a timeout they are charged for the time they 667 * have been in service, to preserve fairness among them, but 668 * without service-domain guarantees). 669 */ 670 unsigned int bfq_timeout; 671 672 /* 673 * Force device idling whenever needed to provide accurate 674 * service guarantees, without caring about throughput 675 * issues. CAVEAT: this may even increase latencies, in case 676 * of useless idling for processes that did stop doing I/O. 677 */ 678 bool strict_guarantees; 679 680 /* 681 * Last time at which a queue entered the current burst of 682 * queues being activated shortly after each other; for more 683 * details about this and the following parameters related to 684 * a burst of activations, see the comments on the function 685 * bfq_handle_burst. 686 */ 687 unsigned long last_ins_in_burst; 688 /* 689 * Reference time interval used to decide whether a queue has 690 * been activated shortly after @last_ins_in_burst. 691 */ 692 unsigned long bfq_burst_interval; 693 /* number of queues in the current burst of queue activations */ 694 int burst_size; 695 696 /* common parent entity for the queues in the burst */ 697 struct bfq_entity *burst_parent_entity; 698 /* Maximum burst size above which the current queue-activation 699 * burst is deemed as 'large'. 700 */ 701 unsigned long bfq_large_burst_thresh; 702 /* true if a large queue-activation burst is in progress */ 703 bool large_burst; 704 /* 705 * Head of the burst list (as for the above fields, more 706 * details in the comments on the function bfq_handle_burst). 707 */ 708 struct hlist_head burst_list; 709 710 /* if set to true, low-latency heuristics are enabled */ 711 bool low_latency; 712 /* 713 * Maximum factor by which the weight of a weight-raised queue 714 * is multiplied. 715 */ 716 unsigned int bfq_wr_coeff; 717 /* maximum duration of a weight-raising period (jiffies) */ 718 unsigned int bfq_wr_max_time; 719 720 /* Maximum weight-raising duration for soft real-time processes */ 721 unsigned int bfq_wr_rt_max_time; 722 /* 723 * Minimum idle period after which weight-raising may be 724 * reactivated for a queue (in jiffies). 725 */ 726 unsigned int bfq_wr_min_idle_time; 727 /* 728 * Minimum period between request arrivals after which 729 * weight-raising may be reactivated for an already busy async 730 * queue (in jiffies). 731 */ 732 unsigned long bfq_wr_min_inter_arr_async; 733 734 /* Max service-rate for a soft real-time queue, in sectors/sec */ 735 unsigned int bfq_wr_max_softrt_rate; 736 /* 737 * Cached value of the product ref_rate*ref_wr_duration, used 738 * for computing the maximum duration of weight raising 739 * automatically. 740 */ 741 u64 rate_dur_prod; 742 743 /* fallback dummy bfqq for extreme OOM conditions */ 744 struct bfq_queue oom_bfqq; 745 746 spinlock_t lock; 747 748 /* 749 * bic associated with the task issuing current bio for 750 * merging. This and the next field are used as a support to 751 * be able to perform the bic lookup, needed by bio-merge 752 * functions, before the scheduler lock is taken, and thus 753 * avoid taking the request-queue lock while the scheduler 754 * lock is being held. 755 */ 756 struct bfq_io_cq *bio_bic; 757 /* bfqq associated with the task issuing current bio for merging */ 758 struct bfq_queue *bio_bfqq; 759 760 /* 761 * Depth limits used in bfq_limit_depth (see comments on the 762 * function) 763 */ 764 unsigned int word_depths[2][2]; 765}; 766 767enum bfqq_state_flags { 768 BFQQF_just_created = 0, /* queue just allocated */ 769 BFQQF_busy, /* has requests or is in service */ 770 BFQQF_wait_request, /* waiting for a request */ 771 BFQQF_non_blocking_wait_rq, /* 772 * waiting for a request 773 * without idling the device 774 */ 775 BFQQF_fifo_expire, /* FIFO checked in this slice */ 776 BFQQF_has_short_ttime, /* queue has a short think time */ 777 BFQQF_sync, /* synchronous queue */ 778 BFQQF_IO_bound, /* 779 * bfqq has timed-out at least once 780 * having consumed at most 2/10 of 781 * its budget 782 */ 783 BFQQF_in_large_burst, /* 784 * bfqq activated in a large burst, 785 * see comments to bfq_handle_burst. 786 */ 787 BFQQF_softrt_update, /* 788 * may need softrt-next-start 789 * update 790 */ 791 BFQQF_coop, /* bfqq is shared */ 792 BFQQF_split_coop, /* shared bfqq will be split */ 793}; 794 795#define BFQ_BFQQ_FNS(name) \ 796void bfq_mark_bfqq_##name(struct bfq_queue *bfqq); \ 797void bfq_clear_bfqq_##name(struct bfq_queue *bfqq); \ 798int bfq_bfqq_##name(const struct bfq_queue *bfqq); 799 800BFQ_BFQQ_FNS(just_created); 801BFQ_BFQQ_FNS(busy); 802BFQ_BFQQ_FNS(wait_request); 803BFQ_BFQQ_FNS(non_blocking_wait_rq); 804BFQ_BFQQ_FNS(fifo_expire); 805BFQ_BFQQ_FNS(has_short_ttime); 806BFQ_BFQQ_FNS(sync); 807BFQ_BFQQ_FNS(IO_bound); 808BFQ_BFQQ_FNS(in_large_burst); 809BFQ_BFQQ_FNS(coop); 810BFQ_BFQQ_FNS(split_coop); 811BFQ_BFQQ_FNS(softrt_update); 812#undef BFQ_BFQQ_FNS 813 814/* Expiration reasons. */ 815enum bfqq_expiration { 816 BFQQE_TOO_IDLE = 0, /* 817 * queue has been idling for 818 * too long 819 */ 820 BFQQE_BUDGET_TIMEOUT, /* budget took too long to be used */ 821 BFQQE_BUDGET_EXHAUSTED, /* budget consumed */ 822 BFQQE_NO_MORE_REQUESTS, /* the queue has no more requests */ 823 BFQQE_PREEMPTED /* preemption in progress */ 824}; 825 826struct bfq_stat { 827 struct percpu_counter cpu_cnt; 828 atomic64_t aux_cnt; 829}; 830 831struct bfqg_stats { 832 /* basic stats */ 833 struct blkg_rwstat bytes; 834 struct blkg_rwstat ios; 835#ifdef CONFIG_BFQ_CGROUP_DEBUG 836 /* number of ios merged */ 837 struct blkg_rwstat merged; 838 /* total time spent on device in ns, may not be accurate w/ queueing */ 839 struct blkg_rwstat service_time; 840 /* total time spent waiting in scheduler queue in ns */ 841 struct blkg_rwstat wait_time; 842 /* number of IOs queued up */ 843 struct blkg_rwstat queued; 844 /* total disk time and nr sectors dispatched by this group */ 845 struct bfq_stat time; 846 /* sum of number of ios queued across all samples */ 847 struct bfq_stat avg_queue_size_sum; 848 /* count of samples taken for average */ 849 struct bfq_stat avg_queue_size_samples; 850 /* how many times this group has been removed from service tree */ 851 struct bfq_stat dequeue; 852 /* total time spent waiting for it to be assigned a timeslice. */ 853 struct bfq_stat group_wait_time; 854 /* time spent idling for this blkcg_gq */ 855 struct bfq_stat idle_time; 856 /* total time with empty current active q with other requests queued */ 857 struct bfq_stat empty_time; 858 /* fields after this shouldn't be cleared on stat reset */ 859 u64 start_group_wait_time; 860 u64 start_idle_time; 861 u64 start_empty_time; 862 uint16_t flags; 863#endif /* CONFIG_BFQ_CGROUP_DEBUG */ 864}; 865 866#ifdef CONFIG_BFQ_GROUP_IOSCHED 867 868/* 869 * struct bfq_group_data - per-blkcg storage for the blkio subsystem. 870 * 871 * @ps: @blkcg_policy_storage that this structure inherits 872 * @weight: weight of the bfq_group 873 */ 874struct bfq_group_data { 875 /* must be the first member */ 876 struct blkcg_policy_data pd; 877 878 unsigned int weight; 879}; 880 881/** 882 * struct bfq_group - per (device, cgroup) data structure. 883 * @entity: schedulable entity to insert into the parent group sched_data. 884 * @sched_data: own sched_data, to contain child entities (they may be 885 * both bfq_queues and bfq_groups). 886 * @bfqd: the bfq_data for the device this group acts upon. 887 * @async_bfqq: array of async queues for all the tasks belonging to 888 * the group, one queue per ioprio value per ioprio_class, 889 * except for the idle class that has only one queue. 890 * @async_idle_bfqq: async queue for the idle class (ioprio is ignored). 891 * @my_entity: pointer to @entity, %NULL for the toplevel group; used 892 * to avoid too many special cases during group creation/ 893 * migration. 894 * @stats: stats for this bfqg. 895 * @active_entities: number of active entities belonging to the group; 896 * unused for the root group. Used to know whether there 897 * are groups with more than one active @bfq_entity 898 * (see the comments to the function 899 * bfq_bfqq_may_idle()). 900 * @rq_pos_tree: rbtree sorted by next_request position, used when 901 * determining if two or more queues have interleaving 902 * requests (see bfq_find_close_cooperator()). 903 * 904 * Each (device, cgroup) pair has its own bfq_group, i.e., for each cgroup 905 * there is a set of bfq_groups, each one collecting the lower-level 906 * entities belonging to the group that are acting on the same device. 907 * 908 * Locking works as follows: 909 * o @bfqd is protected by the queue lock, RCU is used to access it 910 * from the readers. 911 * o All the other fields are protected by the @bfqd queue lock. 912 */ 913struct bfq_group { 914 /* must be the first member */ 915 struct blkg_policy_data pd; 916 917 /* cached path for this blkg (see comments in bfq_bic_update_cgroup) */ 918 char blkg_path[128]; 919 920 /* reference counter (see comments in bfq_bic_update_cgroup) */ 921 int ref; 922 923 struct bfq_entity entity; 924 struct bfq_sched_data sched_data; 925 926 void *bfqd; 927 928 struct bfq_queue *async_bfqq[2][IOPRIO_BE_NR]; 929 struct bfq_queue *async_idle_bfqq; 930 931 struct bfq_entity *my_entity; 932 933 int active_entities; 934 935 struct rb_root rq_pos_tree; 936 937 struct bfqg_stats stats; 938}; 939 940#else 941struct bfq_group { 942 struct bfq_entity entity; 943 struct bfq_sched_data sched_data; 944 945 struct bfq_queue *async_bfqq[2][IOPRIO_BE_NR]; 946 struct bfq_queue *async_idle_bfqq; 947 948 struct rb_root rq_pos_tree; 949}; 950#endif 951 952struct bfq_queue *bfq_entity_to_bfqq(struct bfq_entity *entity); 953 954/* --------------- main algorithm interface ----------------- */ 955 956#define BFQ_SERVICE_TREE_INIT ((struct bfq_service_tree) \ 957 { RB_ROOT, RB_ROOT, NULL, NULL, 0, 0 }) 958 959extern const int bfq_timeout; 960 961struct bfq_queue *bic_to_bfqq(struct bfq_io_cq *bic, bool is_sync); 962void bic_set_bfqq(struct bfq_io_cq *bic, struct bfq_queue *bfqq, bool is_sync); 963struct bfq_data *bic_to_bfqd(struct bfq_io_cq *bic); 964void bfq_pos_tree_add_move(struct bfq_data *bfqd, struct bfq_queue *bfqq); 965void bfq_weights_tree_add(struct bfq_data *bfqd, struct bfq_queue *bfqq, 966 struct rb_root_cached *root); 967void __bfq_weights_tree_remove(struct bfq_data *bfqd, 968 struct bfq_queue *bfqq, 969 struct rb_root_cached *root); 970void bfq_weights_tree_remove(struct bfq_data *bfqd, 971 struct bfq_queue *bfqq); 972void bfq_bfqq_expire(struct bfq_data *bfqd, struct bfq_queue *bfqq, 973 bool compensate, enum bfqq_expiration reason); 974void bfq_put_queue(struct bfq_queue *bfqq); 975void bfq_end_wr_async_queues(struct bfq_data *bfqd, struct bfq_group *bfqg); 976void bfq_release_process_ref(struct bfq_data *bfqd, struct bfq_queue *bfqq); 977void bfq_schedule_dispatch(struct bfq_data *bfqd); 978void bfq_put_async_queues(struct bfq_data *bfqd, struct bfq_group *bfqg); 979 980/* ------------ end of main algorithm interface -------------- */ 981 982/* ---------------- cgroups-support interface ---------------- */ 983 984void bfqg_stats_update_legacy_io(struct request_queue *q, struct request *rq); 985void bfqg_stats_update_io_add(struct bfq_group *bfqg, struct bfq_queue *bfqq, 986 unsigned int op); 987void bfqg_stats_update_io_remove(struct bfq_group *bfqg, unsigned int op); 988void bfqg_stats_update_io_merged(struct bfq_group *bfqg, unsigned int op); 989void bfqg_stats_update_completion(struct bfq_group *bfqg, u64 start_time_ns, 990 u64 io_start_time_ns, unsigned int op); 991void bfqg_stats_update_dequeue(struct bfq_group *bfqg); 992void bfqg_stats_set_start_empty_time(struct bfq_group *bfqg); 993void bfqg_stats_update_idle_time(struct bfq_group *bfqg); 994void bfqg_stats_set_start_idle_time(struct bfq_group *bfqg); 995void bfqg_stats_update_avg_queue_size(struct bfq_group *bfqg); 996void bfq_bfqq_move(struct bfq_data *bfqd, struct bfq_queue *bfqq, 997 struct bfq_group *bfqg); 998 999void bfq_init_entity(struct bfq_entity *entity, struct bfq_group *bfqg); 1000void bfq_bic_update_cgroup(struct bfq_io_cq *bic, struct bio *bio);
1001void bfq_end_wr_async(struct bfq_data *bfqd); 1002struct bfq_group *bfq_find_set_group(struct bfq_data *bfqd, 1003 struct blkcg *blkcg); 1004struct blkcg_gq *bfqg_to_blkg(struct bfq_group *bfqg); 1005struct bfq_group *bfqq_group(struct bfq_queue *bfqq); 1006struct bfq_group *bfq_create_group_hierarchy(struct bfq_data *bfqd, int node); 1007void bfqg_and_blkg_get(struct bfq_group *bfqg); 1008void bfqg_and_blkg_put(struct bfq_group *bfqg); 1009 1010#ifdef CONFIG_BFQ_GROUP_IOSCHED 1011extern struct cftype bfq_blkcg_legacy_files[]; 1012extern struct cftype bfq_blkg_files[]; 1013extern struct blkcg_policy blkcg_policy_bfq; 1014#endif 1015 1016/* ------------- end of cgroups-support interface ------------- */ 1017 1018/* - interface of the internal hierarchical B-WF2Q+ scheduler - */ 1019 1020#ifdef CONFIG_BFQ_GROUP_IOSCHED 1021/* both next loops stop at one of the child entities of the root group */ 1022#define for_each_entity(entity) \ 1023 for (; entity ; entity = entity->parent) 1024 1025/* 1026 * For each iteration, compute parent in advance, so as to be safe if 1027 * entity is deallocated during the iteration. Such a deallocation may 1028 * happen as a consequence of a bfq_put_queue that frees the bfq_queue 1029 * containing entity. 1030 */ 1031#define for_each_entity_safe(entity, parent) \ 1032 for (; entity && ({ parent = entity->parent; 1; }); entity = parent) 1033 1034#else /* CONFIG_BFQ_GROUP_IOSCHED */ 1035/* 1036 * Next two macros are fake loops when cgroups support is not 1037 * enabled. I fact, in such a case, there is only one level to go up 1038 * (to reach the root group). 1039 */ 1040#define for_each_entity(entity) \ 1041 for (; entity ; entity = NULL) 1042 1043#define for_each_entity_safe(entity, parent) \ 1044 for (parent = NULL; entity ; entity = parent) 1045#endif /* CONFIG_BFQ_GROUP_IOSCHED */ 1046 1047struct bfq_group *bfq_bfqq_to_bfqg(struct bfq_queue *bfqq); 1048struct bfq_queue *bfq_entity_to_bfqq(struct bfq_entity *entity); 1049unsigned int bfq_tot_busy_queues(struct bfq_data *bfqd); 1050struct bfq_service_tree *bfq_entity_service_tree(struct bfq_entity *entity); 1051struct bfq_entity *bfq_entity_of(struct rb_node *node); 1052unsigned short bfq_ioprio_to_weight(int ioprio); 1053void bfq_put_idle_entity(struct bfq_service_tree *st, 1054 struct bfq_entity *entity); 1055struct bfq_service_tree * 1056__bfq_entity_update_weight_prio(struct bfq_service_tree *old_st, 1057 struct bfq_entity *entity, 1058 bool update_class_too); 1059void bfq_bfqq_served(struct bfq_queue *bfqq, int served); 1060void bfq_bfqq_charge_time(struct bfq_data *bfqd, struct bfq_queue *bfqq, 1061 unsigned long time_ms); 1062bool __bfq_deactivate_entity(struct bfq_entity *entity, 1063 bool ins_into_idle_tree); 1064bool next_queue_may_preempt(struct bfq_data *bfqd); 1065struct bfq_queue *bfq_get_next_queue(struct bfq_data *bfqd); 1066bool __bfq_bfqd_reset_in_service(struct bfq_data *bfqd); 1067void bfq_deactivate_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq, 1068 bool ins_into_idle_tree, bool expiration); 1069void bfq_activate_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq); 1070void bfq_requeue_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq, 1071 bool expiration); 1072void bfq_del_bfqq_busy(struct bfq_data *bfqd, struct bfq_queue *bfqq, 1073 bool expiration); 1074void bfq_add_bfqq_busy(struct bfq_data *bfqd, struct bfq_queue *bfqq); 1075 1076/* --------------- end of interface of B-WF2Q+ ---------------- */ 1077 1078/* Logging facilities. */ 1079static inline void bfq_pid_to_str(int pid, char *str, int len) 1080{ 1081 if (pid != -1) 1082 snprintf(str, len, "%d", pid); 1083 else 1084 snprintf(str, len, "SHARED-"); 1085} 1086 1087#ifdef CONFIG_BFQ_GROUP_IOSCHED 1088struct bfq_group *bfqq_group(struct bfq_queue *bfqq); 1089 1090#define bfq_log_bfqq(bfqd, bfqq, fmt, args...) do { \ 1091 char pid_str[MAX_PID_STR_LENGTH]; \ 1092 if (likely(!blk_trace_note_message_enabled((bfqd)->queue))) \ 1093 break; \ 1094 bfq_pid_to_str((bfqq)->pid, pid_str, MAX_PID_STR_LENGTH); \ 1095 blk_add_cgroup_trace_msg((bfqd)->queue, \ 1096 bfqg_to_blkg(bfqq_group(bfqq))->blkcg, \ 1097 "bfq%s%c " fmt, pid_str, \ 1098 bfq_bfqq_sync((bfqq)) ? 'S' : 'A', ##args); \ 1099} while (0) 1100 1101#define bfq_log_bfqg(bfqd, bfqg, fmt, args...) do { \ 1102 blk_add_cgroup_trace_msg((bfqd)->queue, \ 1103 bfqg_to_blkg(bfqg)->blkcg, fmt, ##args); \ 1104} while (0) 1105 1106#else /* CONFIG_BFQ_GROUP_IOSCHED */ 1107 1108#define bfq_log_bfqq(bfqd, bfqq, fmt, args...) do { \ 1109 char pid_str[MAX_PID_STR_LENGTH]; \ 1110 if (likely(!blk_trace_note_message_enabled((bfqd)->queue))) \ 1111 break; \ 1112 bfq_pid_to_str((bfqq)->pid, pid_str, MAX_PID_STR_LENGTH); \ 1113 blk_add_trace_msg((bfqd)->queue, "bfq%s%c " fmt, pid_str, \ 1114 bfq_bfqq_sync((bfqq)) ? 'S' : 'A', \ 1115 ##args); \ 1116} while (0) 1117#define bfq_log_bfqg(bfqd, bfqg, fmt, args...) do {} while (0) 1118 1119#endif /* CONFIG_BFQ_GROUP_IOSCHED */ 1120 1121#define bfq_log(bfqd, fmt, args...) \ 1122 blk_add_trace_msg((bfqd)->queue, "bfq " fmt, ##args) 1123 1124#endif /* _BFQ_H */ 1125