1 /* SPDX-License-Identifier: GPL-2.0-or-later */ 2 /* 3 * Header file for the BFQ I/O scheduler: data structures and 4 * prototypes of interface functions among BFQ components. 5 */ 6 #ifndef _BFQ_H 7 #define _BFQ_H 8 9 #include <linux/blktrace_api.h> 10 #include <linux/hrtimer.h> 11 #include <linux/blk-cgroup.h> 12 13 #include "blk-cgroup-rwstat.h" 14 15 #define BFQ_IOPRIO_CLASSES 3 16 #define BFQ_CL_IDLE_TIMEOUT (HZ/5) 17 18 #define BFQ_MIN_WEIGHT 1 19 #define BFQ_MAX_WEIGHT 1000 20 #define BFQ_WEIGHT_CONVERSION_COEFF 10 21 22 #define BFQ_DEFAULT_QUEUE_IOPRIO 4 23 24 #define BFQ_WEIGHT_LEGACY_DFL 100 25 #define BFQ_DEFAULT_GRP_IOPRIO 0 26 #define BFQ_DEFAULT_GRP_CLASS IOPRIO_CLASS_BE 27 28 #define MAX_BFQQ_NAME_LENGTH 16 29 30 /* 31 * Soft real-time applications are extremely more latency sensitive 32 * than interactive ones. Over-raise the weight of the former to 33 * privilege them against the latter. 34 */ 35 #define BFQ_SOFTRT_WEIGHT_FACTOR 100 36 37 struct bfq_entity; 38 39 /** 40 * struct bfq_service_tree - per ioprio_class service tree. 41 * 42 * Each service tree represents a B-WF2Q+ scheduler on its own. Each 43 * ioprio_class has its own independent scheduler, and so its own 44 * bfq_service_tree. All the fields are protected by the queue lock 45 * of the containing bfqd. 46 */ 47 struct bfq_service_tree { 48 /* tree for active entities (i.e., those backlogged) */ 49 struct rb_root active; 50 /* tree for idle entities (i.e., not backlogged, with V < F_i)*/ 51 struct rb_root idle; 52 53 /* idle entity with minimum F_i */ 54 struct bfq_entity *first_idle; 55 /* idle entity with maximum F_i */ 56 struct bfq_entity *last_idle; 57 58 /* scheduler virtual time */ 59 u64 vtime; 60 /* scheduler weight sum; active and idle entities contribute to it */ 61 unsigned long wsum; 62 }; 63 64 /** 65 * struct bfq_sched_data - multi-class scheduler. 66 * 67 * bfq_sched_data is the basic scheduler queue. It supports three 68 * ioprio_classes, and can be used either as a toplevel queue or as an 69 * intermediate queue in a hierarchical setup. 70 * 71 * The supported ioprio_classes are the same as in CFQ, in descending 72 * priority order, IOPRIO_CLASS_RT, IOPRIO_CLASS_BE, IOPRIO_CLASS_IDLE. 73 * Requests from higher priority queues are served before all the 74 * requests from lower priority queues; among requests of the same 75 * queue requests are served according to B-WF2Q+. 76 * 77 * The schedule is implemented by the service trees, plus the field 78 * @next_in_service, which points to the entity on the active trees 79 * that will be served next, if 1) no changes in the schedule occurs 80 * before the current in-service entity is expired, 2) the in-service 81 * queue becomes idle when it expires, and 3) if the entity pointed by 82 * in_service_entity is not a queue, then the in-service child entity 83 * of the entity pointed by in_service_entity becomes idle on 84 * expiration. This peculiar definition allows for the following 85 * optimization, not yet exploited: while a given entity is still in 86 * service, we already know which is the best candidate for next 87 * service among the other active entities in the same parent 88 * entity. We can then quickly compare the timestamps of the 89 * in-service entity with those of such best candidate. 90 * 91 * All fields are protected by the lock of the containing bfqd. 92 */ 93 struct bfq_sched_data { 94 /* entity in service */ 95 struct bfq_entity *in_service_entity; 96 /* head-of-line entity (see comments above) */ 97 struct bfq_entity *next_in_service; 98 /* array of service trees, one per ioprio_class */ 99 struct bfq_service_tree service_tree[BFQ_IOPRIO_CLASSES]; 100 /* last time CLASS_IDLE was served */ 101 unsigned long bfq_class_idle_last_service; 102 103 }; 104 105 /** 106 * struct bfq_weight_counter - counter of the number of all active queues 107 * with a given weight. 108 */ 109 struct bfq_weight_counter { 110 unsigned int weight; /* weight of the queues this counter refers to */ 111 unsigned int num_active; /* nr of active queues with this weight */ 112 /* 113 * Weights tree member (see bfq_data's @queue_weights_tree) 114 */ 115 struct rb_node weights_node; 116 }; 117 118 /** 119 * struct bfq_entity - schedulable entity. 120 * 121 * A bfq_entity is used to represent either a bfq_queue (leaf node in the 122 * cgroup hierarchy) or a bfq_group into the upper level scheduler. Each 123 * entity belongs to the sched_data of the parent group in the cgroup 124 * hierarchy. Non-leaf entities have also their own sched_data, stored 125 * in @my_sched_data. 126 * 127 * Each entity stores independently its priority values; this would 128 * allow different weights on different devices, but this 129 * functionality is not exported to userspace by now. Priorities and 130 * weights are updated lazily, first storing the new values into the 131 * new_* fields, then setting the @prio_changed flag. As soon as 132 * there is a transition in the entity state that allows the priority 133 * update to take place the effective and the requested priority 134 * values are synchronized. 135 * 136 * Unless cgroups are used, the weight value is calculated from the 137 * ioprio to export the same interface as CFQ. When dealing with 138 * "well-behaved" queues (i.e., queues that do not spend too much 139 * time to consume their budget and have true sequential behavior, and 140 * when there are no external factors breaking anticipation) the 141 * relative weights at each level of the cgroups hierarchy should be 142 * guaranteed. All the fields are protected by the queue lock of the 143 * containing bfqd. 144 */ 145 struct bfq_entity { 146 /* service_tree member */ 147 struct rb_node rb_node; 148 149 /* 150 * Flag, true if the entity is on a tree (either the active or 151 * the idle one of its service_tree) or is in service. 152 */ 153 bool on_st_or_in_serv; 154 155 /* B-WF2Q+ start and finish timestamps [sectors/weight] */ 156 u64 start, finish; 157 158 /* tree the entity is enqueued into; %NULL if not on a tree */ 159 struct rb_root *tree; 160 161 /* 162 * minimum start time of the (active) subtree rooted at this 163 * entity; used for O(log N) lookups into active trees 164 */ 165 u64 min_start; 166 167 /* amount of service received during the last service slot */ 168 int service; 169 170 /* budget, used also to calculate F_i: F_i = S_i + @budget / @weight */ 171 int budget; 172 173 /* Number of requests allocated in the subtree of this entity */ 174 int allocated; 175 176 /* device weight, if non-zero, it overrides the default weight of 177 * bfq_group_data */ 178 int dev_weight; 179 /* weight of the queue */ 180 int weight; 181 /* next weight if a change is in progress */ 182 int new_weight; 183 184 /* original weight, used to implement weight boosting */ 185 int orig_weight; 186 187 /* parent entity, for hierarchical scheduling */ 188 struct bfq_entity *parent; 189 190 /* 191 * For non-leaf nodes in the hierarchy, the associated 192 * scheduler queue, %NULL on leaf nodes. 193 */ 194 struct bfq_sched_data *my_sched_data; 195 /* the scheduler queue this entity belongs to */ 196 struct bfq_sched_data *sched_data; 197 198 /* flag, set to request a weight, ioprio or ioprio_class change */ 199 int prio_changed; 200 201 /* flag, set if the entity is counted in groups_with_pending_reqs */ 202 bool in_groups_with_pending_reqs; 203 204 /* last child queue of entity created (for non-leaf entities) */ 205 struct bfq_queue *last_bfqq_created; 206 }; 207 208 struct bfq_group; 209 210 /** 211 * struct bfq_ttime - per process thinktime stats. 212 */ 213 struct bfq_ttime { 214 /* completion time of the last request */ 215 u64 last_end_request; 216 217 /* total process thinktime */ 218 u64 ttime_total; 219 /* number of thinktime samples */ 220 unsigned long ttime_samples; 221 /* average process thinktime */ 222 u64 ttime_mean; 223 }; 224 225 /** 226 * struct bfq_queue - leaf schedulable entity. 227 * 228 * A bfq_queue is a leaf request queue; it can be associated with an 229 * io_context or more, if it is async or shared between cooperating 230 * processes. @cgroup holds a reference to the cgroup, to be sure that it 231 * does not disappear while a bfqq still references it (mostly to avoid 232 * races between request issuing and task migration followed by cgroup 233 * destruction). 234 * All the fields are protected by the queue lock of the containing bfqd. 235 */ 236 struct bfq_queue { 237 /* reference counter */ 238 int ref; 239 /* counter of references from other queues for delayed stable merge */ 240 int stable_ref; 241 /* parent bfq_data */ 242 struct bfq_data *bfqd; 243 244 /* current ioprio and ioprio class */ 245 unsigned short ioprio, ioprio_class; 246 /* next ioprio and ioprio class if a change is in progress */ 247 unsigned short new_ioprio, new_ioprio_class; 248 249 /* last total-service-time sample, see bfq_update_inject_limit() */ 250 u64 last_serv_time_ns; 251 /* limit for request injection */ 252 unsigned int inject_limit; 253 /* last time the inject limit has been decreased, in jiffies */ 254 unsigned long decrease_time_jif; 255 256 /* 257 * Shared bfq_queue if queue is cooperating with one or more 258 * other queues. 259 */ 260 struct bfq_queue *new_bfqq; 261 /* request-position tree member (see bfq_group's @rq_pos_tree) */ 262 struct rb_node pos_node; 263 /* request-position tree root (see bfq_group's @rq_pos_tree) */ 264 struct rb_root *pos_root; 265 266 /* sorted list of pending requests */ 267 struct rb_root sort_list; 268 /* if fifo isn't expired, next request to serve */ 269 struct request *next_rq; 270 /* number of sync and async requests queued */ 271 int queued[2]; 272 /* number of pending metadata requests */ 273 int meta_pending; 274 /* fifo list of requests in sort_list */ 275 struct list_head fifo; 276 277 /* entity representing this queue in the scheduler */ 278 struct bfq_entity entity; 279 280 /* pointer to the weight counter associated with this entity */ 281 struct bfq_weight_counter *weight_counter; 282 283 /* maximum budget allowed from the feedback mechanism */ 284 int max_budget; 285 /* budget expiration (in jiffies) */ 286 unsigned long budget_timeout; 287 288 /* number of requests on the dispatch list or inside driver */ 289 int dispatched; 290 291 /* status flags */ 292 unsigned long flags; 293 294 /* node for active/idle bfqq list inside parent bfqd */ 295 struct list_head bfqq_list; 296 297 /* associated @bfq_ttime struct */ 298 struct bfq_ttime ttime; 299 300 /* when bfqq started to do I/O within the last observation window */ 301 u64 io_start_time; 302 /* how long bfqq has remained empty during the last observ. window */ 303 u64 tot_idle_time; 304 305 /* bit vector: a 1 for each seeky requests in history */ 306 u32 seek_history; 307 308 /* node for the device's burst list */ 309 struct hlist_node burst_list_node; 310 311 /* position of the last request enqueued */ 312 sector_t last_request_pos; 313 314 /* Number of consecutive pairs of request completion and 315 * arrival, such that the queue becomes idle after the 316 * completion, but the next request arrives within an idle 317 * time slice; used only if the queue's IO_bound flag has been 318 * cleared. 319 */ 320 unsigned int requests_within_timer; 321 322 /* pid of the process owning the queue, used for logging purposes */ 323 pid_t pid; 324 325 /* 326 * Pointer to the bfq_io_cq owning the bfq_queue, set to %NULL 327 * if the queue is shared. 328 */ 329 struct bfq_io_cq *bic; 330 331 /* current maximum weight-raising time for this queue */ 332 unsigned long wr_cur_max_time; 333 /* 334 * Minimum time instant such that, only if a new request is 335 * enqueued after this time instant in an idle @bfq_queue with 336 * no outstanding requests, then the task associated with the 337 * queue it is deemed as soft real-time (see the comments on 338 * the function bfq_bfqq_softrt_next_start()) 339 */ 340 unsigned long soft_rt_next_start; 341 /* 342 * Start time of the current weight-raising period if 343 * the @bfq-queue is being weight-raised, otherwise 344 * finish time of the last weight-raising period. 345 */ 346 unsigned long last_wr_start_finish; 347 /* factor by which the weight of this queue is multiplied */ 348 unsigned int wr_coeff; 349 /* 350 * Time of the last transition of the @bfq_queue from idle to 351 * backlogged. 352 */ 353 unsigned long last_idle_bklogged; 354 /* 355 * Cumulative service received from the @bfq_queue since the 356 * last transition from idle to backlogged. 357 */ 358 unsigned long service_from_backlogged; 359 /* 360 * Cumulative service received from the @bfq_queue since its 361 * last transition to weight-raised state. 362 */ 363 unsigned long service_from_wr; 364 365 /* 366 * Value of wr start time when switching to soft rt 367 */ 368 unsigned long wr_start_at_switch_to_srt; 369 370 unsigned long split_time; /* time of last split */ 371 372 unsigned long first_IO_time; /* time of first I/O for this queue */ 373 374 unsigned long creation_time; /* when this queue is created */ 375 376 /* max service rate measured so far */ 377 u32 max_service_rate; 378 379 /* 380 * Pointer to the waker queue for this queue, i.e., to the 381 * queue Q such that this queue happens to get new I/O right 382 * after some I/O request of Q is completed. For details, see 383 * the comments on the choice of the queue for injection in 384 * bfq_select_queue(). 385 */ 386 struct bfq_queue *waker_bfqq; 387 /* pointer to the curr. tentative waker queue, see bfq_check_waker() */ 388 struct bfq_queue *tentative_waker_bfqq; 389 /* number of times the same tentative waker has been detected */ 390 unsigned int num_waker_detections; 391 /* time when we started considering this waker */ 392 u64 waker_detection_started; 393 394 /* node for woken_list, see below */ 395 struct hlist_node woken_list_node; 396 /* 397 * Head of the list of the woken queues for this queue, i.e., 398 * of the list of the queues for which this queue is a waker 399 * queue. This list is used to reset the waker_bfqq pointer in 400 * the woken queues when this queue exits. 401 */ 402 struct hlist_head woken_list; 403 }; 404 405 /** 406 * struct bfq_io_cq - per (request_queue, io_context) structure. 407 */ 408 struct bfq_io_cq { 409 /* associated io_cq structure */ 410 struct io_cq icq; /* must be the first member */ 411 /* array of two process queues, the sync and the async */ 412 struct bfq_queue *bfqq[2]; 413 /* per (request_queue, blkcg) ioprio */ 414 int ioprio; 415 #ifdef CONFIG_BFQ_GROUP_IOSCHED 416 uint64_t blkcg_serial_nr; /* the current blkcg serial */ 417 #endif 418 /* 419 * Snapshot of the has_short_time flag before merging; taken 420 * to remember its value while the queue is merged, so as to 421 * be able to restore it in case of split. 422 */ 423 bool saved_has_short_ttime; 424 /* 425 * Same purpose as the previous two fields for the I/O bound 426 * classification of a queue. 427 */ 428 bool saved_IO_bound; 429 430 u64 saved_io_start_time; 431 u64 saved_tot_idle_time; 432 433 /* 434 * Same purpose as the previous fields for the value of the 435 * field keeping the queue's belonging to a large burst 436 */ 437 bool saved_in_large_burst; 438 /* 439 * True if the queue belonged to a burst list before its merge 440 * with another cooperating queue. 441 */ 442 bool was_in_burst_list; 443 444 /* 445 * Save the weight when a merge occurs, to be able 446 * to restore it in case of split. If the weight is not 447 * correctly resumed when the queue is recycled, 448 * then the weight of the recycled queue could differ 449 * from the weight of the original queue. 450 */ 451 unsigned int saved_weight; 452 453 /* 454 * Similar to previous fields: save wr information. 455 */ 456 unsigned long saved_wr_coeff; 457 unsigned long saved_last_wr_start_finish; 458 unsigned long saved_service_from_wr; 459 unsigned long saved_wr_start_at_switch_to_srt; 460 unsigned int saved_wr_cur_max_time; 461 struct bfq_ttime saved_ttime; 462 463 /* Save also injection state */ 464 u64 saved_last_serv_time_ns; 465 unsigned int saved_inject_limit; 466 unsigned long saved_decrease_time_jif; 467 468 /* candidate queue for a stable merge (due to close creation time) */ 469 struct bfq_queue *stable_merge_bfqq; 470 471 bool stably_merged; /* non splittable if true */ 472 }; 473 474 /** 475 * struct bfq_data - per-device data structure. 476 * 477 * All the fields are protected by @lock. 478 */ 479 struct bfq_data { 480 /* device request queue */ 481 struct request_queue *queue; 482 /* dispatch queue */ 483 struct list_head dispatch; 484 485 /* root bfq_group for the device */ 486 struct bfq_group *root_group; 487 488 /* 489 * rbtree of weight counters of @bfq_queues, sorted by 490 * weight. Used to keep track of whether all @bfq_queues have 491 * the same weight. The tree contains one counter for each 492 * distinct weight associated to some active and not 493 * weight-raised @bfq_queue (see the comments to the functions 494 * bfq_weights_tree_[add|remove] for further details). 495 */ 496 struct rb_root_cached queue_weights_tree; 497 498 /* 499 * Number of groups with at least one descendant process that 500 * has at least one request waiting for completion. Note that 501 * this accounts for also requests already dispatched, but not 502 * yet completed. Therefore this number of groups may differ 503 * (be larger) than the number of active groups, as a group is 504 * considered active only if its corresponding entity has 505 * descendant queues with at least one request queued. This 506 * number is used to decide whether a scenario is symmetric. 507 * For a detailed explanation see comments on the computation 508 * of the variable asymmetric_scenario in the function 509 * bfq_better_to_idle(). 510 * 511 * However, it is hard to compute this number exactly, for 512 * groups with multiple descendant processes. Consider a group 513 * that is inactive, i.e., that has no descendant process with 514 * pending I/O inside BFQ queues. Then suppose that 515 * num_groups_with_pending_reqs is still accounting for this 516 * group, because the group has descendant processes with some 517 * I/O request still in flight. num_groups_with_pending_reqs 518 * should be decremented when the in-flight request of the 519 * last descendant process is finally completed (assuming that 520 * nothing else has changed for the group in the meantime, in 521 * terms of composition of the group and active/inactive state of child 522 * groups and processes). To accomplish this, an additional 523 * pending-request counter must be added to entities, and must 524 * be updated correctly. To avoid this additional field and operations, 525 * we resort to the following tradeoff between simplicity and 526 * accuracy: for an inactive group that is still counted in 527 * num_groups_with_pending_reqs, we decrement 528 * num_groups_with_pending_reqs when the first descendant 529 * process of the group remains with no request waiting for 530 * completion. 531 * 532 * Even this simpler decrement strategy requires a little 533 * carefulness: to avoid multiple decrements, we flag a group, 534 * more precisely an entity representing a group, as still 535 * counted in num_groups_with_pending_reqs when it becomes 536 * inactive. Then, when the first descendant queue of the 537 * entity remains with no request waiting for completion, 538 * num_groups_with_pending_reqs is decremented, and this flag 539 * is reset. After this flag is reset for the entity, 540 * num_groups_with_pending_reqs won't be decremented any 541 * longer in case a new descendant queue of the entity remains 542 * with no request waiting for completion. 543 */ 544 unsigned int num_groups_with_pending_reqs; 545 546 /* 547 * Per-class (RT, BE, IDLE) number of bfq_queues containing 548 * requests (including the queue in service, even if it is 549 * idling). 550 */ 551 unsigned int busy_queues[3]; 552 /* number of weight-raised busy @bfq_queues */ 553 int wr_busy_queues; 554 /* number of queued requests */ 555 int queued; 556 /* number of requests dispatched and waiting for completion */ 557 int rq_in_driver; 558 559 /* true if the device is non rotational and performs queueing */ 560 bool nonrot_with_queueing; 561 562 /* 563 * Maximum number of requests in driver in the last 564 * @hw_tag_samples completed requests. 565 */ 566 int max_rq_in_driver; 567 /* number of samples used to calculate hw_tag */ 568 int hw_tag_samples; 569 /* flag set to one if the driver is showing a queueing behavior */ 570 int hw_tag; 571 572 /* number of budgets assigned */ 573 int budgets_assigned; 574 575 /* 576 * Timer set when idling (waiting) for the next request from 577 * the queue in service. 578 */ 579 struct hrtimer idle_slice_timer; 580 581 /* bfq_queue in service */ 582 struct bfq_queue *in_service_queue; 583 584 /* on-disk position of the last served request */ 585 sector_t last_position; 586 587 /* position of the last served request for the in-service queue */ 588 sector_t in_serv_last_pos; 589 590 /* time of last request completion (ns) */ 591 u64 last_completion; 592 593 /* bfqq owning the last completed rq */ 594 struct bfq_queue *last_completed_rq_bfqq; 595 596 /* last bfqq created, among those in the root group */ 597 struct bfq_queue *last_bfqq_created; 598 599 /* time of last transition from empty to non-empty (ns) */ 600 u64 last_empty_occupied_ns; 601 602 /* 603 * Flag set to activate the sampling of the total service time 604 * of a just-arrived first I/O request (see 605 * bfq_update_inject_limit()). This will cause the setting of 606 * waited_rq when the request is finally dispatched. 607 */ 608 bool wait_dispatch; 609 /* 610 * If set, then bfq_update_inject_limit() is invoked when 611 * waited_rq is eventually completed. 612 */ 613 struct request *waited_rq; 614 /* 615 * True if some request has been injected during the last service hole. 616 */ 617 bool rqs_injected; 618 619 /* time of first rq dispatch in current observation interval (ns) */ 620 u64 first_dispatch; 621 /* time of last rq dispatch in current observation interval (ns) */ 622 u64 last_dispatch; 623 624 /* beginning of the last budget */ 625 ktime_t last_budget_start; 626 /* beginning of the last idle slice */ 627 ktime_t last_idling_start; 628 unsigned long last_idling_start_jiffies; 629 630 /* number of samples in current observation interval */ 631 int peak_rate_samples; 632 /* num of samples of seq dispatches in current observation interval */ 633 u32 sequential_samples; 634 /* total num of sectors transferred in current observation interval */ 635 u64 tot_sectors_dispatched; 636 /* max rq size seen during current observation interval (sectors) */ 637 u32 last_rq_max_size; 638 /* time elapsed from first dispatch in current observ. interval (us) */ 639 u64 delta_from_first; 640 /* 641 * Current estimate of the device peak rate, measured in 642 * [(sectors/usec) / 2^BFQ_RATE_SHIFT]. The left-shift by 643 * BFQ_RATE_SHIFT is performed to increase precision in 644 * fixed-point calculations. 645 */ 646 u32 peak_rate; 647 648 /* maximum budget allotted to a bfq_queue before rescheduling */ 649 int bfq_max_budget; 650 651 /* list of all the bfq_queues active on the device */ 652 struct list_head active_list; 653 /* list of all the bfq_queues idle on the device */ 654 struct list_head idle_list; 655 656 /* 657 * Timeout for async/sync requests; when it fires, requests 658 * are served in fifo order. 659 */ 660 u64 bfq_fifo_expire[2]; 661 /* weight of backward seeks wrt forward ones */ 662 unsigned int bfq_back_penalty; 663 /* maximum allowed backward seek */ 664 unsigned int bfq_back_max; 665 /* maximum idling time */ 666 u32 bfq_slice_idle; 667 668 /* user-configured max budget value (0 for auto-tuning) */ 669 int bfq_user_max_budget; 670 /* 671 * Timeout for bfq_queues to consume their budget; used to 672 * prevent seeky queues from imposing long latencies to 673 * sequential or quasi-sequential ones (this also implies that 674 * seeky queues cannot receive guarantees in the service 675 * domain; after a timeout they are charged for the time they 676 * have been in service, to preserve fairness among them, but 677 * without service-domain guarantees). 678 */ 679 unsigned int bfq_timeout; 680 681 /* 682 * Force device idling whenever needed to provide accurate 683 * service guarantees, without caring about throughput 684 * issues. CAVEAT: this may even increase latencies, in case 685 * of useless idling for processes that did stop doing I/O. 686 */ 687 bool strict_guarantees; 688 689 /* 690 * Last time at which a queue entered the current burst of 691 * queues being activated shortly after each other; for more 692 * details about this and the following parameters related to 693 * a burst of activations, see the comments on the function 694 * bfq_handle_burst. 695 */ 696 unsigned long last_ins_in_burst; 697 /* 698 * Reference time interval used to decide whether a queue has 699 * been activated shortly after @last_ins_in_burst. 700 */ 701 unsigned long bfq_burst_interval; 702 /* number of queues in the current burst of queue activations */ 703 int burst_size; 704 705 /* common parent entity for the queues in the burst */ 706 struct bfq_entity *burst_parent_entity; 707 /* Maximum burst size above which the current queue-activation 708 * burst is deemed as 'large'. 709 */ 710 unsigned long bfq_large_burst_thresh; 711 /* true if a large queue-activation burst is in progress */ 712 bool large_burst; 713 /* 714 * Head of the burst list (as for the above fields, more 715 * details in the comments on the function bfq_handle_burst). 716 */ 717 struct hlist_head burst_list; 718 719 /* if set to true, low-latency heuristics are enabled */ 720 bool low_latency; 721 /* 722 * Maximum factor by which the weight of a weight-raised queue 723 * is multiplied. 724 */ 725 unsigned int bfq_wr_coeff; 726 /* maximum duration of a weight-raising period (jiffies) */ 727 unsigned int bfq_wr_max_time; 728 729 /* Maximum weight-raising duration for soft real-time processes */ 730 unsigned int bfq_wr_rt_max_time; 731 /* 732 * Minimum idle period after which weight-raising may be 733 * reactivated for a queue (in jiffies). 734 */ 735 unsigned int bfq_wr_min_idle_time; 736 /* 737 * Minimum period between request arrivals after which 738 * weight-raising may be reactivated for an already busy async 739 * queue (in jiffies). 740 */ 741 unsigned long bfq_wr_min_inter_arr_async; 742 743 /* Max service-rate for a soft real-time queue, in sectors/sec */ 744 unsigned int bfq_wr_max_softrt_rate; 745 /* 746 * Cached value of the product ref_rate*ref_wr_duration, used 747 * for computing the maximum duration of weight raising 748 * automatically. 749 */ 750 u64 rate_dur_prod; 751 752 /* fallback dummy bfqq for extreme OOM conditions */ 753 struct bfq_queue oom_bfqq; 754 755 spinlock_t lock; 756 757 /* 758 * bic associated with the task issuing current bio for 759 * merging. This and the next field are used as a support to 760 * be able to perform the bic lookup, needed by bio-merge 761 * functions, before the scheduler lock is taken, and thus 762 * avoid taking the request-queue lock while the scheduler 763 * lock is being held. 764 */ 765 struct bfq_io_cq *bio_bic; 766 /* bfqq associated with the task issuing current bio for merging */ 767 struct bfq_queue *bio_bfqq; 768 769 /* 770 * Depth limits used in bfq_limit_depth (see comments on the 771 * function) 772 */ 773 unsigned int word_depths[2][2]; 774 unsigned int full_depth_shift; 775 }; 776 777 enum bfqq_state_flags { 778 BFQQF_just_created = 0, /* queue just allocated */ 779 BFQQF_busy, /* has requests or is in service */ 780 BFQQF_wait_request, /* waiting for a request */ 781 BFQQF_non_blocking_wait_rq, /* 782 * waiting for a request 783 * without idling the device 784 */ 785 BFQQF_fifo_expire, /* FIFO checked in this slice */ 786 BFQQF_has_short_ttime, /* queue has a short think time */ 787 BFQQF_sync, /* synchronous queue */ 788 BFQQF_IO_bound, /* 789 * bfqq has timed-out at least once 790 * having consumed at most 2/10 of 791 * its budget 792 */ 793 BFQQF_in_large_burst, /* 794 * bfqq activated in a large burst, 795 * see comments to bfq_handle_burst. 796 */ 797 BFQQF_softrt_update, /* 798 * may need softrt-next-start 799 * update 800 */ 801 BFQQF_coop, /* bfqq is shared */ 802 BFQQF_split_coop, /* shared bfqq will be split */ 803 }; 804 805 #define BFQ_BFQQ_FNS(name) \ 806 void bfq_mark_bfqq_##name(struct bfq_queue *bfqq); \ 807 void bfq_clear_bfqq_##name(struct bfq_queue *bfqq); \ 808 int bfq_bfqq_##name(const struct bfq_queue *bfqq); 809 810 BFQ_BFQQ_FNS(just_created); 811 BFQ_BFQQ_FNS(busy); 812 BFQ_BFQQ_FNS(wait_request); 813 BFQ_BFQQ_FNS(non_blocking_wait_rq); 814 BFQ_BFQQ_FNS(fifo_expire); 815 BFQ_BFQQ_FNS(has_short_ttime); 816 BFQ_BFQQ_FNS(sync); 817 BFQ_BFQQ_FNS(IO_bound); 818 BFQ_BFQQ_FNS(in_large_burst); 819 BFQ_BFQQ_FNS(coop); 820 BFQ_BFQQ_FNS(split_coop); 821 BFQ_BFQQ_FNS(softrt_update); 822 #undef BFQ_BFQQ_FNS 823 824 /* Expiration reasons. */ 825 enum bfqq_expiration { 826 BFQQE_TOO_IDLE = 0, /* 827 * queue has been idling for 828 * too long 829 */ 830 BFQQE_BUDGET_TIMEOUT, /* budget took too long to be used */ 831 BFQQE_BUDGET_EXHAUSTED, /* budget consumed */ 832 BFQQE_NO_MORE_REQUESTS, /* the queue has no more requests */ 833 BFQQE_PREEMPTED /* preemption in progress */ 834 }; 835 836 struct bfq_stat { 837 struct percpu_counter cpu_cnt; 838 atomic64_t aux_cnt; 839 }; 840 841 struct bfqg_stats { 842 /* basic stats */ 843 struct blkg_rwstat bytes; 844 struct blkg_rwstat ios; 845 #ifdef CONFIG_BFQ_CGROUP_DEBUG 846 /* number of ios merged */ 847 struct blkg_rwstat merged; 848 /* total time spent on device in ns, may not be accurate w/ queueing */ 849 struct blkg_rwstat service_time; 850 /* total time spent waiting in scheduler queue in ns */ 851 struct blkg_rwstat wait_time; 852 /* number of IOs queued up */ 853 struct blkg_rwstat queued; 854 /* total disk time and nr sectors dispatched by this group */ 855 struct bfq_stat time; 856 /* sum of number of ios queued across all samples */ 857 struct bfq_stat avg_queue_size_sum; 858 /* count of samples taken for average */ 859 struct bfq_stat avg_queue_size_samples; 860 /* how many times this group has been removed from service tree */ 861 struct bfq_stat dequeue; 862 /* total time spent waiting for it to be assigned a timeslice. */ 863 struct bfq_stat group_wait_time; 864 /* time spent idling for this blkcg_gq */ 865 struct bfq_stat idle_time; 866 /* total time with empty current active q with other requests queued */ 867 struct bfq_stat empty_time; 868 /* fields after this shouldn't be cleared on stat reset */ 869 u64 start_group_wait_time; 870 u64 start_idle_time; 871 u64 start_empty_time; 872 uint16_t flags; 873 #endif /* CONFIG_BFQ_CGROUP_DEBUG */ 874 }; 875 876 #ifdef CONFIG_BFQ_GROUP_IOSCHED 877 878 /* 879 * struct bfq_group_data - per-blkcg storage for the blkio subsystem. 880 * 881 * @ps: @blkcg_policy_storage that this structure inherits 882 * @weight: weight of the bfq_group 883 */ 884 struct bfq_group_data { 885 /* must be the first member */ 886 struct blkcg_policy_data pd; 887 888 unsigned int weight; 889 }; 890 891 /** 892 * struct bfq_group - per (device, cgroup) data structure. 893 * @entity: schedulable entity to insert into the parent group sched_data. 894 * @sched_data: own sched_data, to contain child entities (they may be 895 * both bfq_queues and bfq_groups). 896 * @bfqd: the bfq_data for the device this group acts upon. 897 * @async_bfqq: array of async queues for all the tasks belonging to 898 * the group, one queue per ioprio value per ioprio_class, 899 * except for the idle class that has only one queue. 900 * @async_idle_bfqq: async queue for the idle class (ioprio is ignored). 901 * @my_entity: pointer to @entity, %NULL for the toplevel group; used 902 * to avoid too many special cases during group creation/ 903 * migration. 904 * @stats: stats for this bfqg. 905 * @active_entities: number of active entities belonging to the group; 906 * unused for the root group. Used to know whether there 907 * are groups with more than one active @bfq_entity 908 * (see the comments to the function 909 * bfq_bfqq_may_idle()). 910 * @rq_pos_tree: rbtree sorted by next_request position, used when 911 * determining if two or more queues have interleaving 912 * requests (see bfq_find_close_cooperator()). 913 * 914 * Each (device, cgroup) pair has its own bfq_group, i.e., for each cgroup 915 * there is a set of bfq_groups, each one collecting the lower-level 916 * entities belonging to the group that are acting on the same device. 917 * 918 * Locking works as follows: 919 * o @bfqd is protected by the queue lock, RCU is used to access it 920 * from the readers. 921 * o All the other fields are protected by the @bfqd queue lock. 922 */ 923 struct bfq_group { 924 /* must be the first member */ 925 struct blkg_policy_data pd; 926 927 /* cached path for this blkg (see comments in bfq_bic_update_cgroup) */ 928 char blkg_path[128]; 929 930 /* reference counter (see comments in bfq_bic_update_cgroup) */ 931 int ref; 932 933 struct bfq_entity entity; 934 struct bfq_sched_data sched_data; 935 936 void *bfqd; 937 938 struct bfq_queue *async_bfqq[2][IOPRIO_NR_LEVELS]; 939 struct bfq_queue *async_idle_bfqq; 940 941 struct bfq_entity *my_entity; 942 943 int active_entities; 944 945 struct rb_root rq_pos_tree; 946 947 struct bfqg_stats stats; 948 }; 949 950 #else 951 struct bfq_group { 952 struct bfq_entity entity; 953 struct bfq_sched_data sched_data; 954 955 struct bfq_queue *async_bfqq[2][IOPRIO_NR_LEVELS]; 956 struct bfq_queue *async_idle_bfqq; 957 958 struct rb_root rq_pos_tree; 959 }; 960 #endif 961 962 /* --------------- main algorithm interface ----------------- */ 963 964 #define BFQ_SERVICE_TREE_INIT ((struct bfq_service_tree) \ 965 { RB_ROOT, RB_ROOT, NULL, NULL, 0, 0 }) 966 967 extern const int bfq_timeout; 968 969 struct bfq_queue *bic_to_bfqq(struct bfq_io_cq *bic, bool is_sync); 970 void bic_set_bfqq(struct bfq_io_cq *bic, struct bfq_queue *bfqq, bool is_sync); 971 struct bfq_data *bic_to_bfqd(struct bfq_io_cq *bic); 972 void bfq_pos_tree_add_move(struct bfq_data *bfqd, struct bfq_queue *bfqq); 973 void bfq_weights_tree_add(struct bfq_data *bfqd, struct bfq_queue *bfqq, 974 struct rb_root_cached *root); 975 void __bfq_weights_tree_remove(struct bfq_data *bfqd, 976 struct bfq_queue *bfqq, 977 struct rb_root_cached *root); 978 void bfq_weights_tree_remove(struct bfq_data *bfqd, 979 struct bfq_queue *bfqq); 980 void bfq_bfqq_expire(struct bfq_data *bfqd, struct bfq_queue *bfqq, 981 bool compensate, enum bfqq_expiration reason); 982 void bfq_put_queue(struct bfq_queue *bfqq); 983 void bfq_end_wr_async_queues(struct bfq_data *bfqd, struct bfq_group *bfqg); 984 void bfq_release_process_ref(struct bfq_data *bfqd, struct bfq_queue *bfqq); 985 void bfq_schedule_dispatch(struct bfq_data *bfqd); 986 void bfq_put_async_queues(struct bfq_data *bfqd, struct bfq_group *bfqg); 987 988 /* ------------ end of main algorithm interface -------------- */ 989 990 /* ---------------- cgroups-support interface ---------------- */ 991 992 void bfqg_stats_update_legacy_io(struct request_queue *q, struct request *rq); 993 void bfqg_stats_update_io_add(struct bfq_group *bfqg, struct bfq_queue *bfqq, 994 unsigned int op); 995 void bfqg_stats_update_io_remove(struct bfq_group *bfqg, unsigned int op); 996 void bfqg_stats_update_io_merged(struct bfq_group *bfqg, unsigned int op); 997 void bfqg_stats_update_completion(struct bfq_group *bfqg, u64 start_time_ns, 998 u64 io_start_time_ns, unsigned int op); 999 void bfqg_stats_update_dequeue(struct bfq_group *bfqg); 1000 void bfqg_stats_set_start_empty_time(struct bfq_group *bfqg); 1001 void bfqg_stats_update_idle_time(struct bfq_group *bfqg); 1002 void bfqg_stats_set_start_idle_time(struct bfq_group *bfqg); 1003 void bfqg_stats_update_avg_queue_size(struct bfq_group *bfqg); 1004 void bfq_bfqq_move(struct bfq_data *bfqd, struct bfq_queue *bfqq, 1005 struct bfq_group *bfqg); 1006 1007 void bfq_init_entity(struct bfq_entity *entity, struct bfq_group *bfqg); 1008 void bfq_bic_update_cgroup(struct bfq_io_cq *bic, struct bio *bio); 1009 void bfq_end_wr_async(struct bfq_data *bfqd); 1010 struct bfq_group *bfq_find_set_group(struct bfq_data *bfqd, 1011 struct blkcg *blkcg); 1012 struct blkcg_gq *bfqg_to_blkg(struct bfq_group *bfqg); 1013 struct bfq_group *bfqq_group(struct bfq_queue *bfqq); 1014 struct bfq_group *bfq_create_group_hierarchy(struct bfq_data *bfqd, int node); 1015 void bfqg_and_blkg_put(struct bfq_group *bfqg); 1016 1017 #ifdef CONFIG_BFQ_GROUP_IOSCHED 1018 extern struct cftype bfq_blkcg_legacy_files[]; 1019 extern struct cftype bfq_blkg_files[]; 1020 extern struct blkcg_policy blkcg_policy_bfq; 1021 #endif 1022 1023 /* ------------- end of cgroups-support interface ------------- */ 1024 1025 /* - interface of the internal hierarchical B-WF2Q+ scheduler - */ 1026 1027 #ifdef CONFIG_BFQ_GROUP_IOSCHED 1028 /* both next loops stop at one of the child entities of the root group */ 1029 #define for_each_entity(entity) \ 1030 for (; entity ; entity = entity->parent) 1031 1032 /* 1033 * For each iteration, compute parent in advance, so as to be safe if 1034 * entity is deallocated during the iteration. Such a deallocation may 1035 * happen as a consequence of a bfq_put_queue that frees the bfq_queue 1036 * containing entity. 1037 */ 1038 #define for_each_entity_safe(entity, parent) \ 1039 for (; entity && ({ parent = entity->parent; 1; }); entity = parent) 1040 1041 #else /* CONFIG_BFQ_GROUP_IOSCHED */ 1042 /* 1043 * Next two macros are fake loops when cgroups support is not 1044 * enabled. I fact, in such a case, there is only one level to go up 1045 * (to reach the root group). 1046 */ 1047 #define for_each_entity(entity) \ 1048 for (; entity ; entity = NULL) 1049 1050 #define for_each_entity_safe(entity, parent) \ 1051 for (parent = NULL; entity ; entity = parent) 1052 #endif /* CONFIG_BFQ_GROUP_IOSCHED */ 1053 1054 struct bfq_group *bfq_bfqq_to_bfqg(struct bfq_queue *bfqq); 1055 struct bfq_queue *bfq_entity_to_bfqq(struct bfq_entity *entity); 1056 unsigned int bfq_tot_busy_queues(struct bfq_data *bfqd); 1057 struct bfq_service_tree *bfq_entity_service_tree(struct bfq_entity *entity); 1058 struct bfq_entity *bfq_entity_of(struct rb_node *node); 1059 unsigned short bfq_ioprio_to_weight(int ioprio); 1060 void bfq_put_idle_entity(struct bfq_service_tree *st, 1061 struct bfq_entity *entity); 1062 struct bfq_service_tree * 1063 __bfq_entity_update_weight_prio(struct bfq_service_tree *old_st, 1064 struct bfq_entity *entity, 1065 bool update_class_too); 1066 void bfq_bfqq_served(struct bfq_queue *bfqq, int served); 1067 void bfq_bfqq_charge_time(struct bfq_data *bfqd, struct bfq_queue *bfqq, 1068 unsigned long time_ms); 1069 bool __bfq_deactivate_entity(struct bfq_entity *entity, 1070 bool ins_into_idle_tree); 1071 bool next_queue_may_preempt(struct bfq_data *bfqd); 1072 struct bfq_queue *bfq_get_next_queue(struct bfq_data *bfqd); 1073 bool __bfq_bfqd_reset_in_service(struct bfq_data *bfqd); 1074 void bfq_deactivate_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq, 1075 bool ins_into_idle_tree, bool expiration); 1076 void bfq_activate_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq); 1077 void bfq_requeue_bfqq(struct bfq_data *bfqd, struct bfq_queue *bfqq, 1078 bool expiration); 1079 void bfq_del_bfqq_busy(struct bfq_data *bfqd, struct bfq_queue *bfqq, 1080 bool expiration); 1081 void bfq_add_bfqq_busy(struct bfq_data *bfqd, struct bfq_queue *bfqq); 1082 1083 /* --------------- end of interface of B-WF2Q+ ---------------- */ 1084 1085 /* Logging facilities. */ 1086 static inline void bfq_bfqq_name(struct bfq_queue *bfqq, char *str, int len) 1087 { 1088 char type = bfq_bfqq_sync(bfqq) ? 'S' : 'A'; 1089 1090 if (bfqq->pid != -1) 1091 snprintf(str, len, "bfq%d%c", bfqq->pid, type); 1092 else 1093 snprintf(str, len, "bfqSHARED-%c", type); 1094 } 1095 1096 #ifdef CONFIG_BFQ_GROUP_IOSCHED 1097 struct bfq_group *bfqq_group(struct bfq_queue *bfqq); 1098 1099 #define bfq_log_bfqq(bfqd, bfqq, fmt, args...) do { \ 1100 char pid_str[MAX_BFQQ_NAME_LENGTH]; \ 1101 if (likely(!blk_trace_note_message_enabled((bfqd)->queue))) \ 1102 break; \ 1103 bfq_bfqq_name((bfqq), pid_str, MAX_BFQQ_NAME_LENGTH); \ 1104 blk_add_cgroup_trace_msg((bfqd)->queue, \ 1105 bfqg_to_blkg(bfqq_group(bfqq))->blkcg, \ 1106 "%s " fmt, pid_str, ##args); \ 1107 } while (0) 1108 1109 #define bfq_log_bfqg(bfqd, bfqg, fmt, args...) do { \ 1110 blk_add_cgroup_trace_msg((bfqd)->queue, \ 1111 bfqg_to_blkg(bfqg)->blkcg, fmt, ##args); \ 1112 } while (0) 1113 1114 #else /* CONFIG_BFQ_GROUP_IOSCHED */ 1115 1116 #define bfq_log_bfqq(bfqd, bfqq, fmt, args...) do { \ 1117 char pid_str[MAX_BFQQ_NAME_LENGTH]; \ 1118 if (likely(!blk_trace_note_message_enabled((bfqd)->queue))) \ 1119 break; \ 1120 bfq_bfqq_name((bfqq), pid_str, MAX_BFQQ_NAME_LENGTH); \ 1121 blk_add_trace_msg((bfqd)->queue, "%s " fmt, pid_str, ##args); \ 1122 } while (0) 1123 #define bfq_log_bfqg(bfqd, bfqg, fmt, args...) do {} while (0) 1124 1125 #endif /* CONFIG_BFQ_GROUP_IOSCHED */ 1126 1127 #define bfq_log(bfqd, fmt, args...) \ 1128 blk_add_trace_msg((bfqd)->queue, "bfq " fmt, ##args) 1129 1130 #endif /* _BFQ_H */ 1131